Bump version and changelog

Make /keys/changes a bit more performant

.gitignore

@@ -24,10 +24,10 @@ homeserver*.yaml
 .coverage
 htmlcov
 
-demo/*.db
-demo/*.log
-demo/*.log.*
-demo/*.pid
+demo/*/*.db
+demo/*/*.log
+demo/*/*.log.*
+demo/*/*.pid
 demo/media_store.*
 demo/etc
 

.travis.yml

@@ -0,0 +1,17 @@
+sudo: false
+language: python
+python: 2.7
+
+# tell travis to cache ~/.cache/pip
+cache: pip
+
+env:
+  - TOX_ENV=packaging
+  - TOX_ENV=pep8
+  - TOX_ENV=py27
+
+install:
+  - pip install tox
+
+script:
+  - tox -e $TOX_ENV

CHANGES.rst

@@ -1,3 +1,613 @@
+Changes in synapse v0.19.0-rc3 (2017-02-03)
+===========================================
+
+* Fix email push in pusher worker (PR #1875)
+* Make presence.get_new_events a bit faster (PR #1876)
+* Make /keys/changes a bit more performant (PR #1877)
+
+
+Changes in synapse v0.19.0-rc2 (2017-02-02)
+===========================================
+
+* Include newly joined users in /keys/changes API (PR #1872)
+
+
+Changes in synapse v0.19.0-rc1 (2017-02-02)
+===========================================
+
+Features:
+
+* Add support for specifying multiple bind addresses (PR #1709, #1712, #1795,
+  #1835). Thanks to @kyrias!
+* Add /account/3pid/delete endpoint (PR #1714)
+* Add config option to configure the Riot URL used in notification emails (PR
+  #1811). Thanks to @aperezdc!
+* Add username and password config options for turn server (PR #1832). Thanks
+  to @xsteadfastx!
+* Implement device lists updates over federation (PR #1857, #1861, #1864)
+* Implement /keys/changes (PR #1869, #1872)
+
+
+Changes:
+
+* Improve IPv6 support (PR #1696). Thanks to @kyrias and @glyph!
+* Log which files we saved attachments to in the media_repository (PR #1791)
+* Linearize updates to membership via PUT /state/ to better handle multiple
+  joins (PR #1787)
+* Limit number of entries to prefill from cache on startup (PR #1792)
+* Remove full_twisted_stacktraces option (PR #1802)
+* Measure size of some caches by sum of the size of cached values (PR #1815)
+* Measure metrics of string_cache (PR #1821)
+* Reduce logging verbosity (PR #1822, #1823, #1824)
+* Don't clobber a displayname or avatar_url if provided by an m.room.member
+  event (PR #1852)
+* Better handle 401/404 response for federation /send/ (PR #1866, #1871)
+
+
+Fixes:
+
+* Fix ability to change password to a non-ascii one (PR #1711)
+* Fix push getting stuck due to looking at the wrong view of state (PR #1820)
+* Fix email address comparison to be case insensitive (PR #1827)
+* Fix occasional inconsistencies of room membership (PR #1836, #1840)
+
+
+Performance:
+
+* Don't block messages sending on bumping presence (PR #1789)
+* Change device_inbox stream index to include user (PR #1793)
+* Optimise state resolution (PR #1818)
+* Use DB cache of joined users for presence (PR #1862)
+* Add an index to make membership queries faster (PR #1867)
+
+
+Changes in synapse v0.18.7 (2017-01-09)
+=======================================
+
+No changes from v0.18.7-rc2
+
+
+Changes in synapse v0.18.7-rc2 (2017-01-07)
+===========================================
+
+Bug fixes:
+
+* Fix error in rc1's discarding invalid inbound traffic logic that was
+  incorrectly discarding missing events
+
+
+Changes in synapse v0.18.7-rc1 (2017-01-06)
+===========================================
+
+Bug fixes:
+
+* Fix error in #PR 1764 to actually fix the nightmare #1753 bug.
+* Improve deadlock logging further
+* Discard inbound federation traffic from invalid domains, to immunise
+  against #1753
+
+
+Changes in synapse v0.18.6 (2017-01-06)
+=======================================
+
+Bug fixes:
+
+* Fix bug when checking if a guest user is allowed to join a room (PR #1772)
+  Thanks to Patrik Oldsberg for diagnosing and the fix!
+
+
+Changes in synapse v0.18.6-rc3 (2017-01-05)
+===========================================
+
+Bug fixes:
+
+* Fix bug where we failed to send ban events to the banned server (PR #1758)
+* Fix bug where we sent event that didn't originate on this server to
+  other servers (PR #1764)
+* Fix bug where processing an event from a remote server took a long time
+  because we were making long HTTP requests (PR #1765, PR #1744)
+
+Changes:
+
+* Improve logging for debugging deadlocks (PR #1766, PR #1767)
+
+
+Changes in synapse v0.18.6-rc2 (2016-12-30)
+===========================================
+
+Bug fixes:
+
+* Fix memory leak in twisted by initialising logging correctly (PR #1731)
+* Fix bug where fetching missing events took an unacceptable amount of time in
+  large rooms (PR #1734)
+
+
+Changes in synapse v0.18.6-rc1 (2016-12-29)
+===========================================
+
+Bug fixes:
+
+* Make sure that outbound connections are closed (PR #1725)
+
+
+Changes in synapse v0.18.5 (2016-12-16)
+=======================================
+
+Bug fixes:
+
+* Fix federation /backfill returning events it shouldn't (PR #1700)
+* Fix crash in url preview (PR #1701)
+
+
+Changes in synapse v0.18.5-rc3 (2016-12-13)
+===========================================
+
+Features:
+
+* Add support for E2E for guests (PR #1653)
+* Add new API appservice specific public room list (PR #1676)
+* Add new room membership APIs (PR #1680)
+
+
+Changes:
+
+* Enable guest access for private rooms by default (PR #653)
+* Limit the number of events that can be created on a given room concurrently
+  (PR #1620)
+* Log the args that we have on UI auth completion (PR #1649)
+* Stop generating refresh_tokens (PR #1654)
+* Stop putting a time caveat on access tokens (PR #1656)
+* Remove unspecced GET endpoints for e2e keys (PR #1694)
+
+
+Bug fixes:
+
+* Fix handling of 500 and 429's over federation (PR #1650)
+* Fix Content-Type header parsing (PR #1660)
+* Fix error when previewing sites that include unicode, thanks to kyrias (PR
+  #1664)
+* Fix some cases where we drop read receipts (PR #1678)
+* Fix bug where calls to ``/sync`` didn't correctly timeout (PR #1683)
+* Fix bug where E2E key query would fail if a single remote host failed (PR
+  #1686)
+
+
+
+Changes in synapse v0.18.5-rc2 (2016-11-24)
+===========================================
+
+Bug fixes:
+
+* Don't send old events over federation, fixes bug in -rc1.
+
+Changes in synapse v0.18.5-rc1 (2016-11-24)
+===========================================
+
+Features:
+
+* Implement "event_fields" in filters (PR #1638)
+
+Changes:
+
+* Use external ldap auth pacakge (PR #1628)
+* Split out federation transaction sending to a worker (PR #1635)
+* Fail with a coherent error message if `/sync?filter=` is invalid (PR #1636)
+* More efficient notif count queries (PR #1644)
+
+
+Changes in synapse v0.18.4 (2016-11-22)
+=======================================
+
+Bug fixes:
+
+* Add workaround for buggy clients that the fail to register (PR #1632)
+
+
+Changes in synapse v0.18.4-rc1 (2016-11-14)
+===========================================
+
+Changes:
+
+* Various database efficiency improvements (PR #1188, #1192)
+* Update default config to blacklist more internal IPs, thanks to Euan Kemp (PR
+  #1198)
+* Allow specifying duration in minutes in config, thanks to Daniel Dent (PR
+  #1625)
+
+
+Bug fixes:
+
+* Fix media repo to set CORs headers on responses (PR #1190)
+* Fix registration to not error on non-ascii passwords (PR #1191)
+* Fix create event code to limit the number of prev_events (PR #1615)
+* Fix bug in transaction ID deduplication (PR #1624)
+
+
+Changes in synapse v0.18.3 (2016-11-08)
+=======================================
+
+SECURITY UPDATE
+
+Explicitly require authentication when using LDAP3. This is the default on
+versions of ``ldap3`` above 1.0, but some distributions will package an older
+version.
+
+If you are using LDAP3 login and have a version of ``ldap3`` older than 1.0 it
+is **CRITICAL to updgrade**.
+
+
+Changes in synapse v0.18.2 (2016-11-01)
+=======================================
+
+No changes since v0.18.2-rc5
+
+
+Changes in synapse v0.18.2-rc5 (2016-10-28)
+===========================================
+
+Bug fixes:
+
+* Fix prometheus process metrics in worker processes (PR #1184)
+
+
+Changes in synapse v0.18.2-rc4 (2016-10-27)
+===========================================
+
+Bug fixes:
+
+* Fix ``user_threepids`` schema delta, which in some instances prevented
+  startup after upgrade (PR #1183)
+
+
+Changes in synapse v0.18.2-rc3 (2016-10-27)
+===========================================
+
+Changes:
+
+* Allow clients to supply access tokens as headers (PR #1098)
+* Clarify error codes for GET /filter/, thanks to Alexander Maznev (PR #1164)
+* Make password reset email field case insensitive (PR #1170)
+* Reduce redundant database work in email pusher (PR #1174)
+* Allow configurable rate limiting per AS (PR #1175)
+* Check whether to ratelimit sooner to avoid work (PR #1176)
+* Standardise prometheus metrics (PR #1177)
+
+
+Bug fixes:
+
+* Fix incredibly slow back pagination query (PR #1178)
+* Fix infinite typing bug (PR #1179)
+
+
+Changes in synapse v0.18.2-rc2 (2016-10-25)
+===========================================
+
+(This release did not include the changes advertised and was identical to RC1)
+
+
+Changes in synapse v0.18.2-rc1 (2016-10-17)
+===========================================
+
+Changes:
+
+* Remove redundant event_auth index (PR #1113)
+* Reduce DB hits for replication (PR #1141)
+* Implement pluggable password auth (PR #1155)
+* Remove rate limiting from app service senders and fix get_or_create_user
+  requester, thanks to Patrik Oldsberg (PR #1157)
+* window.postmessage for Interactive Auth fallback (PR #1159)
+* Use sys.executable instead of hardcoded python, thanks to Pedro Larroy
+  (PR #1162)
+* Add config option for adding additional TLS fingerprints (PR #1167)
+* User-interactive auth on delete device (PR #1168)
+
+
+Bug fixes:
+
+* Fix not being allowed to set your own state_key, thanks to Patrik Oldsberg
+  (PR #1150)
+* Fix interactive auth to return 401 from for incorrect password (PR #1160,
+  #1166)
+* Fix email push notifs being dropped (PR #1169)
+
+
+
+Changes in synapse v0.18.1 (2016-10-05)
+======================================
+
+No changes since v0.18.1-rc1
+
+
+Changes in synapse v0.18.1-rc1 (2016-09-30)
+===========================================
+
+Features:
+
+* Add total_room_count_estimate to ``/publicRooms`` (PR #1133)
+
+
+Changes:
+
+* Time out typing over federation (PR #1140)
+* Restructure LDAP authentication (PR #1153)
+
+
+Bug fixes:
+
+* Fix 3pid invites when server is already in the room (PR #1136)
+* Fix upgrading with SQLite taking lots of CPU for a few days
+  after upgrade (PR #1144)
+* Fix upgrading from very old database versions (PR #1145)
+* Fix port script to work with recently added tables (PR #1146)
+
+
+Changes in synapse v0.18.0 (2016-09-19)
+=======================================
+
+The release includes major changes to the state storage database schemas, which
+significantly reduce database size. Synapse will attempt to upgrade the current
+data in the background. Servers with large SQLite database may experience
+degradation of performance while this upgrade is in progress, therefore you may
+want to consider migrating to using Postgres before upgrading very large SQLite
+databases
+
+
+Changes:
+
+* Make public room search case insensitive (PR #1127)
+
+
+Bug fixes:
+
+* Fix and clean up publicRooms pagination (PR #1129)
+
+
+Changes in synapse v0.18.0-rc1 (2016-09-16)
+===========================================
+
+Features:
+
+* Add ``only=highlight`` on ``/notifications`` (PR #1081)
+* Add server param to /publicRooms (PR #1082)
+* Allow clients to ask for the whole of a single state event (PR #1094)
+* Add is_direct param to /createRoom (PR #1108)
+* Add pagination support to publicRooms (PR #1121)
+* Add very basic filter API to /publicRooms (PR #1126)
+* Add basic direct to device messaging support for E2E (PR #1074, #1084, #1104,
+  #1111)
+
+
+Changes:
+
+* Move to storing state_groups_state as deltas, greatly reducing DB size (PR
+  #1065)
+* Reduce amount of state pulled out of the DB during common requests (PR #1069)
+* Allow PDF to be rendered from media repo (PR #1071)
+* Reindex state_groups_state after pruning (PR #1085)
+* Clobber EDUs in send queue (PR #1095)
+* Conform better to the CAS protocol specification (PR #1100)
+* Limit how often we ask for keys from dead servers (PR #1114)
+
+
+Bug fixes:
+
+* Fix /notifications API when used with ``from`` param (PR #1080)
+* Fix backfill when cannot find an event. (PR #1107)
+
+
+Changes in synapse v0.17.3 (2016-09-09)
+=======================================
+
+This release fixes a major bug that stopped servers from handling rooms with
+over 1000 members.
+
+
+Changes in synapse v0.17.2 (2016-09-08)
+=======================================
+
+This release contains security bug fixes. Please upgrade.
+
+
+No changes since v0.17.2-rc1
+
+
+Changes in synapse v0.17.2-rc1 (2016-09-05)
+===========================================
+
+Features:
+
+* Start adding store-and-forward direct-to-device messaging (PR #1046, #1050,
+  #1062, #1066)
+
+
+Changes:
+
+* Avoid pulling the full state of a room out so often (PR #1047, #1049, #1063,
+  #1068)
+* Don't notify for online to online presence transitions. (PR #1054)
+* Occasionally persist unpersisted presence updates (PR #1055)
+* Allow application services to have an optional 'url' (PR #1056)
+* Clean up old sent transactions from DB (PR #1059)
+
+
+Bug fixes:
+
+* Fix None check in backfill (PR #1043)
+* Fix membership changes to be idempotent (PR #1067)
+* Fix bug in get_pdu where it would sometimes return events with incorrect
+  signature
+
+
+
+Changes in synapse v0.17.1 (2016-08-24)
+=======================================
+
+Changes:
+
+* Delete old received_transactions rows (PR #1038)
+* Pass through user-supplied content in /join/$room_id (PR #1039)
+
+
+Bug fixes:
+
+* Fix bug with backfill (PR #1040)
+
+
+Changes in synapse v0.17.1-rc1 (2016-08-22)
+===========================================
+
+Features:
+
+* Add notification API (PR #1028)
+
+
+Changes:
+
+* Don't print stack traces when failing to get remote keys (PR #996)
+* Various federation /event/ perf improvements (PR #998)
+* Only process one local membership event per room at a time (PR #1005)
+* Move default display name push rule (PR #1011, #1023)
+* Fix up preview URL API. Add tests. (PR #1015)
+* Set ``Content-Security-Policy`` on media repo (PR #1021)
+* Make notify_interested_services faster (PR #1022)
+* Add usage stats to prometheus monitoring (PR #1037)
+
+
+Bug fixes:
+
+* Fix token login (PR #993)
+* Fix CAS login (PR #994, #995)
+* Fix /sync to not clobber status_msg (PR #997)
+* Fix redacted state events to include prev_content (PR #1003)
+* Fix some bugs in the auth/ldap handler (PR #1007)
+* Fix backfill request to limit URI length, so that remotes don't reject the
+  requests due to path length limits (PR #1012)
+* Fix AS push code to not send duplicate events (PR #1025)
+
+
+
+Changes in synapse v0.17.0 (2016-08-08)
+=======================================
+
+This release contains significant security bug fixes regarding authenticating
+events received over federation. PLEASE UPGRADE.
+
+This release changes the LDAP configuration format in a backwards incompatible
+way, see PR #843 for details.
+
+
+Changes:
+
+* Add federation /version API (PR #990)
+* Make psutil dependency optional (PR #992)
+
+
+Bug fixes:
+
+* Fix URL preview API to exclude HTML comments in description (PR #988)
+* Fix error handling of remote joins (PR #991)
+
+
+Changes in synapse v0.17.0-rc4 (2016-08-05)
+===========================================
+
+Changes:
+
+* Change the way we summarize URLs when previewing (PR #973)
+* Add new ``/state_ids/`` federation API (PR #979)
+* Speed up processing of ``/state/`` response (PR #986)
+
+Bug fixes:
+
+* Fix event persistence when event has already been partially persisted
+  (PR #975, #983, #985)
+* Fix port script to also copy across backfilled events (PR #982)
+
+
+Changes in synapse v0.17.0-rc3 (2016-08-02)
+===========================================
+
+Changes:
+
+* Forbid non-ASes from registering users whose names begin with '_' (PR #958)
+* Add some basic admin API docs (PR #963)
+
+
+Bug fixes:
+
+* Send the correct host header when fetching keys (PR #941)
+* Fix joining a room that has missing auth events (PR #964)
+* Fix various push bugs (PR #966, #970)
+* Fix adding emails on registration (PR #968)
+
+
+Changes in synapse v0.17.0-rc2 (2016-08-02)
+===========================================
+
+(This release did not include the changes advertised and was identical to RC1)
+
+
+Changes in synapse v0.17.0-rc1 (2016-07-28)
+===========================================
+
+This release changes the LDAP configuration format in a backwards incompatible
+way, see PR #843 for details.
+
+
+Features:
+
+* Add purge_media_cache admin API (PR #902)
+* Add deactivate account admin API (PR #903)
+* Add optional pepper to password hashing (PR #907, #910 by KentShikama)
+* Add an admin option to shared secret registration (breaks backwards compat)
+  (PR #909)
+* Add purge local room history API (PR #911, #923, #924)
+* Add requestToken endpoints (PR #915)
+* Add an /account/deactivate endpoint (PR #921)
+* Add filter param to /messages. Add 'contains_url' to filter. (PR #922)
+* Add device_id support to /login (PR #929)
+* Add device_id support to /v2/register flow. (PR #937, #942)
+* Add GET /devices endpoint (PR #939, #944)
+* Add GET /device/{deviceId} (PR #943)
+* Add update and delete APIs for devices (PR #949)
+
+
+Changes:
+
+* Rewrite LDAP Authentication against ldap3 (PR #843 by mweinelt)
+* Linearize some federation endpoints based on (origin, room_id) (PR #879)
+* Remove the legacy v0 content upload API. (PR #888)
+* Use similar naming we use in email notifs for push (PR #894)
+* Optionally include password hash in createUser endpoint (PR #905 by
+  KentShikama)
+* Use a query that postgresql optimises better for get_events_around (PR #906)
+* Fall back to 'username' if 'user' is not given for appservice registration.
+  (PR #927 by Half-Shot)
+* Add metrics for psutil derived memory usage (PR #936)
+* Record device_id in client_ips (PR #938)
+* Send the correct host header when fetching keys (PR #941)
+* Log the hostname the reCAPTCHA was completed on (PR #946)
+* Make the device id on e2e key upload optional (PR #956)
+* Add r0.2.0 to the "supported versions" list (PR #960)
+* Don't include name of room for invites in push (PR #961)
+
+
+Bug fixes:
+
+* Fix substitution failure in mail template (PR #887)
+* Put most recent 20 messages in email notif (PR #892)
+* Ensure that the guest user is in the database when upgrading accounts
+  (PR #914)
+* Fix various edge cases in auth handling (PR #919)
+* Fix 500 ISE when sending alias event without a state_key (PR #925)
+* Fix bug where we stored rejections in the state_group, persist all
+  rejections (PR #948)
+* Fix lack of check of if the user is banned when handling 3pid invites
+  (PR #952)
+* Fix a couple of bugs in the transaction and keyring code (PR #954, #955)
+
+
+
 Changes in synapse v0.16.1-r1 (2016-07-08)
 ==========================================
 

MANIFEST.in

@@ -14,6 +14,7 @@ recursive-include docs *
 recursive-include res *
 recursive-include scripts *
 recursive-include scripts-dev *
+recursive-include synapse *.pyi
 recursive-include tests *.py
 
 recursive-include synapse/static *.css
@@ -23,5 +24,7 @@ recursive-include synapse/static *.js
 
 exclude jenkins.sh
 exclude jenkins*.sh
+exclude jenkins*
+recursive-exclude jenkins *.sh
 
 prune demo/etc

README.rst

@@ -11,8 +11,8 @@ VoIP.  The basics you need to know to get up and running are:
   like ``#matrix:matrix.org`` or ``#test:localhost:8448``.
 
 - Matrix user IDs look like ``@matthew:matrix.org`` (although in the future
-  you will normally refer to yourself and others using a 3PID: email
-  address, phone number, etc rather than manipulating Matrix user IDs)
+  you will normally refer to yourself and others using a third party identifier
+  (3PID): email address, phone number, etc rather than manipulating Matrix user IDs)
 
 The overall architecture is::
 
@@ -20,12 +20,13 @@ The overall architecture is::
              https://somewhere.org/_matrix      https://elsewhere.net/_matrix
 
 ``#matrix:matrix.org`` is the official support room for Matrix, and can be
-accessed by any client from https://matrix.org/blog/try-matrix-now or via IRC
-bridge at irc://irc.freenode.net/matrix.
+accessed by any client from https://matrix.org/docs/projects/try-matrix-now or
+via IRC bridge at irc://irc.freenode.net/matrix.
 
 Synapse is currently in rapid development, but as of version 0.5 we believe it
 is sufficiently stable to be run as an internet-facing service for real usage!
 
+
 About Matrix
 ============
 
@@ -52,10 +53,10 @@ generation of fully open and interoperable messaging and VoIP apps for the
 internet.
 
 Synapse is a reference "homeserver" implementation of Matrix from the core
-development team at matrix.org, written in Python/Twisted for clarity and
-simplicity.  It is intended to showcase the concept of Matrix and let folks see
-the spec in the context of a codebase and let you run your own homeserver and
-generally help bootstrap the ecosystem.
+development team at matrix.org, written in Python/Twisted.  It is intended to
+showcase the concept of Matrix and let folks see the spec in the context of a
+codebase and let you run your own homeserver and generally help bootstrap the
+ecosystem.
 
 In Matrix, every user runs one or more Matrix clients, which connect through to
 a Matrix homeserver. The homeserver stores all their personal chat history and
@@ -66,26 +67,16 @@ hosted by someone else (e.g. matrix.org) - there is no single point of control
 or mandatory service provider in Matrix, unlike WhatsApp, Facebook, Hangouts,
 etc.
 
-Synapse ships with two basic demo Matrix clients: webclient (a basic group chat
-web client demo implemented in AngularJS) and cmdclient (a basic Python
-command line utility which lets you easily see what the JSON APIs are up to).
-
-Meanwhile, iOS and Android SDKs and clients are available from:
-
-- https://github.com/matrix-org/matrix-ios-sdk
-- https://github.com/matrix-org/matrix-ios-kit
-- https://github.com/matrix-org/matrix-ios-console
-- https://github.com/matrix-org/matrix-android-sdk
-
 We'd like to invite you to join #matrix:matrix.org (via
-https://matrix.org/blog/try-matrix-now), run a homeserver, take a look at the
-Matrix spec at https://matrix.org/docs/spec and API docs at
-https://matrix.org/docs/api, experiment with the APIs and the demo clients, and
-report any bugs via https://matrix.org/jira.
+https://matrix.org/docs/projects/try-matrix-now), run a homeserver, take a look
+at the `Matrix spec <https://matrix.org/docs/spec>`_, and experiment with the
+`APIs <https://matrix.org/docs/api>`_ and `Client SDKs
+<http://matrix.org/docs/projects/try-matrix-now.html#client-sdks>`_.
 
 Thanks for using Matrix!
 
-[1] End-to-end encryption is currently in development - see https://matrix.org/git/olm
+[1] End-to-end encryption is currently in beta: `blog post <https://matrix.org/blog/2016/11/21/matrixs-olm-end-to-end-encryption-security-assessment-released-and-implemented-cross-platform-on-riot-at-last>`_.
+
 
 Synapse Installation
 ====================
@@ -95,9 +86,14 @@ Synapse is the reference python/twisted Matrix homeserver implementation.
 System requirements:
 - POSIX-compliant system (tested on Linux & OS X)
 - Python 2.7
-- At least 512 MB RAM.
+- At least 1GB of free RAM if you want to join large public rooms like #matrix:matrix.org
 
-Synapse is written in python but some of the libraries is uses are written in
+Installing from source
+----------------------
+(Prebuilt packages are available for some platforms - see `Platform-Specific
+Instructions`_.)
+
+Synapse is written in python but some of the libraries it uses are written in
 C. So before we can install synapse itself we need a working C compiler and the
 header files for python C extensions.
 
@@ -124,6 +120,7 @@ Installing prerequisites on Mac OS X::
     xcode-select --install
     sudo easy_install pip
     sudo pip install virtualenv
+    brew install pkg-config libffi
 
 Installing prerequisites on Raspbian::
 
@@ -134,6 +131,17 @@ Installing prerequisites on Raspbian::
     sudo pip install --upgrade ndg-httpsclient
     sudo pip install --upgrade virtualenv
 
+Installing prerequisites on openSUSE::
+
+    sudo zypper in -t pattern devel_basis
+    sudo zypper in python-pip python-setuptools sqlite3 python-virtualenv \
+                   python-devel libffi-devel libopenssl-devel libjpeg62-devel
+
+Installing prerequisites on OpenBSD::
+
+    doas pkg_add python libffi py-pip py-setuptools sqlite3 py-virtualenv \
+                 libxslt
+
 To install the synapse homeserver run::
 
     virtualenv -p python2.7 ~/.synapse
@@ -145,38 +153,74 @@ This installs synapse, along with the libraries it uses, into a virtual
 environment under ``~/.synapse``.  Feel free to pick a different directory
 if you prefer.
 
-In case of problems, please see the _Troubleshooting section below.
+In case of problems, please see the _`Troubleshooting` section below.
 
 Alternatively, Silvio Fricke has contributed a Dockerfile to automate the
 above in Docker at https://registry.hub.docker.com/u/silviof/docker-matrix/.
 
-Also, Martin Giess has created an auto-deployment process with vagrant/ansible, 
-tested with VirtualBox/AWS/DigitalOcean - see https://github.com/EMnify/matrix-synapse-auto-deploy 
+Also, Martin Giess has created an auto-deployment process with vagrant/ansible,
+tested with VirtualBox/AWS/DigitalOcean - see https://github.com/EMnify/matrix-synapse-auto-deploy
 for details.
 
-To set up your homeserver, run (in your virtualenv, as before)::
+Configuring synapse
+-------------------
+
+Before you can start Synapse, you will need to generate a configuration
+file. To do this, run (in your virtualenv, as before)::
 
     cd ~/.synapse
     python -m synapse.app.homeserver \
-        --server-name machine.my.domain.name \
+        --server-name my.domain.name \
         --config-path homeserver.yaml \
         --generate-config \
         --report-stats=[yes|no]
 
-...substituting your host and domain name as appropriate.
+... substituting an appropriate value for ``--server-name``. The server name
+determines the "domain" part of user-ids for users on your server: these will
+all be of the format ``@user:my.domain.name``. It also determines how other
+matrix servers will reach yours for `Federation`_. For a test configuration,
+set this to the hostname of your server. For a more production-ready setup, you
+will probably want to specify your domain (``example.com``) rather than a
+matrix-specific hostname here (in the same way that your email address is
+probably ``user@example.com`` rather than ``user@email.example.com``) - but
+doing so may require more advanced setup - see `Setting up
+Federation`_. Beware that the server name cannot be changed later.
 
-This will generate you a config file that you can then customise, but it will
+This command will generate you a config file that you can then customise, but it will
 also generate a set of keys for you. These keys will allow your Home Server to
 identify itself to other Home Servers, so don't lose or delete them. It would be
-wise to back them up somewhere safe. If, for whatever reason, you do need to
+wise to back them up somewhere safe. (If, for whatever reason, you do need to
 change your Home Server's keys, you may find that other Home Servers have the
 old key cached. If you update the signing key, you should change the name of the
-key in the <server name>.signing.key file (the second word) to something different.
+key in the ``<server name>.signing.key`` file (the second word) to something
+different. See `the spec`__ for more information on key management.)
 
-By default, registration of new users is disabled. You can either enable
-registration in the config by specifying ``enable_registration: true``
-(it is then recommended to also set up CAPTCHA - see docs/CAPTCHA_SETUP), or
-you can use the command line to register new users::
+.. __: `key_management`_
+
+The default configuration exposes two HTTP ports: 8008 and 8448. Port 8008 is
+configured without TLS; it is not recommended this be exposed outside your
+local network. Port 8448 is configured to use TLS with a self-signed
+certificate. This is fine for testing with but, to avoid your clients
+complaining about the certificate, you will almost certainly want to use
+another certificate for production purposes. (Note that a self-signed
+certificate is fine for `Federation`_). You can do so by changing
+``tls_certificate_path``, ``tls_private_key_path`` and ``tls_dh_params_path``
+in ``homeserver.yaml``; alternatively, you can use a reverse-proxy, but be sure
+to read `Using a reverse proxy with Synapse`_ when doing so.
+
+Apart from port 8448 using TLS, both ports are the same in the default
+configuration.
+
+Registering a user
+------------------
+
+You will need at least one user on your server in order to use a Matrix
+client. Users can be registered either `via a Matrix client`__, or via a
+commandline script.
+
+.. __: `client-user-reg`_
+
+To get started, it is easiest to use the command line to register new users::
 
     $ source ~/.synapse/bin/activate
     $ synctl start # if not already running
@@ -186,8 +230,19 @@ you can use the command line to register new users::
     Confirm password:
     Success!
 
+This process uses a setting ``registration_shared_secret`` in
+``homeserver.yaml``, which is shared between Synapse itself and the
+``register_new_matrix_user`` script. It doesn't matter what it is (a random
+value is generated by ``--generate-config``), but it should be kept secret, as
+anyone with knowledge of it can register users on your server even if
+``enable_registration`` is ``false``.
+
+Setting up a TURN server
+------------------------
+
 For reliable VoIP calls to be routed via this homeserver, you MUST configure
-a TURN server.  See docs/turn-howto.rst for details.
+a TURN server.  See `<docs/turn-howto.rst>`_ for details.
+
 
 Running Synapse
 ===============
@@ -199,29 +254,66 @@ run (e.g. ``~/.synapse``), and::
     source ./bin/activate
     synctl start
 
-Using PostgreSQL
-================
 
-As of Synapse 0.9, `PostgreSQL <http://www.postgresql.org>`_ is supported as an
-alternative to the `SQLite <http://sqlite.org/>`_ database that Synapse has
-traditionally used for convenience and simplicity.
+Connecting to Synapse from a client
+===================================
 
-The advantages of Postgres include:
+The easiest way to try out your new Synapse installation is by connecting to it
+from a web client. The easiest option is probably the one at
+http://riot.im/app. You will need to specify a "Custom server" when you log on
+or register: set this to ``https://localhost:8448`` - remember to specify the
+port (``:8448``) unless you changed the configuration. (Leave the identity
+server as the default - see `Identity servers`_.)
 
-* significant performance improvements due to the superior threading and
-  caching model, smarter query optimiser
-* allowing the DB to be run on separate hardware
-* allowing basic active/backup high-availability with a "hot spare" synapse
-  pointing at the same DB master, as well as enabling DB replication in
-  synapse itself.
+If all goes well you should at least be able to log in, create a room, and
+start sending messages.
 
-The only disadvantage is that the code is relatively new as of April 2015 and
-may have a few regressions relative to SQLite.
+(The homeserver runs a web client by default at https://localhost:8448/, though
+as of the time of writing it is somewhat outdated and not really recommended -
+https://github.com/matrix-org/synapse/issues/1527).
 
-For information on how to install and use PostgreSQL, please see
-`docs/postgres.rst <docs/postgres.rst>`_.
+.. _`client-user-reg`:
 
-Platform Specific Instructions
+Registering a new user from a client
+------------------------------------
+
+By default, registration of new users via Matrix clients is disabled. To enable
+it, specify ``enable_registration: true`` in ``homeserver.yaml``. (It is then
+recommended to also set up CAPTCHA - see `<docs/CAPTCHA_SETUP.rst>`_.)
+
+Once ``enable_registration`` is set to ``true``, it is possible to register a
+user via `riot.im <https://riot.im/app/#/register>`_ or other Matrix clients.
+
+Your new user name will be formed partly from the ``server_name`` (see
+`Configuring synapse`_), and partly from a localpart you specify when you
+create the account. Your name will take the form of::
+
+    @localpart:my.domain.name
+
+(pronounced "at localpart on my dot domain dot name").
+
+As when logging in, you will need to specify a "Custom server".  Specify your
+desired ``localpart`` in the 'User name' box.
+
+
+Security Note
+=============
+
+Matrix serves raw user generated data in some APIs - specifically the `content
+repository endpoints <http://matrix.org/docs/spec/client_server/latest.html#get-matrix-media-r0-download-servername-mediaid>`_.
+
+Whilst we have tried to mitigate against possible XSS attacks (e.g.
+https://github.com/matrix-org/synapse/pull/1021) we recommend running
+matrix homeservers on a dedicated domain name, to limit any malicious user generated
+content served to web browsers a matrix API from being able to attack webapps hosted
+on the same domain.  This is particularly true of sharing a matrix webclient and
+server on the same domain.
+
+See https://github.com/vector-im/vector-web/issues/1977 and
+https://developer.github.com/changes/2014-04-25-user-content-security for more details.
+
+
+Platform-Specific Instructions
 ==============================
 
 Debian
@@ -229,7 +321,7 @@ Debian
 
 Matrix provides official Debian packages via apt from http://matrix.org/packages/debian/.
 Note that these packages do not include a client - choose one from
-https://matrix.org/blog/try-matrix-now/ (or build your own with one of our SDKs :)
+https://matrix.org/docs/projects/try-matrix-now/ (or build your own with one of our SDKs :)
 
 Fedora
 ------
@@ -283,6 +375,32 @@ Synapse can be installed via FreeBSD Ports or Packages contributed by Brendan Mo
  - Ports: ``cd /usr/ports/net/py-matrix-synapse && make install clean``
  - Packages: ``pkg install py27-matrix-synapse``
 
+
+OpenBSD
+-------
+
+There is currently no port for OpenBSD. Additionally, OpenBSD's security
+settings require a slightly more difficult installation process.
+
+1) Create a new directory in ``/usr/local`` called ``_synapse``. Also, create a
+   new user called ``_synapse`` and set that directory as the new user's home.
+   This is required because, by default, OpenBSD only allows binaries which need
+   write and execute permissions on the same memory space to be run from
+   ``/usr/local``.
+2) ``su`` to the new ``_synapse`` user and change to their home directory.
+3) Create a new virtualenv: ``virtualenv -p python2.7 ~/.synapse``
+4) Source the virtualenv configuration located at
+   ``/usr/local/_synapse/.synapse/bin/activate``. This is done in ``ksh`` by
+   using the ``.`` command, rather than ``bash``'s ``source``.
+5) Optionally, use ``pip`` to install ``lxml``, which Synapse needs to parse
+   webpages for their titles.
+6) Use ``pip`` to install this repository: ``pip install
+   https://github.com/matrix-org/synapse/tarball/master``
+7) Optionally, change ``_synapse``'s shell to ``/bin/false`` to reduce the
+   chance of a compromised Synapse server being used to take over your box.
+
+After this, you may proceed with the rest of the install directions.
+
 NixOS
 -----
 
@@ -322,6 +440,7 @@ Troubleshooting:
   you do, you may need to create a symlink to ``libsodium.a`` so ``ld`` can find
   it: ``ln -s /usr/local/lib/libsodium.a /usr/lib/libsodium.a``
 
+
 Troubleshooting
 ===============
 
@@ -395,37 +514,6 @@ you will need to explicitly call Python2.7 - either running as::
 
 ...or by editing synctl with the correct python executable.
 
-Synapse Development
-===================
-
-To check out a synapse for development, clone the git repo into a working
-directory of your choice::
-
-    git clone https://github.com/matrix-org/synapse.git
-    cd synapse
-
-Synapse has a number of external dependencies, that are easiest
-to install using pip and a virtualenv::
-
-    virtualenv env
-    source env/bin/activate
-    python synapse/python_dependencies.py | xargs -n1 pip install
-    pip install setuptools_trial mock
-
-This will run a process of downloading and installing all the needed
-dependencies into a virtual env.
-
-Once this is done, you may wish to run Synapse's unit tests, to
-check that everything is installed as it should be::
-
-    python setup.py test
-
-This should end with a 'PASSED' result::
-
-    Ran 143 tests in 0.601s
-
-    PASSED (successes=143)
-
 
 Upgrading an existing Synapse
 =============================
@@ -436,140 +524,248 @@ versions of synapse.
 
 .. _UPGRADE.rst: UPGRADE.rst
 
+.. _federation:
+
 Setting up Federation
 =====================
 
-In order for other homeservers to send messages to your server, it will need to
-be publicly visible on the internet, and they will need to know its host name.
-You have two choices here, which will influence the form of your Matrix user
-IDs:
+Federation is the process by which users on different servers can participate
+in the same room. For this to work, those other servers must be able to contact
+yours to send messages.
 
-1) Use the machine's own hostname as available on public DNS in the form of
-   its A or AAAA records. This is easier to set up initially, perhaps for
-   testing, but lacks the flexibility of SRV.
+As explained in `Configuring synapse`_, the ``server_name`` in your
+``homeserver.yaml`` file determines the way that other servers will reach
+yours. By default, they will treat it as a hostname and try to connect to
+port 8448. This is easy to set up and will work with the default configuration,
+provided you set the ``server_name`` to match your machine's public DNS
+hostname.
 
-2) Set up a SRV record for your domain name. This requires you create a SRV
-   record in DNS, but gives the flexibility to run the server on your own
-   choice of TCP port, on a machine that might not be the same name as the
-   domain name.
+For a more flexible configuration, you can set up a DNS SRV record. This allows
+you to run your server on a machine that might not have the same name as your
+domain name. For example, you might want to run your server at
+``synapse.example.com``, but have your Matrix user-ids look like
+``@user:example.com``. (A SRV record also allows you to change the port from
+the default 8448. However, if you are thinking of using a reverse-proxy, be
+sure to read `Reverse-proxying the federation port`_ first.)
 
-For the first form, simply pass the required hostname (of the machine) as the
---server-name parameter::
+To use a SRV record, first create your SRV record and publish it in DNS. This
+should have the format ``_matrix._tcp.<yourdomain.com> <ttl> IN SRV 10 0 <port>
+<synapse.server.name>``. The DNS record should then look something like::
+
+    $ dig -t srv _matrix._tcp.example.com
+    _matrix._tcp.example.com. 3600    IN      SRV     10 0 8448 synapse.example.com.
+
+You can then configure your homeserver to use ``<yourdomain.com>`` as the domain in
+its user-ids, by setting ``server_name``::
 
     python -m synapse.app.homeserver \
-        --server-name machine.my.domain.name \
+        --server-name <yourdomain.com> \
         --config-path homeserver.yaml \
         --generate-config
     python -m synapse.app.homeserver --config-path homeserver.yaml
 
-Alternatively, you can run ``synctl start`` to guide you through the process.
-
-For the second form, first create your SRV record and publish it in DNS. This
-needs to be named _matrix._tcp.YOURDOMAIN, and point at at least one hostname
-and port where the server is running.  (At the current time synapse does not
-support clustering multiple servers into a single logical homeserver).  The DNS
-record would then look something like::
-
-    $ dig -t srv _matrix._tcp.machine.my.domain.name
-    _matrix._tcp    IN      SRV     10 0 8448 machine.my.domain.name.
-
-
-At this point, you should then run the homeserver with the hostname of this
-SRV record, as that is the name other machines will expect it to have::
-
-    python -m synapse.app.homeserver \
-        --server-name YOURDOMAIN \
-        --config-path homeserver.yaml \
-        --generate-config
-    python -m synapse.app.homeserver --config-path homeserver.yaml
-
-
-If you've already generated the config file, you need to edit the "server_name"
-in you  ```homeserver.yaml``` file. If you've already started Synapse and a
+If you've already generated the config file, you need to edit the ``server_name``
+in your ``homeserver.yaml`` file. If you've already started Synapse and a
 database has been created, you will have to recreate the database.
 
-You may additionally want to pass one or more "-v" options, in order to
-increase the verbosity of logging output; at least for initial testing.
+If all goes well, you should be able to `connect to your server with a client`__,
+and then join a room via federation. (Try ``#matrix-dev:matrix.org`` as a first
+step. "Matrix HQ"'s sheer size and activity level tends to make even the
+largest boxes pause for thought.)
+
+.. __: `Connecting to Synapse from a client`_
+
+Troubleshooting
+---------------
+The typical failure mode with federation is that when you try to join a room,
+it is rejected with "401: Unauthorized". Generally this means that other
+servers in the room couldn't access yours. (Joining a room over federation is a
+complicated dance which requires connections in both directions).
+
+So, things to check are:
+
+* If you are trying to use a reverse-proxy, read `Reverse-proxying the
+  federation port`_.
+* If you are not using a SRV record, check that your ``server_name`` (the part
+  of your user-id after the ``:``) matches your hostname, and that port 8448 on
+  that hostname is reachable from outside your network.
+* If you *are* using a SRV record, check that it matches your ``server_name``
+  (it should be ``_matrix._tcp.<server_name>``), and that the port and hostname
+  it specifies are reachable from outside your network.
 
 Running a Demo Federation of Synapses
 -------------------------------------
 
 If you want to get up and running quickly with a trio of homeservers in a
-private federation (``localhost:8080``, ``localhost:8081`` and
-``localhost:8082``) which you can then access through the webclient running at
-http://localhost:8080. Simply run::
-
-    demo/start.sh
-
-This is mainly useful just for development purposes.
-
-Running The Demo Web Client
-===========================
-
-The homeserver runs a web client by default at https://localhost:8448/.
-
-If this is the first time you have used the client from that browser (it uses
-HTML5 local storage to remember its config), you will need to log in to your
-account. If you don't yet have an account, because you've just started the
-homeserver for the first time, then you'll need to register one.
+private federation, there is a script in the ``demo`` directory. This is mainly
+useful just for development purposes. See `<demo/README>`_.
 
 
-Registering A New Account
--------------------------
+Using PostgreSQL
+================
 
-Your new user name will be formed partly from the hostname your server is
-running as, and partly from a localpart you specify when you create the
-account. Your name will take the form of::
+As of Synapse 0.9, `PostgreSQL <http://www.postgresql.org>`_ is supported as an
+alternative to the `SQLite <http://sqlite.org/>`_ database that Synapse has
+traditionally used for convenience and simplicity.
 
-    @localpart:my.domain.here
-         (pronounced "at localpart on my dot domain dot here")
+The advantages of Postgres include:
 
-Specify your desired localpart in the topmost box of the "Register for an
-account" form, and click the "Register" button. Hostnames can contain ports if
-required due to lack of SRV records (e.g. @matthew:localhost:8448 on an
-internal synapse sandbox running on localhost).
+* significant performance improvements due to the superior threading and
+  caching model, smarter query optimiser
+* allowing the DB to be run on separate hardware
+* allowing basic active/backup high-availability with a "hot spare" synapse
+  pointing at the same DB master, as well as enabling DB replication in
+  synapse itself.
 
-If registration fails, you may need to enable it in the homeserver (see
-`Synapse Installation`_ above)
+For information on how to install and use PostgreSQL, please see
+`docs/postgres.rst <docs/postgres.rst>`_.
 
 
-Logging In To An Existing Account
----------------------------------
+.. _reverse-proxy:
+
+Using a reverse proxy with Synapse
+==================================
+
+It is possible to put a reverse proxy such as
+`nginx <https://nginx.org/en/docs/http/ngx_http_proxy_module.html>`_,
+`Apache <https://httpd.apache.org/docs/current/mod/mod_proxy_http.html>`_ or
+`HAProxy <http://www.haproxy.org/>`_ in front of Synapse. One advantage of
+doing so is that it means that you can expose the default https port (443) to
+Matrix clients without needing to run Synapse with root privileges.
+
+The most important thing to know here is that Matrix clients and other Matrix
+servers do not necessarily need to connect to your server via the same
+port. Indeed, clients will use port 443 by default, whereas servers default to
+port 8448. Where these are different, we refer to the 'client port' and the
+'federation port'.
+
+The next most important thing to know is that using a reverse-proxy on the
+federation port has a number of pitfalls. It is possible, but be sure to read
+`Reverse-proxying the federation port`_.
+
+The recommended setup is therefore to configure your reverse-proxy on port 443
+for client connections, but to also expose port 8448 for server-server
+connections. All the Matrix endpoints begin ``/_matrix``, so an example nginx
+configuration might look like::
+
+  server {
+      listen 443 ssl;
+      listen [::]:443 ssl;
+      server_name matrix.example.com;
+
+      location /_matrix {
+          proxy_pass http://localhost:8008;
+          proxy_set_header X-Forwarded-For $remote_addr;
+      }
+  }
+
+You will also want to set ``bind_addresses: ['127.0.0.1']`` and ``x_forwarded: true``
+for port 8008 in ``homeserver.yaml`` to ensure that client IP addresses are
+recorded correctly.
+
+Having done so, you can then use ``https://matrix.example.com`` (instead of
+``https://matrix.example.com:8448``) as the "Custom server" when `Connecting to
+Synapse from a client`_.
+
+Reverse-proxying the federation port
+------------------------------------
+
+There are two issues to consider before using a reverse-proxy on the federation
+port:
+
+* Due to the way SSL certificates are managed in the Matrix federation protocol
+  (see `spec`__), Synapse needs to be configured with the path to the SSL
+  certificate, *even if you do not terminate SSL at Synapse*.
+
+  .. __: `key_management`_
+
+* Synapse does not currently support SNI on the federation protocol
+  (`bug #1491 <https://github.com/matrix-org/synapse/issues/1491>`_), which
+  means that using name-based virtual hosting is unreliable.
+
+Furthermore, a number of the normal reasons for using a reverse-proxy do not
+apply:
+
+* Other servers will connect on port 8448 by default, so there is no need to
+  listen on port 443 (for federation, at least), which avoids the need for root
+  privileges and virtual hosting.
+
+* A self-signed SSL certificate is fine for federation, so there is no need to
+  automate renewals. (The certificate generated by ``--generate-config`` is
+  valid for 10 years.)
+
+If you want to set up a reverse-proxy on the federation port despite these
+caveats, you will need to do the following:
+
+* In ``homeserver.yaml``, set ``tls_certificate_path`` to the path to the SSL
+  certificate file used by your reverse-proxy, and set ``no_tls`` to ``True``.
+  (``tls_private_key_path`` will be ignored if ``no_tls`` is ``True``.)
+
+* In your reverse-proxy configuration:
+
+  * If there are other virtual hosts on the same port, make sure that the
+    *default* one uses the certificate configured above.
+
+  * Forward ``/_matrix`` to Synapse.
+
+* If your reverse-proxy is not listening on port 8448, publish a SRV record to
+  tell other servers how to find you. See `Setting up Federation`_.
+
+When updating the SSL certificate, just update the file pointed to by
+``tls_certificate_path``: there is no need to restart synapse. (You may like to
+use a symbolic link to help make this process atomic.)
+
+The most common mistake when setting up federation is not to tell Synapse about
+your SSL certificate. To check it, you can visit
+``https://matrix.org/federationtester/api/report?server_name=<your_server_name>``.
+Unfortunately, there is no UI for this yet, but, you should see
+``"MatchingTLSFingerprint": true``. If not, check that
+``Certificates[0].SHA256Fingerprint`` (the fingerprint of the certificate
+presented by your reverse-proxy) matches ``Keys.tls_fingerprints[0].sha256``
+(the fingerprint of the certificate Synapse is using).
 
-Just enter the ``@localpart:my.domain.here`` Matrix user ID and password into
-the form and click the Login button.
 
 Identity Servers
 ================
 
-The job of authenticating 3PIDs and tracking which 3PIDs are associated with a
-given Matrix user is very security-sensitive, as there is obvious risk of spam
-if it is too easy to sign up for Matrix accounts or harvest 3PID data.
-Meanwhile the job of publishing the end-to-end encryption public keys for
-Matrix users is also very security-sensitive for similar reasons.
+Identity servers have the job of mapping email addresses and other 3rd Party
+IDs (3PIDs) to Matrix user IDs, as well as verifying the ownership of 3PIDs
+before creating that mapping.
 
-Therefore the role of managing trusted identity in the Matrix ecosystem is
-farmed out to a cluster of known trusted ecosystem partners, who run 'Matrix
-Identity Servers' such as ``sydent``, whose role is purely to authenticate and
-track 3PID logins and publish end-user public keys.
+**They are not where accounts or credentials are stored - these live on home
+servers. Identity Servers are just for mapping 3rd party IDs to matrix IDs.**
 
-It's currently early days for identity servers as Matrix is not yet using 3PIDs
-as the primary means of identity and E2E encryption is not complete. As such,
-we are running a single identity server (https://matrix.org) at the current
-time.
+This process is very security-sensitive, as there is obvious risk of spam if it
+is too easy to sign up for Matrix accounts or harvest 3PID data. In the longer
+term, we hope to create a decentralised system to manage it (`matrix-doc #712
+<https://github.com/matrix-org/matrix-doc/issues/712>`_), but in the meantime,
+the role of managing trusted identity in the Matrix ecosystem is farmed out to
+a cluster of known trusted ecosystem partners, who run 'Matrix Identity
+Servers' such as `Sydent <https://github.com/matrix-org/sydent>`_, whose role
+is purely to authenticate and track 3PID logins and publish end-user public
+keys.
+
+You can host your own copy of Sydent, but this will prevent you reaching other
+users in the Matrix ecosystem via their email address, and prevent them finding
+you. We therefore recommend that you use one of the centralised identity servers
+at ``https://matrix.org`` or ``https://vector.im`` for now.
+
+To reiterate: the Identity server will only be used if you choose to associate
+an email address with your account, or send an invite to another user via their
+email address.
 
 
 URL Previews
 ============
 
-Synapse 0.15.0 introduces an experimental new API for previewing URLs at
-/_matrix/media/r0/preview_url.  This is disabled by default.  To turn it on
-you must enable the `url_preview_enabled: True` config parameter and explicitly
-specify the IP ranges that Synapse is not allowed to spider for previewing in
-the `url_preview_ip_range_blacklist` configuration parameter.  This is critical
-from a security perspective to stop arbitrary Matrix users spidering 'internal'
-URLs on your network.  At the very least we recommend that your loopback and
-RFC1918 IP addresses are blacklisted.
+Synapse 0.15.0 introduces a new API for previewing URLs at
+``/_matrix/media/r0/preview_url``.  This is disabled by default.  To turn it on
+you must enable the ``url_preview_enabled: True`` config parameter and
+explicitly specify the IP ranges that Synapse is not allowed to spider for
+previewing in the ``url_preview_ip_range_blacklist`` configuration parameter.
+This is critical from a security perspective to stop arbitrary Matrix users
+spidering 'internal' URLs on your network.  At the very least we recommend that
+your loopback and RFC1918 IP addresses are blacklisted.
 
 This also requires the optional lxml and netaddr python dependencies to be
 installed.
@@ -583,24 +779,54 @@ server, they can request a password-reset token via clients such as Vector.
 
 A manual password reset can be done via direct database access as follows.
 
-First calculate the hash of the new password:
+First calculate the hash of the new password::
 
     $ source ~/.synapse/bin/activate
     $ ./scripts/hash_password
-    Password: 
-    Confirm password: 
+    Password:
+    Confirm password:
     $2a$12$xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
 
-Then update the `users` table in the database:
+Then update the `users` table in the database::
 
     UPDATE users SET password_hash='$2a$12$xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx'
         WHERE name='@test:test.com';
 
-Where's the spec?!
-==================
 
-The source of the matrix spec lives at https://github.com/matrix-org/matrix-doc.
-A recent HTML snapshot of this lives at http://matrix.org/docs/spec
+Synapse Development
+===================
+
+Before setting up a development environment for synapse, make sure you have the
+system dependencies (such as the python header files) installed - see
+`Installing from source`_.
+
+To check out a synapse for development, clone the git repo into a working
+directory of your choice::
+
+    git clone https://github.com/matrix-org/synapse.git
+    cd synapse
+
+Synapse has a number of external dependencies, that are easiest
+to install using pip and a virtualenv::
+
+    virtualenv env
+    source env/bin/activate
+    python synapse/python_dependencies.py | xargs pip install
+    pip install lxml mock
+
+This will run a process of downloading and installing all the needed
+dependencies into a virtual env.
+
+Once this is done, you may wish to run Synapse's unit tests, to
+check that everything is installed as it should be::
+
+    PYTHONPATH="." trial tests
+
+This should end with a 'PASSED' result::
+
+    Ran 143 tests in 0.601s
+
+    PASSED (successes=143)
 
 
 Building Internal API Documentation
@@ -617,7 +843,6 @@ Building internal API documentation::
     python setup.py build_sphinx
 
 
-
 Help!! Synapse eats all my RAM!
 ===============================
 
@@ -633,3 +858,5 @@ around a ~700MB footprint.  You can dial it down further to 0.02 if
 desired, which targets roughly ~512MB.  Conversely you can dial it up if
 you need performance for lots of users and have a box with a lot of RAM.
 
+
+.. _`key_management`: https://matrix.org/docs/spec/server_server/unstable.html#retrieving-server-keys

UPGRADE.rst

@@ -27,7 +27,7 @@ running:
     # Pull the latest version of the master branch.
     git pull
     # Update the versions of synapse's python dependencies.
-    python synapse/python_dependencies.py | xargs -n1 pip install
+    python synapse/python_dependencies.py | xargs -n1 pip install --upgrade
 
 
 Upgrading to v0.15.0

docs/CAPTCHA_SETUP.rst

@@ -10,13 +10,13 @@ https://developers.google.com/recaptcha/
 
 Setting ReCaptcha Keys
 ----------------------
-The keys are a config option on the home server config. If they are not 
-visible, you can generate them via --generate-config. Set the following value:
+The keys are a config option on the home server config. If they are not
+visible, you can generate them via --generate-config. Set the following value::
 
   recaptcha_public_key: YOUR_PUBLIC_KEY
   recaptcha_private_key: YOUR_PRIVATE_KEY
-  
-In addition, you MUST enable captchas via:
+
+In addition, you MUST enable captchas via::
 
   enable_registration_captcha: true
 
@@ -25,7 +25,6 @@ Configuring IP used for auth
 The ReCaptcha API requires that the IP address of the user who solved the
 captcha is sent. If the client is connecting through a proxy or load balancer,
 it may be required to use the X-Forwarded-For (XFF) header instead of the origin
-IP address. This can be configured as an option on the home server like so:
+IP address. This can be configured as an option on the home server like so::
 
   captcha_ip_origin_is_x_forwarded: true
-

docs/admin_api/README.rst

@@ -0,0 +1,12 @@
+Admin APIs
+==========
+
+This directory includes documentation for the various synapse specific admin
+APIs available.
+
+Only users that are server admins can use these APIs. A user can be marked as a
+server admin by updating the database directly, e.g.:
+
+``UPDATE users SET admin = 1 WHERE name = '@foo:bar.com'``
+
+Restarting may be required for the changes to register.

docs/admin_api/purge_history_api.rst

@@ -0,0 +1,15 @@
+Purge History API
+=================
+
+The purge history API allows server admins to purge historic events from their
+database, reclaiming disk space.
+
+Depending on the amount of history being purged a call to the API may take
+several minutes or longer. During this period users will not be able to
+paginate further back in the room from the point being purged from.
+
+The API is simply:
+
+``POST /_matrix/client/r0/admin/purge_history/<room_id>/<event_id>``
+
+including an ``access_token`` of a server admin.

docs/admin_api/purge_remote_media.rst

@@ -0,0 +1,17 @@
+Purge Remote Media API
+======================
+
+The purge remote media API allows server admins to purge old cached remote
+media.
+
+The API is::
+
+    POST /_matrix/client/r0/admin/purge_media_cache?before_ts=<unix_timestamp_in_ms>&access_token=<access_token>
+
+    {}
+
+Which will remove all cached media that was last accessed before
+``<unix_timestamp_in_ms>``.
+
+If the user re-requests purged remote media, synapse will re-request the media
+from the originating server.

docs/metrics-howto.rst

@@ -15,36 +15,45 @@ How to monitor Synapse metrics using Prometheus
 
   Restart synapse
 
-3: Check out synapse-prometheus-config
-  https://github.com/matrix-org/synapse-prometheus-config
+3: Add a prometheus target for synapse. It needs to set the ``metrics_path``
+   to a non-default value::
 
-4: Add ``synapse.html`` and ``synapse.rules``
-  The ``.html`` file needs to appear in prometheus's ``consoles`` directory,
-  and the ``.rules`` file needs to be invoked somewhere in the main config
-  file. A symlink to each from the git checkout into the prometheus directory
-  might be easiest to ensure ``git pull`` keeps it updated.
+    - job_name: "synapse"
+      metrics_path: "/_synapse/metrics"
+      static_configs:
+        - targets:
+            "my.server.here:9092"
 
-5: Add a prometheus target for synapse
-  This is easiest if prometheus runs on the same machine as synapse, as it can
-  then just use localhost::
+Standard Metric Names
+---------------------
 
-    global: {
-      rule_file: "synapse.rules"
-    }
+As of synapse version 0.18.2, the format of the process-wide metrics has been
+changed to fit prometheus standard naming conventions. Additionally the units
+have been changed to seconds, from miliseconds.
 
-    job: {
-      name: "synapse"
+================================== =============================
+New name                           Old name
+---------------------------------- -----------------------------
+process_cpu_user_seconds_total     process_resource_utime / 1000
+process_cpu_system_seconds_total   process_resource_stime / 1000
+process_open_fds (no 'type' label) process_fds
+================================== =============================
 
-      target_group: {
-        target: "http://localhost:9092/"
-      }
-    }
+The python-specific counts of garbage collector performance have been renamed.
 
-6: Start prometheus::
+=========================== ======================
+New name                    Old name
+--------------------------- ----------------------
+python_gc_time              reactor_gc_time      
+python_gc_unreachable_total reactor_gc_unreachable
+python_gc_counts            reactor_gc_counts
+=========================== ======================
 
-   ./prometheus -config.file=prometheus.conf
+The twisted-specific reactor metrics have been renamed.
 
-7: Wait a few seconds for it to start and perform the first scrape,
-   then visit the console:
-
-    http://server-where-prometheus-runs:9090/consoles/synapse.html
+==================================== =====================
+New name                             Old name
+------------------------------------ ---------------------
+python_twisted_reactor_pending_calls reactor_pending_calls
+python_twisted_reactor_tick_time     reactor_tick_time
+==================================== =====================

docs/workers.rst

@@ -0,0 +1,98 @@
+Scaling synapse via workers
+---------------------------
+
+Synapse has experimental support for splitting out functionality into
+multiple separate python processes, helping greatly with scalability.  These
+processes are called 'workers', and are (eventually) intended to scale
+horizontally independently.
+
+All processes continue to share the same database instance, and as such, workers
+only work with postgres based synapse deployments (sharing a single sqlite
+across multiple processes is a recipe for disaster, plus you should be using
+postgres anyway if you care about scalability).
+
+The workers communicate with the master synapse process via a synapse-specific
+HTTP protocol called 'replication' - analogous to MySQL or Postgres style
+database replication; feeding a stream of relevant data to the workers so they
+can be kept in sync with the main synapse process and database state.
+
+To enable workers, you need to add a replication listener to the master synapse, e.g.::
+
+    listeners:
+      - port: 9092
+        bind_address: '127.0.0.1'
+        type: http
+        tls: false
+        x_forwarded: false
+        resources:
+          - names: [replication]
+            compress: false
+
+Under **no circumstances** should this replication API listener be exposed to the
+public internet; it currently implements no authentication whatsoever and is
+unencrypted HTTP.
+
+You then create a set of configs for the various worker processes.  These should be
+worker configuration files should be stored in a dedicated subdirectory, to allow
+synctl to manipulate them.
+
+The current available worker applications are:
+ * synapse.app.pusher - handles sending push notifications to sygnal and email
+ * synapse.app.synchrotron - handles /sync endpoints.  can scales horizontally through multiple instances.
+ * synapse.app.appservice - handles output traffic to Application Services
+ * synapse.app.federation_reader - handles receiving federation traffic (including public_rooms API)
+ * synapse.app.media_repository - handles the media repository.
+ * synapse.app.client_reader - handles client API endpoints like /publicRooms
+
+Each worker configuration file inherits the configuration of the main homeserver
+configuration file.  You can then override configuration specific to that worker,
+e.g. the HTTP listener that it provides (if any); logging configuration; etc.
+You should minimise the number of overrides though to maintain a usable config.
+
+You must specify the type of worker application (worker_app) and the replication
+endpoint that it's talking to on the main synapse process (worker_replication_url).
+
+For instance::
+
+    worker_app: synapse.app.synchrotron
+
+    # The replication listener on the synapse to talk to.
+    worker_replication_url: http://127.0.0.1:9092/_synapse/replication
+
+    worker_listeners:
+     - type: http
+       port: 8083
+       resources:
+         - names:
+           - client
+
+    worker_daemonize: True
+    worker_pid_file: /home/matrix/synapse/synchrotron.pid
+    worker_log_config: /home/matrix/synapse/config/synchrotron_log_config.yaml
+
+...is a full configuration for a synchrotron worker instance, which will expose a
+plain HTTP /sync endpoint on port 8083 separately from the /sync endpoint provided
+by the main synapse.
+
+Obviously you should configure your loadbalancer to route the /sync endpoint to
+the synchrotron instance(s) in this instance.
+
+Finally, to actually run your worker-based synapse, you must pass synctl the -a
+commandline option to tell it to operate on all the worker configurations found
+in the given directory, e.g.::
+
+    synctl -a $CONFIG/workers start
+
+Currently one should always restart all workers when restarting or upgrading
+synapse, unless you explicitly know it's safe not to.  For instance, restarting
+synapse without restarting all the synchrotrons may result in broken typing
+notifications.
+
+To manipulate a specific worker, you pass the -w option to synctl::
+
+    synctl -w $CONFIG/workers/synchrotron.yaml restart
+
+All of the above is highly experimental and subject to change as Synapse evolves,
+but documenting it here to help folks needing highly scalable Synapses similar
+to the one running matrix.org!
+

jenkins-dendron-postgres.sh

@@ -4,84 +4,22 @@ set -eux
 
 : ${WORKSPACE:="$(pwd)"}
 
+export WORKSPACE
 export PYTHONDONTWRITEBYTECODE=yep
 export SYNAPSE_CACHE_FACTOR=1
 
-# Output test results as junit xml
-export TRIAL_FLAGS="--reporter=subunit"
-export TOXSUFFIX="| subunit-1to2 | subunit2junitxml --no-passthrough --output-to=results.xml"
-# Write coverage reports to a separate file for each process
-export COVERAGE_OPTS="-p"
-export DUMP_COVERAGE_COMMAND="coverage help"
+./jenkins/prepare_synapse.sh
+./jenkins/clone.sh sytest https://github.com/matrix-org/sytest.git
+./jenkins/clone.sh dendron https://github.com/matrix-org/dendron.git
+./dendron/jenkins/build_dendron.sh
+./sytest/jenkins/prep_sytest_for_postgres.sh
 
-# Output flake8 violations to violations.flake8.log
-# Don't exit with non-0 status code on Jenkins,
-# so that the build steps continue and a later step can decided whether to
-# UNSTABLE or FAILURE this build.
-export PEP8SUFFIX="--output-file=violations.flake8.log || echo flake8 finished with status code \$?"
-
-rm .coverage* || echo "No coverage files to remove"
-
-tox --notest -e py27
-
-TOX_BIN=$WORKSPACE/.tox/py27/bin
-python synapse/python_dependencies.py | xargs -n1 $TOX_BIN/pip install
-$TOX_BIN/pip install psycopg2
-$TOX_BIN/pip install lxml
-
-: ${GIT_BRANCH:="origin/$(git rev-parse --abbrev-ref HEAD)"}
-
-if [[ ! -e .dendron-base ]]; then
-    git clone https://github.com/matrix-org/dendron.git .dendron-base --mirror
-else
-    (cd .dendron-base; git fetch -p)
-fi
-
-rm -rf dendron
-git clone .dendron-base dendron --shared
-cd dendron
-
-: ${GOPATH:=${WORKSPACE}/.gopath}
-if [[ "${GOPATH}" != *:* ]]; then
-  mkdir -p "${GOPATH}"
-  export PATH="${GOPATH}/bin:${PATH}"
-fi
-export GOPATH
-
-git checkout "${GIT_BRANCH}" || (echo >&2 "No ref ${GIT_BRANCH} found, falling back to develop" ; git checkout develop)
-
-go get github.com/constabulary/gb/...
-gb generate
-gb build
-
-cd ..
-
-
-if [[ ! -e .sytest-base ]]; then
-  git clone https://github.com/matrix-org/sytest.git .sytest-base --mirror
-else
-  (cd .sytest-base; git fetch -p)
-fi
-
-rm -rf sytest
-git clone .sytest-base sytest --shared
-cd sytest
-
-git checkout "${GIT_BRANCH}" || (echo >&2 "No ref ${GIT_BRANCH} found, falling back to develop" ; git checkout develop)
-
-: ${PORT_BASE:=8000}
-: ${PORT_COUNT=20}
-
-./jenkins/prep_sytest_for_postgres.sh
-
-mkdir -p var
-
-echo >&2 "Running sytest with PostgreSQL";
-./jenkins/install_and_run.sh --python $TOX_BIN/python \
-                             --synapse-directory $WORKSPACE \
-                             --dendron $WORKSPACE/dendron/bin/dendron \
-                             --pusher \
-                             --synchrotron \
-                             --port-range ${PORT_BASE}:$((PORT_BASE+PORT_COUNT-1))
-
-cd ..
+./sytest/jenkins/install_and_run.sh \
+    --synapse-directory $WORKSPACE \
+    --dendron $WORKSPACE/dendron/bin/dendron \
+    --pusher \
+    --synchrotron \
+    --federation-reader \
+    --client-reader \
+    --appservice \
+    --federation-sender \

jenkins-postgres.sh

@@ -4,61 +4,14 @@ set -eux
 
 : ${WORKSPACE:="$(pwd)"}
 
+export WORKSPACE
 export PYTHONDONTWRITEBYTECODE=yep
 export SYNAPSE_CACHE_FACTOR=1
 
-# Output test results as junit xml
-export TRIAL_FLAGS="--reporter=subunit"
-export TOXSUFFIX="| subunit-1to2 | subunit2junitxml --no-passthrough --output-to=results.xml"
-# Write coverage reports to a separate file for each process
-export COVERAGE_OPTS="-p"
-export DUMP_COVERAGE_COMMAND="coverage help"
+./jenkins/prepare_synapse.sh
+./jenkins/clone.sh sytest https://github.com/matrix-org/sytest.git
 
-# Output flake8 violations to violations.flake8.log
-# Don't exit with non-0 status code on Jenkins,
-# so that the build steps continue and a later step can decided whether to
-# UNSTABLE or FAILURE this build.
-export PEP8SUFFIX="--output-file=violations.flake8.log || echo flake8 finished with status code \$?"
+./sytest/jenkins/prep_sytest_for_postgres.sh
 
-rm .coverage* || echo "No coverage files to remove"
-
-tox --notest -e py27
-
-TOX_BIN=$WORKSPACE/.tox/py27/bin
-python synapse/python_dependencies.py | xargs -n1 $TOX_BIN/pip install
-$TOX_BIN/pip install psycopg2
-$TOX_BIN/pip install lxml
-
-: ${GIT_BRANCH:="origin/$(git rev-parse --abbrev-ref HEAD)"}
-
-if [[ ! -e .sytest-base ]]; then
-  git clone https://github.com/matrix-org/sytest.git .sytest-base --mirror
-else
-  (cd .sytest-base; git fetch -p)
-fi
-
-rm -rf sytest
-git clone .sytest-base sytest --shared
-cd sytest
-
-git checkout "${GIT_BRANCH}" || (echo >&2 "No ref ${GIT_BRANCH} found, falling back to develop" ; git checkout develop)
-
-: ${PORT_BASE:=8000}
-: ${PORT_COUNT=20}
-
-./jenkins/prep_sytest_for_postgres.sh
-
-echo >&2 "Running sytest with PostgreSQL";
-./jenkins/install_and_run.sh --coverage \
-                             --python $TOX_BIN/python \
-                             --synapse-directory $WORKSPACE \
-                             --port-range ${PORT_BASE}:$((PORT_BASE+PORT_COUNT-1)) \
-
-cd ..
-cp sytest/.coverage.* .
-
-# Combine the coverage reports
-echo "Combining:" .coverage.*
-$TOX_BIN/python -m coverage combine
-# Output coverage to coverage.xml
-$TOX_BIN/coverage xml -o coverage.xml
+./sytest/jenkins/install_and_run.sh \
+    --synapse-directory $WORKSPACE \

jenkins-sqlite.sh

@@ -4,55 +4,12 @@ set -eux
 
 : ${WORKSPACE:="$(pwd)"}
 
+export WORKSPACE
 export PYTHONDONTWRITEBYTECODE=yep
 export SYNAPSE_CACHE_FACTOR=1
 
