You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
lokinet/llarp/service/endpoint.cpp

1670 lines
43 KiB
C++

#include "endpoint.hpp"
#include "auth.hpp"
#include "endpoint_state.hpp"
#include "endpoint_util.hpp"
#include "info.hpp"
#include "outbound_context.hpp"
#include "protocol.hpp"
#include "protocol_type.hpp"
#include <llarp/dht/key.hpp>
#include <llarp/link/contacts.hpp>
#include <llarp/link/tunnel.hpp>
#include <llarp/net/ip.hpp>
#include <llarp/net/ip_range.hpp>
#include <llarp/nodedb.hpp>
#include <llarp/path/path.hpp>
#include <llarp/profiling.hpp>
#include <llarp/router/route_poker.hpp>
#include <llarp/router/router.hpp>
#include <llarp/util/logging.hpp>
#include <llarp/util/priority_queue.hpp>
#include <optional>
#include <type_traits>
#include <utility>
#include <variant>
namespace llarp::service
{
static auto logcat = log::Cat("endpoint");
Endpoint::Endpoint(Router* r, Context* parent)
: path::Builder{r, 3, path::DEFAULT_LEN}
, context{parent}
, _inbound_queue{512}
, _send_queue{512}
, _recv_event_queue{512}
, _introset_lookup_filter{5s}
{
_state = std::make_unique<EndpointState>();
_state->router = r;
_state->name = "endpoint";
_recv_event_queue.enable();
// if (Loop()->MaybeGetUVWLoop())
// _tunnel_manager = std::make_unique<link::TunnelManager>(*this);
}
bool
Endpoint::Configure(const NetworkConfig& conf, [[maybe_unused]] const DnsConfig& dnsConf)
{
if (conf.paths.has_value())
numDesiredPaths = *conf.paths;
if (conf.hops.has_value())
numHops = *conf.hops;
conf.exit_map.ForEachEntry(
[&](const IPRange& range, const service::Address& addr) { MapExitRange(range, addr); });
for (auto [exit, auth] : conf.exit_auths)
{
SetAuthInfoForEndpoint(exit, auth);
}
conf.ons_exit_map.ForEachEntry([&](const IPRange& range, const std::string& name) {
std::optional<AuthInfo> auth;
const auto itr = conf.ons_exit_auths.find(name);
if (itr != conf.ons_exit_auths.end())
auth = itr->second;
_startup_ons_mappings[name] = std::make_pair(range, auth);
});
return _state->Configure(conf);
}
bool
Endpoint::HasPendingPathToService(const Address& addr) const
{
return _state->pending_service_lookups.find(addr) != _state->pending_service_lookups.end();
}
bool
Endpoint::is_ready() const
{
const auto now = llarp::time_now_ms();
if (intro_set().intros.empty())
return false;
if (intro_set().IsExpired(now))
return false;
return true;
}
bool
Endpoint::HasPendingRouterLookup(const RouterID remote) const
{
const auto& routers = _state->pending_routers;
return routers.find(remote) != routers.end();
}
std::optional<std::variant<Address, RouterID>>
Endpoint::GetEndpointWithConvoTag(ConvoTag tag) const
{
auto itr = Sessions().find(tag);
if (itr != Sessions().end())
{
return itr->second.remote.Addr();
}
for (const auto& item : _state->snode_sessions)
{
if (const auto maybe = item.second->CurrentPath())
{
if (ConvoTag{maybe->as_array()} == tag)
return item.first;
}
}
return std::nullopt;
}
void
Endpoint::map_exit(
std::string name,
std::string token,
std::vector<IPRange> ranges,
std::function<void(bool, std::string)> result_handler)
{
if (ranges.empty())
{
result_handler(false, "no ranges provided");
return;
}
lookup_name(
name,
[ptr = std::static_pointer_cast<Endpoint>(GetSelf()),
name,
auth = AuthInfo{token},
ranges,
result_handler,
poker = router()->route_poker()](std::string name_result, bool success) mutable {
if (not success)
{
result_handler(false, "Exit {} not found!"_format(name));
return;
}
if (auto saddr = service::Address(); saddr.FromString(name_result))
{
ptr->SetAuthInfoForEndpoint(saddr, auth);
ptr->MarkAddressOutbound(saddr);
auto result = ptr->EnsurePathToService(
saddr,
[ptr, name, name_result, ranges, result_handler, poker](
auto addr, OutboundContext* ctx) {
if (ctx == nullptr)
{
result_handler(
false, "could not establish flow to {} ({})"_format(name_result, name));
return;
}
// make a lambda that sends the reply after doing auth
auto apply_result = [ptr, poker, addr, result_handler, ranges](
std::string result, bool success) {
if (success)
{
for (const auto& range : ranges)
ptr->MapExitRange(range, addr);
if (poker)
poker->put_up();
}
result_handler(success, result);
};
ctx->send_auth_async(apply_result);
},
ptr->PathAlignmentTimeout());
if (not result)
result_handler(false, "Could not build path to {} ({})"_format(name_result, name));
}
});
}
void
Endpoint::LookupServiceAsync(
std::string name,
std::string service,
std::function<void(std::vector<dns::SRVData>)> resultHandler)
{
// A lookup goes through a chain of events:
// - see if the name is ONS, and if so resolve it to a ADDR.loki
// - once we've resolved to ADDR.loki then initiate a path to it
// - once we have a path, consult the remote's introset to pull out the SRV records
// If we fail along the way (e.g. it's a .snode, we can't build a path, or whatever else) then
// we invoke the resultHandler with an empty vector.
lookup_name(
name,
[this, resultHandler, service = std::move(service)](
std::string name_result, bool success) mutable {
if (!success)
return resultHandler({});
std::string name;
try
{
oxenc::bt_dict_consumer btdc{name_result};
name = btdc.require<std::string>("NAME");
}
catch (...)
