dash/src/llmq/quorums_dkgsessionhandler.cpp
Alexander Block 26fcd3f0bf Also remove m_tx_announced and m_tx_in_flight entries when EraseObjectRequest is called
Otherwise they'll run into false-positive timeouts.
2020-04-07 13:25:29 +02:00

598 lines
22 KiB
C++

// Copyright (c) 2018-2019 The Dash Core developers
// Distributed under the MIT/X11 software license, see the accompanying
// file COPYING or http://www.opensource.org/licenses/mit-license.php.
#include <llmq/quorums_dkgsessionhandler.h>
#include <llmq/quorums_blockprocessor.h>
#include <llmq/quorums_debug.h>
#include <llmq/quorums_init.h>
#include <llmq/quorums_utils.h>
#include <masternode/activemasternode.h>
#include <chainparams.h>
#include <init.h>
#include <net_processing.h>
#include <spork.h>
#include <validation.h>
namespace llmq
{
CDKGPendingMessages::CDKGPendingMessages(size_t _maxMessagesPerNode, int _invType) :
maxMessagesPerNode(_maxMessagesPerNode),
invType(_invType)
{
}
void CDKGPendingMessages::PushPendingMessage(NodeId from, CDataStream& vRecv)
{
// this will also consume the data, even if we bail out early
auto pm = std::make_shared<CDataStream>(std::move(vRecv));
{
LOCK(cs);
if (messagesPerNode[from] >= maxMessagesPerNode) {
// TODO ban?
LogPrint(BCLog::LLMQ_DKG, "CDKGPendingMessages::%s -- too many messages, peer=%d\n", __func__, from);
return;
}
messagesPerNode[from]++;
}
CHashWriter hw(SER_GETHASH, 0);
hw.write(pm->data(), pm->size());
uint256 hash = hw.GetHash();
LOCK2(cs_main, cs);
if (!seenMessages.emplace(hash).second) {
LogPrint(BCLog::LLMQ_DKG, "CDKGPendingMessages::%s -- already seen %s, peer=%d\n", __func__, hash.ToString(), from);
return;
}
EraseObjectRequest(from, CInv(invType, hash));
pendingMessages.emplace_back(std::make_pair(from, std::move(pm)));
}
std::list<CDKGPendingMessages::BinaryMessage> CDKGPendingMessages::PopPendingMessages(size_t maxCount)
{
LOCK(cs);
std::list<BinaryMessage> ret;
while (!pendingMessages.empty() && ret.size() < maxCount) {
ret.emplace_back(std::move(pendingMessages.front()));
pendingMessages.pop_front();
}
return std::move(ret);
}
bool CDKGPendingMessages::HasSeen(const uint256& hash) const
{
LOCK(cs);
return seenMessages.count(hash) != 0;
}
void CDKGPendingMessages::Clear()
{
LOCK(cs);
pendingMessages.clear();
messagesPerNode.clear();
seenMessages.clear();
}
//////
CDKGSessionHandler::CDKGSessionHandler(const Consensus::LLMQParams& _params, ctpl::thread_pool& _messageHandlerPool, CBLSWorker& _blsWorker, CDKGSessionManager& _dkgManager) :
params(_params),
messageHandlerPool(_messageHandlerPool),
blsWorker(_blsWorker),
dkgManager(_dkgManager),
curSession(std::make_shared<CDKGSession>(_params, _blsWorker, _dkgManager)),
pendingContributions((size_t)_params.size * 2, MSG_QUORUM_CONTRIB), // we allow size*2 messages as we need to make sure we see bad behavior (double messages)
pendingComplaints((size_t)_params.size * 2, MSG_QUORUM_COMPLAINT),