-# Output test results as junit xml
-export TRIAL_FLAGS="--reporter=subunit"
-export TOXSUFFIX="| subunit-1to2 | subunit2junitxml --no-passthrough --output-to=results.xml"
-# Write coverage reports to a separate file for each process
-export COVERAGE_OPTS="-p"
-export DUMP_COVERAGE_COMMAND="coverage help"
+./jenkins/prepare_synapse.sh
+./jenkins/clone.sh sytest https://github.com/matrix-org/sytest.git
 
-# Output flake8 violations to violations.flake8.log
-# Don't exit with non-0 status code on Jenkins,
-# so that the build steps continue and a later step can decided whether to
-# UNSTABLE or FAILURE this build.
-export PEP8SUFFIX="--output-file=violations.flake8.log || echo flake8 finished with status code \$?"
-
-rm .coverage* || echo "No coverage files to remove"
-
-tox --notest -e py27
-TOX_BIN=$WORKSPACE/.tox/py27/bin
-python synapse/python_dependencies.py | xargs -n1 $TOX_BIN/pip install
-$TOX_BIN/pip install lxml
-
-: ${GIT_BRANCH:="origin/$(git rev-parse --abbrev-ref HEAD)"}
-
-if [[ ! -e .sytest-base ]]; then
-  git clone https://github.com/matrix-org/sytest.git .sytest-base --mirror
-else
-  (cd .sytest-base; git fetch -p)
-fi
-
-rm -rf sytest
-git clone .sytest-base sytest --shared
-cd sytest
-
-git checkout "${GIT_BRANCH}" || (echo >&2 "No ref ${GIT_BRANCH} found, falling back to develop" ; git checkout develop)
-
-: ${PORT_COUNT=20}
-: ${PORT_BASE:=8000}
-./jenkins/install_and_run.sh --coverage \
-                             --python $TOX_BIN/python \
-                             --synapse-directory $WORKSPACE \
-                             --port-range ${PORT_BASE}:$((PORT_BASE+PORT_COUNT-1)) \
-
-cd ..
-cp sytest/.coverage.* .
-
-# Combine the coverage reports
-echo "Combining:" .coverage.*
-$TOX_BIN/python -m coverage combine
-# Output coverage to coverage.xml
-$TOX_BIN/coverage xml -o coverage.xml
+./sytest/jenkins/install_and_run.sh \
+    --synapse-directory $WORKSPACE \

jenkins-unittests.sh

@@ -22,4 +22,9 @@ export PEP8SUFFIX="--output-file=violations.flake8.log || echo flake8 finished w
 
 rm .coverage* || echo "No coverage files to remove"
 
+tox --notest -e py27
+TOX_BIN=$WORKSPACE/.tox/py27/bin
+python synapse/python_dependencies.py | xargs -n1 $TOX_BIN/pip install
+$TOX_BIN/pip install lxml
+
 tox -e py27

jenkins/clone.sh

@@ -0,0 +1,44 @@
+#! /bin/bash
+
+# This clones a project from github into a named subdirectory
+# If the project has a branch with the same name as this branch
+# then it will checkout that branch after cloning.
+# Otherwise it will checkout "origin/develop."
+# The first argument is the name of the directory to checkout
+# the branch into.
+# The second argument is the URL of the remote repository to checkout.
+# Usually something like https://github.com/matrix-org/sytest.git
+
+set -eux
+
+NAME=$1
+PROJECT=$2
+BASE=".$NAME-base"
+
+# Update our mirror.
+if [ ! -d ".$NAME-base" ]; then
+  # Create a local mirror of the source repository.
+  # This saves us from having to download the entire repository
+  # when this script is next run.
+  git clone "$PROJECT" "$BASE" --mirror
+else
+  # Fetch any updates from the source repository.
+  (cd "$BASE"; git fetch -p)
+fi
+
+# Remove the existing repository so that we have a clean copy
+rm -rf "$NAME"
+# Cloning with --shared means that we will share portions of the
+# .git directory with our local mirror.
+git clone "$BASE" "$NAME" --shared
+
+# Jenkins may have supplied us with the name of the branch in the
+# environment. Otherwise we will have to guess based on the current
+# commit.
+: ${GIT_BRANCH:="origin/$(git rev-parse --abbrev-ref HEAD)"}
+cd "$NAME"
+# check out the relevant branch
+git checkout "${GIT_BRANCH}" || (
+    echo >&2 "No ref ${GIT_BRANCH} found, falling back to develop"
+    git checkout "origin/develop"
+)

jenkins/prepare_synapse.sh

@@ -0,0 +1,20 @@
+#! /bin/bash
+
+cd "`dirname $0`/.."
+
+TOX_DIR=$WORKSPACE/.tox
+
+mkdir -p $TOX_DIR
+
+if ! [ $TOX_DIR -ef .tox ]; then
+    ln -s "$TOX_DIR" .tox
+fi
+
+# set up the virtualenv
+tox -e py27 --notest -v
+
+TOX_BIN=$TOX_DIR/py27/bin
+$TOX_BIN/pip install setuptools
+{ python synapse/python_dependencies.py
+  echo lxml psycopg2
+} | xargs $TOX_BIN/pip install

res/templates/notif_mail.html

@@ -18,7 +18,9 @@
                                 <div class="summarytext">{{ summary_text }}</div>
                             </td>
                             <td class="logo">
-                                {% if app_name == "Vector" %}
+                                {% if app_name == "Riot" %}
+                                    <img src="http://matrix.org/img/riot-logo-email.png" width="83" height="83" alt="[Riot]"/>
+                                {% elif app_name == "Vector" %}
                                     <img src="http://matrix.org/img/vector-logo-email.png" width="64" height="83" alt="[Vector]"/>
                                 {% else %}
                                     <img src="http://matrix.org/img/matrix-120x51.png" width="120" height="51" alt="[matrix]"/>

scripts-dev/federation_client.py

@@ -116,17 +116,19 @@ def get_json(origin_name, origin_key, destination, path):
     authorization_headers = []
 
     for key, sig in signed_json["signatures"][origin_name].items():
-        authorization_headers.append(bytes(
-            "X-Matrix origin=%s,key=\"%s\",sig=\"%s\"" % (
-                origin_name, key, sig,
-            )
-        ))
+        header = "X-Matrix origin=%s,key=\"%s\",sig=\"%s\"" % (
+            origin_name, key, sig,
+        )
+        authorization_headers.append(bytes(header))
+        sys.stderr.write(header)
+        sys.stderr.write("\n")
 
     result = requests.get(
         lookup(destination, path),
         headers={"Authorization": authorization_headers[0]},
         verify=False,
     )
+    sys.stderr.write("Status Code: %d\n" % (result.status_code,))
     return result.json()
 
 
@@ -141,6 +143,7 @@ def main():
     )
 
     json.dump(result, sys.stdout)
+    print ""
 
 if __name__ == "__main__":
     main()

scripts/synapse_port_db

@@ -34,11 +34,12 @@ logger = logging.getLogger("synapse_port_db")
 
 
 BOOLEAN_COLUMNS = {
-    "events": ["processed", "outlier"],
+    "events": ["processed", "outlier", "contains_url"],
     "rooms": ["is_public"],
     "event_edges": ["is_state"],
     "presence_list": ["accepted"],
     "presence_stream": ["currently_active"],
+    "public_room_list_stream": ["visibility"],
 }
 
 
@@ -71,6 +72,14 @@ APPEND_ONLY_TABLES = [
     "event_to_state_groups",
     "rejections",
     "event_search",
+    "presence_stream",
+    "push_rules_stream",
+    "current_state_resets",
+    "ex_outlier_stream",
+    "cache_invalidation_stream",
+    "public_room_list_stream",
+    "state_group_edges",
+    "stream_ordering_to_exterm",
 ]
 
 
@@ -92,8 +101,12 @@ class Store(object):
 
     _simple_select_onecol_txn = SQLBaseStore.__dict__["_simple_select_onecol_txn"]
     _simple_select_onecol = SQLBaseStore.__dict__["_simple_select_onecol"]
+    _simple_select_one = SQLBaseStore.__dict__["_simple_select_one"]
+    _simple_select_one_txn = SQLBaseStore.__dict__["_simple_select_one_txn"]
     _simple_select_one_onecol = SQLBaseStore.__dict__["_simple_select_one_onecol"]
-    _simple_select_one_onecol_txn = SQLBaseStore.__dict__["_simple_select_one_onecol_txn"]
+    _simple_select_one_onecol_txn = SQLBaseStore.__dict__[
+        "_simple_select_one_onecol_txn"
+    ]
 
     _simple_update_one = SQLBaseStore.__dict__["_simple_update_one"]
     _simple_update_one_txn = SQLBaseStore.__dict__["_simple_update_one_txn"]
@@ -158,31 +171,40 @@ class Porter(object):
     def setup_table(self, table):
         if table in APPEND_ONLY_TABLES:
             # It's safe to just carry on inserting.
-            next_chunk = yield self.postgres_store._simple_select_one_onecol(
+            row = yield self.postgres_store._simple_select_one(
                 table="port_from_sqlite3",
                 keyvalues={"table_name": table},
-                retcol="rowid",
+                retcols=("forward_rowid", "backward_rowid"),
                 allow_none=True,
             )
 
             total_to_port = None
-            if next_chunk is None:
+            if row is None:
                 if table == "sent_transactions":
-                    next_chunk, already_ported, total_to_port = (
+                    forward_chunk, already_ported, total_to_port = (
                         yield self._setup_sent_transactions()
                     )
+                    backward_chunk = 0
                 else:
                     yield self.postgres_store._simple_insert(
                         table="port_from_sqlite3",
-                        values={"table_name": table, "rowid": 1}
+                        values={
+                            "table_name": table,
+                            "forward_rowid": 1,
+                            "backward_rowid": 0,
+                        }
                     )
 
-                    next_chunk = 1
+                    forward_chunk = 1
+                    backward_chunk = 0
                     already_ported = 0
+            else:
+                forward_chunk = row["forward_rowid"]
+                backward_chunk = row["backward_rowid"]
 
             if total_to_port is None:
                 already_ported, total_to_port = yield self._get_total_count_to_port(
-                    table, next_chunk
+                    table, forward_chunk, backward_chunk
                 )
         else:
             def delete_all(txn):
@@ -196,46 +218,85 @@ class Porter(object):
 
             yield self.postgres_store._simple_insert(
                 table="port_from_sqlite3",
-                values={"table_name": table, "rowid": 0}
+                values={
+                    "table_name": table,
+                    "forward_rowid": 1,
+                    "backward_rowid": 0,
+                }
             )
 
-            next_chunk = 1
+            forward_chunk = 1
+            backward_chunk = 0
 
             already_ported, total_to_port = yield self._get_total_count_to_port(
-                table, next_chunk
+                table, forward_chunk, backward_chunk
             )
 
-        defer.returnValue((table, already_ported, total_to_port, next_chunk))
+        defer.returnValue(
+            (table, already_ported, total_to_port, forward_chunk, backward_chunk)
+        )
 
     @defer.inlineCallbacks
-    def handle_table(self, table, postgres_size, table_size, next_chunk):
+    def handle_table(self, table, postgres_size, table_size, forward_chunk,
+                     backward_chunk):
         if not table_size:
             return
 
         self.progress.add_table(table, postgres_size, table_size)
 
         if table == "event_search":
-            yield self.handle_search_table(postgres_size, table_size, next_chunk)
+            yield self.handle_search_table(
+                postgres_size, table_size, forward_chunk, backward_chunk
+            )
             return
 
-        select = (
+        forward_select = (
             "SELECT rowid, * FROM %s WHERE rowid >= ? ORDER BY rowid LIMIT ?"
             % (table,)
         )
 
+        backward_select = (
+            "SELECT rowid, * FROM %s WHERE rowid <= ? ORDER BY rowid LIMIT ?"
+            % (table,)
+        )
+
+        do_forward = [True]
+        do_backward = [True]
+
         while True:
             def r(txn):
-                txn.execute(select, (next_chunk, self.batch_size,))
-                rows = txn.fetchall()
-                headers = [column[0] for column in txn.description]
+                forward_rows = []
+                backward_rows = []
+                if do_forward[0]:
+                    txn.execute(forward_select, (forward_chunk, self.batch_size,))
+                    forward_rows = txn.fetchall()
+                    if not forward_rows:
+                        do_forward[0] = False
 
-                return headers, rows
+                if do_backward[0]:
+                    txn.execute(backward_select, (backward_chunk, self.batch_size,))
+                    backward_rows = txn.fetchall()
+                    if not backward_rows:
+                        do_backward[0] = False
 
-            headers, rows = yield self.sqlite_store.runInteraction("select", r)
+                if forward_rows or backward_rows:
+                    headers = [column[0] for column in txn.description]
+                else:
+                    headers = None
 
-            if rows:
-                next_chunk = rows[-1][0] + 1
+                return headers, forward_rows, backward_rows
 
+            headers, frows, brows = yield self.sqlite_store.runInteraction(
+                "select", r
+            )
+
+            if frows or brows:
+                if frows:
+                    forward_chunk = max(row[0] for row in frows) + 1
+                if brows:
+                    backward_chunk = min(row[0] for row in brows) - 1
+
+                rows = frows + brows
                 self._convert_rows(table, headers, rows)
 
                 def insert(txn):
@@ -247,7 +308,10 @@ class Porter(object):
                         txn,
                         table="port_from_sqlite3",
                         keyvalues={"table_name": table},
-                        updatevalues={"rowid": next_chunk},
+                        updatevalues={
+                            "forward_rowid": forward_chunk,
+                            "backward_rowid": backward_chunk,
+                        },
                     )
 
                 yield self.postgres_store.execute(insert)
@@ -259,7 +323,8 @@ class Porter(object):
                 return
 
     @defer.inlineCallbacks
-    def handle_search_table(self, postgres_size, table_size, next_chunk):
+    def handle_search_table(self, postgres_size, table_size, forward_chunk,
+                            backward_chunk):
         select = (
             "SELECT es.rowid, es.*, e.origin_server_ts, e.stream_ordering"
             " FROM event_search as es"
@@ -270,7 +335,7 @@ class Porter(object):
 
         while True:
             def r(txn):
-                txn.execute(select, (next_chunk, self.batch_size,))
+                txn.execute(select, (forward_chunk, self.batch_size,))
                 rows = txn.fetchall()
                 headers = [column[0] for column in txn.description]
 
@@ -279,7 +344,7 @@ class Porter(object):
             headers, rows = yield self.sqlite_store.runInteraction("select", r)
 
             if rows:
-                next_chunk = rows[-1][0] + 1
+                forward_chunk = rows[-1][0] + 1
 
                 # We have to treat event_search differently since it has a
                 # different structure in the two different databases.
@@ -312,7 +377,10 @@ class Porter(object):
                         txn,
                         table="port_from_sqlite3",
                         keyvalues={"table_name": "event_search"},
-                        updatevalues={"rowid": next_chunk},
+                        updatevalues={
+                            "forward_rowid": forward_chunk,
+                            "backward_rowid": backward_chunk,
+                        },
                     )
 
                 yield self.postgres_store.execute(insert)
@@ -324,7 +392,6 @@ class Porter(object):
             else:
                 return
 
-
     def setup_db(self, db_config, database_engine):
         db_conn = database_engine.module.connect(
             **{
@@ -395,10 +462,32 @@ class Porter(object):
                 txn.execute(
                     "CREATE TABLE port_from_sqlite3 ("
                     " table_name varchar(100) NOT NULL UNIQUE,"
-                    " rowid bigint NOT NULL"
+                    " forward_rowid bigint NOT NULL,"
+                    " backward_rowid bigint NOT NULL"
                     ")"
                 )
 
+            # The old port script created a table with just a "rowid" column.
+            # We want people to be able to rerun this script from an old port
+            # so that they can pick up any missing events that were not
+            # ported across.
+            def alter_table(txn):
+                txn.execute(
+                    "ALTER TABLE IF EXISTS port_from_sqlite3"
+                    " RENAME rowid TO forward_rowid"
+                )
+                txn.execute(
+                    "ALTER TABLE IF EXISTS port_from_sqlite3"
+                    " ADD backward_rowid bigint NOT NULL DEFAULT 0"
+                )
+
+            try:
+                yield self.postgres_store.runInteraction(
+                    "alter_table", alter_table
+                )
+            except Exception as e:
+                logger.info("Failed to create port table: %s", e)
+
             try:
                 yield self.postgres_store.runInteraction(
                     "create_port_table", create_port_table
@@ -458,7 +547,7 @@ class Porter(object):
     @defer.inlineCallbacks
     def _setup_sent_transactions(self):
         # Only save things from the last day
-        yesterday = int(time.time()*1000) - 86400000
+        yesterday = int(time.time() * 1000) - 86400000
 
         # And save the max transaction id from each destination
         select = (
@@ -514,7 +603,11 @@ class Porter(object):
 
         yield self.postgres_store._simple_insert(
             table="port_from_sqlite3",
-            values={"table_name": "sent_transactions", "rowid": next_chunk}
+            values={
+                "table_name": "sent_transactions",
+                "forward_rowid": next_chunk,
+                "backward_rowid": 0,
+            }
         )
 
         def get_sent_table_size(txn):
@@ -535,13 +628,18 @@ class Porter(object):
         defer.returnValue((next_chunk, inserted_rows, total_count))
 
     @defer.inlineCallbacks
-    def _get_remaining_count_to_port(self, table, next_chunk):
-        rows = yield self.sqlite_store.execute_sql(
+    def _get_remaining_count_to_port(self, table, forward_chunk, backward_chunk):
+        frows = yield self.sqlite_store.execute_sql(
             "SELECT count(*) FROM %s WHERE rowid >= ?" % (table,),
-            next_chunk,
+            forward_chunk,
         )
 
-        defer.returnValue(rows[0][0])
+        brows = yield self.sqlite_store.execute_sql(
+            "SELECT count(*) FROM %s WHERE rowid <= ?" % (table,),
+            backward_chunk,
+        )
+
+        defer.returnValue(frows[0][0] + brows[0][0])
 
     @defer.inlineCallbacks
     def _get_already_ported_count(self, table):
@@ -552,10 +650,10 @@ class Porter(object):
         defer.returnValue(rows[0][0])
 
     @defer.inlineCallbacks
-    def _get_total_count_to_port(self, table, next_chunk):
+    def _get_total_count_to_port(self, table, forward_chunk, backward_chunk):
         remaining, done = yield defer.gatherResults(
             [
-                self._get_remaining_count_to_port(table, next_chunk),
+                self._get_remaining_count_to_port(table, forward_chunk, backward_chunk),
                 self._get_already_ported_count(table),
             ],
             consumeErrors=True,
@@ -686,7 +784,7 @@ class CursesProgress(Progress):
             color = curses.color_pair(2) if perc == 100 else curses.color_pair(1)
 
             self.stdscr.addstr(
-                i+2, left_margin + max_len - len(table),
+                i + 2, left_margin + max_len - len(table),
                 table,
                 curses.A_BOLD | color,
             )
@@ -694,18 +792,18 @@ class CursesProgress(Progress):
             size = 20
 
             progress = "[%s%s]" % (
-                "#" * int(perc*size/100),
-                " " * (size - int(perc*size/100)),
+                "#" * int(perc * size / 100),
+                " " * (size - int(perc * size / 100)),
             )
 
             self.stdscr.addstr(
-                i+2, left_margin + max_len + middle_space,
+                i + 2, left_margin + max_len + middle_space,
                 "%s %3d%% (%d/%d)" % (progress, perc, data["num_done"], data["total"]),
             )
 
         if self.finished:
             self.stdscr.addstr(
-                rows-1, 0,
+                rows - 1, 0,
                 "Press any key to exit...",
             )
 

setup.cfg

@@ -16,7 +16,5 @@ ignore =
 
 [flake8]
 max-line-length = 90
-ignore = W503 ; W503 requires that binary operators be at the end, not start, of lines. Erik doesn't like it.
-
-[pep8]
-max-line-length = 90
+#  W503 requires that binary operators be at the end, not start, of lines. Erik doesn't like it.
+ignore = W503

setup.py

@@ -23,6 +23,45 @@ import sys
 here = os.path.abspath(os.path.dirname(__file__))
 
 
+# Some notes on `setup.py test`:
+#
+# Once upon a time we used to try to make `setup.py test` run `tox` to run the
+# tests. That's a bad idea for three reasons:
+#
+# 1: `setup.py test` is supposed to find out whether the tests work in the
+#    *current* environmentt, not whatever tox sets up.
+# 2: Empirically, trying to install tox during the test run wasn't working ("No
+#    module named virtualenv").
+# 3: The tox documentation advises against it[1].
+#
+# Even further back in time, we used to use setuptools_trial [2]. That has its
+# own set of issues: for instance, it requires installation of Twisted to build
+# an sdist (because the recommended mode of usage is to add it to
+# `setup_requires`). That in turn means that in order to successfully run tox
+# you have to have the python header files installed for whichever version of
+# python tox uses (which is python3 on recent ubuntus, for example).
+#
+# So, for now at least, we stick with what appears to be the convention among
+# Twisted projects, and don't attempt to do anything when someone runs
+# `setup.py test`; instead we direct people to run `trial` directly if they
+# care.
+#
+# [1]: http://tox.readthedocs.io/en/2.5.0/example/basic.html#integration-with-setup-py-test-command
+# [2]: https://pypi.python.org/pypi/setuptools_trial
+class TestCommand(Command):
+    user_options = []
+
+    def initialize_options(self):
+        pass
+
+    def finalize_options(self):
+        pass
+
+    def run(self):
+        print ("""Synapse's tests cannot be run via setup.py. To run them, try:
+     PYTHONPATH="." trial tests
+""")
+
 def read_file(path_segments):
     """Read a file from the package. Takes a list of strings to join to
     make the path"""
@@ -39,38 +78,6 @@ def exec_file(path_segments):
     return result
 
 
-class Tox(Command):
-    user_options = [('tox-args=', 'a', "Arguments to pass to tox")]
-
-    def initialize_options(self):
-        self.tox_args = None
-
-    def finalize_options(self):
-        self.test_args = []
-        self.test_suite = True
-
-    def run(self):
-        #import here, cause outside the eggs aren't loaded
-        try:
-            import tox
-        except ImportError:
-            try:
-                self.distribution.fetch_build_eggs("tox")
-                import tox
-            except:
-                raise RuntimeError(
-                    "The tests need 'tox' to run. Please install 'tox'."
-                )
-        import shlex
-        args = self.tox_args
-        if args:
-            args = shlex.split(self.tox_args)
-        else:
-            args = []
-        errno = tox.cmdline(args=args)
-        sys.exit(errno)
-
-
 version = exec_file(("synapse", "__init__.py"))["__version__"]
 dependencies = exec_file(("synapse", "python_dependencies.py"))
 long_description = read_file(("README.rst",))
@@ -86,5 +93,5 @@ setup(
     zip_safe=False,
     long_description=long_description,
     scripts=["synctl"] + glob.glob("scripts/*"),
-    cmdclass={'test': Tox},
+    cmdclass={'test': TestCommand},
 )

synapse/__init__.py

@@ -16,4 +16,4 @@
 """ This is a reference implementation of a Matrix home server.
 """
 
-__version__ = "0.16.1-r1"
+__version__ = "0.19.0-rc3"

synapse/api/constants.py

@@ -85,3 +85,8 @@ class RoomCreationPreset(object):
     PRIVATE_CHAT = "private_chat"
     PUBLIC_CHAT = "public_chat"
     TRUSTED_PRIVATE_CHAT = "trusted_private_chat"
+
+
+class ThirdPartyEntityKind(object):
+    USER = "user"
+    LOCATION = "location"

synapse/api/errors.py

@@ -39,10 +39,12 @@ class Codes(object):
     CAPTCHA_NEEDED = "M_CAPTCHA_NEEDED"
     CAPTCHA_INVALID = "M_CAPTCHA_INVALID"
     MISSING_PARAM = "M_MISSING_PARAM"
+    INVALID_PARAM = "M_INVALID_PARAM"
     TOO_LARGE = "M_TOO_LARGE"
     EXCLUSIVE = "M_EXCLUSIVE"
     THREEPID_AUTH_FAILED = "M_THREEPID_AUTH_FAILED"
     THREEPID_IN_USE = "M_THREEPID_IN_USE"
+    THREEPID_NOT_FOUND = "M_THREEPID_NOT_FOUND"
     INVALID_USERNAME = "M_INVALID_USERNAME"
     SERVER_NOT_TRUSTED = "M_SERVER_NOT_TRUSTED"
 

synapse/api/filtering.py

@@ -71,6 +71,21 @@ class Filtering(object):
                 if key in user_filter_json["room"]:
                     self._check_definition(user_filter_json["room"][key])
 
+        if "event_fields" in user_filter_json:
+            if type(user_filter_json["event_fields"]) != list:
+                raise SynapseError(400, "event_fields must be a list of strings")
+            for field in user_filter_json["event_fields"]:
+                if not isinstance(field, basestring):
+                    raise SynapseError(400, "Event field must be a string")
+                # Don't allow '\\' in event field filters. This makes matching
+                # events a lot easier as we can then use a negative lookbehind
+                # assertion to split '\.' If we allowed \\ then it would
+                # incorrectly split '\\.' See synapse.events.utils.serialize_event
+                if r'\\' in field:
+                    raise SynapseError(
+                        400, r'The escape character \ cannot itself be escaped'
+                    )
+
     def _check_definition_room_lists(self, definition):
         """Check that "rooms" and "not_rooms" are lists of room ids if they
         are present
@@ -152,6 +167,7 @@ class FilterCollection(object):
         self.include_leave = filter_json.get("room", {}).get(
             "include_leave", False
         )
+        self.event_fields = filter_json.get("event_fields", [])
 
     def __repr__(self):
         return "<FilterCollection %s>" % (json.dumps(self._filter_json),)
@@ -186,11 +202,51 @@ class FilterCollection(object):
     def filter_room_account_data(self, events):
         return self._room_account_data.filter(self._room_filter.filter(events))
 
+    def blocks_all_presence(self):
+        return (
+            self._presence_filter.filters_all_types() or
+            self._presence_filter.filters_all_senders()
+        )
+
+    def blocks_all_room_ephemeral(self):
+        return (
+            self._room_ephemeral_filter.filters_all_types() or
+            self._room_ephemeral_filter.filters_all_senders() or
+            self._room_ephemeral_filter.filters_all_rooms()
+        )
+
+    def blocks_all_room_timeline(self):
+        return (
+            self._room_timeline_filter.filters_all_types() or
+            self._room_timeline_filter.filters_all_senders() or
+            self._room_timeline_filter.filters_all_rooms()
+        )
+
 
 class Filter(object):
     def __init__(self, filter_json):
         self.filter_json = filter_json
 
+        self.types = self.filter_json.get("types", None)
+        self.not_types = self.filter_json.get("not_types", [])
+
+        self.rooms = self.filter_json.get("rooms", None)
+        self.not_rooms = self.filter_json.get("not_rooms", [])
+
+        self.senders = self.filter_json.get("senders", None)
+        self.not_senders = self.filter_json.get("not_senders", [])
+
+        self.contains_url = self.filter_json.get("contains_url", None)
+
+    def filters_all_types(self):
+        return "*" in self.not_types
+
+    def filters_all_senders(self):
+        return "*" in self.not_senders
+
+    def filters_all_rooms(self):
+        return "*" in self.not_rooms
+
     def check(self, event):
         """Checks whether the filter matches the given event.
 
@@ -209,9 +265,10 @@ class Filter(object):
             event.get("room_id", None),
             sender,
             event.get("type", None),
+            "url" in event.get("content", {})
         )
 
-    def check_fields(self, room_id, sender, event_type):
+    def check_fields(self, room_id, sender, event_type, contains_url):
         """Checks whether the filter matches the given event fields.
 
         Returns:
@@ -225,15 +282,20 @@ class Filter(object):
 
         for name, match_func in literal_keys.items():
             not_name = "not_%s" % (name,)
-            disallowed_values = self.filter_json.get(not_name, [])
+            disallowed_values = getattr(self, not_name)
             if any(map(match_func, disallowed_values)):
                 return False
 
-            allowed_values = self.filter_json.get(name, None)
+            allowed_values = getattr(self, name)
             if allowed_values is not None:
                 if not any(map(match_func, allowed_values)):
                     return False
 
+        contains_url_filter = self.filter_json.get("contains_url")
+        if contains_url_filter is not None:
+            if contains_url_filter != contains_url:
+                return False
+
         return True
 
     def filter_rooms(self, room_ids):

synapse/api/ratelimiting.py

@@ -23,7 +23,7 @@ class Ratelimiter(object):
     def __init__(self):
         self.message_counts = collections.OrderedDict()
 
-    def send_message(self, user_id, time_now_s, msg_rate_hz, burst_count):
+    def send_message(self, user_id, time_now_s, msg_rate_hz, burst_count, update=True):
         """Can the user send a message?
         Args:
             user_id: The user sending a message.
@@ -32,12 +32,15 @@ class Ratelimiter(object):
                 second.
             burst_count: How many messages the user can send before being
                 limited.
+            update (bool): Whether to update the message rates or not. This is
+                useful to check if a message would be allowed to be sent before
+                its ready to be actually sent.
         Returns:
             A pair of a bool indicating if they can send a message now and a
                 time in seconds of when they can next send a message.
         """
         self.prune_message_counts(time_now_s)
-        message_count, time_start, _ignored = self.message_counts.pop(
+        message_count, time_start, _ignored = self.message_counts.get(
             user_id, (0., time_now_s, None),
         )
         time_delta = time_now_s - time_start
@@ -52,9 +55,10 @@ class Ratelimiter(object):
             allowed = True
             message_count += 1
 
-        self.message_counts[user_id] = (
-            message_count, time_start, msg_rate_hz
-        )
+        if update:
+            self.message_counts[user_id] = (
+                message_count, time_start, msg_rate_hz
+            )
 
         if msg_rate_hz > 0:
             time_allowed = (

synapse/api/urls.py

@@ -25,4 +25,3 @@ SERVER_KEY_PREFIX = "/_matrix/key/v1"
 SERVER_KEY_V2_PREFIX = "/_matrix/key/v2"
 MEDIA_PREFIX = "/_matrix/media/r0"
 LEGACY_MEDIA_PREFIX = "/_matrix/media/v1"
-APP_SERVICE_PREFIX = "/_matrix/appservice/v1"

synapse/app/__init__.py

@@ -16,13 +16,11 @@
 import sys
 sys.dont_write_bytecode = True
 
-from synapse.python_dependencies import (
-    check_requirements, MissingRequirementError
-)  # NOQA
+from synapse import python_dependencies   # noqa: E402
 
 try:
-    check_requirements()
-except MissingRequirementError as e:
+    python_dependencies.check_requirements()
+except python_dependencies.MissingRequirementError as e:
     message = "\n".join([
         "Missing Requirement: %s" % (e.message,),
         "To install run:",

synapse/app/appservice.py

@@ -0,0 +1,220 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+# Copyright 2016 OpenMarket Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import synapse
+
+from synapse.server import HomeServer
+from synapse.config._base import ConfigError
+from synapse.config.logger import setup_logging
+from synapse.config.homeserver import HomeServerConfig
+from synapse.http.site import SynapseSite
+from synapse.metrics.resource import MetricsResource, METRICS_PREFIX
+from synapse.replication.slave.storage.directory import DirectoryStore
+from synapse.replication.slave.storage.events import SlavedEventStore
+from synapse.replication.slave.storage.appservice import SlavedApplicationServiceStore
+from synapse.replication.slave.storage.registration import SlavedRegistrationStore
+from synapse.storage.engines import create_engine
+from synapse.util.async import sleep
+from synapse.util.httpresourcetree import create_resource_tree
+from synapse.util.logcontext import LoggingContext
+from synapse.util.manhole import manhole
+from synapse.util.rlimit import change_resource_limit
+from synapse.util.versionstring import get_version_string
+
+from synapse import events
+
+from twisted.internet import reactor, defer
+from twisted.web.resource import Resource
+
+from daemonize import Daemonize
+
+import sys
+import logging
+import gc
+
+logger = logging.getLogger("synapse.app.appservice")
+
+
+class AppserviceSlaveStore(
+    DirectoryStore, SlavedEventStore, SlavedApplicationServiceStore,
+    SlavedRegistrationStore,
+):
+    pass
+
+
+class AppserviceServer(HomeServer):
+    def get_db_conn(self, run_new_connection=True):
+        # Any param beginning with cp_ is a parameter for adbapi, and should
+        # not be passed to the database engine.
+        db_params = {
+            k: v for k, v in self.db_config.get("args", {}).items()
+            if not k.startswith("cp_")
+        }
+        db_conn = self.database_engine.module.connect(**db_params)
+
+        if run_new_connection:
+            self.database_engine.on_new_connection(db_conn)
+        return db_conn
+
+    def setup(self):
+        logger.info("Setting up.")
+        self.datastore = AppserviceSlaveStore(self.get_db_conn(), self)
+        logger.info("Finished setting up.")
+
+    def _listen_http(self, listener_config):
+        port = listener_config["port"]
+        bind_addresses = listener_config["bind_addresses"]
+        site_tag = listener_config.get("tag", port)
+        resources = {}
+        for res in listener_config["resources"]:
+            for name in res["names"]:
+                if name == "metrics":
+                    resources[METRICS_PREFIX] = MetricsResource(self)
+
+        root_resource = create_resource_tree(resources, Resource())
+
+        for address in bind_addresses:
+            reactor.listenTCP(
+                port,
+                SynapseSite(
+                    "synapse.access.http.%s" % (site_tag,),
+                    site_tag,
+                    listener_config,
+                    root_resource,
+                ),
+                interface=address
+            )
+
+        logger.info("Synapse appservice now listening on port %d", port)
+
+    def start_listening(self, listeners):
+        for listener in listeners:
+            if listener["type"] == "http":
+                self._listen_http(listener)
+            elif listener["type"] == "manhole":
+                bind_addresses = listener["bind_addresses"]
+
+                for address in bind_addresses:
+                    reactor.listenTCP(
+                        listener["port"],
+                        manhole(
+                            username="matrix",
+                            password="rabbithole",
+                            globals={"hs": self},
+                        ),
+                        interface=address
+                    )
+            else:
+                logger.warn("Unrecognized listener type: %s", listener["type"])
+
+    @defer.inlineCallbacks
+    def replicate(self):
+        http_client = self.get_simple_http_client()
+        store = self.get_datastore()
+        replication_url = self.config.worker_replication_url
+        appservice_handler = self.get_application_service_handler()
+
+        @defer.inlineCallbacks
+        def replicate(results):
+            stream = results.get("events")
+            if stream:
+                max_stream_id = stream["position"]
+                yield appservice_handler.notify_interested_services(max_stream_id)
+
+        while True:
+            try:
+                args = store.stream_positions()
+                args["timeout"] = 30000
+                result = yield http_client.get_json(replication_url, args=args)
+                yield store.process_replication(result)
+                replicate(result)
+            except:
+                logger.exception("Error replicating from %r", replication_url)
+                yield sleep(30)
+
+
+def start(config_options):
+    try:
+        config = HomeServerConfig.load_config(
+            "Synapse appservice", config_options
+        )
+    except ConfigError as e:
+        sys.stderr.write("\n" + e.message + "\n")
+        sys.exit(1)
+
+    assert config.worker_app == "synapse.app.appservice"
+
+    setup_logging(config.worker_log_config, config.worker_log_file)
+
+    events.USE_FROZEN_DICTS = config.use_frozen_dicts
+
+    database_engine = create_engine(config.database_config)
+
+    if config.notify_appservices:
+        sys.stderr.write(
+            "\nThe appservices must be disabled in the main synapse process"
+            "\nbefore they can be run in a separate worker."
+            "\nPlease add ``notify_appservices: false`` to the main config"
+            "\n"
+        )
+        sys.exit(1)
+
+    # Force the pushers to start since they will be disabled in the main config
+    config.notify_appservices = True
+
+    ps = AppserviceServer(
+        config.server_name,
+        db_config=config.database_config,
+        config=config,
+        version_string="Synapse/" + get_version_string(synapse),
+        database_engine=database_engine,
+    )
+
+    ps.setup()
+    ps.start_listening(config.worker_listeners)
+
+    def run():
+        with LoggingContext("run"):
+            logger.info("Running")
+            change_resource_limit(config.soft_file_limit)
+            if config.gc_thresholds:
+                gc.set_threshold(*config.gc_thresholds)
+            reactor.run()
+
+    def start():
+        ps.replicate()
+        ps.get_datastore().start_profiling()
+        ps.get_state_handler().start_caching()
+
+    reactor.callWhenRunning(start)
+
+    if config.worker_daemonize:
+        daemon = Daemonize(
+            app="synapse-appservice",
+            pid=config.worker_pid_file,
+            action=run,
+            auto_close_fds=False,
+            verbose=True,
+            logger=logger,
+        )
+        daemon.start()
+    else:
+        run()
+
+
+if __name__ == '__main__':
+    with LoggingContext("main"):
+        start(sys.argv[1:])

synapse/app/client_reader.py

@@ -0,0 +1,226 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+# Copyright 2016 OpenMarket Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import synapse
+
+from synapse.config._base import ConfigError
+from synapse.config.homeserver import HomeServerConfig
+from synapse.config.logger import setup_logging
+from synapse.http.site import SynapseSite
+from synapse.http.server import JsonResource
+from synapse.metrics.resource import MetricsResource, METRICS_PREFIX
+from synapse.replication.slave.storage._base import BaseSlavedStore
+from synapse.replication.slave.storage.appservice import SlavedApplicationServiceStore
+from synapse.replication.slave.storage.events import SlavedEventStore
+from synapse.replication.slave.storage.keys import SlavedKeyStore
+from synapse.replication.slave.storage.room import RoomStore
+from synapse.replication.slave.storage.directory import DirectoryStore
+from synapse.replication.slave.storage.registration import SlavedRegistrationStore
+from synapse.rest.client.v1.room import PublicRoomListRestServlet
+from synapse.server import HomeServer
+from synapse.storage.client_ips import ClientIpStore
+from synapse.storage.engines import create_engine
+from synapse.util.async import sleep
+from synapse.util.httpresourcetree import create_resource_tree
+from synapse.util.logcontext import LoggingContext
+from synapse.util.manhole import manhole
+from synapse.util.rlimit import change_resource_limit
+from synapse.util.versionstring import get_version_string
+from synapse.crypto import context_factory
+
+from synapse import events
+
+
+from twisted.internet import reactor, defer
+from twisted.web.resource import Resource
+
+from daemonize import Daemonize
+
+import sys
+import logging
+import gc
+
+logger = logging.getLogger("synapse.app.client_reader")
+
+
+class ClientReaderSlavedStore(
+    SlavedEventStore,
+    SlavedKeyStore,
+    RoomStore,
+    DirectoryStore,
+    SlavedApplicationServiceStore,
+    SlavedRegistrationStore,
+    BaseSlavedStore,
+    ClientIpStore,  # After BaseSlavedStore because the constructor is different
+):
+    pass
+
+
+class ClientReaderServer(HomeServer):
+    def get_db_conn(self, run_new_connection=True):
+        # Any param beginning with cp_ is a parameter for adbapi, and should
+        # not be passed to the database engine.
+        db_params = {
+            k: v for k, v in self.db_config.get("args", {}).items()
+            if not k.startswith("cp_")
+        }
+        db_conn = self.database_engine.module.connect(**db_params)
+
+        if run_new_connection:
+            self.database_engine.on_new_connection(db_conn)
+        return db_conn
+
+    def setup(self):
+        logger.info("Setting up.")
+        self.datastore = ClientReaderSlavedStore(self.get_db_conn(), self)
+        logger.info("Finished setting up.")
+
+    def _listen_http(self, listener_config):
+        port = listener_config["port"]
+        bind_addresses = listener_config["bind_addresses"]
+        site_tag = listener_config.get("tag", port)
+        resources = {}
+        for res in listener_config["resources"]:
+            for name in res["names"]:
+                if name == "metrics":
+                    resources[METRICS_PREFIX] = MetricsResource(self)
+                elif name == "client":
+                    resource = JsonResource(self, canonical_json=False)
+                    PublicRoomListRestServlet(self).register(resource)
+                    resources.update({
+                        "/_matrix/client/r0": resource,
+                        "/_matrix/client/unstable": resource,
+                        "/_matrix/client/v2_alpha": resource,
+                        "/_matrix/client/api/v1": resource,
+                    })
+
+        root_resource = create_resource_tree(resources, Resource())
+
+        for address in bind_addresses:
+            reactor.listenTCP(
+                port,
+                SynapseSite(
+                    "synapse.access.http.%s" % (site_tag,),
+                    site_tag,
+                    listener_config,
+                    root_resource,
+                ),
+                interface=address
+            )
+
+        logger.info("Synapse client reader now listening on port %d", port)
+
+    def start_listening(self, listeners):
+        for listener in listeners:
+            if listener["type"] == "http":
+                self._listen_http(listener)
+            elif listener["type"] == "manhole":
+                bind_addresses = listener["bind_addresses"]
+
+                for address in bind_addresses:
+                    reactor.listenTCP(
+                        listener["port"],
+                        manhole(
+                            username="matrix",
+                            password="rabbithole",
+                            globals={"hs": self},
+                        ),
+                        interface=address
+                    )
+            else:
+                logger.warn("Unrecognized listener type: %s", listener["type"])
+
+    @defer.inlineCallbacks
+    def replicate(self):
+        http_client = self.get_simple_http_client()
+        store = self.get_datastore()
+        replication_url = self.config.worker_replication_url
+
+        while True:
+            try:
+                args = store.stream_positions()
+                args["timeout"] = 30000
+                result = yield http_client.get_json(replication_url, args=args)
+                yield store.process_replication(result)
+            except:
+                logger.exception("Error replicating from %r", replication_url)
+                yield sleep(5)
+
+
+def start(config_options):
+    try:
+        config = HomeServerConfig.load_config(
+            "Synapse client reader", config_options
+        )
+    except ConfigError as e:
+        sys.stderr.write("\n" + e.message + "\n")
+        sys.exit(1)
+
+    assert config.worker_app == "synapse.app.client_reader"
+
+    setup_logging(config.worker_log_config, config.worker_log_file)
+
+    events.USE_FROZEN_DICTS = config.use_frozen_dicts
+
+    database_engine = create_engine(config.database_config)
+
+    tls_server_context_factory = context_factory.ServerContextFactory(config)
+
+    ss = ClientReaderServer(
+        config.server_name,
+        db_config=config.database_config,
+        tls_server_context_factory=tls_server_context_factory,
+        config=config,
+        version_string="Synapse/" + get_version_string(synapse),
+        database_engine=database_engine,
+    )
+
+    ss.setup()
+    ss.get_handlers()
+    ss.start_listening(config.worker_listeners)
+
+    def run():
+        with LoggingContext("run"):
+            logger.info("Running")
+            change_resource_limit(config.soft_file_limit)
+            if config.gc_thresholds:
+                gc.set_threshold(*config.gc_thresholds)
+            reactor.run()
+
+    def start():
+        ss.get_state_handler().start_caching()
+        ss.get_datastore().start_profiling()
+        ss.replicate()
+
+    reactor.callWhenRunning(start)
+
+    if config.worker_daemonize:
+        daemon = Daemonize(
+            app="synapse-client-reader",
+            pid=config.worker_pid_file,
+            action=run,
+            auto_close_fds=False,
+            verbose=True,
+            logger=logger,
+        )
+        daemon.start()
+    else:
+        run()
+
+
+if __name__ == '__main__':
+    with LoggingContext("main"):
+        start(sys.argv[1:])

synapse/app/federation_reader.py

@@ -0,0 +1,217 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+# Copyright 2016 OpenMarket Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import synapse
+
+from synapse.config._base import ConfigError
+from synapse.config.homeserver import HomeServerConfig
+from synapse.config.logger import setup_logging
+from synapse.http.site import SynapseSite
+from synapse.metrics.resource import MetricsResource, METRICS_PREFIX
+from synapse.replication.slave.storage._base import BaseSlavedStore
+from synapse.replication.slave.storage.events import SlavedEventStore
+from synapse.replication.slave.storage.keys import SlavedKeyStore
+from synapse.replication.slave.storage.room import RoomStore
+from synapse.replication.slave.storage.transactions import TransactionStore
+from synapse.replication.slave.storage.directory import DirectoryStore
+from synapse.server import HomeServer
+from synapse.storage.engines import create_engine
+from synapse.util.async import sleep
+from synapse.util.httpresourcetree import create_resource_tree
+from synapse.util.logcontext import LoggingContext
+from synapse.util.manhole import manhole
+from synapse.util.rlimit import change_resource_limit
+from synapse.util.versionstring import get_version_string
+from synapse.api.urls import FEDERATION_PREFIX
+from synapse.federation.transport.server import TransportLayerServer
+from synapse.crypto import context_factory
+
+from synapse import events
+
+
+from twisted.internet import reactor, defer
+from twisted.web.resource import Resource
+
+from daemonize import Daemonize
+
+import sys
+import logging
+import gc
+
+logger = logging.getLogger("synapse.app.federation_reader")
+
+
+class FederationReaderSlavedStore(
+    SlavedEventStore,
+    SlavedKeyStore,
+    RoomStore,
+    DirectoryStore,
+    TransactionStore,
+    BaseSlavedStore,
+):
+    pass
+
+
+class FederationReaderServer(HomeServer):
+    def get_db_conn(self, run_new_connection=True):
+        # Any param beginning with cp_ is a parameter for adbapi, and should
+        # not be passed to the database engine.
+        db_params = {
+            k: v for k, v in self.db_config.get("args", {}).items()
+            if not k.startswith("cp_")
+        }
+        db_conn = self.database_engine.module.connect(**db_params)
+
+        if run_new_connection:
+            self.database_engine.on_new_connection(db_conn)
+        return db_conn
+
+    def setup(self):
+        logger.info("Setting up.")
+        self.datastore = FederationReaderSlavedStore(self.get_db_conn(), self)
+        logger.info("Finished setting up.")
+
+    def _listen_http(self, listener_config):
+        port = listener_config["port"]
+        bind_addresses = listener_config["bind_addresses"]
+        site_tag = listener_config.get("tag", port)
+        resources = {}
+        for res in listener_config["resources"]:
+            for name in res["names"]:
+                if name == "metrics":
+                    resources[METRICS_PREFIX] = MetricsResource(self)
+                elif name == "federation":
+                    resources.update({
+                        FEDERATION_PREFIX: TransportLayerServer(self),
+                    })
+
+        root_resource = create_resource_tree(resources, Resource())
+
+        for address in bind_addresses:
+            reactor.listenTCP(
+                port,
+                SynapseSite(
+                    "synapse.access.http.%s" % (site_tag,),
+                    site_tag,
+                    listener_config,
+                    root_resource,
+                ),
+                interface=address
+            )
+
+        logger.info("Synapse federation reader now listening on port %d", port)
+
+    def start_listening(self, listeners):
+        for listener in listeners:
+            if listener["type"] == "http":
+                self._listen_http(listener)
+            elif listener["type"] == "manhole":
+                bind_addresses = listener["bind_addresses"]
+
+                for address in bind_addresses:
+                    reactor.listenTCP(
+                        listener["port"],
+                        manhole(
+                            username="matrix",
+                            password="rabbithole",
+                            globals={"hs": self},
+                        ),
+                        interface=address
+                    )
+            else:
+                logger.warn("Unrecognized listener type: %s", listener["type"])
+
+    @defer.inlineCallbacks
+    def replicate(self):
+        http_client = self.get_simple_http_client()
+        store = self.get_datastore()
+        replication_url = self.config.worker_replication_url
+
+        while True:
+            try:
+                args = store.stream_positions()
+                args["timeout"] = 30000
+                result = yield http_client.get_json(replication_url, args=args)
+                yield store.process_replication(result)
+            except:
+                logger.exception("Error replicating from %r", replication_url)
+                yield sleep(5)
+
+
+def start(config_options):
+    try:
+        config = HomeServerConfig.load_config(
+            "Synapse federation reader", config_options
+        )
+    except ConfigError as e:
+        sys.stderr.write("\n" + e.message + "\n")
+        sys.exit(1)
+
+    assert config.worker_app == "synapse.app.federation_reader"
+
+    setup_logging(config.worker_log_config, config.worker_log_file)
+
+    events.USE_FROZEN_DICTS = config.use_frozen_dicts
+
+    database_engine = create_engine(config.database_config)
+
+    tls_server_context_factory = context_factory.ServerContextFactory(config)
+
+    ss = FederationReaderServer(
+        config.server_name,
+        db_config=config.database_config,
+        tls_server_context_factory=tls_server_context_factory,
+        config=config,
+        version_string="Synapse/" + get_version_string(synapse),
+        database_engine=database_engine,
+    )
+
+    ss.setup()
+    ss.get_handlers()
+    ss.start_listening(config.worker_listeners)
+
+    def run():
+        with LoggingContext("run"):
+            logger.info("Running")
+            change_resource_limit(config.soft_file_limit)
+            if config.gc_thresholds:
+                gc.set_threshold(*config.gc_thresholds)
+            reactor.run()
+
+    def start():
+        ss.get_state_handler().start_caching()
+        ss.get_datastore().start_profiling()
+        ss.replicate()
+
+    reactor.callWhenRunning(start)
+
+    if config.worker_daemonize:
+        daemon = Daemonize(
+            app="synapse-federation-reader",
+            pid=config.worker_pid_file,
+            action=run,
+            auto_close_fds=False,
+            verbose=True,
+            logger=logger,
+        )
+        daemon.start()
+    else:
+        run()
+
+
+if __name__ == '__main__':
+    with LoggingContext("main"):
+        start(sys.argv[1:])

synapse/app/federation_sender.py

@@ -0,0 +1,338 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+# Copyright 2016 OpenMarket Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import synapse
+
+from synapse.server import HomeServer
+from synapse.config._base import ConfigError
+from synapse.config.logger import setup_logging
+from synapse.config.homeserver import HomeServerConfig
+from synapse.crypto import context_factory
+from synapse.http.site import SynapseSite
+from synapse.federation import send_queue
+from synapse.federation.units import Edu
+from synapse.metrics.resource import MetricsResource, METRICS_PREFIX
+from synapse.replication.slave.storage.deviceinbox import SlavedDeviceInboxStore
+from synapse.replication.slave.storage.events import SlavedEventStore
+from synapse.replication.slave.storage.receipts import SlavedReceiptsStore
+from synapse.replication.slave.storage.registration import SlavedRegistrationStore
+from synapse.replication.slave.storage.transactions import TransactionStore
+from synapse.replication.slave.storage.devices import SlavedDeviceStore
+from synapse.storage.engines import create_engine
+from synapse.storage.presence import UserPresenceState
+from synapse.util.async import sleep
+from synapse.util.httpresourcetree import create_resource_tree
+from synapse.util.logcontext import LoggingContext
+from synapse.util.manhole import manhole
+from synapse.util.rlimit import change_resource_limit
+from synapse.util.versionstring import get_version_string
+
+from synapse import events
+
+from twisted.internet import reactor, defer
+from twisted.web.resource import Resource
+
+from daemonize import Daemonize
+
+import sys
+import logging
+import gc
+import ujson as json
+
+logger = logging.getLogger("synapse.app.appservice")
+
+
+class FederationSenderSlaveStore(
+    SlavedDeviceInboxStore, TransactionStore, SlavedReceiptsStore, SlavedEventStore,
+    SlavedRegistrationStore, SlavedDeviceStore,
+):
+    pass
+
+
+class FederationSenderServer(HomeServer):
+    def get_db_conn(self, run_new_connection=True):
+        # Any param beginning with cp_ is a parameter for adbapi, and should
+        # not be passed to the database engine.
+        db_params = {
+            k: v for k, v in self.db_config.get("args", {}).items()
+            if not k.startswith("cp_")
+        }
+        db_conn = self.database_engine.module.connect(**db_params)
+
+        if run_new_connection:
+            self.database_engine.on_new_connection(db_conn)
+        return db_conn
+
+    def setup(self):
+        logger.info("Setting up.")
+        self.datastore = FederationSenderSlaveStore(self.get_db_conn(), self)
+        logger.info("Finished setting up.")
+
+    def _listen_http(self, listener_config):
+        port = listener_config["port"]
+        bind_addresses = listener_config["bind_addresses"]
+        site_tag = listener_config.get("tag", port)
+        resources = {}
+        for res in listener_config["resources"]:
+            for name in res["names"]:
+                if name == "metrics":
+                    resources[METRICS_PREFIX] = MetricsResource(self)
+
+        root_resource = create_resource_tree(resources, Resource())
+
+        for address in bind_addresses:
+            reactor.listenTCP(
+                port,
+                SynapseSite(
+                    "synapse.access.http.%s" % (site_tag,),
+                    site_tag,
+                    listener_config,
+                    root_resource,
+                ),
+                interface=address
+            )
+
+        logger.info("Synapse federation_sender now listening on port %d", port)
+
+    def start_listening(self, listeners):
+        for listener in listeners:
+            if listener["type"] == "http":
+                self._listen_http(listener)
+            elif listener["type"] == "manhole":
+                bind_addresses = listener["bind_addresses"]
+
+                for address in bind_addresses:
+                    reactor.listenTCP(
+                        listener["port"],
+                        manhole(
+                            username="matrix",
+                            password="rabbithole",
+                            globals={"hs": self},
+                        ),
+                        interface=address
+                    )
+            else:
+                logger.warn("Unrecognized listener type: %s", listener["type"])
+
+    @defer.inlineCallbacks
+    def replicate(self):
+        http_client = self.get_simple_http_client()
+        store = self.get_datastore()
+        replication_url = self.config.worker_replication_url
+        send_handler = FederationSenderHandler(self)
+
+        send_handler.on_start()
+
+        while True:
+            try:
+                args = store.stream_positions()
+                args.update((yield send_handler.stream_positions()))
+                args["timeout"] = 30000
+                result = yield http_client.get_json(replication_url, args=args)
+                yield store.process_replication(result)
+                yield send_handler.process_replication(result)
+            except:
+                logger.exception("Error replicating from %r", replication_url)
+                yield sleep(30)
+
+
+def start(config_options):
+    try:
+        config = HomeServerConfig.load_config(
+            "Synapse federation sender", config_options
+        )
+    except ConfigError as e:
+        sys.stderr.write("\n" + e.message + "\n")
+        sys.exit(1)
+
+    assert config.worker_app == "synapse.app.federation_sender"
+
+    setup_logging(config.worker_log_config, config.worker_log_file)
+
+    events.USE_FROZEN_DICTS = config.use_frozen_dicts
+
+    database_engine = create_engine(config.database_config)
+
+    if config.send_federation:
+        sys.stderr.write(
+            "\nThe send_federation must be disabled in the main synapse process"
+            "\nbefore they can be run in a separate worker."
+            "\nPlease add ``send_federation: false`` to the main config"
+            "\n"
+        )
+        sys.exit(1)
+
+    # Force the pushers to start since they will be disabled in the main config
+    config.send_federation = True
+
+    tls_server_context_factory = context_factory.ServerContextFactory(config)
+
+    ps = FederationSenderServer(
+        config.server_name,
+        db_config=config.database_config,
+        tls_server_context_factory=tls_server_context_factory,
+        config=config,
+        version_string="Synapse/" + get_version_string(synapse),
+        database_engine=database_engine,
+    )
+
+    ps.setup()
+    ps.start_listening(config.worker_listeners)
+
+    def run():
+        with LoggingContext("run"):
+            logger.info("Running")
+            change_resource_limit(config.soft_file_limit)
+            if config.gc_thresholds:
+                gc.set_threshold(*config.gc_thresholds)
+            reactor.run()
+
+    def start():
+        ps.replicate()
+        ps.get_datastore().start_profiling()
+        ps.get_state_handler().start_caching()
+
+    reactor.callWhenRunning(start)
+
+    if config.worker_daemonize:
+        daemon = Daemonize(
+            app="synapse-federation-sender",
+            pid=config.worker_pid_file,
+            action=run,
+            auto_close_fds=False,
+            verbose=True,
+            logger=logger,
+        )
+        daemon.start()
+    else:
+        run()
+
+
+class FederationSenderHandler(object):
+    """Processes the replication stream and forwards the appropriate entries
+    to the federation sender.
+    """
+    def __init__(self, hs):
+        self.store = hs.get_datastore()
+        self.federation_sender = hs.get_federation_sender()
+
+        self._room_serials = {}
+        self._room_typing = {}
+
+    def on_start(self):
+        # There may be some events that are persisted but haven't been sent,
+        # so send them now.
+        self.federation_sender.notify_new_events(
+            self.store.get_room_max_stream_ordering()
+        )
+
+    @defer.inlineCallbacks
+    def stream_positions(self):
+        stream_id = yield self.store.get_federation_out_pos("federation")
+        defer.returnValue({
+            "federation": stream_id,
+
+            # Ack stuff we've "processed", this should only be called from
+            # one process.
+            "federation_ack": stream_id,
+        })
+
+    @defer.inlineCallbacks
+    def process_replication(self, result):
+        # The federation stream contains things that we want to send out, e.g.
+        # presence, typing, etc.
+        fed_stream = result.get("federation")
+        if fed_stream:
+            latest_id = int(fed_stream["position"])
+
+            # The federation stream containis a bunch of different types of
+            # rows that need to be handled differently. We parse the rows, put
+            # them into the appropriate collection and then send them off.
+            presence_to_send = {}
+            keyed_edus = {}
+            edus = {}
+            failures = {}
+            device_destinations = set()
+
+            # Parse the rows in the stream
+            for row in fed_stream["rows"]:
+                position, typ, content_js = row
+                content = json.loads(content_js)
+
+                if typ == send_queue.PRESENCE_TYPE:
+                    destination = content["destination"]
+                    state = UserPresenceState.from_dict(content["state"])
+
+                    presence_to_send.setdefault(destination, []).append(state)
+                elif typ == send_queue.KEYED_EDU_TYPE:
+                    key = content["key"]
+                    edu = Edu(**content["edu"])
+
+                    keyed_edus.setdefault(
+                        edu.destination, {}
+                    )[(edu.destination, tuple(key))] = edu
+                elif typ == send_queue.EDU_TYPE:
+                    edu = Edu(**content)
+
+                    edus.setdefault(edu.destination, []).append(edu)
+                elif typ == send_queue.FAILURE_TYPE:
+                    destination = content["destination"]
+                    failure = content["failure"]
+
+                    failures.setdefault(destination, []).append(failure)
+                elif typ == send_queue.DEVICE_MESSAGE_TYPE:
+                    device_destinations.add(content["destination"])
+                else:
+                    raise Exception("Unrecognised federation type: %r", typ)
+
+            # We've finished collecting, send everything off
+            for destination, states in presence_to_send.items():
+                self.federation_sender.send_presence(destination, states)
+
+            for destination, edu_map in keyed_edus.items():
+                for key, edu in edu_map.items():
+                    self.federation_sender.send_edu(
+                        edu.destination, edu.edu_type, edu.content, key=key,
+                    )
+
+            for destination, edu_list in edus.items():
+                for edu in edu_list:
+                    self.federation_sender.send_edu(
+                        edu.destination, edu.edu_type, edu.content, key=None,
+                    )
+
+            for destination, failure_list in failures.items():
+                for failure in failure_list:
+                    self.federation_sender.send_failure(destination, failure)
+
+            for destination in device_destinations:
+                self.federation_sender.send_device_messages(destination)
+
+            # Record where we are in the stream.
+            yield self.store.update_federation_out_pos(
+                "federation", latest_id
+            )
+
+        # We also need to poke the federation sender when new events happen
+        event_stream = result.get("events")
+        if event_stream:
+            latest_pos = event_stream["position"]
+            self.federation_sender.notify_new_events(latest_pos)
+
+
+if __name__ == '__main__':
+    with LoggingContext("main"):
+        start(sys.argv[1:])

synapse/app/homeserver.py

@@ -51,6 +51,7 @@ from synapse.api.urls import (
 from synapse.config.homeserver import HomeServerConfig
 from synapse.crypto import context_factory
 from synapse.util.logcontext import LoggingContext
+from synapse.metrics import register_memory_metrics, get_metrics_for
 from synapse.metrics.resource import MetricsResource, METRICS_PREFIX
 from synapse.replication.resource import ReplicationResource, REPLICATION_PREFIX
 from synapse.federation.transport.server import TransportLayerServer
@@ -106,7 +107,7 @@ def build_resource_for_web_client(hs):
 class SynapseHomeServer(HomeServer):
     def _listener_http(self, config, listener_config):
         port = listener_config["port"]
-        bind_address = listener_config.get("bind_address", "")
+        bind_addresses = listener_config["bind_addresses"]
         tls = listener_config.get("tls", False)
         site_tag = listener_config.get("tag", port)
 
@@ -172,29 +173,32 @@ class SynapseHomeServer(HomeServer):
             root_resource = Resource()
 
         root_resource = create_resource_tree(resources, root_resource)
+
         if tls:
-            reactor.listenSSL(
-                port,
-                SynapseSite(
-                    "synapse.access.https.%s" % (site_tag,),
-                    site_tag,
-                    listener_config,
-                    root_resource,
-                ),
-                self.tls_server_context_factory,
-                interface=bind_address
-            )
+            for address in bind_addresses:
+                reactor.listenSSL(
+                    port,
+                    SynapseSite(
+                        "synapse.access.https.%s" % (site_tag,),
+                        site_tag,
+                        listener_config,
+                        root_resource,
+                    ),
+                    self.tls_server_context_factory,
+                    interface=address
+                )
         else:
-            reactor.listenTCP(
-                port,
-                SynapseSite(
-                    "synapse.access.http.%s" % (site_tag,),
-                    site_tag,
-                    listener_config,
-                    root_resource,
-                ),
-                interface=bind_address
-            )
+            for address in bind_addresses:
+                reactor.listenTCP(
+                    port,
+                    SynapseSite(
+                        "synapse.access.http.%s" % (site_tag,),
+                        site_tag,
+                        listener_config,
+                        root_resource,
+                    ),
+                    interface=address
+                )
         logger.info("Synapse now listening on port %d", port)
 
     def start_listening(self):
@@ -204,15 +208,18 @@ class SynapseHomeServer(HomeServer):
             if listener["type"] == "http":
                 self._listener_http(config, listener)
             elif listener["type"] == "manhole":
-                reactor.listenTCP(
-                    listener["port"],
-                    manhole(
-                        username="matrix",
-                        password="rabbithole",
-                        globals={"hs": self},
-                    ),
-                    interface=listener.get("bind_address", '127.0.0.1')
-                )
+                bind_addresses = listener["bind_addresses"]
+
+                for address in bind_addresses:
+                    reactor.listenTCP(
+                        listener["port"],
+                        manhole(
+                            username="matrix",
+                            password="rabbithole",
+                            globals={"hs": self},
+                        ),
+                        interface=address
+                    )
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
@@ -284,7 +291,7 @@ def setup(config_options):
     # check any extra requirements we have now we have a config
     check_requirements(config)
 
-    version_string = get_version_string("Synapse", synapse)
+    version_string = "Synapse/" + get_version_string(synapse)
 
     logger.info("Server hostname: %s", config.server_name)
     logger.info("Server version: %s", version_string)
@@ -335,6 +342,8 @@ def setup(config_options):
         hs.get_datastore().start_doing_background_updates()
         hs.get_replication_layer().start_get_pdu_cache()
 
+        register_memory_metrics(hs)
+
     reactor.callWhenRunning(start)
 
     return hs
@@ -382,6 +391,8 @@ def run(hs):
 
     start_time = hs.get_clock().time()
 
+    stats = {}
+
     @defer.inlineCallbacks
     def phone_stats_home():
         logger.info("Gathering stats for reporting")
@@ -390,7 +401,10 @@ def run(hs):
         if uptime < 0:
             uptime = 0
 
-        stats = {}
+        # If the stats directory is empty then this is the first time we've
+        # reported stats.
+        first_time = not stats
+
         stats["homeserver"] = hs.config.server_name
         stats["timestamp"] = now
         stats["uptime_seconds"] = uptime
@@ -403,6 +417,25 @@ def run(hs):
         daily_messages = yield hs.get_datastore().count_daily_messages()
         if daily_messages is not None:
             stats["daily_messages"] = daily_messages
+        else:
+            stats.pop("daily_messages", None)
+
+        if first_time:
+            # Add callbacks to report the synapse stats as metrics whenever
+            # prometheus requests them, typically every 30s.
+            # As some of the stats are expensive to calculate we only update
+            # them when synapse phones home to matrix.org every 24 hours.
+            metrics = get_metrics_for("synapse.usage")
+            metrics.add_callback("timestamp", lambda: stats["timestamp"])
+            metrics.add_callback("uptime_seconds", lambda: stats["uptime_seconds"])
+            metrics.add_callback("total_users", lambda: stats["total_users"])
+            metrics.add_callback("total_room_count", lambda: stats["total_room_count"])
+            metrics.add_callback(
+                "daily_active_users", lambda: stats["daily_active_users"]
+            )
+            metrics.add_callback(
+                "daily_messages", lambda: stats.get("daily_messages", 0)
+            )
 
         logger.info("Reporting stats to matrix.org: %s" % (stats,))
         try:

synapse/app/media_repository.py

@@ -0,0 +1,223 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+# Copyright 2016 OpenMarket Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import synapse
+
+from synapse.config._base import ConfigError
+from synapse.config.homeserver import HomeServerConfig
+from synapse.config.logger import setup_logging
+from synapse.http.site import SynapseSite
+from synapse.metrics.resource import MetricsResource, METRICS_PREFIX
+from synapse.replication.slave.storage._base import BaseSlavedStore
+from synapse.replication.slave.storage.appservice import SlavedApplicationServiceStore
+from synapse.replication.slave.storage.registration import SlavedRegistrationStore
+from synapse.rest.media.v0.content_repository import ContentRepoResource
+from synapse.rest.media.v1.media_repository import MediaRepositoryResource
+from synapse.server import HomeServer
+from synapse.storage.client_ips import ClientIpStore
+from synapse.storage.engines import create_engine
+from synapse.storage.media_repository import MediaRepositoryStore
+from synapse.util.async import sleep
+from synapse.util.httpresourcetree import create_resource_tree
+from synapse.util.logcontext import LoggingContext
+from synapse.util.manhole import manhole
+from synapse.util.rlimit import change_resource_limit
+from synapse.util.versionstring import get_version_string
+from synapse.api.urls import (
+    CONTENT_REPO_PREFIX, LEGACY_MEDIA_PREFIX, MEDIA_PREFIX
+)
+from synapse.crypto import context_factory
+
+from synapse import events
+
+
+from twisted.internet import reactor, defer
+from twisted.web.resource import Resource
+
+from daemonize import Daemonize
+
+import sys
+import logging
+import gc
+
+logger = logging.getLogger("synapse.app.media_repository")
+
+
+class MediaRepositorySlavedStore(
+    SlavedApplicationServiceStore,
+    SlavedRegistrationStore,
+    BaseSlavedStore,
+    MediaRepositoryStore,
+    ClientIpStore,
+):
+    pass
+
+
+class MediaRepositoryServer(HomeServer):
+    def get_db_conn(self, run_new_connection=True):
+        # Any param beginning with cp_ is a parameter for adbapi, and should
+        # not be passed to the database engine.
+        db_params = {
+            k: v for k, v in self.db_config.get("args", {}).items()
+            if not k.startswith("cp_")
+        }
+        db_conn = self.database_engine.module.connect(**db_params)
+
+        if run_new_connection:
+            self.database_engine.on_new_connection(db_conn)
+        return db_conn
+
+    def setup(self):
+        logger.info("Setting up.")
+        self.datastore = MediaRepositorySlavedStore(self.get_db_conn(), self)
+        logger.info("Finished setting up.")
+
+    def _listen_http(self, listener_config):
+        port = listener_config["port"]
+        bind_addresses = listener_config["bind_addresses"]
+        site_tag = listener_config.get("tag", port)
+        resources = {}
+        for res in listener_config["resources"]:
+            for name in res["names"]:
+                if name == "metrics":
+                    resources[METRICS_PREFIX] = MetricsResource(self)
+                elif name == "media":
+                    media_repo = MediaRepositoryResource(self)
+                    resources.update({
+                        MEDIA_PREFIX: media_repo,
+                        LEGACY_MEDIA_PREFIX: media_repo,
+                        CONTENT_REPO_PREFIX: ContentRepoResource(
+                            self, self.config.uploads_path
+                        ),
+                    })
+
+        root_resource = create_resource_tree(resources, Resource())
+
+        for address in bind_addresses:
+            reactor.listenTCP(
+                port,
+                SynapseSite(
+                    "synapse.access.http.%s" % (site_tag,),
+                    site_tag,
+                    listener_config,
+                    root_resource,
+                ),
+                interface=address
+            )
+
+        logger.info("Synapse media repository now listening on port %d", port)
+
+    def start_listening(self, listeners):
+        for listener in listeners:
+            if listener["type"] == "http":
+                self._listen_http(listener)
+            elif listener["type"] == "manhole":
+                bind_addresses = listener["bind_addresses"]
+
+                for address in bind_addresses:
+                    reactor.listenTCP(
+                        listener["port"],
+                        manhole(
+                            username="matrix",
+                            password="rabbithole",
+                            globals={"hs": self},
+                        ),
+                        interface=address
+                    )
+            else:
+                logger.warn("Unrecognized listener type: %s", listener["type"])
+
+    @defer.inlineCallbacks
+    def replicate(self):
+        http_client = self.get_simple_http_client()
+        store = self.get_datastore()
+        replication_url = self.config.worker_replication_url
+
+        while True:
+            try:
+                args = store.stream_positions()
+                args["timeout"] = 30000
+                result = yield http_client.get_json(replication_url, args=args)
+                yield store.process_replication(result)
+            except:
+                logger.exception("Error replicating from %r", replication_url)
+                yield sleep(5)
+
+
+def start(config_options):
+    try:
+        config = HomeServerConfig.load_config(
+            "Synapse media repository", config_options
+        )
+    except ConfigError as e:
+        sys.stderr.write("\n" + e.message + "\n")
+        sys.exit(1)
+
+    assert config.worker_app == "synapse.app.media_repository"
+
+    setup_logging(config.worker_log_config, config.worker_log_file)
+
+    events.USE_FROZEN_DICTS = config.use_frozen_dicts
+
+    database_engine = create_engine(config.database_config)
+
+    tls_server_context_factory = context_factory.ServerContextFactory(config)
+
+    ss = MediaRepositoryServer(
+        config.server_name,
+        db_config=config.database_config,
+        tls_server_context_factory=tls_server_context_factory,
+        config=config,
+        version_string="Synapse/" + get_version_string(synapse),
+        database_engine=database_engine,
+    )
+
+    ss.setup()
+    ss.get_handlers()
+    ss.start_listening(config.worker_listeners)
+
+    def run():
+        with LoggingContext("run"):
+            logger.info("Running")
+            change_resource_limit(config.soft_file_limit)
+            if config.gc_thresholds:
+                gc.set_threshold(*config.gc_thresholds)
+            reactor.run()
+
+    def start():
+        ss.get_state_handler().start_caching()
+        ss.get_datastore().start_profiling()
+        ss.replicate()
+
+    reactor.callWhenRunning(start)
+
+    if config.worker_daemonize:
+        daemon = Daemonize(
+            app="synapse-media-repository",
+            pid=config.worker_pid_file,
+            action=run,
+            auto_close_fds=False,
+            verbose=True,
+            logger=logger,
+        )
+        daemon.start()
+    else:
+        run()
+
+
+if __name__ == '__main__':
+    with LoggingContext("main"):
+        start(sys.argv[1:])

synapse/app/pusher.py

@@ -36,6 +36,8 @@ from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.versionstring import get_version_string
 
