Merge branch 'develop' into ximinez/fix-getledger

* XRPLF/develop:
  chore: Run clang_tidy_check with `pass_filenames: false` from pre-commit (7800)
  feat: Add delegate filter param for account_tx RPC (6126)
  refactor: Move `jss.h` `include` out of `Indexes.h` (7799)
  test: Add tests for check doxygen style (7795)
  style: Add pre-commit hook to check doxygen style (7794)
  style: Unify style for all Doxygen comments (7776)
  fix: Improve Number addition/subtraction rounding (7369)
  feat: XLS-68: Sponsor, 5887 continuation  (7350)

include/xrpl/basics/CanProcess.h

@@ -0,0 +1,113 @@
+#pragma once
+
+#include <functional>
+#include <mutex>
+#include <set>
+
+/**
+ * RAII class to check if an Item is already being processed on another thread,
+ * as indicated by it's presence in a Collection.
+ *
+ * If the Item is not in the Collection, it will be added under lock in the
+ * ctor, and removed under lock in the dtor. The object will be considered
+ * "usable" and evaluate to `true`.
+ *
+ * If the Item is in the Collection, no changes will be made to the collection,
+ * and the CanProcess object will be considered "unusable".
+ *
+ * It's up to the caller to decide what "usable" and "unusable" mean. (e.g.
+ * Process or skip a block of code, or set a flag.)
+ *
+ * The current use is to avoid lock contention that would be involved in
+ * processing something associated with the Item.
+ *
+ * Examples:
+ *
+ * void IncomingLedgers::acquireAsync(LedgerHash const& hash, ...)
+ * {
+ *    if (CanProcess check{acquiresMutex_, pendingAcquires_, hash})
+ *    {
+ *      acquire(hash, ...);
+ *    }
+ * }
+ *
+ * bool
+ * NetworkOPsImp::recvValidation(
+ *   std::shared_ptr<STValidation> const& val,
+ *   std::string const& source)
+ *   {
+ *       CanProcess check(
+ *           validationsMutex_, pendingValidations_, val->getLedgerHash());
+ *       BypassAccept bypassAccept =
+ *           check ? BypassAccept::no : BypassAccept::yes;
+ *       handleNewValidation(app_, val, source, bypassAccept, m_journal);
+ *   }
+ *
+ */
+class CanProcess
+{
+public:
+    template <class Mutex, class Collection, class Item>
+    CanProcess(Mutex& mtx, Collection& collection, Item const& item)
+        : cleanup_(insert(mtx, collection, item))
+    {
+    }
+
+    ~CanProcess()
+    {
+        if (cleanup_)
+            cleanup_();
+    }
+
+    explicit
+    operator bool() const
+    {
+        return static_cast<bool>(cleanup_);
+    }
+
+private:
+    template <bool useIterator, class Mutex, class Collection, class Item>
+    std::function<void()>
+    doInsert(Mutex& mtx, Collection& collection, Item const& item)
+    {
+        std::unique_lock<Mutex> lock(mtx);
+        // TODO: Use structured binding once LLVM 16 is the minimum supported
+        // version. See also: https://github.com/llvm/llvm-project/issues/48582
+        // https://github.com/llvm/llvm-project/commit/127bf44385424891eb04cff8e52d3f157fc2cb7c
+        auto const insertResult = collection.insert(item);
+        auto const it = insertResult.first;
+        if (!insertResult.second)
+            return {};
+        if constexpr (useIterator)
+            return [&, it]() {
+                std::unique_lock<Mutex> lock(mtx);
+                collection.erase(it);
+            };
+        else
+            return [&]() {
+                std::unique_lock<Mutex> lock(mtx);
+                collection.erase(item);
+            };
+    }
+
+    // Generic insert() function doesn't use iterators because they may get
+    // invalidated
+    template <class Mutex, class Collection, class Item>
+    std::function<void()>
+    insert(Mutex& mtx, Collection& collection, Item const& item)
+    {
+        return doInsert<false>(mtx, collection, item);
+    }
+
+    // Specialize insert() for std::set, which does not invalidate iterators for
+    // insert and erase
+    template <class Mutex, class Item>
+    std::function<void()>
+    insert(Mutex& mtx, std::set<Item>& collection, Item const& item)
+    {
+        return doInsert<true>(mtx, collection, item);
+    }
+
+    // If set, then the item is "usable"
+    std::function<void()> cleanup_;
+};

include/xrpl/beast/hash/hash_append.h

@@ -44,7 +44,7 @@ template <class T>
 inline void
 maybeReverseBytes(T& t, std::true_type)
 {
-    reverse_bytes(t);
+    reverseBytes(t);
 }
 
 template <class T, class Hasher>

include/xrpl/core/HashRouter.h

@@ -154,6 +154,15 @@ private:
             return std::move(peers_);
         }
 
+        /**
+         * Return set of peers waiting for reply. Leaves list unchanged.
+         */
+        std::set<PeerShortID> const&
+        peekPeerSet()
+        {
+            return peers_;
+        }
+
         /**
          * Return seated relay time point if the message has been relayed
          */
@@ -188,6 +197,20 @@ private:
             return true;
         }
 