pendingJustifications((size_t)_params.size * 2, MSG_QUORUM_JUSTIFICATION),
pendingPrematureCommitments((size_t)_params.size * 2, MSG_QUORUM_PREMATURE_COMMITMENT)
{
phaseHandlerThread = std::thread([this] {
RenameThread(strprintf("dash-q-phase-%d", (uint8_t)params.type).c_str());
PhaseHandlerThread();
});
}
CDKGSessionHandler::~CDKGSessionHandler()
{
stopRequested = true;
if (phaseHandlerThread.joinable()) {
phaseHandlerThread.join();
}
}
void CDKGSessionHandler::UpdatedBlockTip(const CBlockIndex* pindexNew)
{
LOCK(cs);
int quorumStageInt = pindexNew->nHeight % params.dkgInterval;
const CBlockIndex* pindexQuorum = pindexNew->GetAncestor(pindexNew->nHeight - quorumStageInt);
currentHeight = pindexNew->nHeight;
quorumHeight = pindexQuorum->nHeight;
quorumHash = pindexQuorum->GetBlockHash();
bool fNewPhase = (quorumStageInt % params.dkgPhaseBlocks) == 0;
int phaseInt = quorumStageInt / params.dkgPhaseBlocks + 1;
QuorumPhase oldPhase = phase;
if (fNewPhase && phaseInt >= QuorumPhase_Initialized && phaseInt <= QuorumPhase_Idle) {
phase = static_cast<QuorumPhase>(phaseInt);
}
LogPrint(BCLog::LLMQ_DKG, "CDKGSessionHandler::%s -- %s - currentHeight=%d, quorumHeight=%d, oldPhase=%d, newPhase=%d\n", __func__,
params.name, currentHeight, quorumHeight, oldPhase, phase);
}
void CDKGSessionHandler::ProcessMessage(CNode* pfrom, const std::string& strCommand, CDataStream& vRecv, CConnman& connman)
{
// We don't handle messages in the calling thread as deserialization/processing of these would block everything
if (strCommand == NetMsgType::QCONTRIB) {
pendingContributions.PushPendingMessage(pfrom->GetId(), vRecv);
} else if (strCommand == NetMsgType::QCOMPLAINT) {
pendingComplaints.PushPendingMessage(pfrom->GetId(), vRecv);
} else if (strCommand == NetMsgType::QJUSTIFICATION) {
pendingJustifications.PushPendingMessage(pfrom->GetId(), vRecv);
} else if (strCommand == NetMsgType::QPCOMMITMENT) {
pendingPrematureCommitments.PushPendingMessage(pfrom->GetId(), vRecv);
}
}
bool CDKGSessionHandler::InitNewQuorum(const CBlockIndex* pindexQuorum)
{
//AssertLockHeld(cs_main);
const auto& consensus = Params().GetConsensus();
curSession = std::make_shared<CDKGSession>(params, blsWorker, dkgManager);
if (!deterministicMNManager->IsDIP3Enforced(pindexQuorum->nHeight)) {
return false;
}
auto mns = CLLMQUtils::GetAllQuorumMembers(params.type, pindexQuorum);
if (!curSession->Init(pindexQuorum, mns, activeMasternodeInfo.proTxHash)) {
LogPrintf("CDKGSessionManager::%s -- quorum initialiation failed for %s\n", __func__, curSession->params.name);
return false;
}
return true;
}
std::pair<QuorumPhase, uint256> CDKGSessionHandler::GetPhaseAndQuorumHash() const
{
LOCK(cs);
return std::make_pair(phase, quorumHash);
}
class AbortPhaseException : public std::exception {
};
void CDKGSessionHandler::WaitForNextPhase(QuorumPhase curPhase,
QuorumPhase nextPhase,
const uint256& expectedQuorumHash,