+from synapse import events
+
 from twisted.internet import reactor, defer
 from twisted.web.resource import Resource
 
@@ -80,11 +82,6 @@ class PusherSlaveStore(
         DataStore.get_profile_displayname.__func__
     )
 
-    # XXX: This is a bit broken because we don't persist forgotten rooms
-    # in a way that they can be streamed. This means that we don't have a
-    # way to invalidate the forgotten rooms cache correctly.
-    # For now we expire the cache every 10 minutes.
-    BROKEN_CACHE_EXPIRY_MS = 60 * 60 * 1000
     who_forgot_in_room = (
         RoomMemberStore.__dict__["who_forgot_in_room"]
     )
@@ -124,7 +121,7 @@ class PusherServer(HomeServer):
 
     def _listen_http(self, listener_config):
         port = listener_config["port"]
-        bind_address = listener_config.get("bind_address", "")
+        bind_addresses = listener_config["bind_addresses"]
         site_tag = listener_config.get("tag", port)
         resources = {}
         for res in listener_config["resources"]:
@@ -133,16 +130,19 @@ class PusherServer(HomeServer):
                     resources[METRICS_PREFIX] = MetricsResource(self)
 
         root_resource = create_resource_tree(resources, Resource())
-        reactor.listenTCP(
-            port,
-            SynapseSite(
-                "synapse.access.http.%s" % (site_tag,),
-                site_tag,
-                listener_config,
-                root_resource,
-            ),
-            interface=bind_address
-        )
+
+        for address in bind_addresses:
+            reactor.listenTCP(
+                port,
+                SynapseSite(
+                    "synapse.access.http.%s" % (site_tag,),
+                    site_tag,
+                    listener_config,
+                    root_resource,
+                ),
+                interface=address
+            )
+
         logger.info("Synapse pusher now listening on port %d", port)
 
     def start_listening(self, listeners):
@@ -150,15 +150,18 @@ class PusherServer(HomeServer):
             if listener["type"] == "http":
                 self._listen_http(listener)
             elif listener["type"] == "manhole":
-                reactor.listenTCP(
-                    listener["port"],
-                    manhole(
-                        username="matrix",
-                        password="rabbithole",
-                        globals={"hs": self},
-                    ),
-                    interface=listener.get("bind_address", '127.0.0.1')
-                )
+                bind_addresses = listener["bind_addresses"]
+
+                for address in bind_addresses:
+                    reactor.listenTCP(
+                        listener["port"],
+                        manhole(
+                            username="matrix",
+                            password="rabbithole",
+                            globals={"hs": self},
+                        ),
+                        interface=address
+                    )
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
@@ -168,7 +171,6 @@ class PusherServer(HomeServer):
         store = self.get_datastore()
         replication_url = self.config.worker_replication_url
         pusher_pool = self.get_pusherpool()
-        clock = self.get_clock()
 
         def stop_pusher(user_id, app_id, pushkey):
             key = "%s:%s" % (app_id, pushkey)
@@ -203,7 +205,7 @@ class PusherServer(HomeServer):
                     yield start_pusher(user_id, app_id, pushkey)
 
             stream = results.get("events")
-            if stream:
+            if stream and stream["rows"]:
                 min_stream_id = stream["rows"][0][0]
                 max_stream_id = stream["position"]
                 preserve_fn(pusher_pool.on_new_notifications)(
@@ -211,7 +213,7 @@ class PusherServer(HomeServer):
                 )
 
             stream = results.get("receipts")
-            if stream:
+            if stream and stream["rows"]:
                 rows = stream["rows"]
                 affected_room_ids = set(row[1] for row in rows)
                 min_stream_id = rows[0][0]
@@ -220,21 +222,11 @@ class PusherServer(HomeServer):
                     min_stream_id, max_stream_id, affected_room_ids
                 )
 
-        def expire_broken_caches():
-            store.who_forgot_in_room.invalidate_all()
-
-        next_expire_broken_caches_ms = 0
         while True:
             try:
                 args = store.stream_positions()
                 args["timeout"] = 30000
                 result = yield http_client.get_json(replication_url, args=args)
-                now_ms = clock.time_msec()
-                if now_ms > next_expire_broken_caches_ms:
-                    expire_broken_caches()
-                    next_expire_broken_caches_ms = (
-                        now_ms + store.BROKEN_CACHE_EXPIRY_MS
-                    )
                 yield store.process_replication(result)
                 poke_pushers(result)
             except:
@@ -255,6 +247,8 @@ def start(config_options):
 
     setup_logging(config.worker_log_config, config.worker_log_file)
 
+    events.USE_FROZEN_DICTS = config.use_frozen_dicts
+
     if config.start_pushers:
         sys.stderr.write(
             "\nThe pushers must be disabled in the main synapse process"
@@ -273,7 +267,7 @@ def start(config_options):
         config.server_name,
         db_config=config.database_config,
         config=config,
-        version_string=get_version_string("Synapse", synapse),
+        version_string="Synapse/" + get_version_string(synapse),
         database_engine=database_engine,
     )
 
@@ -292,6 +286,7 @@ def start(config_options):
         ps.replicate()
         ps.get_pusherpool().start()
         ps.get_datastore().start_profiling()
+        ps.get_state_handler().start_caching()
 
     reactor.callWhenRunning(start)
 

synapse/app/synchrotron.py

@@ -26,6 +26,9 @@ from synapse.http.site import SynapseSite
 from synapse.http.server import JsonResource
 from synapse.metrics.resource import MetricsResource, METRICS_PREFIX
 from synapse.rest.client.v2_alpha import sync
+from synapse.rest.client.v1 import events
+from synapse.rest.client.v1.room import RoomInitialSyncRestServlet
+from synapse.rest.client.v1.initial_sync import InitialSyncRestServlet
 from synapse.replication.slave.storage._base import BaseSlavedStore
 from synapse.replication.slave.storage.events import SlavedEventStore
 from synapse.replication.slave.storage.receipts import SlavedReceiptsStore
@@ -35,6 +38,9 @@ from synapse.replication.slave.storage.registration import SlavedRegistrationSto
 from synapse.replication.slave.storage.filtering import SlavedFilteringStore
 from synapse.replication.slave.storage.push_rule import SlavedPushRuleStore
 from synapse.replication.slave.storage.presence import SlavedPresenceStore
+from synapse.replication.slave.storage.deviceinbox import SlavedDeviceInboxStore
+from synapse.replication.slave.storage.devices import SlavedDeviceStore
+from synapse.replication.slave.storage.room import RoomStore
 from synapse.server import HomeServer
 from synapse.storage.client_ips import ClientIpStore
 from synapse.storage.engines import create_engine
@@ -71,14 +77,12 @@ class SynchrotronSlavedStore(
     SlavedRegistrationStore,
     SlavedFilteringStore,
     SlavedPresenceStore,
+    SlavedDeviceInboxStore,
+    SlavedDeviceStore,
+    RoomStore,
     BaseSlavedStore,
     ClientIpStore,  # After BaseSlavedStore because the constructor is different
 ):
-    # XXX: This is a bit broken because we don't persist forgotten rooms
-    # in a way that they can be streamed. This means that we don't have a
-    # way to invalidate the forgotten rooms cache correctly.
-    # For now we expire the cache every 10 minutes.
-    BROKEN_CACHE_EXPIRY_MS = 60 * 60 * 1000
     who_forgot_in_room = (
         RoomMemberStore.__dict__["who_forgot_in_room"]
     )
@@ -89,17 +93,23 @@ class SynchrotronSlavedStore(
     get_presence_list_accepted = PresenceStore.__dict__[
         "get_presence_list_accepted"
     ]
+    get_presence_list_observers_accepted = PresenceStore.__dict__[
+        "get_presence_list_observers_accepted"
+    ]
+
 
 UPDATE_SYNCING_USERS_MS = 10 * 1000
 
 
 class SynchrotronPresence(object):
     def __init__(self, hs):
+        self.is_mine_id = hs.is_mine_id
         self.http_client = hs.get_simple_http_client()
         self.store = hs.get_datastore()
         self.user_to_num_current_syncs = {}
         self.syncing_users_url = hs.config.worker_replication_url + "/syncing_users"
         self.clock = hs.get_clock()
+        self.notifier = hs.get_notifier()
 
         active_presence = self.store.take_presence_startup_info()
         self.user_to_current_state = {
@@ -119,11 +129,13 @@ class SynchrotronPresence(object):
 
         reactor.addSystemEventTrigger("before", "shutdown", self._on_shutdown)
 
-    def set_state(self, user, state):
+    def set_state(self, user, state, ignore_status_msg=False):
         # TODO Hows this supposed to work?
         pass
 
     get_states = PresenceHandler.get_states.__func__
+    get_state = PresenceHandler.get_state.__func__
+    _get_interested_parties = PresenceHandler._get_interested_parties.__func__
     current_state_for_users = PresenceHandler.current_state_for_users.__func__
 
     @defer.inlineCallbacks
@@ -194,19 +206,39 @@ class SynchrotronPresence(object):
             self._need_to_send_sync = False
             yield self._send_syncing_users_now()
 
+    @defer.inlineCallbacks
+    def notify_from_replication(self, states, stream_id):
+        parties = yield self._get_interested_parties(
+            states, calculate_remote_hosts=False
+        )
+        room_ids_to_states, users_to_states, _ = parties
+
+        self.notifier.on_new_event(
+            "presence_key", stream_id, rooms=room_ids_to_states.keys(),
+            users=users_to_states.keys()
+        )
+
+    @defer.inlineCallbacks
     def process_replication(self, result):
         stream = result.get("presence", {"rows": []})
+        states = []
         for row in stream["rows"]:
             (
                 position, user_id, state, last_active_ts,
                 last_federation_update_ts, last_user_sync_ts, status_msg,
                 currently_active
             ) = row
-            self.user_to_current_state[user_id] = UserPresenceState(
+            state = UserPresenceState(
                 user_id, state, last_active_ts,
                 last_federation_update_ts, last_user_sync_ts, status_msg,
                 currently_active
             )
+            self.user_to_current_state[user_id] = state
+            states.append(state)
+
+        if states and "position" in stream:
+            stream_id = int(stream["position"])
+            yield self.notify_from_replication(states, stream_id)
 
 
 class SynchrotronTyping(object):
@@ -216,6 +248,9 @@ class SynchrotronTyping(object):
         self._room_typing = {}
 
     def stream_positions(self):
+        # We must update this typing token from the response of the previous
+        # sync. In particular, the stream id may "reset" back to zero/a low
+        # value which we *must* use for the next replication request.
         return {"typing": self._latest_room_serial}
 
     def process_replication(self, result):
@@ -256,7 +291,7 @@ class SynchrotronServer(HomeServer):
 
     def _listen_http(self, listener_config):
         port = listener_config["port"]
-        bind_address = listener_config.get("bind_address", "")
+        bind_addresses = listener_config["bind_addresses"]
         site_tag = listener_config.get("tag", port)
         resources = {}
         for res in listener_config["resources"]:
@@ -266,23 +301,30 @@ class SynchrotronServer(HomeServer):
                 elif name == "client":
                     resource = JsonResource(self, canonical_json=False)
                     sync.register_servlets(self, resource)
+                    events.register_servlets(self, resource)
+                    InitialSyncRestServlet(self).register(resource)
+                    RoomInitialSyncRestServlet(self).register(resource)
                     resources.update({
                         "/_matrix/client/r0": resource,
                         "/_matrix/client/unstable": resource,
                         "/_matrix/client/v2_alpha": resource,
+                        "/_matrix/client/api/v1": resource,
                     })
 
         root_resource = create_resource_tree(resources, Resource())
-        reactor.listenTCP(
-            port,
-            SynapseSite(
-                "synapse.access.http.%s" % (site_tag,),
-                site_tag,
-                listener_config,
-                root_resource,
-            ),
-            interface=bind_address
-        )
+
+        for address in bind_addresses:
+            reactor.listenTCP(
+                port,
+                SynapseSite(
+                    "synapse.access.http.%s" % (site_tag,),
+                    site_tag,
+                    listener_config,
+                    root_resource,
+                ),
+                interface=address
+            )
+
         logger.info("Synapse synchrotron now listening on port %d", port)
 
     def start_listening(self, listeners):
@@ -290,15 +332,18 @@ class SynchrotronServer(HomeServer):
             if listener["type"] == "http":
                 self._listen_http(listener)
             elif listener["type"] == "manhole":
-                reactor.listenTCP(
-                    listener["port"],
-                    manhole(
-                        username="matrix",
-                        password="rabbithole",
-                        globals={"hs": self},
-                    ),
-                    interface=listener.get("bind_address", '127.0.0.1')
-                )
+                bind_addresses = listener["bind_addresses"]
+
+                for address in bind_addresses:
+                    reactor.listenTCP(
+                        listener["port"],
+                        manhole(
+                            username="matrix",
+                            password="rabbithole",
+                            globals={"hs": self},
+                        ),
+                        interface=address
+                    )
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
@@ -307,15 +352,10 @@ class SynchrotronServer(HomeServer):
         http_client = self.get_simple_http_client()
         store = self.get_datastore()
         replication_url = self.config.worker_replication_url
-        clock = self.get_clock()
         notifier = self.get_notifier()
         presence_handler = self.get_presence_handler()
         typing_handler = self.get_typing_handler()
 
-        def expire_broken_caches():
-            store.who_forgot_in_room.invalidate_all()
-            store.get_presence_list_accepted.invalidate_all()
-
         def notify_from_stream(
             result, stream_name, stream_key, room=None, user=None
         ):
@@ -342,6 +382,27 @@ class SynchrotronServer(HomeServer):
                         stream_key, position, users=users, rooms=rooms
                     )
 
+        @defer.inlineCallbacks
+        def notify_device_list_update(result):
+            stream = result.get("device_lists")
+            if not stream:
+                return
+
+            position_index = stream["field_names"].index("position")
+            user_index = stream["field_names"].index("user_id")
+
+            for row in stream["rows"]:
+                position = row[position_index]
+                user_id = row[user_index]
+
+                rooms = yield store.get_rooms_for_user(user_id)
+                room_ids = [r.room_id for r in rooms]
+
+                notifier.on_new_event(
+                    "device_list_key", position, rooms=room_ids,
+                )
+
+        @defer.inlineCallbacks
         def notify(result):
             stream = result.get("events")
             if stream:
@@ -376,24 +437,21 @@ class SynchrotronServer(HomeServer):
             notify_from_stream(
                 result, "typing", "typing_key", room="room_id"
             )
+            notify_from_stream(
+                result, "to_device", "to_device_key", user="user_id"
+            )
+            yield notify_device_list_update(result)
 
-        next_expire_broken_caches_ms = 0
         while True:
             try:
                 args = store.stream_positions()
                 args.update(typing_handler.stream_positions())
                 args["timeout"] = 30000
                 result = yield http_client.get_json(replication_url, args=args)
-                now_ms = clock.time_msec()
-                if now_ms > next_expire_broken_caches_ms:
-                    expire_broken_caches()
-                    next_expire_broken_caches_ms = (
-                        now_ms + store.BROKEN_CACHE_EXPIRY_MS
-                    )
                 yield store.process_replication(result)
                 typing_handler.process_replication(result)
-                presence_handler.process_replication(result)
-                notify(result)
+                yield presence_handler.process_replication(result)
+                yield notify(result)
             except:
                 logger.exception("Error replicating from %r", replication_url)
                 yield sleep(5)
@@ -418,13 +476,15 @@ def start(config_options):
 
     setup_logging(config.worker_log_config, config.worker_log_file)
 
+    synapse.events.USE_FROZEN_DICTS = config.use_frozen_dicts
+
     database_engine = create_engine(config.database_config)
 
     ss = SynchrotronServer(
         config.server_name,
         db_config=config.database_config,
         config=config,
-        version_string=get_version_string("Synapse", synapse),
+        version_string="Synapse/" + get_version_string(synapse),
         database_engine=database_engine,
         application_service_handler=SynchrotronApplicationService(),
     )
@@ -443,6 +503,7 @@ def start(config_options):
     def start():
         ss.get_datastore().start_profiling()
         ss.replicate()
+        ss.get_state_handler().start_caching()
 
     reactor.callWhenRunning(start)
 

synapse/app/synctl.py

@@ -24,7 +24,7 @@ import subprocess
 import sys
 import yaml
 
-SYNAPSE = ["python", "-B", "-m", "synapse.app.homeserver"]
+SYNAPSE = [sys.executable, "-B", "-m", "synapse.app.homeserver"]
 
 GREEN = "\x1b[1;32m"
 RED = "\x1b[1;31m"

synapse/appservice/__init__.py

@@ -14,6 +14,8 @@
 # limitations under the License.
 from synapse.api.constants import EventTypes
 
+from twisted.internet import defer
+
 import logging
 import re
 
@@ -79,7 +81,7 @@ class ApplicationService(object):
     NS_LIST = [NS_USERS, NS_ALIASES, NS_ROOMS]
 
     def __init__(self, token, url=None, namespaces=None, hs_token=None,
-                 sender=None, id=None):
+                 sender=None, id=None, protocols=None, rate_limited=True):
         self.token = token
         self.url = url
         self.hs_token = hs_token
@@ -87,6 +89,17 @@ class ApplicationService(object):
         self.namespaces = self._check_namespaces(namespaces)
         self.id = id
 
+        if "|" in self.id:
+            raise Exception("application service ID cannot contain '|' character")
+
+        # .protocols is a publicly visible field
+        if protocols:
+            self.protocols = set(protocols)
+        else:
+            self.protocols = set()
+
+        self.rate_limited = rate_limited
+
     def _check_namespaces(self, namespaces):
         # Sanity check that it is of the form:
         # {
@@ -138,65 +151,66 @@ class ApplicationService(object):
             return regex_obj["exclusive"]
         return False
 
-    def _matches_user(self, event, member_list):
-        if (hasattr(event, "sender") and
-                self.is_interested_in_user(event.sender)):
-            return True
+    @defer.inlineCallbacks
+    def _matches_user(self, event, store):
+        if not event:
+            defer.returnValue(False)
+
+        if self.is_interested_in_user(event.sender):
+            defer.returnValue(True)
         # also check m.room.member state key
-        if (hasattr(event, "type") and event.type == EventTypes.Member
-                and hasattr(event, "state_key")
-                and self.is_interested_in_user(event.state_key)):
-            return True
+        if (event.type == EventTypes.Member and
+                self.is_interested_in_user(event.state_key)):
+            defer.returnValue(True)
+
+        if not store:
+            defer.returnValue(False)
+
+        member_list = yield store.get_users_in_room(event.room_id)
+
         # check joined member events
         for user_id in member_list:
             if self.is_interested_in_user(user_id):
-                return True
-        return False
+                defer.returnValue(True)
+        defer.returnValue(False)
 
     def _matches_room_id(self, event):
         if hasattr(event, "room_id"):
             return self.is_interested_in_room(event.room_id)
         return False
 
-    def _matches_aliases(self, event, alias_list):
+    @defer.inlineCallbacks
+    def _matches_aliases(self, event, store):
+        if not store or not event:
+            defer.returnValue(False)
+
+        alias_list = yield store.get_aliases_for_room(event.room_id)
         for alias in alias_list:
             if self.is_interested_in_alias(alias):
-                return True
-        return False
+                defer.returnValue(True)
+        defer.returnValue(False)
 
-    def is_interested(self, event, restrict_to=None, aliases_for_event=None,
-                      member_list=None):
+    @defer.inlineCallbacks
+    def is_interested(self, event, store=None):
         """Check if this service is interested in this event.
 
         Args:
             event(Event): The event to check.
-            restrict_to(str): The namespace to restrict regex tests to.
-            aliases_for_event(list): A list of all the known room aliases for
-            this event.
-            member_list(list): A list of all joined user_ids in this room.
+            store(DataStore)
         Returns:
             bool: True if this service would like to know about this event.
         """
-        if aliases_for_event is None:
-            aliases_for_event = []
-        if member_list is None:
-            member_list = []
+        # Do cheap checks first
+        if self._matches_room_id(event):
+            defer.returnValue(True)
 
-        if restrict_to and restrict_to not in ApplicationService.NS_LIST:
-            # this is a programming error, so fail early and raise a general
-            # exception
-            raise Exception("Unexpected restrict_to value: %s". restrict_to)
+        if (yield self._matches_aliases(event, store)):
+            defer.returnValue(True)
 
-        if not restrict_to:
-            return (self._matches_user(event, member_list)
-                    or self._matches_aliases(event, aliases_for_event)
-                    or self._matches_room_id(event))
-        elif restrict_to == ApplicationService.NS_ALIASES:
-            return self._matches_aliases(event, aliases_for_event)
-        elif restrict_to == ApplicationService.NS_ROOMS:
-            return self._matches_room_id(event)
-        elif restrict_to == ApplicationService.NS_USERS:
-            return self._matches_user(event, member_list)
+        if (yield self._matches_user(event, store)):
+            defer.returnValue(True)
+
+        defer.returnValue(False)
 
     def is_interested_in_user(self, user_id):
         return (
@@ -216,11 +230,17 @@ class ApplicationService(object):
             or user_id == self.sender
         )
 
+    def is_interested_in_protocol(self, protocol):
+        return protocol in self.protocols
+
     def is_exclusive_alias(self, alias):
         return self._is_exclusive(ApplicationService.NS_ALIASES, alias)
 
     def is_exclusive_room(self, room_id):
         return self._is_exclusive(ApplicationService.NS_ROOMS, room_id)
 
+    def is_rate_limited(self):
+        return self.rate_limited
+
     def __str__(self):
         return "ApplicationService: %s" % (self.__dict__,)

synapse/appservice/api.py

@@ -14,9 +14,12 @@
 # limitations under the License.
 from twisted.internet import defer
 
+from synapse.api.constants import ThirdPartyEntityKind
 from synapse.api.errors import CodeMessageException
 from synapse.http.client import SimpleHttpClient
 from synapse.events.utils import serialize_event
+from synapse.util.caches.response_cache import ResponseCache
+from synapse.types import ThirdPartyInstanceID
 
 import logging
 import urllib
@@ -24,6 +27,42 @@ import urllib
 logger = logging.getLogger(__name__)
 
 
+HOUR_IN_MS = 60 * 60 * 1000
+
+
+APP_SERVICE_PREFIX = "/_matrix/app/unstable"
+
+
+def _is_valid_3pe_metadata(info):
+    if "instances" not in info:
+        return False
+    if not isinstance(info["instances"], list):
+        return False
+    return True
+
+
+def _is_valid_3pe_result(r, field):
+    if not isinstance(r, dict):
+        return False
+
+    for k in (field, "protocol"):
+        if k not in r:
+            return False
+        if not isinstance(r[k], str):
+            return False
+
+    if "fields" not in r:
+        return False
+    fields = r["fields"]
+    if not isinstance(fields, dict):
+        return False
+    for k in fields.keys():
+        if not isinstance(fields[k], str):
+            return False
+
+    return True
+
+
 class ApplicationServiceApi(SimpleHttpClient):
     """This class manages HS -> AS communications, including querying and
     pushing.
@@ -33,8 +72,12 @@ class ApplicationServiceApi(SimpleHttpClient):
         super(ApplicationServiceApi, self).__init__(hs)
         self.clock = hs.get_clock()
 
+        self.protocol_meta_cache = ResponseCache(hs, timeout_ms=HOUR_IN_MS)
+
     @defer.inlineCallbacks
     def query_user(self, service, user_id):
+        if service.url is None:
+            defer.returnValue(False)
         uri = service.url + ("/users/%s" % urllib.quote(user_id))
         response = None
         try:
@@ -54,6 +97,8 @@ class ApplicationServiceApi(SimpleHttpClient):
 
     @defer.inlineCallbacks
     def query_alias(self, service, alias):
+        if service.url is None:
+            defer.returnValue(False)
         uri = service.url + ("/rooms/%s" % urllib.quote(alias))
         response = None
         try:
@@ -71,8 +116,91 @@ class ApplicationServiceApi(SimpleHttpClient):
             logger.warning("query_alias to %s threw exception %s", uri, ex)
         defer.returnValue(False)
 
+    @defer.inlineCallbacks
+    def query_3pe(self, service, kind, protocol, fields):
+        if kind == ThirdPartyEntityKind.USER:
+            required_field = "userid"
+        elif kind == ThirdPartyEntityKind.LOCATION:
+            required_field = "alias"
+        else:
+            raise ValueError(
+                "Unrecognised 'kind' argument %r to query_3pe()", kind
+            )
+        if service.url is None:
+            defer.returnValue([])
+
+        uri = "%s%s/thirdparty/%s/%s" % (
+            service.url,
+            APP_SERVICE_PREFIX,
+            kind,
+            urllib.quote(protocol)
+        )
+        try:
+            response = yield self.get_json(uri, fields)
+            if not isinstance(response, list):
+                logger.warning(
+                    "query_3pe to %s returned an invalid response %r",
+                    uri, response
+                )
+                defer.returnValue([])
+
+            ret = []
+            for r in response:
+                if _is_valid_3pe_result(r, field=required_field):
+                    ret.append(r)
+                else:
+                    logger.warning(
+                        "query_3pe to %s returned an invalid result %r",
+                        uri, r
+                    )
+
+            defer.returnValue(ret)
+        except Exception as ex:
+            logger.warning("query_3pe to %s threw exception %s", uri, ex)
+            defer.returnValue([])
+
+    def get_3pe_protocol(self, service, protocol):
+        if service.url is None:
+            defer.returnValue({})
+
+        @defer.inlineCallbacks
+        def _get():
+            uri = "%s%s/thirdparty/protocol/%s" % (
+                service.url,
+                APP_SERVICE_PREFIX,
+                urllib.quote(protocol)
+            )
+            try:
+                info = yield self.get_json(uri, {})
+
+                if not _is_valid_3pe_metadata(info):
+                    logger.warning("query_3pe_protocol to %s did not return a"
+                                   " valid result", uri)
+                    defer.returnValue(None)
+
+                for instance in info.get("instances", []):
+                    network_id = instance.get("network_id", None)
+                    if network_id is not None:
+                        instance["instance_id"] = ThirdPartyInstanceID(
+                            service.id, network_id,
+                        ).to_string()
+
+                defer.returnValue(info)
+            except Exception as ex:
+                logger.warning("query_3pe_protocol to %s threw exception %s",
+                               uri, ex)
+                defer.returnValue(None)
+
+        key = (service.id, protocol)
+        return self.protocol_meta_cache.get(key) or (
+            self.protocol_meta_cache.set(key, _get())
+        )
+
     @defer.inlineCallbacks
     def push_bulk(self, service, events, txn_id=None):
+        if service.url is None:
+            defer.returnValue(True)
+
         events = self._serialize(events)
 
         if txn_id is None:

synapse/appservice/scheduler.py

@@ -48,9 +48,12 @@ UP & quit           +---------- YES                       SUCCESS
 This is all tied together by the AppServiceScheduler which DIs the required
 components.
 """
+from twisted.internet import defer
 
 from synapse.appservice import ApplicationServiceState
-from twisted.internet import defer
+from synapse.util.logcontext import preserve_fn
+from synapse.util.metrics import Measure
+
 import logging
 
 logger = logging.getLogger(__name__)
@@ -73,7 +76,7 @@ class ApplicationServiceScheduler(object):
         self.txn_ctrl = _TransactionController(
             self.clock, self.store, self.as_api, create_recoverer
         )
-        self.queuer = _ServiceQueuer(self.txn_ctrl)
+        self.queuer = _ServiceQueuer(self.txn_ctrl, self.clock)
 
     @defer.inlineCallbacks
     def start(self):
@@ -94,38 +97,36 @@ class _ServiceQueuer(object):
     this schedules any other events in the queue to run.
     """
 
-    def __init__(self, txn_ctrl):
+    def __init__(self, txn_ctrl, clock):
         self.queued_events = {}  # dict of {service_id: [events]}
-        self.pending_requests = {}  # dict of {service_id: Deferred}
+        self.requests_in_flight = set()
         self.txn_ctrl = txn_ctrl
+        self.clock = clock
 
     def enqueue(self, service, event):
         # if this service isn't being sent something
-        if not self.pending_requests.get(service.id):
-            self._send_request(service, [event])
-        else:
-            # add to queue for this service
-            if service.id not in self.queued_events:
-                self.queued_events[service.id] = []
-            self.queued_events[service.id].append(event)
+        self.queued_events.setdefault(service.id, []).append(event)
+        preserve_fn(self._send_request)(service)
 
-    def _send_request(self, service, events):
-        # send request and add callbacks
-        d = self.txn_ctrl.send(service, events)
-        d.addBoth(self._on_request_finish)
-        d.addErrback(self._on_request_fail)
-        self.pending_requests[service.id] = d
+    @defer.inlineCallbacks
+    def _send_request(self, service):
+        if service.id in self.requests_in_flight:
+            return
 
-    def _on_request_finish(self, service):
-        self.pending_requests[service.id] = None
-        # if there are queued events, then send them.
-        if (service.id in self.queued_events
-                and len(self.queued_events[service.id]) > 0):
-            self._send_request(service, self.queued_events[service.id])
-            self.queued_events[service.id] = []
+        self.requests_in_flight.add(service.id)
+        try:
+            while True:
+                events = self.queued_events.pop(service.id, [])
+                if not events:
+                    return
 
-    def _on_request_fail(self, err):
-        logger.error("AS request failed: %s", err)
+                with Measure(self.clock, "servicequeuer.send"):
+                    try:
+                        yield self.txn_ctrl.send(service, events)
+                    except:
+                        logger.exception("AS request failed")
+        finally:
+            self.requests_in_flight.discard(service.id)
 
 
 class _TransactionController(object):
@@ -149,14 +150,12 @@ class _TransactionController(object):
             if service_is_up:
                 sent = yield txn.send(self.as_api)
                 if sent:
-                    txn.complete(self.store)
+                    yield txn.complete(self.store)
                 else:
-                    self._start_recoverer(service)
+                    preserve_fn(self._start_recoverer)(service)
         except Exception as e:
             logger.exception(e)
-            self._start_recoverer(service)
-        # request has finished
-        defer.returnValue(service)
+            preserve_fn(self._start_recoverer)(service)
 
     @defer.inlineCallbacks
     def on_recovered(self, recoverer):

synapse/config/_base.py

@@ -64,11 +64,12 @@ class Config(object):
         if isinstance(value, int) or isinstance(value, long):
             return value
         second = 1000
-        hour = 60 * 60 * second
+        minute = 60 * second
+        hour = 60 * minute
         day = 24 * hour
         week = 7 * day
         year = 365 * day
-        sizes = {"s": second, "h": hour, "d": day, "w": week, "y": year}
+        sizes = {"s": second, "m": minute, "h": hour, "d": day, "w": week, "y": year}
         size = 1
         suffix = value[-1]
         if suffix in sizes:

synapse/config/appservice.py

@@ -28,6 +28,7 @@ class AppServiceConfig(Config):
 
     def read_config(self, config):
         self.app_service_config_files = config.get("app_service_config_files", [])
+        self.notify_appservices = config.get("notify_appservices", True)
 
     def default_config(cls, **kwargs):
         return """\
@@ -85,7 +86,7 @@ def load_appservices(hostname, config_files):
 
 def _load_appservice(hostname, as_info, config_filename):
     required_string_fields = [
-        "id", "url", "as_token", "hs_token", "sender_localpart"
+        "id", "as_token", "hs_token", "sender_localpart"
     ]
     for field in required_string_fields:
         if not isinstance(as_info.get(field), basestring):
@@ -93,6 +94,14 @@ def _load_appservice(hostname, as_info, config_filename):
                 field, config_filename,
             ))
 
+    # 'url' must either be a string or explicitly null, not missing
+    # to avoid accidentally turning off push for ASes.
+    if (not isinstance(as_info.get("url"), basestring) and
+            as_info.get("url", "") is not None):
+        raise KeyError(
+            "Required string field or explicit null: 'url' (%s)" % (config_filename,)
+        )
+
     localpart = as_info["sender_localpart"]
     if urllib.quote(localpart) != localpart:
         raise ValueError(
@@ -101,6 +110,11 @@ def _load_appservice(hostname, as_info, config_filename):
     user = UserID(localpart, hostname)
     user_id = user.to_string()
 
+    # Rate limiting for users of this AS is on by default (excludes sender)
+    rate_limited = True
+    if isinstance(as_info.get("rate_limited"), bool):
+        rate_limited = as_info.get("rate_limited")
+
     # namespace checks
     if not isinstance(as_info.get("namespaces"), dict):
         raise KeyError("Requires 'namespaces' object.")
@@ -122,6 +136,22 @@ def _load_appservice(hostname, as_info, config_filename):
                     raise ValueError(
                         "Missing/bad type 'exclusive' key in %s", regex_obj
                     )
+    # protocols check
+    protocols = as_info.get("protocols")
+    if protocols:
+        # Because strings are lists in python
+        if isinstance(protocols, str) or not isinstance(protocols, list):
+            raise KeyError("Optional 'protocols' must be a list if present.")
+        for p in protocols:
+            if not isinstance(p, str):
+                raise KeyError("Bad value for 'protocols' item")
+
+    if as_info["url"] is None:
+        logger.info(
+            "(%s) Explicitly empty 'url' provided. This application service"
+            " will not receive events or queries.",
+            config_filename,
+        )
     return ApplicationService(
         token=as_info["as_token"],
         url=as_info["url"],
@@ -129,4 +159,6 @@ def _load_appservice(hostname, as_info, config_filename):
         hs_token=as_info["hs_token"],
         sender=user_id,
         id=as_info["id"],
+        protocols=protocols,
+        rate_limited=rate_limited
     )

synapse/config/emailconfig.py

@@ -68,6 +68,9 @@ class EmailConfig(Config):
             self.email_notif_for_new_users = email_config.get(
                 "notif_for_new_users", True
             )
+            self.email_riot_base_url = email_config.get(
+                "riot_base_url", None
+            )
             if "app_name" in email_config:
                 self.email_app_name = email_config["app_name"]
             else:
@@ -85,6 +88,9 @@ class EmailConfig(Config):
     def default_config(self, config_dir_path, server_name, **kwargs):
         return """
         # Enable sending emails for notification events
+        # Defining a custom URL for Riot is only needed if email notifications
+        # should contain links to a self-hosted installation of Riot; when set
+        # the "app_name" setting is ignored.
         #email:
         #   enable_notifs: false
         #   smtp_host: "localhost"
@@ -95,4 +101,5 @@ class EmailConfig(Config):
         #   notif_template_html: notif_mail.html
         #   notif_template_text: notif_mail.txt
         #   notif_for_new_users: True
+        #   riot_base_url: "http://localhost/riot"
         """

synapse/config/homeserver.py

@@ -30,7 +30,7 @@ from .saml2 import SAML2Config
 from .cas import CasConfig
 from .password import PasswordConfig
 from .jwt import JWTConfig
-from .ldap import LDAPConfig
+from .password_auth_providers import PasswordAuthProviderConfig
 from .emailconfig import EmailConfig
 from .workers import WorkerConfig
 
@@ -39,8 +39,8 @@ class HomeServerConfig(TlsConfig, ServerConfig, DatabaseConfig, LoggingConfig,
                        RatelimitConfig, ContentRepositoryConfig, CaptchaConfig,
                        VoipConfig, RegistrationConfig, MetricsConfig, ApiConfig,
                        AppServiceConfig, KeyConfig, SAML2Config, CasConfig,
-                       JWTConfig, LDAPConfig, PasswordConfig, EmailConfig,
-                       WorkerConfig,):
+                       JWTConfig, PasswordConfig, EmailConfig,
+                       WorkerConfig, PasswordAuthProviderConfig,):
     pass
 
 

synapse/config/ldap.py

@@ -1,100 +0,0 @@
-# -*- coding: utf-8 -*-
-# Copyright 2015 Niklas Riekenbrauck
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from ._base import Config, ConfigError
-
-
-MISSING_LDAP3 = (
-    "Missing ldap3 library. This is required for LDAP Authentication."
-)
-
-
-class LDAPMode(object):
-    SIMPLE = "simple",
-    SEARCH = "search",
-
-    LIST = (SIMPLE, SEARCH)
-
-
-class LDAPConfig(Config):
-    def read_config(self, config):
-        ldap_config = config.get("ldap_config", {})
-
-        self.ldap_enabled = ldap_config.get("enabled", False)
-
-        if self.ldap_enabled:
-            # verify dependencies are available
-            try:
-                import ldap3
-                ldap3  # to stop unused lint
-            except ImportError:
-                raise ConfigError(MISSING_LDAP3)
-
-            self.ldap_mode = LDAPMode.SIMPLE
-
-            # verify config sanity
-            self.require_keys(ldap_config, [
-                "uri",
-                "base",
-                "attributes",
-            ])
-
-            self.ldap_uri = ldap_config["uri"]
-            self.ldap_start_tls = ldap_config.get("start_tls", False)
-            self.ldap_base = ldap_config["base"]
-            self.ldap_attributes = ldap_config["attributes"]
-
-            if "bind_dn" in ldap_config:
-                self.ldap_mode = LDAPMode.SEARCH
-                self.require_keys(ldap_config, [
-                    "bind_dn",
-                    "bind_password",
-                ])
-
-                self.ldap_bind_dn = ldap_config["bind_dn"]
-                self.ldap_bind_password = ldap_config["bind_password"]
-                self.ldap_filter = ldap_config.get("filter", None)
-
-            # verify attribute lookup
-            self.require_keys(ldap_config['attributes'], [
-                "uid",
-                "name",
-                "mail",
-            ])
-
-    def require_keys(self, config, required):
-        missing = [key for key in required if key not in config]
-        if missing:
-            raise ConfigError(
-                "LDAP enabled but missing required config values: {}".format(
-                    ", ".join(missing)
-                )
-            )
-
-    def default_config(self, **kwargs):
-        return """\
-        # ldap_config:
-        #   enabled: true
-        #   uri: "ldap://ldap.example.com:389"
-        #   start_tls: true
-        #   base: "ou=users,dc=example,dc=com"
-        #   attributes:
-        #      uid: "cn"
-        #      mail: "email"
-        #      name: "givenName"
-        #   #bind_dn:
-        #   #bind_password:
-        #   #filter: "(objectClass=posixAccount)"
-        """

synapse/config/logger.py

@@ -15,14 +15,13 @@
 
 from ._base import Config
 from synapse.util.logcontext import LoggingContextFilter
-from twisted.python.log import PythonLoggingObserver
+from twisted.logger import globalLogBeginner, STDLibLogObserver
 import logging
 import logging.config
 import yaml
 from string import Template
 import os
 import signal
-from synapse.util.debug import debug_deferreds
 
 
 DEFAULT_LOG_CONFIG = Template("""
@@ -50,6 +49,7 @@ handlers:
   console:
     class: logging.StreamHandler
     formatter: precise
+    filters: [context]
 
 loggers:
     synapse:
@@ -70,8 +70,6 @@ class LoggingConfig(Config):
         self.verbosity = config.get("verbose", 0)
         self.log_config = self.abspath(config.get("log_config"))
         self.log_file = self.abspath(config.get("log_file"))
-        if config.get("full_twisted_stacktraces"):
-            debug_deferreds()
 
     def default_config(self, config_dir_path, server_name, **kwargs):
         log_file = self.abspath("homeserver.log")
@@ -87,11 +85,6 @@ class LoggingConfig(Config):
 
         # A yaml python logging config file
         log_config: "%(log_config)s"
-
-        # Stop twisted from discarding the stack traces of exceptions in
-        # deferreds by waiting a reactor tick before running a deferred's
-        # callbacks.
-        # full_twisted_stacktraces: true
         """ % locals()
 
     def read_arguments(self, args):
@@ -179,5 +172,15 @@ def setup_logging(log_config=None, log_file=None, verbosity=None):
         with open(log_config, 'r') as f:
             logging.config.dictConfig(yaml.load(f))
 
-    observer = PythonLoggingObserver()
-    observer.start()
+    # It's critical to point twisted's internal logging somewhere, otherwise it
+    # stacks up and leaks kup to 64K object;
+    # see: https://twistedmatrix.com/trac/ticket/8164
+    #
+    # Routing to the python logging framework could be a performance problem if
+    # the handlers blocked for a long time as python.logging is a blocking API
+    # see https://twistedmatrix.com/documents/current/core/howto/logger.html
+    # filed as https://github.com/matrix-org/synapse/issues/1727
+    #
+    # However this may not be too much of a problem if we are just writing to a file.
+    observer = STDLibLogObserver()
+    globalLogBeginner.beginLoggingTo([observer])

synapse/config/password_auth_providers.py

@@ -0,0 +1,72 @@
+# -*- coding: utf-8 -*-
+# Copyright 2016 Openmarket
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from ._base import Config, ConfigError
+
+import importlib
+
+
+class PasswordAuthProviderConfig(Config):
+    def read_config(self, config):
+        self.password_providers = []
+
+        # We want to be backwards compatible with the old `ldap_config`
+        # param.
+        ldap_config = config.get("ldap_config", {})
+        self.ldap_enabled = ldap_config.get("enabled", False)
+        if self.ldap_enabled:
+            from ldap_auth_provider import LdapAuthProvider
+            parsed_config = LdapAuthProvider.parse_config(ldap_config)
+            self.password_providers.append((LdapAuthProvider, parsed_config))
+
+        providers = config.get("password_providers", [])
+        for provider in providers:
+            # This is for backwards compat when the ldap auth provider resided
+            # in this package.
+            if provider['module'] == "synapse.util.ldap_auth_provider.LdapAuthProvider":
+                from ldap_auth_provider import LdapAuthProvider
+                provider_class = LdapAuthProvider
+            else:
+                # We need to import the module, and then pick the class out of
+                # that, so we split based on the last dot.
+                module, clz = provider['module'].rsplit(".", 1)
+                module = importlib.import_module(module)
+                provider_class = getattr(module, clz)
+
+            try:
+                provider_config = provider_class.parse_config(provider["config"])
+            except Exception as e:
+                raise ConfigError(
+                    "Failed to parse config for %r: %r" % (provider['module'], e)
+                )
+            self.password_providers.append((provider_class, provider_config))
+
+    def default_config(self, **kwargs):
+        return """\
+        # password_providers:
+        #     - module: "ldap_auth_provider.LdapAuthProvider"
+        #       config:
+        #         enabled: true
+        #         uri: "ldap://ldap.example.com:389"
+        #         start_tls: true
+        #         base: "ou=users,dc=example,dc=com"
+        #         attributes:
+        #            uid: "cn"
+        #            mail: "email"
+        #            name: "givenName"
+        #         #bind_dn:
+        #         #bind_password:
+        #         #filter: "(objectClass=posixAccount)"
+        """

synapse/config/registration.py

@@ -32,7 +32,6 @@ class RegistrationConfig(Config):
             )
 
         self.registration_shared_secret = config.get("registration_shared_secret")
-        self.user_creation_max_duration = int(config["user_creation_max_duration"])
 
         self.bcrypt_rounds = config.get("bcrypt_rounds", 12)
         self.trusted_third_party_id_servers = config["trusted_third_party_id_servers"]
@@ -55,11 +54,6 @@ class RegistrationConfig(Config):
         # secret, even if registration is otherwise disabled.
         registration_shared_secret: "%(registration_shared_secret)s"
 
-        # Sets the expiry for the short term user creation in
-        # milliseconds. For instance the bellow duration is two weeks
-        # in milliseconds.
-        user_creation_max_duration: 1209600000
-
         # Set the number of bcrypt rounds used to generate password hash.
         # Larger numbers increase the work factor needed to generate the hash.
         # The default number of rounds is 12.

synapse/config/repository.py

@@ -167,6 +167,8 @@ class ContentRepositoryConfig(Config):
         # - '10.0.0.0/8'
         # - '172.16.0.0/12'
         # - '192.168.0.0/16'
+        # - '100.64.0.0/10'
+        # - '169.254.0.0/16'
         #
         # List of IP address CIDR ranges that the URL preview spider is allowed
         # to access even if they are specified in url_preview_ip_range_blacklist.

synapse/config/server.py

@@ -29,7 +29,11 @@ class ServerConfig(Config):
         self.user_agent_suffix = config.get("user_agent_suffix")
         self.use_frozen_dicts = config.get("use_frozen_dicts", False)
         self.public_baseurl = config.get("public_baseurl")
-        self.secondary_directory_servers = config.get("secondary_directory_servers", [])
+
+        # Whether to send federation traffic out in this process. This only
+        # applies to some federation traffic, and so shouldn't be used to
+        # "disable" federation
+        self.send_federation = config.get("send_federation", True)
 
         if self.public_baseurl is not None:
             if self.public_baseurl[-1] != '/':
@@ -38,6 +42,15 @@ class ServerConfig(Config):
 
         self.listeners = config.get("listeners", [])
 
+        for listener in self.listeners:
+            bind_address = listener.pop("bind_address", None)
+            bind_addresses = listener.setdefault("bind_addresses", [])
+
+            if bind_address:
+                bind_addresses.append(bind_address)
+            elif not bind_addresses:
+                bind_addresses.append('')
+
         self.gc_thresholds = read_gc_thresholds(config.get("gc_thresholds", None))
 
         bind_port = config.get("bind_port")
@@ -50,7 +63,7 @@ class ServerConfig(Config):
 
             self.listeners.append({
                 "port": bind_port,
-                "bind_address": bind_host,
+                "bind_addresses": [bind_host],
                 "tls": True,
                 "type": "http",
                 "resources": [
@@ -69,7 +82,7 @@ class ServerConfig(Config):
             if unsecure_port:
                 self.listeners.append({
                     "port": unsecure_port,
-                    "bind_address": bind_host,
+                    "bind_addresses": [bind_host],
                     "tls": False,
                     "type": "http",
                     "resources": [
@@ -88,7 +101,7 @@ class ServerConfig(Config):
         if manhole:
             self.listeners.append({
                 "port": manhole,
-                "bind_address": "127.0.0.1",
+                "bind_addresses": ["127.0.0.1"],
                 "type": "manhole",
             })
 
@@ -96,7 +109,7 @@ class ServerConfig(Config):
         if metrics_port:
             self.listeners.append({
                 "port": metrics_port,
-                "bind_address": config.get("metrics_bind_host", "127.0.0.1"),
+                "bind_addresses": [config.get("metrics_bind_host", "127.0.0.1")],
                 "tls": False,
                 "type": "http",
                 "resources": [
@@ -142,14 +155,6 @@ class ServerConfig(Config):
         # The GC threshold parameters to pass to `gc.set_threshold`, if defined
         # gc_thresholds: [700, 10, 10]
 
-        # A list of other Home Servers to fetch the public room directory from
-        # and include in the public room directory of this home server
-        # This is a temporary stopgap solution to populate new server with a
-        # list of rooms until there exists a good solution of a decentralized
-        # room directory.
-        # secondary_directory_servers:
-        #     - matrix.org
-
         # List of ports that Synapse should listen on, their purpose and their
         # configuration.
         listeners:
@@ -159,9 +164,14 @@ class ServerConfig(Config):
             # The port to listen for HTTPS requests on.
             port: %(bind_port)s
 
-            # Local interface to listen on.
-            # The empty string will cause synapse to listen on all interfaces.
-            bind_address: ''
+            # Local addresses to listen on.
+            # This will listen on all IPv4 addresses by default.
+            bind_addresses:
+              - '0.0.0.0'
+              # Uncomment to listen on all IPv6 interfaces
+              # N.B: On at least Linux this will also listen on all IPv4
+              # addresses, so you will need to comment out the line above.
+              # - '::'
 
             # This is a 'http' listener, allows us to specify 'resources'.
             type: http
@@ -192,7 +202,7 @@ class ServerConfig(Config):
           # For when matrix traffic passes through loadbalancer that unwraps TLS.
           - port: %(unsecure_port)s
             tls: false
-            bind_address: ''
+            bind_addresses: ['0.0.0.0']
             type: http
 
             x_forwarded: false

synapse/config/tls.py

@@ -19,6 +19,9 @@ from OpenSSL import crypto
 import subprocess
 import os
 
+from hashlib import sha256
+from unpaddedbase64 import encode_base64
+
 GENERATE_DH_PARAMS = False
 
 
@@ -42,6 +45,19 @@ class TlsConfig(Config):
             config.get("tls_dh_params_path"), "tls_dh_params"
         )
 
+        self.tls_fingerprints = config["tls_fingerprints"]
+
+        # Check that our own certificate is included in the list of fingerprints
+        # and include it if it is not.
+        x509_certificate_bytes = crypto.dump_certificate(
+            crypto.FILETYPE_ASN1,
+            self.tls_certificate
+        )
+        sha256_fingerprint = encode_base64(sha256(x509_certificate_bytes).digest())
+        sha256_fingerprints = set(f["sha256"] for f in self.tls_fingerprints)
+        if sha256_fingerprint not in sha256_fingerprints:
+            self.tls_fingerprints.append({u"sha256": sha256_fingerprint})
+
         # This config option applies to non-federation HTTP clients
         # (e.g. for talking to recaptcha, identity servers, and such)
         # It should never be used in production, and is intended for
@@ -73,6 +89,28 @@ class TlsConfig(Config):
 
         # Don't bind to the https port
         no_tls: False
+
+        # List of allowed TLS fingerprints for this server to publish along
+        # with the signing keys for this server. Other matrix servers that
+        # make HTTPS requests to this server will check that the TLS
+        # certificates returned by this server match one of the fingerprints.
+        #
+        # Synapse automatically adds its the fingerprint of its own certificate
+        # to the list. So if federation traffic is handle directly by synapse
+        # then no modification to the list is required.
+        #
+        # If synapse is run behind a load balancer that handles the TLS then it
+        # will be necessary to add the fingerprints of the certificates used by
+        # the loadbalancers to this list if they are different to the one
+        # synapse is using.
+        #
+        # Homeservers are permitted to cache the list of TLS fingerprints
+        # returned in the key responses up to the "valid_until_ts" returned in
+        # key. It may be necessary to publish the fingerprints of a new
+        # certificate and wait until the "valid_until_ts" of the previous key
+        # responses have passed before deploying it.
+        tls_fingerprints: []
+        # tls_fingerprints: [{"sha256": "<base64_encoded_sha256_fingerprint>"}]
         """ % locals()
 
     def read_tls_certificate(self, cert_path):

synapse/config/voip.py

@@ -19,7 +19,9 @@ class VoipConfig(Config):
 
     def read_config(self, config):
         self.turn_uris = config.get("turn_uris", [])
-        self.turn_shared_secret = config["turn_shared_secret"]
+        self.turn_shared_secret = config.get("turn_shared_secret")
+        self.turn_username = config.get("turn_username")
+        self.turn_password = config.get("turn_password")
         self.turn_user_lifetime = self.parse_duration(config["turn_user_lifetime"])
 
     def default_config(self, **kwargs):
@@ -32,6 +34,11 @@ class VoipConfig(Config):
         # The shared secret used to compute passwords for the TURN server
         turn_shared_secret: "YOUR_SHARED_SECRET"
 
+        # The Username and password if the TURN server needs them and
+        # does not use a token
+        #turn_username: "TURNSERVER_USERNAME"
+        #turn_password: "TURNSERVER_PASSWORD"
+
         # How long generated TURN credentials last
         turn_user_lifetime: "1h"
         """

synapse/config/workers.py

@@ -29,3 +29,13 @@ class WorkerConfig(Config):
         self.worker_log_file = config.get("worker_log_file")
         self.worker_log_config = config.get("worker_log_config")
         self.worker_replication_url = config.get("worker_replication_url")
+
+        if self.worker_listeners:
+            for listener in self.worker_listeners:
+                bind_address = listener.pop("bind_address", None)
+                bind_addresses = listener.setdefault("bind_addresses", [])
+
+                if bind_address:
+                    bind_addresses.append(bind_address)
+                elif not bind_addresses:
+                    bind_addresses.append('')

synapse/crypto/keyclient.py

@@ -77,10 +77,12 @@ class SynapseKeyClientProtocol(HTTPClient):
     def __init__(self):
         self.remote_key = defer.Deferred()
         self.host = None
+        self._peer = None
 
     def connectionMade(self):
-        self.host = self.transport.getHost()
-        logger.debug("Connected to %s", self.host)
+        self._peer = self.transport.getPeer()
+        logger.debug("Connected to %s", self._peer)
+
         self.sendCommand(b"GET", self.path)
         if self.host:
             self.sendHeader(b"Host", self.host)
@@ -124,7 +126,10 @@ class SynapseKeyClientProtocol(HTTPClient):
         self.timer.cancel()
 
     def on_timeout(self):
-        logger.debug("Timeout waiting for response from %s", self.host)
+        logger.debug(
+            "Timeout waiting for response from %s: %s",
+            self.host, self._peer,
+        )
         self.errback(IOError("Timeout waiting for response"))
         self.transport.abortConnection()
 
@@ -133,4 +138,5 @@ class SynapseKeyClientFactory(Factory):
     def protocol(self):
         protocol = SynapseKeyClientProtocol()
         protocol.path = self.path
+        protocol.host = self.host
         return protocol

synapse/crypto/keyring.py

@@ -22,6 +22,7 @@ from synapse.util.logcontext import (
     preserve_context_over_deferred, preserve_context_over_fn, PreserveLoggingContext,
     preserve_fn
 )
+from synapse.util.metrics import Measure
 
 from twisted.internet import defer
 
@@ -44,7 +45,25 @@ import logging
 logger = logging.getLogger(__name__)
 
 
-KeyGroup = namedtuple("KeyGroup", ("server_name", "group_id", "key_ids"))
+VerifyKeyRequest = namedtuple("VerifyRequest", (
+    "server_name", "key_ids", "json_object", "deferred"
+))
+"""
+A request for a verify key to verify a JSON object.
+
+Attributes:
+    server_name(str): The name of the server to verify against.
+    key_ids(set(str)): The set of key_ids to that could be used to verify the
+        JSON object
+    json_object(dict): The JSON object to verify.
+    deferred(twisted.internet.defer.Deferred):
+        A deferred (server_name, key_id, verify_key) tuple that resolves when
+        a verify key has been fetched
+"""
+
+
+class KeyLookupError(ValueError):
+    pass
 
 
 class Keyring(object):
@@ -74,39 +93,32 @@ class Keyring(object):
             list of deferreds indicating success or failure to verify each
             json object's signature for the given server_name.
         """
-        group_id_to_json = {}
-        group_id_to_group = {}
-        group_ids = []
-
-        next_group_id = 0
-        deferreds = {}
+        verify_requests = []
 
         for server_name, json_object in server_and_json:
             logger.debug("Verifying for %s", server_name)
-            group_id = next_group_id
-            next_group_id += 1
-            group_ids.append(group_id)
 
             key_ids = signature_ids(json_object, server_name)
             if not key_ids:
-                deferreds[group_id] = defer.fail(SynapseError(
+                deferred = defer.fail(SynapseError(
                     400,
                     "Not signed with a supported algorithm",
                     Codes.UNAUTHORIZED,
                 ))
             else:
-                deferreds[group_id] = defer.Deferred()
+                deferred = defer.Deferred()
 
-            group = KeyGroup(server_name, group_id, key_ids)
+            verify_request = VerifyKeyRequest(
+                server_name, key_ids, json_object, deferred
+            )
 
-            group_id_to_group[group_id] = group
-            group_id_to_json[group_id] = json_object
+            verify_requests.append(verify_request)
 
         @defer.inlineCallbacks
-        def handle_key_deferred(group, deferred):
-            server_name = group.server_name
+        def handle_key_deferred(verify_request):
+            server_name = verify_request.server_name
             try:
-                _, _, key_id, verify_key = yield deferred
+                _, key_id, verify_key = yield verify_request.deferred
             except IOError as e:
                 logger.warn(
                     "Got IOError when downloading keys for %s: %s %s",
@@ -128,7 +140,7 @@ class Keyring(object):
                     Codes.UNAUTHORIZED,
                 )
 
-            json_object = group_id_to_json[group.group_id]
+            json_object = verify_request.json_object
 
             try:
                 verify_signed_json(json_object, server_name, verify_key)
@@ -157,36 +169,34 @@ class Keyring(object):
 
             # Actually start fetching keys.
             wait_on_deferred.addBoth(
-                lambda _: self.get_server_verify_keys(group_id_to_group, deferreds)
+                lambda _: self.get_server_verify_keys(verify_requests)
             )
 
             # When we've finished fetching all the keys for a given server_name,
             # resolve the deferred passed to `wait_for_previous_lookups` so that
             # any lookups waiting will proceed.
-            server_to_gids = {}
+            server_to_request_ids = {}
 
-            def remove_deferreds(res, server_name, group_id):
-                server_to_gids[server_name].discard(group_id)
-                if not server_to_gids[server_name]:
+            def remove_deferreds(res, server_name, verify_request):
+                request_id = id(verify_request)
+                server_to_request_ids[server_name].discard(request_id)
+                if not server_to_request_ids[server_name]:
                     d = server_to_deferred.pop(server_name, None)
                     if d:
                         d.callback(None)
                 return res
 
-            for g_id, deferred in deferreds.items():
-                server_name = group_id_to_group[g_id].server_name
-                server_to_gids.setdefault(server_name, set()).add(g_id)
-                deferred.addBoth(remove_deferreds, server_name, g_id)
+            for verify_request in verify_requests:
+                server_name = verify_request.server_name
+                request_id = id(verify_request)
+                server_to_request_ids.setdefault(server_name, set()).add(request_id)
+                deferred.addBoth(remove_deferreds, server_name, verify_request)
 
         # Pass those keys to handle_key_deferred so that the json object
         # signatures can be verified
         return [
-            preserve_context_over_fn(
-                handle_key_deferred,
-                group_id_to_group[g_id],
-                deferreds[g_id],
-            )
-            for g_id in group_ids
+            preserve_context_over_fn(handle_key_deferred, verify_request)
+            for verify_request in verify_requests
         ]
 
     @defer.inlineCallbacks
@@ -220,7 +230,7 @@ class Keyring(object):
 
             d.addBoth(rm, server_name)
 
-    def get_server_verify_keys(self, group_id_to_group, group_id_to_deferred):
+    def get_server_verify_keys(self, verify_requests):
         """Takes a dict of KeyGroups and tries to find at least one key for
         each group.
         """
@@ -234,76 +244,79 @@ class Keyring(object):
 
         @defer.inlineCallbacks
         def do_iterations():
-            merged_results = {}
+            with Measure(self.clock, "get_server_verify_keys"):
+                merged_results = {}
 
-            missing_keys = {}
-            for group in group_id_to_group.values():
-                missing_keys.setdefault(group.server_name, set()).update(
-                    group.key_ids
-                )
-
-            for fn in key_fetch_fns:
-                results = yield fn(missing_keys.items())
-                merged_results.update(results)
-
-                # We now need to figure out which groups we have keys for
-                # and which we don't
-                missing_groups = {}
-                for group in group_id_to_group.values():
-                    for key_id in group.key_ids:
-                        if key_id in merged_results[group.server_name]:
-                            with PreserveLoggingContext():
-                                group_id_to_deferred[group.group_id].callback((
-                                    group.group_id,
-                                    group.server_name,
-                                    key_id,
-                                    merged_results[group.server_name][key_id],
-                                ))
-                            break
-                    else:
-                        missing_groups.setdefault(
-                            group.server_name, []
-                        ).append(group)
-
-                if not missing_groups:
-                    break
-
-                missing_keys = {
-                    server_name: set(
-                        key_id for group in groups for key_id in group.key_ids
+                missing_keys = {}
+                for verify_request in verify_requests:
+                    missing_keys.setdefault(verify_request.server_name, set()).update(
+                        verify_request.key_ids
                     )
-                    for server_name, groups in missing_groups.items()
-                }
 
-            for group in missing_groups.values():
-                group_id_to_deferred[group.group_id].errback(SynapseError(
-                    401,
-                    "No key for %s with id %s" % (
-                        group.server_name, group.key_ids,
-                    ),
-                    Codes.UNAUTHORIZED,
-                ))
+                for fn in key_fetch_fns:
+                    results = yield fn(missing_keys.items())
+                    merged_results.update(results)
+
+                    # We now need to figure out which verify requests we have keys
+                    # for and which we don't
+                    missing_keys = {}
+                    requests_missing_keys = []
+                    for verify_request in verify_requests:
+                        server_name = verify_request.server_name
+                        result_keys = merged_results[server_name]
+
+                        if verify_request.deferred.called:
+                            # We've already called this deferred, which probably
+                            # means that we've already found a key for it.
+                            continue
+
+                        for key_id in verify_request.key_ids:
+                            if key_id in result_keys:
+                                with PreserveLoggingContext():
+                                    verify_request.deferred.callback((
+                                        server_name,
+                                        key_id,
+                                        result_keys[key_id],
+                                    ))
+                                break
+                        else:
+                            # The else block is only reached if the loop above
+                            # doesn't break.
+                            missing_keys.setdefault(server_name, set()).update(
+                                verify_request.key_ids
+                            )
+                            requests_missing_keys.append(verify_request)
+
+                    if not missing_keys:
+                        break
+
+                for verify_request in requests_missing_keys.values():
+                    verify_request.deferred.errback(SynapseError(
+                        401,
+                        "No key for %s with id %s" % (
+                            verify_request.server_name, verify_request.key_ids,
+                        ),
+                        Codes.UNAUTHORIZED,
+                    ))
 
         def on_err(err):
-            for deferred in group_id_to_deferred.values():
-                if not deferred.called:
-                    deferred.errback(err)
+            for verify_request in verify_requests:
+                if not verify_request.deferred.called:
+                    verify_request.deferred.errback(err)
 
         do_iterations().addErrback(on_err)
 
-        return group_id_to_deferred
-
     @defer.inlineCallbacks
     def get_keys_from_store(self, server_name_and_key_ids):
-        res = yield defer.gatherResults(
+        res = yield preserve_context_over_deferred(defer.gatherResults(
             [
-                self.store.get_server_verify_keys(
+                preserve_fn(self.store.get_server_verify_keys)(
                     server_name, key_ids
                 ).addCallback(lambda ks, server: (server, ks), server_name)
                 for server_name, key_ids in server_name_and_key_ids
             ],
             consumeErrors=True,
-        ).addErrback(unwrapFirstError)
+        )).addErrback(unwrapFirstError)
 
         defer.returnValue(dict(res))
 
@@ -324,13 +337,13 @@ class Keyring(object):
                 )
                 defer.returnValue({})
 
-        results = yield defer.gatherResults(
+        results = yield preserve_context_over_deferred(defer.gatherResults(
             [
-                get_key(p_name, p_keys)
+                preserve_fn(get_key)(p_name, p_keys)
                 for p_name, p_keys in self.perspective_servers.items()
             ],
             consumeErrors=True,
-        ).addErrback(unwrapFirstError)
+        )).addErrback(unwrapFirstError)
 
         union_of_keys = {}
         for result in results:
@@ -356,7 +369,7 @@ class Keyring(object):
                     )
                 except Exception as e:
                     logger.info(
-                        "Unable to getting key %r for %r directly: %s %s",
+                        "Unable to get key %r for %r directly: %s %s",
                         key_ids, server_name,
                         type(e).__name__, str(e.message),
                     )
@@ -370,13 +383,13 @@ class Keyring(object):
 
             defer.returnValue(keys)
 
-        results = yield defer.gatherResults(
+        results = yield preserve_context_over_deferred(defer.gatherResults(
             [
-                get_key(server_name, key_ids)
+                preserve_fn(get_key)(server_name, key_ids)
                 for server_name, key_ids in server_name_and_key_ids
             ],
             consumeErrors=True,
-        ).addErrback(unwrapFirstError)
+        )).addErrback(unwrapFirstError)
 
         merged = {}
         for result in results:
@@ -418,7 +431,7 @@ class Keyring(object):
         for response in responses:
             if (u"signatures" not in response
                     or perspective_name not in response[u"signatures"]):
-                raise ValueError(
+                raise KeyLookupError(
                     "Key response not signed by perspective server"
                     " %r" % (perspective_name,)
                 )
@@ -441,21 +454,21 @@ class Keyring(object):
                     list(response[u"signatures"][perspective_name]),
                     list(perspective_keys)
                 )
-                raise ValueError(
+                raise KeyLookupError(
                     "Response not signed with a known key for perspective"
                     " server %r" % (perspective_name,)
                 )
 
             processed_response = yield self.process_v2_response(
-                perspective_name, response
+                perspective_name, response, only_from_server=False
             )
 
             for server_name, response_keys in processed_response.items():
                 keys.setdefault(server_name, {}).update(response_keys)
 
-        yield defer.gatherResults(
+        yield preserve_context_over_deferred(defer.gatherResults(
             [
-                self.store_keys(
+                preserve_fn(self.store_keys)(
                     server_name=server_name,
                     from_server=perspective_name,
                     verify_keys=response_keys,
@@ -463,7 +476,7 @@ class Keyring(object):
                 for server_name, response_keys in keys.items()
             ],
             consumeErrors=True
-        ).addErrback(unwrapFirstError)
+        )).addErrback(unwrapFirstError)
 
         defer.returnValue(keys)
 
@@ -484,10 +497,10 @@ class Keyring(object):
 
             if (u"signatures" not in response
                     or server_name not in response[u"signatures"]):
-                raise ValueError("Key response not signed by remote server")
+                raise KeyLookupError("Key response not signed by remote server")
 
             if "tls_fingerprints" not in response:
-                raise ValueError("Key response missing TLS fingerprints")
+                raise KeyLookupError("Key response missing TLS fingerprints")
 
             certificate_bytes = crypto.dump_certificate(
                 crypto.FILETYPE_ASN1, tls_certificate
@@ -501,7 +514,7 @@ class Keyring(object):
                     response_sha256_fingerprints.add(fingerprint[u"sha256"])
 
             if sha256_fingerprint_b64 not in response_sha256_fingerprints:
-                raise ValueError("TLS certificate not allowed by fingerprints")
+                raise KeyLookupError("TLS certificate not allowed by fingerprints")
 
             response_keys = yield self.process_v2_response(
                 from_server=server_name,
@@ -511,7 +524,7 @@ class Keyring(object):
 
             keys.update(response_keys)
 
-        yield defer.gatherResults(
+        yield preserve_context_over_deferred(defer.gatherResults(
             [
                 preserve_fn(self.store_keys)(
                     server_name=key_server_name,
@@ -521,13 +534,13 @@ class Keyring(object):
                 for key_server_name, verify_keys in keys.items()
             ],
             consumeErrors=True
-        ).addErrback(unwrapFirstError)
+        )).addErrback(unwrapFirstError)
 
         defer.returnValue(keys)
 
     @defer.inlineCallbacks
     def process_v2_response(self, from_server, response_json,
-                            requested_ids=[]):
+                            requested_ids=[], only_from_server=True):
         time_now_ms = self.clock.time_msec()
         response_keys = {}
         verify_keys = {}
@@ -551,9 +564,16 @@ class Keyring(object):
 
         results = {}
         server_name = response_json["server_name"]
+        if only_from_server:
+            if server_name != from_server:
+                raise KeyLookupError(
+                    "Expected a response for server %r not %r" % (
+                        from_server, server_name
+                    )
+                )
         for key_id in response_json["signatures"].get(server_name, {}):
             if key_id not in response_json["verify_keys"]:
-                raise ValueError(
+                raise KeyLookupError(
                     "Key response must include verification keys for all"
                     " signatures"
                 )
@@ -580,7 +600,7 @@ class Keyring(object):
         response_keys.update(verify_keys)
         response_keys.update(old_verify_keys)
 
