0
0
Fork 0
mirror of https://github.com/matrix-construct/construct synced 2024-12-27 16:04:15 +01:00
construct/matrix/room_bootstrap.cc

832 lines
17 KiB
C++
Raw Normal View History

// Matrix Construct
//
// Copyright (C) Matrix Construct Developers, Authors & Contributors
// Copyright (C) 2016-2019 Jason Volk <jason@zemos.net>
//
// Permission to use, copy, modify, and/or distribute this software for any
// purpose with or without fee is hereby granted, provided that the above
// copyright notice and this permission notice is present in all copies. The
// full license for this software is available in the LICENSE file.
namespace ircd::m::roomstrap
{
struct pkg;
using send_join_response = std::tuple<json::object, unique_buffer<mutable_buffer>>;
static event::id::buf make_join(const string_view &host, const room::id &, const user::id &, const mutable_buffer &);
static send_join_response send_join(const string_view &host, const room::id &, const event::id &, const json::object &event);
static void broadcast_join(const room &, const event &, const string_view &exclude);
static void fetch_keys(const json::array &events);
static void eval_auth_chain(const json::array &auth_chain, vm::opts);
static void eval_state(const json::array &state, vm::opts);
static void backfill(const string_view &host, const room::id &, const event::id &, vm::opts);
static void worker(pkg);
extern conf::item<seconds> make_join_timeout;
extern conf::item<seconds> send_join_timeout;
extern conf::item<seconds> backfill_timeout;
extern conf::item<size_t> backfill_limit;
extern log::log log;
}
struct ircd::m::roomstrap::pkg
{
std::string event;
std::string event_id;
std::string host;
std::string room_version;
};
decltype(ircd::m::roomstrap::log)
ircd::m::roomstrap::log
{
"m.room.bootstrap"
};
decltype(ircd::m::roomstrap::backfill_limit)
ircd::m::roomstrap::backfill_limit
{
{ "name", "ircd.client.rooms.join.backfill.limit" },
{ "default", 64L },
{ "description",
R"(
The number of events to request on initial backfill. Specapse may limit
this to 50, but it also may not. Either way, a good choice is enough to
fill a client's timeline quickly with a little headroom.
)"}
};
decltype(ircd::m::roomstrap::backfill_timeout)
ircd::m::roomstrap::backfill_timeout
{
{ "name", "ircd.client.rooms.join.backfill.timeout" },
{ "default", 15L },
};
decltype(ircd::m::roomstrap::send_join_timeout)
ircd::m::roomstrap::send_join_timeout
{
{ "name", "ircd.client.rooms.join.send_join.timeout" },
{ "default", 90L /* spinappse */ },
};
decltype(ircd::m::roomstrap::make_join_timeout)
ircd::m::roomstrap::make_join_timeout
{
{ "name", "ircd.client.rooms.join.make_join.timeout" },
{ "default", 15L },
};
//
// m::room::bootstrap
//
ircd::m::room::bootstrap::bootstrap(m::event::id::buf &event_id_buf,
const m::room::id &room_id,
const m::user::id &user_id,
const vector_view<const string_view> &hosts)
{
const auto member_event_idx
{
m::room(room_id).get(std::nothrow, "m.room.member", user_id)
};
const bool existing_join
{
m::membership(member_event_idx, "join")
};
char room_version_buf[64];
string_view room_version
{
m::version(room_version_buf, room_id, std::nothrow)
};
//TODO: try more hosts?
const auto &host
{
hosts.empty()?
