1 files changed, 129 insertions, 0 deletions

diff --git a/src/net_processing.cpp b/src/net_processing.cpp
index 61f98ca747..50ac76924b 100644
--- a/src/net_processing.cpp
+++ b/src/net_processing.cpp
@@ -124,6 +124,9 @@ namespace {
     /** Number of peers from which we're downloading blocks. */
     int nPeersWithValidatedDownloads = 0;
 
+    /** Number of outbound peers with m_chain_sync.m_protect. */
+    int g_outbound_peers_with_protect_from_disconnect = 0;
+
     /** Relay map, protected by cs_main. */
     typedef std::map<uint256, CTransactionRef> MapRelay;
     MapRelay mapRelay;
@@ -201,6 +204,33 @@ struct CNodeState {
      */
     bool fSupportsDesiredCmpctVersion;
 
+    /** State used to enforce CHAIN_SYNC_TIMEOUT
+      * Only in effect for outbound, non-manual connections, with
+      * m_protect == false
+      * Algorithm: if a peer's best known block has less work than our tip,
+      * set a timeout CHAIN_SYNC_TIMEOUT seconds in the future:
+      *   - If at timeout their best known block now has more work than our tip
+      *     when the timeout was set, then either reset the timeout or clear it
+      *     (after comparing against our current tip's work)
+      *   - If at timeout their best known block still has less work than our
+      *     tip did when the timeout was set, then send a getheaders message,
+      *     and set a shorter timeout, HEADERS_RESPONSE_TIME seconds in future.
+      *     If their best known block is still behind when that new timeout is
+      *     reached, disconnect.
+      */
+    struct ChainSyncTimeoutState {
+        //! A timeout used for checking whether our peer has sufficiently synced
+        int64_t m_timeout;
+        //! A header with the work we require on our peer's chain
+        const CBlockIndex * m_work_header;
+        //! After timeout is reached, set to true after sending getheaders
+        bool m_sent_getheaders;
+        //! Whether this peer is protected from disconnection due to a bad/slow chain
+        bool m_protect;
+    };
+
+    ChainSyncTimeoutState m_chain_sync;
+
     CNodeState(CAddress addrIn, std::string addrNameIn) : address(addrIn), name(addrNameIn) {
         fCurrentlyConnected = false;
         nMisbehavior = 0;
@@ -223,6 +253,7 @@ struct CNodeState {
         fHaveWitness = false;
         fWantsCmpctWitness = false;
         fSupportsDesiredCmpctVersion = false;
+        m_chain_sync = { 0, nullptr, false, false };
     }
 };
 
@@ -502,6 +533,13 @@ void FindNextBlocksToDownload(NodeId nodeid, unsigned int count, std::vector<con
 
 } // namespace
 
+// Returns true for outbound peers, excluding manual connections, feelers, and
+// one-shots
+bool IsOutboundDisconnectionCandidate(const CNode *node)
+{
+    return !(node->fInbound || node->m_manual_connection || node->fFeeler || node->fOneShot);
+}
+
 void PeerLogicValidation::InitializeNode(CNode *pnode) {
     CAddress addr = pnode->addr;
     std::string addrName = pnode->GetAddrName();
@@ -534,6 +572,8 @@ void PeerLogicValidation::FinalizeNode(NodeId nodeid, bool& fUpdateConnectionTim
     nPreferredDownload -= state->fPreferredDownload;
     nPeersWithValidatedDownloads -= (state->nBlocksInFlightValidHeaders != 0);
     assert(nPeersWithValidatedDownloads >= 0);
+    g_outbound_peers_with_protect_from_disconnect -= state->m_chain_sync.m_protect;
+    assert(g_outbound_peers_with_protect_from_disconnect >= 0);
 
     mapNodeState.erase(nodeid);
 
@@ -542,6 +582,7 @@ void PeerLogicValidation::FinalizeNode(NodeId nodeid, bool& fUpdateConnectionTim
         assert(mapBlocksInFlight.empty());
         assert(nPreferredDownload == 0);
         assert(nPeersWithValidatedDownloads == 0);
+        assert(g_outbound_peers_with_protect_from_disconnect == 0);
     }
     LogPrint(BCLog::NET, "Cleared nodestate for peer=%d\n", nodeid);
 }
@@ -2337,6 +2378,10 @@ bool static ProcessMessage(CNode* pfrom, const std::string& strCommand, CDataStr
         assert(pindexLast);
         UpdateBlockAvailability(pfrom->GetId(), pindexLast->GetBlockHash());
 