-        yield defer.gatherResults(
+        yield preserve_context_over_deferred(defer.gatherResults(
             [
                 preserve_fn(self.store.store_server_keys_json)(
                     server_name=server_name,
@@ -593,7 +613,7 @@ class Keyring(object):
                 for key_id in updated_key_ids
             ],
             consumeErrors=True,
-        ).addErrback(unwrapFirstError)
+        )).addErrback(unwrapFirstError)
 
         results[server_name] = response_keys
 
@@ -621,15 +641,15 @@ class Keyring(object):
 
         if ("signatures" not in response
                 or server_name not in response["signatures"]):
-            raise ValueError("Key response not signed by remote server")
+            raise KeyLookupError("Key response not signed by remote server")
 
         if "tls_certificate" not in response:
-            raise ValueError("Key response missing TLS certificate")
+            raise KeyLookupError("Key response missing TLS certificate")
 
         tls_certificate_b64 = response["tls_certificate"]
 
         if encode_base64(x509_certificate_bytes) != tls_certificate_b64:
-            raise ValueError("TLS certificate doesn't match")
+            raise KeyLookupError("TLS certificate doesn't match")
 
         # Cache the result in the datastore.
 
@@ -645,7 +665,7 @@ class Keyring(object):
 
         for key_id in response["signatures"][server_name]:
             if key_id not in response["verify_keys"]:
-                raise ValueError(
+                raise KeyLookupError(
                     "Key response must include verification keys for all"
                     " signatures"
                 )
@@ -682,7 +702,7 @@ class Keyring(object):
             A deferred that completes when the keys are stored.
         """
         # TODO(markjh): Store whether the keys have expired.
-        yield defer.gatherResults(
+        yield preserve_context_over_deferred(defer.gatherResults(
             [
                 preserve_fn(self.store.store_server_verify_key)(
                     server_name, server_name, key.time_added, key
@@ -690,4 +710,4 @@ class Keyring(object):
                 for key_id, key in verify_keys.items()
             ],
             consumeErrors=True,
-        ).addErrback(unwrapFirstError)
+        )).addErrback(unwrapFirstError)

synapse/event_auth.py

@@ -0,0 +1,678 @@
+# -*- coding: utf-8 -*-
+# Copyright 2014 - 2016 OpenMarket Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+
+from canonicaljson import encode_canonical_json
+from signedjson.key import decode_verify_key_bytes
+from signedjson.sign import verify_signed_json, SignatureVerifyException
+from unpaddedbase64 import decode_base64
+
+from synapse.api.constants import EventTypes, Membership, JoinRules
+from synapse.api.errors import AuthError, SynapseError, EventSizeError
+from synapse.types import UserID, get_domain_from_id
+
+logger = logging.getLogger(__name__)
+
+
+def check(event, auth_events, do_sig_check=True, do_size_check=True):
+    """ Checks if this event is correctly authed.
+
+    Args:
+        event: the event being checked.
+        auth_events (dict: event-key -> event): the existing room state.
+
+
+    Returns:
+        True if the auth checks pass.
+    """
+    if do_size_check:
+        _check_size_limits(event)
+
+    if not hasattr(event, "room_id"):
+        raise AuthError(500, "Event has no room_id: %s" % event)
+
+    if do_sig_check:
+        sender_domain = get_domain_from_id(event.sender)
+        event_id_domain = get_domain_from_id(event.event_id)
+
+        is_invite_via_3pid = (
+            event.type == EventTypes.Member
+            and event.membership == Membership.INVITE
+            and "third_party_invite" in event.content
+        )
+
+        # Check the sender's domain has signed the event
+        if not event.signatures.get(sender_domain):
+            # We allow invites via 3pid to have a sender from a different
+            # HS, as the sender must match the sender of the original
+            # 3pid invite. This is checked further down with the
+            # other dedicated membership checks.
+            if not is_invite_via_3pid:
+                raise AuthError(403, "Event not signed by sender's server")
+
+        # Check the event_id's domain has signed the event
+        if not event.signatures.get(event_id_domain):
+            raise AuthError(403, "Event not signed by sending server")
+
+    if auth_events is None:
+        # Oh, we don't know what the state of the room was, so we
+        # are trusting that this is allowed (at least for now)
+        logger.warn("Trusting event: %s", event.event_id)
+        return True
+
+    if event.type == EventTypes.Create:
+        room_id_domain = get_domain_from_id(event.room_id)
+        if room_id_domain != sender_domain:
+            raise AuthError(
+                403,
+                "Creation event's room_id domain does not match sender's"
+            )
+        # FIXME
+        return True
+
+    creation_event = auth_events.get((EventTypes.Create, ""), None)
+
+    if not creation_event:
+        raise SynapseError(
+            403,
+            "Room %r does not exist" % (event.room_id,)
+        )
+
+    creating_domain = get_domain_from_id(event.room_id)
+    originating_domain = get_domain_from_id(event.sender)
+    if creating_domain != originating_domain:
+        if not _can_federate(event, auth_events):
+            raise AuthError(
+                403,
+                "This room has been marked as unfederatable."
+            )
+
+    # FIXME: Temp hack
+    if event.type == EventTypes.Aliases:
+        if not event.is_state():
+            raise AuthError(
+                403,
+                "Alias event must be a state event",
+            )
+        if not event.state_key:
+            raise AuthError(
+                403,
+                "Alias event must have non-empty state_key"
+            )
+        sender_domain = get_domain_from_id(event.sender)
+        if event.state_key != sender_domain:
+            raise AuthError(
+                403,
+                "Alias event's state_key does not match sender's domain"
+            )
+        return True
+
+    if logger.isEnabledFor(logging.DEBUG):
+        logger.debug(
+            "Auth events: %s",
+            [a.event_id for a in auth_events.values()]
+        )
+
+    if event.type == EventTypes.Member:
+        allowed = _is_membership_change_allowed(
+            event, auth_events
+        )
+        if allowed:
+            logger.debug("Allowing! %s", event)
+        else:
+            logger.debug("Denying! %s", event)
+        return allowed
+
+    _check_event_sender_in_room(event, auth_events)
+
+    # Special case to allow m.room.third_party_invite events wherever
+    # a user is allowed to issue invites.  Fixes
+    # https://github.com/vector-im/vector-web/issues/1208 hopefully
+    if event.type == EventTypes.ThirdPartyInvite:
+        user_level = get_user_power_level(event.user_id, auth_events)
+        invite_level = _get_named_level(auth_events, "invite", 0)
+
+        if user_level < invite_level:
+            raise AuthError(
+                403, (
+                    "You cannot issue a third party invite for %s." %
+                    (event.content.display_name,)
+                )
+            )
+        else:
+            return True
+
+    _can_send_event(event, auth_events)
+
+    if event.type == EventTypes.PowerLevels:
+        _check_power_levels(event, auth_events)
+
+    if event.type == EventTypes.Redaction:
+        check_redaction(event, auth_events)
+
+    logger.debug("Allowing! %s", event)
+
+
+def _check_size_limits(event):
+    def too_big(field):
+        raise EventSizeError("%s too large" % (field,))
+
+    if len(event.user_id) > 255:
+        too_big("user_id")
+    if len(event.room_id) > 255:
+        too_big("room_id")
+    if event.is_state() and len(event.state_key) > 255:
+        too_big("state_key")
+    if len(event.type) > 255:
+        too_big("type")
+    if len(event.event_id) > 255:
+        too_big("event_id")
+    if len(encode_canonical_json(event.get_pdu_json())) > 65536:
+        too_big("event")
+
+
+def _can_federate(event, auth_events):
+    creation_event = auth_events.get((EventTypes.Create, ""))
+
+    return creation_event.content.get("m.federate", True) is True
+
+
+def _is_membership_change_allowed(event, auth_events):
+    membership = event.content["membership"]
+
+    # Check if this is the room creator joining:
+    if len(event.prev_events) == 1 and Membership.JOIN == membership:
+        # Get room creation event:
+        key = (EventTypes.Create, "", )
+        create = auth_events.get(key)
+        if create and event.prev_events[0][0] == create.event_id:
+            if create.content["creator"] == event.state_key:
+                return True
+
+    target_user_id = event.state_key
+
+    creating_domain = get_domain_from_id(event.room_id)
+    target_domain = get_domain_from_id(target_user_id)
+    if creating_domain != target_domain:
+        if not _can_federate(event, auth_events):
+            raise AuthError(
+                403,
+                "This room has been marked as unfederatable."
+            )
+
+    # get info about the caller
+    key = (EventTypes.Member, event.user_id, )
+    caller = auth_events.get(key)
+
+    caller_in_room = caller and caller.membership == Membership.JOIN
+    caller_invited = caller and caller.membership == Membership.INVITE
+
+    # get info about the target
+    key = (EventTypes.Member, target_user_id, )
+    target = auth_events.get(key)
+
+    target_in_room = target and target.membership == Membership.JOIN
+    target_banned = target and target.membership == Membership.BAN
+
+    key = (EventTypes.JoinRules, "", )
+    join_rule_event = auth_events.get(key)
+    if join_rule_event:
+        join_rule = join_rule_event.content.get(
+            "join_rule", JoinRules.INVITE
+        )
+    else:
+        join_rule = JoinRules.INVITE
+
+    user_level = get_user_power_level(event.user_id, auth_events)
+    target_level = get_user_power_level(
+        target_user_id, auth_events
+    )
+
+    # FIXME (erikj): What should we do here as the default?
+    ban_level = _get_named_level(auth_events, "ban", 50)
+
+    logger.debug(
+        "_is_membership_change_allowed: %s",
+        {
+            "caller_in_room": caller_in_room,
+            "caller_invited": caller_invited,
+            "target_banned": target_banned,
+            "target_in_room": target_in_room,
+            "membership": membership,
+            "join_rule": join_rule,
+            "target_user_id": target_user_id,
+            "event.user_id": event.user_id,
+        }
+    )
+
+    if Membership.INVITE == membership and "third_party_invite" in event.content:
+        if not _verify_third_party_invite(event, auth_events):
+            raise AuthError(403, "You are not invited to this room.")
+        if target_banned:
+            raise AuthError(
+                403, "%s is banned from the room" % (target_user_id,)
+            )
+        return True
+
+    if Membership.JOIN != membership:
+        if (caller_invited
+                and Membership.LEAVE == membership
+                and target_user_id == event.user_id):
+            return True
+
+        if not caller_in_room:  # caller isn't joined
+            raise AuthError(
+                403,
+                "%s not in room %s." % (event.user_id, event.room_id,)
+            )
+
+    if Membership.INVITE == membership:
+        # TODO (erikj): We should probably handle this more intelligently
+        # PRIVATE join rules.
+
+        # Invites are valid iff caller is in the room and target isn't.
+        if target_banned:
+            raise AuthError(
+                403, "%s is banned from the room" % (target_user_id,)
+            )
+        elif target_in_room:  # the target is already in the room.
+            raise AuthError(403, "%s is already in the room." %
+                                 target_user_id)
+        else:
+            invite_level = _get_named_level(auth_events, "invite", 0)
+
+            if user_level < invite_level:
+                raise AuthError(
+                    403, "You cannot invite user %s." % target_user_id
+                )
+    elif Membership.JOIN == membership:
+        # Joins are valid iff caller == target and they were:
+        # invited: They are accepting the invitation
+        # joined: It's a NOOP
+        if event.user_id != target_user_id:
+            raise AuthError(403, "Cannot force another user to join.")
+        elif target_banned:
+            raise AuthError(403, "You are banned from this room")
+        elif join_rule == JoinRules.PUBLIC:
+            pass
+        elif join_rule == JoinRules.INVITE:
+            if not caller_in_room and not caller_invited:
+                raise AuthError(403, "You are not invited to this room.")
+        else:
+            # TODO (erikj): may_join list
+            # TODO (erikj): private rooms
+            raise AuthError(403, "You are not allowed to join this room")
+    elif Membership.LEAVE == membership:
+        # TODO (erikj): Implement kicks.
+        if target_banned and user_level < ban_level:
+            raise AuthError(
+                403, "You cannot unban user &s." % (target_user_id,)
+            )
+        elif target_user_id != event.user_id:
+            kick_level = _get_named_level(auth_events, "kick", 50)
+
+            if user_level < kick_level or user_level <= target_level:
+                raise AuthError(
+                    403, "You cannot kick user %s." % target_user_id
+                )
+    elif Membership.BAN == membership:
+        if user_level < ban_level or user_level <= target_level:
+            raise AuthError(403, "You don't have permission to ban")
+    else:
+        raise AuthError(500, "Unknown membership %s" % membership)
+
+    return True
+
+
+def _check_event_sender_in_room(event, auth_events):
+    key = (EventTypes.Member, event.user_id, )
+    member_event = auth_events.get(key)
+
+    return _check_joined_room(
+        member_event,
+        event.user_id,
+        event.room_id
+    )
+
+
+def _check_joined_room(member, user_id, room_id):
+    if not member or member.membership != Membership.JOIN:
+        raise AuthError(403, "User %s not in room %s (%s)" % (
+            user_id, room_id, repr(member)
+        ))
+
+
+def get_send_level(etype, state_key, auth_events):
+    key = (EventTypes.PowerLevels, "", )
+    send_level_event = auth_events.get(key)
+    send_level = None
+    if send_level_event:
+        send_level = send_level_event.content.get("events", {}).get(
+            etype
+        )
+        if send_level is None:
+            if state_key is not None:
+                send_level = send_level_event.content.get(
+                    "state_default", 50
+                )
+            else:
+                send_level = send_level_event.content.get(
+                    "events_default", 0
+                )
+
+    if send_level:
+        send_level = int(send_level)
+    else:
+        send_level = 0
+
+    return send_level
+
+
+def _can_send_event(event, auth_events):
+    send_level = get_send_level(
+        event.type, event.get("state_key", None), auth_events
+    )
+    user_level = get_user_power_level(event.user_id, auth_events)
+
+    if user_level < send_level:
+        raise AuthError(
+            403,
+            "You don't have permission to post that to the room. " +
+            "user_level (%d) < send_level (%d)" % (user_level, send_level)
+        )
+
+    # Check state_key
+    if hasattr(event, "state_key"):
+        if event.state_key.startswith("@"):
+            if event.state_key != event.user_id:
+                raise AuthError(
+                    403,
+                    "You are not allowed to set others state"
+                )
+
+    return True
+
+
+def check_redaction(event, auth_events):
+    """Check whether the event sender is allowed to redact the target event.
+
+    Returns:
+        True if the the sender is allowed to redact the target event if the
+        target event was created by them.
+        False if the sender is allowed to redact the target event with no
+        further checks.
+
+    Raises:
+        AuthError if the event sender is definitely not allowed to redact
+        the target event.
+    """
+    user_level = get_user_power_level(event.user_id, auth_events)
+
+    redact_level = _get_named_level(auth_events, "redact", 50)
+
+    if user_level >= redact_level:
+        return False
+
+    redacter_domain = get_domain_from_id(event.event_id)
+    redactee_domain = get_domain_from_id(event.redacts)
+    if redacter_domain == redactee_domain:
+        return True
+
+    raise AuthError(
+        403,
+        "You don't have permission to redact events"
+    )
+
+
+def _check_power_levels(event, auth_events):
+    user_list = event.content.get("users", {})
+    # Validate users
+    for k, v in user_list.items():
+        try:
+            UserID.from_string(k)
+        except:
+            raise SynapseError(400, "Not a valid user_id: %s" % (k,))
+
+        try:
+            int(v)
+        except:
+            raise SynapseError(400, "Not a valid power level: %s" % (v,))
+
+    key = (event.type, event.state_key, )
+    current_state = auth_events.get(key)
+
+    if not current_state:
+        return
+
+    user_level = get_user_power_level(event.user_id, auth_events)
+
+    # Check other levels:
+    levels_to_check = [
+        ("users_default", None),
+        ("events_default", None),
+        ("state_default", None),
+        ("ban", None),
+        ("redact", None),
+        ("kick", None),
+        ("invite", None),
+    ]
+
+    old_list = current_state.content.get("users")
+    for user in set(old_list.keys() + user_list.keys()):
+        levels_to_check.append(
+            (user, "users")
+        )
+
+    old_list = current_state.content.get("events")
+    new_list = event.content.get("events")
+    for ev_id in set(old_list.keys() + new_list.keys()):
+        levels_to_check.append(
+            (ev_id, "events")
+        )
+
+    old_state = current_state.content
+    new_state = event.content
+
+    for level_to_check, dir in levels_to_check:
+        old_loc = old_state
+        new_loc = new_state
+        if dir:
+            old_loc = old_loc.get(dir, {})
+            new_loc = new_loc.get(dir, {})
+
+        if level_to_check in old_loc:
+            old_level = int(old_loc[level_to_check])
+        else:
+            old_level = None
+
+        if level_to_check in new_loc:
+            new_level = int(new_loc[level_to_check])
+        else:
+            new_level = None
+
+        if new_level is not None and old_level is not None:
+            if new_level == old_level:
+                continue
+
+        if dir == "users" and level_to_check != event.user_id:
+            if old_level == user_level:
+                raise AuthError(
+                    403,
+                    "You don't have permission to remove ops level equal "
+                    "to your own"
+                )
+
+        if old_level > user_level or new_level > user_level:
+            raise AuthError(
+                403,
+                "You don't have permission to add ops level greater "
+                "than your own"
+            )
+
+
+def _get_power_level_event(auth_events):
+    key = (EventTypes.PowerLevels, "", )
+    return auth_events.get(key)
+
+
+def get_user_power_level(user_id, auth_events):
+    power_level_event = _get_power_level_event(auth_events)
+
+    if power_level_event:
+        level = power_level_event.content.get("users", {}).get(user_id)
+        if not level:
+            level = power_level_event.content.get("users_default", 0)
+
+        if level is None:
+            return 0
+        else:
+            return int(level)
+    else:
+        key = (EventTypes.Create, "", )
+        create_event = auth_events.get(key)
+        if (create_event is not None and
+                create_event.content["creator"] == user_id):
+            return 100
+        else:
+            return 0
+
+
+def _get_named_level(auth_events, name, default):
+    power_level_event = _get_power_level_event(auth_events)
+
+    if not power_level_event:
+        return default
+
+    level = power_level_event.content.get(name, None)
+    if level is not None:
+        return int(level)
+    else:
+        return default
+
+
+def _verify_third_party_invite(event, auth_events):
+    """
+    Validates that the invite event is authorized by a previous third-party invite.
+
+    Checks that the public key, and keyserver, match those in the third party invite,
+    and that the invite event has a signature issued using that public key.
+
+    Args:
+        event: The m.room.member join event being validated.
+        auth_events: All relevant previous context events which may be used
+            for authorization decisions.
+
+    Return:
+        True if the event fulfills the expectations of a previous third party
+        invite event.
+    """
+    if "third_party_invite" not in event.content:
+        return False
+    if "signed" not in event.content["third_party_invite"]:
+        return False
+    signed = event.content["third_party_invite"]["signed"]
+    for key in {"mxid", "token"}:
+        if key not in signed:
+            return False
+
+    token = signed["token"]
+
+    invite_event = auth_events.get(
+        (EventTypes.ThirdPartyInvite, token,)
+    )
+    if not invite_event:
+        return False
+
+    if invite_event.sender != event.sender:
+        return False
+
+    if event.user_id != invite_event.user_id:
+        return False
+
+    if signed["mxid"] != event.state_key:
+        return False
+    if signed["token"] != token:
+        return False
+
+    for public_key_object in get_public_keys(invite_event):
+        public_key = public_key_object["public_key"]
+        try:
+            for server, signature_block in signed["signatures"].items():
+                for key_name, encoded_signature in signature_block.items():
+                    if not key_name.startswith("ed25519:"):
+                        continue
+                    verify_key = decode_verify_key_bytes(
+                        key_name,
+                        decode_base64(public_key)
+                    )
+                    verify_signed_json(signed, server, verify_key)
+
+                    # We got the public key from the invite, so we know that the
+                    # correct server signed the signed bundle.
+                    # The caller is responsible for checking that the signing
+                    # server has not revoked that public key.
+                    return True
+        except (KeyError, SignatureVerifyException,):
+            continue
+    return False
+
+
+def get_public_keys(invite_event):
+    public_keys = []
+    if "public_key" in invite_event.content:
+        o = {
+            "public_key": invite_event.content["public_key"],
+        }
+        if "key_validity_url" in invite_event.content:
+            o["key_validity_url"] = invite_event.content["key_validity_url"]
+        public_keys.append(o)
+    public_keys.extend(invite_event.content.get("public_keys", []))
+    return public_keys
+
+
+def auth_types_for_event(event):
+    """Given an event, return a list of (EventType, StateKey) that may be
+    needed to auth the event. The returned list may be a superset of what
+    would actually be required depending on the full state of the room.
+
+    Used to limit the number of events to fetch from the database to
+    actually auth the event.
+    """
+    if event.type == EventTypes.Create:
+        return []
+
+    auth_types = []
+
+    auth_types.append((EventTypes.PowerLevels, "", ))
+    auth_types.append((EventTypes.Member, event.user_id, ))
+    auth_types.append((EventTypes.Create, "", ))
+
+    if event.type == EventTypes.Member:
+        membership = event.content["membership"]
+        if membership in [Membership.JOIN, Membership.INVITE]:
+            auth_types.append((EventTypes.JoinRules, "", ))
+
+        auth_types.append((EventTypes.Member, event.state_key, ))
+
+        if membership == Membership.INVITE:
+            if "third_party_invite" in event.content:
+                key = (
+                    EventTypes.ThirdPartyInvite,
+                    event.content["third_party_invite"]["signed"]["token"]
+                )
+                auth_types.append(key)
+
+    return auth_types

synapse/events/__init__.py

@@ -36,6 +36,15 @@ class _EventInternalMetadata(object):
     def is_invite_from_remote(self):
         return getattr(self, "invite_from_remote", False)
 
+    def get_send_on_behalf_of(self):
+        """Whether this server should send the event on behalf of another server.
+        This is used by the federation "send_join" API to forward the initial join
+        event for a server in the room.
+
+        returns a str with the name of the server this event is sent on behalf of.
+        """
+        return getattr(self, "send_on_behalf_of", None)
+
 
 def _event_dict_property(key):
     def getter(self):
@@ -70,7 +79,6 @@ class EventBase(object):
     auth_events = _event_dict_property("auth_events")
     depth = _event_dict_property("depth")
     content = _event_dict_property("content")
-    event_id = _event_dict_property("event_id")
     hashes = _event_dict_property("hashes")
     origin = _event_dict_property("origin")
     origin_server_ts = _event_dict_property("origin_server_ts")
@@ -79,8 +87,6 @@ class EventBase(object):
     redacts = _event_dict_property("redacts")
     room_id = _event_dict_property("room_id")
     sender = _event_dict_property("sender")
-    state_key = _event_dict_property("state_key")
-    type = _event_dict_property("type")
     user_id = _event_dict_property("sender")
 
     @property
@@ -99,7 +105,7 @@ class EventBase(object):
 
         return d
 
-    def get(self, key, default):
+    def get(self, key, default=None):
         return self._event_dict.get(key, default)
 
     def get_internal_metadata_dict(self):
@@ -153,6 +159,11 @@ class FrozenEvent(EventBase):
         else:
             frozen_dict = event_dict
 
+        self.event_id = event_dict["event_id"]
+        self.type = event_dict["type"]
+        if "state_key" in event_dict:
+            self.state_key = event_dict["state_key"]
+
         super(FrozenEvent, self).__init__(
             frozen_dict,
             signatures=signatures,

synapse/events/builder.py

@@ -13,7 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from . import EventBase, FrozenEvent
+from . import EventBase, FrozenEvent, _event_dict_property
 
 from synapse.types import EventID
 
@@ -34,6 +34,10 @@ class EventBuilder(EventBase):
             internal_metadata_dict=internal_metadata_dict,
         )
 
+    event_id = _event_dict_property("event_id")
+    state_key = _event_dict_property("state_key")
+    type = _event_dict_property("type")
+
     def build(self):
         return FrozenEvent.from_event(self)
 

synapse/events/snapshot.py

@@ -15,9 +15,30 @@
 
 
 class EventContext(object):
+    __slots__ = [
+        "current_state_ids",
+        "prev_state_ids",
+        "state_group",
+        "rejected",
+        "push_actions",
+        "prev_group",
+        "delta_ids",
+        "prev_state_events",
+    ]
 
-    def __init__(self, current_state=None):
-        self.current_state = current_state
+    def __init__(self):
+        # The current state including the current event
+        self.current_state_ids = None
+        # The current state excluding the current event
+        self.prev_state_ids = None
         self.state_group = None
+
         self.rejected = False
         self.push_actions = []
+
+        # A previously persisted state group and a delta between that
+        # and this state.
+        self.prev_group = None
+        self.delta_ids = None
+
+        self.prev_state_events = None

synapse/events/utils.py

@@ -16,6 +16,17 @@
 from synapse.api.constants import EventTypes
 from . import EventBase
 
+from frozendict import frozendict
+
+import re
+
+# Split strings on "." but not "\." This uses a negative lookbehind assertion for '\'
+# (?<!stuff) matches if the current position in the string is not preceded
+# by a match for 'stuff'.
+# TODO: This is fast, but fails to handle "foo\\.bar" which should be treated as
+#       the literal fields "foo\" and "bar" but will instead be treated as "foo\\.bar"
+SPLIT_FIELD_REGEX = re.compile(r'(?<!\\)\.')
+
 
 def prune_event(event):
     """ Returns a pruned version of the given event, which removes all keys we
@@ -88,6 +99,8 @@ def prune_event(event):
 
     if "age_ts" in event.unsigned:
         allowed_fields["unsigned"]["age_ts"] = event.unsigned["age_ts"]
+    if "replaces_state" in event.unsigned:
+        allowed_fields["unsigned"]["replaces_state"] = event.unsigned["replaces_state"]
 
     return type(event)(
         allowed_fields,
@@ -95,6 +108,83 @@ def prune_event(event):
     )
 
 
+def _copy_field(src, dst, field):
+    """Copy the field in 'src' to 'dst'.
+
+    For example, if src={"foo":{"bar":5}} and dst={}, and field=["foo","bar"]
+    then dst={"foo":{"bar":5}}.
+
+    Args:
+        src(dict): The dict to read from.
+        dst(dict): The dict to modify.
+        field(list<str>): List of keys to drill down to in 'src'.
+    """
+    if len(field) == 0:  # this should be impossible
+        return
+    if len(field) == 1:  # common case e.g. 'origin_server_ts'
+        if field[0] in src:
+            dst[field[0]] = src[field[0]]
+        return
+
+    # Else is a nested field e.g. 'content.body'
+    # Pop the last field as that's the key to move across and we need the
+    # parent dict in order to access the data. Drill down to the right dict.
+    key_to_move = field.pop(-1)
+    sub_dict = src
+    for sub_field in field:  # e.g. sub_field => "content"
+        if sub_field in sub_dict and type(sub_dict[sub_field]) in [dict, frozendict]:
+            sub_dict = sub_dict[sub_field]
+        else:
+            return
+
+    if key_to_move not in sub_dict:
+        return
+
+    # Insert the key into the output dictionary, creating nested objects
+    # as required. We couldn't do this any earlier or else we'd need to delete
+    # the empty objects if the key didn't exist.
+    sub_out_dict = dst
+    for sub_field in field:
+        sub_out_dict = sub_out_dict.setdefault(sub_field, {})
+    sub_out_dict[key_to_move] = sub_dict[key_to_move]
+
+
+def only_fields(dictionary, fields):
+    """Return a new dict with only the fields in 'dictionary' which are present
+    in 'fields'.
+
+    If there are no event fields specified then all fields are included.
+    The entries may include '.' charaters to indicate sub-fields.
+    So ['content.body'] will include the 'body' field of the 'content' object.
+    A literal '.' character in a field name may be escaped using a '\'.
+
+    Args:
+        dictionary(dict): The dictionary to read from.
+        fields(list<str>): A list of fields to copy over. Only shallow refs are
+        taken.
+    Returns:
+        dict: A new dictionary with only the given fields. If fields was empty,
+        the same dictionary is returned.
+    """
+    if len(fields) == 0:
+        return dictionary
+
+    # for each field, convert it:
+    # ["content.body.thing\.with\.dots"] => [["content", "body", "thing\.with\.dots"]]
+    split_fields = [SPLIT_FIELD_REGEX.split(f) for f in fields]
+
+    # for each element of the output array of arrays:
+    # remove escaping so we can use the right key names.
+    split_fields[:] = [
+        [f.replace(r'\.', r'.') for f in field_array] for field_array in split_fields
+    ]
+
+    output = {}
+    for field_array in split_fields:
+        _copy_field(dictionary, output, field_array)
+    return output
+
+
 def format_event_raw(d):
     return d
 
@@ -135,7 +225,7 @@ def format_event_for_client_v2_without_room_id(d):
 
 def serialize_event(e, time_now_ms, as_client_event=True,
                     event_format=format_event_for_client_v1,
-                    token_id=None):
+                    token_id=None, only_event_fields=None):
     # FIXME(erikj): To handle the case of presence events and the like
     if not isinstance(e, EventBase):
         return e
@@ -162,6 +252,12 @@ def serialize_event(e, time_now_ms, as_client_event=True,
                 d["unsigned"]["transaction_id"] = txn_id
 
     if as_client_event:
-        return event_format(d)
-    else:
-        return d
+        d = event_format(d)
+
+    if only_event_fields:
+        if (not isinstance(only_event_fields, list) or
+                not all(isinstance(f, basestring) for f in only_event_fields)):
+            raise TypeError("only_event_fields must be a list of strings")
+        d = only_fields(d, only_event_fields)
+
+    return d

synapse/federation/__init__.py

@@ -17,10 +17,9 @@
 """
 
 from .replication import ReplicationLayer
-from .transport.client import TransportLayerClient
 
 
-def initialize_http_replication(homeserver):
-    transport = TransportLayerClient(homeserver)
+def initialize_http_replication(hs):
+    transport = hs.get_federation_transport_client()
 
-    return ReplicationLayer(homeserver, transport)
+    return ReplicationLayer(hs, transport)

synapse/federation/federation_base.py

@@ -23,6 +23,7 @@ from synapse.crypto.event_signing import check_event_content_hash
 from synapse.api.errors import SynapseError
 
 from synapse.util import unwrapFirstError
+from synapse.util.logcontext import preserve_fn, preserve_context_over_deferred
 
 import logging
 
@@ -102,10 +103,10 @@ class FederationBase(object):
                 warn, pdu
             )
 
-        valid_pdus = yield defer.gatherResults(
+        valid_pdus = yield preserve_context_over_deferred(defer.gatherResults(
             deferreds,
             consumeErrors=True
-        ).addErrback(unwrapFirstError)
+        )).addErrback(unwrapFirstError)
 
         if include_none:
             defer.returnValue(valid_pdus)
@@ -129,7 +130,7 @@ class FederationBase(object):
             for pdu in pdus
         ]
 
-        deferreds = self.keyring.verify_json_objects_for_server([
+        deferreds = preserve_fn(self.keyring.verify_json_objects_for_server)([
             (p.origin, p.get_pdu_json())
             for p in redacted_pdus
         ])

synapse/federation/federation_client.py

@@ -18,16 +18,15 @@ from twisted.internet import defer
 
 from .federation_base import FederationBase
 from synapse.api.constants import Membership
-from .units import Edu
 
 from synapse.api.errors import (
     CodeMessageException, HttpResponseException, SynapseError,
 )
 from synapse.util import unwrapFirstError
-from synapse.util.async import concurrently_execute
 from synapse.util.caches.expiringcache import ExpiringCache
 from synapse.util.logutils import log_function
-from synapse.events import FrozenEvent
+from synapse.util.logcontext import preserve_fn, preserve_context_over_deferred
+from synapse.events import FrozenEvent, builder
 import synapse.metrics
 
 from synapse.util.retryutils import get_retry_limiter, NotRetryingDestination
@@ -44,17 +43,38 @@ logger = logging.getLogger(__name__)
 # synapse.federation.federation_client is a silly name
 metrics = synapse.metrics.get_metrics_for("synapse.federation.client")
 
-sent_pdus_destination_dist = metrics.register_distribution("sent_pdu_destinations")
-
-sent_edus_counter = metrics.register_counter("sent_edus")
-
 sent_queries_counter = metrics.register_counter("sent_queries", labels=["type"])
 
 
+PDU_RETRY_TIME_MS = 1 * 60 * 1000
+
+
 class FederationClient(FederationBase):
     def __init__(self, hs):
         super(FederationClient, self).__init__(hs)
 
+        self.pdu_destination_tried = {}
+        self._clock.looping_call(
+            self._clear_tried_cache, 60 * 1000,
+        )
+        self.state = hs.get_state_handler()
+
+    def _clear_tried_cache(self):
+        """Clear pdu_destination_tried cache"""
+        now = self._clock.time_msec()
+
+        old_dict = self.pdu_destination_tried
+        self.pdu_destination_tried = {}
+
+        for event_id, destination_dict in old_dict.items():
+            destination_dict = {
+                dest: time
+                for dest, time in destination_dict.items()
+                if time + PDU_RETRY_TIME_MS > now
+            }
+            if destination_dict:
+                self.pdu_destination_tried[event_id] = destination_dict
+
     def start_get_pdu_cache(self):
         self._get_pdu_cache = ExpiringCache(
             cache_name="get_pdu_cache",
@@ -66,55 +86,6 @@ class FederationClient(FederationBase):
 
         self._get_pdu_cache.start()
 
-    @log_function
-    def send_pdu(self, pdu, destinations):
-        """Informs the replication layer about a new PDU generated within the
-        home server that should be transmitted to others.
-
-        TODO: Figure out when we should actually resolve the deferred.
-
-        Args:
-            pdu (Pdu): The new Pdu.
-
-        Returns:
-            Deferred: Completes when we have successfully processed the PDU
-            and replicated it to any interested remote home servers.
-        """
-        order = self._order
-        self._order += 1
-
-        sent_pdus_destination_dist.inc_by(len(destinations))
-
-        logger.debug("[%s] transaction_layer.enqueue_pdu... ", pdu.event_id)
-
-        # TODO, add errback, etc.
-        self._transaction_queue.enqueue_pdu(pdu, destinations, order)
-
-        logger.debug(
-            "[%s] transaction_layer.enqueue_pdu... done",
-            pdu.event_id
-        )
-
-    @log_function
-    def send_edu(self, destination, edu_type, content):
-        edu = Edu(
-            origin=self.server_name,
-            destination=destination,
-            edu_type=edu_type,
-            content=content,
-        )
-
-        sent_edus_counter.inc()
-
-        # TODO, add errback, etc.
-        self._transaction_queue.enqueue_edu(edu)
-        return defer.succeed(None)
-
-    @log_function
-    def send_failure(self, failure, destination):
-        self._transaction_queue.enqueue_failure(failure, destination)
-        return defer.succeed(None)
-
     @log_function
     def make_query(self, destination, query_type, args,
                    retry_on_dns_fail=False):
@@ -139,7 +110,7 @@ class FederationClient(FederationBase):
         )
 
     @log_function
-    def query_client_keys(self, destination, content):
+    def query_client_keys(self, destination, content, timeout):
         """Query device keys for a device hosted on a remote server.
 
         Args:
@@ -151,10 +122,22 @@ class FederationClient(FederationBase):
             response
         """
         sent_queries_counter.inc("client_device_keys")
-        return self.transport_layer.query_client_keys(destination, content)
+        return self.transport_layer.query_client_keys(
+            destination, content, timeout
+        )
 
     @log_function
-    def claim_client_keys(self, destination, content):
+    def query_user_devices(self, destination, user_id, timeout=30000):
+        """Query the device keys for a list of user ids hosted on a remote
+        server.
+        """
+        sent_queries_counter.inc("user_devices")
+        return self.transport_layer.query_user_devices(
+            destination, user_id, timeout
+        )
+
+    @log_function
+    def claim_client_keys(self, destination, content, timeout):
         """Claims one-time keys for a device hosted on a remote server.
 
         Args:
@@ -166,7 +149,9 @@ class FederationClient(FederationBase):
             response
         """
         sent_queries_counter.inc("client_one_time_keys")
-        return self.transport_layer.claim_client_keys(destination, content)
+        return self.transport_layer.claim_client_keys(
+            destination, content, timeout
+        )
 
     @defer.inlineCallbacks
     @log_function
@@ -201,10 +186,10 @@ class FederationClient(FederationBase):
         ]
 
         # FIXME: We should handle signature failures more gracefully.
-        pdus[:] = yield defer.gatherResults(
+        pdus[:] = yield preserve_context_over_deferred(defer.gatherResults(
             self._check_sigs_and_hashes(pdus),
             consumeErrors=True,
-        ).addErrback(unwrapFirstError)
+        )).addErrback(unwrapFirstError)
 
         defer.returnValue(pdus)
 
@@ -236,12 +221,19 @@ class FederationClient(FederationBase):
         # TODO: Rate limit the number of times we try and get the same event.
 
         if self._get_pdu_cache:
-            e = self._get_pdu_cache.get(event_id)
-            if e:
-                defer.returnValue(e)
+            ev = self._get_pdu_cache.get(event_id)
+            if ev:
+                defer.returnValue(ev)
 
-        pdu = None
+        pdu_attempts = self.pdu_destination_tried.setdefault(event_id, {})
+
+        signed_pdu = None
         for destination in destinations:
+            now = self._clock.time_msec()
+            last_attempt = pdu_attempts.get(destination, 0)
+            if last_attempt + PDU_RETRY_TIME_MS > now:
+                continue
+
             try:
                 limiter = yield get_retry_limiter(
                     destination,
@@ -265,39 +257,33 @@ class FederationClient(FederationBase):
                         pdu = pdu_list[0]
 
                         # Check signatures are correct.
-                        pdu = yield self._check_sigs_and_hashes([pdu])[0]
+                        signed_pdu = yield self._check_sigs_and_hashes([pdu])[0]
 
                         break
 
-            except SynapseError:
-                logger.info(
-                    "Failed to get PDU %s from %s because %s",
-                    event_id, destination, e,
-                )
-                continue
-            except CodeMessageException as e:
-                if 400 <= e.code < 500:
-                    raise
+                pdu_attempts[destination] = now
 
+            except SynapseError as e:
                 logger.info(
                     "Failed to get PDU %s from %s because %s",
                     event_id, destination, e,
                 )
-                continue
             except NotRetryingDestination as e:
                 logger.info(e.message)
                 continue
             except Exception as e:
+                pdu_attempts[destination] = now
+
                 logger.info(
                     "Failed to get PDU %s from %s because %s",
                     event_id, destination, e,
                 )
                 continue
 
-        if self._get_pdu_cache is not None and pdu:
-            self._get_pdu_cache[event_id] = pdu
+        if self._get_pdu_cache is not None and signed_pdu:
+            self._get_pdu_cache[event_id] = signed_pdu
 
-        defer.returnValue(pdu)
+        defer.returnValue(signed_pdu)
 
     @defer.inlineCallbacks
     @log_function
@@ -314,6 +300,42 @@ class FederationClient(FederationBase):
             Deferred: Results in a list of PDUs.
         """
 
+        try:
+            # First we try and ask for just the IDs, as thats far quicker if
+            # we have most of the state and auth_chain already.
+            # However, this may 404 if the other side has an old synapse.
+            result = yield self.transport_layer.get_room_state_ids(
+                destination, room_id, event_id=event_id,
+            )
+
+            state_event_ids = result["pdu_ids"]
+            auth_event_ids = result.get("auth_chain_ids", [])
+
+            fetched_events, failed_to_fetch = yield self.get_events(
+                [destination], room_id, set(state_event_ids + auth_event_ids)
+            )
+
+            if failed_to_fetch:
+                logger.warn("Failed to get %r", failed_to_fetch)
+
+            event_map = {
+                ev.event_id: ev for ev in fetched_events
+            }
+
+            pdus = [event_map[e_id] for e_id in state_event_ids if e_id in event_map]
+            auth_chain = [
+                event_map[e_id] for e_id in auth_event_ids if e_id in event_map
+            ]
+
+            auth_chain.sort(key=lambda e: e.depth)
+
+            defer.returnValue((pdus, auth_chain))
+        except HttpResponseException as e:
+            if e.code == 400 or e.code == 404:
+                logger.info("Failed to use get_room_state_ids API, falling back")
+            else:
+                raise e
+
         result = yield self.transport_layer.get_room_state(
             destination, room_id, event_id=event_id,
         )
@@ -327,18 +349,95 @@ class FederationClient(FederationBase):
             for p in result.get("auth_chain", [])
         ]
 
+        seen_events = yield self.store.get_events([
+            ev.event_id for ev in itertools.chain(pdus, auth_chain)
+        ])
+
         signed_pdus = yield self._check_sigs_and_hash_and_fetch(
-            destination, pdus, outlier=True
+            destination,
+            [p for p in pdus if p.event_id not in seen_events],
+            outlier=True
+        )
+        signed_pdus.extend(
+            seen_events[p.event_id] for p in pdus if p.event_id in seen_events
         )
 
         signed_auth = yield self._check_sigs_and_hash_and_fetch(
-            destination, auth_chain, outlier=True
+            destination,
+            [p for p in auth_chain if p.event_id not in seen_events],
+            outlier=True
+        )
+        signed_auth.extend(
+            seen_events[p.event_id] for p in auth_chain if p.event_id in seen_events
         )
 
         signed_auth.sort(key=lambda e: e.depth)
 
         defer.returnValue((signed_pdus, signed_auth))
 
+    @defer.inlineCallbacks
+    def get_events(self, destinations, room_id, event_ids, return_local=True):
+        """Fetch events from some remote destinations, checking if we already
+        have them.
+
+        Args:
+            destinations (list)
+            room_id (str)
+            event_ids (list)
+            return_local (bool): Whether to include events we already have in
+                the DB in the returned list of events
+
+        Returns:
+            Deferred: A deferred resolving to a 2-tuple where the first is a list of
+            events and the second is a list of event ids that we failed to fetch.
+        """
+        if return_local:
+            seen_events = yield self.store.get_events(event_ids, allow_rejected=True)
+            signed_events = seen_events.values()
+        else:
+            seen_events = yield self.store.have_events(event_ids)
+            signed_events = []
+
+        failed_to_fetch = set()
+
+        missing_events = set(event_ids)
+        for k in seen_events:
+            missing_events.discard(k)
+
+        if not missing_events:
+            defer.returnValue((signed_events, failed_to_fetch))
+
+        def random_server_list():
+            srvs = list(destinations)
+            random.shuffle(srvs)
+            return srvs
+
+        batch_size = 20
+        missing_events = list(missing_events)
+        for i in xrange(0, len(missing_events), batch_size):
+            batch = set(missing_events[i:i + batch_size])
+
+            deferreds = [
+                preserve_fn(self.get_pdu)(
+                    destinations=random_server_list(),
+                    event_id=e_id,
+                )
+                for e_id in batch
+            ]
+
+            res = yield preserve_context_over_deferred(
+                defer.DeferredList(deferreds, consumeErrors=True)
+            )
+            for success, result in res:
+                if success and result:
+                    signed_events.append(result)
+                    batch.discard(result.event_id)
+
+            # We removed all events we successfully fetched from `batch`
+            failed_to_fetch.update(batch)
+
+        defer.returnValue((signed_events, failed_to_fetch))
+
     @defer.inlineCallbacks
     @log_function
     def get_event_auth(self, destination, room_id, event_id):
@@ -410,18 +509,25 @@ class FederationClient(FederationBase):
                 if "prev_state" not in pdu_dict:
                     pdu_dict["prev_state"] = []
 
+                ev = builder.EventBuilder(pdu_dict)
+
                 defer.returnValue(
-                    (destination, self.event_from_pdu_json(pdu_dict))
+                    (destination, ev)
                 )
                 break
-            except CodeMessageException:
-                raise
+            except CodeMessageException as e:
+                if not 500 <= e.code < 600:
+                    raise
+                else:
+                    logger.warn(
+                        "Failed to make_%s via %s: %s",
+                        membership, destination, e.message
+                    )
             except Exception as e:
                 logger.warn(
                     "Failed to make_%s via %s: %s",
                     membership, destination, e.message
                 )
-                raise
 
         raise RuntimeError("Failed to send to any server.")
 
@@ -493,8 +599,14 @@ class FederationClient(FederationBase):
                     "auth_chain": signed_auth,
                     "origin": destination,
                 })
-            except CodeMessageException:
-                raise
+            except CodeMessageException as e:
+                if not 500 <= e.code < 600:
+                    raise
+                else:
+                    logger.exception(
+                        "Failed to send_join via %s: %s",
+                        destination, e.message
+                    )
             except Exception as e:
                 logger.exception(
                     "Failed to send_join via %s: %s",
@@ -553,24 +665,17 @@ class FederationClient(FederationBase):
 
         raise RuntimeError("Failed to send to any server.")
 
-    @defer.inlineCallbacks
-    def get_public_rooms(self, destinations):
-        results_by_server = {}
+    def get_public_rooms(self, destination, limit=None, since_token=None,
+                         search_filter=None, include_all_networks=False,
+                         third_party_instance_id=None):
+        if destination == self.server_name:
+            return
 
-        @defer.inlineCallbacks
-        def _get_result(s):
-            if s == self.server_name:
-                defer.returnValue()
-
-            try:
-                result = yield self.transport_layer.get_public_rooms(s)
-                results_by_server[s] = result
-            except:
-                logger.exception("Error getting room list from server %r", s)
-
-        yield concurrently_execute(_get_result, destinations, 3)
-
-        defer.returnValue(results_by_server)
+        return self.transport_layer.get_public_rooms(
+            destination, limit, since_token, search_filter,
+            include_all_networks=include_all_networks,
+            third_party_instance_id=third_party_instance_id,
+        )
 
     @defer.inlineCallbacks
     def query_auth(self, destination, room_id, event_id, local_auth):
@@ -614,7 +719,7 @@ class FederationClient(FederationBase):
 
     @defer.inlineCallbacks
     def get_missing_events(self, destination, room_id, earliest_events_ids,
-                           latest_events, limit, min_depth):
+                           latest_events, limit, min_depth, timeout):
         """Tries to fetch events we are missing. This is called when we receive
         an event without having received all of its ancestors.
 
@@ -628,6 +733,7 @@ class FederationClient(FederationBase):
                 have all previous events for.
             limit (int): Maximum number of events to return.
             min_depth (int): Minimum depth of events tor return.
+            timeout (int): Max time to wait in ms
         """
         try:
             content = yield self.transport_layer.get_missing_events(
@@ -637,6 +743,7 @@ class FederationClient(FederationBase):
                 latest_events=[e.event_id for e in latest_events],
                 limit=limit,
                 min_depth=min_depth,
+                timeout=timeout,
             )
 
             events = [
@@ -647,8 +754,6 @@ class FederationClient(FederationBase):
             signed_events = yield self._check_sigs_and_hash_and_fetch(
                 destination, events, outlier=False
             )
-
-            have_gotten_all_from_destination = True
         except HttpResponseException as e:
             if not e.code == 400:
                 raise
@@ -656,69 +761,6 @@ class FederationClient(FederationBase):
             # We are probably hitting an old server that doesn't support
             # get_missing_events
             signed_events = []
-            have_gotten_all_from_destination = False
-
-        if len(signed_events) >= limit:
-            defer.returnValue(signed_events)
-
-        servers = yield self.store.get_joined_hosts_for_room(room_id)
-
-        servers = set(servers)
-        servers.discard(self.server_name)
-
-        failed_to_fetch = set()
-
-        while len(signed_events) < limit:
-            # Are we missing any?
-
-            seen_events = set(earliest_events_ids)
-            seen_events.update(e.event_id for e in signed_events if e)
-
-            missing_events = {}
-            for e in itertools.chain(latest_events, signed_events):
-                if e.depth > min_depth:
-                    missing_events.update({
-                        e_id: e.depth for e_id, _ in e.prev_events
-                        if e_id not in seen_events
-                        and e_id not in failed_to_fetch
-                    })
-
-            if not missing_events:
-                break
-
-            have_seen = yield self.store.have_events(missing_events)
-
-            for k in have_seen:
-                missing_events.pop(k, None)
-
-            if not missing_events:
-                break
-
-            # Okay, we haven't gotten everything yet. Lets get them.
-            ordered_missing = sorted(missing_events.items(), key=lambda x: x[0])
-
-            if have_gotten_all_from_destination:
-                servers.discard(destination)
-
-            def random_server_list():
-                srvs = list(servers)
-                random.shuffle(srvs)
-                return srvs
-
-            deferreds = [
-                self.get_pdu(
-                    destinations=random_server_list(),
-                    event_id=e_id,
-                )
-                for e_id, depth in ordered_missing[:limit - len(signed_events)]
-            ]
-
-            res = yield defer.DeferredList(deferreds, consumeErrors=True)
-            for (result, val), (e_id, _) in zip(res, ordered_missing):
-                if result and val:
-                    signed_events.append(val)
-                else:
-                    failed_to_fetch.add(e_id)
 
         defer.returnValue(signed_events)
 

synapse/federation/federation_server.py

@@ -21,10 +21,12 @@ from .units import Transaction, Edu
 
 from synapse.util.async import Linearizer
 from synapse.util.logutils import log_function
+from synapse.util.caches.response_cache import ResponseCache
 from synapse.events import FrozenEvent
+from synapse.types import get_domain_from_id
 import synapse.metrics
 
-from synapse.api.errors import FederationError, SynapseError
+from synapse.api.errors import AuthError, FederationError, SynapseError
 
 from synapse.crypto.event_signing import compute_event_signature
 
@@ -48,8 +50,14 @@ class FederationServer(FederationBase):
     def __init__(self, hs):
         super(FederationServer, self).__init__(hs)
 
-        self._room_pdu_linearizer = Linearizer()
-        self._server_linearizer = Linearizer()
+        self.auth = hs.get_auth()
+
+        self._room_pdu_linearizer = Linearizer("fed_room_pdu")
+        self._server_linearizer = Linearizer("fed_server")
+
+        # We cache responses to state queries, as they take a while and often
+        # come in waves.
+        self._state_resp_cache = ResponseCache(hs, timeout_ms=30000)
 
     def set_handler(self, handler):
         """Sets the handler that the replication layer will use to communicate
@@ -125,7 +133,7 @@ class FederationServer(FederationBase):
 
         if response:
             logger.debug(
-                "[%s] We've already responed to this request",
+                "[%s] We've already responded to this request",
                 transaction.transaction_id
             )
             defer.returnValue(response)
@@ -136,6 +144,26 @@ class FederationServer(FederationBase):
         results = []
 
         for pdu in pdu_list:
+            # check that it's actually being sent from a valid destination to
+            # workaround bug #1753 in 0.18.5 and 0.18.6
+            if transaction.origin != get_domain_from_id(pdu.event_id):
+                if not (
+                    pdu.type == 'm.room.member' and
+                    pdu.content and
+                    pdu.content.get("membership", None) == 'join' and
+                    self.hs.is_mine_id(pdu.state_key)
+                ):
+                    logger.info(
+                        "Discarding PDU %s from invalid origin %s",
+                        pdu.event_id, transaction.origin
+                    )
+                    continue
+                else:
+                    logger.info(
+                        "Accepting join PDU %s from %s",
+                        pdu.event_id, transaction.origin
+                    )
+
             try:
                 yield self._handle_new_pdu(transaction.origin, pdu)
                 results.append({})
@@ -181,40 +209,76 @@ class FederationServer(FederationBase):
             except SynapseError as e:
                 logger.info("Failed to handle edu %r: %r", edu_type, e)
             except Exception as e:
-                logger.exception("Failed to handle edu %r", edu_type, e)
+                logger.exception("Failed to handle edu %r", edu_type)
         else:
             logger.warn("Received EDU of type %s with no handler", edu_type)
 
     @defer.inlineCallbacks
     @log_function
     def on_context_state_request(self, origin, room_id, event_id):
-        with (yield self._server_linearizer.queue((origin, room_id))):
-            if event_id:
-                pdus = yield self.handler.get_state_for_pdu(
-                    origin, room_id, event_id,
-                )
-                auth_chain = yield self.store.get_auth_chain(
-                    [pdu.event_id for pdu in pdus]
-                )
+        if not event_id:
+            raise NotImplementedError("Specify an event")
 
-                for event in auth_chain:
-                    # We sign these again because there was a bug where we
-                    # incorrectly signed things the first time round
-                    if self.hs.is_mine_id(event.event_id):
-                        event.signatures.update(
-                            compute_event_signature(
-                                event,
-                                self.hs.hostname,
-                                self.hs.config.signing_key[0]
-                            )
-                        )
-            else:
-                raise NotImplementedError("Specify an event")
+        in_room = yield self.auth.check_host_in_room(room_id, origin)
+        if not in_room:
+            raise AuthError(403, "Host not in room.")
+
+        result = self._state_resp_cache.get((room_id, event_id))
+        if not result:
+            with (yield self._server_linearizer.queue((origin, room_id))):
+                resp = yield self._state_resp_cache.set(
+                    (room_id, event_id),
+                    self._on_context_state_request_compute(room_id, event_id)
+                )
+        else:
+            resp = yield result
+
+        defer.returnValue((200, resp))
+
+    @defer.inlineCallbacks
+    def on_state_ids_request(self, origin, room_id, event_id):
+        if not event_id:
+            raise NotImplementedError("Specify an event")
+
+        in_room = yield self.auth.check_host_in_room(room_id, origin)
+        if not in_room:
+            raise AuthError(403, "Host not in room.")
+
+        state_ids = yield self.handler.get_state_ids_for_pdu(
+            room_id, event_id,
+        )
+        auth_chain_ids = yield self.store.get_auth_chain_ids(state_ids)
 
         defer.returnValue((200, {
+            "pdu_ids": state_ids,
+            "auth_chain_ids": auth_chain_ids,
+        }))
+
+    @defer.inlineCallbacks
+    def _on_context_state_request_compute(self, room_id, event_id):
+        pdus = yield self.handler.get_state_for_pdu(
+            room_id, event_id,
+        )
+        auth_chain = yield self.store.get_auth_chain(
+            [pdu.event_id for pdu in pdus]
+        )
+
+        for event in auth_chain:
+            # We sign these again because there was a bug where we
+            # incorrectly signed things the first time round
+            if self.hs.is_mine_id(event.event_id):
+                event.signatures.update(
+                    compute_event_signature(
+                        event,
+                        self.hs.hostname,
+                        self.hs.config.signing_key[0]
+                    )
+                )
+
+        defer.returnValue({
             "pdus": [pdu.get_pdu_json() for pdu in pdus],
             "auth_chain": [pdu.get_pdu_json() for pdu in auth_chain],
-        }))
+        })
 
     @defer.inlineCallbacks
     @log_function
@@ -348,27 +412,12 @@ class FederationServer(FederationBase):
             (200, send_content)
         )
 
-    @defer.inlineCallbacks
     @log_function
     def on_query_client_keys(self, origin, content):
-        query = []
-        for user_id, device_ids in content.get("device_keys", {}).items():
-            if not device_ids:
-                query.append((user_id, None))
-            else:
-                for device_id in device_ids:
-                    query.append((user_id, device_id))
+        return self.on_query_request("client_keys", content)
 
-        results = yield self.store.get_e2e_device_keys(query)
-
-        json_result = {}
-        for user_id, device_keys in results.items():
-            for device_id, json_bytes in device_keys.items():
-                json_result.setdefault(user_id, {})[device_id] = json.loads(
-                    json_bytes
-                )
-
-        defer.returnValue({"device_keys": json_result})
+    def on_query_user_devices(self, origin, user_id):
+        return self.on_query_request("user_devices", user_id)
 
     @defer.inlineCallbacks
     @log_function
@@ -400,6 +449,7 @@ class FederationServer(FederationBase):
                 " limit: %d, min_depth: %d",
                 earliest_events, latest_events, limit, min_depth
             )
+
             missing_events = yield self.handler.on_get_missing_events(
                 origin, room_id, earliest_events, latest_events, limit, min_depth
             )
@@ -449,6 +499,7 @@ class FederationServer(FederationBase):
     @defer.inlineCallbacks
     @log_function
     def _handle_new_pdu(self, origin, pdu, get_missing=True):
+
         # We reprocess pdus when we have seen them only as outliers
         existing = yield self._get_persisted_pdu(
             origin, pdu.event_id, do_auth=False
@@ -513,7 +564,16 @@ class FederationServer(FederationBase):
                 if get_missing and prevs - seen:
                     # If we're missing stuff, ensure we only fetch stuff one
                     # at a time.
+                    logger.info(
+                        "Acquiring lock for room %r to fetch %d missing events: %r...",
+                        pdu.room_id, len(prevs - seen), list(prevs - seen)[:5],
+                    )
                     with (yield self._room_pdu_linearizer.queue(pdu.room_id)):
+                        logger.info(
+                            "Acquired lock for room %r to fetch %d missing events",
+                            pdu.room_id, len(prevs - seen),
+                        )
+
                         # We recalculate seen, since it may have changed.
                         have_seen = yield self.store.have_events(prevs)
                         seen = set(have_seen.keys())
@@ -533,6 +593,25 @@ class FederationServer(FederationBase):
                                 len(prevs - seen), pdu.room_id, list(prevs - seen)[:5]
                             )
 
+                            # XXX: we set timeout to 10s to help workaround
+                            # https://github.com/matrix-org/synapse/issues/1733.
+                            # The reason is to avoid holding the linearizer lock
+                            # whilst processing inbound /send transactions, causing
+                            # FDs to stack up and block other inbound transactions
+                            # which empirically can currently take up to 30 minutes.
+                            #
+                            # N.B. this explicitly disables retry attempts.
+                            #
+                            # N.B. this also increases our chances of falling back to
+                            # fetching fresh state for the room if the missing event
+                            # can't be found, which slightly reduces our security.
+                            # it may also increase our DAG extremity count for the room,
+                            # causing additional state resolution?  See #1760.
+                            # However, fetching state doesn't hold the linearizer lock
+                            # apparently.
+                            #
+                            # see https://github.com/matrix-org/synapse/pull/1744
+
                             missing_events = yield self.get_missing_events(
                                 origin,
                                 pdu.room_id,
@@ -540,6 +619,7 @@ class FederationServer(FederationBase):
                                 latest_events=[pdu],
                                 limit=10,
                                 min_depth=min_depth,
+                                timeout=10000,
                             )
 
                             # We want to sort these by depth so we process them and
@@ -578,7 +658,7 @@ class FederationServer(FederationBase):
                     origin, pdu.room_id, pdu.event_id,
                 )
             except:
-                logger.warn("Failed to get state for event: %s", pdu.event_id)
+                logger.exception("Failed to get state for event: %s", pdu.event_id)
 
         yield self.handler.on_receive_pdu(
             origin,

synapse/federation/replication.py

@@ -20,8 +20,6 @@ a given transport.
 from .federation_client import FederationClient
 from .federation_server import FederationServer
 
-from .transaction_queue import TransactionQueue
-
 from .persistence import TransactionActions
 
 import logging
@@ -66,9 +64,6 @@ class ReplicationLayer(FederationClient, FederationServer):
         self._clock = hs.get_clock()
 
         self.transaction_actions = TransactionActions(self.store)
-        self._transaction_queue = TransactionQueue(hs, transport_layer)
-
-        self._order = 0
 
         self.hs = hs
 

synapse/federation/send_queue.py

@@ -0,0 +1,298 @@
+# -*- coding: utf-8 -*-
+# Copyright 2014-2016 OpenMarket Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""A federation sender that forwards things to be sent across replication to
+a worker process.
+
+It assumes there is a single worker process feeding off of it.
+
+Each row in the replication stream consists of a type and some json, where the
+types indicate whether they are presence, or edus, etc.
+
+Ephemeral or non-event data are queued up in-memory. When the worker requests
+updates since a particular point, all in-memory data since before that point is
+dropped. We also expire things in the queue after 5 minutes, to ensure that a
+dead worker doesn't cause the queues to grow limitlessly.
+
+Events are replicated via a separate events stream.
+"""
+
+from .units import Edu
+
+from synapse.util.metrics import Measure
+import synapse.metrics
+
+from blist import sorteddict
+import ujson
+
+
+metrics = synapse.metrics.get_metrics_for(__name__)
+
+
+PRESENCE_TYPE = "p"
+KEYED_EDU_TYPE = "k"
+EDU_TYPE = "e"
+FAILURE_TYPE = "f"
+DEVICE_MESSAGE_TYPE = "d"
+
+
+class FederationRemoteSendQueue(object):
+    """A drop in replacement for TransactionQueue"""
+
+    def __init__(self, hs):
+        self.server_name = hs.hostname
+        self.clock = hs.get_clock()
+
+        self.presence_map = {}
+        self.presence_changed = sorteddict()
+
+        self.keyed_edu = {}
+        self.keyed_edu_changed = sorteddict()
+
+        self.edus = sorteddict()
+
+        self.failures = sorteddict()
+
+        self.device_messages = sorteddict()
+
+        self.pos = 1
+        self.pos_time = sorteddict()
+
+        # EVERYTHING IS SAD. In particular, python only makes new scopes when
+        # we make a new function, so we need to make a new function so the inner
+        # lambda binds to the queue rather than to the name of the queue which
+        # changes. ARGH.
+        def register(name, queue):
+            metrics.register_callback(
+                queue_name + "_size",
+                lambda: len(queue),
+            )
+
+        for queue_name in [
+            "presence_map", "presence_changed", "keyed_edu", "keyed_edu_changed",
+            "edus", "failures", "device_messages", "pos_time",
+        ]:
+            register(queue_name, getattr(self, queue_name))
+
+        self.clock.looping_call(self._clear_queue, 30 * 1000)
+
+    def _next_pos(self):
+        pos = self.pos
+        self.pos += 1
+        self.pos_time[self.clock.time_msec()] = pos
+        return pos
+
+    def _clear_queue(self):
+        """Clear the queues for anything older than N minutes"""
+
+        FIVE_MINUTES_AGO = 5 * 60 * 1000
+        now = self.clock.time_msec()
+
+        keys = self.pos_time.keys()
+        time = keys.bisect_left(now - FIVE_MINUTES_AGO)
+        if not keys[:time]:
+            return
+
+        position_to_delete = max(keys[:time])
+        for key in keys[:time]:
+            del self.pos_time[key]
+
+        self._clear_queue_before_pos(position_to_delete)
+
+    def _clear_queue_before_pos(self, position_to_delete):
+        """Clear all the queues from before a given position"""
+        with Measure(self.clock, "send_queue._clear"):
+            # Delete things out of presence maps
+            keys = self.presence_changed.keys()
+            i = keys.bisect_left(position_to_delete)
+            for key in keys[:i]:
+                del self.presence_changed[key]
+
+            user_ids = set(
+                user_id for uids in self.presence_changed.values() for _, user_id in uids
+            )
+
+            to_del = [
+                user_id for user_id in self.presence_map if user_id not in user_ids
+            ]
+            for user_id in to_del:
+                del self.presence_map[user_id]
+
+            # Delete things out of keyed edus
+            keys = self.keyed_edu_changed.keys()
+            i = keys.bisect_left(position_to_delete)
+            for key in keys[:i]:
+                del self.keyed_edu_changed[key]
+
+            live_keys = set()
+            for edu_key in self.keyed_edu_changed.values():
+                live_keys.add(edu_key)
+
+            to_del = [edu_key for edu_key in self.keyed_edu if edu_key not in live_keys]
+            for edu_key in to_del:
+                del self.keyed_edu[edu_key]
+
+            # Delete things out of edu map
+            keys = self.edus.keys()
+            i = keys.bisect_left(position_to_delete)
+            for key in keys[:i]:
+                del self.edus[key]
+
+            # Delete things out of failure map
+            keys = self.failures.keys()
+            i = keys.bisect_left(position_to_delete)
+            for key in keys[:i]:
+                del self.failures[key]
+
+            # Delete things out of device map
+            keys = self.device_messages.keys()
+            i = keys.bisect_left(position_to_delete)
+            for key in keys[:i]:
+                del self.device_messages[key]
+
+    def notify_new_events(self, current_id):
+        """As per TransactionQueue"""
+        # We don't need to replicate this as it gets sent down a different
+        # stream.
+        pass
+
+    def send_edu(self, destination, edu_type, content, key=None):
+        """As per TransactionQueue"""
+        pos = self._next_pos()
+
+        edu = Edu(
+            origin=self.server_name,
+            destination=destination,
+            edu_type=edu_type,
+            content=content,
+        )
+
+        if key:
+            assert isinstance(key, tuple)
+            self.keyed_edu[(destination, key)] = edu
+            self.keyed_edu_changed[pos] = (destination, key)
+        else:
+            self.edus[pos] = edu
+
+    def send_presence(self, destination, states):
+        """As per TransactionQueue"""
+        pos = self._next_pos()
+
+        self.presence_map.update({
+            state.user_id: state
+            for state in states
+        })
+
+        self.presence_changed[pos] = [
+            (destination, state.user_id) for state in states
+        ]
+
+    def send_failure(self, failure, destination):
+        """As per TransactionQueue"""
+        pos = self._next_pos()
+
+        self.failures[pos] = (destination, str(failure))
+
+    def send_device_messages(self, destination):
+        """As per TransactionQueue"""
+        pos = self._next_pos()
+        self.device_messages[pos] = destination
+
+    def get_current_token(self):
+        return self.pos - 1
+
+    def get_replication_rows(self, token, limit, federation_ack=None):
+        """
+        Args:
+            token (int)
+            limit (int)
+            federation_ack (int): Optional. The position where the worker is
+                explicitly acknowledged it has handled. Allows us to drop
+                data from before that point
+        """
+        # TODO: Handle limit.
+
+        # To handle restarts where we wrap around
+        if token > self.pos:
+            token = -1
+
+        rows = []
+
+        # There should be only one reader, so lets delete everything its
+        # acknowledged its seen.
+        if federation_ack:
+            self._clear_queue_before_pos(federation_ack)
+
+        # Fetch changed presence
+        keys = self.presence_changed.keys()
+        i = keys.bisect_right(token)
+        dest_user_ids = set(
+            (pos, dest_user_id)
+            for pos in keys[i:]
+            for dest_user_id in self.presence_changed[pos]
+        )
+
+        for (key, (dest, user_id)) in dest_user_ids:
+            rows.append((key, PRESENCE_TYPE, ujson.dumps({
+                "destination": dest,
+                "state": self.presence_map[user_id].as_dict(),
+            })))
+
+        # Fetch changes keyed edus
+        keys = self.keyed_edu_changed.keys()
+        i = keys.bisect_right(token)
+        keyed_edus = set((k, self.keyed_edu_changed[k]) for k in keys[i:])
+
+        for (pos, (destination, edu_key)) in keyed_edus:
+            rows.append(
+                (pos, KEYED_EDU_TYPE, ujson.dumps({
+                    "key": edu_key,
+                    "edu": self.keyed_edu[(destination, edu_key)].get_internal_dict(),
+                }))
+            )
+
+        # Fetch changed edus
+        keys = self.edus.keys()
+        i = keys.bisect_right(token)
+        edus = set((k, self.edus[k]) for k in keys[i:])
+
+        for (pos, edu) in edus:
+            rows.append((pos, EDU_TYPE, ujson.dumps(edu.get_internal_dict())))
+
+        # Fetch changed failures
+        keys = self.failures.keys()
+        i = keys.bisect_right(token)
+        failures = set((k, self.failures[k]) for k in keys[i:])
+
+        for (pos, (destination, failure)) in failures:
+            rows.append((pos, FAILURE_TYPE, ujson.dumps({
+                "destination": destination,
+                "failure": failure,
+            })))
+
+        # Fetch changed device messages
+        keys = self.device_messages.keys()
+        i = keys.bisect_right(token)
+        device_messages = set((k, self.device_messages[k]) for k in keys[i:])
+
+        for (pos, destination) in device_messages:
+            rows.append((pos, DEVICE_MESSAGE_TYPE, ujson.dumps({
+                "destination": destination,
+            })))
+
+        # Sort rows based on pos
+        rows.sort()
+
+        return rows

synapse/federation/transaction_queue.py

@@ -17,15 +17,17 @@
 from twisted.internet import defer
 
 from .persistence import TransactionActions
-from .units import Transaction
+from .units import Transaction, Edu
 
 from synapse.api.errors import HttpResponseException
 from synapse.util.async import run_on_reactor
-from synapse.util.logutils import log_function
-from synapse.util.logcontext import PreserveLoggingContext
+from synapse.util.logcontext import preserve_context_over_fn
 from synapse.util.retryutils import (
     get_retry_limiter, NotRetryingDestination,
 )
+from synapse.util.metrics import measure_func
+from synapse.types import get_domain_from_id
+from synapse.handlers.presence import format_user_presence_state
 import synapse.metrics
 
 import logging
@@ -35,6 +37,12 @@ logger = logging.getLogger(__name__)
 
 metrics = synapse.metrics.get_metrics_for(__name__)
 
+client_metrics = synapse.metrics.get_metrics_for("synapse.federation.client")
+sent_pdus_destination_dist = client_metrics.register_distribution(
+    "sent_pdu_destinations"
+)
+sent_edus_counter = client_metrics.register_counter("sent_edus")
+
 
 class TransactionQueue(object):
     """This class makes sure we only have one transaction in flight at
@@ -43,15 +51,17 @@ class TransactionQueue(object):
     It batches pending PDUs into single transactions.
     """
 
-    def __init__(self, hs, transport_layer):
+    def __init__(self, hs):
         self.server_name = hs.hostname
 
         self.store = hs.get_datastore()
+        self.state = hs.get_state_handler()
         self.transaction_actions = TransactionActions(self.store)
 
-        self.transport_layer = transport_layer
+        self.transport_layer = hs.get_federation_transport_client()
 
-        self._clock = hs.get_clock()
+        self.clock = hs.get_clock()
+        self.is_mine_id = hs.is_mine_id
 
         # Is a mapping from destinations -> deferreds. Used to keep track
         # of which destinations have transactions in flight and when they are
@@ -69,20 +79,36 @@ class TransactionQueue(object):
         # destination -> list of tuple(edu, deferred)
         self.pending_edus_by_dest = edus = {}
 
+        # Presence needs to be separate as we send single aggragate EDUs
+        self.pending_presence_by_dest = presence = {}
+        self.pending_edus_keyed_by_dest = edus_keyed = {}
+
         metrics.register_callback(
             "pending_pdus",
             lambda: sum(map(len, pdus.values())),
         )
         metrics.register_callback(
             "pending_edus",
-            lambda: sum(map(len, edus.values())),
+            lambda: (
+                sum(map(len, edus.values()))
+                + sum(map(len, presence.values()))
+                + sum(map(len, edus_keyed.values()))
+            ),
         )
 
         # destination -> list of tuple(failure, deferred)
         self.pending_failures_by_dest = {}
 
+        self.last_device_stream_id_by_dest = {}
+        self.last_device_list_stream_id_by_dest = {}
+
         # HACK to get unique tx id
-        self._next_txn_id = int(self._clock.time_msec())
+        self._next_txn_id = int(self.clock.time_msec())
+
+        self._order = 1
+
+        self._is_processing = False
+        self._last_poked_id = -1
 
     def can_send_to(self, destination):
         """Can we send messages to the given server?
@@ -104,11 +130,76 @@ class TransactionQueue(object):
         else:
             return not destination.startswith("localhost")
 
-    def enqueue_pdu(self, pdu, destinations, order):
+    @defer.inlineCallbacks
+    def notify_new_events(self, current_id):
+        """This gets called when we have some new events we might want to
+        send out to other servers.
+        """
+        self._last_poked_id = max(current_id, self._last_poked_id)
+
+        if self._is_processing:
+            return
+
+        try:
+            self._is_processing = True
+            while True:
+                last_token = yield self.store.get_federation_out_pos("events")
+                next_token, events = yield self.store.get_all_new_events_stream(
+                    last_token, self._last_poked_id, limit=20,
+                )
+
+                logger.debug("Handling %s -> %s", last_token, next_token)
+
+                if not events and next_token >= self._last_poked_id:
+                    break
+
+                for event in events:
+                    # Only send events for this server.
+                    send_on_behalf_of = event.internal_metadata.get_send_on_behalf_of()
+                    is_mine = self.is_mine_id(event.event_id)
+                    if not is_mine and send_on_behalf_of is None:
+                        continue
+
+                    # Get the state from before the event.
+                    # We need to make sure that this is the state from before
+                    # the event and not from after it.
+                    # Otherwise if the last member on a server in a room is
+                    # banned then it won't receive the event because it won't
+                    # be in the room after the ban.
+                    users_in_room = yield self.state.get_current_user_in_room(
+                        event.room_id, latest_event_ids=[
+                            prev_id for prev_id, _ in event.prev_events
+                        ],
+                    )
+
+                    destinations = set(
+                        get_domain_from_id(user_id) for user_id in users_in_room
+                    )
+                    if send_on_behalf_of is not None:
+                        # If we are sending the event on behalf of another server
+                        # then it already has the event and there is no reason to
+                        # send the event to it.
+                        destinations.discard(send_on_behalf_of)
+
+                    logger.debug("Sending %s to %r", event, destinations)
+
+                    self._send_pdu(event, destinations)
+
+                yield self.store.update_federation_out_pos(
+                    "events", next_token
+                )
+
+        finally:
+            self._is_processing = False
+
+    def _send_pdu(self, pdu, destinations):
         # We loop through all destinations to see whether we already have
         # a transaction in progress. If we do, stick it in the pending_pdus
         # table and we'll get back to it later.
 
