0
0
Fork 0
mirror of https://github.com/matrix-construct/construct synced 2024-12-27 07:54:05 +01:00
construct/matrix/vm_fetch.cc
2022-07-02 23:56:11 -07:00

797 lines
16 KiB
C++

// The Construct
//
// Copyright (C) The Construct Developers, Authors & Contributors
// Copyright (C) 2016-2020 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::vm::fetch
{
static void prev_check(const event &, vm::eval &);
static bool prev_wait(const event &, vm::eval &);
static std::forward_list<ctx::future<m::fetch::result>> prev_fetch(const event &, vm::eval &, const room &);
static void prev(const event &, vm::eval &, const room &);
static std::forward_list<ctx::future<m::fetch::result>> state_fetch(const event &, vm::eval &, const room &);
static void state(const event &, vm::eval &, const room &);
static void auth_chain_eval(const event &, vm::eval &, const room &, const json::array &, const string_view &);
static void auth_chain(const event &, vm::eval &, const room &);
static void auth(const event &, vm::eval &, const room &);
static void handle(const event &, vm::eval &);
extern conf::item<milliseconds> prev_fetch_check_interval;
extern conf::item<milliseconds> prev_wait_time;
extern conf::item<size_t> prev_wait_count;
extern conf::item<size_t> prev_backfill_limit;
extern conf::item<seconds> event_timeout;
extern conf::item<seconds> state_timeout;
extern conf::item<seconds> auth_timeout;
extern conf::item<bool> enable;
extern hookfn<vm::eval &> auth_hook;
extern hookfn<vm::eval &> prev_hook;
extern hookfn<vm::eval &> state_hook;
extern log::log log;
}
decltype(ircd::m::vm::fetch::log)
ircd::m::vm::fetch::log
{
"m.vm.fetch"
};
decltype(ircd::m::vm::fetch::auth_hook)
ircd::m::vm::fetch::auth_hook
{
handle,
{
{ "_site", "vm.fetch.auth" }
}
};
decltype(ircd::m::vm::fetch::prev_hook)
ircd::m::vm::fetch::prev_hook
{
handle,
{
{ "_site", "vm.fetch.prev" }
}
};
decltype(ircd::m::vm::fetch::state_hook)
ircd::m::vm::fetch::state_hook
{
handle,
{
{ "_site", "vm.fetch.state" }
}
};
decltype(ircd::m::vm::fetch::enable)
ircd::m::vm::fetch::enable
{
{ "name", "ircd.m.vm.fetch.enable" },
{ "default", true },
};
decltype(ircd::m::vm::fetch::auth_timeout)
ircd::m::vm::fetch::auth_timeout
{
{ "name", "ircd.m.vm.fetch.auth.timeout" },
{ "default", 15L },
};
decltype(ircd::m::vm::fetch::state_timeout)
ircd::m::vm::fetch::state_timeout
{
{ "name", "ircd.m.vm.fetch.state.timeout" },
{ "default", 20L },
};
decltype(ircd::m::vm::fetch::event_timeout)
ircd::m::vm::fetch::event_timeout
{
{ "name", "ircd.m.vm.fetch.event.timeout" },
{ "default", 10L },
};
decltype(ircd::m::vm::fetch::prev_backfill_limit)
ircd::m::vm::fetch::prev_backfill_limit
{
{ "name", "ircd.m.vm.fetch.prev.backfill.limit" },
{ "default", 128L },
};
decltype(ircd::m::vm::fetch::prev_wait_count)
ircd::m::vm::fetch::prev_wait_count
{
{ "name", "ircd.m.vm.fetch.prev.wait.count" },
{ "default", 4L },
};
decltype(ircd::m::vm::fetch::prev_wait_time)
ircd::m::vm::fetch::prev_wait_time
{
{ "name", "ircd.m.vm.fetch.prev.wait.time" },
{ "default", 200L },
};
decltype(ircd::m::vm::fetch::prev_fetch_check_interval)
ircd::m::vm::fetch::prev_fetch_check_interval
{
{ "name", "ircd.m.vm.fetch.prev.fetch.check_interval" },
{ "default", 500L },
};
//
// fetch_phase
//
void
ircd::m::vm::fetch::handle(const event &event,
vm::eval &eval)
try
{
if(eval.room_internal)
return;
assert(eval.opts);
const auto &opts
{
*eval.opts
};
const auto &type
{
at<"type"_>(event)
};
if(type == "m.room.create")
return;
const m::event::id &event_id
{
event.event_id
};
const m::room::id &room_id
{
at<"room_id"_>(event)
};
// Can't construct m::room with the event_id argument because it
// won't be found (we're evaluating that event here!) so we just set
// the member manually to make further use of the room struct.