room_id.host():
hosts[0]
};
log::info
{
log, "Starting in %s for %s to '%s' joined:%b ver:%s",
string_view{room_id},
string_view{user_id},
host,
existing_join,
room_version,
};
if(existing_join)
event_id_buf = m::event_id(std::nothrow, member_event_idx);
if(!event_id_buf)
event_id_buf = m::roomstrap::make_join(host, room_id, user_id, room_version_buf);
if(!room_version)
m::roomstrap::make_join(host, room_id, user_id, room_version_buf);
assert(event_id_buf);
// asynchronous; returns quickly
room::bootstrap
{
event_id_buf, host, room_version
};
}
ircd::m::room::bootstrap::bootstrap(const m::event::id &event_id,
const string_view &host,
const string_view &room_version)
try
{
static const context::flags flags
{
context::POST | context::DETACH
};
static const auto stack_sz
{
128_KiB
};
const m::event::fetch event
{
event_id
};
assert(event.valid);
assert(event.source);
m::roomstrap::pkg pkg
{
std::string(event.source),
event.event_id,
host,
room_version,
};
context
{
"bootstrap",
stack_sz,
flags,
std::bind(&ircd::m::roomstrap::worker, std::move(pkg))
};
}
catch(const std::exception &e)
{
log::error
{
log, "Failed to bootstrap for %s to %s :%s",
string_view{event_id},
host,
e.what(),
};
}
ircd::m::room::bootstrap::bootstrap(const m::event &event,
const string_view &host,
const string_view &room_version)
try
{
const m::event::id &event_id
{
event.event_id
};
const m::room::id &room_id
{
at<"room_id"_>(event)
};
const m::user::id &user_id
{
at<"sender"_>(event)
};
const m::room room
{
room_id, event_id
};
log::info
{
log, "Sending in %s (version %s) for %s at %s to '%s'",
string_view{room_id},
room_version,
string_view{user_id},
string_view{event_id},
host
};
assert(event.source);
const auto &[response, buf]
{
m::roomstrap::send_join(host, room_id, event_id, event.source)
};
const json::array &auth_chain
{
response["auth_chain"]
};
const json::array &state
{
response["state"]
};
log::info
{
log, "Joined to %s for %s at %s to '%s' state:%zu auth_chain:%zu",
string_view{room_id},
string_view{user_id},
string_view{event_id},
host,
state.size(),
auth_chain.size(),
};
m::vm::opts vmopts;
vmopts.infolog_accept = false;
vmopts.warnlog &= ~vm::fault::EXISTS;
vmopts.nothrows = -1;
vmopts.room_version = room_version;
vmopts.phase.reset(m::vm::phase::FETCH_PREV);
vmopts.phase.reset(m::vm::phase::FETCH_STATE);
m::roomstrap::fetch_keys(auth_chain);
m::roomstrap::eval_auth_chain(auth_chain, vmopts);
m::roomstrap::fetch_keys(state);
m::roomstrap::eval_state(state, vmopts);
m::roomstrap::backfill(host, room_id, event_id, vmopts);
// After we just received and processed all of this state with only a
// recent backfill our system doesn't know if state events which are
// unreferenced are simply referenced by events we just don't have. They
// will all be added to the room::head and each future event we transmit
// to the room will drain that list little by little. But the cost of all
// these references is too high. We take the easy route here and simply
// clear the head of every event except our own join event.
const size_t num_reset
{
m::room::head::reset(room)
};
// At this point we have only transmitted the join event to one bootstrap
// server. Now that we have processed the state we know of more servers.
// They don't know about our join event though, so we conduct a synchronous
// broadcast to the room now manually.
m::roomstrap::broadcast_join(room, event, host);
log::notice
{
log, "Joined to %s for %s at %s reset:%zu complete",
string_view{room_id},
string_view{user_id},
string_view{event_id},
num_reset,
};
}
catch(const std::exception &e)
{
log::error
{
log, "Join %s with %s to %s :%s",
json::get<"room_id"_>(event),
string_view{event.event_id},
string(host),
e.what()
};
}
//
// m::roomstrap
//
void
ircd::m::roomstrap::worker(pkg pkg)
try
{
assert(!empty(pkg.event));
assert(!empty(pkg.event_id));
const m::event event
{
pkg.event, pkg.event_id
};
assert(!empty(pkg.host));
room::bootstrap
{
event, pkg.host, pkg.room_version
};
}
catch(const http::error &e)
{
log::error
{
log, "(worker) Failed to bootstrap for %s to %s :%s :%s",
pkg.event_id,
pkg.host,
e.what(),
e.content,
};
}
catch(const std::exception &e)
{
log::error
{
log, "(worker) Failed to bootstrap for %s to %s :%s",
pkg.event_id,
pkg.host,
e.what(),
};
}
void
ircd::m::roomstrap::broadcast_join(const m::room &room,