+        order = self._order
+        self._order += 1
+
         destinations = set(destinations)
         destinations = set(
             dest for dest in destinations if self.can_send_to(dest)
@@ -119,89 +210,84 @@ class TransactionQueue(object):
         if not destinations:
             return
 
-        deferreds = []
+        sent_pdus_destination_dist.inc_by(len(destinations))
 
         for destination in destinations:
-            deferred = defer.Deferred()
             self.pending_pdus_by_dest.setdefault(destination, []).append(
-                (pdu, deferred, order)
+                (pdu, order)
             )
 
-            def chain(failure):
-                if not deferred.called:
-                    deferred.errback(failure)
+            preserve_context_over_fn(
+                self._attempt_new_transaction, destination
+            )
 
-            def log_failure(f):
-                logger.warn("Failed to send pdu to %s: %s", destination, f.value)
+    def send_presence(self, destination, states):
+        if not self.can_send_to(destination):
+            return
 
-            deferred.addErrback(log_failure)
+        self.pending_presence_by_dest.setdefault(destination, {}).update({
+            state.user_id: state for state in states
+        })
 
-            with PreserveLoggingContext():
-                self._attempt_new_transaction(destination).addErrback(chain)
+        preserve_context_over_fn(
+            self._attempt_new_transaction, destination
+        )
 
-            deferreds.append(deferred)
-
-    # NO inlineCallbacks
-    def enqueue_edu(self, edu):
-        destination = edu.destination
+    def send_edu(self, destination, edu_type, content, key=None):
+        edu = Edu(
+            origin=self.server_name,
+            destination=destination,
+            edu_type=edu_type,
+            content=content,
+        )
 
         if not self.can_send_to(destination):
             return
 
-        deferred = defer.Deferred()
-        self.pending_edus_by_dest.setdefault(destination, []).append(
-            (edu, deferred)
+        sent_edus_counter.inc()
+
+        if key:
+            self.pending_edus_keyed_by_dest.setdefault(
+                destination, {}
+            )[(edu.edu_type, key)] = edu
+        else:
+            self.pending_edus_by_dest.setdefault(destination, []).append(edu)
+
+        preserve_context_over_fn(
+            self._attempt_new_transaction, destination
         )
 
-        def chain(failure):
-            if not deferred.called:
-                deferred.errback(failure)
-
-        def log_failure(f):
-            logger.warn("Failed to send edu to %s: %s", destination, f.value)
-
-        deferred.addErrback(log_failure)
-
-        with PreserveLoggingContext():
-            self._attempt_new_transaction(destination).addErrback(chain)
-
-        return deferred
-
-    @defer.inlineCallbacks
-    def enqueue_failure(self, failure, destination):
+    def send_failure(self, failure, destination):
         if destination == self.server_name or destination == "localhost":
             return
 
-        deferred = defer.Deferred()
-
         if not self.can_send_to(destination):
             return
 
         self.pending_failures_by_dest.setdefault(
             destination, []
-        ).append(
-            (failure, deferred)
+        ).append(failure)
+
+        preserve_context_over_fn(
+            self._attempt_new_transaction, destination
         )
 
-        def chain(f):
-            if not deferred.called:
-                deferred.errback(f)
+    def send_device_messages(self, destination):
+        if destination == self.server_name or destination == "localhost":
+            return
 
-        def log_failure(f):
-            logger.warn("Failed to send failure to %s: %s", destination, f.value)
+        if not self.can_send_to(destination):
+            return
 
-        deferred.addErrback(log_failure)
+        preserve_context_over_fn(
+            self._attempt_new_transaction, destination
+        )
 
-        with PreserveLoggingContext():
-            self._attempt_new_transaction(destination).addErrback(chain)
-
-        yield deferred
+    def get_current_token(self):
+        return 0
 
     @defer.inlineCallbacks
-    @log_function
     def _attempt_new_transaction(self, destination):
-        yield run_on_reactor()
-
         # list of (pending_pdu, deferred, order)
         if destination in self.pending_transactions:
             # XXX: pending_transactions can get stuck on by a never-ending
@@ -214,55 +300,154 @@ class TransactionQueue(object):
             )
             return
 
-        pending_pdus = self.pending_pdus_by_dest.pop(destination, [])
-        pending_edus = self.pending_edus_by_dest.pop(destination, [])
-        pending_failures = self.pending_failures_by_dest.pop(destination, [])
-
-        if pending_pdus:
-            logger.debug("TX [%s] len(pending_pdus_by_dest[dest]) = %d",
-                         destination, len(pending_pdus))
-
-        if not pending_pdus and not pending_edus and not pending_failures:
-            logger.debug("TX [%s] Nothing to send", destination)
-            return
-
         try:
             self.pending_transactions[destination] = 1
 
+            yield run_on_reactor()
+
+            while True:
+                pending_pdus = self.pending_pdus_by_dest.pop(destination, [])
+                pending_edus = self.pending_edus_by_dest.pop(destination, [])
+                pending_presence = self.pending_presence_by_dest.pop(destination, {})
+                pending_failures = self.pending_failures_by_dest.pop(destination, [])
+
+                pending_edus.extend(
+                    self.pending_edus_keyed_by_dest.pop(destination, {}).values()
+                )
+
+                limiter = yield get_retry_limiter(
+                    destination,
+                    self.clock,
+                    self.store,
+                    backoff_on_404=True,  # If we get a 404 the other side has gone
+                )
+
+                device_message_edus, device_stream_id, dev_list_id = (
+                    yield self._get_new_device_messages(destination)
+                )
+
+                pending_edus.extend(device_message_edus)
+                if pending_presence:
+                    pending_edus.append(
+                        Edu(
+                            origin=self.server_name,
+                            destination=destination,
+                            edu_type="m.presence",
+                            content={
+                                "push": [
+                                    format_user_presence_state(
+                                        presence, self.clock.time_msec()
+                                    )
+                                    for presence in pending_presence.values()
+                                ]
+                            },
+                        )
+                    )
+
+                if pending_pdus:
+                    logger.debug("TX [%s] len(pending_pdus_by_dest[dest]) = %d",
+                                 destination, len(pending_pdus))
+
+                if not pending_pdus and not pending_edus and not pending_failures:
+                    logger.debug("TX [%s] Nothing to send", destination)
+                    self.last_device_stream_id_by_dest[destination] = (
+                        device_stream_id
+                    )
+                    return
+
+                success = yield self._send_new_transaction(
+                    destination, pending_pdus, pending_edus, pending_failures,
+                    limiter=limiter,
+                )
+                if success:
+                    # Remove the acknowledged device messages from the database
+                    # Only bother if we actually sent some device messages
+                    if device_message_edus:
+                        yield self.store.delete_device_msgs_for_remote(
+                            destination, device_stream_id
+                        )
+                        logger.info("Marking as sent %r %r", destination, dev_list_id)
+                        yield self.store.mark_as_sent_devices_by_remote(
+                            destination, dev_list_id
+                        )
+
+                    self.last_device_stream_id_by_dest[destination] = device_stream_id
+                    self.last_device_list_stream_id_by_dest[destination] = dev_list_id
+                else:
+                    break
+        except NotRetryingDestination:
+            logger.debug(
+                "TX [%s] not ready for retry yet - "
+                "dropping transaction for now",
+                destination,
+            )
+        finally:
+            # We want to be *very* sure we delete this after we stop processing
+            self.pending_transactions.pop(destination, None)
+
+    @defer.inlineCallbacks
+    def _get_new_device_messages(self, destination):
+        last_device_stream_id = self.last_device_stream_id_by_dest.get(destination, 0)
+        to_device_stream_id = self.store.get_to_device_stream_token()
+        contents, stream_id = yield self.store.get_new_device_msgs_for_remote(
+            destination, last_device_stream_id, to_device_stream_id
+        )
+        edus = [
+            Edu(
+                origin=self.server_name,
+                destination=destination,
+                edu_type="m.direct_to_device",
+                content=content,
+            )
+            for content in contents
+        ]
+
+        last_device_list = self.last_device_list_stream_id_by_dest.get(destination, 0)
+        now_stream_id, results = yield self.store.get_devices_by_remote(
+            destination, last_device_list
+        )
+        edus.extend(
+            Edu(
+                origin=self.server_name,
+                destination=destination,
+                edu_type="m.device_list_update",
+                content=content,
+            )
+            for content in results
+        )
+        defer.returnValue((edus, stream_id, now_stream_id))
+
+    @measure_func("_send_new_transaction")
+    @defer.inlineCallbacks
+    def _send_new_transaction(self, destination, pending_pdus, pending_edus,
+                              pending_failures, limiter):
+
+        # Sort based on the order field
+        pending_pdus.sort(key=lambda t: t[1])
+        pdus = [x[0] for x in pending_pdus]
+        edus = pending_edus
+        failures = [x.get_dict() for x in pending_failures]
+
+        success = True
+
+        try:
             logger.debug("TX [%s] _attempt_new_transaction", destination)
 
-            # Sort based on the order field
-            pending_pdus.sort(key=lambda t: t[2])
-
-            pdus = [x[0] for x in pending_pdus]
-            edus = [x[0] for x in pending_edus]
-            failures = [x[0].get_dict() for x in pending_failures]
-            deferreds = [
-                x[1]
-                for x in pending_pdus + pending_edus + pending_failures
-            ]
-
             txn_id = str(self._next_txn_id)
 
-            limiter = yield get_retry_limiter(
-                destination,
-                self._clock,
-                self.store,
-            )
-
             logger.debug(
                 "TX [%s] {%s} Attempting new transaction"
                 " (pdus: %d, edus: %d, failures: %d)",
                 destination, txn_id,
-                len(pending_pdus),
-                len(pending_edus),
-                len(pending_failures)
+                len(pdus),
+                len(edus),
+                len(failures)
             )
 
             logger.debug("TX [%s] Persisting transaction...", destination)
 
             transaction = Transaction.create_new(
-                origin_server_ts=int(self._clock.time_msec()),
+                origin_server_ts=int(self.clock.time_msec()),
                 transaction_id=txn_id,
                 origin=self.server_name,
                 destination=destination,
@@ -281,9 +466,9 @@ class TransactionQueue(object):
                 " (PDUs: %d, EDUs: %d, failures: %d)",
                 destination, txn_id,
                 transaction.transaction_id,
-                len(pending_pdus),
-                len(pending_edus),
-                len(pending_failures),
+                len(pdus),
+                len(edus),
+                len(failures),
             )
 
             with limiter:
@@ -293,7 +478,7 @@ class TransactionQueue(object):
                 # keys work
                 def json_data_cb():
                     data = transaction.get_dict()
-                    now = int(self._clock.time_msec())
+                    now = int(self.clock.time_msec())
                     if "pdus" in data:
                         for p in data["pdus"]:
                             if "age_ts" in p:
@@ -319,6 +504,13 @@ class TransactionQueue(object):
                     code = e.code
                     response = e.response
 
+                    if e.code in (401, 404, 429) or 500 <= e.code:
+                        logger.info(
+                            "TX [%s] {%s} got %d response",
+                            destination, txn_id, code
+                        )
+                        raise e
+
                 logger.info(
                     "TX [%s] {%s} got %d response",
                     destination, txn_id, code
@@ -333,28 +525,12 @@ class TransactionQueue(object):
 
             logger.debug("TX [%s] Marked as delivered", destination)
 
-            logger.debug("TX [%s] Yielding to callbacks...", destination)
-
-            for deferred in deferreds:
-                if code == 200:
-                    deferred.callback(None)
-                else:
-                    deferred.errback(RuntimeError("Got status %d" % code))
-
-                # Ensures we don't continue until all callbacks on that
-                # deferred have fired
-                try:
-                    yield deferred
-                except:
-                    pass
-
-            logger.debug("TX [%s] Yielded to callbacks", destination)
-        except NotRetryingDestination:
-            logger.info(
-                "TX [%s] not ready for retry yet - "
-                "dropping transaction for now",
-                destination,
-            )
+            if code != 200:
+                for p in pdus:
+                    logger.info(
+                        "Failed to send event %s to %s", p.event_id, destination
+                    )
+                success = False
         except RuntimeError as e:
             # We capture this here as there as nothing actually listens
             # for this finishing functions deferred.
@@ -363,6 +539,11 @@ class TransactionQueue(object):
                 destination,
                 e,
             )
+
+            success = False
+
+            for p in pdus:
+                logger.info("Failed to send event %s to %s", p.event_id, destination)
         except Exception as e:
             # We capture this here as there as nothing actually listens
             # for this finishing functions deferred.
@@ -372,13 +553,9 @@ class TransactionQueue(object):
                 e,
             )
 
-            for deferred in deferreds:
-                if not deferred.called:
-                    deferred.errback(e)
+            success = False
 
-        finally:
-            # We want to be *very* sure we delete this after we stop processing
-            self.pending_transactions.pop(destination, None)
+            for p in pdus:
+                logger.info("Failed to send event %s to %s", p.event_id, destination)
 
-            # Check to see if there is anything else to send.
-            self._attempt_new_transaction(destination)
+        defer.returnValue(success)

synapse/federation/transport/client.py

@@ -54,6 +54,28 @@ class TransportLayerClient(object):
             destination, path=path, args={"event_id": event_id},
         )
 
+    @log_function
+    def get_room_state_ids(self, destination, room_id, event_id):
+        """ Requests all state for a given room from the given server at the
+        given event. Returns the state's event_id's
+
+        Args:
+            destination (str): The host name of the remote home server we want
+                to get the state from.
+            context (str): The name of the context we want the state of
+            event_id (str): The event we want the context at.
+
+        Returns:
+            Deferred: Results in a dict received from the remote homeserver.
+        """
+        logger.debug("get_room_state_ids dest=%s, room=%s",
+                     destination, room_id)
+
+        path = PREFIX + "/state_ids/%s/" % room_id
+        return self.client.get_json(
+            destination, path=path, args={"event_id": event_id},
+        )
+
     @log_function
     def get_event(self, destination, event_id, timeout=None):
         """ Requests the pdu with give id and origin from the given server.
@@ -226,12 +248,27 @@ class TransportLayerClient(object):
 
     @defer.inlineCallbacks
     @log_function
-    def get_public_rooms(self, remote_server):
+    def get_public_rooms(self, remote_server, limit, since_token,
+                         search_filter=None, include_all_networks=False,
+                         third_party_instance_id=None):
         path = PREFIX + "/publicRooms"
 
+        args = {
+            "include_all_networks": "true" if include_all_networks else "false",
+        }
+        if third_party_instance_id:
+            args["third_party_instance_id"] = third_party_instance_id,
+        if limit:
+            args["limit"] = [str(limit)]
+        if since_token:
+            args["since"] = [since_token]
+
+        # TODO(erikj): Actually send the search_filter across federation.
+
         response = yield self.client.get_json(
             destination=remote_server,
             path=path,
+            args=args,
         )
 
         defer.returnValue(response)
@@ -276,7 +313,7 @@ class TransportLayerClient(object):
 
     @defer.inlineCallbacks
     @log_function
-    def query_client_keys(self, destination, query_content):
+    def query_client_keys(self, destination, query_content, timeout):
         """Query the device keys for a list of user ids hosted on a remote
         server.
 
@@ -305,12 +342,39 @@ class TransportLayerClient(object):
             destination=destination,
             path=path,
             data=query_content,
+            timeout=timeout,
         )
         defer.returnValue(content)
 
     @defer.inlineCallbacks
     @log_function
-    def claim_client_keys(self, destination, query_content):
+    def query_user_devices(self, destination, user_id, timeout):
+        """Query the devices for a user id hosted on a remote server.
+
+        Response:
+            {
+              "stream_id": "...",
+              "devices": [ { ... } ]
+            }
+
+        Args:
+            destination(str): The server to query.
+            query_content(dict): The user ids to query.
+        Returns:
+            A dict containg the device keys.
+        """
+        path = PREFIX + "/user/devices/" + user_id
+
+        content = yield self.client.get_json(
+            destination=destination,
+            path=path,
+            timeout=timeout,
+        )
+        defer.returnValue(content)
+
+    @defer.inlineCallbacks
+    @log_function
+    def claim_client_keys(self, destination, query_content, timeout):
         """Claim one-time keys for a list of devices hosted on a remote server.
 
         Request:
@@ -341,13 +405,14 @@ class TransportLayerClient(object):
             destination=destination,
             path=path,
             data=query_content,
+            timeout=timeout,
         )
         defer.returnValue(content)
 
     @defer.inlineCallbacks
     @log_function
     def get_missing_events(self, destination, room_id, earliest_events,
-                           latest_events, limit, min_depth):
+                           latest_events, limit, min_depth, timeout):
         path = PREFIX + "/get_missing_events/%s" % (room_id,)
 
         content = yield self.client.post_json(
@@ -358,7 +423,8 @@ class TransportLayerClient(object):
                 "min_depth": int(min_depth),
                 "earliest_events": earliest_events,
                 "latest_events": latest_events,
-            }
+            },
+            timeout=timeout,
         )
 
         defer.returnValue(content)

synapse/federation/transport/server.py

@@ -18,13 +18,18 @@ from twisted.internet import defer
 from synapse.api.urls import FEDERATION_PREFIX as PREFIX
 from synapse.api.errors import Codes, SynapseError
 from synapse.http.server import JsonResource
-from synapse.http.servlet import parse_json_object_from_request, parse_string
+from synapse.http.servlet import (
+    parse_json_object_from_request, parse_integer_from_args, parse_string_from_args,
+    parse_boolean_from_args,
+)
 from synapse.util.ratelimitutils import FederationRateLimiter
+from synapse.util.versionstring import get_version_string
+from synapse.types import ThirdPartyInstanceID
 
 import functools
 import logging
-import simplejson as json
 import re
+import synapse
 
 
 logger = logging.getLogger(__name__)
@@ -60,6 +65,16 @@ class TransportLayerServer(JsonResource):
         )
 
 
+class AuthenticationError(SynapseError):
+    """There was a problem authenticating the request"""
+    pass
+
+
+class NoAuthenticationError(AuthenticationError):
+    """The request had no authentication information"""
+    pass
+
+
 class Authenticator(object):
     def __init__(self, hs):
         self.keyring = hs.get_keyring()
@@ -67,7 +82,7 @@ class Authenticator(object):
 
     # A method just so we can pass 'self' as the authenticator to the Servlets
     @defer.inlineCallbacks
-    def authenticate_request(self, request):
+    def authenticate_request(self, request, content):
         json_request = {
             "method": request.method,
             "uri": request.uri,
@@ -75,17 +90,10 @@ class Authenticator(object):
             "signatures": {},
         }
 
-        content = None
-        origin = None
+        if content is not None:
+            json_request["content"] = content
 
-        if request.method in ["PUT", "POST"]:
-            # TODO: Handle other method types? other content types?
-            try:
-                content_bytes = request.content.read()
-                content = json.loads(content_bytes)
-                json_request["content"] = content
-            except:
-                raise SynapseError(400, "Unable to parse JSON", Codes.BAD_JSON)
+        origin = None
 
         def parse_auth_header(header_str):
             try:
@@ -103,14 +111,14 @@ class Authenticator(object):
                 sig = strip_quotes(param_dict["sig"])
                 return (origin, key, sig)
             except:
-                raise SynapseError(
+                raise AuthenticationError(
                     400, "Malformed Authorization header", Codes.UNAUTHORIZED
                 )
 
         auth_headers = request.requestHeaders.getRawHeaders(b"Authorization")
 
         if not auth_headers:
-            raise SynapseError(
+            raise NoAuthenticationError(
                 401, "Missing Authorization headers", Codes.UNAUTHORIZED,
             )
 
@@ -121,7 +129,7 @@ class Authenticator(object):
                 json_request["signatures"].setdefault(origin, {})[key] = sig
 
         if not json_request["signatures"]:
-            raise SynapseError(
+            raise NoAuthenticationError(
                 401, "Missing Authorization headers", Codes.UNAUTHORIZED,
             )
 
@@ -130,10 +138,12 @@ class Authenticator(object):
         logger.info("Request from %s", origin)
         request.authenticated_entity = origin
 
-        defer.returnValue((origin, content))
+        defer.returnValue(origin)
 
 
 class BaseFederationServlet(object):
+    REQUIRE_AUTH = True
+
     def __init__(self, handler, authenticator, ratelimiter, server_name,
                  room_list_handler):
         self.handler = handler
@@ -141,29 +151,46 @@ class BaseFederationServlet(object):
         self.ratelimiter = ratelimiter
         self.room_list_handler = room_list_handler
 
-    def _wrap(self, code):
+    def _wrap(self, func):
         authenticator = self.authenticator
         ratelimiter = self.ratelimiter
 
         @defer.inlineCallbacks
-        @functools.wraps(code)
-        def new_code(request, *args, **kwargs):
+        @functools.wraps(func)
+        def new_func(request, *args, **kwargs):
+            content = None
+            if request.method in ["PUT", "POST"]:
+                # TODO: Handle other method types? other content types?
+                content = parse_json_object_from_request(request)
+
             try:
-                (origin, content) = yield authenticator.authenticate_request(request)
-                with ratelimiter.ratelimit(origin) as d:
-                    yield d
-                    response = yield code(
-                        origin, content, request.args, *args, **kwargs
-                    )
+                origin = yield authenticator.authenticate_request(request, content)
+            except NoAuthenticationError:
+                origin = None
+                if self.REQUIRE_AUTH:
+                    logger.exception("authenticate_request failed")
+                    raise
             except:
                 logger.exception("authenticate_request failed")
                 raise
+
+            if origin:
+                with ratelimiter.ratelimit(origin) as d:
+                    yield d
+                    response = yield func(
+                        origin, content, request.args, *args, **kwargs
+                    )
+            else:
+                response = yield func(
+                    origin, content, request.args, *args, **kwargs
+                )
+
             defer.returnValue(response)
 
         # Extra logic that functools.wraps() doesn't finish
-        new_code.__self__ = code.__self__
+        new_func.__self__ = func.__self__
 
-        return new_code
+        return new_func
 
     def register(self, server):
         pattern = re.compile("^" + PREFIX + self.PATH + "$")
@@ -271,6 +298,17 @@ class FederationStateServlet(BaseFederationServlet):
         )
 
 
+class FederationStateIdsServlet(BaseFederationServlet):
+    PATH = "/state_ids/(?P<room_id>[^/]*)/"
+
+    def on_GET(self, origin, content, query, room_id):
+        return self.handler.on_state_ids_request(
+            origin,
+            room_id,
+            query.get("event_id", [None])[0],
+        )
+
+
 class FederationBackfillServlet(BaseFederationServlet):
     PATH = "/backfill/(?P<context>[^/]*)/"
 
@@ -367,10 +405,15 @@ class FederationThirdPartyInviteExchangeServlet(BaseFederationServlet):
 class FederationClientKeysQueryServlet(BaseFederationServlet):
     PATH = "/user/keys/query"
 
-    @defer.inlineCallbacks
     def on_POST(self, origin, content, query):
-        response = yield self.handler.on_query_client_keys(origin, content)
-        defer.returnValue((200, response))
+        return self.handler.on_query_client_keys(origin, content)
+
+
+class FederationUserDevicesQueryServlet(BaseFederationServlet):
+    PATH = "/user/devices/(?P<user_id>[^/]*)"
+
+    def on_GET(self, origin, content, query, user_id):
+        return self.handler.on_query_user_devices(origin, user_id)
 
 
 class FederationClientKeysClaimServlet(BaseFederationServlet):
@@ -420,9 +463,10 @@ class FederationGetMissingEventsServlet(BaseFederationServlet):
 class On3pidBindServlet(BaseFederationServlet):
     PATH = "/3pid/onbind"
 
+    REQUIRE_AUTH = False
+
     @defer.inlineCallbacks
-    def on_POST(self, request):
-        content = parse_json_object_from_request(request)
+    def on_POST(self, origin, content, query):
         if "invites" in content:
             last_exception = None
             for invite in content["invites"]:
@@ -444,11 +488,6 @@ class On3pidBindServlet(BaseFederationServlet):
                 raise last_exception
         defer.returnValue((200, {}))
 
-    # Avoid doing remote HS authorization checks which are done by default by
-    # BaseFederationServlet.
-    def _wrap(self, code):
-        return code
-
 
 class OpenIdUserInfo(BaseFederationServlet):
     """
@@ -469,9 +508,11 @@ class OpenIdUserInfo(BaseFederationServlet):
 
     PATH = "/openid/userinfo"
 
+    REQUIRE_AUTH = False
+
     @defer.inlineCallbacks
-    def on_GET(self, request):
-        token = parse_string(request, "access_token")
+    def on_GET(self, origin, content, query):
+        token = query.get("access_token", [None])[0]
         if token is None:
             defer.returnValue((401, {
                 "errcode": "M_MISSING_TOKEN", "error": "Access Token required"
@@ -488,11 +529,6 @@ class OpenIdUserInfo(BaseFederationServlet):
 
         defer.returnValue((200, {"sub": user_id}))
 
-    # Avoid doing remote HS authorization checks which are done by default by
-    # BaseFederationServlet.
-    def _wrap(self, code):
-        return code
-
 
 class PublicRoomList(BaseFederationServlet):
     """
@@ -529,15 +565,49 @@ class PublicRoomList(BaseFederationServlet):
 
     @defer.inlineCallbacks
     def on_GET(self, origin, content, query):
-        data = yield self.room_list_handler.get_local_public_room_list()
+        limit = parse_integer_from_args(query, "limit", 0)
+        since_token = parse_string_from_args(query, "since", None)
+        include_all_networks = parse_boolean_from_args(
+            query, "include_all_networks", False
+        )
+        third_party_instance_id = parse_string_from_args(
+            query, "third_party_instance_id", None
+        )
+
+        if include_all_networks:
+            network_tuple = None
+        elif third_party_instance_id:
+            network_tuple = ThirdPartyInstanceID.from_string(third_party_instance_id)
+        else:
+            network_tuple = ThirdPartyInstanceID(None, None)
+
+        data = yield self.room_list_handler.get_local_public_room_list(
+            limit, since_token,
+            network_tuple=network_tuple
+        )
         defer.returnValue((200, data))
 
 
+class FederationVersionServlet(BaseFederationServlet):
+    PATH = "/version"
+
+    REQUIRE_AUTH = False
+
+    def on_GET(self, origin, content, query):
+        return defer.succeed((200, {
+            "server": {
+                "name": "Synapse",
+                "version": get_version_string(synapse)
+            },
+        }))
+
+
 SERVLET_CLASSES = (
     FederationSendServlet,
     FederationPullServlet,
     FederationEventServlet,
     FederationStateServlet,
+    FederationStateIdsServlet,
     FederationBackfillServlet,
     FederationQueryServlet,
     FederationMakeJoinServlet,
@@ -550,11 +620,13 @@ SERVLET_CLASSES = (
     FederationGetMissingEventsServlet,
     FederationEventAuthServlet,
     FederationClientKeysQueryServlet,
+    FederationUserDevicesQueryServlet,
     FederationClientKeysClaimServlet,
     FederationThirdPartyInviteExchangeServlet,
     On3pidBindServlet,
     OpenIdUserInfo,
     PublicRoomList,
+    FederationVersionServlet,
 )
 
 

synapse/handlers/__init__.py

@@ -19,22 +19,31 @@ from .room import (
 )
 from .room_member import RoomMemberHandler
 from .message import MessageHandler
-from .events import EventStreamHandler, EventHandler
 from .federation import FederationHandler
 from .profile import ProfileHandler
 from .directory import DirectoryHandler
 from .admin import AdminHandler
 from .identity import IdentityHandler
-from .receipts import ReceiptsHandler
 from .search import SearchHandler
 
 
 class Handlers(object):
 
-    """ A collection of all the event handlers.
+    """ Deprecated. A collection of handlers.
 
-    There's no need to lazily create these; we'll just make them all eagerly
-    at construction time.
+    At some point most of the classes whose name ended "Handler" were
+    accessed through this class.
+
+    However this makes it painful to unit test the handlers and to run cut
+    down versions of synapse that only use specific handlers because using a
+    single handler required creating all of the handlers. So some of the
+    handlers have been lifted out of the Handlers object and are now accessed
+    directly through the homeserver object itself.
+
+    Any new handlers should follow the new pattern of being accessed through
+    the homeserver object and should not be added to the Handlers object.
+
+    The remaining handlers should be moved out of the handlers object.
     """
 
     def __init__(self, hs):
@@ -42,13 +51,10 @@ class Handlers(object):
         self.message_handler = MessageHandler(hs)
         self.room_creation_handler = RoomCreationHandler(hs)
         self.room_member_handler = RoomMemberHandler(hs)
-        self.event_stream_handler = EventStreamHandler(hs)
-        self.event_handler = EventHandler(hs)
         self.federation_handler = FederationHandler(hs)
         self.profile_handler = ProfileHandler(hs)
         self.directory_handler = DirectoryHandler(hs)
         self.admin_handler = AdminHandler(hs)
-        self.receipts_handler = ReceiptsHandler(hs)
         self.identity_handler = IdentityHandler(hs)
         self.search_handler = SearchHandler(hs)
         self.room_context_handler = RoomContextHandler(hs)

synapse/handlers/_base.py

@@ -13,14 +13,14 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import logging
+
 from twisted.internet import defer
 
-from synapse.api.errors import LimitExceededError
+import synapse.types
 from synapse.api.constants import Membership, EventTypes
-from synapse.types import UserID, Requester
-
-
-import logging
+from synapse.api.errors import LimitExceededError
+from synapse.types import UserID
 
 
 logger = logging.getLogger(__name__)
@@ -31,11 +31,15 @@ class BaseHandler(object):
     Common base class for the event handlers.
 
     Attributes:
-        store (synapse.storage.events.StateStore):
+        store (synapse.storage.DataStore):
         state_handler (synapse.state.StateHandler):
     """
 
     def __init__(self, hs):
+        """
+        Args:
+            hs (synapse.server.HomeServer):
+        """
         self.store = hs.get_datastore()
         self.auth = hs.get_auth()
         self.notifier = hs.get_notifier()
@@ -51,8 +55,20 @@ class BaseHandler(object):
 
     def ratelimit(self, requester):
         time_now = self.clock.time()
+        user_id = requester.user.to_string()
+
+        # The AS user itself is never rate limited.
+        app_service = self.store.get_app_service_by_user_id(user_id)
+        if app_service is not None:
+            return  # do not ratelimit app service senders
+
+        # Disable rate limiting of users belonging to any AS that is configured
+        # not to be rate limited in its registration file (rate_limited: true|false).
+        if requester.app_service and not requester.app_service.is_rate_limited():
+            return
+
         allowed, time_allowed = self.ratelimiter.send_message(
-            requester.user.to_string(), time_now,
+            user_id, time_now,
             msg_rate_hz=self.hs.config.rc_messages_per_second,
             burst_count=self.hs.config.rc_message_burst_count,
         )
@@ -61,33 +77,25 @@ class BaseHandler(object):
                 retry_after_ms=int(1000 * (time_allowed - time_now)),
             )
 
-    def is_host_in_room(self, current_state):
-        room_members = [
-            (state_key, event.membership)
-            for ((event_type, state_key), event) in current_state.items()
-            if event_type == EventTypes.Member
-        ]
-        if len(room_members) == 0:
-            # Have we just created the room, and is this about to be the very
-            # first member event?
-            create_event = current_state.get(("m.room.create", ""))
-            if create_event:
-                return True
-        for (state_key, membership) in room_members:
-            if (
-                self.hs.is_mine_id(state_key)
-                and membership == Membership.JOIN
-            ):
-                return True
-        return False
-
     @defer.inlineCallbacks
-    def maybe_kick_guest_users(self, event, current_state):
+    def maybe_kick_guest_users(self, event, context=None):
         # Technically this function invalidates current_state by changing it.
         # Hopefully this isn't that important to the caller.
         if event.type == EventTypes.GuestAccess:
             guest_access = event.content.get("guest_access", "forbidden")
             if guest_access != "can_join":
+                if context:
+                    current_state = yield self.store.get_events(
+                        context.current_state_ids.values()
+                    )
+                else:
+                    current_state = yield self.state_handler.get_current_state(
+                        event.room_id
+                    )
+
+                current_state = current_state.values()
+
+                logger.info("maybe_kick_guest_users %r", current_state)
                 yield self.kick_guest_users(current_state)
 
     @defer.inlineCallbacks
@@ -120,7 +128,8 @@ class BaseHandler(object):
                 # and having homeservers have their own users leave keeps more
                 # of that decision-making and control local to the guest-having
                 # homeserver.
-                requester = Requester(target_user, "", True)
+                requester = synapse.types.create_requester(
+                    target_user, is_guest=True)
                 handler = self.hs.get_handlers().room_member_handler
                 yield handler.update_membership(
                     requester,

synapse/handlers/appservice.py

@@ -16,7 +16,8 @@
 from twisted.internet import defer
 
 from synapse.api.constants import EventTypes
-from synapse.appservice import ApplicationService
+from synapse.util.metrics import Measure
+from synapse.util.logcontext import preserve_fn, preserve_context_over_deferred
 
 import logging
 
@@ -42,36 +43,73 @@ class ApplicationServicesHandler(object):
         self.appservice_api = hs.get_application_service_api()
         self.scheduler = hs.get_application_service_scheduler()
         self.started_scheduler = False
+        self.clock = hs.get_clock()
+        self.notify_appservices = hs.config.notify_appservices
+
+        self.current_max = 0
+        self.is_processing = False
 
     @defer.inlineCallbacks
-    def notify_interested_services(self, event):
+    def notify_interested_services(self, current_id):
         """Notifies (pushes) all application services interested in this event.
 
         Pushing is done asynchronously, so this method won't block for any
         prolonged length of time.
 
         Args:
-            event(Event): The event to push out to interested services.
+            current_id(int): The current maximum ID.
         """
-        # Gather interested services
-        services = yield self._get_services_for_event(event)
-        if len(services) == 0:
-            return  # no services need notifying
+        services = self.store.get_app_services()
+        if not services or not self.notify_appservices:
+            return
 
-        # Do we know this user exists? If not, poke the user query API for
-        # all services which match that user regex. This needs to block as these
-        # user queries need to be made BEFORE pushing the event.
-        yield self._check_user_exists(event.sender)
-        if event.type == EventTypes.Member:
-            yield self._check_user_exists(event.state_key)
+        self.current_max = max(self.current_max, current_id)
+        if self.is_processing:
+            return
 
-        if not self.started_scheduler:
-            self.scheduler.start().addErrback(log_failure)
-            self.started_scheduler = True
+        with Measure(self.clock, "notify_interested_services"):
+            self.is_processing = True
+            try:
+                upper_bound = self.current_max
+                limit = 100
+                while True:
+                    upper_bound, events = yield self.store.get_new_events_for_appservice(
+                        upper_bound, limit
+                    )
 
-        # Fork off pushes to these services
-        for service in services:
-            self.scheduler.submit_event_for_as(service, event)
+                    if not events:
+                        break
+
+                    for event in events:
+                        # Gather interested services
+                        services = yield self._get_services_for_event(event)
+                        if len(services) == 0:
+                            continue  # no services need notifying
+
+                        # Do we know this user exists? If not, poke the user
+                        # query API for all services which match that user regex.
+                        # This needs to block as these user queries need to be
+                        # made BEFORE pushing the event.
+                        yield self._check_user_exists(event.sender)
+                        if event.type == EventTypes.Member:
+                            yield self._check_user_exists(event.state_key)
+
+                        if not self.started_scheduler:
+                            self.scheduler.start().addErrback(log_failure)
+                            self.started_scheduler = True
+
+                        # Fork off pushes to these services
+                        for service in services:
+                            preserve_fn(self.scheduler.submit_event_for_as)(
+                                service, event
+                            )
+
+                    yield self.store.set_appservice_last_pos(upper_bound)
+
+                    if len(events) < limit:
+                        break
+            finally:
+                self.is_processing = False
 
     @defer.inlineCallbacks
     def query_user_exists(self, user_id):
@@ -104,11 +142,12 @@ class ApplicationServicesHandler(object):
             association can be found.
         """
         room_alias_str = room_alias.to_string()
-        alias_query_services = yield self._get_services_for_event(
-            event=None,
-            restrict_to=ApplicationService.NS_ALIASES,
-            alias_list=[room_alias_str]
-        )
+        services = self.store.get_app_services()
+        alias_query_services = [
+            s for s in services if (
+                s.is_interested_in_alias(room_alias_str)
+            )
+        ]
         for alias_service in alias_query_services:
             is_known_alias = yield self.appservice_api.query_alias(
                 alias_service, room_alias_str
@@ -121,47 +160,93 @@ class ApplicationServicesHandler(object):
                 defer.returnValue(result)
 
     @defer.inlineCallbacks
-    def _get_services_for_event(self, event, restrict_to="", alias_list=None):
+    def query_3pe(self, kind, protocol, fields):
+        services = yield self._get_services_for_3pn(protocol)
+
+        results = yield preserve_context_over_deferred(defer.DeferredList([
+            preserve_fn(self.appservice_api.query_3pe)(service, kind, protocol, fields)
+            for service in services
+        ], consumeErrors=True))
+
+        ret = []
+        for (success, result) in results:
+            if success:
+                ret.extend(result)
+
+        defer.returnValue(ret)
+
+    @defer.inlineCallbacks
+    def get_3pe_protocols(self, only_protocol=None):
+        services = self.store.get_app_services()
+        protocols = {}
+
+        # Collect up all the individual protocol responses out of the ASes
+        for s in services:
+            for p in s.protocols:
+                if only_protocol is not None and p != only_protocol:
+                    continue
+
+                if p not in protocols:
+                    protocols[p] = []
+
+                info = yield self.appservice_api.get_3pe_protocol(s, p)
+
+                if info is not None:
+                    protocols[p].append(info)
+
+        def _merge_instances(infos):
+            if not infos:
+                return {}
+
+            # Merge the 'instances' lists of multiple results, but just take
+            # the other fields from the first as they ought to be identical
+            # copy the result so as not to corrupt the cached one
+            combined = dict(infos[0])
+            combined["instances"] = list(combined["instances"])
+
+            for info in infos[1:]:
+                combined["instances"].extend(info["instances"])
+
+            return combined
+
+        for p in protocols.keys():
+            protocols[p] = _merge_instances(protocols[p])
+
+        defer.returnValue(protocols)
+
+    @defer.inlineCallbacks
+    def _get_services_for_event(self, event):
         """Retrieve a list of application services interested in this event.
 
         Args:
             event(Event): The event to check. Can be None if alias_list is not.
-            restrict_to(str): The namespace to restrict regex tests to.
-            alias_list: A list of aliases to get services for. If None, this
-            list is obtained from the database.
         Returns:
             list<ApplicationService>: A list of services interested in this
             event based on the service regex.
         """
-        member_list = None
-        if hasattr(event, "room_id"):
-            # We need to know the aliases associated with this event.room_id,
-            # if any.
-            if not alias_list:
-                alias_list = yield self.store.get_aliases_for_room(
-                    event.room_id
-                )
-            # We need to know the members associated with this event.room_id,
-            # if any.
-            member_list = yield self.store.get_users_in_room(event.room_id)
-
-        services = yield self.store.get_app_services()
+        services = self.store.get_app_services()
         interested_list = [
             s for s in services if (
-                s.is_interested(event, restrict_to, alias_list, member_list)
+                yield s.is_interested(event, self.store)
             )
         ]
         defer.returnValue(interested_list)
 
-    @defer.inlineCallbacks
     def _get_services_for_user(self, user_id):
-        services = yield self.store.get_app_services()
+        services = self.store.get_app_services()
         interested_list = [
             s for s in services if (
                 s.is_interested_in_user(user_id)
             )
         ]
-        defer.returnValue(interested_list)
+        return defer.succeed(interested_list)
+
+    def _get_services_for_3pn(self, protocol):
+        services = self.store.get_app_services()
+        interested_list = [
+            s for s in services if s.is_interested_in_protocol(protocol)
+        ]
+        return defer.succeed(interested_list)
 
     @defer.inlineCallbacks
     def _is_unknown_user(self, user_id):
@@ -177,7 +262,7 @@ class ApplicationServicesHandler(object):
             return
 
         # user not found; could be the AS though, so check.
-        services = yield self.store.get_app_services()
+        services = self.store.get_app_services()
         service_list = [s for s in services if s.sender == user_id]
         defer.returnValue(len(service_list) == 0)
 

synapse/handlers/auth.py

@@ -20,7 +20,6 @@ from synapse.api.constants import LoginType
 from synapse.types import UserID
 from synapse.api.errors import AuthError, LoginError, Codes, StoreError, SynapseError
 from synapse.util.async import run_on_reactor
-from synapse.config.ldap import LDAPMode
 
 from twisted.web.client import PartialDownloadError
 
@@ -29,12 +28,6 @@ import bcrypt
 import pymacaroons
 import simplejson
 
-try:
-    import ldap3
-except ImportError:
-    ldap3 = None
-    pass
-
 import synapse.util.stringutils as stringutils
 
 
@@ -45,6 +38,10 @@ class AuthHandler(BaseHandler):
     SESSION_EXPIRE_MS = 48 * 60 * 60 * 1000
 
     def __init__(self, hs):
+        """
+        Args:
+            hs (synapse.server.HomeServer):
+        """
         super(AuthHandler, self).__init__(hs)
         self.checkers = {
             LoginType.PASSWORD: self._check_password_auth,
@@ -54,25 +51,21 @@ class AuthHandler(BaseHandler):
         }
         self.bcrypt_rounds = hs.config.bcrypt_rounds
         self.sessions = {}
-        self.INVALID_TOKEN_HTTP_STATUS = 401
 
-        self.ldap_enabled = hs.config.ldap_enabled
-        if self.ldap_enabled:
-            if not ldap3:
-                raise RuntimeError(
-                    'Missing ldap3 library. This is required for LDAP Authentication.'
-                )
-            self.ldap_mode = hs.config.ldap_mode
-            self.ldap_uri = hs.config.ldap_uri
-            self.ldap_start_tls = hs.config.ldap_start_tls
-            self.ldap_base = hs.config.ldap_base
-            self.ldap_filter = hs.config.ldap_filter
-            self.ldap_attributes = hs.config.ldap_attributes
-            if self.ldap_mode == LDAPMode.SEARCH:
-                self.ldap_bind_dn = hs.config.ldap_bind_dn
-                self.ldap_bind_password = hs.config.ldap_bind_password
+        account_handler = _AccountHandler(
+            hs, check_user_exists=self.check_user_exists
+        )
+
+        self.password_providers = [
+            module(config=config, account_handler=account_handler)
+            for module, config in hs.config.password_providers
+        ]
+
+        logger.info("Extra password_providers: %r", self.password_providers)
 
         self.hs = hs  # FIXME better possibility to access registrationHandler later?
+        self.device_handler = hs.get_device_handler()
+        self.macaroon_gen = hs.get_macaroon_generator()
 
     @defer.inlineCallbacks
     def check_auth(self, flows, clientdict, clientip):
@@ -143,21 +136,47 @@ class AuthHandler(BaseHandler):
         creds = session['creds']
 
         # check auth type currently being presented
+        errordict = {}
         if 'type' in authdict:
-            if authdict['type'] not in self.checkers:
+            login_type = authdict['type']
+            if login_type not in self.checkers:
                 raise LoginError(400, "", Codes.UNRECOGNIZED)
-            result = yield self.checkers[authdict['type']](authdict, clientip)
-            if result:
-                creds[authdict['type']] = result
-                self._save_session(session)
+            try:
+                result = yield self.checkers[login_type](authdict, clientip)
+                if result:
+                    creds[login_type] = result
+                    self._save_session(session)
+            except LoginError, e:
+                if login_type == LoginType.EMAIL_IDENTITY:
+                    # riot used to have a bug where it would request a new
+                    # validation token (thus sending a new email) each time it
+                    # got a 401 with a 'flows' field.
+                    # (https://github.com/vector-im/vector-web/issues/2447).
+                    #
+                    # Grandfather in the old behaviour for now to avoid
+                    # breaking old riot deployments.
+                    raise e
+
+                # this step failed. Merge the error dict into the response
+                # so that the client can have another go.
+                errordict = e.error_dict()
 
         for f in flows:
             if len(set(f) - set(creds.keys())) == 0:
-                logger.info("Auth completed with creds: %r", creds)
+                # it's very useful to know what args are stored, but this can
+                # include the password in the case of registering, so only log
+                # the keys (confusingly, clientdict may contain a password
+                # param, creds is just what the user authed as for UI auth
+                # and is not sensitive).
+                logger.info(
+                    "Auth completed with creds: %r. Client dict has keys: %r",
+                    creds, clientdict.keys()
+                )
                 defer.returnValue((True, creds, clientdict, session['id']))
 
         ret = self._auth_dict_for_flows(flows, session)
         ret['completed'] = creds.keys()
+        ret.update(errordict)
         defer.returnValue((False, ret, clientdict, session['id']))
 
     @defer.inlineCallbacks
@@ -230,7 +249,6 @@ class AuthHandler(BaseHandler):
         sess = self._get_session_info(session_id)
         return sess.setdefault('serverdict', {}).get(key, default)
 
-    @defer.inlineCallbacks
     def _check_password_auth(self, authdict, _):
         if "user" not in authdict or "password" not in authdict:
             raise LoginError(400, "", Codes.MISSING_PARAM)
@@ -240,11 +258,7 @@ class AuthHandler(BaseHandler):
         if not user_id.startswith('@'):
             user_id = UserID.create(user_id, self.hs.hostname).to_string()
 
-        if not (yield self._check_password(user_id, password)):
-            logger.warn("Failed password login for user %s", user_id)
-            raise LoginError(403, "", errcode=Codes.FORBIDDEN)
-
-        defer.returnValue(user_id)
+        return self._check_password(user_id, password)
 
     @defer.inlineCallbacks
     def _check_recaptcha(self, authdict, clientip):
@@ -280,8 +294,17 @@ class AuthHandler(BaseHandler):
             data = pde.response
             resp_body = simplejson.loads(data)
 
-        if 'success' in resp_body and resp_body['success']:
-            defer.returnValue(True)
+        if 'success' in resp_body:
+            # Note that we do NOT check the hostname here: we explicitly
+            # intend the CAPTCHA to be presented by whatever client the
+            # user is using, we just care that they have completed a CAPTCHA.
+            logger.info(
+                "%s reCAPTCHA from hostname %s",
+                "Successful" if resp_body['success'] else "Failed",
+                resp_body.get('hostname')
+            )
+            if resp_body['success']:
+                defer.returnValue(True)
         raise LoginError(401, "", errcode=Codes.UNAUTHORIZED)
 
     @defer.inlineCallbacks
@@ -348,362 +371,185 @@ class AuthHandler(BaseHandler):
 
         return self.sessions[session_id]
 
-    @defer.inlineCallbacks
-    def login_with_password(self, user_id, password):
+    def validate_password_login(self, user_id, password):
         """
         Authenticates the user with their username and password.
 
         Used only by the v1 login API.
 
         Args:
-            user_id (str): User ID
+            user_id (str): complete @user:id
             password (str): Password
         Returns:
-            A tuple of:
-              The user's ID.
-              The access token for the user's session.
-              The refresh token for the user's session.
+            defer.Deferred: (str) canonical user id
         Raises:
-            StoreError if there was a problem storing the token.
+            StoreError if there was a problem accessing the database
             LoginError if there was an authentication problem.
         """
-
-        if not (yield self._check_password(user_id, password)):
-            logger.warn("Failed password login for user %s", user_id)
-            raise LoginError(403, "", errcode=Codes.FORBIDDEN)
-
-        logger.info("Logging in user %s", user_id)
-        access_token = yield self.issue_access_token(user_id)
-        refresh_token = yield self.issue_refresh_token(user_id)
-        defer.returnValue((user_id, access_token, refresh_token))
+        return self._check_password(user_id, password)
 
     @defer.inlineCallbacks
-    def get_login_tuple_for_user_id(self, user_id):
+    def get_access_token_for_user_id(self, user_id, device_id=None,
+                                     initial_display_name=None):
         """
-        Gets login tuple for the user with the given user ID.
+        Creates a new access token for the user with the given user ID.
+
         The user is assumed to have been authenticated by some other
-        machanism (e.g. CAS)
+        machanism (e.g. CAS), and the user_id converted to the canonical case.
+
+        The device will be recorded in the table if it is not there already.
 
         Args:
-            user_id (str): User ID
+            user_id (str): canonical User ID
+            device_id (str|None): the device ID to associate with the tokens.
+               None to leave the tokens unassociated with a device (deprecated:
+               we should always have a device ID)
+            initial_display_name (str): display name to associate with the
+               device if it needs re-registering
         Returns:
-            A tuple of:
-              The user's ID.
               The access token for the user's session.
-              The refresh token for the user's session.
         Raises:
             StoreError if there was a problem storing the token.
             LoginError if there was an authentication problem.
         """
-        user_id, ignored = yield self._find_user_id_and_pwd_hash(user_id)
+        logger.info("Logging in user %s on device %s", user_id, device_id)
+        access_token = yield self.issue_access_token(user_id, device_id)
 
-        logger.info("Logging in user %s", user_id)
-        access_token = yield self.issue_access_token(user_id)
-        refresh_token = yield self.issue_refresh_token(user_id)
-        defer.returnValue((user_id, access_token, refresh_token))
+        # the device *should* have been registered before we got here; however,
+        # it's possible we raced against a DELETE operation. The thing we
+        # really don't want is active access_tokens without a record of the
+        # device, so we double-check it here.
+        if device_id is not None:
+            yield self.device_handler.check_device_registered(
+                user_id, device_id, initial_display_name
+            )
+
+        defer.returnValue(access_token)
 
     @defer.inlineCallbacks
-    def does_user_exist(self, user_id):
-        try:
-            yield self._find_user_id_and_pwd_hash(user_id)
-            defer.returnValue(True)
-        except LoginError:
-            defer.returnValue(False)
+    def check_user_exists(self, user_id):
+        """
+        Checks to see if a user with the given id exists. Will check case
+        insensitively, but return None if there are multiple inexact matches.
+
+        Args:
+            (str) user_id: complete @user:id
+
+        Returns:
+            defer.Deferred: (str) canonical_user_id, or None if zero or
+            multiple matches
+        """
+        res = yield self._find_user_id_and_pwd_hash(user_id)
+        if res is not None:
+            defer.returnValue(res[0])
+        defer.returnValue(None)
 
     @defer.inlineCallbacks
     def _find_user_id_and_pwd_hash(self, user_id):
         """Checks to see if a user with the given id exists. Will check case
-        insensitively, but will throw if there are multiple inexact matches.
+        insensitively, but will return None if there are multiple inexact
+        matches.
 
         Returns:
             tuple: A 2-tuple of `(canonical_user_id, password_hash)`
+            None: if there is not exactly one match
         """
         user_infos = yield self.store.get_users_by_id_case_insensitive(user_id)
+
+        result = None
         if not user_infos:
             logger.warn("Attempted to login as %s but they do not exist", user_id)
-            raise LoginError(403, "", errcode=Codes.FORBIDDEN)
-
-        if len(user_infos) > 1:
-            if user_id not in user_infos:
-                logger.warn(
-                    "Attempted to login as %s but it matches more than one user "
-                    "inexactly: %r",
-                    user_id, user_infos.keys()
-                )
-                raise LoginError(403, "", errcode=Codes.FORBIDDEN)
-
-            defer.returnValue((user_id, user_infos[user_id]))
+        elif len(user_infos) == 1:
+            # a single match (possibly not exact)
+            result = user_infos.popitem()
+        elif user_id in user_infos:
+            # multiple matches, but one is exact
+            result = (user_id, user_infos[user_id])
         else:
-            defer.returnValue(user_infos.popitem())
+            # multiple matches, none of them exact
+            logger.warn(
+                "Attempted to login as %s but it matches more than one user "
+                "inexactly: %r",
+                user_id, user_infos.keys()
+            )
+        defer.returnValue(result)
 
     @defer.inlineCallbacks
     def _check_password(self, user_id, password):
-        """
+        """Authenticate a user against the LDAP and local databases.
+
+        user_id is checked case insensitively against the local database, but
+        will throw if there are multiple inexact matches.
+
+        Args:
+            user_id (str): complete @user:id
         Returns:
-            True if the user_id successfully authenticated
+            (str) the canonical_user_id
+        Raises:
+            LoginError if login fails
         """
-        valid_ldap = yield self._check_ldap_password(user_id, password)
-        if valid_ldap:
-            defer.returnValue(True)
+        for provider in self.password_providers:
+            is_valid = yield provider.check_password(user_id, password)
+            if is_valid:
+                defer.returnValue(user_id)
 
-        valid_local_password = yield self._check_local_password(user_id, password)
-        if valid_local_password:
-            defer.returnValue(True)
+        canonical_user_id = yield self._check_local_password(user_id, password)
 
-        defer.returnValue(False)
+        if canonical_user_id:
+            defer.returnValue(canonical_user_id)
+
+        # unknown username or invalid password. We raise a 403 here, but note
+        # that if we're doing user-interactive login, it turns all LoginErrors
+        # into a 401 anyway.
+        raise LoginError(
+            403, "Invalid password",
+            errcode=Codes.FORBIDDEN
+        )
 
     @defer.inlineCallbacks
     def _check_local_password(self, user_id, password):
-        try:
-            user_id, password_hash = yield self._find_user_id_and_pwd_hash(user_id)
-            defer.returnValue(self.validate_hash(password, password_hash))
-        except LoginError:
-            defer.returnValue(False)
+        """Authenticate a user against the local password database.
 
-    @defer.inlineCallbacks
-    def _check_ldap_password(self, user_id, password):
-        """ Attempt to authenticate a user against an LDAP Server
-            and register an account if none exists.
+        user_id is checked case insensitively, but will return None if there are
+        multiple inexact matches.
 
-            Returns:
-                True if authentication against LDAP was successful
+        Args:
+            user_id (str): complete @user:id
+        Returns:
+            (str) the canonical_user_id, or None if unknown user / bad password
         """
-
-        if not ldap3 or not self.ldap_enabled:
-            defer.returnValue(False)
-
-        if self.ldap_mode not in LDAPMode.LIST:
-            raise RuntimeError(
-                'Invalid ldap mode specified: {mode}'.format(
-                    mode=self.ldap_mode
-                )
-            )
-
-        try:
-            server = ldap3.Server(self.ldap_uri)
-            logger.debug(
-                "Attempting ldap connection with %s",
-                self.ldap_uri
-            )
-
-            localpart = UserID.from_string(user_id).localpart
-            if self.ldap_mode == LDAPMode.SIMPLE:
-                # bind with the the local users ldap credentials
-                bind_dn = "{prop}={value},{base}".format(
-                    prop=self.ldap_attributes['uid'],
-                    value=localpart,
-                    base=self.ldap_base
-                )
-                conn = ldap3.Connection(server, bind_dn, password)
-                logger.debug(
-                    "Established ldap connection in simple mode: %s",
-                    conn
-                )
-
-                if self.ldap_start_tls:
-                    conn.start_tls()
-                    logger.debug(
-                        "Upgraded ldap connection in simple mode through StartTLS: %s",
-                        conn
-                    )
-
-                conn.bind()
-
-            elif self.ldap_mode == LDAPMode.SEARCH:
-                # connect with preconfigured credentials and search for local user
-                conn = ldap3.Connection(
-                    server,
-                    self.ldap_bind_dn,
-                    self.ldap_bind_password
-                )
-                logger.debug(
-                    "Established ldap connection in search mode: %s",
-                    conn
-                )
-
-                if self.ldap_start_tls:
-                    conn.start_tls()
-                    logger.debug(
-                        "Upgraded ldap connection in search mode through StartTLS: %s",
-                        conn
-                    )
-
-                conn.bind()
-
-                # find matching dn
-                query = "({prop}={value})".format(
-                    prop=self.ldap_attributes['uid'],
-                    value=localpart
-                )
-                if self.ldap_filter:
-                    query = "(&{query}{filter})".format(
-                        query=query,
-                        filter=self.ldap_filter
-                    )
-                logger.debug("ldap search filter: %s", query)
-                result = conn.search(self.ldap_base, query)
-
-                if result and len(conn.response) == 1:
-                    # found exactly one result
-                    user_dn = conn.response[0]['dn']
-                    logger.debug('ldap search found dn: %s', user_dn)
-
-                    # unbind and reconnect, rebind with found dn
-                    conn.unbind()
-                    conn = ldap3.Connection(
-                        server,
-                        user_dn,
-                        password,
-                        auto_bind=True
-                    )
-                else:
-                    # found 0 or > 1 results, abort!
-                    logger.warn(
-                        "ldap search returned unexpected (%d!=1) amount of results",
-                        len(conn.response)
-                    )
-                    defer.returnValue(False)
-
-            logger.info(
-                "User authenticated against ldap server: %s",
-                conn
-            )
-
-            # check for existing account, if none exists, create one
-            if not (yield self.does_user_exist(user_id)):
-                # query user metadata for account creation
-                query = "({prop}={value})".format(
-                    prop=self.ldap_attributes['uid'],
-                    value=localpart
-                )
-
-                if self.ldap_mode == LDAPMode.SEARCH and self.ldap_filter:
-                    query = "(&{filter}{user_filter})".format(
-                        filter=query,
-                        user_filter=self.ldap_filter
-                    )
-                logger.debug("ldap registration filter: %s", query)
-
-                result = conn.search(
-                    search_base=self.ldap_base,
-                    search_filter=query,
-                    attributes=[
-                        self.ldap_attributes['name'],
-                        self.ldap_attributes['mail']
-                    ]
-                )
-
-                if len(conn.response) == 1:
-                    attrs = conn.response[0]['attributes']
-                    mail = attrs[self.ldap_attributes['mail']][0]
-                    name = attrs[self.ldap_attributes['name']][0]
-
-                    # create account
-                    registration_handler = self.hs.get_handlers().registration_handler
-                    user_id, access_token = (
-                        yield registration_handler.register(localpart=localpart)
-                    )
-
-                    # TODO: bind email, set displayname with data from ldap directory
-
-                    logger.info(
-                        "ldap registration successful: %d: %s (%s, %)",
-                        user_id,
-                        localpart,
-                        name,
-                        mail
-                    )
-                else:
-                    logger.warn(
-                        "ldap registration failed: unexpected (%d!=1) amount of results",
-                        len(result)
-                    )
-                    defer.returnValue(False)
-
-            defer.returnValue(True)
-        except ldap3.core.exceptions.LDAPException as e:
-            logger.warn("Error during ldap authentication: %s", e)
-            defer.returnValue(False)
+        lookupres = yield self._find_user_id_and_pwd_hash(user_id)
+        if not lookupres:
+            defer.returnValue(None)
+        (user_id, password_hash) = lookupres
+        result = self.validate_hash(password, password_hash)
+        if not result:
+            logger.warn("Failed password login for user %s", user_id)
+            defer.returnValue(None)
+        defer.returnValue(user_id)
 
     @defer.inlineCallbacks
-    def issue_access_token(self, user_id):
-        access_token = self.generate_access_token(user_id)
-        yield self.store.add_access_token_to_user(user_id, access_token)
+    def issue_access_token(self, user_id, device_id=None):
+        access_token = self.macaroon_gen.generate_access_token(user_id)
+        yield self.store.add_access_token_to_user(user_id, access_token,
+                                                  device_id)
         defer.returnValue(access_token)
 
-    @defer.inlineCallbacks
-    def issue_refresh_token(self, user_id):
-        refresh_token = self.generate_refresh_token(user_id)
-        yield self.store.add_refresh_token_to_user(user_id, refresh_token)
-        defer.returnValue(refresh_token)
-
-    def generate_access_token(self, user_id, extra_caveats=None):
-        extra_caveats = extra_caveats or []
-        macaroon = self._generate_base_macaroon(user_id)
-        macaroon.add_first_party_caveat("type = access")
-        now = self.hs.get_clock().time_msec()
-        expiry = now + (60 * 60 * 1000)
-        macaroon.add_first_party_caveat("time < %d" % (expiry,))
-        for caveat in extra_caveats:
-            macaroon.add_first_party_caveat(caveat)
-        return macaroon.serialize()
-
-    def generate_refresh_token(self, user_id):
-        m = self._generate_base_macaroon(user_id)
-        m.add_first_party_caveat("type = refresh")
-        # Important to add a nonce, because otherwise every refresh token for a
-        # user will be the same.
-        m.add_first_party_caveat("nonce = %s" % (
-            stringutils.random_string_with_symbols(16),
-        ))
-        return m.serialize()
-
-    def generate_short_term_login_token(self, user_id, duration_in_ms=(2 * 60 * 1000)):
-        macaroon = self._generate_base_macaroon(user_id)
-        macaroon.add_first_party_caveat("type = login")
-        now = self.hs.get_clock().time_msec()
-        expiry = now + duration_in_ms
-        macaroon.add_first_party_caveat("time < %d" % (expiry,))
-        return macaroon.serialize()
-
-    def generate_delete_pusher_token(self, user_id):
-        macaroon = self._generate_base_macaroon(user_id)
-        macaroon.add_first_party_caveat("type = delete_pusher")
-        return macaroon.serialize()
-
     def validate_short_term_login_token_and_get_user_id(self, login_token):
+        auth_api = self.hs.get_auth()
         try:
             macaroon = pymacaroons.Macaroon.deserialize(login_token)
-            auth_api = self.hs.get_auth()
-            auth_api.validate_macaroon(macaroon, "login", True)
-            return self.get_user_from_macaroon(macaroon)
-        except (pymacaroons.exceptions.MacaroonException, TypeError, ValueError):
-            raise AuthError(401, "Invalid token", errcode=Codes.UNKNOWN_TOKEN)
-
-    def _generate_base_macaroon(self, user_id):
-        macaroon = pymacaroons.Macaroon(
-            location=self.hs.config.server_name,
-            identifier="key",
-            key=self.hs.config.macaroon_secret_key)
-        macaroon.add_first_party_caveat("gen = 1")
-        macaroon.add_first_party_caveat("user_id = %s" % (user_id,))
-        return macaroon
-
-    def get_user_from_macaroon(self, macaroon):
-        user_prefix = "user_id = "
-        for caveat in macaroon.caveats:
-            if caveat.caveat_id.startswith(user_prefix):
-                return caveat.caveat_id[len(user_prefix):]
-        raise AuthError(
-            self.INVALID_TOKEN_HTTP_STATUS, "No user_id found in token",
-            errcode=Codes.UNKNOWN_TOKEN
-        )
+            user_id = auth_api.get_user_id_from_macaroon(macaroon)
+            auth_api.validate_macaroon(macaroon, "login", True, user_id)
+            return user_id
+        except Exception:
+            raise AuthError(403, "Invalid token", errcode=Codes.FORBIDDEN)
 
     @defer.inlineCallbacks
     def set_password(self, user_id, newpassword, requester=None):
         password_hash = self.hash(newpassword)
 
-        except_access_token_ids = [requester.access_token_id] if requester else []
+        except_access_token_id = requester.access_token_id if requester else None
 
         try:
             yield self.store.user_set_password_hash(user_id, password_hash)
@@ -712,19 +558,42 @@ class AuthHandler(BaseHandler):
                 raise SynapseError(404, "Unknown user", Codes.NOT_FOUND)
             raise e
         yield self.store.user_delete_access_tokens(
-            user_id, except_access_token_ids
+            user_id, except_access_token_id
         )
         yield self.hs.get_pusherpool().remove_pushers_by_user(
-            user_id, except_access_token_ids
+            user_id, except_access_token_id
         )
 
     @defer.inlineCallbacks
     def add_threepid(self, user_id, medium, address, validated_at):
+        # 'Canonicalise' email addresses down to lower case.
+        # We've now moving towards the Home Server being the entity that
+        # is responsible for validating threepids used for resetting passwords
+        # on accounts, so in future Synapse will gain knowledge of specific
+        # types (mediums) of threepid. For now, we still use the existing
+        # infrastructure, but this is the start of synapse gaining knowledge
+        # of specific types of threepid (and fixes the fact that checking
+        # for the presence of an email address during password reset was
+        # case sensitive).
+        if medium == 'email':
+            address = address.lower()
+
         yield self.store.user_add_threepid(
             user_id, medium, address, validated_at,
             self.hs.get_clock().time_msec()
         )
 