{
log::warning(link_cat, "Failed to parse find name response!");
return resultHandler({});
}
auto saddr = service::Address();
if (!saddr.FromString(name))
return resultHandler({}); // Not a regular ADDR.loki so doesn't support SRV
// initiate path build
const auto build_started = EnsurePathTo(
saddr,
[this, address = std::move(saddr), resultHandler, service = std::move(service)](
auto maybe_tag) {
if (not maybe_tag)
return resultHandler({});
// we can probably get this info before we have a path to them but we do this after
// we have a path so when we send the DNS response back they can talk to them
// immediately
const auto& container = _state->remote_sessions;
if (auto itr = container.find(address); itr != container.end())
// parse the stuff we need from this guy
resultHandler(itr->second->GetCurrentIntroSet().GetMatchingSRVRecords(service));
else
resultHandler({});
},
PathAlignmentTimeout());
// on path build start fail short circuit
if (not build_started)
resultHandler({});
});
}
bool
Endpoint::IntrosetIsStale() const
{
return intro_set().HasExpiredIntros(llarp::time_now_ms());
}
util::StatusObject
Endpoint::ExtractStatus() const
{
auto obj = path::Builder::ExtractStatus();
obj["exitMap"] = _exit_map.ExtractStatus();
obj["identity"] = _identity.pub.Addr().ToString();
obj["networkReady"] = ReadyForNetwork();
util::StatusObject authCodes;
for (const auto& [service, info] : _remote_auth_infos)
{
authCodes[service.ToString()] = info.token;
}
obj["authCodes"] = authCodes;
return _state->ExtractStatus(obj);
}
void
Endpoint::Tick(llarp_time_t)
{
const auto now = llarp::time_now_ms();
path::Builder::Tick(now);
// publish descriptors
if (ShouldPublishDescriptors(now))
{
regen_and_publish_introset();
}
// decay introset lookup filter
_introset_lookup_filter.Decay(now);
// expire name cache
_state->nameCache.Decay(now);
// expire snode sessions
EndpointUtil::ExpireSNodeSessions(now, _state->snode_sessions);
// expire pending router lookups
EndpointUtil::ExpirePendingRouterLookups(now, _state->pending_routers);
// deregister dead sessions
EndpointUtil::DeregisterDeadSessions(now, _state->dead_sessions);
// tick remote sessions
EndpointUtil::TickRemoteSessions(
now, _state->remote_sessions, _state->dead_sessions, Sessions());
// expire convotags
EndpointUtil::ExpireConvoSessions(now, Sessions());
if (NumInStatus(path::ESTABLISHED) > 1)
{
for (const auto& item : _startup_ons_mappings)
{
auto& name = item.first;
lookup_name(
name, [this, name, info = item.second](std::string name_result, bool success) mutable {
if (not success)
return;
const auto maybe_range = info.first;
const auto maybe_auth = info.second;
_startup_ons_mappings.erase(name);
if (auto saddr = service::Address(); saddr.FromString(name_result))
{
if (maybe_range.has_value())
_exit_map.Insert(*maybe_range, saddr);
if (maybe_auth.has_value())
SetAuthInfoForEndpoint(saddr, *maybe_auth);
}
});
}
}
}
bool
Endpoint::Stop()
{
// stop remote sessions
log::debug(logcat, "Endpoint stopping remote sessions.");
EndpointUtil::StopRemoteSessions(_state->remote_sessions);
// stop snode sessions
log::debug(logcat, "Endpoint stopping snode sessions.");
EndpointUtil::StopSnodeSessions(_state->snode_sessions);
log::debug(logcat, "Endpoint stopping its path builder.");
return path::Builder::Stop();
}
uint64_t
Endpoint::GenTXID()
{
return randint();
}
std::string
Endpoint::Name() const
{
return _state->name + ":" + _identity.pub.Name();
}
bool
Endpoint::HasInboundConvo(const Address& addr) const
{
for (const auto& item : Sessions())
{
if (item.second.remote.Addr() == addr and item.second.inbound)
return true;
}
return false;
}
bool
Endpoint::HasOutboundConvo(const Address& addr) const
{
for (const auto& item : Sessions())
{
if (item.second.remote.Addr() == addr && not item.second.inbound)
return true;
}
return false;
}
void
Endpoint::PutSenderFor(const ConvoTag& tag, const ServiceInfo& info, bool inbound)
{
if (info.Addr().IsZero())
{
LogError(Name(), " cannot put invalid service info ", info, " T=", tag);
return;
}
auto itr = Sessions().find(tag);
if (itr == Sessions().end())
{
if (WantsOutboundSession(info.Addr()) and inbound)
{
LogWarn(
Name(),
" not adding sender for ",
info.Addr(),
" session is inbound and we want outbound T=",
tag);
return;
}
itr = Sessions().emplace(tag, Session{}).first;
itr->second.inbound = inbound;
itr->second.remote = info;
}
}
size_t
Endpoint::RemoveAllConvoTagsFor(service::Address remote)
{
size_t removed = 0;
auto& sessions = Sessions();
auto itr = sessions.begin();
while (itr != sessions.end())
{
if (itr->second.remote.Addr() == remote)
{
itr = sessions.erase(itr);
removed++;
}
else
++itr;
}
return removed;
}
bool