const WhileWaitFunc& runWhileWaiting)
{
LogPrint(BCLog::LLMQ_DKG, "CDKGSessionManager::%s -- %s - starting, curPhase=%d, nextPhase=%d\n", __func__, params.name, curPhase, nextPhase);
while (true) {
if (stopRequested || ShutdownRequested()) {
LogPrint(BCLog::LLMQ_DKG, "CDKGSessionManager::%s -- %s - aborting due to stop/shutdown requested\n", __func__, params.name);
throw AbortPhaseException();
}
auto p = GetPhaseAndQuorumHash();
if (!expectedQuorumHash.IsNull() && p.second != expectedQuorumHash) {
LogPrint(BCLog::LLMQ_DKG, "CDKGSessionManager::%s -- %s - aborting due unexpected expectedQuorumHash change\n", __func__, params.name);
throw AbortPhaseException();
}
if (p.first == nextPhase) {
break;
}
if (curPhase != QuorumPhase_None && p.first != curPhase) {
LogPrint(BCLog::LLMQ_DKG, "CDKGSessionManager::%s -- %s - aborting due unexpected phase change\n", __func__, params.name);
throw AbortPhaseException();
}
if (!runWhileWaiting()) {
MilliSleep(100);
}
}
LogPrint(BCLog::LLMQ_DKG, "CDKGSessionManager::%s -- %s - done, curPhase=%d, nextPhase=%d\n", __func__, params.name, curPhase, nextPhase);
if (nextPhase == QuorumPhase_Initialized) {
quorumDKGDebugManager->ResetLocalSessionStatus(params.type);
} else {
quorumDKGDebugManager->UpdateLocalSessionStatus(params.type, [&](CDKGDebugSessionStatus& status) {
bool changed = status.phase != (uint8_t) nextPhase;
status.phase = (uint8_t) nextPhase;
return changed;
});
}
}
void CDKGSessionHandler::WaitForNewQuorum(const uint256& oldQuorumHash)
{
LogPrint(BCLog::LLMQ_DKG, "CDKGSessionManager::%s -- %s - starting\n", __func__, params.name);
while (true) {
if (stopRequested || ShutdownRequested()) {
LogPrint(BCLog::LLMQ_DKG, "CDKGSessionManager::%s -- %s - aborting due to stop/shutdown requested\n", __func__, params.name);
throw AbortPhaseException();
}
auto p = GetPhaseAndQuorumHash();
if (p.second != oldQuorumHash) {
break;
}
MilliSleep(100);
}
LogPrint(BCLog::LLMQ_DKG, "CDKGSessionManager::%s -- %s - done\n", __func__, params.name);
}
// Sleep some time to not fully overload the whole network
void CDKGSessionHandler::SleepBeforePhase(QuorumPhase curPhase,
const uint256& expectedQuorumHash,
double randomSleepFactor,
const WhileWaitFunc& runWhileWaiting)
{
if (!curSession->AreWeMember()) {
// Non-members do not participate and do not create any network load, no need to sleep.
return;
}
if (Params().MineBlocksOnDemand()) {
// On regtest, blocks can be mined on demand without any significant time passing between these.
// We shouldn't wait before phases in this case.
return;
}
// Two blocks can come very close to each other, this happens pretty regularly. We don't want to be
// left behind and marked as a bad member. This means that we should not count the last block of the
// phase as a safe one to keep sleeping, that's why we calculate the phase sleep time as a time of
// the full phase minus one block here.