+    @defer.inlineCallbacks
+    def delete_threepid(self, user_id, medium, address):
+        # 'Canonicalise' email addresses as per above
+        if medium == 'email':
+            address = address.lower()
+
+        ret = yield self.store.user_delete_threepid(
+            user_id, medium, address,
+        )
+        defer.returnValue(ret)
+
     def _save_session(self, session):
         # TODO: Persistent storage
         logger.debug("Saving session %s", session)
@@ -750,7 +619,7 @@ class AuthHandler(BaseHandler):
         Returns:
             Hashed password (str).
         """
-        return bcrypt.hashpw(password + self.hs.config.password_pepper,
+        return bcrypt.hashpw(password.encode('utf8') + self.hs.config.password_pepper,
                              bcrypt.gensalt(self.bcrypt_rounds))
 
     def validate_hash(self, password, stored_hash):
@@ -764,7 +633,76 @@ class AuthHandler(BaseHandler):
             Whether self.hash(password) == stored_hash (bool).
         """
         if stored_hash:
-            return bcrypt.hashpw(password + self.hs.config.password_pepper,
-                                 stored_hash.encode('utf-8')) == stored_hash
+            return bcrypt.hashpw(password.encode('utf8') + self.hs.config.password_pepper,
+                                 stored_hash.encode('utf8')) == stored_hash
         else:
             return False
+
+
+class MacaroonGeneartor(object):
+    def __init__(self, hs):
+        self.clock = hs.get_clock()
+        self.server_name = hs.config.server_name
+        self.macaroon_secret_key = hs.config.macaroon_secret_key
+
+    def generate_access_token(self, user_id, extra_caveats=None):
+        extra_caveats = extra_caveats or []
+        macaroon = self._generate_base_macaroon(user_id)
+        macaroon.add_first_party_caveat("type = access")
+        # Include a nonce, to make sure that each login gets a different
+        # access token.
+        macaroon.add_first_party_caveat("nonce = %s" % (
+            stringutils.random_string_with_symbols(16),
+        ))
+        for caveat in extra_caveats:
+            macaroon.add_first_party_caveat(caveat)
+        return macaroon.serialize()
+
+    def generate_short_term_login_token(self, user_id, duration_in_ms=(2 * 60 * 1000)):
+        macaroon = self._generate_base_macaroon(user_id)
+        macaroon.add_first_party_caveat("type = login")
+        now = self.clock.time_msec()
+        expiry = now + duration_in_ms
+        macaroon.add_first_party_caveat("time < %d" % (expiry,))
+        return macaroon.serialize()
+
+    def generate_delete_pusher_token(self, user_id):
+        macaroon = self._generate_base_macaroon(user_id)
+        macaroon.add_first_party_caveat("type = delete_pusher")
+        return macaroon.serialize()
+
+    def _generate_base_macaroon(self, user_id):
+        macaroon = pymacaroons.Macaroon(
+            location=self.server_name,
+            identifier="key",
+            key=self.macaroon_secret_key)
+        macaroon.add_first_party_caveat("gen = 1")
+        macaroon.add_first_party_caveat("user_id = %s" % (user_id,))
+        return macaroon
+
+
+class _AccountHandler(object):
+    """A proxy object that gets passed to password auth providers so they
+    can register new users etc if necessary.
+    """
+    def __init__(self, hs, check_user_exists):
+        self.hs = hs
+
+        self._check_user_exists = check_user_exists
+
+    def check_user_exists(self, user_id):
+        """Check if user exissts.
+
+        Returns:
+            Deferred(bool)
+        """
+        return self._check_user_exists(user_id)
+
+    def register(self, localpart):
+        """Registers a new user with given localpart
+
+        Returns:
+            Deferred: a 2-tuple of (user_id, access_token)
+        """
+        reg = self.hs.get_handlers().registration_handler
+        return reg.register(localpart=localpart)

synapse/handlers/device.py

@@ -0,0 +1,354 @@
+# -*- coding: utf-8 -*-
+# Copyright 2016 OpenMarket Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from synapse.api import errors
+from synapse.api.constants import EventTypes
+from synapse.util import stringutils
+from synapse.util.async import Linearizer
+from synapse.types import get_domain_from_id, RoomStreamToken
+from twisted.internet import defer
+from ._base import BaseHandler
+
+import logging
+
+logger = logging.getLogger(__name__)
+
+
+class DeviceHandler(BaseHandler):
+    def __init__(self, hs):
+        super(DeviceHandler, self).__init__(hs)
+
+        self.hs = hs
+        self.state = hs.get_state_handler()
+        self.federation_sender = hs.get_federation_sender()
+        self.federation = hs.get_replication_layer()
+        self._remote_edue_linearizer = Linearizer(name="remote_device_list")
+
+        self.federation.register_edu_handler(
+            "m.device_list_update", self._incoming_device_list_update,
+        )
+        self.federation.register_query_handler(
+            "user_devices", self.on_federation_query_user_devices,
+        )
+
+        hs.get_distributor().observe("user_left_room", self.user_left_room)
+
+    @defer.inlineCallbacks
+    def check_device_registered(self, user_id, device_id,
+                                initial_device_display_name=None):
+        """
+        If the given device has not been registered, register it with the
+        supplied display name.
+
+        If no device_id is supplied, we make one up.
+
+        Args:
+            user_id (str):  @user:id
+            device_id (str | None): device id supplied by client
+            initial_device_display_name (str | None): device display name from
+                 client
+        Returns:
+            str: device id (generated if none was supplied)
+        """
+        if device_id is not None:
+            new_device = yield self.store.store_device(
+                user_id=user_id,
+                device_id=device_id,
+                initial_device_display_name=initial_device_display_name,
+            )
+            if new_device:
+                yield self.notify_device_update(user_id, [device_id])
+            defer.returnValue(device_id)
+
+        # if the device id is not specified, we'll autogen one, but loop a few
+        # times in case of a clash.
+        attempts = 0
+        while attempts < 5:
+            device_id = stringutils.random_string(10).upper()
+            new_device = yield self.store.store_device(
+                user_id=user_id,
+                device_id=device_id,
+                initial_device_display_name=initial_device_display_name,
+            )
+            if new_device:
+                yield self.notify_device_update(user_id, [device_id])
+                defer.returnValue(device_id)
+            attempts += 1
+
+        raise errors.StoreError(500, "Couldn't generate a device ID.")
+
+    @defer.inlineCallbacks
+    def get_devices_by_user(self, user_id):
+        """
+        Retrieve the given user's devices
+
+        Args:
+            user_id (str):
+        Returns:
+            defer.Deferred: list[dict[str, X]]: info on each device
+        """
+
+        device_map = yield self.store.get_devices_by_user(user_id)
+
+        ips = yield self.store.get_last_client_ip_by_device(
+            devices=((user_id, device_id) for device_id in device_map.keys())
+        )
+
+        devices = device_map.values()
+        for device in devices:
+            _update_device_from_client_ips(device, ips)
+
+        defer.returnValue(devices)
+
+    @defer.inlineCallbacks
+    def get_device(self, user_id, device_id):
+        """ Retrieve the given device
+
+        Args:
+            user_id (str):
+            device_id (str):
+
+        Returns:
+            defer.Deferred: dict[str, X]: info on the device
+        Raises:
+            errors.NotFoundError: if the device was not found
+        """
+        try:
+            device = yield self.store.get_device(user_id, device_id)
+        except errors.StoreError:
+            raise errors.NotFoundError
+        ips = yield self.store.get_last_client_ip_by_device(
+            devices=((user_id, device_id),)
+        )
+        _update_device_from_client_ips(device, ips)
+        defer.returnValue(device)
+
+    @defer.inlineCallbacks
+    def delete_device(self, user_id, device_id):
+        """ Delete the given device
+
+        Args:
+            user_id (str):
+            device_id (str):
+
+        Returns:
+            defer.Deferred:
+        """
+
+        try:
+            yield self.store.delete_device(user_id, device_id)
+        except errors.StoreError, e:
+            if e.code == 404:
+                # no match
+                pass
+            else:
+                raise
+
+        yield self.store.user_delete_access_tokens(
+            user_id, device_id=device_id,
+            delete_refresh_tokens=True,
+        )
+
+        yield self.store.delete_e2e_keys_by_device(
+            user_id=user_id, device_id=device_id
+        )
+
+        yield self.notify_device_update(user_id, [device_id])
+
+    @defer.inlineCallbacks
+    def update_device(self, user_id, device_id, content):
+        """ Update the given device
+
+        Args:
+            user_id (str):
+            device_id (str):
+            content (dict): body of update request
+
+        Returns:
+            defer.Deferred:
+        """
+
+        try:
+            yield self.store.update_device(
+                user_id,
+                device_id,
+                new_display_name=content.get("display_name")
+            )
+            yield self.notify_device_update(user_id, [device_id])
+        except errors.StoreError, e:
+            if e.code == 404:
+                raise errors.NotFoundError()
+            else:
+                raise
+
+    @defer.inlineCallbacks
+    def notify_device_update(self, user_id, device_ids):
+        """Notify that a user's device(s) has changed. Pokes the notifier, and
+        remote servers if the user is local.
+        """
+        users_who_share_room = yield self.store.get_users_who_share_room_with_user(
+            user_id
+        )
+
+        hosts = set()
+        if self.hs.is_mine_id(user_id):
+            hosts.update(get_domain_from_id(u) for u in users_who_share_room)
+            hosts.discard(self.server_name)
+
+        position = yield self.store.add_device_change_to_streams(
+            user_id, device_ids, list(hosts)
+        )
+
+        rooms = yield self.store.get_rooms_for_user(user_id)
+        room_ids = [r.room_id for r in rooms]
+
+        yield self.notifier.on_new_event(
+            "device_list_key", position, rooms=room_ids,
+        )
+
+        if hosts:
+            logger.info("Sending device list update notif to: %r", hosts)
+            for host in hosts:
+                self.federation_sender.send_device_messages(host)
+
+    @defer.inlineCallbacks
+    def get_user_ids_changed(self, user_id, from_token):
+        """Get list of users that have had the devices updated, or have newly
+        joined a room, that `user_id` may be interested in.
+
+        Args:
+            user_id (str)
+            from_token (StreamToken)
+        """
+        rooms = yield self.store.get_rooms_for_user(user_id)
+        room_ids = set(r.room_id for r in rooms)
+
+        # First we check if any devices have changed
+        changed = yield self.store.get_user_whose_devices_changed(
+            from_token.device_list_key
+        )
+
+        # Then work out if any users have since joined
+        rooms_changed = self.store.get_rooms_that_changed(room_ids, from_token.room_key)
+
+        possibly_changed = set(changed)
+        for room_id in rooms_changed:
+            # Fetch  the current state at the time.
+            stream_ordering = RoomStreamToken.parse_stream_token(from_token.room_key)
+
+            try:
+                event_ids = yield self.store.get_forward_extremeties_for_room(
+                    room_id, stream_ordering=stream_ordering
+                )
+                prev_state_ids = yield self.store.get_state_ids_for_events(event_ids)
+            except:
+                prev_state_ids = {}
+
+            current_state_ids = yield self.state.get_current_state_ids(room_id)
+
+            # If there has been any change in membership, include them in the
+            # possibly changed list. We'll check if they are joined below,
+            # and we're not toooo worried about spuriously adding users.
+            for key, event_id in current_state_ids.iteritems():
+                etype, state_key = key
+                if etype == EventTypes.Member:
+                    prev_event_id = prev_state_ids.get(key, None)
+                    if not prev_event_id or prev_event_id != event_id:
+                        possibly_changed.add(state_key)
+
+        users_who_share_room = yield self.store.get_users_who_share_room_with_user(
+            user_id
+        )
+
+        # Take the intersection of the users whose devices may have changed
+        # and those that actually still share a room with the user
+        defer.returnValue(users_who_share_room & possibly_changed)
+
+    @defer.inlineCallbacks
+    def _incoming_device_list_update(self, origin, edu_content):
+        user_id = edu_content["user_id"]
+        device_id = edu_content["device_id"]
+        stream_id = edu_content["stream_id"]
+        prev_ids = edu_content.get("prev_id", [])
+
+        if get_domain_from_id(user_id) != origin:
+            # TODO: Raise?
+            logger.warning("Got device list update edu for %r from %r", user_id, origin)
+            return
+
+        rooms = yield self.store.get_rooms_for_user(user_id)
+        if not rooms:
+            # We don't share any rooms with this user. Ignore update, as we
+            # probably won't get any further updates.
+            return
+
+        with (yield self._remote_edue_linearizer.queue(user_id)):
+            # If the prev id matches whats in our cache table, then we don't need
+            # to resync the users device list, otherwise we do.
+            resync = True
+            if len(prev_ids) == 1:
+                extremity = yield self.store.get_device_list_last_stream_id_for_remote(
+                    user_id
+                )
+                logger.info("Extrem: %r, prev_ids: %r", extremity, prev_ids)
+                if str(extremity) == str(prev_ids[0]):
+                    resync = False
+
+            if resync:
+                # Fetch all devices for the user.
+                result = yield self.federation.query_user_devices(origin, user_id)
+                stream_id = result["stream_id"]
+                devices = result["devices"]
+                yield self.store.update_remote_device_list_cache(
+                    user_id, devices, stream_id,
+                )
+                device_ids = [device["device_id"] for device in devices]
+                yield self.notify_device_update(user_id, device_ids)
+            else:
+                # Simply update the single device, since we know that is the only
+                # change (becuase of the single prev_id matching the current cache)
+                content = dict(edu_content)
+                for key in ("user_id", "device_id", "stream_id", "prev_ids"):
+                    content.pop(key, None)
+                yield self.store.update_remote_device_list_cache_entry(
+                    user_id, device_id, content, stream_id,
+                )
+                yield self.notify_device_update(user_id, [device_id])
+
+    @defer.inlineCallbacks
+    def on_federation_query_user_devices(self, user_id):
+        stream_id, devices = yield self.store.get_devices_with_keys_by_user(user_id)
+        defer.returnValue({
+            "user_id": user_id,
+            "stream_id": stream_id,
+            "devices": devices,
+        })
+
+    @defer.inlineCallbacks
+    def user_left_room(self, user, room_id):
+        user_id = user.to_string()
+        rooms = yield self.store.get_rooms_for_user(user_id)
+        if not rooms:
+            # We no longer share rooms with this user, so we'll no longer
+            # receive device updates. Mark this in DB.
+            yield self.store.mark_remote_user_device_list_as_unsubscribed(user_id)
+
+
+def _update_device_from_client_ips(device, client_ips):
+    ip = client_ips.get((device["user_id"], device["device_id"]), {})
+    device.update({
+        "last_seen_ts": ip.get("last_seen"),
+        "last_seen_ip": ip.get("ip"),
+    })

synapse/handlers/devicemessage.py

@@ -0,0 +1,117 @@
+# -*- coding: utf-8 -*-
+# Copyright 2016 OpenMarket Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+
+from twisted.internet import defer
+
+from synapse.types import get_domain_from_id
+from synapse.util.stringutils import random_string
+
+
+logger = logging.getLogger(__name__)
+
+
+class DeviceMessageHandler(object):
+
+    def __init__(self, hs):
+        """
+        Args:
+            hs (synapse.server.HomeServer): server
+        """
+        self.store = hs.get_datastore()
+        self.notifier = hs.get_notifier()
+        self.is_mine_id = hs.is_mine_id
+        self.federation = hs.get_federation_sender()
+
+        hs.get_replication_layer().register_edu_handler(
+            "m.direct_to_device", self.on_direct_to_device_edu
+        )
+
+    @defer.inlineCallbacks
+    def on_direct_to_device_edu(self, origin, content):
+        local_messages = {}
+        sender_user_id = content["sender"]
+        if origin != get_domain_from_id(sender_user_id):
+            logger.warn(
+                "Dropping device message from %r with spoofed sender %r",
+                origin, sender_user_id
+            )
+        message_type = content["type"]
+        message_id = content["message_id"]
+        for user_id, by_device in content["messages"].items():
+            messages_by_device = {
+                device_id: {
+                    "content": message_content,
+                    "type": message_type,
+                    "sender": sender_user_id,
+                }
+                for device_id, message_content in by_device.items()
+            }
+            if messages_by_device:
+                local_messages[user_id] = messages_by_device
+
+        stream_id = yield self.store.add_messages_from_remote_to_device_inbox(
+            origin, message_id, local_messages
+        )
+
+        self.notifier.on_new_event(
+            "to_device_key", stream_id, users=local_messages.keys()
+        )
+
+    @defer.inlineCallbacks
+    def send_device_message(self, sender_user_id, message_type, messages):
+
+        local_messages = {}
+        remote_messages = {}
+        for user_id, by_device in messages.items():
+            if self.is_mine_id(user_id):
+                messages_by_device = {
+                    device_id: {
+                        "content": message_content,
+                        "type": message_type,
+                        "sender": sender_user_id,
+                    }
+                    for device_id, message_content in by_device.items()
+                }
+                if messages_by_device:
+                    local_messages[user_id] = messages_by_device
+            else:
+                destination = get_domain_from_id(user_id)
+                remote_messages.setdefault(destination, {})[user_id] = by_device
+
+        message_id = random_string(16)
+
+        remote_edu_contents = {}
+        for destination, messages in remote_messages.items():
+            remote_edu_contents[destination] = {
+                "messages": messages,
+                "sender": sender_user_id,
+                "type": message_type,
+                "message_id": message_id,
+            }
+
+        stream_id = yield self.store.add_messages_to_device_inbox(
+            local_messages, remote_edu_contents
+        )
+
+        self.notifier.on_new_event(
+            "to_device_key", stream_id, users=local_messages.keys()
+        )
+
+        for destination in remote_messages.keys():
+            # Enqueue a new federation transaction to send the new
+            # device messages to each remote destination.
+            self.federation.send_device_messages(destination)

synapse/handlers/directory.py

@@ -19,7 +19,7 @@ from ._base import BaseHandler
 
 from synapse.api.errors import SynapseError, Codes, CodeMessageException, AuthError
 from synapse.api.constants import EventTypes
-from synapse.types import RoomAlias, UserID
+from synapse.types import RoomAlias, UserID, get_domain_from_id
 
 import logging
 import string
@@ -55,7 +55,8 @@ class DirectoryHandler(BaseHandler):
         # TODO(erikj): Add transactions.
         # TODO(erikj): Check if there is a current association.
         if not servers:
-            servers = yield self.store.get_joined_hosts_for_room(room_id)
+            users = yield self.state.get_current_user_in_room(room_id)
+            servers = set(get_domain_from_id(u) for u in users)
 
         if not servers:
             raise SynapseError(400, "Failed to get server list")
@@ -193,7 +194,8 @@ class DirectoryHandler(BaseHandler):
                 Codes.NOT_FOUND
             )
 
-        extra_servers = yield self.store.get_joined_hosts_for_room(room_id)
+        users = yield self.state.get_current_user_in_room(room_id)
+        extra_servers = set(get_domain_from_id(u) for u in users)
         servers = set(extra_servers) | set(servers)
 
         # If this server is in the list of servers, return it first.
@@ -286,13 +288,12 @@ class DirectoryHandler(BaseHandler):
             result = yield as_handler.query_room_alias_exists(room_alias)
         defer.returnValue(result)
 
-    @defer.inlineCallbacks
     def can_modify_alias(self, alias, user_id=None):
         # Any application service "interested" in an alias they are regexing on
         # can modify the alias.
         # Users can only modify the alias if ALL the interested services have
         # non-exclusive locks on the alias (or there are no interested services)
-        services = yield self.store.get_app_services()
+        services = self.store.get_app_services()
         interested_services = [
             s for s in services if s.is_interested_in_alias(alias.to_string())
         ]
@@ -300,14 +301,12 @@ class DirectoryHandler(BaseHandler):
         for service in interested_services:
             if user_id == service.sender:
                 # this user IS the app service so they can do whatever they like
-                defer.returnValue(True)
-                return
+                return defer.succeed(True)
             elif service.is_exclusive_alias(alias.to_string()):
                 # another service has an exclusive lock on this alias.
-                defer.returnValue(False)
-                return
+                return defer.succeed(False)
         # either no interested services, or no service with an exclusive lock
-        defer.returnValue(True)
+        return defer.succeed(True)
 
     @defer.inlineCallbacks
     def _user_can_delete_alias(self, alias, user_id):
@@ -340,3 +339,22 @@ class DirectoryHandler(BaseHandler):
         yield self.auth.check_can_change_room_list(room_id, requester.user)
 
         yield self.store.set_room_is_public(room_id, visibility == "public")
+
+    @defer.inlineCallbacks
+    def edit_published_appservice_room_list(self, appservice_id, network_id,
+                                            room_id, visibility):
+        """Add or remove a room from the appservice/network specific public
+        room list.
+
+        Args:
+            appservice_id (str): ID of the appservice that owns the list
+            network_id (str): The ID of the network the list is associated with
+            room_id (str)
+            visibility (str): either "public" or "private"
+        """
+        if visibility not in ["public", "private"]:
+            raise SynapseError(400, "Invalid visibility setting")
+
+        yield self.store.set_room_is_public_appservice(
+            room_id, appservice_id, network_id, visibility == "public"
+        )

synapse/handlers/e2e_keys.py

@@ -0,0 +1,323 @@
+# -*- coding: utf-8 -*-
+# Copyright 2016 OpenMarket Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import ujson as json
+import logging
+
+from canonicaljson import encode_canonical_json
+from twisted.internet import defer
+
+from synapse.api.errors import SynapseError, CodeMessageException
+from synapse.types import get_domain_from_id
+from synapse.util.logcontext import preserve_fn, preserve_context_over_deferred
+from synapse.util.retryutils import get_retry_limiter, NotRetryingDestination
+
+logger = logging.getLogger(__name__)
+
+
+class E2eKeysHandler(object):
+    def __init__(self, hs):
+        self.store = hs.get_datastore()
+        self.federation = hs.get_replication_layer()
+        self.device_handler = hs.get_device_handler()
+        self.is_mine_id = hs.is_mine_id
+        self.clock = hs.get_clock()
+
+        # doesn't really work as part of the generic query API, because the
+        # query request requires an object POST, but we abuse the
+        # "query handler" interface.
+        self.federation.register_query_handler(
+            "client_keys", self.on_federation_query_client_keys
+        )
+
+    @defer.inlineCallbacks
+    def query_devices(self, query_body, timeout):
+        """ Handle a device key query from a client
+
+        {
+            "device_keys": {
+                "<user_id>": ["<device_id>"]
+            }
+        }
+        ->
+        {
+            "device_keys": {
+                "<user_id>": {
+                    "<device_id>": {
+                        ...
+                    }
+                }
+            }
+        }
+        """
+        device_keys_query = query_body.get("device_keys", {})
+
+        # separate users by domain.
+        # make a map from domain to user_id to device_ids
+        local_query = {}
+        remote_queries = {}
+
+        for user_id, device_ids in device_keys_query.items():
+            if self.is_mine_id(user_id):
+                local_query[user_id] = device_ids
+            else:
+                remote_queries[user_id] = device_ids
+
+        # Firt get local devices.
+        failures = {}
+        results = {}
+        if local_query:
+            local_result = yield self.query_local_devices(local_query)
+            for user_id, keys in local_result.items():
+                if user_id in local_query:
+                    results[user_id] = keys
+
+        # Now attempt to get any remote devices from our local cache.
+        remote_queries_not_in_cache = {}
+        if remote_queries:
+            query_list = []
+            for user_id, device_ids in remote_queries.iteritems():
+                if device_ids:
+                    query_list.extend((user_id, device_id) for device_id in device_ids)
+                else:
+                    query_list.append((user_id, None))
+
+            user_ids_not_in_cache, remote_results = (
+                yield self.store.get_user_devices_from_cache(
+                    query_list
+                )
+            )
+            for user_id, devices in remote_results.iteritems():
+                user_devices = results.setdefault(user_id, {})
+                for device_id, device in devices.iteritems():
+                    keys = device.get("keys", None)
+                    device_display_name = device.get("device_display_name", None)
+                    if keys:
+                        result = dict(keys)
+                        unsigned = result.setdefault("unsigned", {})
+                        if device_display_name:
+                            unsigned["device_display_name"] = device_display_name
+                        user_devices[device_id] = result
+
+            for user_id in user_ids_not_in_cache:
+                domain = get_domain_from_id(user_id)
+                r = remote_queries_not_in_cache.setdefault(domain, {})
+                r[user_id] = remote_queries[user_id]
+
+        # Now fetch any devices that we don't have in our cache
+        @defer.inlineCallbacks
+        def do_remote_query(destination):
+            destination_query = remote_queries_not_in_cache[destination]
+            try:
+                limiter = yield get_retry_limiter(
+                    destination, self.clock, self.store
+                )
+                with limiter:
+                    remote_result = yield self.federation.query_client_keys(
+                        destination,
+                        {"device_keys": destination_query},
+                        timeout=timeout
+                    )
+
+                for user_id, keys in remote_result["device_keys"].items():
+                    if user_id in destination_query:
+                        results[user_id] = keys
+
+            except CodeMessageException as e:
+                failures[destination] = {
+                    "status": e.code, "message": e.message
+                }
+            except NotRetryingDestination as e:
+                failures[destination] = {
+                    "status": 503, "message": "Not ready for retry",
+                }
+            except Exception as e:
+                # include ConnectionRefused and other errors
+                failures[destination] = {
+                    "status": 503, "message": e.message
+                }
+
+        yield preserve_context_over_deferred(defer.gatherResults([
+            preserve_fn(do_remote_query)(destination)
+            for destination in remote_queries_not_in_cache
+        ]))
+
+        defer.returnValue({
+            "device_keys": results, "failures": failures,
+        })
+
+    @defer.inlineCallbacks
+    def query_local_devices(self, query):
+        """Get E2E device keys for local users
+
+        Args:
+            query (dict[string, list[string]|None): map from user_id to a list
+                 of devices to query (None for all devices)
+
+        Returns:
+            defer.Deferred: (resolves to dict[string, dict[string, dict]]):
+                 map from user_id -> device_id -> device details
+        """
+        local_query = []
+
+        result_dict = {}
+        for user_id, device_ids in query.items():
+            if not self.is_mine_id(user_id):
+                logger.warning("Request for keys for non-local user %s",
+                               user_id)
+                raise SynapseError(400, "Not a user here")
+
+            if not device_ids:
+                local_query.append((user_id, None))
+            else:
+                for device_id in device_ids:
+                    local_query.append((user_id, device_id))
+
+            # make sure that each queried user appears in the result dict
+            result_dict[user_id] = {}
+
+        results = yield self.store.get_e2e_device_keys(local_query)
+
+        # Build the result structure, un-jsonify the results, and add the
+        # "unsigned" section
+        for user_id, device_keys in results.items():
+            for device_id, device_info in device_keys.items():
+                r = dict(device_info["keys"])
+                r["unsigned"] = {}
+                display_name = device_info["device_display_name"]
+                if display_name is not None:
+                    r["unsigned"]["device_display_name"] = display_name
+                result_dict[user_id][device_id] = r
+
+        defer.returnValue(result_dict)
+
+    @defer.inlineCallbacks
+    def on_federation_query_client_keys(self, query_body):
+        """ Handle a device key query from a federated server
+        """
+        device_keys_query = query_body.get("device_keys", {})
+        res = yield self.query_local_devices(device_keys_query)
+        defer.returnValue({"device_keys": res})
+
+    @defer.inlineCallbacks
+    def claim_one_time_keys(self, query, timeout):
+        local_query = []
+        remote_queries = {}
+
+        for user_id, device_keys in query.get("one_time_keys", {}).items():
+            if self.is_mine_id(user_id):
+                for device_id, algorithm in device_keys.items():
+                    local_query.append((user_id, device_id, algorithm))
+            else:
+                domain = get_domain_from_id(user_id)
+                remote_queries.setdefault(domain, {})[user_id] = device_keys
+
+        results = yield self.store.claim_e2e_one_time_keys(local_query)
+
+        json_result = {}
+        failures = {}
+        for user_id, device_keys in results.items():
+            for device_id, keys in device_keys.items():
+                for key_id, json_bytes in keys.items():
+                    json_result.setdefault(user_id, {})[device_id] = {
+                        key_id: json.loads(json_bytes)
+                    }
+
+        @defer.inlineCallbacks
+        def claim_client_keys(destination):
+            device_keys = remote_queries[destination]
+            try:
+                limiter = yield get_retry_limiter(
+                    destination, self.clock, self.store
+                )
+                with limiter:
+                    remote_result = yield self.federation.claim_client_keys(
+                        destination,
+                        {"one_time_keys": device_keys},
+                        timeout=timeout
+                    )
+                    for user_id, keys in remote_result["one_time_keys"].items():
+                        if user_id in device_keys:
+                            json_result[user_id] = keys
+            except CodeMessageException as e:
+                failures[destination] = {
+                    "status": e.code, "message": e.message
+                }
+            except NotRetryingDestination as e:
+                failures[destination] = {
+                    "status": 503, "message": "Not ready for retry",
+                }
+            except Exception as e:
+                # include ConnectionRefused and other errors
+                failures[destination] = {
+                    "status": 503, "message": e.message
+                }
+
+        yield preserve_context_over_deferred(defer.gatherResults([
+            preserve_fn(claim_client_keys)(destination)
+            for destination in remote_queries
+        ]))
+
+        defer.returnValue({
+            "one_time_keys": json_result,
+            "failures": failures
+        })
+
+    @defer.inlineCallbacks
+    def upload_keys_for_user(self, user_id, device_id, keys):
+        time_now = self.clock.time_msec()
+
+        # TODO: Validate the JSON to make sure it has the right keys.
+        device_keys = keys.get("device_keys", None)
+        if device_keys:
+            logger.info(
+                "Updating device_keys for device %r for user %s at %d",
+                device_id, user_id, time_now
+            )
+            # TODO: Sign the JSON with the server key
+            changed = yield self.store.set_e2e_device_keys(
+                user_id, device_id, time_now, device_keys,
+            )
+            if changed:
+                # Only notify about device updates *if* the keys actually changed
+                yield self.device_handler.notify_device_update(user_id, [device_id])
+
+        one_time_keys = keys.get("one_time_keys", None)
+        if one_time_keys:
+            logger.info(
+                "Adding %d one_time_keys for device %r for user %r at %d",
+                len(one_time_keys), device_id, user_id, time_now
+            )
+            key_list = []
+            for key_id, key_json in one_time_keys.items():
+                algorithm, key_id = key_id.split(":")
+                key_list.append((
+                    algorithm, key_id, encode_canonical_json(key_json)
+                ))
+
+            yield self.store.add_e2e_one_time_keys(
+                user_id, device_id, time_now, key_list
+            )
+
+        # the device should have been registered already, but it may have been
+        # deleted due to a race with a DELETE request. Or we may be using an
+        # old access_token without an associated device_id. Either way, we
+        # need to double-check the device is registered to avoid ending up with
+        # keys without a corresponding device.
+        self.device_handler.check_device_registered(user_id, device_id)
+
+        result = yield self.store.count_e2e_one_time_keys(user_id, device_id)
+
+        defer.returnValue({"one_time_key_counts": result})

synapse/handlers/events.py

@@ -47,6 +47,7 @@ class EventStreamHandler(BaseHandler):
         self.clock = hs.get_clock()
 
         self.notifier = hs.get_notifier()
+        self.state = hs.get_state_handler()
 
     @defer.inlineCallbacks
     @log_function
@@ -90,7 +91,7 @@ class EventStreamHandler(BaseHandler):
                     # Send down presence.
                     if event.state_key == auth_user_id:
                         # Send down presence for everyone in the room.
-                        users = yield self.store.get_users_in_room(event.room_id)
+                        users = yield self.state.get_current_user_in_room(event.room_id)
                         states = yield presence_handler.get_states(
                             users,
                             as_event=True,

synapse/handlers/federation.py

@@ -26,7 +26,10 @@ from synapse.api.errors import (
 from synapse.api.constants import EventTypes, Membership, RejectedReason
 from synapse.events.validator import EventValidator
 from synapse.util import unwrapFirstError
-from synapse.util.logcontext import PreserveLoggingContext, preserve_fn
+from synapse.util.logcontext import (
+    PreserveLoggingContext, preserve_fn, preserve_context_over_deferred
+)
+from synapse.util.metrics import measure_func
 from synapse.util.logutils import log_function
 from synapse.util.async import run_on_reactor
 from synapse.util.frozenutils import unfreeze
@@ -77,27 +80,14 @@ class FederationHandler(BaseHandler):
         # When joining a room we need to queue any events for that room up
         self.room_queues = {}
 
-    def handle_new_event(self, event, destinations):
-        """ Takes in an event from the client to server side, that has already
-        been authed and handled by the state module, and sends it to any
-        remote home servers that may be interested.
-
-        Args:
-            event: The event to send
-            destinations: A list of destinations to send it to
-
-        Returns:
-            Deferred: Resolved when it has successfully been queued for
-            processing.
-        """
-
-        return self.replication_layer.send_pdu(event, destinations)
-
     @log_function
     @defer.inlineCallbacks
     def on_receive_pdu(self, origin, pdu, state=None, auth_chain=None):
         """ Called by the ReplicationLayer when we have a new pdu. We need to
         do auth checks and put it through the StateHandler.
+
+        auth_chain and state are None if we already have the necessary state
+        and prev_events in the db
         """
         event = pdu
 
@@ -115,16 +105,25 @@ class FederationHandler(BaseHandler):
 
         # FIXME (erikj): Awful hack to make the case where we are not currently
         # in the room work
-        is_in_room = yield self.auth.check_host_in_room(
-            event.room_id,
-            self.server_name
-        )
-        if not is_in_room and not event.internal_metadata.is_outlier():
-            logger.debug("Got event for room we're not in.")
+        # If state and auth_chain are None, then we don't need to do this check
+        # as we already know we have enough state in the DB to handle this
+        # event.
+        if state and auth_chain and not event.internal_metadata.is_outlier():
+            is_in_room = yield self.auth.check_host_in_room(
+                event.room_id,
+                self.server_name
+            )
+        else:
+            is_in_room = True
+        if not is_in_room:
+            logger.info(
+                "Got event for room we're not in: %r %r",
+                event.room_id, event.event_id
+            )
 
             try:
                 event_stream_id, max_stream_id = yield self._persist_auth_tree(
-                    auth_chain, state, event
+                    origin, auth_chain, state, event
                 )
             except AuthError as e:
                 raise FederationError(
@@ -215,17 +214,28 @@ class FederationHandler(BaseHandler):
 
         if event.type == EventTypes.Member:
             if event.membership == Membership.JOIN:
-                prev_state = context.current_state.get((event.type, event.state_key))
-                if not prev_state or prev_state.membership != Membership.JOIN:
-                    # Only fire user_joined_room if the user has acutally
-                    # joined the room. Don't bother if the user is just
-                    # changing their profile info.
+                # Only fire user_joined_room if the user has acutally
+                # joined the room. Don't bother if the user is just
+                # changing their profile info.
+                newly_joined = True
+                prev_state_id = context.prev_state_ids.get(
+                    (event.type, event.state_key)
+                )
+                if prev_state_id:
+                    prev_state = yield self.store.get_event(
+                        prev_state_id, allow_none=True,
+                    )
+                    if prev_state and prev_state.membership == Membership.JOIN:
+                        newly_joined = False
+
+                if newly_joined:
                     user = UserID.from_string(event.state_key)
                     yield user_joined_room(self.distributor, user, event.room_id)
 
+    @measure_func("_filter_events_for_server")
     @defer.inlineCallbacks
     def _filter_events_for_server(self, server_name, room_id, events):
-        event_to_state = yield self.store.get_state_for_events(
+        event_to_state_ids = yield self.store.get_state_ids_for_events(
             frozenset(e.event_id for e in events),
             types=(
                 (EventTypes.RoomHistoryVisibility, ""),
@@ -233,6 +243,33 @@ class FederationHandler(BaseHandler):
             )
         )
 
+        # We only want to pull out member events that correspond to the
+        # server's domain.
+
+        def check_match(id):
+            try:
+                return server_name == get_domain_from_id(id)
+            except:
+                return False
+
+        # Parses mapping `event_id -> (type, state_key) -> state event_id`
+        # to get all state ids that we're interested in.
+        event_map = yield self.store.get_events([
+            e_id
+            for key_to_eid in event_to_state_ids.values()
+            for key, e_id in key_to_eid.items()
+            if key[0] != EventTypes.Member or check_match(key[1])
+        ])
+
+        event_to_state = {
+            e_id: {
+                key: event_map[inner_e_id]
+                for key, inner_e_id in key_to_eid.items()
+                if inner_e_id in event_map
+            }
+            for e_id, key_to_eid in event_to_state_ids.items()
+        }
+
         def redact_disallowed(event, state):
             if not state:
                 return event
@@ -249,7 +286,7 @@ class FederationHandler(BaseHandler):
                         if ev.type != EventTypes.Member:
                             continue
                         try:
-                            domain = UserID.from_string(ev.state_key).domain
+                            domain = get_domain_from_id(ev.state_key)
                         except:
                             continue
 
@@ -274,7 +311,7 @@ class FederationHandler(BaseHandler):
 
     @log_function
     @defer.inlineCallbacks
-    def backfill(self, dest, room_id, limit, extremities=[]):
+    def backfill(self, dest, room_id, limit, extremities):
         """ Trigger a backfill request to `dest` for the given `room_id`
 
         This will attempt to get more events from the remote. This may return
@@ -284,9 +321,6 @@ class FederationHandler(BaseHandler):
         if dest == self.server_name:
             raise SynapseError(400, "Can't backfill from self.")
 
-        if not extremities:
-            extremities = yield self.store.get_oldest_events_in_room(room_id)
-
         events = yield self.replication_layer.backfill(
             dest,
             room_id,
@@ -335,32 +369,61 @@ class FederationHandler(BaseHandler):
             state_events.update({s.event_id: s for s in state})
             events_to_state[e_id] = state
 
+        required_auth = set(
+            a_id
+            for event in events + state_events.values() + auth_events.values()
+            for a_id, _ in event.auth_events
+        )
+        auth_events.update({
+            e_id: event_map[e_id] for e_id in required_auth if e_id in event_map
+        })
+        missing_auth = required_auth - set(auth_events)
+        failed_to_fetch = set()
+
+        # Try and fetch any missing auth events from both DB and remote servers.
+        # We repeatedly do this until we stop finding new auth events.
+        while missing_auth - failed_to_fetch:
+            logger.info("Missing auth for backfill: %r", missing_auth)
+            ret_events = yield self.store.get_events(missing_auth - failed_to_fetch)
+            auth_events.update(ret_events)
+
+            required_auth.update(
+                a_id for event in ret_events.values() for a_id, _ in event.auth_events
+            )
+            missing_auth = required_auth - set(auth_events)
+
+            if missing_auth - failed_to_fetch:
+                logger.info(
+                    "Fetching missing auth for backfill: %r",
+                    missing_auth - failed_to_fetch
+                )
+
+                results = yield preserve_context_over_deferred(defer.gatherResults(
+                    [
+                        preserve_fn(self.replication_layer.get_pdu)(
+                            [dest],
+                            event_id,
+                            outlier=True,
+                            timeout=10000,
+                        )
+                        for event_id in missing_auth - failed_to_fetch
+                    ],
+                    consumeErrors=True
+                )).addErrback(unwrapFirstError)
+                auth_events.update({a.event_id: a for a in results if a})
+                required_auth.update(
+                    a_id
+                    for event in results if event
+                    for a_id, _ in event.auth_events
+                )
+                missing_auth = required_auth - set(auth_events)
+
+                failed_to_fetch = missing_auth - set(auth_events)
+
         seen_events = yield self.store.have_events(
             set(auth_events.keys()) | set(state_events.keys())
         )
 
-        all_events = events + state_events.values() + auth_events.values()
-        required_auth = set(
-            a_id for event in all_events for a_id, _ in event.auth_events
-        )
-
-        missing_auth = required_auth - set(auth_events)
-        if missing_auth:
-            logger.info("Missing auth for backfill: %r", missing_auth)
-            results = yield defer.gatherResults(
-                [
-                    self.replication_layer.get_pdu(
-                        [dest],
-                        event_id,
-                        outlier=True,
-                        timeout=10000,
-                    )
-                    for event_id in missing_auth
-                ],
-                consumeErrors=True
-            ).addErrback(unwrapFirstError)
-            auth_events.update({a.event_id: a for a in results})
-
         ev_infos = []
         for a in auth_events.values():
             if a.event_id in seen_events:
@@ -372,6 +435,7 @@ class FederationHandler(BaseHandler):
                     (auth_events[a_id].type, auth_events[a_id].state_key):
                     auth_events[a_id]
                     for a_id, _ in a.auth_events
+                    if a_id in auth_events
                 }
             })
 
@@ -383,6 +447,7 @@ class FederationHandler(BaseHandler):
                     (auth_events[a_id].type, auth_events[a_id].state_key):
                     auth_events[a_id]
                     for a_id, _ in event_map[e_id].auth_events
+                    if a_id in auth_events
                 }
             })
 
@@ -426,6 +491,10 @@ class FederationHandler(BaseHandler):
         )
         max_depth = sorted_extremeties_tuple[0][1]
 
+        # We don't want to specify too many extremities as it causes the backfill
+        # request URI to be too long.
+        extremities = dict(sorted_extremeties_tuple[:5])
+
         if current_depth > max_depth:
             logger.debug(
                 "Not backfilling as we don't need to. %d < %d",
@@ -522,11 +591,24 @@ class FederationHandler(BaseHandler):
 
         event_ids = list(extremities.keys())
 
-        states = yield defer.gatherResults([
-            self.state_handler.resolve_state_groups(room_id, [e])
+        logger.debug("calling resolve_state_groups in _maybe_backfill")
+        states = yield preserve_context_over_deferred(defer.gatherResults([
+            preserve_fn(self.state_handler.resolve_state_groups)(room_id, [e])
             for e in event_ids
-        ])
-        states = dict(zip(event_ids, [s[1] for s in states]))
+        ]))
+        states = dict(zip(event_ids, [s.state for s in states]))
+
+        state_map = yield self.store.get_events(
+            [e_id for ids in states.values() for e_id in ids],
+            get_prev_content=False
+        )
+        states = {
+            key: {
+                k: state_map[e_id]
+                for k, e_id in state_dict.items()
+                if e_id in state_map
+            } for key, state_dict in states.items()
+        }
 
         for e_id, _ in sorted_extremeties_tuple:
             likely_domains = get_domains_from_state(states[e_id])
@@ -637,7 +719,7 @@ class FederationHandler(BaseHandler):
                 pass
 
             event_stream_id, max_stream_id = yield self._persist_auth_tree(
-                auth_chain, state, event
+                origin, auth_chain, state, event
             )
 
             with PreserveLoggingContext():
@@ -688,7 +770,9 @@ class FederationHandler(BaseHandler):
             logger.warn("Failed to create join %r because %s", event, e)
             raise e
 
-        self.auth.check(event, auth_events=context.current_state)
+        # The remote hasn't signed it yet, obviously. We'll do the full checks
+        # when we get the event back in `on_send_join_request`
+        yield self.auth.check_from_context(event, context, do_sig_check=False)
 
         defer.returnValue(event)
 
@@ -707,6 +791,10 @@ class FederationHandler(BaseHandler):
         )
 
         event.internal_metadata.outlier = False
+        # Send this event on behalf of the origin server since they may not
+        # have an up to data view of the state of the room at this event so
+        # will not know which servers to send the event to.
+        event.internal_metadata.send_on_behalf_of = origin
 
         context, event_stream_id, max_stream_id = yield self._handle_new_event(
             origin, event
@@ -734,37 +822,15 @@ class FederationHandler(BaseHandler):
                 user = UserID.from_string(event.state_key)
                 yield user_joined_room(self.distributor, user, event.room_id)
 
-        new_pdu = event
-
-        destinations = set()
-
-        for k, s in context.current_state.items():
-            try:
-                if k[0] == EventTypes.Member:
-                    if s.content["membership"] == Membership.JOIN:
-                        destinations.add(get_domain_from_id(s.state_key))
-            except:
-                logger.warn(
-                    "Failed to get destination from event %s", s.event_id
-                )
-
-        destinations.discard(origin)
-
-        logger.debug(
-            "on_send_join_request: Sending event: %s, signatures: %s",
-            event.event_id,
-            event.signatures,
-        )
-
-        self.replication_layer.send_pdu(new_pdu, destinations)
-
-        state_ids = [e.event_id for e in context.current_state.values()]
+        state_ids = context.prev_state_ids.values()
         auth_chain = yield self.store.get_auth_chain(set(
             [event.event_id] + state_ids
         ))
 
+        state = yield self.store.get_events(context.prev_state_ids.values())
+
         defer.returnValue({
-            "state": context.current_state.values(),
+            "state": state.values(),
             "auth_chain": auth_chain,
         })
 
@@ -918,7 +984,9 @@ class FederationHandler(BaseHandler):
         )
 
         try:
-            self.auth.check(event, auth_events=context.current_state)
+            # The remote hasn't signed it yet, obviously. We'll do the full checks
+            # when we get the event back in `on_send_leave_request`
+            yield self.auth.check_from_context(event, context, do_sig_check=False)
         except AuthError as e:
             logger.warn("Failed to create new leave %r because %s", event, e)
             raise e
@@ -960,41 +1028,14 @@ class FederationHandler(BaseHandler):
                 event, event_stream_id, max_stream_id, extra_users=extra_users
             )
 
-        new_pdu = event
-
-        destinations = set()
-
-        for k, s in context.current_state.items():
-            try:
-                if k[0] == EventTypes.Member:
-                    if s.content["membership"] == Membership.LEAVE:
-                        destinations.add(get_domain_from_id(s.state_key))
-            except:
-                logger.warn(
-                    "Failed to get destination from event %s", s.event_id
-                )
-
-        destinations.discard(origin)
-
-        logger.debug(
-            "on_send_leave_request: Sending event: %s, signatures: %s",
-            event.event_id,
-            event.signatures,
-        )
-
-        self.replication_layer.send_pdu(new_pdu, destinations)
-
         defer.returnValue(None)
 
     @defer.inlineCallbacks
-    def get_state_for_pdu(self, origin, room_id, event_id, do_auth=True):
+    def get_state_for_pdu(self, room_id, event_id):
+        """Returns the state at the event. i.e. not including said event.
+        """
         yield run_on_reactor()
 
-        if do_auth:
-            in_room = yield self.auth.check_host_in_room(room_id, origin)
-            if not in_room:
-                raise AuthError(403, "Host not in room.")
-
         state_groups = yield self.store.get_state_groups(
             room_id, [event_id]
         )
@@ -1033,6 +1074,34 @@ class FederationHandler(BaseHandler):
         else:
             defer.returnValue([])
 
+    @defer.inlineCallbacks
+    def get_state_ids_for_pdu(self, room_id, event_id):
+        """Returns the state at the event. i.e. not including said event.
+        """
+        yield run_on_reactor()
+
+        state_groups = yield self.store.get_state_groups_ids(
+            room_id, [event_id]
+        )
+
+        if state_groups:
+            _, state = state_groups.items().pop()
+            results = state
+
+            event = yield self.store.get_event(event_id)
+            if event and event.is_state():
+                # Get previous state
+                if "replaces_state" in event.unsigned:
+                    prev_id = event.unsigned["replaces_state"]
+                    if prev_id != event.event_id:
+                        results[(event.type, event.state_key)] = prev_id
+                else:
+                    del results[(event.type, event.state_key)]
+
+            defer.returnValue(results.values())
+        else:
+            defer.returnValue([])
+
     @defer.inlineCallbacks
     @log_function
     def on_backfill_request(self, origin, room_id, pdu_list, limit):
@@ -1065,16 +1134,17 @@ class FederationHandler(BaseHandler):
         )
 
         if event:
-            # FIXME: This is a temporary work around where we occasionally
-            # return events slightly differently than when they were
-            # originally signed
-            event.signatures.update(
-                compute_event_signature(
-                    event,
-                    self.hs.hostname,
-                    self.hs.config.signing_key[0]
+            if self.hs.is_mine_id(event.event_id):
+                # FIXME: This is a temporary work around where we occasionally
+                # return events slightly differently than when they were
+                # originally signed
+                event.signatures.update(
+                    compute_event_signature(
+                        event,
+                        self.hs.hostname,
+                        self.hs.config.signing_key[0]
+                    )
                 )
-            )
 
             if do_auth:
                 in_room = yield self.auth.check_host_in_room(
@@ -1084,6 +1154,12 @@ class FederationHandler(BaseHandler):
                 if not in_room:
                     raise AuthError(403, "Host not in room.")
 
+                events = yield self._filter_events_for_server(
+                    origin, event.room_id, [event]
+                )
+
+                event = events[0]
+
             defer.returnValue(event)
         else:
             defer.returnValue(None)
@@ -1114,11 +1190,12 @@ class FederationHandler(BaseHandler):
             backfilled=backfilled,
         )
 
-        # this intentionally does not yield: we don't care about the result
-        # and don't need to wait for it.
-        preserve_fn(self.hs.get_pusherpool().on_new_notifications)(
-            event_stream_id, max_stream_id
-        )
+        if not backfilled:
+            # this intentionally does not yield: we don't care about the result
+            # and don't need to wait for it.
+            preserve_fn(self.hs.get_pusherpool().on_new_notifications)(
+                event_stream_id, max_stream_id
+            )
 
         defer.returnValue((context, event_stream_id, max_stream_id))
 
@@ -1129,9 +1206,9 @@ class FederationHandler(BaseHandler):
         a bunch of outliers, but not a chunk of individual events that depend
         on each other for state calculations.
         """
-        contexts = yield defer.gatherResults(
+        contexts = yield preserve_context_over_deferred(defer.gatherResults(
             [
-                self._prep_event(
+                preserve_fn(self._prep_event)(
                     origin,
                     ev_info["event"],
                     state=ev_info.get("state"),
@@ -1139,7 +1216,7 @@ class FederationHandler(BaseHandler):
                 )
                 for ev_info in event_infos
             ]
-        )
+        ))
 
         yield self.store.persist_events(
             [
@@ -1150,11 +1227,19 @@ class FederationHandler(BaseHandler):
         )
 
     @defer.inlineCallbacks
-    def _persist_auth_tree(self, auth_events, state, event):
+    def _persist_auth_tree(self, origin, auth_events, state, event):
         """Checks the auth chain is valid (and passes auth checks) for the
         state and event. Then persists the auth chain and state atomically.
         Persists the event seperately.
 
+        Will attempt to fetch missing auth events.
+
+        Args:
+            origin (str): Where the events came from
+            auth_events (list)
+            state (list)
+            event (Event)
+
         Returns:
             2-tuple of (event_stream_id, max_stream_id) from the persist_event
             call for `event`
@@ -1167,7 +1252,7 @@ class FederationHandler(BaseHandler):
 
         event_map = {
             e.event_id: e
-            for e in auth_events
+            for e in itertools.chain(auth_events, state, [event])
         }
 
         create_event = None
@@ -1176,10 +1261,29 @@ class FederationHandler(BaseHandler):
                 create_event = e
                 break
 
+        missing_auth_events = set()
+        for e in itertools.chain(auth_events, state, [event]):
+            for e_id, _ in e.auth_events:
+                if e_id not in event_map:
+                    missing_auth_events.add(e_id)
+
+        for e_id in missing_auth_events:
+            m_ev = yield self.replication_layer.get_pdu(
+                [origin],
+                e_id,
+                outlier=True,
+                timeout=10000,
+            )
+            if m_ev and m_ev.event_id == e_id:
+                event_map[e_id] = m_ev
+            else:
+                logger.info("Failed to find auth event %r", e_id)
+
         for e in itertools.chain(auth_events, state, [event]):
             auth_for_e = {
                 (event_map[e_id].type, event_map[e_id].state_key): event_map[e_id]
                 for e_id, _ in e.auth_events
+                if e_id in event_map
             }
             if create_event:
                 auth_for_e[(EventTypes.Create, "")] = create_event
@@ -1215,7 +1319,6 @@ class FederationHandler(BaseHandler):
 
         event_stream_id, max_stream_id = yield self.store.persist_event(
             event, new_event_context,
-            current_state=state,
         )
 
         defer.returnValue((event_stream_id, max_stream_id))
@@ -1228,7 +1331,13 @@ class FederationHandler(BaseHandler):
         )
 
         if not auth_events:
-            auth_events = context.current_state
+            auth_events_ids = yield self.auth.compute_auth_events(
+                event, context.prev_state_ids, for_verification=True,
+            )
+            auth_events = yield self.store.get_events(auth_events_ids)
+            auth_events = {
+                (e.type, e.state_key): e for e in auth_events.values()
+            }
 
         # This is a hack to fix some old rooms where the initial join event
         # didn't reference the create event in its auth events.
@@ -1254,8 +1363,7 @@ class FederationHandler(BaseHandler):
             context.rejected = RejectedReason.AUTH_ERROR
 
         if event.type == EventTypes.GuestAccess:
-            full_context = yield self.store.get_current_state(room_id=event.room_id)
-            yield self.maybe_kick_guest_users(event, full_context)
+            yield self.maybe_kick_guest_users(event)
 
         defer.returnValue(context)
 
@@ -1323,6 +1431,11 @@ class FederationHandler(BaseHandler):
         current_state = set(e.event_id for e in auth_events.values())
         event_auth_events = set(e_id for e_id, _ in event.auth_events)
 
+        if event.is_state():
+            event_key = (event.type, event.state_key)
+        else:
+            event_key = None
+
         if event_auth_events - current_state:
             have_events = yield self.store.have_events(
                 event_auth_events - current_state
@@ -1396,9 +1509,9 @@ class FederationHandler(BaseHandler):
             # Do auth conflict res.
             logger.info("Different auth: %s", different_auth)
 
-            different_events = yield defer.gatherResults(
+            different_events = yield preserve_context_over_deferred(defer.gatherResults(
                 [
-                    self.store.get_event(
+                    preserve_fn(self.store.get_event)(
                         d,
                         allow_none=True,
                         allow_rejected=False,
@@ -1407,7 +1520,7 @@ class FederationHandler(BaseHandler):
                     if d in have_events and not have_events[d]
                 ],
                 consumeErrors=True
-            ).addErrback(unwrapFirstError)
+            )).addErrback(unwrapFirstError)
 
             if different_events:
                 local_view = dict(auth_events)
@@ -1416,7 +1529,7 @@ class FederationHandler(BaseHandler):
                     (d.type, d.state_key): d for d in different_events if d
                 })
 
-                new_state, prev_state = self.state_handler.resolve_events(
+                new_state = self.state_handler.resolve_events(
                     [local_view.values(), remote_view.values()],
                     event
                 )
@@ -1426,8 +1539,16 @@ class FederationHandler(BaseHandler):
                 current_state = set(e.event_id for e in auth_events.values())
                 different_auth = event_auth_events - current_state
 
-                context.current_state.update(auth_events)
-                context.state_group = None
+                context.current_state_ids = dict(context.current_state_ids)
+                context.current_state_ids.update({
+                    k: a.event_id for k, a in auth_events.items()
+                    if k != event_key
+                })
+                context.prev_state_ids = dict(context.prev_state_ids)
+                context.prev_state_ids.update({
+                    k: a.event_id for k, a in auth_events.items()
+                })
+                context.state_group = self.store.get_next_state_group()
 
         if different_auth and not event.internal_metadata.is_outlier():
             logger.info("Different auth after resolution: %s", different_auth)
@@ -1448,8 +1569,8 @@ class FederationHandler(BaseHandler):
 
             if do_resolution:
                 # 1. Get what we think is the auth chain.
-                auth_ids = self.auth.compute_auth_events(
-                    event, context.current_state
+                auth_ids = yield self.auth.compute_auth_events(
+                    event, context.prev_state_ids
                 )
                 local_auth_chain = yield self.store.get_auth_chain(auth_ids)
 
@@ -1505,8 +1626,16 @@ class FederationHandler(BaseHandler):
                 # 4. Look at rejects and their proofs.
                 # TODO.
 
-                context.current_state.update(auth_events)
-                context.state_group = None
+                context.current_state_ids = dict(context.current_state_ids)
+                context.current_state_ids.update({
+                    k: a.event_id for k, a in auth_events.items()
+                    if k != event_key
+                })
+                context.prev_state_ids = dict(context.prev_state_ids)
+                context.prev_state_ids.update({
+                    k: a.event_id for k, a in auth_events.items()
+                })
+                context.state_group = self.store.get_next_state_group()
 
         try:
             self.auth.check(event, auth_events=auth_events)
@@ -1692,12 +1821,12 @@ class FederationHandler(BaseHandler):
             )
 
             try:
-                self.auth.check(event, context.current_state)
+                yield self.auth.check_from_context(event, context)
             except AuthError as e:
                 logger.warn("Denying new third party invite %r because %s", event, e)
                 raise e
 
-            yield self._check_signature(event, auth_events=context.current_state)
+            yield self._check_signature(event, context)
             member_handler = self.hs.get_handlers().room_member_handler
             yield member_handler.send_membership_event(None, event, context)
         else:
@@ -1723,11 +1852,11 @@ class FederationHandler(BaseHandler):
         )
 
         try:
-            self.auth.check(event, auth_events=context.current_state)
+            self.auth.check_from_context(event, context)
         except AuthError as e:
             logger.warn("Denying third party invite %r because %s", event, e)
             raise e
-        yield self._check_signature(event, auth_events=context.current_state)
+        yield self._check_signature(event, context)
 
         returned_invite = yield self.send_invite(origin, event)
         # TODO: Make sure the signatures actually are correct.
@@ -1741,16 +1870,24 @@ class FederationHandler(BaseHandler):
             EventTypes.ThirdPartyInvite,
             event.content["third_party_invite"]["signed"]["token"]
         )
-        original_invite = context.current_state.get(key)
-        if not original_invite:
-            logger.info(
-                "Could not find invite event for third_party_invite - "
-                "discarding: %s" % (event_dict,)
+        original_invite = None
+        original_invite_id = context.prev_state_ids.get(key)
+        if original_invite_id:
+            original_invite = yield self.store.get_event(
+                original_invite_id, allow_none=True
             )
-            return
+        if original_invite:
+            display_name = original_invite.content["display_name"]
+            event_dict["content"]["third_party_invite"]["display_name"] = display_name
+        else:
+            logger.info(
+                "Could not find invite event for third_party_invite: %r",
+                event_dict
+            )
+            # We don't discard here as this is not the appropriate place to do
+            # auth checks. If we need the invite and don't have it then the
+            # auth check code will explode appropriately.
 
-        display_name = original_invite.content["display_name"]
-        event_dict["content"]["third_party_invite"]["display_name"] = display_name
         builder = self.event_builder_factory.new(event_dict)
         EventValidator().validate_new(builder)
         message_handler = self.hs.get_handlers().message_handler
@@ -1758,13 +1895,13 @@ class FederationHandler(BaseHandler):
         defer.returnValue((event, context))
 
     @defer.inlineCallbacks
-    def _check_signature(self, event, auth_events):
+    def _check_signature(self, event, context):
         """
         Checks that the signature in the event is consistent with its invite.
 
         Args:
             event (Event): The m.room.member event to check
-            auth_events (dict<(event type, state_key), event>):
+            context (EventContext):
 
         Raises:
             AuthError: if signature didn't match any keys, or key has been
@@ -1775,10 +1912,14 @@ class FederationHandler(BaseHandler):
         signed = event.content["third_party_invite"]["signed"]
         token = signed["token"]
 
-        invite_event = auth_events.get(
+        invite_event_id = context.prev_state_ids.get(
             (EventTypes.ThirdPartyInvite, token,)
         )
 
+        invite_event = None
+        if invite_event_id:
+            invite_event = yield self.store.get_event(invite_event_id, allow_none=True)
+
         if not invite_event:
             raise AuthError(403, "Could not find invite")
 

synapse/handlers/initial_sync.py

@@ -0,0 +1,444 @@
+# -*- coding: utf-8 -*-
+# Copyright 2016 OpenMarket Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from twisted.internet import defer
+
+from synapse.api.constants import EventTypes, Membership
+from synapse.api.errors import AuthError, Codes
+from synapse.events.utils import serialize_event
+from synapse.events.validator import EventValidator
+from synapse.streams.config import PaginationConfig
+from synapse.types import (
+    UserID, StreamToken,
+)
+from synapse.util import unwrapFirstError
+from synapse.util.async import concurrently_execute
+from synapse.util.caches.snapshot_cache import SnapshotCache
+from synapse.util.logcontext import preserve_fn, preserve_context_over_deferred
+from synapse.visibility import filter_events_for_client
+
+from ._base import BaseHandler
+
+import logging
+
+
+logger = logging.getLogger(__name__)
+
+
+class InitialSyncHandler(BaseHandler):
+    def __init__(self, hs):
+        super(InitialSyncHandler, self).__init__(hs)
+        self.hs = hs
+        self.state = hs.get_state_handler()
+        self.clock = hs.get_clock()
+        self.validator = EventValidator()
+        self.snapshot_cache = SnapshotCache()
+
+    def snapshot_all_rooms(self, user_id=None, pagin_config=None,
+                           as_client_event=True, include_archived=False):
+        """Retrieve a snapshot of all rooms the user is invited or has joined.
+
+        This snapshot may include messages for all rooms where the user is
+        joined, depending on the pagination config.
+
+        Args:
+            user_id (str): The ID of the user making the request.
+            pagin_config (synapse.api.streams.PaginationConfig): The pagination
+            config used to determine how many messages *PER ROOM* to return.
+            as_client_event (bool): True to get events in client-server format.
+            include_archived (bool): True to get rooms that the user has left
+        Returns:
+            A list of dicts with "room_id" and "membership" keys for all rooms
+            the user is currently invited or joined in on. Rooms where the user
+            is joined on, may return a "messages" key with messages, depending
+            on the specified PaginationConfig.
+        """
+        key = (
+            user_id,
+            pagin_config.from_token,
+            pagin_config.to_token,
+            pagin_config.direction,
+            pagin_config.limit,
+            as_client_event,
+            include_archived,
+        )
+        now_ms = self.clock.time_msec()
+        result = self.snapshot_cache.get(now_ms, key)
+        if result is not None:
+            return result
+
+        return self.snapshot_cache.set(now_ms, key, self._snapshot_all_rooms(
+            user_id, pagin_config, as_client_event, include_archived
+        ))
+
+    @defer.inlineCallbacks
+    def _snapshot_all_rooms(self, user_id=None, pagin_config=None,
+                            as_client_event=True, include_archived=False):
+
+        memberships = [Membership.INVITE, Membership.JOIN]
+        if include_archived:
+            memberships.append(Membership.LEAVE)
+
+        room_list = yield self.store.get_rooms_for_user_where_membership_is(
+            user_id=user_id, membership_list=memberships
+        )
+
+        user = UserID.from_string(user_id)
+
+        rooms_ret = []
+
+        now_token = yield self.hs.get_event_sources().get_current_token()
+
+        presence_stream = self.hs.get_event_sources().sources["presence"]
+        pagination_config = PaginationConfig(from_token=now_token)
+        presence, _ = yield presence_stream.get_pagination_rows(
+            user, pagination_config.get_source_config("presence"), None
+        )
+
+        receipt_stream = self.hs.get_event_sources().sources["receipt"]
+        receipt, _ = yield receipt_stream.get_pagination_rows(
+            user, pagination_config.get_source_config("receipt"), None
+        )
+
+        tags_by_room = yield self.store.get_tags_for_user(user_id)
+
+        account_data, account_data_by_room = (
+            yield self.store.get_account_data_for_user(user_id)
+        )
+
+        public_room_ids = yield self.store.get_public_room_ids()
+
+        limit = pagin_config.limit
+        if limit is None:
+            limit = 10
+
+        @defer.inlineCallbacks
+        def handle_room(event):
+            d = {
+                "room_id": event.room_id,
+                "membership": event.membership,
+                "visibility": (
+                    "public" if event.room_id in public_room_ids
+                    else "private"
+                ),
+            }
+
+            if event.membership == Membership.INVITE:
+                time_now = self.clock.time_msec()
+                d["inviter"] = event.sender
+
+                invite_event = yield self.store.get_event(event.event_id)
+                d["invite"] = serialize_event(invite_event, time_now, as_client_event)
+
+            rooms_ret.append(d)
+
+            if event.membership not in (Membership.JOIN, Membership.LEAVE):
+                return
+
+            try:
+                if event.membership == Membership.JOIN:
+                    room_end_token = now_token.room_key
+                    deferred_room_state = self.state_handler.get_current_state(
+                        event.room_id
+                    )
+                elif event.membership == Membership.LEAVE:
+                    room_end_token = "s%d" % (event.stream_ordering,)
+                    deferred_room_state = self.store.get_state_for_events(
+                        [event.event_id], None
+                    )
+                    deferred_room_state.addCallback(
+                        lambda states: states[event.event_id]
+                    )
+
+                (messages, token), current_state = yield preserve_context_over_deferred(
+                    defer.gatherResults(
+                        [
+                            preserve_fn(self.store.get_recent_events_for_room)(
+                                event.room_id,
+                                limit=limit,
+                                end_token=room_end_token,
+                            ),
+                            deferred_room_state,
+                        ]
+                    )
+                ).addErrback(unwrapFirstError)
+
+                messages = yield filter_events_for_client(
+                    self.store, user_id, messages
+                )
+
+                start_token = now_token.copy_and_replace("room_key", token[0])
+                end_token = now_token.copy_and_replace("room_key", token[1])
+                time_now = self.clock.time_msec()
+
+                d["messages"] = {
+                    "chunk": [
+                        serialize_event(m, time_now, as_client_event)
+                        for m in messages
+                    ],
+                    "start": start_token.to_string(),
+                    "end": end_token.to_string(),
+                }
+
+                d["state"] = [
+                    serialize_event(c, time_now, as_client_event)
+                    for c in current_state.values()
+                ]
+
+                account_data_events = []
+                tags = tags_by_room.get(event.room_id)
+                if tags:
+                    account_data_events.append({
+                        "type": "m.tag",
+                        "content": {"tags": tags},
+                    })
+
+                account_data = account_data_by_room.get(event.room_id, {})
+                for account_data_type, content in account_data.items():
+                    account_data_events.append({
+                        "type": account_data_type,
+                        "content": content,
+                    })
+
+                d["account_data"] = account_data_events
+            except:
+                logger.exception("Failed to get snapshot")
+
+        yield concurrently_execute(handle_room, room_list, 10)
+
+        account_data_events = []
+        for account_data_type, content in account_data.items():
+            account_data_events.append({
+                "type": account_data_type,
+                "content": content,
+            })
+
+        ret = {
+            "rooms": rooms_ret,
+            "presence": presence,
+            "account_data": account_data_events,
+            "receipts": receipt,
+            "end": now_token.to_string(),
+        }
+
+        defer.returnValue(ret)
+
+    @defer.inlineCallbacks
+    def room_initial_sync(self, requester, room_id, pagin_config=None):
+        """Capture the a snapshot of a room. If user is currently a member of
+        the room this will be what is currently in the room. If the user left
+        the room this will be what was in the room when they left.
+
+        Args:
+            requester(Requester): The user to get a snapshot for.
+            room_id(str): The room to get a snapshot of.
+            pagin_config(synapse.streams.config.PaginationConfig):
+                The pagination config used to determine how many messages to
+                return.
+        Raises:
+            AuthError if the user wasn't in the room.
+        Returns:
+            A JSON serialisable dict with the snapshot of the room.
+        """
+
+        user_id = requester.user.to_string()
+
+        membership, member_event_id = yield self._check_in_room_or_world_readable(
+            room_id, user_id,
+        )
+        is_peeking = member_event_id is None
+
+        if membership == Membership.JOIN:
+            result = yield self._room_initial_sync_joined(
+                user_id, room_id, pagin_config, membership, is_peeking
+            )
+        elif membership == Membership.LEAVE:
+            result = yield self._room_initial_sync_parted(
+                user_id, room_id, pagin_config, membership, member_event_id, is_peeking
+            )
+
+        account_data_events = []
+        tags = yield self.store.get_tags_for_room(user_id, room_id)
+        if tags:
+            account_data_events.append({
+                "type": "m.tag",
+                "content": {"tags": tags},
+            })
+
+        account_data = yield self.store.get_account_data_for_room(user_id, room_id)
+        for account_data_type, content in account_data.items():
+            account_data_events.append({
+                "type": account_data_type,
+                "content": content,
+            })
+
+        result["account_data"] = account_data_events
+
+        defer.returnValue(result)
+
+    @defer.inlineCallbacks
+    def _room_initial_sync_parted(self, user_id, room_id, pagin_config,
+                                  membership, member_event_id, is_peeking):
+        room_state = yield self.store.get_state_for_events(
+            [member_event_id], None
+        )
+
+        room_state = room_state[member_event_id]
+
+        limit = pagin_config.limit if pagin_config else None
+        if limit is None:
+            limit = 10
+
+        stream_token = yield self.store.get_stream_token_for_event(
+            member_event_id
+        )
+
+        messages, token = yield self.store.get_recent_events_for_room(
+            room_id,
+            limit=limit,
+            end_token=stream_token
+        )
+
+        messages = yield filter_events_for_client(
+            self.store, user_id, messages, is_peeking=is_peeking
+        )
+
+        start_token = StreamToken.START.copy_and_replace("room_key", token[0])
+        end_token = StreamToken.START.copy_and_replace("room_key", token[1])
+
+        time_now = self.clock.time_msec()
+
+        defer.returnValue({
+            "membership": membership,
+            "room_id": room_id,
+            "messages": {
+                "chunk": [serialize_event(m, time_now) for m in messages],
+                "start": start_token.to_string(),
+                "end": end_token.to_string(),
+            },
+            "state": [serialize_event(s, time_now) for s in room_state.values()],
+            "presence": [],
+            "receipts": [],
+        })
+
+    @defer.inlineCallbacks
+    def _room_initial_sync_joined(self, user_id, room_id, pagin_config,
+                                  membership, is_peeking):
+        current_state = yield self.state.get_current_state(
+            room_id=room_id,
+        )
+
+        # TODO: These concurrently
+        time_now = self.clock.time_msec()
+        state = [
+            serialize_event(x, time_now)
+            for x in current_state.values()
+        ]
+
+        now_token = yield self.hs.get_event_sources().get_current_token()
+
+        limit = pagin_config.limit if pagin_config else None
+        if limit is None:
+            limit = 10
+
+        room_members = [
+            m for m in current_state.values()
+            if m.type == EventTypes.Member
+            and m.content["membership"] == Membership.JOIN
+        ]
+
+        presence_handler = self.hs.get_presence_handler()
+
+        @defer.inlineCallbacks
+        def get_presence():
+            states = yield presence_handler.get_states(
+                [m.user_id for m in room_members],
+                as_event=True,
+            )
+
+            defer.returnValue(states)
+
+        @defer.inlineCallbacks
+        def get_receipts():
+            receipts = yield self.store.get_linearized_receipts_for_room(
+                room_id,
+                to_key=now_token.receipt_key,
+            )
+            if not receipts:
+                receipts = []
+            defer.returnValue(receipts)
+
+        presence, receipts, (messages, token) = yield defer.gatherResults(
+            [
+                preserve_fn(get_presence)(),
+                preserve_fn(get_receipts)(),
+                preserve_fn(self.store.get_recent_events_for_room)(
+                    room_id,
+                    limit=limit,
+                    end_token=now_token.room_key,
+                )
+            ],
+            consumeErrors=True,
+        ).addErrback(unwrapFirstError)
+
+        messages = yield filter_events_for_client(
+            self.store, user_id, messages, is_peeking=is_peeking,
+        )
+
+        start_token = now_token.copy_and_replace("room_key", token[0])
+        end_token = now_token.copy_and_replace("room_key", token[1])
+
+        time_now = self.clock.time_msec()
+
+        ret = {
+            "room_id": room_id,
+            "messages": {
+                "chunk": [serialize_event(m, time_now) for m in messages],
+                "start": start_token.to_string(),
+                "end": end_token.to_string(),
+            },
+            "state": state,
+            "presence": presence,
+            "receipts": receipts,
+        }
+        if not is_peeking:
+            ret["membership"] = membership
+
+        defer.returnValue(ret)
+
+    @defer.inlineCallbacks
+    def _check_in_room_or_world_readable(self, room_id, user_id):
+        try:
+            # check_user_was_in_room will return the most recent membership
+            # event for the user if:
+            #  * The user is a non-guest user, and was ever in the room
+            #  * The user is a guest user, and has joined the room
+            # else it will throw.
+            member_event = yield self.auth.check_user_was_in_room(room_id, user_id)
+            defer.returnValue((member_event.membership, member_event.event_id))
+            return
+        except AuthError:
+            visibility = yield self.state_handler.get_current_state(
+                room_id, EventTypes.RoomHistoryVisibility, ""
+            )
+            if (
+                visibility and
+                visibility.content["history_visibility"] == "world_readable"
+            ):
+                defer.returnValue((Membership.JOIN, None))
+                return
+            raise AuthError(
+                403, "Guest access not allowed", errcode=Codes.GUEST_ACCESS_FORBIDDEN
+            )

synapse/handlers/message.py

@@ -16,19 +16,17 @@
 from twisted.internet import defer
 
 from synapse.api.constants import EventTypes, Membership
-from synapse.api.errors import AuthError, Codes, SynapseError
+from synapse.api.errors import AuthError, Codes, SynapseError, LimitExceededError
 from synapse.crypto.event_signing import add_hashes_and_signatures
 from synapse.events.utils import serialize_event
 from synapse.events.validator import EventValidator
 from synapse.push.action_generator import ActionGenerator
-from synapse.streams.config import PaginationConfig
 from synapse.types import (
-    UserID, RoomAlias, RoomStreamToken, StreamToken, get_domain_from_id
+    UserID, RoomAlias, RoomStreamToken,
 )
-from synapse.util import unwrapFirstError
-from synapse.util.async import concurrently_execute, run_on_reactor, ReadWriteLock
-from synapse.util.caches.snapshot_cache import SnapshotCache
+from synapse.util.async import run_on_reactor, ReadWriteLock, Limiter
 from synapse.util.logcontext import preserve_fn
+from synapse.util.metrics import measure_func
 from synapse.visibility import filter_events_for_client
 
 from ._base import BaseHandler
@@ -36,6 +34,7 @@ from ._base import BaseHandler
 from canonicaljson import encode_canonical_json
 
 import logging
+import random
 
 logger = logging.getLogger(__name__)
 
@@ -48,10 +47,13 @@ class MessageHandler(BaseHandler):
         self.state = hs.get_state_handler()
         self.clock = hs.get_clock()
         self.validator = EventValidator()
-        self.snapshot_cache = SnapshotCache()
 
         self.pagination_lock = ReadWriteLock()
 
+        # We arbitrarily limit concurrent event creation for a room to 5.
+        # This is to stop us from diverging history *too* much.
+        self.limiter = Limiter(max_count=5)
+
     @defer.inlineCallbacks
     def purge_history(self, room_id, event_id):
         event = yield self.store.get_event(event_id)
@@ -66,7 +68,7 @@ class MessageHandler(BaseHandler):
 
     @defer.inlineCallbacks
     def get_messages(self, requester, room_id=None, pagin_config=None,
-                     as_client_event=True):
+                     as_client_event=True, event_filter=None):
         """Get messages in a room.
 
