Merge #26326: net: don't lock cs_main while reading blocks in net pro…

…cessing

75d27fe net: reduce LOCK(cs_main) scope in ProcessGetBlockData (Andrew Toth)
613a45c net: reduce LOCK(cs_main) scope in GETBLOCKTXN (Andrew Toth)

Pull request description:

  Inspired by #11913 and #26308.

  `cs_main` doesn't need to be locked while reading blocks. This removes the locks in `net_processing`.

ACKs for top commit:
  sr-gi:
    ACK [75d27fe](75d27fe)
  achow101:
    ACK 75d27fe
  furszy:
    ACK 75d27fe with a non-blocking nit.
  mzumsande:
    Code Review ACK 75d27fe
  TheCharlatan:
    ACK 75d27fe

Tree-SHA512: 79b85f748f68ecfb2f2afd3267857dd41b8e76dd482c9c922037399dcbce7b1e5d4c708a4f5fd17c3fb6699b0d88f26a17cc1d92db115dd43c8d4392ae27cec4

src/net_processing.cpp

@@ -118,6 +118,7 @@ static const unsigned int MAX_HEADERS_RESULTS = 2000;
 static const int MAX_CMPCTBLOCK_DEPTH = 5;
 /** Maximum depth of blocks we're willing to respond to GETBLOCKTXN requests for. */
 static const int MAX_BLOCKTXN_DEPTH = 10;
+static_assert(MAX_BLOCKTXN_DEPTH <= MIN_BLOCKS_TO_KEEP, "MAX_BLOCKTXN_DEPTH too high");
 /** Size of the "block download window": how far ahead of our current height do we fetch?
  *  Larger windows tolerate larger download speed differences between peer, but increase the potential
  *  degree of disordering of blocks on disk (which make reindexing and pruning harder). We'll probably
@@ -2420,55 +2421,77 @@ void PeerManagerImpl::ProcessGetBlockData(CNode& pfrom, Peer& peer, const CInv&
         }
     }
 