double phaseSleepTime = (params.dkgPhaseBlocks - 1) * Params().GetConsensus().nPowTargetSpacing * 1000;
// Expected phase sleep time per member
double phaseSleepTimePerMember = phaseSleepTime / params.size;
// Don't expect perfect block times and thus reduce the phase time to be on the secure side (caller chooses factor)
double adjustedPhaseSleepTimePerMember = phaseSleepTimePerMember * randomSleepFactor;
int64_t sleepTime = (int64_t)(adjustedPhaseSleepTimePerMember * curSession->GetMyMemberIndex());
int64_t endTime = GetTimeMillis() + sleepTime;
int heightTmp{-1};
int heightStart{-1};
{
LOCK(cs);
heightTmp = heightStart = currentHeight;
}
LogPrint(BCLog::LLMQ_DKG, "CDKGSessionManager::%s -- %s - starting sleep for %d ms, curPhase=%d\n", __func__, params.name, sleepTime, curPhase);
while (GetTimeMillis() < endTime) {
if (stopRequested || ShutdownRequested()) {
LogPrint(BCLog::LLMQ_DKG, "CDKGSessionManager::%s -- %s - aborting due to stop/shutdown requested\n", __func__, params.name);
throw AbortPhaseException();
}
{
LOCK(cs);
if (currentHeight > heightTmp) {
// New block(s) just came in
int64_t expectedBlockTime = (currentHeight - heightStart) * Params().GetConsensus().nPowTargetSpacing * 1000;
if (expectedBlockTime > sleepTime) {
// Blocks came faster than we expected, jump into the phase func asap
break;
}
heightTmp = currentHeight;
}
if (phase != curPhase || quorumHash != expectedQuorumHash) {
// Smth went wrong and/or we missed quite a few blocks and it's just too late now
LogPrint(BCLog::LLMQ_DKG, "CDKGSessionManager::%s -- %s - aborting due unexpected phase/expectedQuorumHash change\n", __func__, params.name);
throw AbortPhaseException();
}
}
if (!runWhileWaiting()) {
MilliSleep(100);
}
}
LogPrint(BCLog::LLMQ_DKG, "CDKGSessionManager::%s -- %s - done, curPhase=%d\n", __func__, params.name, curPhase);
}
void CDKGSessionHandler::HandlePhase(QuorumPhase curPhase,
QuorumPhase nextPhase,
const uint256& expectedQuorumHash,
double randomSleepFactor,
const StartPhaseFunc& startPhaseFunc,
const WhileWaitFunc& runWhileWaiting)
{
LogPrint(BCLog::LLMQ_DKG, "CDKGSessionManager::%s -- %s - starting, curPhase=%d, nextPhase=%d\n", __func__, params.name, curPhase, nextPhase);
SleepBeforePhase(curPhase, expectedQuorumHash, randomSleepFactor, runWhileWaiting);
startPhaseFunc();
WaitForNextPhase(curPhase, nextPhase, expectedQuorumHash, runWhileWaiting);
LogPrint(BCLog::LLMQ_DKG, "CDKGSessionManager::%s -- %s - done, curPhase=%d, nextPhase=%d\n", __func__, params.name, curPhase, nextPhase);
}
// returns a set of NodeIds which sent invalid messages
template<typename Message>
std::set<NodeId> BatchVerifyMessageSigs(CDKGSession& session, const std::vector<std::pair<NodeId, std::shared_ptr<Message>>>& messages)
{
if (messages.empty()) {
return {};
}
std::set<NodeId> ret;
bool revertToSingleVerification = false;
CBLSSignature aggSig;
std::vector<CBLSPublicKey> pubKeys;
std::vector<uint256> messageHashes;
std::set<uint256> messageHashesSet;
pubKeys.reserve(messages.size());
messageHashes.reserve(messages.size());
bool first = true;
for (const auto& p : messages ) {
const auto& msg = *p.second;
auto member = session.GetMember(msg.proTxHash);
if (!member) {
// should not happen as it was verified before
ret.emplace(p.first);
continue;
}
if (first) {
aggSig = msg.sig;
} else {
aggSig.AggregateInsecure(msg.sig);
}
first = false;
auto msgHash = msg.GetSignHash();
if (!messageHashesSet.emplace(msgHash).second) {
// can only happen in 2 cases:
// 1. Someone sent us the same message twice but with differing signature, meaning that at least one of them
// must be invalid. In this case, we'd have to revert to single message verification nevertheless
// 2. Someone managed to find a way to create two different binary representations of a message that deserializes
// to the same object representation. This would be some form of malleability. However, this shouldn't be
// possible as only deterministic/unique BLS signatures and very simple data types are involved
revertToSingleVerification = true;
break;
}
pubKeys.emplace_back(member->dmn->pdmnState->pubKeyOperator.Get());
messageHashes.emplace_back(msgHash);
}
if (!revertToSingleVerification) {
bool valid = aggSig.VerifyInsecureAggregated(pubKeys, messageHashes);
if (valid) {
// all good
return ret;
}
// are all messages from the same node?