Endpoint::GetSenderFor(const ConvoTag& tag, ServiceInfo& si) const
{
auto itr = Sessions().find(tag);
if (itr == Sessions().end())
return false;
si = itr->second.remote;
si.UpdateAddr();
return true;
}
void
Endpoint::PutIntroFor(const ConvoTag& tag, const Introduction& intro)
{
auto& s = Sessions()[tag];
s.intro = intro;
}
bool
Endpoint::GetIntroFor(const ConvoTag& tag, Introduction& intro) const
{
auto itr = Sessions().find(tag);
if (itr == Sessions().end())
return false;
intro = itr->second.intro;
return true;
}
void
Endpoint::PutReplyIntroFor(const ConvoTag& tag, const Introduction& intro)
{
auto itr = Sessions().find(tag);
if (itr == Sessions().end())
{
return;
}
itr->second.replyIntro = intro;
}
bool
Endpoint::GetReplyIntroFor(const ConvoTag& tag, Introduction& intro) const
{
auto itr = Sessions().find(tag);
if (itr == Sessions().end())
return false;
intro = itr->second.replyIntro;
return true;
}
bool
Endpoint::GetConvoTagsForService(const Address& addr, std::set<ConvoTag>& tags) const
{
return EndpointUtil::GetConvoTagsForService(Sessions(), addr, tags);
}
bool
Endpoint::GetCachedSessionKeyFor(const ConvoTag& tag, SharedSecret& secret) const
{
auto itr = Sessions().find(tag);
if (itr == Sessions().end())
return false;
secret = itr->second.sharedKey;
return true;
}
void
Endpoint::PutCachedSessionKeyFor(const ConvoTag& tag, const SharedSecret& k)
{
auto itr = Sessions().find(tag);
if (itr == Sessions().end())
{
itr = Sessions().emplace(tag, Session{}).first;
}
itr->second.sharedKey = k;
}
void
Endpoint::ConvoTagTX(const ConvoTag& tag)
{
if (Sessions().count(tag))
Sessions()[tag].TX();
}
void
Endpoint::ConvoTagRX(const ConvoTag& tag)
{
if (Sessions().count(tag))
Sessions()[tag].RX();
}
bool
Endpoint::LoadKeyFile()
{
const auto& keyfile = _state->key_file;
if (!keyfile.empty())
{
_identity.EnsureKeys(keyfile, router()->key_manager()->needs_backup());
}
else
{
_identity.RegenerateKeys();
}
return true;
}
bool
Endpoint::Start()
{
// this does network isolation
while (_state->on_init_callbacks.size())
{
if (_state->on_init_callbacks.front()())
_state->on_init_callbacks.pop_front();
else
{
LogWarn("Can't call init of network isolation");
return false;
}
}
return true;
}
// Keep this here (rather than the header) so that we don't need to include endpoint_state.hpp
// in endpoint.hpp for the unique_ptr member destructor.
Endpoint::~Endpoint() = default;
void
Endpoint::regen_and_publish_introset()
{
const auto now = llarp::time_now_ms();
_last_introset_regen_attempt = now;
std::set<Introduction, CompareIntroTimestamp> intros;
if (const auto maybe =
GetCurrentIntroductionsWithFilter([now](const service::Introduction& intro) -> bool {
return not intro.ExpiresSoon(now, path::INTRO_STALE_THRESHOLD);
}))
{
intros.insert(maybe->begin(), maybe->end());
}
else
{
LogWarn(
"could not publish descriptors for endpoint ",
Name(),
" because we couldn't get enough valid introductions");
BuildOne();
return;
}
intro_set().supported_protocols.clear();
// add supported ethertypes
if (HasIfAddr())
{
if (IPRange::V4MappedRange().Contains(GetIfAddr()))
{
intro_set().supported_protocols.push_back(ProtocolType::TrafficV4);
}
else
{
intro_set().supported_protocols.push_back(ProtocolType::TrafficV6);
}
// exit related stuffo
if (_state->is_exit_enabled)
{
intro_set().supported_protocols.push_back(ProtocolType::Exit);
intro_set().exit_policy = GetExitPolicy();
intro_set().owned_ranges = GetOwnedRanges();
}
}
// add quic ethertype if we have listeners set up
// if (auto* quic = GetQUICTunnel())
// {
// TODO:
// if (quic->hasListeners())
// intro_set().supported_protocols.push_back(ProtocolType::QUIC);
// }
intro_set().intros.clear();
for (auto& intro : intros)
{
if (intro_set().intros.size() < numDesiredPaths)
intro_set().intros.emplace_back(std::move(intro));
}
if (intro_set().intros.empty())
{
LogWarn("not enough intros to publish introset for ", Name());
if (ShouldBuildMore(now))
ManualRebuild(1);
return;
}
auto maybe = _identity.encrypt_and_sign_introset(intro_set(), now);
if (not maybe)
{
LogWarn("failed to generate introset for endpoint ", Name());
return;
}
if (publish_introset(*maybe))
{
LogInfo("(re)publishing introset for endpoint ", Name());
}
else
{
LogWarn("failed to publish intro set for endpoint ", Name());
}
}
bool
Endpoint::publish_introset(const EncryptedIntroSet& introset)
{
const auto paths = GetManyPathsWithUniqueEndpoints(
this, INTROSET_RELAY_REDUNDANCY, dht::Key_t{introset.derivedSigningKey.as_array()});
if (paths.size() != INTROSET_RELAY_REDUNDANCY)
{
LogWarn(
"Cannot publish intro set because we only have ",
paths.size(),
" paths, but need ",
INTROSET_RELAY_REDUNDANCY);
return false;
}