+        // From here, pindexBestKnownBlock should be guaranteed to be non-null,
+        // because it is set in UpdateBlockAvailability. Some nullptr checks
+        // are still present, however, as belt-and-suspenders.
+
         if (nCount == MAX_HEADERS_RESULTS) {
             // Headers message had its maximum size; the peer may have more headers.
             // TODO: optimize: if pindexLast is an ancestor of chainActive.Tip or pindexBestHeader, continue
@@ -2396,6 +2441,35 @@ bool static ProcessMessage(CNode* pfrom, const std::string& strCommand, CDataStr
                 }
             }
         }
+        // If we're in IBD, we want outbound peers that will serve us a useful
+        // chain. Disconnect peers that are on chains with insufficient work.
+        if (IsInitialBlockDownload() && nCount != MAX_HEADERS_RESULTS) {
+            // When nCount < MAX_HEADERS_RESULTS, we know we have no more
+            // headers to fetch from this peer.
+            if (nodestate->pindexBestKnownBlock && nodestate->pindexBestKnownBlock->nChainWork < nMinimumChainWork) {
+                // This peer has too little work on their headers chain to help
+                // us sync -- disconnect if using an outbound slot (unless
+                // whitelisted or addnode).
+                // Note: We compare their tip to nMinimumChainWork (rather than
+                // chainActive.Tip()) because we won't start block download
+                // until we have a headers chain that has at least
+                // nMinimumChainWork, even if a peer has a chain past our tip,
+                // as an anti-DoS measure.
+                if (IsOutboundDisconnectionCandidate(pfrom)) {
+                    LogPrintf("Disconnecting outbound peer %d -- headers chain has insufficient work\n", pfrom->GetId());
+                    pfrom->fDisconnect = true;
+                }
+            }
+        }
+
+        if (!pfrom->fDisconnect && IsOutboundDisconnectionCandidate(pfrom) && nodestate->pindexBestKnownBlock != nullptr) {
+            // If this is an outbound peer, check to see if we should protect
+            // it from the bad/lagging chain logic.
+            if (g_outbound_peers_with_protect_from_disconnect < MAX_OUTBOUND_PEERS_TO_PROTECT_FROM_DISCONNECT && nodestate->pindexBestKnownBlock->nChainWork >= chainActive.Tip()->nChainWork && !nodestate->m_chain_sync.m_protect) {
+                nodestate->m_chain_sync.m_protect = true;
+                ++g_outbound_peers_with_protect_from_disconnect;
+            }
+        }
         }
     }
 
@@ -2794,6 +2868,58 @@ bool PeerLogicValidation::ProcessMessages(CNode* pfrom, std::atomic<bool>& inter
     return fMoreWork;
 }
 
+void PeerLogicValidation::ConsiderEviction(CNode *pto, int64_t time_in_seconds)
+{
+    AssertLockHeld(cs_main);
+
+    CNodeState &state = *State(pto->GetId());
+    const CNetMsgMaker msgMaker(pto->GetSendVersion());
+
+    if (!state.m_chain_sync.m_protect && IsOutboundDisconnectionCandidate(pto) && state.fSyncStarted) {
+        // This is an outbound peer subject to disconnection if they don't
+        // announce a block with as much work as the current tip within
+        // CHAIN_SYNC_TIMEOUT + HEADERS_RESPONSE_TIME seconds (note: if
+        // their chain has more work than ours, we should sync to it,
+        // unless it's invalid, in which case we should find that out and
+        // disconnect from them elsewhere).
+        if (state.pindexBestKnownBlock != nullptr && state.pindexBestKnownBlock->nChainWork >= chainActive.Tip()->nChainWork) {
+            if (state.m_chain_sync.m_timeout != 0) {
+                state.m_chain_sync.m_timeout = 0;
+                state.m_chain_sync.m_work_header = nullptr;
+                state.m_chain_sync.m_sent_getheaders = false;
+            }
+        } else if (state.m_chain_sync.m_timeout == 0 || (state.m_chain_sync.m_work_header != nullptr && state.pindexBestKnownBlock != nullptr && state.pindexBestKnownBlock->nChainWork >= state.m_chain_sync.m_work_header->nChainWork)) {
+            // Our best block known by this peer is behind our tip, and we're either noticing
+            // that for the first time, OR this peer was able to catch up to some earlier point
+            // where we checked against our tip.
+            // Either way, set a new timeout based on current tip.
+            state.m_chain_sync.m_timeout = time_in_seconds + CHAIN_SYNC_TIMEOUT;
+            state.m_chain_sync.m_work_header = chainActive.Tip();
+            state.m_chain_sync.m_sent_getheaders = false;
+        } else if (state.m_chain_sync.m_timeout > 0 && time_in_seconds > state.m_chain_sync.m_timeout) {
+            // No evidence yet that our peer has synced to a chain with work equal to that
+            // of our tip, when we first detected it was behind. Send a single getheaders
+            // message to give the peer a chance to update us.
+            if (state.m_chain_sync.m_sent_getheaders) {
+                // They've run out of time to catch up!
+                LogPrintf("Disconnecting outbound peer %d for old chain, best known block = %s\n", pto->GetId(), state.pindexBestKnownBlock != nullptr ? state.pindexBestKnownBlock->GetBlockHash().ToString() : "<none>");
+                pto->fDisconnect = true;
+            } else {
+                LogPrint(BCLog::NET, "sending getheaders to outbound peer=%d to verify chain work (current best known block:%s, benchmark blockhash: %s)\n", pto->GetId(), state.pindexBestKnownBlock != nullptr ? state.pindexBestKnownBlock->GetBlockHash().ToString() : "<none>", state.m_chain_sync.m_work_header->GetBlockHash().ToString());
+                connman->PushMessage(pto, msgMaker.Make(NetMsgType::GETHEADERS, chainActive.GetLocator(state.m_chain_sync.m_work_header->pprev), uint256()));
+                state.m_chain_sync.m_sent_getheaders = true;
+                constexpr int64_t HEADERS_RESPONSE_TIME = 120; // 2 minutes
+                // Bump the timeout to allow a response, which could clear the timeout
+                // (if the response shows the peer has synced), reset the timeout (if
+                // the peer syncs to the required work but not to our tip), or result
+                // in disconnect (if we advance to the timeout and pindexBestKnownBlock
+                // has not sufficiently progressed)
+                state.m_chain_sync.m_timeout = time_in_seconds + HEADERS_RESPONSE_TIME;
+            }
+        }
+    }
+}
+
 class CompareInvMempoolOrder
 {
     CTxMemPool *mp;
@@ -3260,6 +3386,9 @@ bool PeerLogicValidation::SendMessages(CNode* pto, std::atomic<bool>& interruptM
             }
         }
 
+        // Check that outbound peers have reasonable chains
+        // GetTime() is used by this anti-DoS logic so we can test this using mocktime
+        ConsiderEviction(pto, GetTime());
 
         //
         // Message: getdata (blocks)