NodeId firstNodeId;
first = true;
bool nodeIdsAllSame = true;
for (auto it = messages.begin(); it != messages.end(); ++it) {
if (first) {
firstNodeId = it->first;
} else {
first = false;
if (it->first != firstNodeId) {
nodeIdsAllSame = false;
break;
}
}
}
// if yes, take a short path and return a set with only him
if (nodeIdsAllSame) {
ret.emplace(firstNodeId);
return ret;
}
// different nodes, let's figure out who are the bad ones
}
for (const auto& p : messages) {
if (ret.count(p.first)) {
continue;
}
const auto& msg = *p.second;
auto member = session.GetMember(msg.proTxHash);
bool valid = msg.sig.VerifyInsecure(member->dmn->pdmnState->pubKeyOperator.Get(), msg.GetSignHash());
if (!valid) {
ret.emplace(p.first);
}
}
return ret;
}
template<typename Message, int MessageType>
bool ProcessPendingMessageBatch(CDKGSession& session, CDKGPendingMessages& pendingMessages, size_t maxCount)
{
auto msgs = pendingMessages.PopAndDeserializeMessages<Message>(maxCount);
if (msgs.empty()) {
return false;
}
std::vector<uint256> hashes;
std::vector<std::pair<NodeId, std::shared_ptr<Message>>> preverifiedMessages;
hashes.reserve(msgs.size());
preverifiedMessages.reserve(msgs.size());
for (const auto& p : msgs) {
if (!p.second) {
LogPrint(BCLog::LLMQ_DKG, "%s -- failed to deserialize message, peer=%d\n", __func__, p.first);
{
LOCK(cs_main);
Misbehaving(p.first, 100);
}
continue;
}
const auto& msg = *p.second;
auto hash = ::SerializeHash(msg);
{
LOCK(cs_main);
EraseObjectRequest(p.first, CInv(MessageType, hash));
}
bool ban = false;
if (!session.PreVerifyMessage(hash, msg, ban)) {
if (ban) {
LogPrint(BCLog::LLMQ_DKG, "%s -- banning node due to failed preverification, peer=%d\n", __func__, p.first);
{
LOCK(cs_main);
Misbehaving(p.first, 100);
}
}
LogPrint(BCLog::LLMQ_DKG, "%s -- skipping message due to failed preverification, peer=%d\n", __func__, p.first);
continue;
}
hashes.emplace_back(hash);
preverifiedMessages.emplace_back(p);
}
if (preverifiedMessages.empty()) {
return true;
}
auto badNodes = BatchVerifyMessageSigs(session, preverifiedMessages);
if (!badNodes.empty()) {
LOCK(cs_main);
for (auto nodeId : badNodes) {
LogPrint(BCLog::LLMQ_DKG, "%s -- failed to verify signature, peer=%d\n", __func__, nodeId);
Misbehaving(nodeId, 100);
}
}
for (size_t i = 0; i < preverifiedMessages.size(); i++) {
NodeId nodeId = preverifiedMessages[i].first;
if (badNodes.count(nodeId)) {
continue;
}
const auto& msg = *preverifiedMessages[i].second;
bool ban = false;
session.ReceiveMessage(hashes[i], msg, ban);
if (ban) {
LogPrint(BCLog::LLMQ_DKG, "%s -- banning node after ReceiveMessage failed, peer=%d\n", __func__, nodeId);
LOCK(cs_main);
Misbehaving(nodeId, 100);
badNodes.emplace(nodeId);
}
}
return true;
}
void CDKGSessionHandler::HandleDKGRound()
{
uint256 curQuorumHash;
int curQuorumHeight;
WaitForNextPhase(QuorumPhase_None, QuorumPhase_Initialized, uint256(), []{return false;});
{
LOCK(cs);
pendingContributions.Clear();
pendingComplaints.Clear();
pendingJustifications.Clear();
pendingPrematureCommitments.Clear();
curQuorumHash = quorumHash;