const m::event &event,
const string_view &exclude) //TODO: XX
{
const m::room::origins origins
{
room
};
log::info
{
log, "Broadcasting %s to %s estimated servers:%zu",
string_view{event.event_id},
string_view{room.room_id},
origins.count(),
};
const json::value pdu
{
event.source
};
const vector_view<const json::value> pdus
{
&pdu, 1
};
const auto txn
{
m::txn::create(pdus)
};
char idbuf[128];
const auto txnid
{
m::txn::create_id(idbuf, txn)
};
m::feds::opts opts;
opts.op = feds::op::send;
opts.exclude_myself = true;
opts.room_id = room;
opts.arg[0] = txnid;
opts.arg[1] = txn;
size_t good(0), fail(0);
m::feds::execute(opts, [&event, &good, &fail]
(const auto &result)
{
if(result.eptr)
log::derror
{
log, "Failed to broadcast %s to %s :%s",
string_view{event.event_id},
result.origin,
what(result.eptr),
};
fail += bool(result.eptr);
good += !result.eptr;
return true;
});
log::info
{
log, "Broadcast %s to %s good:%zu fail:%zu servers:%zu online:%zu error:%zu",
string_view{event.event_id},
string_view{room.room_id},
good,
fail,
origins.count(),
origins.count_online(),
origins.count_error(),
};
}
void
ircd::m::roomstrap::backfill(const string_view &host,
const m::room::id &room_id,
const m::event::id &event_id,
vm::opts vmopts)
try
{
log::info
{
log, "Requesting recent events for %s from %s at %s",
string_view{room_id},
host,
string_view{event_id},
};
const unique_buffer<mutable_buffer> buf
{
16_KiB // headers in and out
};
m::fed::backfill::opts opts;
opts.remote = host;
opts.event_id = event_id;
opts.limit = size_t(backfill_limit);
m::fed::backfill request
{
room_id, buf, std::move(opts)
};
request.wait(seconds(backfill_timeout));
const auto code
{
request.get()
};
const json::object &response
{
request.in.content
};
const json::array &pdus
{
response["pdus"]
};
log::info
{
log, "Processing backfill for %s from %s at %s events:%zu",
string_view{room_id},
host,
string_view{event_id},
pdus.size(),
};
m::vm::eval
{
pdus, vmopts
};
}
catch(const std::exception &e)
{
log::error
{
log, "%s backfill @ %s from %s :%s",
string_view{room_id},
string_view{event_id},
string(host),
e.what(),
};
// Backfill errors are not propagated further, thus they won't stop the
// bootstrap process. The timeline won't have any readable messages, but
// we can remedy that later.
//throw;
}
void
ircd::m::roomstrap::eval_state(const json::array &state,
vm::opts vmopts)
try
{
log::info
{
log, "Evaluating %zu state events...",
state.size(),
};
m::vm::eval
{
state, vmopts
};
}
catch(const std::exception &e)
{
log::error
{
log, "eval state :%s", e.what(),
};
// State errors are not propagated further, thus they won't stop the
// bootstrap process. The room state will be incomplete, but we can
// remedy that later.
//throw;
}
void
ircd::m::roomstrap::eval_auth_chain(const json::array &auth_chain,
vm::opts vmopts)
try
{
log::info
{
log, "Evaluating %zu authentication events...",
auth_chain.size(),
};
vmopts.nothrows = vm::fault::EXISTS;
vmopts.fetch = false;
m::vm::eval
{
auth_chain, vmopts
};
}
catch(const std::exception &e)
{
log::error
{
log, "eval auth_chain :%s", e.what(),
};
// This needs to rethrow because any failure coming out of vm::eval to
// process the auth_chain is a showstopper.
throw;
}
void
ircd::m::roomstrap::fetch_keys(const json::array &events)
try
{
std::vector<m::fed::key::server_key> queries;
queries.reserve(events.size());
for(const json::object event : events)
for(const auto &[server_name, signatures] : json::object(event["signatures"]))
for(const auto &[key_id, signature] : json::object(signatures))
queries.emplace_back(unquote(event.at("origin")), key_id);
std::sort(begin(queries), end(queries));
queries.erase(std::unique(begin(queries), end(queries)), end(queries));
log::info
{
log, "Fetching %zu keys for %zu events...",
queries.size(),
events.size(),
};
const size_t fetched
{
m::keys::fetch(queries)
};
log::info
{
log, "Fetched %zu of %zu keys for %zu events",
fetched,
queries.size(),
events.size(),
};
}
catch(const std::exception &e)
{
log::error
{
log, "Error when fetching keys for %zu events :%s",
events.size(),
};
// All errors for the parallel key fetch are logged and then suppressed
// here. This operation is an optimization; if there's an unexpected
// failure here keys will just be fetched in the eval loop and bootstrap
// will just be really slow.