for (const auto& path : paths)
{
for (size_t i = 0; i < INTROSET_REQS_PER_RELAY; ++i)
{
router()->send_control_message(path->upstream(), "publish_intro", introset.bt_encode());
}
}
return true;
}
size_t
Endpoint::UniqueEndpoints() const
{
return _state->remote_sessions.size() + _state->snode_sessions.size();
}
[[maybe_unused]] constexpr auto PublishIntrosetTimeout = 20s;
void
Endpoint::ResetInternalState()
{
path::Builder::ResetInternalState();
static auto resetState = [](auto& container, auto getter) {
std::for_each(container.begin(), container.end(), [getter](auto& item) {
getter(item)->ResetInternalState();
});
};
resetState(_state->remote_sessions, [](const auto& item) { return item.second; });
resetState(_state->snode_sessions, [](const auto& item) { return item.second; });
}
bool
Endpoint::ShouldPublishDescriptors(llarp_time_t now) const
{
if (not _publish_introset)
return false;
const auto lastEventAt = std::max(_state->last_publish_attempt, _state->last_publish);
const auto next_pub = lastEventAt
+ (_state->local_introset.HasStaleIntros(now, path::INTRO_STALE_THRESHOLD)
? IntrosetPublishRetryCooldown
: IntrosetPublishInterval);
return now >= next_pub;
}
std::optional<std::vector<RemoteRC>>
Endpoint::GetHopsForBuild()
{
std::unordered_set<RouterID> exclude;
ForEachPath([&exclude](auto path) { exclude.insert(path->Endpoint()); });
auto hook = [exclude, r = router()](const RemoteRC& rc) -> bool {
const auto& rid = rc.router_id();
return not(exclude.count(rid) || r->router_profiling().IsBadForPath(rid));
};
if (auto maybe = router()->node_db()->get_random_rc_conditional(hook))
return GetHopsForBuildWithEndpoint(maybe->router_id());
return std::nullopt;
}
std::optional<std::vector<RemoteRC>>
Endpoint::GetHopsForBuildWithEndpoint(RouterID endpoint)
{
return path::Builder::GetHopsAlignedToForBuild(endpoint, SnodeBlacklist());
}
constexpr auto MaxOutboundContextPerRemote = 1;
void
Endpoint::PutNewOutboundContext(const service::IntroSet& introset, llarp_time_t)
{
const Address addr{introset.address_keys.Addr()};
auto& remoteSessions = _state->remote_sessions;
if (remoteSessions.count(addr) < MaxOutboundContextPerRemote)
{
remoteSessions.emplace(addr, std::make_shared<OutboundContext>(introset, this));
LogInfo("Created New outbound context for ", addr.ToString());
}
auto sessionRange = remoteSessions.equal_range(addr);
for (auto itr = sessionRange.first; itr != sessionRange.second; ++itr)
{
// TODO:
// itr->second->AddReadyHook(
// [addr, this](auto session) { InformPathToService(addr, session); }, left);
}
}
bool
Endpoint::HasExit() const
{
for (const auto& [name, info] : _startup_ons_mappings)
{
if (info.first.has_value())
return true;
}
return not _exit_map.Empty();
}
auto
Endpoint::GetUniqueEndpointsForLookup() const
{
path::UniqueEndpointSet_t paths;
ForEachPath([&paths](auto path) {
if (path and path->IsReady())
paths.insert(path);
});
return paths;
}
bool
Endpoint::ReadyForNetwork() const
{
return is_ready() and ReadyToDoLookup(GetUniqueEndpointsForLookup().size());
}
bool
Endpoint::ReadyToDoLookup(size_t num_paths) const
{
// Currently just checks the number of paths, but could do more checks in the future.
return num_paths >= MIN_ONS_LOOKUP_ENDPOINTS;
}
void
Endpoint::lookup_name(std::string name, std::function<void(std::string, bool)> func)
{
// TODO: so fuck all this?
// if (not is_valid_name(name))
// {
// handler(parse_address(name));
// return;
// }
// auto& cache = _state->nameCache;
// const auto maybe = cache.Get(name);
// if (maybe.has_value())
// {
// handler(maybe);
// return;
// }
log::info(link_cat, "{} looking up ONS name {}", Name(), name);
auto paths = GetUniqueEndpointsForLookup();
// // not enough paths
// if (not ReadyToDoLookup(paths.size()))
// {
// LogWarn(
// Name(),
// " not enough paths for lns lookup, have ",
// paths.size(),
// " need ",
// MIN_ONS_LOOKUP_ENDPOINTS);
// handler(std::nullopt);
// return;
// }
// pick up to max_unique_lns_endpoints random paths to do lookups from
std::vector<path::Path_ptr> chosenpaths;
chosenpaths.insert(chosenpaths.begin(), paths.begin(), paths.end());
std::shuffle(chosenpaths.begin(), chosenpaths.end(), llarp::csrng);
chosenpaths.resize(std::min(paths.size(), MAX_ONS_LOOKUP_ENDPOINTS));
// TODO: only want one successful response to call the callback, or failed if all fail
auto response_cb = [func = std::move(func)](std::string resp) {
std::string name{};
try
{
oxenc::bt_dict_consumer btdc{resp};
auto status = btdc.require<std::string_view>(messages::STATUS_KEY);
if (status != "OK"sv)
{
log::info(link_cat, "Error on ONS lookup: {}", status);
func(std::string{status}, false);
}
name = btdc.require<std::string>("NAME");
}
catch (...)