curQuorumHeight = quorumHeight;
}
const CBlockIndex* pindexQuorum;
{
LOCK(cs_main);
pindexQuorum = mapBlockIndex.at(curQuorumHash);
}
if (!InitNewQuorum(pindexQuorum)) {
// should actually never happen
WaitForNewQuorum(curQuorumHash);
throw AbortPhaseException();
}
quorumDKGDebugManager->UpdateLocalSessionStatus(params.type, [&](CDKGDebugSessionStatus& status) {
bool changed = status.phase != (uint8_t) QuorumPhase_Initialized;
status.phase = (uint8_t) QuorumPhase_Initialized;
return changed;
});
CLLMQUtils::EnsureQuorumConnections(params.type, pindexQuorum, curSession->myProTxHash, gArgs.GetBoolArg("-watchquorums", DEFAULT_WATCH_QUORUMS));
if (curSession->AreWeMember() && sporkManager.IsSporkActive(SPORK_21_QUORUM_ALL_CONNECTED)) {
CLLMQUtils::AddQuorumProbeConnections(params.type, pindexQuorum, curSession->myProTxHash);
}
WaitForNextPhase(QuorumPhase_Initialized, QuorumPhase_Contribute, curQuorumHash, []{return false;});
// Contribute
auto fContributeStart = [this]() {
curSession->Contribute(pendingContributions);
};
auto fContributeWait = [this] {
return ProcessPendingMessageBatch<CDKGContribution, MSG_QUORUM_CONTRIB>(*curSession, pendingContributions, 8);
};
HandlePhase(QuorumPhase_Contribute, QuorumPhase_Complain, curQuorumHash, 0.05, fContributeStart, fContributeWait);
// Complain
auto fComplainStart = [this]() {
curSession->VerifyAndComplain(pendingComplaints);
};
auto fComplainWait = [this] {
return ProcessPendingMessageBatch<CDKGComplaint, MSG_QUORUM_COMPLAINT>(*curSession, pendingComplaints, 8);
};
HandlePhase(QuorumPhase_Complain, QuorumPhase_Justify, curQuorumHash, 0.05, fComplainStart, fComplainWait);
// Justify
auto fJustifyStart = [this]() {
curSession->VerifyAndJustify(pendingJustifications);
};
auto fJustifyWait = [this] {
return ProcessPendingMessageBatch<CDKGJustification, MSG_QUORUM_JUSTIFICATION>(*curSession, pendingJustifications, 8);
};
HandlePhase(QuorumPhase_Justify, QuorumPhase_Commit, curQuorumHash, 0.05, fJustifyStart, fJustifyWait);
// Commit
auto fCommitStart = [this]() {
curSession->VerifyAndCommit(pendingPrematureCommitments);
};
auto fCommitWait = [this] {
return ProcessPendingMessageBatch<CDKGPrematureCommitment, MSG_QUORUM_PREMATURE_COMMITMENT>(*curSession, pendingPrematureCommitments, 8);
};
HandlePhase(QuorumPhase_Commit, QuorumPhase_Finalize, curQuorumHash, 0.1, fCommitStart, fCommitWait);
auto finalCommitments = curSession->FinalizeCommitments();
for (const auto& fqc : finalCommitments) {
quorumBlockProcessor->AddMinableCommitment(fqc);
}
}
void CDKGSessionHandler::PhaseHandlerThread()
{
while (!stopRequested && !ShutdownRequested()) {
try {
LogPrint(BCLog::LLMQ_DKG, "CDKGSessionHandler::%s -- %s - starting HandleDKGRound\n", __func__, params.name);
HandleDKGRound();
} catch (AbortPhaseException& e) {
quorumDKGDebugManager->UpdateLocalSessionStatus(params.type, [&](CDKGDebugSessionStatus& status) {
status.aborted = true;
return true;
});
LogPrint(BCLog::LLMQ_DKG, "CDKGSessionHandler::%s -- %s - aborted current DKG session\n", __func__, params.name);
}
}
}
} // namespace llmq