-    LOCK(cs_main);
-    const CBlockIndex* pindex = m_chainman.m_blockman.LookupBlockIndex(inv.hash);
-    if (!pindex) {
-        return;
-    }
-    if (!BlockRequestAllowed(pindex)) {
-        LogPrint(BCLog::NET, "%s: ignoring request from peer=%i for old block that isn't in the main chain\n", __func__, pfrom.GetId());
-        return;
-    }
-    // disconnect node in case we have reached the outbound limit for serving historical blocks
-    if (m_connman.OutboundTargetReached(true) &&
-        (((m_chainman.m_best_header != nullptr) && (m_chainman.m_best_header->GetBlockTime() - pindex->GetBlockTime() > HISTORICAL_BLOCK_AGE)) || inv.IsMsgFilteredBlk()) &&
-        !pfrom.HasPermission(NetPermissionFlags::Download) // nodes with the download permission may exceed target
-    ) {
-        LogPrint(BCLog::NET, "historical block serving limit reached, disconnect peer=%d\n", pfrom.GetId());
-        pfrom.fDisconnect = true;
-        return;
-    }
-    // Avoid leaking prune-height by never sending blocks below the NODE_NETWORK_LIMITED threshold
-    if (!pfrom.HasPermission(NetPermissionFlags::NoBan) && (
-            (((peer.m_our_services & NODE_NETWORK_LIMITED) == NODE_NETWORK_LIMITED) && ((peer.m_our_services & NODE_NETWORK) != NODE_NETWORK) && (m_chainman.ActiveChain().Tip()->nHeight - pindex->nHeight > (int)NODE_NETWORK_LIMITED_MIN_BLOCKS + 2 /* add two blocks buffer extension for possible races */) )
-       )) {
-        LogPrint(BCLog::NET, "Ignore block request below NODE_NETWORK_LIMITED threshold, disconnect peer=%d\n", pfrom.GetId());
-        //disconnect node and prevent it from stalling (would otherwise wait for the missing block)
-        pfrom.fDisconnect = true;
-        return;
-    }
-    // Pruned nodes may have deleted the block, so check whether
-    // it's available before trying to send.
-    if (!(pindex->nStatus & BLOCK_HAVE_DATA)) {
-        return;
+    const CBlockIndex* pindex{nullptr};
+    const CBlockIndex* tip{nullptr};
+    bool can_direct_fetch{false};
+    FlatFilePos block_pos{};
+    {
+        LOCK(cs_main);
+        pindex = m_chainman.m_blockman.LookupBlockIndex(inv.hash);
+        if (!pindex) {
+            return;
+        }
+        if (!BlockRequestAllowed(pindex)) {
+            LogPrint(BCLog::NET, "%s: ignoring request from peer=%i for old block that isn't in the main chain\n", __func__, pfrom.GetId());
+            return;
+        }
+        // disconnect node in case we have reached the outbound limit for serving historical blocks
+        if (m_connman.OutboundTargetReached(true) &&
+            (((m_chainman.m_best_header != nullptr) && (m_chainman.m_best_header->GetBlockTime() - pindex->GetBlockTime() > HISTORICAL_BLOCK_AGE)) || inv.IsMsgFilteredBlk()) &&
+            !pfrom.HasPermission(NetPermissionFlags::Download) // nodes with the download permission may exceed target
+        ) {
+            LogPrint(BCLog::NET, "historical block serving limit reached, disconnect peer=%d\n", pfrom.GetId());
+            pfrom.fDisconnect = true;
+            return;
+        }
+        tip = m_chainman.ActiveChain().Tip();
+        // Avoid leaking prune-height by never sending blocks below the NODE_NETWORK_LIMITED threshold
+        if (!pfrom.HasPermission(NetPermissionFlags::NoBan) && (
+                (((peer.m_our_services & NODE_NETWORK_LIMITED) == NODE_NETWORK_LIMITED) && ((peer.m_our_services & NODE_NETWORK) != NODE_NETWORK) && (tip->nHeight - pindex->nHeight > (int)NODE_NETWORK_LIMITED_MIN_BLOCKS + 2 /* add two blocks buffer extension for possible races */) )
+           )) {
+            LogPrint(BCLog::NET, "Ignore block request below NODE_NETWORK_LIMITED threshold, disconnect peer=%d\n", pfrom.GetId());
+            //disconnect node and prevent it from stalling (would otherwise wait for the missing block)
+            pfrom.fDisconnect = true;
+            return;
+        }
+        // Pruned nodes may have deleted the block, so check whether
+        // it's available before trying to send.
+        if (!(pindex->nStatus & BLOCK_HAVE_DATA)) {
+            return;
+        }
+        can_direct_fetch = CanDirectFetch();
+        block_pos = pindex->GetBlockPos();
     }
+
     std::shared_ptr<const CBlock> pblock;
     if (a_recent_block && a_recent_block->GetHash() == pindex->GetBlockHash()) {
         pblock = a_recent_block;
     } else if (inv.IsMsgWitnessBlk()) {
         // Fast-path: in this case it is possible to serve the block directly from disk,