{
log::warning(link_cat, "Failed to parse find name response!");
func("ERROR"s, false);
}
func(std::move(name), true);
};
for (const auto& path : chosenpaths)
{
log::info(link_cat, "{} lookup {} from {}", Name(), name, path->Endpoint());
path->find_name(name, response_cb);
}
}
void
Endpoint::HandlePathBuilt(path::Path_ptr p)
{
// p->SetDataHandler(util::memFn(&Endpoint::HandleHiddenServiceFrame, this));
// p->SetDropHandler(util::memFn(&Endpoint::HandleDataDrop, this));
// p->SetDeadChecker(util::memFn(&Endpoint::CheckPathIsDead, this));
path::Builder::HandlePathBuilt(p);
}
bool
Endpoint::HandleDataDrop(path::Path_ptr p, const PathID_t& dst, uint64_t seq)
{
LogWarn(Name(), " message ", seq, " dropped by endpoint ", p->Endpoint(), " via ", dst);
return true;
}
std::unordered_map<std::string, std::string>
Endpoint::NotifyParams() const
{
return {{"LOKINET_ADDR", _identity.pub.Addr().ToString()}};
}
void
Endpoint::FlushRecvData()
{
while (auto maybe = _recv_event_queue.tryPopFront())
{
auto& ev = *maybe;
ProtocolMessage::ProcessAsync(ev.fromPath, ev.pathid, ev.msg);
}
}
void
Endpoint::QueueRecvData(RecvDataEvent ev)
{
_recv_event_queue.tryPushBack(std::move(ev));
router()->TriggerPump();
}
bool
Endpoint::HandleDataMessage(
path::Path_ptr p, const PathID_t from, std::shared_ptr<ProtocolMessage> msg)
{
PutSenderFor(msg->tag, msg->sender, true);
Introduction intro = msg->introReply;
if (HasInboundConvo(msg->sender.Addr()))
{
intro.path_id = from;
intro.router = p->Endpoint();
}
PutReplyIntroFor(msg->tag, intro);
ConvoTagRX(msg->tag);
return ProcessDataMessage(msg);
}
bool
Endpoint::HasPathToSNode(const RouterID ident) const
{
auto range = _state->snode_sessions.equal_range(ident);
auto itr = range.first;
while (itr != range.second)
{
if (itr->second->IsReady())
{
return true;
}
++itr;
}
return false;
}
EndpointBase::AddressVariant_t
Endpoint::LocalAddress() const
{
return _identity.pub.Addr();
}
std::optional<EndpointBase::SendStat>
Endpoint::GetStatFor(AddressVariant_t) const
{
// TODO: implement me
return std::nullopt;
}
std::unordered_set<EndpointBase::AddressVariant_t>
Endpoint::AllRemoteEndpoints() const
{
std::unordered_set<AddressVariant_t> remote;
for (const auto& item : Sessions())
{
remote.insert(item.second.remote.Addr());
}
for (const auto& item : _state->snode_sessions)
{
remote.insert(item.first);
}
return remote;
}
bool
Endpoint::ProcessDataMessage(std::shared_ptr<ProtocolMessage> msg)
{
if ((msg->proto == ProtocolType::Exit
&& (_state->is_exit_enabled || _exit_map.ContainsValue(msg->sender.Addr())))
|| msg->proto == ProtocolType::TrafficV4 || msg->proto == ProtocolType::TrafficV6
|| (msg->proto == ProtocolType::QUIC and _tunnel_manager))
{
_inbound_queue.tryPushBack(std::move(msg));
router()->TriggerPump();
return true;
}
if (msg->proto == ProtocolType::Control)
{
// TODO: implement me (?)