         Args:
@@ -75,18 +77,18 @@ class MessageHandler(BaseHandler):
             pagin_config (synapse.api.streams.PaginationConfig): The pagination
                 config rules to apply, if any.
             as_client_event (bool): True to get events in client-server format.
+            event_filter (Filter): Filter to apply to results or None
         Returns:
             dict: Pagination API results
         """
         user_id = requester.user.to_string()
-        data_source = self.hs.get_event_sources().sources["room"]
 
         if pagin_config.from_token:
             room_token = pagin_config.from_token.room_key
         else:
             pagin_config.from_token = (
-                yield self.hs.get_event_sources().get_current_token(
-                    direction='b'
+                yield self.hs.get_event_sources().get_current_token_for_room(
+                    room_id=room_id
                 )
             )
             room_token = pagin_config.from_token.room_key
@@ -129,8 +131,13 @@ class MessageHandler(BaseHandler):
                     room_id, max_topo
                 )
 
-            events, next_key = yield data_source.get_pagination_rows(
-                requester.user, source_config, room_id
+            events, next_key = yield self.store.paginate_room_events(
+                room_id=room_id,
+                from_key=source_config.from_key,
+                to_key=source_config.to_key,
+                direction=source_config.direction,
+                limit=source_config.limit,
+                event_filter=event_filter,
             )
 
             next_token = pagin_config.from_token.copy_and_replace(
@@ -144,6 +151,9 @@ class MessageHandler(BaseHandler):
                 "end": next_token.to_string(),
             })
 
+        if event_filter:
+            events = event_filter.filter(events)
+
         events = yield filter_events_for_client(
             self.store,
             user_id,
@@ -185,36 +195,40 @@ class MessageHandler(BaseHandler):
         """
         builder = self.event_builder_factory.new(event_dict)
 
-        self.validator.validate_new(builder)
+        with (yield self.limiter.queue(builder.room_id)):
+            self.validator.validate_new(builder)
 
-        if builder.type == EventTypes.Member:
-            membership = builder.content.get("membership", None)
-            target = UserID.from_string(builder.state_key)
+            if builder.type == EventTypes.Member:
+                membership = builder.content.get("membership", None)
+                target = UserID.from_string(builder.state_key)
 
-            if membership in {Membership.JOIN, Membership.INVITE}:
-                # If event doesn't include a display name, add one.
-                profile = self.hs.get_handlers().profile_handler
-                content = builder.content
+                if membership in {Membership.JOIN, Membership.INVITE}:
+                    # If event doesn't include a display name, add one.
+                    profile = self.hs.get_handlers().profile_handler
+                    content = builder.content
 
-                try:
-                    content["displayname"] = yield profile.get_displayname(target)
-                    content["avatar_url"] = yield profile.get_avatar_url(target)
-                except Exception as e:
-                    logger.info(
-                        "Failed to get profile information for %r: %s",
-                        target, e
-                    )
+                    try:
+                        if "displayname" not in content:
+                            content["displayname"] = yield profile.get_displayname(target)
+                        if "avatar_url" not in content:
+                            content["avatar_url"] = yield profile.get_avatar_url(target)
+                    except Exception as e:
+                        logger.info(
+                            "Failed to get profile information for %r: %s",
+                            target, e
+                        )
 
-        if token_id is not None:
-            builder.internal_metadata.token_id = token_id
+            if token_id is not None:
+                builder.internal_metadata.token_id = token_id
 
-        if txn_id is not None:
-            builder.internal_metadata.txn_id = txn_id
+            if txn_id is not None:
+                builder.internal_metadata.txn_id = txn_id
+
+            event, context = yield self._create_new_client_event(
+                builder=builder,
+                prev_event_ids=prev_event_ids,
+            )
 
-        event, context = yield self._create_new_client_event(
-            builder=builder,
-            prev_event_ids=prev_event_ids,
-        )
         defer.returnValue((event, context))
 
     @defer.inlineCallbacks
@@ -234,12 +248,27 @@ class MessageHandler(BaseHandler):
                 "Tried to send member event through non-member codepath"
             )
 
+        # We check here if we are currently being rate limited, so that we
+        # don't do unnecessary work. We check again just before we actually
+        # send the event.
+        time_now = self.clock.time()
+        allowed, time_allowed = self.ratelimiter.send_message(
+            event.sender, time_now,
+            msg_rate_hz=self.hs.config.rc_messages_per_second,
+            burst_count=self.hs.config.rc_message_burst_count,
+            update=False,
+        )
+        if not allowed:
+            raise LimitExceededError(
+                retry_after_ms=int(1000 * (time_allowed - time_now)),
+            )
+
         user = UserID.from_string(event.sender)
 
         assert self.hs.is_mine(user), "User must be our own: %s" % (user,)
 
         if event.is_state():
-            prev_state = self.deduplicate_state_event(event, context)
+            prev_state = yield self.deduplicate_state_event(event, context)
             if prev_state is not None:
                 defer.returnValue(prev_state)
 
@@ -252,8 +281,11 @@ class MessageHandler(BaseHandler):
 
         if event.type == EventTypes.Message:
             presence = self.hs.get_presence_handler()
-            yield presence.bump_presence_active_time(user)
+            # We don't want to block sending messages on any presence code. This
+            # matters as sometimes presence code can take a while.
+            preserve_fn(presence.bump_presence_active_time)(user)
 
+    @defer.inlineCallbacks
     def deduplicate_state_event(self, event, context):
         """
         Checks whether event is in the latest resolved state in context.
@@ -261,13 +293,17 @@ class MessageHandler(BaseHandler):
         If so, returns the version of the event in context.
         Otherwise, returns None.
         """
-        prev_event = context.current_state.get((event.type, event.state_key))
+        prev_event_id = context.prev_state_ids.get((event.type, event.state_key))
+        prev_event = yield self.store.get_event(prev_event_id, allow_none=True)
+        if not prev_event:
+            return
+
         if prev_event and event.user_id == prev_event.user_id:
             prev_content = encode_canonical_json(prev_event.content)
             next_content = encode_canonical_json(event.content)
             if prev_content == next_content:
-                return prev_event
-        return None
+                defer.returnValue(prev_event)
+        return
 
     @defer.inlineCallbacks
     def create_and_send_nonmember_event(
@@ -378,375 +414,7 @@ class MessageHandler(BaseHandler):
             [serialize_event(c, now) for c in room_state.values()]
         )
 
-    def snapshot_all_rooms(self, user_id=None, pagin_config=None,
-                           as_client_event=True, include_archived=False):
-        """Retrieve a snapshot of all rooms the user is invited or has joined.
-
-        This snapshot may include messages for all rooms where the user is
-        joined, depending on the pagination config.
-
-        Args:
-            user_id (str): The ID of the user making the request.
-            pagin_config (synapse.api.streams.PaginationConfig): The pagination
-            config used to determine how many messages *PER ROOM* to return.
-            as_client_event (bool): True to get events in client-server format.
-            include_archived (bool): True to get rooms that the user has left
-        Returns:
-            A list of dicts with "room_id" and "membership" keys for all rooms
-            the user is currently invited or joined in on. Rooms where the user
-            is joined on, may return a "messages" key with messages, depending
-            on the specified PaginationConfig.
-        """
-        key = (
-            user_id,
-            pagin_config.from_token,
-            pagin_config.to_token,
-            pagin_config.direction,
-            pagin_config.limit,
-            as_client_event,
-            include_archived,
-        )
-        now_ms = self.clock.time_msec()
-        result = self.snapshot_cache.get(now_ms, key)
-        if result is not None:
-            return result
-
-        return self.snapshot_cache.set(now_ms, key, self._snapshot_all_rooms(
-            user_id, pagin_config, as_client_event, include_archived
-        ))
-
-    @defer.inlineCallbacks
-    def _snapshot_all_rooms(self, user_id=None, pagin_config=None,
-                            as_client_event=True, include_archived=False):
-
-        memberships = [Membership.INVITE, Membership.JOIN]
-        if include_archived:
-            memberships.append(Membership.LEAVE)
-
-        room_list = yield self.store.get_rooms_for_user_where_membership_is(
-            user_id=user_id, membership_list=memberships
-        )
-
-        user = UserID.from_string(user_id)
-
-        rooms_ret = []
-
-        now_token = yield self.hs.get_event_sources().get_current_token()
-
-        presence_stream = self.hs.get_event_sources().sources["presence"]
-        pagination_config = PaginationConfig(from_token=now_token)
-        presence, _ = yield presence_stream.get_pagination_rows(
-            user, pagination_config.get_source_config("presence"), None
-        )
-
-        receipt_stream = self.hs.get_event_sources().sources["receipt"]
-        receipt, _ = yield receipt_stream.get_pagination_rows(
-            user, pagination_config.get_source_config("receipt"), None
-        )
-
-        tags_by_room = yield self.store.get_tags_for_user(user_id)
-
-        account_data, account_data_by_room = (
-            yield self.store.get_account_data_for_user(user_id)
-        )
-
-        public_room_ids = yield self.store.get_public_room_ids()
-
-        limit = pagin_config.limit
-        if limit is None:
-            limit = 10
-
-        @defer.inlineCallbacks
-        def handle_room(event):
-            d = {
-                "room_id": event.room_id,
-                "membership": event.membership,
-                "visibility": (
-                    "public" if event.room_id in public_room_ids
-                    else "private"
-                ),
-            }
-
-            if event.membership == Membership.INVITE:
-                time_now = self.clock.time_msec()
-                d["inviter"] = event.sender
-
-                invite_event = yield self.store.get_event(event.event_id)
-                d["invite"] = serialize_event(invite_event, time_now, as_client_event)
-
-            rooms_ret.append(d)
-
-            if event.membership not in (Membership.JOIN, Membership.LEAVE):
-                return
-
-            try:
-                if event.membership == Membership.JOIN:
-                    room_end_token = now_token.room_key
-                    deferred_room_state = self.state_handler.get_current_state(
-                        event.room_id
-                    )
-                elif event.membership == Membership.LEAVE:
-                    room_end_token = "s%d" % (event.stream_ordering,)
-                    deferred_room_state = self.store.get_state_for_events(
-                        [event.event_id], None
-                    )
-                    deferred_room_state.addCallback(
-                        lambda states: states[event.event_id]
-                    )
-
-                (messages, token), current_state = yield defer.gatherResults(
-                    [
-                        self.store.get_recent_events_for_room(
-                            event.room_id,
-                            limit=limit,
-                            end_token=room_end_token,
-                        ),
-                        deferred_room_state,
-                    ]
-                ).addErrback(unwrapFirstError)
-
-                messages = yield filter_events_for_client(
-                    self.store, user_id, messages
-                )
-
-                start_token = now_token.copy_and_replace("room_key", token[0])
-                end_token = now_token.copy_and_replace("room_key", token[1])
-                time_now = self.clock.time_msec()
-
-                d["messages"] = {
-                    "chunk": [
-                        serialize_event(m, time_now, as_client_event)
-                        for m in messages
-                    ],
-                    "start": start_token.to_string(),
-                    "end": end_token.to_string(),
-                }
-
-                d["state"] = [
-                    serialize_event(c, time_now, as_client_event)
-                    for c in current_state.values()
-                ]
-
-                account_data_events = []
-                tags = tags_by_room.get(event.room_id)
-                if tags:
-                    account_data_events.append({
-                        "type": "m.tag",
-                        "content": {"tags": tags},
-                    })
-
-                account_data = account_data_by_room.get(event.room_id, {})
-                for account_data_type, content in account_data.items():
-                    account_data_events.append({
-                        "type": account_data_type,
-                        "content": content,
-                    })
-
-                d["account_data"] = account_data_events
-            except:
-                logger.exception("Failed to get snapshot")
-
-        yield concurrently_execute(handle_room, room_list, 10)
-
-        account_data_events = []
-        for account_data_type, content in account_data.items():
-            account_data_events.append({
-                "type": account_data_type,
-                "content": content,
-            })
-
-        ret = {
-            "rooms": rooms_ret,
-            "presence": presence,
-            "account_data": account_data_events,
-            "receipts": receipt,
-            "end": now_token.to_string(),
-        }
-
-        defer.returnValue(ret)
-
-    @defer.inlineCallbacks
-    def room_initial_sync(self, requester, room_id, pagin_config=None):
-        """Capture the a snapshot of a room. If user is currently a member of
-        the room this will be what is currently in the room. If the user left
-        the room this will be what was in the room when they left.
-
-        Args:
-            requester(Requester): The user to get a snapshot for.
-            room_id(str): The room to get a snapshot of.
-            pagin_config(synapse.streams.config.PaginationConfig):
-                The pagination config used to determine how many messages to
-                return.
-        Raises:
-            AuthError if the user wasn't in the room.
-        Returns:
-            A JSON serialisable dict with the snapshot of the room.
-        """
-
-        user_id = requester.user.to_string()
-
-        membership, member_event_id = yield self._check_in_room_or_world_readable(
-            room_id, user_id,
-        )
-        is_peeking = member_event_id is None
-
-        if membership == Membership.JOIN:
-            result = yield self._room_initial_sync_joined(
-                user_id, room_id, pagin_config, membership, is_peeking
-            )
-        elif membership == Membership.LEAVE:
-            result = yield self._room_initial_sync_parted(
-                user_id, room_id, pagin_config, membership, member_event_id, is_peeking
-            )
-
-        account_data_events = []
-        tags = yield self.store.get_tags_for_room(user_id, room_id)
-        if tags:
-            account_data_events.append({
-                "type": "m.tag",
-                "content": {"tags": tags},
-            })
-
-        account_data = yield self.store.get_account_data_for_room(user_id, room_id)
-        for account_data_type, content in account_data.items():
-            account_data_events.append({
-                "type": account_data_type,
-                "content": content,
-            })
-
-        result["account_data"] = account_data_events
-
-        defer.returnValue(result)
-
-    @defer.inlineCallbacks
-    def _room_initial_sync_parted(self, user_id, room_id, pagin_config,
-                                  membership, member_event_id, is_peeking):
-        room_state = yield self.store.get_state_for_events(
-            [member_event_id], None
-        )
-
-        room_state = room_state[member_event_id]
-
-        limit = pagin_config.limit if pagin_config else None
-        if limit is None:
-            limit = 10
-
-        stream_token = yield self.store.get_stream_token_for_event(
-            member_event_id
-        )
-
-        messages, token = yield self.store.get_recent_events_for_room(
-            room_id,
-            limit=limit,
-            end_token=stream_token
-        )
-
-        messages = yield filter_events_for_client(
-            self.store, user_id, messages, is_peeking=is_peeking
-        )
-
-        start_token = StreamToken.START.copy_and_replace("room_key", token[0])
-        end_token = StreamToken.START.copy_and_replace("room_key", token[1])
-
-        time_now = self.clock.time_msec()
-
-        defer.returnValue({
-            "membership": membership,
-            "room_id": room_id,
-            "messages": {
-                "chunk": [serialize_event(m, time_now) for m in messages],
-                "start": start_token.to_string(),
-                "end": end_token.to_string(),
-            },
-            "state": [serialize_event(s, time_now) for s in room_state.values()],
-            "presence": [],
-            "receipts": [],
-        })
-
-    @defer.inlineCallbacks
-    def _room_initial_sync_joined(self, user_id, room_id, pagin_config,
-                                  membership, is_peeking):
-        current_state = yield self.state.get_current_state(
-            room_id=room_id,
-        )
-
-        # TODO: These concurrently
-        time_now = self.clock.time_msec()
-        state = [
-            serialize_event(x, time_now)
-            for x in current_state.values()
-        ]
-
-        now_token = yield self.hs.get_event_sources().get_current_token()
-
-        limit = pagin_config.limit if pagin_config else None
-        if limit is None:
-            limit = 10
-
-        room_members = [
-            m for m in current_state.values()
-            if m.type == EventTypes.Member
-            and m.content["membership"] == Membership.JOIN
-        ]
-
-        presence_handler = self.hs.get_presence_handler()
-
-        @defer.inlineCallbacks
-        def get_presence():
-            states = yield presence_handler.get_states(
-                [m.user_id for m in room_members],
-                as_event=True,
-            )
-
-            defer.returnValue(states)
-
-        @defer.inlineCallbacks
-        def get_receipts():
-            receipts_handler = self.hs.get_handlers().receipts_handler
-            receipts = yield receipts_handler.get_receipts_for_room(
-                room_id,
-                now_token.receipt_key
-            )
-            defer.returnValue(receipts)
-
-        presence, receipts, (messages, token) = yield defer.gatherResults(
-            [
-                get_presence(),
-                get_receipts(),
-                self.store.get_recent_events_for_room(
-                    room_id,
-                    limit=limit,
-                    end_token=now_token.room_key,
-                )
-            ],
-            consumeErrors=True,
-        ).addErrback(unwrapFirstError)
-
-        messages = yield filter_events_for_client(
-            self.store, user_id, messages, is_peeking=is_peeking,
-        )
-
-        start_token = now_token.copy_and_replace("room_key", token[0])
-        end_token = now_token.copy_and_replace("room_key", token[1])
-
-        time_now = self.clock.time_msec()
-
-        ret = {
-            "room_id": room_id,
-            "messages": {
-                "chunk": [serialize_event(m, time_now) for m in messages],
-                "start": start_token.to_string(),
-                "end": end_token.to_string(),
-            },
-            "state": state,
-            "presence": presence,
-            "receipts": receipts,
-        }
-        if not is_peeking:
-            ret["membership"] = membership
-
-        defer.returnValue(ret)
-
+    @measure_func("_create_new_client_event")
     @defer.inlineCallbacks
     def _create_new_client_event(self, builder, prev_event_ids=None):
         if prev_event_ids:
@@ -758,6 +426,20 @@ class MessageHandler(BaseHandler):
                 builder.room_id,
             )
 
+            # We want to limit the max number of prev events we point to in our
+            # new event
+            if len(latest_ret) > 10:
+                # Sort by reverse depth, so we point to the most recent.
+                latest_ret.sort(key=lambda a: -a[2])
+                new_latest_ret = latest_ret[:5]
+
+                # We also randomly point to some of the older events, to make
+                # sure that we don't completely ignore the older events.
+                if latest_ret[5:]:
+                    sample_size = min(5, len(latest_ret[5:]))
+                    new_latest_ret.extend(random.sample(latest_ret[5:], sample_size))
+                latest_ret = new_latest_ret
+
             if latest_ret:
                 depth = max([d for _, _, d in latest_ret]) + 1
             else:
@@ -790,14 +472,15 @@ class MessageHandler(BaseHandler):
         event = builder.build()
 
         logger.debug(
-            "Created event %s with current state: %s",
-            event.event_id, context.current_state,
+            "Created event %s with state: %s",
+            event.event_id, context.prev_state_ids,
         )
 
         defer.returnValue(
             (event, context,)
         )
 
+    @measure_func("handle_new_client_event")
     @defer.inlineCallbacks
     def handle_new_client_event(
         self,
@@ -813,12 +496,12 @@ class MessageHandler(BaseHandler):
             self.ratelimit(requester)
 
         try:
-            self.auth.check(event, auth_events=context.current_state)
+            yield self.auth.check_from_context(event, context)
         except AuthError as err:
             logger.warn("Denying new event %r because %s", event, err)
             raise err
 
-        yield self.maybe_kick_guest_users(event, context.current_state.values())
+        yield self.maybe_kick_guest_users(event, context)
 
         if event.type == EventTypes.CanonicalAlias:
             # Check the alias is acually valid (at this time at least)
@@ -846,6 +529,15 @@ class MessageHandler(BaseHandler):
                         e.sender == event.sender
                     )
 
+                state_to_include_ids = [
+                    e_id
+                    for k, e_id in context.current_state_ids.items()
+                    if k[0] in self.hs.config.room_invite_state_types
+                    or k[0] == EventTypes.Member and k[1] == event.sender
+                ]
+
+                state_to_include = yield self.store.get_events(state_to_include_ids)
+
                 event.unsigned["invite_room_state"] = [
                     {
                         "type": e.type,
@@ -853,9 +545,7 @@ class MessageHandler(BaseHandler):
                         "content": e.content,
                         "sender": e.sender,
                     }
-                    for k, e in context.current_state.items()
-                    if e.type in self.hs.config.room_invite_state_types
-                    or is_inviter_member_event(e)
+                    for e in state_to_include.values()
                 ]
 
                 invitee = UserID.from_string(event.state_key)
@@ -877,7 +567,14 @@ class MessageHandler(BaseHandler):
                     )
 
         if event.type == EventTypes.Redaction:
-            if self.auth.check_redaction(event, auth_events=context.current_state):
+            auth_events_ids = yield self.auth.compute_auth_events(
+                event, context.prev_state_ids, for_verification=True,
+            )
+            auth_events = yield self.store.get_events(auth_events_ids)
+            auth_events = {
+                (e.type, e.state_key): e for e in auth_events.values()
+            }
+            if self.auth.check_redaction(event, auth_events=auth_events):
                 original_event = yield self.store.get_event(
                     event.redacts,
                     check_redacted=False,
@@ -891,7 +588,7 @@ class MessageHandler(BaseHandler):
                         "You don't have permission to redact events"
                     )
 
-        if event.type == EventTypes.Create and context.current_state:
+        if event.type == EventTypes.Create and context.prev_state_ids:
             raise AuthError(
                 403,
                 "Changing the room create event is forbidden",
@@ -912,21 +609,10 @@ class MessageHandler(BaseHandler):
             event_stream_id, max_stream_id
         )
 
-        destinations = set()
-        for k, s in context.current_state.items():
-            try:
-                if k[0] == EventTypes.Member:
-                    if s.content["membership"] == Membership.JOIN:
-                        destinations.add(get_domain_from_id(s.state_key))
-            except SynapseError:
-                logger.warn(
-                    "Failed to get destination from event %s", s.event_id
-                )
-
         @defer.inlineCallbacks
         def _notify():
             yield run_on_reactor()
-            self.notifier.on_new_room_event(
+            yield self.notifier.on_new_room_event(
                 event, event_stream_id, max_stream_id,
                 extra_users=extra_users
             )
@@ -935,7 +621,3 @@ class MessageHandler(BaseHandler):
 
         # If invite, remove room_state from unsigned before sending.
         event.unsigned.pop("invite_room_state", None)
-
-        federation_handler.handle_new_event(
-            event, destinations=destinations,
-        )

synapse/handlers/presence.py

@@ -52,6 +52,11 @@ bump_active_time_counter = metrics.register_counter("bump_active_time")
 
 get_updates_counter = metrics.register_counter("get_updates", labels=["type"])
 
+notify_reason_counter = metrics.register_counter("notify_reason", labels=["reason"])
+state_transition_counter = metrics.register_counter(
+    "state_transition", labels=["from", "to"]
+)
+
 
 # If a user was last active in the last LAST_ACTIVE_GRANULARITY, consider them
 # "currently_active"
@@ -86,26 +91,29 @@ class PresenceHandler(object):
         self.store = hs.get_datastore()
         self.wheel_timer = WheelTimer()
         self.notifier = hs.get_notifier()
-        self.federation = hs.get_replication_layer()
+        self.replication = hs.get_replication_layer()
+        self.federation = hs.get_federation_sender()
 
-        self.federation.register_edu_handler(
+        self.state = hs.get_state_handler()
+
+        self.replication.register_edu_handler(
             "m.presence", self.incoming_presence
         )
-        self.federation.register_edu_handler(
+        self.replication.register_edu_handler(
             "m.presence_invite",
             lambda origin, content: self.invite_presence(
                 observed_user=UserID.from_string(content["observed_user"]),
                 observer_user=UserID.from_string(content["observer_user"]),
             )
         )
-        self.federation.register_edu_handler(
+        self.replication.register_edu_handler(
             "m.presence_accept",
             lambda origin, content: self.accept_presence(
                 observed_user=UserID.from_string(content["observed_user"]),
                 observer_user=UserID.from_string(content["observer_user"]),
             )
         )
-        self.federation.register_edu_handler(
+        self.replication.register_edu_handler(
             "m.presence_deny",
             lambda origin, content: self.deny_presence(
                 observed_user=UserID.from_string(content["observed_user"]),
@@ -189,6 +197,13 @@ class PresenceHandler(object):
             5000,
         )
 
+        self.clock.call_later(
+            60,
+            self.clock.looping_call,
+            self._persist_unpersisted_changes,
+            60 * 1000,
+        )
+
         metrics.register_callback("wheel_timer_size", lambda: len(self.wheel_timer))
 
     @defer.inlineCallbacks
@@ -203,7 +218,7 @@ class PresenceHandler(object):
         is some spurious presence changes that will self-correct.
         """
         logger.info(
-            "Performing _on_shutdown. Persiting %d unpersisted changes",
+            "Performing _on_shutdown. Persisting %d unpersisted changes",
             len(self.user_to_current_state)
         )
 
@@ -214,6 +229,27 @@ class PresenceHandler(object):
             ])
         logger.info("Finished _on_shutdown")
 
+    @defer.inlineCallbacks
+    def _persist_unpersisted_changes(self):
+        """We periodically persist the unpersisted changes, as otherwise they
+        may stack up and slow down shutdown times.
+        """
+        logger.info(
+            "Performing _persist_unpersisted_changes. Persisting %d unpersisted changes",
+            len(self.unpersisted_users_changes)
+        )
+
+        unpersisted = self.unpersisted_users_changes
+        self.unpersisted_users_changes = set()
+
+        if unpersisted:
+            yield self.store.update_presence([
+                self.user_to_current_state[user_id]
+                for user_id in unpersisted
+            ])
+
+        logger.info("Finished _persist_unpersisted_changes")
+
     @defer.inlineCallbacks
     def _update_states(self, new_states):
         """Updates presence of users. Sets the appropriate timeouts. Pokes
@@ -230,6 +266,12 @@ class PresenceHandler(object):
             to_notify = {}  # Changes we want to notify everyone about
             to_federation_ping = {}  # These need sending keep-alives
 
+            # Only bother handling the last presence change for each user
+            new_states_dict = {}
+            for new_state in new_states:
+                new_states_dict[new_state.user_id] = new_state
+            new_state = new_states_dict.values()
+
             for new_state in new_states:
                 user_id = new_state.user_id
 
@@ -503,7 +545,7 @@ class PresenceHandler(object):
         defer.returnValue(states)
 
     @defer.inlineCallbacks
-    def _get_interested_parties(self, states):
+    def _get_interested_parties(self, states, calculate_remote_hosts=True):
         """Given a list of states return which entities (rooms, users, servers)
         are interested in the given states.
 
@@ -526,14 +568,17 @@ class PresenceHandler(object):
             users_to_states.setdefault(state.user_id, []).append(state)
 
         hosts_to_states = {}
-        for room_id, states in room_ids_to_states.items():
-            local_states = filter(lambda s: self.is_mine_id(s.user_id), states)
-            if not local_states:
-                continue
+        if calculate_remote_hosts:
+            for room_id, states in room_ids_to_states.items():
+                local_states = filter(lambda s: self.is_mine_id(s.user_id), states)
+                if not local_states:
+                    continue
 
-            hosts = yield self.store.get_joined_hosts_for_room(room_id)
-            for host in hosts:
-                hosts_to_states.setdefault(host, []).extend(local_states)
+                users = yield self.store.get_users_in_room(room_id)
+                hosts = set(get_domain_from_id(u) for u in users)
+
+                for host in hosts:
+                    hosts_to_states.setdefault(host, []).extend(local_states)
 
         for user_id, states in users_to_states.items():
             local_states = filter(lambda s: self.is_mine_id(s.user_id), states)
@@ -565,24 +610,24 @@ class PresenceHandler(object):
 
         self._push_to_remotes(hosts_to_states)
 
+    @defer.inlineCallbacks
+    def notify_for_states(self, state, stream_id):
+        parties = yield self._get_interested_parties([state])
+        room_ids_to_states, users_to_states, hosts_to_states = parties
+
+        self.notifier.on_new_event(
+            "presence_key", stream_id, rooms=room_ids_to_states.keys(),
+            users=[UserID.from_string(u) for u in users_to_states.keys()]
+        )
+
     def _push_to_remotes(self, hosts_to_states):
         """Sends state updates to remote servers.
 
         Args:
             hosts_to_states (dict): Mapping `server_name` -> `[UserPresenceState]`
         """
-        now = self.clock.time_msec()
         for host, states in hosts_to_states.items():
-            self.federation.send_edu(
-                destination=host,
-                edu_type="m.presence",
-                content={
-                    "push": [
-                        _format_user_presence_state(state, now)
-                        for state in states
-                    ]
-                }
-            )
+            self.federation.send_presence(host, states)
 
     @defer.inlineCallbacks
     def incoming_presence(self, origin, content):
@@ -603,6 +648,13 @@ class PresenceHandler(object):
                 )
                 continue
 
+            if get_domain_from_id(user_id) != origin:
+                logger.info(
+                    "Got presence update from %r with bad 'user_id': %r",
+                    origin, user_id,
+                )
+                continue
+
             presence_state = push.get("presence", None)
             if not presence_state:
                 logger.info(
@@ -662,17 +714,17 @@ class PresenceHandler(object):
             defer.returnValue([
                 {
                     "type": "m.presence",
-                    "content": _format_user_presence_state(state, now),
+                    "content": format_user_presence_state(state, now),
                 }
                 for state in updates
             ])
         else:
             defer.returnValue([
-                _format_user_presence_state(state, now) for state in updates
+                format_user_presence_state(state, now) for state in updates
             ])
 
     @defer.inlineCallbacks
-    def set_state(self, target_user, state):
+    def set_state(self, target_user, state, ignore_status_msg=False):
         """Set the presence state of the user.
         """
         status_msg = state.get("status_msg", None)
@@ -689,10 +741,13 @@ class PresenceHandler(object):
         prev_state = yield self.current_state_for_user(user_id)
 
         new_fields = {
-            "state": presence,
-            "status_msg": status_msg if presence != PresenceState.OFFLINE else None
+            "state": presence
         }
 
+        if not ignore_status_msg:
+            msg = status_msg if presence != PresenceState.OFFLINE else None
+            new_fields["status_msg"] = msg
+
         if presence == PresenceState.ONLINE:
             new_fields["last_active_ts"] = self.clock.time_msec()
 
@@ -711,13 +766,13 @@ class PresenceHandler(object):
         # don't need to send to local clients here, as that is done as part
         # of the event stream/sync.
         # TODO: Only send to servers not already in the room.
+        user_ids = yield self.store.get_users_in_room(room_id)
         if self.is_mine(user):
             state = yield self.current_state_for_user(user.to_string())
 
-            hosts = yield self.store.get_joined_hosts_for_room(room_id)
+            hosts = set(get_domain_from_id(u) for u in user_ids)
             self._push_to_remotes({host: (state,) for host in hosts})
         else:
-            user_ids = yield self.store.get_users_in_room(room_id)
             user_ids = filter(self.is_mine_id, user_ids)
 
             states = yield self.current_state_for_users(user_ids)
@@ -893,28 +948,38 @@ class PresenceHandler(object):
 def should_notify(old_state, new_state):
     """Decides if a presence state change should be sent to interested parties.
     """
+    if old_state == new_state:
+        return False
+
     if old_state.status_msg != new_state.status_msg:
-        return True
-
-    if old_state.state == PresenceState.ONLINE:
-        if new_state.state != PresenceState.ONLINE:
-            # Always notify for online -> anything
-            return True
-
-        if new_state.currently_active != old_state.currently_active:
-            return True
-
-    if new_state.last_active_ts - old_state.last_active_ts > LAST_ACTIVE_GRANULARITY:
-        # Always notify for a transition where last active gets bumped.
+        notify_reason_counter.inc("status_msg_change")
         return True
 
     if old_state.state != new_state.state:
+        notify_reason_counter.inc("state_change")
+        state_transition_counter.inc(old_state.state, new_state.state)
+        return True
+
+    if old_state.state == PresenceState.ONLINE:
+        if new_state.currently_active != old_state.currently_active:
+            notify_reason_counter.inc("current_active_change")
+            return True
+
+        if new_state.last_active_ts - old_state.last_active_ts > LAST_ACTIVE_GRANULARITY:
+            # Only notify about last active bumps if we're not currently acive
+            if not new_state.currently_active:
+                notify_reason_counter.inc("last_active_change_online")
+                return True
+
+    elif new_state.last_active_ts - old_state.last_active_ts > LAST_ACTIVE_GRANULARITY:
+        # Always notify for a transition where last active gets bumped.
+        notify_reason_counter.inc("last_active_change_not_online")
         return True
 
     return False
 
 
-def _format_user_presence_state(state, now):
+def format_user_presence_state(state, now):
     """Convert UserPresenceState to a format that can be sent down to clients
     and to other servers.
     """
@@ -941,11 +1006,12 @@ class PresenceEventSource(object):
         self.get_presence_handler = hs.get_presence_handler
         self.clock = hs.get_clock()
         self.store = hs.get_datastore()
+        self.state = hs.get_state_handler()
 
     @defer.inlineCallbacks
     @log_function
     def get_new_events(self, user, from_key, room_ids=None, include_offline=True,
-                       **kwargs):
+                       explicit_room_id=None, **kwargs):
         # The process for getting presence events are:
         #  1. Get the rooms the user is in.
         #  2. Get the list of user in the rooms.
@@ -962,22 +1028,24 @@ class PresenceEventSource(object):
             user_id = user.to_string()
             if from_key is not None:
                 from_key = int(from_key)
-            room_ids = room_ids or []
 
             presence = self.get_presence_handler()
             stream_change_cache = self.store.presence_stream_cache
 
-            if not room_ids:
-                rooms = yield self.store.get_rooms_for_user(user_id)
-                room_ids = set(e.room_id for e in rooms)
-            else:
-                room_ids = set(room_ids)
-
             max_token = self.store.get_current_presence_token()
 
             plist = yield self.store.get_presence_list_accepted(user.localpart)
-            friends = set(row["observed_user_id"] for row in plist)
-            friends.add(user_id)  # So that we receive our own presence
+            users_interested_in = set(row["observed_user_id"] for row in plist)
+            users_interested_in.add(user_id)  # So that we receive our own presence
+
+            users_who_share_room = yield self.store.get_users_who_share_room_with_user(
+                user_id
+            )
+            users_interested_in.update(users_who_share_room)
+
+            if explicit_room_id:
+                user_ids = yield self.store.get_users_in_room(explicit_room_id)
+                users_interested_in.update(user_ids)
 
             user_ids_changed = set()
             changed = None
@@ -989,35 +1057,19 @@ class PresenceEventSource(object):
                 # work out if we share a room or they're in our presence list
                 get_updates_counter.inc("stream")
                 for other_user_id in changed:
-                    if other_user_id in friends:
+                    if other_user_id in users_interested_in:
                         user_ids_changed.add(other_user_id)
-                        continue
-                    other_rooms = yield self.store.get_rooms_for_user(other_user_id)
-                    if room_ids.intersection(e.room_id for e in other_rooms):
-                        user_ids_changed.add(other_user_id)
-                        continue
             else:
                 # Too many possible updates. Find all users we can see and check
                 # if any of them have changed.
                 get_updates_counter.inc("full")
 
-                user_ids_to_check = set()
-                for room_id in room_ids:
-                    users = yield self.store.get_users_in_room(room_id)
-                    user_ids_to_check.update(users)
-
-                user_ids_to_check.update(friends)
-
-                # Always include yourself. Only really matters for when the user is
-                # not in any rooms, but still.
-                user_ids_to_check.add(user_id)
-
                 if from_key:
                     user_ids_changed = stream_change_cache.get_entities_changed(
-                        user_ids_to_check, from_key,
+                        users_interested_in, from_key,
                     )
                 else:
-                    user_ids_changed = user_ids_to_check
+                    user_ids_changed = users_interested_in
 
             updates = yield presence.current_state_for_users(user_ids_changed)
 
@@ -1026,7 +1078,7 @@ class PresenceEventSource(object):
         defer.returnValue(([
             {
                 "type": "m.presence",
-                "content": _format_user_presence_state(s, now),
+                "content": format_user_presence_state(s, now),
             }
             for s in updates.values()
             if include_offline or s.state != PresenceState.OFFLINE

synapse/handlers/profile.py

@@ -13,15 +13,15 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import logging
+
 from twisted.internet import defer
 
+import synapse.types
 from synapse.api.errors import SynapseError, AuthError, CodeMessageException
-from synapse.types import UserID, Requester
-
+from synapse.types import UserID
 from ._base import BaseHandler
 
-import logging
-
 
 logger = logging.getLogger(__name__)
 
@@ -65,13 +65,13 @@ class ProfileHandler(BaseHandler):
                 defer.returnValue(result["displayname"])
 
     @defer.inlineCallbacks
-    def set_displayname(self, target_user, requester, new_displayname):
+    def set_displayname(self, target_user, requester, new_displayname, by_admin=False):
         """target_user is the user whose displayname is to be changed;
         auth_user is the user attempting to make this change."""
         if not self.hs.is_mine(target_user):
             raise SynapseError(400, "User is not hosted on this Home Server")
 
-        if target_user != requester.user:
+        if not by_admin and target_user != requester.user:
             raise AuthError(400, "Cannot set another user's displayname")
 
         if new_displayname == '':
@@ -111,13 +111,13 @@ class ProfileHandler(BaseHandler):
             defer.returnValue(result["avatar_url"])
 
     @defer.inlineCallbacks
-    def set_avatar_url(self, target_user, requester, new_avatar_url):
+    def set_avatar_url(self, target_user, requester, new_avatar_url, by_admin=False):
         """target_user is the user whose avatar_url is to be changed;
         auth_user is the user attempting to make this change."""
         if not self.hs.is_mine(target_user):
             raise SynapseError(400, "User is not hosted on this Home Server")
 
-        if target_user != requester.user:
+        if not by_admin and target_user != requester.user:
             raise AuthError(400, "Cannot set another user's avatar_url")
 
         yield self.store.set_profile_avatar_url(
@@ -165,7 +165,9 @@ class ProfileHandler(BaseHandler):
             try:
                 # Assume the user isn't a guest because we don't let guests set
                 # profile or avatar data.
-                requester = Requester(user, "", False)
+                # XXX why are we recreating `requester` here for each room?
+                # what was wrong with the `requester` we were passed?
+                requester = synapse.types.create_requester(user)
                 yield handler.update_membership(
                     requester,
                     user,

synapse/handlers/receipts.py

@@ -18,6 +18,7 @@ from ._base import BaseHandler
 from twisted.internet import defer
 
 from synapse.util.logcontext import PreserveLoggingContext
+from synapse.types import get_domain_from_id
 
 import logging
 
@@ -32,11 +33,12 @@ class ReceiptsHandler(BaseHandler):
         self.server_name = hs.config.server_name
         self.store = hs.get_datastore()
         self.hs = hs
-        self.federation = hs.get_replication_layer()
-        self.federation.register_edu_handler(
+        self.federation = hs.get_federation_sender()
+        hs.get_replication_layer().register_edu_handler(
             "m.receipt", self._received_remote_receipt
         )
         self.clock = self.hs.get_clock()
+        self.state = hs.get_state_handler()
 
     @defer.inlineCallbacks
     def received_client_receipt(self, room_id, receipt_type, user_id,
@@ -98,7 +100,7 @@ class ReceiptsHandler(BaseHandler):
 
             if not res:
                 # res will be None if this read receipt is 'old'
-                defer.returnValue(False)
+                continue
 
             stream_id, max_persisted_id = res
 
@@ -107,6 +109,10 @@ class ReceiptsHandler(BaseHandler):
             if max_batch_id is None or max_persisted_id > max_batch_id:
                 max_batch_id = max_persisted_id
 
+        if min_batch_id is None:
+            # no new receipts
+            defer.returnValue(False)
+
         affected_room_ids = list(set([r["room_id"] for r in receipts]))
 
         with PreserveLoggingContext():
@@ -133,7 +139,8 @@ class ReceiptsHandler(BaseHandler):
             event_ids = receipt["event_ids"]
             data = receipt["data"]
 
-            remotedomains = yield self.store.get_joined_hosts_for_room(room_id)
+            users = yield self.state.get_current_user_in_room(room_id)
+            remotedomains = set(get_domain_from_id(u) for u in users)
             remotedomains = remotedomains.copy()
             remotedomains.discard(self.server_name)
 
@@ -153,6 +160,7 @@ class ReceiptsHandler(BaseHandler):
                             }
                         },
                     },
+                    key=(room_id, receipt_type, user_id),
                 )
 
     @defer.inlineCallbacks

synapse/handlers/register.py

@@ -14,18 +14,18 @@
 # limitations under the License.
 
 """Contains functions for registering clients."""
+import logging
+import urllib
+
 from twisted.internet import defer
 
-from synapse.types import UserID, Requester
 from synapse.api.errors import (
     AuthError, Codes, SynapseError, RegistrationError, InvalidCaptchaError
 )
-from ._base import BaseHandler
-from synapse.util.async import run_on_reactor
 from synapse.http.client import CaptchaServerHttpClient
-
-import logging
-import urllib
+from synapse.types import UserID
+from synapse.util.async import run_on_reactor
+from ._base import BaseHandler
 
 logger = logging.getLogger(__name__)
 
@@ -40,6 +40,8 @@ class RegistrationHandler(BaseHandler):
 
         self._next_generated_user_id = None
 
+        self.macaroon_gen = hs.get_macaroon_generator()
+
     @defer.inlineCallbacks
     def check_username(self, localpart, guest_access_token=None,
                        assigned_user_id=None):
@@ -52,6 +54,13 @@ class RegistrationHandler(BaseHandler):
                 Codes.INVALID_USERNAME
             )
 
+        if localpart[0] == '_':
+            raise SynapseError(
+                400,
+                "User ID may not begin with _",
+                Codes.INVALID_USERNAME
+            )
+
         user = UserID(localpart, self.hs.hostname)
         user_id = user.to_string()
 
@@ -74,7 +83,7 @@ class RegistrationHandler(BaseHandler):
                     "User ID already taken.",
                     errcode=Codes.USER_IN_USE,
                 )
-            user_data = yield self.auth.get_user_from_macaroon(guest_access_token)
+            user_data = yield self.auth.get_user_by_access_token(guest_access_token)
             if not user_data["is_guest"] or user_data["user"].localpart != localpart:
                 raise AuthError(
                     403,
@@ -99,8 +108,13 @@ class RegistrationHandler(BaseHandler):
             localpart : The local part of the user ID to register. If None,
               one will be generated.
             password (str) : The password to assign to this user so they can
-            login again. This can be None which means they cannot login again
-            via a password (e.g. the user is an application service user).
+              login again. This can be None which means they cannot login again
+              via a password (e.g. the user is an application service user).
+            generate_token (bool): Whether a new access token should be
+              generated. Having this be True should be considered deprecated,
+              since it offers no means of associating a device_id with the
+              access_token. Instead you should call auth_handler.issue_access_token
+              after registration.
         Returns:
             A tuple of (user_id, access_token).
         Raises:
@@ -131,7 +145,7 @@ class RegistrationHandler(BaseHandler):
 
             token = None
             if generate_token:
-                token = self.auth_handler().generate_access_token(user_id)
+                token = self.macaroon_gen.generate_access_token(user_id)
             yield self.store.register(
                 user_id=user_id,
                 token=token,
@@ -155,7 +169,7 @@ class RegistrationHandler(BaseHandler):
                 user_id = user.to_string()
                 yield self.check_user_id_not_appservice_exclusive(user_id)
                 if generate_token:
-                    token = self.auth_handler().generate_access_token(user_id)
+                    token = self.macaroon_gen.generate_access_token(user_id)
                 try:
                     yield self.store.register(
                         user_id=user_id,
@@ -181,7 +195,7 @@ class RegistrationHandler(BaseHandler):
     def appservice_register(self, user_localpart, as_token):
         user = UserID(user_localpart, self.hs.hostname)
         user_id = user.to_string()
-        service = yield self.store.get_app_service_by_token(as_token)
+        service = self.store.get_app_service_by_token(as_token)
         if not service:
             raise AuthError(403, "Invalid application service token.")
         if not service.is_interested_in_user(user_id):
@@ -196,15 +210,13 @@ class RegistrationHandler(BaseHandler):
             user_id, allowed_appservice=service
         )
 
-        token = self.auth_handler().generate_access_token(user_id)
         yield self.store.register(
             user_id=user_id,
-            token=token,
             password_hash="",
             appservice_id=service_id,
             create_profile_with_localpart=user.localpart,
         )
-        defer.returnValue((user_id, token))
+        defer.returnValue(user_id)
 
     @defer.inlineCallbacks
     def check_recaptcha(self, ip, private_key, challenge, response):
@@ -244,7 +256,7 @@ class RegistrationHandler(BaseHandler):
         user_id = user.to_string()
 
         yield self.check_user_id_not_appservice_exclusive(user_id)
-        token = self.auth_handler().generate_access_token(user_id)
+        token = self.macaroon_gen.generate_access_token(user_id)
         try:
             yield self.store.register(
                 user_id=user_id,
@@ -294,11 +306,10 @@ class RegistrationHandler(BaseHandler):
             # XXX: This should be a deferred list, shouldn't it?
             yield identity_handler.bind_threepid(c, user_id)
 
-    @defer.inlineCallbacks
     def check_user_id_not_appservice_exclusive(self, user_id, allowed_appservice=None):
         # valid user IDs must not clash with any user ID namespaces claimed by
         # application services.
-        services = yield self.store.get_app_services()
+        services = self.store.get_app_services()
         interested_services = [
             s for s in services
             if s.is_interested_in_user(user_id)
@@ -360,7 +371,7 @@ class RegistrationHandler(BaseHandler):
         defer.returnValue(data)
 
     @defer.inlineCallbacks
-    def get_or_create_user(self, localpart, displayname, duration_seconds,
+    def get_or_create_user(self, requester, localpart, displayname,
                            password_hash=None):
         """Creates a new user if the user does not exist,
         else revokes all previous access tokens and generates a new one.
@@ -390,8 +401,7 @@ class RegistrationHandler(BaseHandler):
 
         user = UserID(localpart, self.hs.hostname)
         user_id = user.to_string()
-        token = self.auth_handler().generate_short_term_login_token(
-            user_id, duration_seconds)
+        token = self.macaroon_gen.generate_access_token(user_id)
 
         if need_register:
             yield self.store.register(
@@ -408,7 +418,7 @@ class RegistrationHandler(BaseHandler):
             logger.info("setting user display name: %s -> %s", user_id, displayname)
             profile_handler = self.hs.get_handlers().profile_handler
             yield profile_handler.set_displayname(
-                user, Requester(user, token, False), displayname
+                user, requester, displayname, by_admin=True,
             )
 
         defer.returnValue((user_id, token))

synapse/handlers/room.py

@@ -20,12 +20,10 @@ from ._base import BaseHandler
 
 from synapse.types import UserID, RoomAlias, RoomID, RoomStreamToken
 from synapse.api.constants import (
-    EventTypes, JoinRules, RoomCreationPreset, Membership,
+    EventTypes, JoinRules, RoomCreationPreset
 )
 from synapse.api.errors import AuthError, StoreError, SynapseError
 from synapse.util import stringutils
-from synapse.util.async import concurrently_execute
-from synapse.util.caches.response_cache import ResponseCache
 from synapse.visibility import filter_events_for_client
 
 from collections import OrderedDict
@@ -36,8 +34,6 @@ import string
 
 logger = logging.getLogger(__name__)
 
-REMOTE_ROOM_LIST_POLL_INTERVAL = 60 * 1000
-
 id_server_scheme = "https://"
 
 
@@ -48,16 +44,19 @@ class RoomCreationHandler(BaseHandler):
             "join_rules": JoinRules.INVITE,
             "history_visibility": "shared",
             "original_invitees_have_ops": False,
+            "guest_can_join": True,
         },
         RoomCreationPreset.TRUSTED_PRIVATE_CHAT: {
             "join_rules": JoinRules.INVITE,
             "history_visibility": "shared",
             "original_invitees_have_ops": True,
+            "guest_can_join": True,
         },
         RoomCreationPreset.PUBLIC_CHAT: {
             "join_rules": JoinRules.PUBLIC,
             "history_visibility": "shared",
             "original_invitees_have_ops": False,
+            "guest_can_join": False,
         },
     }
 
@@ -196,6 +195,11 @@ class RoomCreationHandler(BaseHandler):
                 },
                 ratelimit=False)
 
+        content = {}
+        is_direct = config.get("is_direct", None)
+        if is_direct:
+            content["is_direct"] = is_direct
+
         for invitee in invite_list:
             yield room_member_handler.update_membership(
                 requester,
@@ -203,6 +207,7 @@ class RoomCreationHandler(BaseHandler):
                 room_id,
                 "invite",
                 ratelimit=False,
+                content=content,
             )
 
         for invite_3pid in invite_3pid_list:
@@ -334,6 +339,13 @@ class RoomCreationHandler(BaseHandler):
                 content={"history_visibility": config["history_visibility"]}
             )
 
+        if config["guest_can_join"]:
+            if (EventTypes.GuestAccess, '') not in initial_state:
+                yield send(
+                    etype=EventTypes.GuestAccess,
+                    content={"guest_access": "can_join"}
+                )
+
         for (etype, state_key), content in initial_state.items():
             yield send(
                 etype=etype,
@@ -342,149 +354,6 @@ class RoomCreationHandler(BaseHandler):
             )
 
 
-class RoomListHandler(BaseHandler):
-    def __init__(self, hs):
-        super(RoomListHandler, self).__init__(hs)
-        self.response_cache = ResponseCache()
-        self.remote_list_request_cache = ResponseCache()
-        self.remote_list_cache = {}
-        self.fetch_looping_call = hs.get_clock().looping_call(
-            self.fetch_all_remote_lists, REMOTE_ROOM_LIST_POLL_INTERVAL
-        )
-        self.fetch_all_remote_lists()
-
-    def get_local_public_room_list(self):
-        result = self.response_cache.get(())
-        if not result:
-            result = self.response_cache.set((), self._get_public_room_list())
-        return result
-
-    @defer.inlineCallbacks
-    def _get_public_room_list(self):
-        room_ids = yield self.store.get_public_room_ids()
-
-        results = []
-
-        @defer.inlineCallbacks
-        def handle_room(room_id):
-            current_state = yield self.state_handler.get_current_state(room_id)
-
-            # Double check that this is actually a public room.
-            join_rules_event = current_state.get((EventTypes.JoinRules, ""))
-            if join_rules_event:
-                join_rule = join_rules_event.content.get("join_rule", None)
-                if join_rule and join_rule != JoinRules.PUBLIC:
-                    defer.returnValue(None)
-
-            result = {"room_id": room_id}
-
-            num_joined_users = len([
-                1 for _, event in current_state.items()
-                if event.type == EventTypes.Member
-                and event.membership == Membership.JOIN
-            ])
-            if num_joined_users == 0:
-                return
-
-            result["num_joined_members"] = num_joined_users
-
-            aliases = yield self.store.get_aliases_for_room(room_id)
-            if aliases:
-                result["aliases"] = aliases
-
-            name_event = yield current_state.get((EventTypes.Name, ""))
-            if name_event:
-                name = name_event.content.get("name", None)
-                if name:
-                    result["name"] = name
-
-            topic_event = current_state.get((EventTypes.Topic, ""))
-            if topic_event:
-                topic = topic_event.content.get("topic", None)
-                if topic:
-                    result["topic"] = topic
-
-            canonical_event = current_state.get((EventTypes.CanonicalAlias, ""))
-            if canonical_event:
-                canonical_alias = canonical_event.content.get("alias", None)
-                if canonical_alias:
-                    result["canonical_alias"] = canonical_alias
-
-            visibility_event = current_state.get((EventTypes.RoomHistoryVisibility, ""))
-            visibility = None
-            if visibility_event:
-                visibility = visibility_event.content.get("history_visibility", None)
-            result["world_readable"] = visibility == "world_readable"
-
-            guest_event = current_state.get((EventTypes.GuestAccess, ""))
-            guest = None
-            if guest_event:
-                guest = guest_event.content.get("guest_access", None)
-            result["guest_can_join"] = guest == "can_join"
-
-            avatar_event = current_state.get(("m.room.avatar", ""))
-            if avatar_event:
-                avatar_url = avatar_event.content.get("url", None)
-                if avatar_url:
-                    result["avatar_url"] = avatar_url
-
-            results.append(result)
-
-        yield concurrently_execute(handle_room, room_ids, 10)
-
-        # FIXME (erikj): START is no longer a valid value
-        defer.returnValue({"start": "START", "end": "END", "chunk": results})
-
-    @defer.inlineCallbacks
-    def fetch_all_remote_lists(self):
-        deferred = self.hs.get_replication_layer().get_public_rooms(
-            self.hs.config.secondary_directory_servers
-        )
-        self.remote_list_request_cache.set((), deferred)
-        self.remote_list_cache = yield deferred
-
-    @defer.inlineCallbacks
-    def get_aggregated_public_room_list(self):
-        """
-        Get the public room list from this server and the servers
-        specified in the secondary_directory_servers config option.
-        XXX: Pagination...
-        """
-        # We return the results from out cache which is updated by a looping call,
-        # unless we're missing a cache entry, in which case wait for the result
-        # of the fetch if there's one in progress. If not, omit that server.
-        wait = False
-        for s in self.hs.config.secondary_directory_servers:
-            if s not in self.remote_list_cache:
-                logger.warn("No cached room list from %s: waiting for fetch", s)
-                wait = True
-                break
-
-        if wait and self.remote_list_request_cache.get(()):
-            yield self.remote_list_request_cache.get(())
-
-        public_rooms = yield self.get_local_public_room_list()
-
-        # keep track of which room IDs we've seen so we can de-dup
-        room_ids = set()
-
-        # tag all the ones in our list with our server name.
-        # Also add the them to the de-deping set
-        for room in public_rooms['chunk']:
-            room["server_name"] = self.hs.hostname
-            room_ids.add(room["room_id"])
-
-        # Now add the results from federation
-        for server_name, server_result in self.remote_list_cache.items():
-            for room in server_result["chunk"]:
-                if room["room_id"] not in room_ids:
-                    room["server_name"] = server_name
-                    public_rooms["chunk"].append(room)
-                    room_ids.add(room["room_id"])
-
-        defer.returnValue(public_rooms)
-
-
 class RoomContextHandler(BaseHandler):
     @defer.inlineCallbacks
     def get_event_context(self, user, room_id, event_id, limit, is_guest):
@@ -568,6 +437,7 @@ class RoomEventSource(object):
             limit,
             room_ids,
             is_guest,
+            explicit_room_id=None,
     ):
         # We just ignore the key for now.
 
@@ -578,7 +448,7 @@ class RoomEventSource(object):
             logger.warn("Stream has topological part!!!! %r", from_key)
             from_key = "s%s" % (from_token.stream,)
 
-        app_service = yield self.store.get_app_service_by_user_id(
+        app_service = self.store.get_app_service_by_user_id(
             user.to_string()
         )
         if app_service:
@@ -616,8 +486,11 @@ class RoomEventSource(object):
 
         defer.returnValue((events, end_key))
 
-    def get_current_key(self, direction='f'):
-        return self.store.get_room_events_max_id(direction)
+    def get_current_key(self):
+        return self.store.get_room_events_max_id()
+
+    def get_current_key_for_room(self, room_id):
+        return self.store.get_room_events_max_id(room_id)
 
     @defer.inlineCallbacks
     def get_pagination_rows(self, user, config, key):

synapse/handlers/room_list.py

@@ -0,0 +1,442 @@
+# -*- coding: utf-8 -*-
+# Copyright 2014 - 2016 OpenMarket Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from twisted.internet import defer
+
+from ._base import BaseHandler
+
+from synapse.api.constants import (
+    EventTypes, JoinRules,
+)
+from synapse.util.async import concurrently_execute
+from synapse.util.caches.response_cache import ResponseCache
+from synapse.types import ThirdPartyInstanceID
+
+from collections import namedtuple
+from unpaddedbase64 import encode_base64, decode_base64
+
+import logging
+import msgpack
+
+logger = logging.getLogger(__name__)
+
+REMOTE_ROOM_LIST_POLL_INTERVAL = 60 * 1000
+
+
+# This is used to indicate we should only return rooms published to the main list.
+EMTPY_THIRD_PARTY_ID = ThirdPartyInstanceID(None, None)
+
+
+class RoomListHandler(BaseHandler):
+    def __init__(self, hs):
+        super(RoomListHandler, self).__init__(hs)
+        self.response_cache = ResponseCache(hs)
+        self.remote_response_cache = ResponseCache(hs, timeout_ms=30 * 1000)
+
+    def get_local_public_room_list(self, limit=None, since_token=None,
+                                   search_filter=None,
+                                   network_tuple=EMTPY_THIRD_PARTY_ID,):
+        """Generate a local public room list.
+
+        There are multiple different lists: the main one plus one per third
+        party network. A client can ask for a specific list or to return all.
+
+        Args:
+            limit (int)
+            since_token (str)
+            search_filter (dict)
+            network_tuple (ThirdPartyInstanceID): Which public list to use.
+                This can be (None, None) to indicate the main list, or a particular
+                appservice and network id to use an appservice specific one.
+                Setting to None returns all public rooms across all lists.
+        """
+        if search_filter:
+            # We explicitly don't bother caching searches or requests for
+            # appservice specific lists.
+            return self._get_public_room_list(
+                limit, since_token, search_filter, network_tuple=network_tuple,
+            )
+
+        key = (limit, since_token, network_tuple)
+        result = self.response_cache.get(key)
+        if not result:
+            result = self.response_cache.set(
+                key,
+                self._get_public_room_list(
+                    limit, since_token, network_tuple=network_tuple
+                )
+            )
+        return result
+
+    @defer.inlineCallbacks
+    def _get_public_room_list(self, limit=None, since_token=None,
+                              search_filter=None,
+                              network_tuple=EMTPY_THIRD_PARTY_ID,):
+        if since_token and since_token != "END":
+            since_token = RoomListNextBatch.from_token(since_token)
+        else:
+            since_token = None
+
+        rooms_to_order_value = {}
+        rooms_to_num_joined = {}
+        rooms_to_latest_event_ids = {}
+
+        newly_visible = []
+        newly_unpublished = []
+        if since_token:
+            stream_token = since_token.stream_ordering
+            current_public_id = yield self.store.get_current_public_room_stream_id()
+            public_room_stream_id = since_token.public_room_stream_id
+            newly_visible, newly_unpublished = yield self.store.get_public_room_changes(
+                public_room_stream_id, current_public_id,
+                network_tuple=network_tuple,
+            )
+        else:
+            stream_token = yield self.store.get_room_max_stream_ordering()
+            public_room_stream_id = yield self.store.get_current_public_room_stream_id()
+
+        room_ids = yield self.store.get_public_room_ids_at_stream_id(
+            public_room_stream_id, network_tuple=network_tuple,
+        )
+
+        # We want to return rooms in a particular order: the number of joined
+        # users. We then arbitrarily use the room_id as a tie breaker.
+
+        @defer.inlineCallbacks
+        def get_order_for_room(room_id):
+            latest_event_ids = rooms_to_latest_event_ids.get(room_id, None)
+            if not latest_event_ids:
+                latest_event_ids = yield self.store.get_forward_extremeties_for_room(
+                    room_id, stream_token
+                )
+                rooms_to_latest_event_ids[room_id] = latest_event_ids
+
+            if not latest_event_ids:
+                return
+
+            joined_users = yield self.state_handler.get_current_user_in_room(
+                room_id, latest_event_ids,
+            )
+            num_joined_users = len(joined_users)
+            rooms_to_num_joined[room_id] = num_joined_users
+
+            if num_joined_users == 0:
+                return
+
+            # We want larger rooms to be first, hence negating num_joined_users
+            rooms_to_order_value[room_id] = (-num_joined_users, room_id)
+
+        yield concurrently_execute(get_order_for_room, room_ids, 10)
+
+        sorted_entries = sorted(rooms_to_order_value.items(), key=lambda e: e[1])
+        sorted_rooms = [room_id for room_id, _ in sorted_entries]
+
+        # `sorted_rooms` should now be a list of all public room ids that is
+        # stable across pagination. Therefore, we can use indices into this
+        # list as our pagination tokens.
+
+        # Filter out rooms that we don't want to return
+        rooms_to_scan = [
+            r for r in sorted_rooms
+            if r not in newly_unpublished and rooms_to_num_joined[room_id] > 0
+        ]
+
+        total_room_count = len(rooms_to_scan)
+
+        if since_token:
+            # Filter out rooms we've already returned previously
+            # `since_token.current_limit` is the index of the last room we
+            # sent down, so we exclude it and everything before/after it.
+            if since_token.direction_is_forward:
+                rooms_to_scan = rooms_to_scan[since_token.current_limit + 1:]
+            else:
+                rooms_to_scan = rooms_to_scan[:since_token.current_limit]
+                rooms_to_scan.reverse()
+
+        # Actually generate the entries. _generate_room_entry will append to
+        # chunk but will stop if len(chunk) > limit
+        chunk = []
+        if limit and not search_filter:
+            step = limit + 1
+            for i in xrange(0, len(rooms_to_scan), step):
+                # We iterate here because the vast majority of cases we'll stop
+                # at first iteration, but occaisonally _generate_room_entry
+                # won't append to the chunk and so we need to loop again.
+                # We don't want to scan over the entire range either as that
+                # would potentially waste a lot of work.
+                yield concurrently_execute(
+                    lambda r: self._generate_room_entry(
+                        r, rooms_to_num_joined[r],
+                        chunk, limit, search_filter
+                    ),
+                    rooms_to_scan[i:i + step], 10
+                )
+                if len(chunk) >= limit + 1:
+                    break
+        else:
+            yield concurrently_execute(
+                lambda r: self._generate_room_entry(
+                    r, rooms_to_num_joined[r],
+                    chunk, limit, search_filter
+                ),
+                rooms_to_scan, 5
+            )
+
+        chunk.sort(key=lambda e: (-e["num_joined_members"], e["room_id"]))
+
+        # Work out the new limit of the batch for pagination, or None if we
+        # know there are no more results that would be returned.
+        # i.e., [since_token.current_limit..new_limit] is the batch of rooms
+        # we've returned (or the reverse if we paginated backwards)
+        # We tried to pull out limit + 1 rooms above, so if we have <= limit
+        # then we know there are no more results to return
+        new_limit = None
+        if chunk and (not limit or len(chunk) > limit):
+
+            if not since_token or since_token.direction_is_forward:
+                if limit:
+                    chunk = chunk[:limit]
+                last_room_id = chunk[-1]["room_id"]
+            else:
+                if limit:
+                    chunk = chunk[-limit:]
+                last_room_id = chunk[0]["room_id"]
+
+            new_limit = sorted_rooms.index(last_room_id)
+
+        results = {
+            "chunk": chunk,
+            "total_room_count_estimate": total_room_count,
+        }
+
+        if since_token:
+            results["new_rooms"] = bool(newly_visible)
+
+        if not since_token or since_token.direction_is_forward:
+            if new_limit is not None:
+                results["next_batch"] = RoomListNextBatch(
+                    stream_ordering=stream_token,
+                    public_room_stream_id=public_room_stream_id,
+                    current_limit=new_limit,
+                    direction_is_forward=True,
+                ).to_token()
+
+            if since_token:
+                results["prev_batch"] = since_token.copy_and_replace(
+                    direction_is_forward=False,
+                    current_limit=since_token.current_limit + 1,
+                ).to_token()
+        else:
+            if new_limit is not None:
+                results["prev_batch"] = RoomListNextBatch(
+                    stream_ordering=stream_token,
+                    public_room_stream_id=public_room_stream_id,
+                    current_limit=new_limit,
+                    direction_is_forward=False,
+                ).to_token()
+
+            if since_token:
+                results["next_batch"] = since_token.copy_and_replace(
+                    direction_is_forward=True,
+                    current_limit=since_token.current_limit - 1,
+                ).to_token()
+
+        defer.returnValue(results)
+
+    @defer.inlineCallbacks
+    def _generate_room_entry(self, room_id, num_joined_users, chunk, limit,
+                             search_filter):
+        if limit and len(chunk) > limit + 1:
+            # We've already got enough, so lets just drop it.
+            return
+
+        result = {
+            "room_id": room_id,
+            "num_joined_members": num_joined_users,
+        }
+
+        current_state_ids = yield self.state_handler.get_current_state_ids(room_id)
+
+        event_map = yield self.store.get_events([
+            event_id for key, event_id in current_state_ids.items()
+            if key[0] in (
+                EventTypes.JoinRules,
+                EventTypes.Name,
+                EventTypes.Topic,
+                EventTypes.CanonicalAlias,
+                EventTypes.RoomHistoryVisibility,
+                EventTypes.GuestAccess,
+                "m.room.avatar",
+            )
+        ])
+
+        current_state = {
+            (ev.type, ev.state_key): ev
+            for ev in event_map.values()
+        }
+
+        # Double check that this is actually a public room.
+        join_rules_event = current_state.get((EventTypes.JoinRules, ""))
+        if join_rules_event:
+            join_rule = join_rules_event.content.get("join_rule", None)
+            if join_rule and join_rule != JoinRules.PUBLIC:
+                defer.returnValue(None)
+
+        aliases = yield self.store.get_aliases_for_room(room_id)
+        if aliases:
+            result["aliases"] = aliases
+
+        name_event = yield current_state.get((EventTypes.Name, ""))
+        if name_event:
+            name = name_event.content.get("name", None)
+            if name:
+                result["name"] = name
+
+        topic_event = current_state.get((EventTypes.Topic, ""))
+        if topic_event:
+            topic = topic_event.content.get("topic", None)
+            if topic:
+                result["topic"] = topic
+
+        canonical_event = current_state.get((EventTypes.CanonicalAlias, ""))
+        if canonical_event:
+            canonical_alias = canonical_event.content.get("alias", None)
+            if canonical_alias:
+                result["canonical_alias"] = canonical_alias
+
+        visibility_event = current_state.get((EventTypes.RoomHistoryVisibility, ""))
+        visibility = None
+        if visibility_event:
+            visibility = visibility_event.content.get("history_visibility", None)
+        result["world_readable"] = visibility == "world_readable"
+
+        guest_event = current_state.get((EventTypes.GuestAccess, ""))
+        guest = None
+        if guest_event:
+            guest = guest_event.content.get("guest_access", None)
+        result["guest_can_join"] = guest == "can_join"
+
+        avatar_event = current_state.get(("m.room.avatar", ""))
+        if avatar_event:
+            avatar_url = avatar_event.content.get("url", None)
+            if avatar_url:
+                result["avatar_url"] = avatar_url
+
+        if _matches_room_entry(result, search_filter):
+            chunk.append(result)
+
+    @defer.inlineCallbacks
+    def get_remote_public_room_list(self, server_name, limit=None, since_token=None,
+                                    search_filter=None, include_all_networks=False,
+                                    third_party_instance_id=None,):
+        if search_filter:
+            # We currently don't support searching across federation, so we have
+            # to do it manually without pagination
+            limit = None
+            since_token = None
+
+        res = yield self._get_remote_list_cached(
+            server_name, limit=limit, since_token=since_token,
+            include_all_networks=include_all_networks,
+            third_party_instance_id=third_party_instance_id,
+        )
+
+        if search_filter:
+            res = {"chunk": [
+                entry
+                for entry in list(res.get("chunk", []))
+                if _matches_room_entry(entry, search_filter)
+            ]}
+
+        defer.returnValue(res)
+
+    def _get_remote_list_cached(self, server_name, limit=None, since_token=None,
+                                search_filter=None, include_all_networks=False,
+                                third_party_instance_id=None,):
+        repl_layer = self.hs.get_replication_layer()
+        if search_filter:
+            # We can't cache when asking for search
+            return repl_layer.get_public_rooms(
+                server_name, limit=limit, since_token=since_token,
+                search_filter=search_filter, include_all_networks=include_all_networks,
+                third_party_instance_id=third_party_instance_id,
+            )
+
+        key = (
+            server_name, limit, since_token, include_all_networks,
+            third_party_instance_id,
+        )
+        result = self.remote_response_cache.get(key)
+        if not result:
+            result = self.remote_response_cache.set(
+                key,
+                repl_layer.get_public_rooms(
+                    server_name, limit=limit, since_token=since_token,
+                    search_filter=search_filter,
+                    include_all_networks=include_all_networks,
+                    third_party_instance_id=third_party_instance_id,
+                )
+            )
+        return result
+
+
+class RoomListNextBatch(namedtuple("RoomListNextBatch", (
+    "stream_ordering",  # stream_ordering of the first public room list
+    "public_room_stream_id",  # public room stream id for first public room list
+    "current_limit",  # The number of previous rooms returned
+    "direction_is_forward",  # Bool if this is a next_batch, false if prev_batch
+))):
+
+    KEY_DICT = {
+        "stream_ordering": "s",
+        "public_room_stream_id": "p",
+        "current_limit": "n",
+        "direction_is_forward": "d",
+    }
+
+    REVERSE_KEY_DICT = {v: k for k, v in KEY_DICT.items()}
+
+    @classmethod
+    def from_token(cls, token):
+        return RoomListNextBatch(**{
+            cls.REVERSE_KEY_DICT[key]: val
+            for key, val in msgpack.loads(decode_base64(token)).items()
+        })
+
+    def to_token(self):
+        return encode_base64(msgpack.dumps({
+            self.KEY_DICT[key]: val
+            for key, val in self._asdict().items()
+        }))
+
+    def copy_and_replace(self, **kwds):
+        return self._replace(
+            **kwds
+        )
+
+
+def _matches_room_entry(room_entry, search_filter):
+    if search_filter and search_filter.get("generic_search_term", None):
+        generic_search_term = search_filter["generic_search_term"].upper()
+        if generic_search_term in room_entry.get("name", "").upper():
+            return True
+        elif generic_search_term in room_entry.get("topic", "").upper():
+            return True
+        elif generic_search_term in room_entry.get("canonical_alias", "").upper():
+            return True
+    else:
+        return True
+
+    return False

synapse/handlers/room_member.py

@@ -14,24 +14,22 @@
 # limitations under the License.
 