         // as the network format matches the format on disk
         std::vector<uint8_t> block_data;
-        if (!m_chainman.m_blockman.ReadRawBlockFromDisk(block_data, pindex->GetBlockPos())) {
-            assert(!"cannot load block from disk");
+        if (!m_chainman.m_blockman.ReadRawBlockFromDisk(block_data, block_pos)) {
+            if (WITH_LOCK(m_chainman.GetMutex(), return m_chainman.m_blockman.IsBlockPruned(*pindex))) {
+                LogPrint(BCLog::NET, "Block was pruned before it could be read, disconnect peer=%s\n", pfrom.GetId());
+            } else {
+                LogError("Cannot load block from disk, disconnect peer=%d\n", pfrom.GetId());
+            }
+            pfrom.fDisconnect = true;
+            return;
         }
         MakeAndPushMessage(pfrom, NetMsgType::BLOCK, Span{block_data});
         // Don't set pblock as we've sent the block
     } else {
         // Send block from disk
         std::shared_ptr<CBlock> pblockRead = std::make_shared<CBlock>();
-        if (!m_chainman.m_blockman.ReadBlockFromDisk(*pblockRead, *pindex)) {
-            assert(!"cannot load block from disk");
+        if (!m_chainman.m_blockman.ReadBlockFromDisk(*pblockRead, block_pos)) {
+            if (WITH_LOCK(m_chainman.GetMutex(), return m_chainman.m_blockman.IsBlockPruned(*pindex))) {
+                LogPrint(BCLog::NET, "Block was pruned before it could be read, disconnect peer=%s\n", pfrom.GetId());
+            } else {
+                LogError("Cannot load block from disk, disconnect peer=%d\n", pfrom.GetId());
+            }
+            pfrom.fDisconnect = true;
+            return;
         }
         pblock = pblockRead;
     }
@@ -2506,7 +2529,7 @@ void PeerManagerImpl::ProcessGetBlockData(CNode& pfrom, Peer& peer, const CInv&
             // they won't have a useful mempool to match against a compact block,
             // and we don't feel like constructing the object for them, so
             // instead we respond with the full, non-compact block.
-            if (CanDirectFetch() && pindex->nHeight >= m_chainman.ActiveChain().Height() - MAX_CMPCTBLOCK_DEPTH) {
+            if (can_direct_fetch && pindex->nHeight >= tip->nHeight - MAX_CMPCTBLOCK_DEPTH) {
                 if (a_recent_compact_block && a_recent_compact_block->header.GetHash() == pindex->GetBlockHash()) {
                     MakeAndPushMessage(pfrom, NetMsgType::CMPCTBLOCK, *a_recent_compact_block);
                 } else {
@@ -2527,7 +2550,7 @@ void PeerManagerImpl::ProcessGetBlockData(CNode& pfrom, Peer& peer, const CInv&
             // and we want it right after the last block so they don't
             // wait for other stuff first.
             std::vector<CInv> vInv;
-            vInv.emplace_back(MSG_BLOCK, m_chainman.ActiveChain().Tip()->GetBlockHash());
+            vInv.emplace_back(MSG_BLOCK, tip->GetBlockHash());
             MakeAndPushMessage(pfrom, NetMsgType::INV, vInv);
             peer.m_continuation_block.SetNull();
         }
@@ -4366,6 +4389,7 @@ void PeerManagerImpl::ProcessMessage(CNode& pfrom, const std::string& msg_type,
             return;
         }
 
+        FlatFilePos block_pos{};
         {
             LOCK(cs_main);
 
@@ -4376,15 +4400,21 @@ void PeerManagerImpl::ProcessMessage(CNode& pfrom, const std::string& msg_type,
             }
 
             if (pindex->nHeight >= m_chainman.ActiveChain().Height() - MAX_BLOCKTXN_DEPTH) {
-                CBlock block;
-                const bool ret{m_chainman.m_blockman.ReadBlockFromDisk(block, *pindex)};
-                assert(ret);
-
-                SendBlockTransactions(pfrom, *peer, block, req);
-                return;
+                block_pos = pindex->GetBlockPos();
             }
         }
 
+        if (!block_pos.IsNull()) {
+            CBlock block;
+            const bool ret{m_chainman.m_blockman.ReadBlockFromDisk(block, block_pos)};
+            // If height is above MAX_BLOCKTXN_DEPTH then this block cannot get
+            // pruned after we release cs_main above, so this read should never fail.
+            assert(ret);
+
+            SendBlockTransactions(pfrom, *peer, block, req);
+            return;
+        }
+
         // If an older block is requested (should never happen in practice,
         // but can happen in tests) send a block response instead of a
         // blocktxn response. Sending a full block response instead of a