// right now it's just random noise
return true;
}
return false;
}
void
Endpoint::AsyncProcessAuthMessage(
std::shared_ptr<ProtocolMessage> msg, std::function<void(std::string, bool)> hook)
{
if (_auth_policy)
{
if (not _auth_policy->auth_async_pending(msg->tag))
{
// do 1 authentication attempt and drop everything else
_auth_policy->authenticate_async(std::move(msg), std::move(hook));
}
}
else
{
router()->loop()->call([h = std::move(hook)] { h("OK", true); });
}
}
void
Endpoint::SendAuthResult(
path::Path_ptr path, PathID_t /* replyPath */, ConvoTag tag, std::string result, bool success)
{
// not applicable because we are not an exit or don't have an endpoint auth policy
if ((not _state->is_exit_enabled) or _auth_policy == nullptr)
return;
ProtocolFrameMessage f{};
f.flag = int(not success);
f.convo_tag = tag;
f.path_id = path->intro.path_id;
f.nonce.Randomize();
if (success)
{
ProtocolMessage msg;
msg.put_buffer(result);
if (_auth_policy)
msg.proto = ProtocolType::Auth;
else
msg.proto = ProtocolType::Control;
if (not GetReplyIntroFor(tag, msg.introReply))
{
LogError("Failed to send auth reply: no reply intro");
return;
}
msg.sender = _identity.pub;
SharedSecret sessionKey{};
if (not GetCachedSessionKeyFor(tag, sessionKey))
{
LogError("failed to send auth reply: no cached session key");
return;
}
if (not f.EncryptAndSign(msg, sessionKey, _identity))
{
LogError("Failed to encrypt and sign auth reply");
return;
}
}
else
{
if (not f.Sign(_identity))
{
LogError("failed to sign auth reply result");
return;
}
}
// TODO:
// _send_queue.tryPushBack(
// SendEvent{std::make_shared<routing::PathTransferMessage>(f, replyPath), path});
}
void
Endpoint::RemoveConvoTag(const ConvoTag& t)
{
Sessions().erase(t);
}
void
Endpoint::ResetConvoTag(ConvoTag tag, path::Path_ptr p, PathID_t /* from */)
{
// send reset convo tag message
ProtocolFrameMessage f{};
f.flag = 1;
f.convo_tag = tag;
f.path_id = p->intro.path_id;
f.Sign(_identity);
{
LogWarn("invalidating convotag T=", tag);
RemoveConvoTag(tag);
// TODO:
// _send_queue.tryPushBack(
// SendEvent{std::make_shared<routing::PathTransferMessage>(f, from), p});
}
}
bool
Endpoint::HandleHiddenServiceFrame(path::Path_ptr p, const ProtocolFrameMessage& frame)
{
if (frame.flag)
{
// handle discard
ServiceInfo si;
if (!GetSenderFor(frame.convo_tag, si))
return false;
// verify source
if (!frame.Verify(si))
return false;
// remove convotag it doesn't exist
LogWarn("remove convotag T=", frame.convo_tag, " R=", frame.flag, " from ", si.Addr());
RemoveConvoTag(frame.convo_tag);
return true;
}
if (not frame.AsyncDecryptAndVerify(router()->loop(), p, _identity, this))
{
ResetConvoTag(frame.convo_tag, p, frame.path_id);
}
return true;
}
void
Endpoint::HandlePathDied(path::Path_ptr p)
{
router()->router_profiling().MarkPathTimeout(p.get());
ManualRebuild(1);
path::Builder::HandlePathDied(p);
regen_and_publish_introset();
}
bool
Endpoint::CheckPathIsDead(path::Path_ptr, llarp_time_t dlt)
{
return dlt > path::ALIVE_TIMEOUT;
}
void
Endpoint::MarkAddressOutbound(service::Address addr)
{
_state->m_OutboundSessions.insert(addr);
}
bool
Endpoint::WantsOutboundSession(const Address& addr) const
{
return _state->m_OutboundSessions.count(addr) > 0;
}
void
Endpoint::InformPathToService(const Address remote, OutboundContext* ctx)
{
auto& serviceLookups = _state->pending_service_lookups;
auto range = serviceLookups.equal_range(remote);
auto itr = range.first;
while (itr != range.second)
{
itr->second(remote, ctx);
++itr;
}
serviceLookups.erase(remote);
}
bool
Endpoint::EnsurePathTo(
std::variant<Address, RouterID> addr,
std::function<void(std::optional<ConvoTag>)> hook,
llarp_time_t timeout)
{
if (auto ptr = std::get_if<Address>(&addr))
{
if (*ptr == _identity.pub.Addr())
{
ConvoTag tag{};
if (auto maybe = GetBestConvoTagFor(*ptr))
tag = *maybe;
else
tag.Randomize();
PutSenderFor(tag, _identity.pub, true);
ConvoTagTX(tag);
Sessions()[tag].forever = true;
Loop()->call_soon([tag, hook]() { hook(tag); });
return true;
}
if (not WantsOutboundSession(*ptr))
{
// we don't want to connect back to inbound sessions
hook(std::nullopt);
return true;
}
return EnsurePathToService(
*ptr,
[hook](auto, auto* ctx) -> bool {
if (ctx)
{
hook(ctx->get_current_tag());
return true;
}
hook(std::nullopt);
return false;
},
timeout);
}
if (auto ptr = std::get_if<RouterID>(&addr))
{
return EnsurePathToSNode(*ptr, [hook](auto, auto session, auto tag) {
if (session)
{
hook(tag);
}
else
{
hook(std::nullopt);
}
});
}
return false;
}
bool
Endpoint::EnsurePathToSNode(
const RouterID snode,
std::function<void(const RouterID, exit::BaseSession_ptr, ConvoTag)> hook)
{
auto& nodeSessions = _state->snode_sessions;
using namespace std::placeholders;
if (nodeSessions.count(snode) == 0)
{
const auto src = xhtonl(net::TruncateV6(GetIfAddr()));
const auto dst = xhtonl(net::TruncateV6(ObtainIPForAddr(snode)));
auto session = std::make_shared<exit::SNodeSession>(
snode,
[=](const llarp_buffer_t& buf) -> bool {
net::IPPacket pkt;
if (not pkt.Load(buf))
return false;
pkt.UpdateIPv4Address(src, dst);
/// TODO: V6
auto itr = _state->snode_sessions.find(snode);
if (itr == _state->snode_sessions.end())
return false;
if (const auto maybe = itr->second->CurrentPath())
return HandleInboundPacket(
ConvoTag{maybe->as_array()}, pkt.ConstBuffer(), ProtocolType::TrafficV4, 0);
return false;
},
router(),
1,
numHops,
false,
this);
_state->snode_sessions[snode] = session;
}
if (not router()->node_db()->has_rc(snode))
return false;
auto range = nodeSessions.equal_range(snode);
auto itr = range.first;
while (itr != range.second)
{
if (itr->second->IsReady())
hook(snode, itr->second, ConvoTag{itr->second->CurrentPath()->as_array()});
else
{
itr->second->AddReadyHook([hook, snode](auto session) {
if (session)
{
hook(snode, session, ConvoTag{session->CurrentPath()->as_array()});
}
else
{
hook(snode, nullptr, ConvoTag{});
}
});
if (not itr->second->BuildCooldownHit(Now()))
itr->second->BuildOne();
}
++itr;
}
return true;
}
bool
Endpoint::EnsurePathToService(
const Address remote,
std::function<void(Address, OutboundContext*)> hook,
[[maybe_unused]] llarp_time_t timeout)
{
if (not WantsOutboundSession(remote))
{
// we don't want to ensure paths to addresses that are inbound
// inform fail right away in that case
hook(remote, nullptr);
return false;
}
/// how many routers to use for lookups
static constexpr size_t NumParallelLookups = 2;
// add response hook to list for address.