 
+import logging
+
+from signedjson.key import decode_verify_key_bytes
+from signedjson.sign import verify_signed_json
 from twisted.internet import defer
+from unpaddedbase64 import decode_base64
 
-from ._base import BaseHandler
-
-from synapse.types import UserID, RoomID, Requester
+import synapse.types
 from synapse.api.constants import (
     EventTypes, Membership,
 )
 from synapse.api.errors import AuthError, SynapseError, Codes
+from synapse.types import UserID, RoomID
 from synapse.util.async import Linearizer
 from synapse.util.distributor import user_left_room, user_joined_room
-
-from signedjson.sign import verify_signed_json
-from signedjson.key import decode_verify_key_bytes
-
-from unpaddedbase64 import decode_base64
-
-import logging
+from ._base import BaseHandler
 
 logger = logging.getLogger(__name__)
 
@@ -47,7 +45,7 @@ class RoomMemberHandler(BaseHandler):
     def __init__(self, hs):
         super(RoomMemberHandler, self).__init__(hs)
 
-        self.member_linearizer = Linearizer()
+        self.member_linearizer = Linearizer(name="member")
 
         self.clock = hs.get_clock()
 
@@ -61,10 +59,13 @@ class RoomMemberHandler(BaseHandler):
         prev_event_ids,
         txn_id=None,
         ratelimit=True,
+        content=None,
     ):
+        if content is None:
+            content = {}
         msg_handler = self.hs.get_handlers().message_handler
 
-        content = {"membership": membership}
+        content["membership"] = membership
         if requester.is_guest:
             content["kind"] = "guest"
 
@@ -84,6 +85,12 @@ class RoomMemberHandler(BaseHandler):
             prev_event_ids=prev_event_ids,
         )
 
+        # Check if this event matches the previous membership event for the user.
+        duplicate = yield msg_handler.deduplicate_state_event(event, context)
+        if duplicate is not None:
+            # Discard the new event since this membership change is a no-op.
+            defer.returnValue(duplicate)
+
         yield msg_handler.handle_new_client_event(
             requester,
             event,
@@ -92,20 +99,28 @@ class RoomMemberHandler(BaseHandler):
             ratelimit=ratelimit,
         )
 
-        prev_member_event = context.current_state.get(
+        prev_member_event_id = context.prev_state_ids.get(
             (EventTypes.Member, target.to_string()),
             None
         )
 
         if event.membership == Membership.JOIN:
-            if not prev_member_event or prev_member_event.membership != Membership.JOIN:
-                # Only fire user_joined_room if the user has acutally joined the
-                # room. Don't bother if the user is just changing their profile
-                # info.
+            # Only fire user_joined_room if the user has acutally joined the
+            # room. Don't bother if the user is just changing their profile
+            # info.
+            newly_joined = True
+            if prev_member_event_id:
+                prev_member_event = yield self.store.get_event(prev_member_event_id)
+                newly_joined = prev_member_event.membership != Membership.JOIN
+            if newly_joined:
                 yield user_joined_room(self.distributor, target, room_id)
         elif event.membership == Membership.LEAVE:
-            if prev_member_event and prev_member_event.membership == Membership.JOIN:
-                user_left_room(self.distributor, target, room_id)
+            if prev_member_event_id:
+                prev_member_event = yield self.store.get_event(prev_member_event_id)
+                if prev_member_event.membership == Membership.JOIN:
+                    user_left_room(self.distributor, target, room_id)
+
+        defer.returnValue(event)
 
     @defer.inlineCallbacks
     def remote_join(self, remote_room_hosts, room_id, user, content):
@@ -142,8 +157,9 @@ class RoomMemberHandler(BaseHandler):
             remote_room_hosts=None,
             third_party_signed=None,
             ratelimit=True,
+            content=None,
     ):
-        key = (target, room_id,)
+        key = (room_id,)
 
         with (yield self.member_linearizer.queue(key)):
             result = yield self._update_membership(
@@ -155,6 +171,7 @@ class RoomMemberHandler(BaseHandler):
                 remote_room_hosts=remote_room_hosts,
                 third_party_signed=third_party_signed,
                 ratelimit=ratelimit,
+                content=content,
             )
 
         defer.returnValue(result)
@@ -170,7 +187,12 @@ class RoomMemberHandler(BaseHandler):
             remote_room_hosts=None,
             third_party_signed=None,
             ratelimit=True,
+            content=None,
     ):
+        content_specified = bool(content)
+        if content is None:
+            content = {}
+
         effective_membership_state = action
         if action in ["kick", "unban"]:
             effective_membership_state = "leave"
@@ -188,43 +210,56 @@ class RoomMemberHandler(BaseHandler):
             remote_room_hosts = []
 
         latest_event_ids = yield self.store.get_latest_event_ids_in_room(room_id)
-        current_state = yield self.state_handler.get_current_state(
+        current_state_ids = yield self.state_handler.get_current_state_ids(
             room_id, latest_event_ids=latest_event_ids,
         )
 
-        old_state = current_state.get((EventTypes.Member, target.to_string()))
-        old_membership = old_state.content.get("membership") if old_state else None
-        if action == "unban" and old_membership != "ban":
-            raise SynapseError(
-                403,
-                "Cannot unban user who was not banned (membership=%s)" % old_membership,
-                errcode=Codes.BAD_STATE
-            )
-        if old_membership == "ban" and action != "unban":
-            raise SynapseError(
-                403,
-                "Cannot %s user who was banned" % (action,),
-                errcode=Codes.BAD_STATE
-            )
+        old_state_id = current_state_ids.get((EventTypes.Member, target.to_string()))
+        if old_state_id:
+            old_state = yield self.store.get_event(old_state_id, allow_none=True)
+            old_membership = old_state.content.get("membership") if old_state else None
+            if action == "unban" and old_membership != "ban":
+                raise SynapseError(
+                    403,
+                    "Cannot unban user who was not banned"
+                    " (membership=%s)" % old_membership,
+                    errcode=Codes.BAD_STATE
+                )
+            if old_membership == "ban" and action != "unban":
+                raise SynapseError(
+                    403,
+                    "Cannot %s user who was banned" % (action,),
+                    errcode=Codes.BAD_STATE
+                )
 
-        is_host_in_room = self.is_host_in_room(current_state)
+            if old_state:
+                same_content = content == old_state.content
+                same_membership = old_membership == effective_membership_state
+                same_sender = requester.user.to_string() == old_state.sender
+                if same_sender and same_membership and same_content:
+                    defer.returnValue(old_state)
+
+        is_host_in_room = yield self._is_host_in_room(current_state_ids)
 
         if effective_membership_state == Membership.JOIN:
-            if requester.is_guest and not self._can_guest_join(current_state):
-                # This should be an auth check, but guests are a local concept,
-                # so don't really fit into the general auth process.
-                raise AuthError(403, "Guest access not allowed")
+            if requester.is_guest:
+                guest_can_join = yield self._can_guest_join(current_state_ids)
+                if not guest_can_join:
+                    # This should be an auth check, but guests are a local concept,
+                    # so don't really fit into the general auth process.
+                    raise AuthError(403, "Guest access not allowed")
 
             if not is_host_in_room:
                 inviter = yield self.get_inviter(target.to_string(), room_id)
                 if inviter and not self.hs.is_mine(inviter):
                     remote_room_hosts.append(inviter.domain)
 
-                content = {"membership": Membership.JOIN}
+                content["membership"] = Membership.JOIN
 
                 profile = self.hs.get_handlers().profile_handler
-                content["displayname"] = yield profile.get_displayname(target)
-                content["avatar_url"] = yield profile.get_avatar_url(target)
+                if not content_specified:
+                    content["displayname"] = yield profile.get_displayname(target)
+                    content["avatar_url"] = yield profile.get_avatar_url(target)
 
                 if requester.is_guest:
                     content["kind"] = "guest"
@@ -266,7 +301,7 @@ class RoomMemberHandler(BaseHandler):
 
                         defer.returnValue({})
 
-        yield self._local_membership_update(
+        res = yield self._local_membership_update(
             requester=requester,
             target=target,
             room_id=room_id,
@@ -274,7 +309,9 @@ class RoomMemberHandler(BaseHandler):
             txn_id=txn_id,
             ratelimit=ratelimit,
             prev_event_ids=latest_event_ids,
+            content=content,
         )
+        defer.returnValue(res)
 
     @defer.inlineCallbacks
     def send_membership_event(
@@ -315,18 +352,20 @@ class RoomMemberHandler(BaseHandler):
             )
             assert self.hs.is_mine(sender), "Sender must be our own: %s" % (sender,)
         else:
-            requester = Requester(target_user, None, False)
+            requester = synapse.types.create_requester(target_user)
 
         message_handler = self.hs.get_handlers().message_handler
-        prev_event = message_handler.deduplicate_state_event(event, context)
+        prev_event = yield message_handler.deduplicate_state_event(event, context)
         if prev_event is not None:
             return
 
         if event.membership == Membership.JOIN:
-            if requester.is_guest and not self._can_guest_join(context.current_state):
-                # This should be an auth check, but guests are a local concept,
-                # so don't really fit into the general auth process.
-                raise AuthError(403, "Guest access not allowed")
+            if requester.is_guest:
+                guest_can_join = yield self._can_guest_join(context.prev_state_ids)
+                if not guest_can_join:
+                    # This should be an auth check, but guests are a local concept,
+                    # so don't really fit into the general auth process.
+                    raise AuthError(403, "Guest access not allowed")
 
         yield message_handler.handle_new_client_event(
             requester,
@@ -336,27 +375,39 @@ class RoomMemberHandler(BaseHandler):
             ratelimit=ratelimit,
         )
 
-        prev_member_event = context.current_state.get(
-            (EventTypes.Member, target_user.to_string()),
+        prev_member_event_id = context.prev_state_ids.get(
+            (EventTypes.Member, event.state_key),
             None
         )
 
         if event.membership == Membership.JOIN:
-            if not prev_member_event or prev_member_event.membership != Membership.JOIN:
-                # Only fire user_joined_room if the user has acutally joined the
-                # room. Don't bother if the user is just changing their profile
-                # info.
+            # Only fire user_joined_room if the user has acutally joined the
+            # room. Don't bother if the user is just changing their profile
+            # info.
+            newly_joined = True
+            if prev_member_event_id:
+                prev_member_event = yield self.store.get_event(prev_member_event_id)
+                newly_joined = prev_member_event.membership != Membership.JOIN
+            if newly_joined:
                 yield user_joined_room(self.distributor, target_user, room_id)
         elif event.membership == Membership.LEAVE:
-            if prev_member_event and prev_member_event.membership == Membership.JOIN:
-                user_left_room(self.distributor, target_user, room_id)
+            if prev_member_event_id:
+                prev_member_event = yield self.store.get_event(prev_member_event_id)
+                if prev_member_event.membership == Membership.JOIN:
+                    user_left_room(self.distributor, target_user, room_id)
 
-    def _can_guest_join(self, current_state):
+    @defer.inlineCallbacks
+    def _can_guest_join(self, current_state_ids):
         """
         Returns whether a guest can join a room based on its current state.
         """
-        guest_access = current_state.get((EventTypes.GuestAccess, ""), None)
-        return (
+        guest_access_id = current_state_ids.get((EventTypes.GuestAccess, ""), None)
+        if not guest_access_id:
+            defer.returnValue(False)
+
+        guest_access = yield self.store.get_event(guest_access_id)
+
+        defer.returnValue(
             guest_access
             and guest_access.content
             and "guest_access" in guest_access.content
@@ -675,3 +726,24 @@ class RoomMemberHandler(BaseHandler):
 
         if membership:
             yield self.store.forget(user_id, room_id)
+
+    @defer.inlineCallbacks
+    def _is_host_in_room(self, current_state_ids):
+        # Have we just created the room, and is this about to be the very
+        # first member event?
+        create_event_id = current_state_ids.get(("m.room.create", ""))
+        if len(current_state_ids) == 1 and create_event_id:
+            defer.returnValue(self.hs.is_mine_id(create_event_id))
+
+        for (etype, state_key), event_id in current_state_ids.items():
+            if etype != EventTypes.Member or not self.hs.is_mine_id(state_key):
+                continue
+
+            event = yield self.store.get_event(event_id, allow_none=True)
+            if not event:
+                continue
+
+            if event.membership == Membership.JOIN:
+                defer.returnValue(True)
+
+        defer.returnValue(False)

synapse/handlers/sync.py

@@ -35,6 +35,7 @@ SyncConfig = collections.namedtuple("SyncConfig", [
     "filter_collection",
     "is_guest",
     "request_key",
+    "device_id",
 ])
 
 
@@ -113,6 +114,8 @@ class SyncResult(collections.namedtuple("SyncResult", [
     "joined",  # JoinedSyncResult for each joined room.
     "invited",  # InvitedSyncResult for each invited room.
     "archived",  # ArchivedSyncResult for each archived room.
+    "to_device",  # List of direct messages for the device.
+    "device_lists",  # List of user_ids whose devices have chanegd
 ])):
     __slots__ = []
 
@@ -126,7 +129,9 @@ class SyncResult(collections.namedtuple("SyncResult", [
             self.joined or
             self.invited or
             self.archived or
-            self.account_data
+            self.account_data or
+            self.to_device or
+            self.device_lists
         )
 
 
@@ -138,7 +143,8 @@ class SyncHandler(object):
         self.presence_handler = hs.get_presence_handler()
         self.event_sources = hs.get_event_sources()
         self.clock = hs.get_clock()
-        self.response_cache = ResponseCache()
+        self.response_cache = ResponseCache(hs)
+        self.state = hs.get_state_handler()
 
     def wait_for_sync_for_user(self, sync_config, since_token=None, timeout=0,
                                full_state=False):
@@ -273,6 +279,7 @@ class SyncHandler(object):
         """
         with Measure(self.clock, "load_filtered_recents"):
             timeline_limit = sync_config.filter_collection.timeline_limit()
+            block_all_timeline = sync_config.filter_collection.blocks_all_room_timeline()
 
             if recents is None or newly_joined_room or timeline_limit < len(recents):
                 limited = True
@@ -289,7 +296,7 @@ class SyncHandler(object):
             else:
                 recents = []
 
-            if not limited:
+            if not limited or block_all_timeline:
                 defer.returnValue(TimelineBatch(
                     events=recents,
                     prev_batch=now_token,
@@ -355,11 +362,11 @@ class SyncHandler(object):
         Returns:
             A Deferred map from ((type, state_key)->Event)
         """
-        state = yield self.store.get_state_for_event(event.event_id)
+        state_ids = yield self.store.get_state_ids_for_event(event.event_id)
         if event.is_state():
-            state = state.copy()
-            state[(event.type, event.state_key)] = event
-        defer.returnValue(state)
+            state_ids = state_ids.copy()
+            state_ids[(event.type, event.state_key)] = event.event_id
+        defer.returnValue(state_ids)
 
     @defer.inlineCallbacks
     def get_state_at(self, room_id, stream_position):
@@ -412,62 +419,66 @@ class SyncHandler(object):
         with Measure(self.clock, "compute_state_delta"):
             if full_state:
                 if batch:
-                    current_state = yield self.store.get_state_for_event(
+                    current_state_ids = yield self.store.get_state_ids_for_event(
                         batch.events[-1].event_id
                     )
 
-                    state = yield self.store.get_state_for_event(
+                    state_ids = yield self.store.get_state_ids_for_event(
                         batch.events[0].event_id
                     )
                 else:
-                    current_state = yield self.get_state_at(
+                    current_state_ids = yield self.get_state_at(
                         room_id, stream_position=now_token
                     )
 
-                    state = current_state
+                    state_ids = current_state_ids
 
                 timeline_state = {
-                    (event.type, event.state_key): event
+                    (event.type, event.state_key): event.event_id
                     for event in batch.events if event.is_state()
                 }
 
-                state = _calculate_state(
+                state_ids = _calculate_state(
                     timeline_contains=timeline_state,
-                    timeline_start=state,
+                    timeline_start=state_ids,
                     previous={},
-                    current=current_state,
+                    current=current_state_ids,
                 )
             elif batch.limited:
                 state_at_previous_sync = yield self.get_state_at(
                     room_id, stream_position=since_token
                 )
 
-                current_state = yield self.store.get_state_for_event(
+                current_state_ids = yield self.store.get_state_ids_for_event(
                     batch.events[-1].event_id
                 )
 
-                state_at_timeline_start = yield self.store.get_state_for_event(
+                state_at_timeline_start = yield self.store.get_state_ids_for_event(
                     batch.events[0].event_id
                 )
 
                 timeline_state = {
-                    (event.type, event.state_key): event
+                    (event.type, event.state_key): event.event_id
                     for event in batch.events if event.is_state()
                 }
 
-                state = _calculate_state(
+                state_ids = _calculate_state(
                     timeline_contains=timeline_state,
                     timeline_start=state_at_timeline_start,
                     previous=state_at_previous_sync,
-                    current=current_state,
+                    current=current_state_ids,
                 )
             else:
-                state = {}
+                state_ids = {}
 
-            defer.returnValue({
-                (e.type, e.state_key): e
-                for e in sync_config.filter_collection.filter_room_state(state.values())
-            })
+        state = {}
+        if state_ids:
+            state = yield self.store.get_events(state_ids.values())
+
+        defer.returnValue({
+            (e.type, e.state_key): e
+            for e in sync_config.filter_collection.filter_room_state(state.values())
+        })
 
     @defer.inlineCallbacks
     def unread_notifs_for_room_id(self, room_id, sync_config):
@@ -485,9 +496,9 @@ class SyncHandler(object):
                 )
                 defer.returnValue(notifs)
 
-            # There is no new information in this period, so your notification
-            # count is whatever it was last time.
-            defer.returnValue(None)
+        # There is no new information in this period, so your notification
+        # count is whatever it was last time.
+        defer.returnValue(None)
 
     @defer.inlineCallbacks
     def generate_sync_result(self, sync_config, since_token=None, full_state=False):
@@ -501,6 +512,7 @@ class SyncHandler(object):
         Returns:
             Deferred(SyncResult)
         """
+        logger.info("Calculating sync response for %r", sync_config.user)
 
         # NB: The now_token gets changed by some of the generate_sync_* methods,
         # this is due to some of the underlying streams not supporting the ability
@@ -523,8 +535,19 @@ class SyncHandler(object):
         )
         newly_joined_rooms, newly_joined_users = res
 
-        yield self._generate_sync_entry_for_presence(
-            sync_result_builder, newly_joined_rooms, newly_joined_users
+        block_all_presence_data = (
+            since_token is None and
+            sync_config.filter_collection.blocks_all_presence()
+        )
+        if not block_all_presence_data:
+            yield self._generate_sync_entry_for_presence(
+                sync_result_builder, newly_joined_rooms, newly_joined_users
+            )
+
+        yield self._generate_sync_entry_for_to_device(sync_result_builder)
+
+        device_lists = yield self._generate_sync_entry_for_device_list(
+            sync_result_builder
         )
 
         defer.returnValue(SyncResult(
@@ -533,9 +556,76 @@ class SyncHandler(object):
             joined=sync_result_builder.joined,
             invited=sync_result_builder.invited,
             archived=sync_result_builder.archived,
+            to_device=sync_result_builder.to_device,
+            device_lists=device_lists,
             next_batch=sync_result_builder.now_token,
         ))
 
+    @defer.inlineCallbacks
+    def _generate_sync_entry_for_device_list(self, sync_result_builder):
+        user_id = sync_result_builder.sync_config.user.to_string()
+        since_token = sync_result_builder.since_token
+
+        if since_token and since_token.device_list_key:
+            rooms = yield self.store.get_rooms_for_user(user_id)
+            room_ids = set(r.room_id for r in rooms)
+
+            user_ids_changed = set()
+            changed = yield self.store.get_user_whose_devices_changed(
+                since_token.device_list_key
+            )
+            for other_user_id in changed:
+                other_rooms = yield self.store.get_rooms_for_user(other_user_id)
+                if room_ids.intersection(e.room_id for e in other_rooms):
+                    user_ids_changed.add(other_user_id)
+
+            defer.returnValue(user_ids_changed)
+        else:
+            defer.returnValue([])
+
+    @defer.inlineCallbacks
+    def _generate_sync_entry_for_to_device(self, sync_result_builder):
+        """Generates the portion of the sync response. Populates
+        `sync_result_builder` with the result.
+
+        Args:
+            sync_result_builder(SyncResultBuilder)
+
+        Returns:
+            Deferred(dict): A dictionary containing the per room account data.
+        """
+        user_id = sync_result_builder.sync_config.user.to_string()
+        device_id = sync_result_builder.sync_config.device_id
+        now_token = sync_result_builder.now_token
+        since_stream_id = 0
+        if sync_result_builder.since_token is not None:
+            since_stream_id = int(sync_result_builder.since_token.to_device_key)
+
+        if since_stream_id != int(now_token.to_device_key):
+            # We only delete messages when a new message comes in, but that's
+            # fine so long as we delete them at some point.
+
+            deleted = yield self.store.delete_messages_for_device(
+                user_id, device_id, since_stream_id
+            )
+            logger.info("Deleted %d to-device messages up to %d",
+                        deleted, since_stream_id)
+
+            messages, stream_id = yield self.store.get_new_messages_for_device(
+                user_id, device_id, since_stream_id, now_token.to_device_key
+            )
+
+            logger.info(
+                "Returning %d to-device messages between %d and %d (current token: %d)",
+                len(messages), since_stream_id, stream_id, now_token.to_device_key
+            )
+            sync_result_builder.now_token = now_token.copy_and_replace(
+                "to_device_key", stream_id
+            )
+            sync_result_builder.to_device = messages
+        else:
+            sync_result_builder.to_device = []
+
     @defer.inlineCallbacks
     def _generate_sync_entry_for_account_data(self, sync_result_builder):
         """Generates the account data portion of the sync response. Populates
@@ -626,7 +716,7 @@ class SyncHandler(object):
 
         extra_users_ids = set(newly_joined_users)
         for room_id in newly_joined_rooms:
-            users = yield self.store.get_users_in_room(room_id)
+            users = yield self.state.get_current_user_in_room(room_id)
             extra_users_ids.update(users)
         extra_users_ids.discard(user.to_string())
 
@@ -659,13 +749,20 @@ class SyncHandler(object):
             `(newly_joined_rooms, newly_joined_users)`
         """
         user_id = sync_result_builder.sync_config.user.to_string()
-
-        now_token, ephemeral_by_room = yield self.ephemeral_by_room(
-            sync_result_builder.sync_config,
-            now_token=sync_result_builder.now_token,
-            since_token=sync_result_builder.since_token,
+        block_all_room_ephemeral = (
+            sync_result_builder.since_token is None and
+            sync_result_builder.sync_config.filter_collection.blocks_all_room_ephemeral()
         )
-        sync_result_builder.now_token = now_token
+
+        if block_all_room_ephemeral:
+            ephemeral_by_room = {}
+        else:
+            now_token, ephemeral_by_room = yield self.ephemeral_by_room(
+                sync_result_builder.sync_config,
+                now_token=sync_result_builder.now_token,
+                since_token=sync_result_builder.since_token,
+            )
+            sync_result_builder.now_token = now_token
 
         ignored_account_data = yield self.store.get_global_account_data_by_type_for_user(
             "m.ignored_user_list", user_id=user_id,
@@ -738,7 +835,7 @@ class SyncHandler(object):
 
         assert since_token
 
-        app_service = yield self.store.get_app_service_by_user_id(user_id)
+        app_service = self.store.get_app_service_by_user_id(user_id)
         if app_service:
             rooms = yield self.store.get_app_service_rooms(app_service)
             joined_room_ids = set(r.room_id for r in rooms)
@@ -766,8 +863,13 @@ class SyncHandler(object):
             # the last sync (even if we have since left). This is to make sure
             # we do send down the room, and with full state, where necessary
             if room_id in joined_room_ids or has_join:
-                old_state = yield self.get_state_at(room_id, since_token)
-                old_mem_ev = old_state.get((EventTypes.Member, user_id), None)
+                old_state_ids = yield self.get_state_at(room_id, since_token)
+                old_mem_ev_id = old_state_ids.get((EventTypes.Member, user_id), None)
+                old_mem_ev = None
+                if old_mem_ev_id:
+                    old_mem_ev = yield self.store.get_event(
+                        old_mem_ev_id, allow_none=True
+                    )
                 if not old_mem_ev or old_mem_ev.membership != Membership.JOIN:
                     newly_joined_rooms.append(room_id)
 
@@ -1059,27 +1161,25 @@ def _calculate_state(timeline_contains, timeline_start, previous, current):
     Returns:
         dict
     """
-    event_id_to_state = {
-        e.event_id: e
-        for e in itertools.chain(
-            timeline_contains.values(),
-            previous.values(),
-            timeline_start.values(),
-            current.values(),
+    event_id_to_key = {
+        e: key
+        for key, e in itertools.chain(
+            timeline_contains.items(),
+            previous.items(),
+            timeline_start.items(),
+            current.items(),
         )
     }
 
-    c_ids = set(e.event_id for e in current.values())
-    tc_ids = set(e.event_id for e in timeline_contains.values())
-    p_ids = set(e.event_id for e in previous.values())
-    ts_ids = set(e.event_id for e in timeline_start.values())
+    c_ids = set(e for e in current.values())
+    tc_ids = set(e for e in timeline_contains.values())
+    p_ids = set(e for e in previous.values())
+    ts_ids = set(e for e in timeline_start.values())
 
     state_ids = ((c_ids | ts_ids) - p_ids) - tc_ids
 
-    evs = (event_id_to_state[e] for e in state_ids)
     return {
-        (e.type, e.state_key): e
-        for e in evs
+        event_id_to_key[e]: e for e in state_ids
     }
 
 
@@ -1103,6 +1203,7 @@ class SyncResultBuilder(object):
         self.joined = []
         self.invited = []
         self.archived = []
+        self.device = []
 
 
 class RoomSyncResultBuilder(object):

synapse/handlers/typing.py

@@ -16,9 +16,10 @@
 from twisted.internet import defer
 
 from synapse.api.errors import SynapseError, AuthError
-from synapse.util.logcontext import PreserveLoggingContext
+from synapse.util.logcontext import preserve_fn
 from synapse.util.metrics import Measure
-from synapse.types import UserID
+from synapse.util.wheel_timer import WheelTimer
+from synapse.types import UserID, get_domain_from_id
 
 import logging
 
@@ -33,6 +34,13 @@ logger = logging.getLogger(__name__)
 RoomMember = namedtuple("RoomMember", ("room_id", "user_id"))
 
 
+# How often we expect remote servers to resend us presence.
+FEDERATION_TIMEOUT = 60 * 1000
+
+# How often to resend typing across federation.
+FEDERATION_PING_INTERVAL = 40 * 1000
+
+
 class TypingHandler(object):
     def __init__(self, hs):
         self.store = hs.get_datastore()
@@ -40,17 +48,21 @@ class TypingHandler(object):
         self.auth = hs.get_auth()
         self.is_mine_id = hs.is_mine_id
         self.notifier = hs.get_notifier()
+        self.state = hs.get_state_handler()
+
+        self.hs = hs
 
         self.clock = hs.get_clock()
+        self.wheel_timer = WheelTimer(bucket_size=5000)
 
-        self.federation = hs.get_replication_layer()
+        self.federation = hs.get_federation_sender()
 
-        self.federation.register_edu_handler("m.typing", self._recv_edu)
+        hs.get_replication_layer().register_edu_handler("m.typing", self._recv_edu)
 
         hs.get_distributor().observe("user_left_room", self.user_left_room)
 
         self._member_typing_until = {}  # clock time we expect to stop
-        self._member_typing_timer = {}  # deferreds to manage theabove
+        self._member_last_federation_poke = {}
 
         # map room IDs to serial numbers
         self._room_serials = {}
@@ -58,12 +70,49 @@ class TypingHandler(object):
         # map room IDs to sets of users currently typing
         self._room_typing = {}
 
-    def tearDown(self):
-        """Cancels all the pending timers.
-        Normally this shouldn't be needed, but it's required from unit tests
-        to avoid a "Reactor was unclean" warning."""
-        for t in self._member_typing_timer.values():
-            self.clock.cancel_call_later(t)
+        self.clock.looping_call(
+            self._handle_timeouts,
+            5000,
+        )
+
+    def _handle_timeouts(self):
+        logger.info("Checking for typing timeouts")
+
+        now = self.clock.time_msec()
+
+        members = set(self.wheel_timer.fetch(now))
+
+        for member in members:
+            if not self.is_typing(member):
+                # Nothing to do if they're no longer typing
+                continue
+
+            until = self._member_typing_until.get(member, None)
+            if not until or until <= now:
+                logger.info("Timing out typing for: %s", member.user_id)
+                preserve_fn(self._stopped_typing)(member)
+                continue
+
+            # Check if we need to resend a keep alive over federation for this
+            # user.
+            if self.hs.is_mine_id(member.user_id):
+                last_fed_poke = self._member_last_federation_poke.get(member, None)
+                if not last_fed_poke or last_fed_poke + FEDERATION_PING_INTERVAL <= now:
+                    preserve_fn(self._push_remote)(
+                        member=member,
+                        typing=True
+                    )
+
+            # Add a paranoia timer to ensure that we always have a timer for
+            # each person typing.
+            self.wheel_timer.insert(
+                now=now,
+                obj=member,
+                then=now + 60 * 1000,
+            )
+
+    def is_typing(self, member):
+        return member.user_id in self._room_typing.get(member.room_id, [])
 
     @defer.inlineCallbacks
     def started_typing(self, target_user, auth_user, room_id, timeout):
@@ -82,23 +131,17 @@ class TypingHandler(object):
             "%s has started typing in %s", target_user_id, room_id
         )
 
-        until = self.clock.time_msec() + timeout
         member = RoomMember(room_id=room_id, user_id=target_user_id)
 
-        was_present = member in self._member_typing_until
+        was_present = member.user_id in self._room_typing.get(room_id, set())
 
-        if member in self._member_typing_timer:
-            self.clock.cancel_call_later(self._member_typing_timer[member])
+        now = self.clock.time_msec()
+        self._member_typing_until[member] = now + timeout
 
-        def _cb():
-            logger.debug(
-                "%s has timed out in %s", target_user.to_string(), room_id
-            )
-            self._stopped_typing(member)
-
-        self._member_typing_until[member] = until
-        self._member_typing_timer[member] = self.clock.call_later(
-            timeout / 1000.0, _cb
+        self.wheel_timer.insert(
+            now=now,
+            obj=member,
+            then=now + timeout,
         )
 
         if was_present:
@@ -106,8 +149,7 @@ class TypingHandler(object):
             defer.returnValue(None)
 
         yield self._push_update(
-            room_id=room_id,
-            user_id=target_user_id,
+            member=member,
             typing=True,
         )
 
@@ -130,10 +172,6 @@ class TypingHandler(object):
 
         member = RoomMember(room_id=room_id, user_id=target_user_id)
 
-        if member in self._member_typing_timer:
-            self.clock.cancel_call_later(self._member_typing_timer[member])
-            del self._member_typing_timer[member]
-
         yield self._stopped_typing(member)
 
     @defer.inlineCallbacks
@@ -145,79 +183,101 @@ class TypingHandler(object):
 
     @defer.inlineCallbacks
     def _stopped_typing(self, member):
-        if member not in self._member_typing_until:
+        if member.user_id not in self._room_typing.get(member.room_id, set()):
             # No point
             defer.returnValue(None)
 
+        self._member_typing_until.pop(member, None)
+        self._member_last_federation_poke.pop(member, None)
+
         yield self._push_update(
-            room_id=member.room_id,
-            user_id=member.user_id,
+            member=member,
             typing=False,
         )
 
-        del self._member_typing_until[member]
+    @defer.inlineCallbacks
+    def _push_update(self, member, typing):
+        if self.hs.is_mine_id(member.user_id):
+            # Only send updates for changes to our own users.
+            yield self._push_remote(member, typing)
 
-        if member in self._member_typing_timer:
-            # Don't cancel it - either it already expired, or the real
-            # stopped_typing() will cancel it
-            del self._member_typing_timer[member]
+        self._push_update_local(
+            member=member,
+            typing=typing
+        )
 
     @defer.inlineCallbacks
-    def _push_update(self, room_id, user_id, typing):
-        domains = yield self.store.get_joined_hosts_for_room(room_id)
+    def _push_remote(self, member, typing):
+        users = yield self.state.get_current_user_in_room(member.room_id)
+        self._member_last_federation_poke[member] = self.clock.time_msec()
 
-        deferreds = []
-        for domain in domains:
-            if domain == self.server_name:
-                self._push_update_local(
-                    room_id=room_id,
-                    user_id=user_id,
-                    typing=typing
-                )
-            else:
-                deferreds.append(self.federation.send_edu(
+        now = self.clock.time_msec()
+        self.wheel_timer.insert(
+            now=now,
+            obj=member,
+            then=now + FEDERATION_PING_INTERVAL,
+        )
+
+        for domain in set(get_domain_from_id(u) for u in users):
+            if domain != self.server_name:
+                self.federation.send_edu(
                     destination=domain,
                     edu_type="m.typing",
                     content={
-                        "room_id": room_id,
-                        "user_id": user_id,
+                        "room_id": member.room_id,
+                        "user_id": member.user_id,
                         "typing": typing,
                     },
-                ))
-
-        yield defer.DeferredList(deferreds, consumeErrors=True)
+                    key=member,
+                )
 
     @defer.inlineCallbacks
     def _recv_edu(self, origin, content):
         room_id = content["room_id"]
         user_id = content["user_id"]
 
-        # Check that the string is a valid user id
-        UserID.from_string(user_id)
+        member = RoomMember(user_id=user_id, room_id=room_id)
 
-        domains = yield self.store.get_joined_hosts_for_room(room_id)
+        # Check that the string is a valid user id
+        user = UserID.from_string(user_id)
+
+        if user.domain != origin:
+            logger.info(
+                "Got typing update from %r with bad 'user_id': %r",
+                origin, user_id,
+            )
+            return
+
+        users = yield self.state.get_current_user_in_room(room_id)
+        domains = set(get_domain_from_id(u) for u in users)
 
         if self.server_name in domains:
+            logger.info("Got typing update from %s: %r", user_id, content)
+            now = self.clock.time_msec()
+            self._member_typing_until[member] = now + FEDERATION_TIMEOUT
+            self.wheel_timer.insert(
+                now=now,
+                obj=member,
+                then=now + FEDERATION_TIMEOUT,
+            )
             self._push_update_local(
-                room_id=room_id,
-                user_id=user_id,
+                member=member,
                 typing=content["typing"]
             )
 
-    def _push_update_local(self, room_id, user_id, typing):
-        room_set = self._room_typing.setdefault(room_id, set())
+    def _push_update_local(self, member, typing):
+        room_set = self._room_typing.setdefault(member.room_id, set())
         if typing:
-            room_set.add(user_id)
+            room_set.add(member.user_id)
         else:
-            room_set.discard(user_id)
+            room_set.discard(member.user_id)
 
         self._latest_room_serial += 1
-        self._room_serials[room_id] = self._latest_room_serial
+        self._room_serials[member.room_id] = self._latest_room_serial
 
-        with PreserveLoggingContext():
-            self.notifier.on_new_event(
-                "typing_key", self._latest_room_serial, rooms=[room_id]
-            )
+        self.notifier.on_new_event(
+            "typing_key", self._latest_room_serial, rooms=[member.room_id]
+        )
 
     def get_all_typing_updates(self, last_id, current_id):
         # TODO: Work out a way to do this without scanning the entire state.

synapse/http/client.py

@@ -25,7 +25,7 @@ from synapse.http.endpoint import SpiderEndpoint
 from canonicaljson import encode_canonical_json
 
 from twisted.internet import defer, reactor, ssl, protocol, task
-from twisted.internet.endpoints import SSL4ClientEndpoint, TCP4ClientEndpoint
+from twisted.internet.endpoints import HostnameEndpoint, wrapClientTLS
 from twisted.web.client import (
     BrowserLikeRedirectAgent, ContentDecoderAgent, GzipDecoder, Agent,
     readBody, PartialDownloadError,
@@ -386,26 +386,23 @@ class SpiderEndpointFactory(object):
 
     def endpointForURI(self, uri):
         logger.info("Getting endpoint for %s", uri.toBytes())
+
         if uri.scheme == "http":
-            return SpiderEndpoint(
-                reactor, uri.host, uri.port, self.blacklist, self.whitelist,
-                endpoint=TCP4ClientEndpoint,
-                endpoint_kw_args={
-                    'timeout': 15
-                },
-            )
+            endpoint_factory = HostnameEndpoint
         elif uri.scheme == "https":
-            tlsPolicy = self.policyForHTTPS.creatorForNetloc(uri.host, uri.port)
-            return SpiderEndpoint(
-                reactor, uri.host, uri.port, self.blacklist, self.whitelist,
-                endpoint=SSL4ClientEndpoint,
-                endpoint_kw_args={
-                    'sslContextFactory': tlsPolicy,
-                    'timeout': 15
-                },
-            )
+            tlsCreator = self.policyForHTTPS.creatorForNetloc(uri.host, uri.port)
+
+            def endpoint_factory(reactor, host, port, **kw):
+                return wrapClientTLS(
+                    tlsCreator,
+                    HostnameEndpoint(reactor, host, port, **kw))
         else:
             logger.warn("Can't get endpoint for unrecognised scheme %s", uri.scheme)
+            return None
+        return SpiderEndpoint(
+            reactor, uri.host, uri.port, self.blacklist, self.whitelist,
+            endpoint=endpoint_factory, endpoint_kw_args=dict(timeout=15),
+        )
 
 
 class SpiderHttpClient(SimpleHttpClient):

synapse/http/endpoint.py

@@ -13,8 +13,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from twisted.internet.endpoints import SSL4ClientEndpoint, TCP4ClientEndpoint
-from twisted.internet import defer
+from twisted.internet.endpoints import HostnameEndpoint, wrapClientTLS
+from twisted.internet import defer, reactor
 from twisted.internet.error import ConnectError
 from twisted.names import client, dns
 from twisted.names.error import DNSNameError, DomainError
@@ -58,21 +58,85 @@ def matrix_federation_endpoint(reactor, destination, ssl_context_factory=None,
         endpoint_kw_args.update(timeout=timeout)
 
     if ssl_context_factory is None:
-        transport_endpoint = TCP4ClientEndpoint
+        transport_endpoint = HostnameEndpoint
         default_port = 8008
     else:
-        transport_endpoint = SSL4ClientEndpoint
-        endpoint_kw_args.update(sslContextFactory=ssl_context_factory)
+        def transport_endpoint(reactor, host, port, timeout):
+            return wrapClientTLS(
+                ssl_context_factory,
+                HostnameEndpoint(reactor, host, port, timeout=timeout))
         default_port = 8448
 
     if port is None:
-        return SRVClientEndpoint(
+        return _WrappingEndpointFac(SRVClientEndpoint(
             reactor, "matrix", domain, protocol="tcp",
             default_port=default_port, endpoint=transport_endpoint,
             endpoint_kw_args=endpoint_kw_args
-        )
+        ))
     else:
-        return transport_endpoint(reactor, domain, port, **endpoint_kw_args)
+        return _WrappingEndpointFac(transport_endpoint(
+            reactor, domain, port, **endpoint_kw_args
+        ))
+
+
+class _WrappingEndpointFac(object):
+    def __init__(self, endpoint_fac):
+        self.endpoint_fac = endpoint_fac
+
+    @defer.inlineCallbacks
+    def connect(self, protocolFactory):
+        conn = yield self.endpoint_fac.connect(protocolFactory)
+        conn = _WrappedConnection(conn)
+        defer.returnValue(conn)
+
+
+class _WrappedConnection(object):
+    """Wraps a connection and calls abort on it if it hasn't seen any action
+    for 2.5-3 minutes.
+    """
+    __slots__ = ["conn", "last_request"]
+
+    def __init__(self, conn):
+        object.__setattr__(self, "conn", conn)
+        object.__setattr__(self, "last_request", time.time())
+
+    def __getattr__(self, name):
+        return getattr(self.conn, name)
+
+    def __setattr__(self, name, value):
+        setattr(self.conn, name, value)
+
+    def _time_things_out_maybe(self):
+        # We use a slightly shorter timeout here just in case the callLater is
+        # triggered early. Paranoia ftw.
+        # TODO: Cancel the previous callLater rather than comparing time.time()?
+        if time.time() - self.last_request >= 2.5 * 60:
+            self.abort()
+            # Abort the underlying TLS connection. The abort() method calls
+            # loseConnection() on the underlying TLS connection which tries to
+            # shutdown the connection cleanly. We call abortConnection()
+            # since that will promptly close the underlying TCP connection.
+            self.transport.abortConnection()
+
+    def request(self, request):
+        self.last_request = time.time()
+
+        # Time this connection out if we haven't send a request in the last
+        # N minutes
+        # TODO: Cancel the previous callLater?
+        reactor.callLater(3 * 60, self._time_things_out_maybe)
+
+        d = self.conn.request(request)
+
+        def update_request_time(res):
+            self.last_request = time.time()
+            # TODO: Cancel the previous callLater?
+            reactor.callLater(3 * 60, self._time_things_out_maybe)
+            return res
+
+        d.addCallback(update_request_time)
+
+        return d
 
 
 class SpiderEndpoint(object):
@@ -80,7 +144,7 @@ class SpiderEndpoint(object):
     Implements twisted.internet.interfaces.IStreamClientEndpoint.
     """
     def __init__(self, reactor, host, port, blacklist, whitelist,
-                 endpoint=TCP4ClientEndpoint, endpoint_kw_args={}):
+                 endpoint=HostnameEndpoint, endpoint_kw_args={}):
         self.reactor = reactor
         self.host = host
         self.port = port
@@ -118,7 +182,7 @@ class SRVClientEndpoint(object):
     """
 
     def __init__(self, reactor, service, domain, protocol="tcp",
-                 default_port=None, endpoint=TCP4ClientEndpoint,
+                 default_port=None, endpoint=HostnameEndpoint,
                  endpoint_kw_args={}):
         self.reactor = reactor
         self.service_name = "_%s._%s.%s" % (service, protocol, domain)

synapse/http/matrixfederationclient.py

@@ -33,6 +33,7 @@ from synapse.api.errors import (
 
 from signedjson.sign import sign_json
 
+import cgi
 import simplejson as json
 import logging
 import random
@@ -87,7 +88,8 @@ class MatrixFederationHttpClient(object):
         self.signing_key = hs.config.signing_key[0]
         self.server_name = hs.hostname
         pool = HTTPConnectionPool(reactor)
-        pool.maxPersistentPerHost = 10
+        pool.maxPersistentPerHost = 5
+        pool.cachedConnectionTimeout = 2 * 60
         self.agent = Agent.usingEndpointFactory(
             reactor, MatrixFederationEndpointFactory(hs), pool=pool
         )
@@ -155,9 +157,7 @@ class MatrixFederationHttpClient(object):
                             time_out=timeout / 1000. if timeout else 60,
                         )
 
-                    response = yield preserve_context_over_fn(
-                        send_request,
-                    )
+                    response = yield preserve_context_over_fn(send_request)
 
                     log_result = "%d %s" % (response.code, response.phrase,)
                     break
@@ -248,7 +248,7 @@ class MatrixFederationHttpClient(object):
 
     @defer.inlineCallbacks
     def put_json(self, destination, path, data={}, json_data_callback=None,
-                 long_retries=False):
+                 long_retries=False, timeout=None):
         """ Sends the specifed json data using PUT
 
         Args:
@@ -261,6 +261,8 @@ class MatrixFederationHttpClient(object):
                 use as the request body.
             long_retries (bool): A boolean that indicates whether we should
                 retry for a short or long time.
+            timeout(int): How long to try (in ms) the destination for before
+                giving up. None indicates no timeout.
 
         Returns:
             Deferred: Succeeds when we get a 2xx HTTP response. The result
@@ -287,22 +289,19 @@ class MatrixFederationHttpClient(object):
             body_callback=body_callback,
             headers_dict={"Content-Type": ["application/json"]},
             long_retries=long_retries,
+            timeout=timeout,
         )
 
         if 200 <= response.code < 300:
             # We need to update the transactions table to say it was sent?
-            c_type = response.headers.getRawHeaders("Content-Type")
-
-            if "application/json" not in c_type:
-                raise RuntimeError(
-                    "Content-Type not application/json"
-                )
+            check_content_type_is_json(response.headers)
 
         body = yield preserve_context_over_fn(readBody, response)
         defer.returnValue(json.loads(body))
 
     @defer.inlineCallbacks
-    def post_json(self, destination, path, data={}, long_retries=True):
+    def post_json(self, destination, path, data={}, long_retries=False,
+                  timeout=None):
         """ Sends the specifed json data using POST
 
         Args:
@@ -313,6 +312,8 @@ class MatrixFederationHttpClient(object):
                 the request body. This will be encoded as JSON.
             long_retries (bool): A boolean that indicates whether we should
                 retry for a short or long time.
+            timeout(int): How long to try (in ms) the destination for before
+                giving up. None indicates no timeout.
 
         Returns:
             Deferred: Succeeds when we get a 2xx HTTP response. The result
@@ -332,17 +333,13 @@ class MatrixFederationHttpClient(object):
             path.encode("ascii"),
             body_callback=body_callback,
             headers_dict={"Content-Type": ["application/json"]},
-            long_retries=True,
+            long_retries=long_retries,
+            timeout=timeout,
         )
 
         if 200 <= response.code < 300:
             # We need to update the transactions table to say it was sent?
-            c_type = response.headers.getRawHeaders("Content-Type")
-
-            if "application/json" not in c_type:
-                raise RuntimeError(
-                    "Content-Type not application/json"
-                )
+            check_content_type_is_json(response.headers)
 
         body = yield preserve_context_over_fn(readBody, response)
 
@@ -395,12 +392,7 @@ class MatrixFederationHttpClient(object):
 
         if 200 <= response.code < 300:
             # We need to update the transactions table to say it was sent?
-            c_type = response.headers.getRawHeaders("Content-Type")
-
-            if "application/json" not in c_type:
-                raise RuntimeError(
-                    "Content-Type not application/json"
-                )
+            check_content_type_is_json(response.headers)
 
         body = yield preserve_context_over_fn(readBody, response)
 
@@ -520,3 +512,29 @@ def _flatten_response_never_received(e):
         )
     else:
         return "%s: %s" % (type(e).__name__, e.message,)
+
+
+def check_content_type_is_json(headers):
+    """
+    Check that a set of HTTP headers have a Content-Type header, and that it
+    is application/json.
+
+    Args:
+        headers (twisted.web.http_headers.Headers): headers to check
+
+    Raises:
+        RuntimeError if the
+
+    """
+    c_type = headers.getRawHeaders("Content-Type")
+    if c_type is None:
+        raise RuntimeError(
+            "No Content-Type header"
+        )
+
+    c_type = c_type[0]  # only the first header
+    val, options = cgi.parse_header(c_type)
+    if val != "application/json":
+        raise RuntimeError(
+            "Content-Type not application/json: was '%s'" % c_type
+        )

synapse/http/server.py

@@ -19,6 +19,7 @@ from synapse.api.errors import (
 )
 from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
 from synapse.util.caches import intern_dict
+from synapse.util.metrics import Measure
 import synapse.metrics
 import synapse.events
 
@@ -74,12 +75,12 @@ response_db_txn_duration = metrics.register_distribution(
 _next_request_id = 0
 
 
-def request_handler(report_metrics=True):
+def request_handler(include_metrics=False):
     """Decorator for ``wrap_request_handler``"""
-    return lambda request_handler: wrap_request_handler(request_handler, report_metrics)
+    return lambda request_handler: wrap_request_handler(request_handler, include_metrics)
 
 
-def wrap_request_handler(request_handler, report_metrics):
+def wrap_request_handler(request_handler, include_metrics=False):
     """Wraps a method that acts as a request handler with the necessary logging
     and exception handling.
 
@@ -103,54 +104,56 @@ def wrap_request_handler(request_handler, report_metrics):
         _next_request_id += 1
 
         with LoggingContext(request_id) as request_context:
-            if report_metrics:
+            with Measure(self.clock, "wrapped_request_handler"):
                 request_metrics = RequestMetrics()
-                request_metrics.start(self.clock)
+                request_metrics.start(self.clock, name=self.__class__.__name__)
 
-            request_context.request = request_id
-            with request.processing():
-                try:
-                    with PreserveLoggingContext(request_context):
-                        yield request_handler(self, request)
-                except CodeMessageException as e:
-                    code = e.code
-                    if isinstance(e, SynapseError):
-                        logger.info(
-                            "%s SynapseError: %s - %s", request, code, e.msg
-                        )
-                    else:
-                        logger.exception(e)
-                    outgoing_responses_counter.inc(request.method, str(code))
-                    respond_with_json(
-                        request, code, cs_exception(e), send_cors=True,
-                        pretty_print=_request_user_agent_is_curl(request),
-                        version_string=self.version_string,
-                    )
-                except:
-                    logger.exception(
-                        "Failed handle request %s.%s on %r: %r",
-                        request_handler.__module__,
-                        request_handler.__name__,
-                        self,
-                        request
-                    )
-                    respond_with_json(
-                        request,
-                        500,
-                        {
-                            "error": "Internal server error",
-                            "errcode": Codes.UNKNOWN,
-                        },
-                        send_cors=True
-                    )
-                finally:
+                request_context.request = request_id
+                with request.processing():
                     try:
-                        if report_metrics:
-                            request_metrics.stop(
-                                self.clock, request, self.__class__.__name__
+                        with PreserveLoggingContext(request_context):
+                            if include_metrics:
+                                yield request_handler(self, request, request_metrics)
+                            else:
+                                yield request_handler(self, request)
+                    except CodeMessageException as e:
+                        code = e.code
+                        if isinstance(e, SynapseError):
+                            logger.info(
+                                "%s SynapseError: %s - %s", request, code, e.msg
                             )
+                        else:
+                            logger.exception(e)
+                        outgoing_responses_counter.inc(request.method, str(code))
+                        respond_with_json(
+                            request, code, cs_exception(e), send_cors=True,
+                            pretty_print=_request_user_agent_is_curl(request),
+                            version_string=self.version_string,
+                        )
                     except:
-                        pass
+                        logger.exception(
+                            "Failed handle request %s.%s on %r: %r",
+                            request_handler.__module__,
+                            request_handler.__name__,
+                            self,
+                            request
+                        )
+                        respond_with_json(
+                            request,
+                            500,
+                            {
+                                "error": "Internal server error",
+                                "errcode": Codes.UNKNOWN,
+                            },
+                            send_cors=True
+                        )
+                    finally:
+                        try:
+                            request_metrics.stop(
+                                self.clock, request
+                            )
+                        except Exception as e:
+                            logger.warn("Failed to stop metrics: %r", e)
     return wrapped_request_handler
 
 
@@ -205,6 +208,7 @@ class JsonResource(HttpServer, resource.Resource):
 
     def register_paths(self, method, path_patterns, callback):
         for path_pattern in path_patterns:
+            logger.debug("Registering for %s %s", method, path_pattern.pattern)
             self.path_regexs.setdefault(method, []).append(
                 self._PathEntry(path_pattern, callback)
             )
@@ -219,9 +223,9 @@ class JsonResource(HttpServer, resource.Resource):
     # It does its own metric reporting because _async_render dispatches to
     # a callback and it's the class name of that callback we want to report
     # against rather than the JsonResource itself.
-    @request_handler(report_metrics=False)
+    @request_handler(include_metrics=True)
     @defer.inlineCallbacks
-    def _async_render(self, request):
+    def _async_render(self, request, request_metrics):
         """ This gets called from render() every time someone sends us a request.
             This checks if anyone has registered a callback for that method and
             path.
@@ -230,9 +234,6 @@ class JsonResource(HttpServer, resource.Resource):
             self._send_response(request, 200, {})
             return
 
-        request_metrics = RequestMetrics()
-        request_metrics.start(self.clock)
-
         # Loop through all the registered callbacks to check if the method
         # and path regex match
         for path_entry in self.path_regexs.get(request.method, []):
@@ -246,12 +247,6 @@ class JsonResource(HttpServer, resource.Resource):
 
             callback = path_entry.callback
 
-            servlet_instance = getattr(callback, "__self__", None)
-            if servlet_instance is not None:
-                servlet_classname = servlet_instance.__class__.__name__
-            else:
-                servlet_classname = "%r" % callback
-
             kwargs = intern_dict({
                 name: urllib.unquote(value).decode("UTF-8") if value else value
                 for name, value in m.groupdict().items()
@@ -262,10 +257,13 @@ class JsonResource(HttpServer, resource.Resource):
                 code, response = callback_return
                 self._send_response(request, code, response)
 
-            try:
-                request_metrics.stop(self.clock, request, servlet_classname)
-            except:
-                pass
+            servlet_instance = getattr(callback, "__self__", None)
+            if servlet_instance is not None:
+                servlet_classname = servlet_instance.__class__.__name__
+            else:
+                servlet_classname = "%r" % callback
+
+            request_metrics.name = servlet_classname
 
             return
 
@@ -297,11 +295,12 @@ class JsonResource(HttpServer, resource.Resource):
 
 
 class RequestMetrics(object):
-    def start(self, clock):
+    def start(self, clock, name):
         self.start = clock.time_msec()
         self.start_context = LoggingContext.current_context()
+        self.name = name
 
-    def stop(self, clock, request, servlet_classname):
+    def stop(self, clock, request):
         context = LoggingContext.current_context()
 
         tag = ""
@@ -315,26 +314,26 @@ class RequestMetrics(object):
                 )
                 return
 
-        incoming_requests_counter.inc(request.method, servlet_classname, tag)
+        incoming_requests_counter.inc(request.method, self.name, tag)
 
         response_timer.inc_by(
             clock.time_msec() - self.start, request.method,
-            servlet_classname, tag
+            self.name, tag
         )
 
         ru_utime, ru_stime = context.get_resource_usage()
 
         response_ru_utime.inc_by(
-            ru_utime, request.method, servlet_classname, tag
+            ru_utime, request.method, self.name, tag
         )
         response_ru_stime.inc_by(
-            ru_stime, request.method, servlet_classname, tag
+            ru_stime, request.method, self.name, tag
         )
         response_db_txn_count.inc_by(
-            context.db_txn_count, request.method, servlet_classname, tag
+            context.db_txn_count, request.method, self.name, tag
         )
         response_db_txn_duration.inc_by(
-            context.db_txn_duration, request.method, servlet_classname, tag
+            context.db_txn_duration, request.method, self.name, tag
         )
 
 
@@ -393,17 +392,30 @@ def respond_with_json_bytes(request, code, json_bytes, send_cors=False,
     request.setHeader(b"Content-Length", b"%d" % (len(json_bytes),))
 
     if send_cors:
-        request.setHeader("Access-Control-Allow-Origin", "*")
-        request.setHeader("Access-Control-Allow-Methods",
-                          "GET, POST, PUT, DELETE, OPTIONS")
-        request.setHeader("Access-Control-Allow-Headers",
-                          "Origin, X-Requested-With, Content-Type, Accept")
+        set_cors_headers(request)
 
     request.write(json_bytes)
     finish_request(request)
     return NOT_DONE_YET
 
 
+def set_cors_headers(request):
+    """Set the CORs headers so that javascript running in a web browsers can
+    use this API
+
+    Args:
+        request (twisted.web.http.Request): The http request to add CORs to.
+    """
+    request.setHeader("Access-Control-Allow-Origin", "*")
+    request.setHeader(
+        "Access-Control-Allow-Methods", "GET, POST, PUT, DELETE, OPTIONS"
+    )
+    request.setHeader(
+        "Access-Control-Allow-Headers",
+        "Origin, X-Requested-With, Content-Type, Accept"
+    )
+
+
 def finish_request(request):
     """ Finish writing the response to the request.
 

synapse/http/servlet.py

@@ -41,9 +41,13 @@ def parse_integer(request, name, default=None, required=False):
         SynapseError: if the parameter is absent and required, or if the
             parameter is present and not an integer.
     """
-    if name in request.args:
+    return parse_integer_from_args(request.args, name, default, required)
+
+
+def parse_integer_from_args(args, name, default=None, required=False):
+    if name in args:
         try:
-            return int(request.args[name][0])
+            return int(args[name][0])
         except:
             message = "Query parameter %r must be an integer" % (name,)
             raise SynapseError(400, message)
@@ -74,12 +78,16 @@ def parse_boolean(request, name, default=None, required=False):
             parameter is present and not one of "true" or "false".
     """
 
-    if name in request.args:
+    return parse_boolean_from_args(request.args, name, default, required)
+
+
+def parse_boolean_from_args(args, name, default=None, required=False):
+    if name in args:
         try:
             return {
                 "true": True,
                 "false": False,
-            }[request.args[name][0]]
+            }[args[name][0]]
         except:
             message = (
                 "Boolean query parameter %r must be one of"
@@ -116,9 +124,15 @@ def parse_string(request, name, default=None, required=False,
             parameter is present, must be one of a list of allowed values and
             is not one of those allowed values.
     """
+    return parse_string_from_args(
+        request.args, name, default, required, allowed_values, param_type,
+    )
 
-    if name in request.args:
-        value = request.args[name][0]
+
+def parse_string_from_args(args, name, default=None, required=False,
+                           allowed_values=None, param_type="string"):
+    if name in args:
+        value = args[name][0]
         if allowed_values is not None and value not in allowed_values:
             message = "Query parameter %r must be one of [%s]" % (
                 name, ", ".join(repr(v) for v in allowed_values)

synapse/metrics/__init__.py

@@ -13,28 +13,25 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Because otherwise 'resource' collides with synapse.metrics.resource
-from __future__ import absolute_import
-
 import logging
-from resource import getrusage, RUSAGE_SELF
 import functools
-import os
-import stat
 import time
 import gc
 
 from twisted.internet import reactor
 
 from .metric import (
-    CounterMetric, CallbackMetric, DistributionMetric, CacheMetric
+    CounterMetric, CallbackMetric, DistributionMetric, CacheMetric,
+    MemoryUsageMetric,
 )
+from .process_collector import register_process_collector
 
 
 logger = logging.getLogger(__name__)
 
 
 all_metrics = []
+all_collectors = []
 
 
 class Metrics(object):
@@ -45,6 +42,12 @@ class Metrics(object):
     def __init__(self, name):
         self.name_prefix = name
 
+    def make_subspace(self, name):
+        return Metrics("%s_%s" % (self.name_prefix, name))
+
+    def register_collector(self, func):
+        all_collectors.append(func)
+
     def _register(self, metric_class, name, *args, **kwargs):
         full_name = "%s_%s" % (self.name_prefix, name)
 
@@ -66,6 +69,21 @@ class Metrics(object):
         return self._register(CacheMetric, *args, **kwargs)
 
 
+def register_memory_metrics(hs):
+    try:
+        import psutil
+        process = psutil.Process()
+        process.memory_info().rss
+    except (ImportError, AttributeError):
+        logger.warn(
+            "psutil is not installed or incorrect version."
+            " Disabling memory metrics."
+        )
+        return
+    metric = MemoryUsageMetric(hs, psutil)
+    all_metrics.append(metric)
+
+
 def get_metrics_for(pkg_name):
     """ Returns a Metrics instance for conveniently creating metrics
     namespaced with the given name prefix. """
@@ -78,8 +96,8 @@ def get_metrics_for(pkg_name):
 def render_all():
     strs = []
 
-    # TODO(paul): Internal hack
-    update_resource_metrics()
+    for collector in all_collectors:
+        collector()
 
     for metric in all_metrics:
         try:
@@ -93,73 +111,21 @@ def render_all():
     return "\n".join(strs)
 
 
-# Now register some standard process-wide state metrics, to give indications of
-# process resource usage
-
-rusage = None
+register_process_collector(get_metrics_for("process"))
 
 
-def update_resource_metrics():
-    global rusage
-    rusage = getrusage(RUSAGE_SELF)
+python_metrics = get_metrics_for("python")
 
-resource_metrics = get_metrics_for("process.resource")
-
-# msecs
-resource_metrics.register_callback("utime", lambda: rusage.ru_utime * 1000)
-resource_metrics.register_callback("stime", lambda: rusage.ru_stime * 1000)
-
-# kilobytes
-resource_metrics.register_callback("maxrss", lambda: rusage.ru_maxrss * 1024)
-
-TYPES = {
-    stat.S_IFSOCK: "SOCK",
-    stat.S_IFLNK: "LNK",
-    stat.S_IFREG: "REG",
-    stat.S_IFBLK: "BLK",
-    stat.S_IFDIR: "DIR",
-    stat.S_IFCHR: "CHR",
-    stat.S_IFIFO: "FIFO",
-}
-
-
-def _process_fds():
-    counts = {(k,): 0 for k in TYPES.values()}
-    counts[("other",)] = 0
-
-    # Not every OS will have a /proc/self/fd directory
-    if not os.path.exists("/proc/self/fd"):
-        return counts
-
-    for fd in os.listdir("/proc/self/fd"):
-        try:
-            s = os.stat("/proc/self/fd/%s" % (fd))
-            fmt = stat.S_IFMT(s.st_mode)
-            if fmt in TYPES:
-                t = TYPES[fmt]
-            else:
-                t = "other"
-
-            counts[(t,)] += 1
-        except OSError:
-            # the dirh itself used by listdir() is usually missing by now
-            pass
-
-    return counts
-
-get_metrics_for("process").register_callback("fds", _process_fds, labels=["type"])
-
-reactor_metrics = get_metrics_for("reactor")
-tick_time = reactor_metrics.register_distribution("tick_time")
-pending_calls_metric = reactor_metrics.register_distribution("pending_calls")
-
-gc_time = reactor_metrics.register_distribution("gc_time", labels=["gen"])
-gc_unreachable = reactor_metrics.register_counter("gc_unreachable", labels=["gen"])
-
-reactor_metrics.register_callback(
+gc_time = python_metrics.register_distribution("gc_time", labels=["gen"])
+gc_unreachable = python_metrics.register_counter("gc_unreachable_total", labels=["gen"])
+python_metrics.register_callback(
     "gc_counts", lambda: {(i,): v for i, v in enumerate(gc.get_count())}, labels=["gen"]
 )
 
+reactor_metrics = get_metrics_for("python.twisted.reactor")
+tick_time = reactor_metrics.register_distribution("tick_time")
+pending_calls_metric = reactor_metrics.register_distribution("pending_calls")
+
 
 def runUntilCurrentTimer(func):
 

synapse/metrics/metric.py

@@ -98,9 +98,9 @@ class CallbackMetric(BaseMetric):
         value = self.callback()
 
         if self.is_scalar():
-            return ["%s %d" % (self.name, value)]
+            return ["%s %.12g" % (self.name, value)]
 
-        return ["%s%s %d" % (self.name, self._render_key(k), value[k])
+        return ["%s%s %.12g" % (self.name, self._render_key(k), value[k])
                 for k in sorted(value.keys())]
 
 
@@ -153,3 +153,43 @@ class CacheMetric(object):
             """%s:total{name="%s"} %d""" % (self.name, self.cache_name, total),
             """%s:size{name="%s"} %d""" % (self.name, self.cache_name, size),
         ]
+
+
+class MemoryUsageMetric(object):
+    """Keeps track of the current memory usage, using psutil.
+
+    The class will keep the current min/max/sum/counts of rss over the last
+    WINDOW_SIZE_SEC, by polling UPDATE_HZ times per second
+    """
+
+    UPDATE_HZ = 2  # number of times to get memory per second
+    WINDOW_SIZE_SEC = 30  # the size of the window in seconds
+
+    def __init__(self, hs, psutil):
+        clock = hs.get_clock()
+        self.memory_snapshots = []
+
+        self.process = psutil.Process()
+
+        clock.looping_call(self._update_curr_values, 1000 / self.UPDATE_HZ)
+
+    def _update_curr_values(self):
+        max_size = self.UPDATE_HZ * self.WINDOW_SIZE_SEC
+        self.memory_snapshots.append(self.process.memory_info().rss)
+        self.memory_snapshots[:] = self.memory_snapshots[-max_size:]
+
+    def render(self):
+        if not self.memory_snapshots:
+            return []
+
+        max_rss = max(self.memory_snapshots)
+        min_rss = min(self.memory_snapshots)
+        sum_rss = sum(self.memory_snapshots)
+        len_rss = len(self.memory_snapshots)
+
+        return [
+            "process_psutil_rss:max %d" % max_rss,
+            "process_psutil_rss:min %d" % min_rss,
+            "process_psutil_rss:total %d" % sum_rss,
+            "process_psutil_rss:count %d" % len_rss,
+        ]