//throw;
}
ircd::m::roomstrap::send_join_response
ircd::m::roomstrap::send_join(const string_view &host,
const m::room::id &room_id,
const m::event::id &event_id,
const json::object &event)
try
{
const unique_buffer<mutable_buffer> buf
{
16_KiB // headers in and out
};
m::fed::send_join::opts opts{host};
m::fed::send_join send_join
{
room_id, event_id, event, buf, std::move(opts)
};
send_join.wait(seconds(send_join_timeout));
const auto send_join_code
{
send_join.get()
};
const json::array send_join_response
{
send_join
};
const uint more_send_join_code
{
send_join_response.at<uint>(0)
};
const json::object &send_join_response_data
{
send_join_response[1]
};
assert(!!send_join.in.dynamic);
return
{
send_join_response_data,
std::move(send_join.in.dynamic)
};
}
catch(const std::exception &e)
{
log::error
{
log, "Bootstrap %s @ %s send_join to %s :%s",
string_view{room_id},
string_view{event_id},
string(host),
e.what(),
};
// This needs to rethrow because if there's any error in the send_join
// request we won't have the response data for the rest of the bootstrap
// process.
throw;
}
ircd::m::event::id::buf
ircd::m::roomstrap::make_join(const string_view &host,
const m::room::id &room_id,
const m::user::id &user_id,
const mutable_buffer &room_version_buf)
try
{
const unique_buffer<mutable_buffer> buf
{
16_KiB // headers in and out
};
m::fed::make_join::opts opts{host};
m::fed::make_join request
{
room_id, user_id, buf, std::move(opts)
};
request.wait(seconds(make_join_timeout));
const auto code
{
request.get()
};
const json::object &response
{
request.in.content
};
const json::string &room_version
{
response.get("room_version", "1"_sv)
};
const json::object &proto
{
response.at("event")
};
const json::array &auth_events
{
proto.get("auth_events")
};
const json::array &prev_events
{
proto.get("prev_events")
};
json::iov event;
json::iov content;
const json::iov::push push[]
{
{ event, { "type", "m.room.member" }},
{ event, { "sender", user_id }},
{ event, { "state_key", user_id }},
{ content, { "membership", "join" }},
{ event, { "prev_events", prev_events }},
{ event, { "auth_events", auth_events }},
{ event, { "prev_state", "[]" }},
{ event, { "depth", proto.get<long>("depth") }},
{ event, { "room_id", room_id }},
};
const m::user user{user_id};
const m::user::profile profile{user};
char displayname_buf[256];
const string_view displayname
{
profile.get(displayname_buf, "displayname")
};
char avatar_url_buf[256];
const string_view avatar_url
{
profile.get(avatar_url_buf, "avatar_url")
};
const json::iov::add _displayname
{
content, !empty(displayname),
{
"displayname", [&displayname]() -> json::value
{
return displayname;
}
}
};
const json::iov::add _avatar_url
{
content, !empty(avatar_url),
{
"avatar_url", [&avatar_url]() -> json::value
{
return avatar_url;
}
}
};
m::vm::copts vmopts;
vmopts.infolog_accept = true;
vmopts.room_version = room_version;
vmopts.user_id = user_id;
vmopts.fetch = false;
vmopts.auth = false;
const vm::eval eval
{
event, content, vmopts
};
strlcpy(room_version_buf, room_version);
assert(eval.event_id);
return eval.event_id;
}
catch(const std::exception &e)
{
log::error
{
log, "Bootstrap %s for %s make_join to %s :%s",
string_view{room_id},
string_view{user_id},
string(host),
e.what(),
};
// This needs to rethrow because if the make_join doesn't complete we
// won't have enough information about the room to further continue the
// bootstrap process.
throw;
}
bool
ircd::m::room::bootstrap::required(const id &room_id)
{
// No bootstrap for my rooms
//TODO: issue for clustering
if(my(room_id))
return false;
// We have nothing for the room
if(!exists(room_id))
return true;
// No users are currently joined from this server;
//TODO: bootstrap shouldn't have to be used to re-sync a room where we have
//TODO: some partial state, so this condition should be eliminated.
if(local_joined(room_id) == 0)
return true;
return false;
}