_state->pending_service_lookups.emplace(remote, hook);
auto& sessions = _state->remote_sessions;
{
auto range = sessions.equal_range(remote);
auto itr = range.first;
while (itr != range.second)
{
if (itr->second->ReadyToSend())
{
InformPathToService(remote, itr->second.get());
return true;
}
++itr;
}
}
/// check replay filter
if (not _introset_lookup_filter.Insert(remote))
return true;
const auto paths = GetManyPathsWithUniqueEndpoints(this, NumParallelLookups);
const dht::Key_t location = remote.ToKey();
// flag to only add callback to list of callbacks for
// address once.
bool hookAdded = false;
auto got_it = std::make_shared<bool>(false);
// TODO: if all requests fail, call callback with failure?
for (const auto& path : paths)
{
path->find_intro(location, false, 0, [hook, got_it, this](std::string resp) mutable {
// asking many, use only first successful
if (*got_it)
return;
std::string introset;
try
{
oxenc::bt_dict_consumer btdc{resp};
auto status = btdc.require<std::string_view>(messages::STATUS_KEY);
if (status != "OK"sv)
{
log::info(link_cat, "Error in find intro set response: {}", status);
return;
}
introset = btdc.require<std::string>("INTROSET");
}
catch (...)
{
log::warning(link_cat, "Failed to parse find name response!");
throw;
}
service::EncryptedIntroSet enc{introset};
router()->contacts().put_intro(std::move(enc));
// TODO: finish this
/*
if (good)
*got_it = true;
*/
});
}
return hookAdded;
}
void
Endpoint::SRVRecordsChanged()
{
auto& introset = intro_set();
introset.SRVs.clear();
for (const auto& srv : SRVRecords())
introset.SRVs.emplace_back(srv.toTuple());
regen_and_publish_introset();
}
bool
Endpoint::send_to(ConvoTag tag, std::string payload)
{
if (tag.IsZero())
{
log::warning(link_cat, "SendToOrQueue failed: convo tag is zero");
return false;
}
log::debug(link_cat, "{} sending {} bytes (Tag: {})", Name(), payload.size(), tag);
if (auto maybe = GetEndpointWithConvoTag(tag))
{
if (auto rid = std::get_if<RouterID>(&*maybe))
{
return router()->send_data_message(*rid, payload);
}
if (auto saddr = std::get_if<Address>(&*maybe))
{
return router()->send_data_message(saddr->ToRouter(), payload);
}
}
log::debug(link_cat, "SendToOrQueue failed: no endpoint for convo tag {}", tag);
return false;
}
void
Endpoint::Pump(llarp_time_t)
{
FlushRecvData();
// send downstream packets to user for snode
for (const auto& [router, session] : _state->snode_sessions)
session->FlushDownstream();
// handle inbound traffic sorted
util::ascending_priority_queue<ProtocolMessage> queue;
while (not _inbound_queue.empty())
{
// suck it out
queue.emplace(std::move(*_inbound_queue.popFront()));
}
while (not queue.empty())
{
const auto& msg = queue.top();
log::debug(
logcat,
"{} handling inbound packet (size {}B) on tag {}",
Name(),
msg.tag,
msg.payload.size());
// if (HandleInboundPacket(msg.tag, msg.payload, msg.proto, msg.seqno))
// {
// ConvoTagRX(msg.tag);
// }
// else
// {
// LogWarn("Failed to handle inbound message");
// }
queue.pop();
}
// auto r = router();
// TODO: locking on this container
// for (const auto& [addr, outctx] : _state->remote_sessions)
// {
// outctx->FlushUpstream();
// outctx->Pump(now);
// }
// // TODO: locking on this container
// for (const auto& [r, session] : _state->snode_sessions)
// session->FlushUpstream();
// // send queue flush
// while (not _send_queue.empty())
// {
// SendEvent item = _send_queue.popFront();
// item.first->sequence_number = item.second->NextSeqNo();
// if (item.second->SendRoutingMessage(*item.first, r))
// ConvoTagTX(item.first->protocol_frame_msg.convo_tag);
// }
}
std::optional<ConvoTag>
Endpoint::GetBestConvoTagFor(std::variant<Address, RouterID> remote) const
{
// get convotag with lowest estimated RTT
if (auto ptr = std::get_if<Address>(&remote))
{
llarp_time_t rtt = 30s;
std::optional<ConvoTag> ret = std::nullopt;
for (const auto& [tag, session] : Sessions())
{
if (tag.IsZero())
continue;