m::room room{room_id};
room.event_id = event_id;
if(eval.phase == phase::FETCH_AUTH)
return auth(event, eval, room);
if(eval.phase == phase::FETCH_PREV)
return prev(event, eval, room);
if(eval.phase == phase::FETCH_STATE)
return state(event, eval, room);
}
catch(const std::exception &e)
{
log::derror
{
log, "%s :%s",
loghead(eval),
e.what(),
};
throw;
}
//
// auth_events handler stack
//
void
ircd::m::vm::fetch::auth(const event &event,
vm::eval &eval,
const room &room)
{
// Count how many of the auth_events provided exist locally.
const auto &opts{*eval.opts};
const event::auth prev
{
event
};
const size_t auth_count
{
prev.auth_events_count()
};
const size_t auth_exists
{
prev.auth_events_exist()
};
assert(auth_exists <= auth_count);
if(auth_exists != auth_count) try
{
log::dwarning
{
log, "%s auth_events:%zu miss:%zu",
loghead(eval),
auth_count,
auth_count - auth_exists,
};
if(!bool(m::vm::fetch::enable))
throw vm::error
{
vm::fault::AUTH, "Fetching auth_events disabled by configuration",
};
// This is a blocking call to recursively fetch and evaluate the auth_chain
// for this event. Upon return all of the auth_events for this event will
// have themselves been fetched and auth'ed recursively.
auth_chain(event, eval, room);
}
catch(const std::exception &e)
{
const ctx::exception_handler eh;
throw vm::error
{
vm::fault::AUTH, "Failed to fetch %zu of %zu auth_events :%s",
auth_count - prev.auth_events_exist(),
auth_count,
e.what()
};
}
}
void
ircd::m::vm::fetch::auth_chain(const event &event,
vm::eval &eval,
const room &room)
try
{
assert(eval.opts);
// Figure out a remote hint as the primary target to request the missing
// auth events from; if provided, m::fetch will ask this remote first. We
// try to use the eval.node_id, which is set to a server that is conducting
// the eval (i.e in a /send/ or when processing some response data from
// them); next we try the origin of the event itself. These remotes are
// most likely to provide a satisfying response.
const string_view hint
{
eval.opts->node_id && !my_host(eval.opts->node_id)?
eval.opts->node_id:
event.event_id.host() && !my_host(event.event_id.host())?
event.event_id.host():
!my_host(json::get<"origin"_>(event))?
string_view(json::get<"origin"_>(event)):
room.room_id.host() && !my_host(room.room_id.host())?
room.room_id.host():
string_view{}
};
log::debug
{
log, "Fetching auth chain for %s in %s hint:%s",
string_view{room.event_id},
string_view{room.room_id},
hint,
};
// send
auto future
{
m::fetch::start(
{
.op = m::fetch::op::auth,
.room_id = room.room_id,
.event_id = room.event_id,
.hint = hint,
.check_hashes = false,
})
};
// recv
const auto result
{
future.get(seconds(auth_timeout))
};
const json::object response
{
result
};
// parse
const json::array &auth_chain
{
response["auth_chain"]
};
auth_chain_eval(event, eval, room, auth_chain, result.origin);
}
catch(const vm::error &e)
{
throw;
}
catch(const std::exception &e)
{
log::error
{
log, "Fetching auth chain for %s in %s :%s",
string_view{room.event_id},
string_view{room.room_id},
e.what(),
};
throw;
}
void
ircd::m::vm::fetch::auth_chain_eval(const event &event,
vm::eval &eval,
const room &room,
const json::array &auth_chain_,
const string_view &origin)
try
{
assert(eval.opts);
auto opts(*eval.opts);
opts.fetch = false;
opts.infolog_accept = true;
opts.warnlog &= ~vm::fault::EXISTS;
opts.notify_servers = false;
opts.node_id = origin;
std::vector<m::event> auth_chain
(
std::begin(auth_chain_), std::end(auth_chain_)
);
// pre-sort here and indicate that to eval.