if (session.remote.Addr() == *ptr)
{
if (*ptr == _identity.pub.Addr())
{
return tag;
}
if (session.inbound)
{
auto path = GetPathByRouter(session.replyIntro.router);
// if we have no path to the remote router that's fine still use it just in case this
// is the ONLY one we have
if (path == nullptr)
{
ret = tag;
continue;
}
if (path and path->IsReady())
{
const auto rttEstimate = (session.replyIntro.latency + path->intro.latency) * 2;
if (rttEstimate < rtt)
{
ret = tag;
rtt = rttEstimate;
}
}
}
else
{
auto range = _state->remote_sessions.equal_range(*ptr);
auto itr = range.first;
while (itr != range.second)
{
// TODO:
// if (itr->second->ReadyToSend() and itr->second->estimatedRTT > 0s)
// {
// if (itr->second->estimatedRTT < rtt)
// {
// ret = tag;
// rtt = itr->second->estimatedRTT;
// }
// }
itr++;
}
}
}
}
return ret;
}
if (auto* ptr = std::get_if<RouterID>(&remote))
{
auto itr = _state->snode_sessions.find(*ptr);
if (itr == _state->snode_sessions.end())
return std::nullopt;
if (auto maybe = itr->second->CurrentPath())
return ConvoTag{maybe->as_array()};
}
return std::nullopt;
}
bool
Endpoint::HasConvoTag(const ConvoTag& t) const
{
return Sessions().find(t) != Sessions().end();
}
std::optional<uint64_t>
Endpoint::GetSeqNoForConvo(const ConvoTag& tag)
{
auto itr = Sessions().find(tag);
if (itr == Sessions().end())
return std::nullopt;
return itr->second.seqno++;
}
bool
Endpoint::ShouldBuildMore(llarp_time_t now) const
{
if (BuildCooldownHit(now))
return false;
const auto requiredPaths = std::max(numDesiredPaths, path::MIN_INTRO_PATHS);
if (NumInStatus(path::BUILDING) >= requiredPaths)
return false;
return NumPathsExistingAt(now + (path::DEFAULT_LIFETIME - path::INTRO_PATH_SPREAD))
< requiredPaths;
}
Router*
Endpoint::router()
{
return _state->router;
}
const EventLoop_ptr&
Endpoint::Loop()
{
return router()->loop();
}
void
Endpoint::BlacklistSNode(const RouterID snode)
{
_state->snode_blacklist.insert(snode);
}
const std::set<RouterID>&
Endpoint::SnodeBlacklist() const
{
return _state->snode_blacklist;
}
const IntroSet&
Endpoint::intro_set() const
{
return _state->local_introset;
}
IntroSet&
Endpoint::intro_set()
{
return _state->local_introset;
}
const std::unordered_map<ConvoTag, Session>&
Endpoint::Sessions() const
{
return _state->m_Sessions;
}
std::unordered_map<ConvoTag, Session>&
Endpoint::Sessions()
{
return _state->m_Sessions;
}
void
Endpoint::SetAuthInfoForEndpoint(Address addr, AuthInfo info)
{
if (info.token.empty())
{
_remote_auth_infos.erase(addr);
return;
}
_remote_auth_infos[addr] = std::move(info);
}
void
Endpoint::MapExitRange(IPRange range, Address exit)
{
if (not exit.IsZero())
LogInfo(Name(), " map ", range, " to exit at ", exit);
_exit_map.Insert(range, exit);
}
bool
Endpoint::HasFlowToService(Address addr) const
{
return HasOutboundConvo(addr) or HasInboundConvo(addr);
}
void
Endpoint::UnmapExitRange(IPRange range)
{
// unmap all ranges that fit in the range we gave
_exit_map.RemoveIf([&](const auto& item) -> bool {
if (not range.Contains(item.first))
return false;
LogInfo(Name(), " unmap ", item.first, " exit range mapping");
return true;
});
if (_exit_map.Empty())
router()->route_poker()->put_down();
}
void
Endpoint::UnmapRangeByExit(IPRange range, std::string exit)
{
// unmap all ranges that match the given exit when hot swapping
_exit_map.RemoveIf([&](const auto& item) -> bool {
if ((range.Contains(item.first)) and (item.second.ToString() == exit))
{
log::info(logcat, "{} unmap {} range mapping to exit node {}", Name(), item.first, exit);
return true;
}
return false;
});
if (_exit_map.Empty())
router()->route_poker()->put_down();
}
std::optional<AuthInfo>
Endpoint::MaybeGetAuthInfoForEndpoint(Address remote)
{
const auto itr = _remote_auth_infos.find(remote);
if (itr == _remote_auth_infos.end())
return std::nullopt;
return itr->second;
}
link::TunnelManager*
Endpoint::GetQUICTunnel()
{
return _tunnel_manager.get();
}
} // namespace llarp::service