std::sort(begin(auth_chain), end(auth_chain));
opts.ordered = true;
log::debug
{
log, "Evaluating auth chain for %s in %s events:%zu",
string_view{room.event_id},
string_view{room.room_id},
auth_chain.size(),
};
// eval
m::vm::eval
{
auth_chain, opts
};
}
catch(const std::exception &e)
{
log::error
{
log, "Evaluating auth chain for %s in %s :%s",
string_view{room.event_id},
string_view{room.room_id},
e.what(),
};
throw;
}
//
// state handler stack
//
void
ircd::m::vm::fetch::state(const event &event,
vm::eval &eval,
const room &room)
try
{
const auto &opts{*eval.opts};
const event::prev prev
{
event
};
const bool prev_exist
{
prev.prev_exist()
};
if(opts.fetch_state_any)
if(prev_exist && prev.prev_events_count() == prev.prev_events_exist())
return;
if(likely(!opts.fetch_state_any))
if(prev_exist)
return;
if(likely(!opts.fetch_state_shallow))
{
const auto &[sounding_depth, sounding_idx]
{
m::sounding(room.room_id)
};
if(at<"depth"_>(event) > sounding_depth)
return;
}
log::dwarning
{
log, "%s fetching possible missing state in %s",
loghead(eval),
string_view{room.room_id},
};
struct m::acquire::opts acq_opts;
acq_opts.room = room;
acq_opts.head = false;
acq_opts.history = false;
acq_opts.state = true;
acq_opts.hint = opts.node_id;
m::acquire acq
{
acq_opts
};
//TODO: XXX
log::info
{
log, "%s evaluated missing state in %s fetched:-- good:-- fail:--",
loghead(eval),
string_view{room.room_id},
0,
0,
0,
};
}
catch(const std::exception &e)
{
log::error
{
log, "%s state fetch in %s :%s",
loghead(eval),
string_view{room.room_id},
e.what(),
};
throw;
}
//
// prev_events handler stack
//
void
ircd::m::vm::fetch::prev(const event &event,
vm::eval &eval,
const room &room)
{
const auto &opts{*eval.opts};
const event::prev prev{event};
const size_t prev_count
{
prev.prev_events_count()
};
const size_t prev_exists
{
prev.prev_events_exist()
};
assert(prev_exists <= prev_count);
if(prev_count == prev_exists)
return;
// Attempt to wait for missing prev_events without issuing fetches here.
if(prev_wait(event, eval))
return;
if(!m::vm::fetch::enable)
{
prev_check(event, eval);
return;
}
auto futures
{
prev_fetch(event, eval, room)
};
// At this point one or more prev_events are missing; the fetches were
// launched asynchronously if the options allowed for it.
log::dwarning
{
log, "%s depth:%ld prev_events:%zu miss:%zu fetching:%zu fetching ...",
loghead(eval),
at<"depth"_>(event),
prev_count,
prev_count - prev_exists,
std::distance(begin(futures), end(futures)),
};
auto fetching
{
ctx::when_all(begin(futures), end(futures))
};
const auto timeout
{
now<system_point>() + seconds(event_timeout)
};
const milliseconds &check_interval
{
prev_fetch_check_interval
};
// Rather than waiting for all of the events to arrive or for the entire
// timeout to expire, we check if the sought events made it to the server
// in the meantime. If so we can drop these requests and bail.
//TODO: Ideally should be replaced with listener/notification/hook on the
//TODO: events arriving rather than this coarse sleep cycles.
while(now<system_point>() < timeout)
{
// Wait for an interval to give this loop some iterations.
if(fetching.wait(check_interval, std::nothrow))
break;
// Check for satisfaction.
if(prev.prev_events_exist() == prev_count)
return;
}
// evaluate results
for(auto &future : futures) try
{
m::fetch::result result
{
future.get()
};
const json::object content
{
result
};
const json::array &pdus
{
content["pdus"]
};
auto opts(*eval.opts);
opts.phase.set(m::vm::phase::FETCH_PREV, false);
opts.phase.set(m::vm::phase::FETCH_STATE, false);
opts.notify_servers = false;
opts.node_id = result.origin;
log::debug
{
log, "%s fetched %zu pdus; evaluating...",
loghead(eval),
pdus.size(),
};
vm::eval
{
pdus, opts
};
}
catch(const ctx::interrupted &)
{
throw;
}
catch(const std::exception &e)
{
log::derror
{
log, "%s prev fetch/eval :%s",
loghead(eval),
e.what(),
};
}
// check if result evals have satisfied this eval now; or throw
prev_check(event, eval);
}
std::forward_list
<
ircd::ctx::future<ircd::m::fetch::result>
>
ircd::m::vm::fetch::prev_fetch(const event &event,
vm::eval &eval,
const room &room)
{
const long room_depth
{
m::depth(std::nothrow, room)
};
const long viewport_depth
{
room_depth - room::events::viewport_size
};
std::forward_list
<
ctx::future<m::fetch::result>
>
ret;
const event::prev prev{event};
for(size_t i(0); i < prev.prev_events_count(); ++i) try
{
const auto &prev_id
{
prev.prev_event(i)
};
if(m::exists(prev_id))
continue;
const long depth_gap
{
std::max(std::abs(at<"depth"_>(event) - room_depth), 1L)
};
const auto backfill_max
{
std::min(size_t(prev_backfill_limit), eval.opts->fetch_prev_limit)
};
const auto backfill_limit
{
std::min(size_t(depth_gap), backfill_max)
};
const string_view hint
{
eval.opts->node_id && !my_host(eval.opts->node_id)?
eval.opts->node_id:
prev_id.host() && !my_host(prev_id.host())?
prev_id.host():
!my_host(json::get<"origin"_>(event))?
string_view(json::get<"origin"_>(event)):
room.room_id.host() && !my_host(room.room_id.host())?
room.room_id.host():
string_view{}
};
log::debug
{
log, "%s requesting backfill off %s; "
"depth:%ld viewport:%ld room:%ld gap:%ld limit:%zu hint:%s",
loghead(eval),
string_view{prev_id},
at<"depth"_>(event),
viewport_depth,
room_depth,
depth_gap,
backfill_limit,
hint,
};
ret.emplace_front(m::fetch::start(
{
.op = m::fetch::op::backfill,
.room_id = room.room_id,
.event_id = prev_id,
.hint = hint,
.backfill_limit = backfill_limit,
}));
}
catch(const ctx::interrupted &)
{
throw;
}
catch(const std::exception &e)
{
log::derror
{
log, "%s requesting backfill off prev %s; depth:%ld :%s",
loghead(eval),
string_view{prev.prev_event(i)},
json::get<"depth"_>(event),
e.what(),
};
}
return ret;
}
//TODO: Adjust when PDU lookahead/lookaround is fixed in the vm::eval iface.
//TODO: Wait on another eval completion instead of just coarse sleep()'s.
bool
ircd::m::vm::fetch::prev_wait(const event &event,
vm::eval &eval)
{
const auto &opts(*eval.opts);
const event::prev prev(event);
const size_t prev_count
{
prev.prev_events_count()
};
const size_t &wait_count
{
ssize_t(opts.fetch_prev_wait_count) >= 0?
opts.fetch_prev_wait_count:
size_t(prev_wait_count)
};
const milliseconds &wait_time
{
opts.fetch_prev_wait_time >= 0ms?
opts.fetch_prev_wait_time:
milliseconds(prev_wait_time)
};
size_t i(0); while(i < wait_count)
{
sleep(milliseconds(++i * wait_time));
if(prev_count == prev.prev_events_exist())
return true;
}
return false;
}
void
ircd::m::vm::fetch::prev_check(const event &event,
vm::eval &eval)
{
const auto &opts
{
*eval.opts
};
const event::prev prev
{
event
};
const size_t prev_exists
{
prev.prev_events_exist()
};
// Aborts this event if the options want us to guarantee at least one
// prev_event was fetched and evaluated for this event. This is generally
// used in conjunction with the fetch_prev_wait option to be effective.
if(opts.fetch_prev_any && !prev_exists)
throw vm::error
{
vm::fault::EVENT, "Failed to fetch any of the %zu prev_events for %s in %s",
prev.prev_events_count(),
string_view{event.event_id},
json::get<"room_id"_>(event)
};
// Aborts this event if the options want us to guarantee ALL of the
// prev_events were fetched and evaluated for this event.
if(opts.fetch_prev_all && prev_exists < prev.prev_events_count())
throw vm::error
{
vm::fault::EVENT, "Missing %zu of %zu required prev_events for %s in %s",
prev_exists,
prev.prev_events_count(),
string_view{event.event_id},
json::get<"room_id"_>(event)
};
}