+        bool
+        shouldProcessForPeer(
+            PeerShortID peer,
+            Stopwatch::time_point now,
+            std::chrono::seconds interval)
+        {
+            if (peerProcessed_.contains(peer) && ((peerProcessed_[peer] + interval) > now))
+                return false;
+            // Peer may already be in the list, but adding it again doesn't hurt
+            addPeer(peer);
+            peerProcessed_[peer] = now;
+            return true;
+        }
+
     private:
         HashRouterFlags flags_ = HashRouterFlags::UNDEFINED;
         std::set<PeerShortID> peers_;
@@ -195,6 +218,7 @@ private:
         // than one flag needs to expire independently.
         std::optional<Stopwatch::time_point> relayed_;
         std::optional<Stopwatch::time_point> processed_;
+        std::map<PeerShortID, Stopwatch::time_point> peerProcessed_;
     };
 
 public:
@@ -218,7 +242,7 @@ public:
     /**
      * Add a suppression peer and get message's relay status.
      * Return pair:
-     * element 1: true if the peer is added.
+     * element 1: true if the key is added.
      * element 2: optional is seated to the relay time point or
      * is unseated if has not relayed yet.
      */
@@ -236,6 +260,16 @@ public:
         HashRouterFlags& flags,
         std::chrono::seconds txInterval);
 
+    /**
+     * Determines whether the hashed item should be processed for the given peer. Could be an
+     * incoming or outgoing message.
+     *
+     * Items filtered with this function should only be processed for the given peer once. Unlike
+     * shouldProcess, it can be processed for other peers.
+     */
+    bool
+    shouldProcessForPeer(uint256 const& key, PeerShortID peer, std::chrono::seconds interval);
+
     /**
      * Set the flags on a hash.
      *
@@ -263,6 +297,12 @@ public:
     std::optional<std::set<PeerShortID>>
     shouldRelay(uint256 const& key);
 
+    /**
+     * Returns a copy of the set of peers in the Entry for the key
+     */
+    std::set<PeerShortID>
+    getPeers(uint256 const& key);
+
 private:
     // pair.second indicates whether the entry was created
     std::pair<Entry&, bool>

include/xrpl/proto/xrpl.proto

@@ -288,8 +288,18 @@ message TMLedgerData {
   required uint32 ledgerSeq = 2;
   required TMLedgerInfoType type = 3;
   repeated TMLedgerNode nodes = 4;
+  // If the peer supports "responseCookies", this field will
+  // never be populated.
   optional uint32 requestCookie = 5;
   optional TMReplyError error = 6;
+  // The old field is called "requestCookie", but this is
+  // a response, so this name makes more sense
+  repeated uint32 responseCookies = 7;
+  // If a TMGetLedger request was received without a "requestCookie",
+  // and the peer supports it, this flag will be set to true to
+  // indicate that the receiver should process the result in addition
+  // to forwarding it to its "responseCookies" peers.
+  optional bool directResponse = 8;
 }
 
 message TMPing {

include/xrpl/protocol/LedgerHeader.h

@@ -40,6 +40,8 @@ struct LedgerHeader
 
     // If validated is false, it means "not yet validated."
     // Once validated is true, it will never be set false at a later time.
+    // NOTE: If you are accessing this directly, you are probably doing it
+    //   wrong. Use LedgerMaster::isValidated().
     // VFALCO TODO Make this not mutable
     bool mutable validated = false;
     bool accepted = false;

include/xrpl/server/NetworkOPs.h

@@ -196,7 +196,7 @@ public:
     virtual bool
     isFull() = 0;
     virtual void
-    setMode(OperatingMode om) = 0;
+    setMode(OperatingMode om, char const* reason) = 0;
     virtual bool
     isBlocked() = 0;
     virtual bool

src/libxrpl/core/HashRouter.cpp

@@ -82,6 +82,19 @@ HashRouter::shouldProcess(
     return s.shouldProcess(suppressionMap_.clock().now(), txInterval);
 }
 
+bool
+HashRouter::shouldProcessForPeer(
+    uint256 const& key,
+    PeerShortID peer,
+    std::chrono::seconds interval)
+{
+    std::lock_guard lock(mutex_);
+
+    auto& entry = emplace(key).first;
+
+    return entry.shouldProcessForPeer(peer, suppressionMap_.clock().now(), interval);
+}
+
 HashRouterFlags
 HashRouter::getFlags(uint256 const& key)
 {
@@ -119,4 +132,13 @@ HashRouter::shouldRelay(uint256 const& key) -> std::optional<std::set<PeerShortI
     return s.releasePeerSet();
 }
 
+auto
+HashRouter::getPeers(uint256 const& key) -> std::set<PeerShortID>
+{
+    std::lock_guard lock(mutex_);
+
+    auto& s = emplace(key).first;
+    return s.peekPeerSet();
+}
+
 }  // namespace xrpl

src/libxrpl/protocol/BuildInfo.cpp

@@ -23,7 +23,7 @@ namespace {
 //------------------------------------------------------------------------------
 // clang-format off
 // NOLINTNEXTLINE(readability-identifier-naming)
-char const* const versionString = "3.3.0-rc1"
+char const* const versionString = "3.3.0-b1"
     // clang-format on
     ;
 

src/test/app/HashRouter_test.cpp

@@ -395,6 +395,33 @@ class HashRouter_test : public beast::unit_test::Suite
         BEAST_EXPECT(!any(HF::UNDEFINED));
     }
 
+    void
+    testProcessPeer()
+    {
+        using namespace std::chrono_literals;
+        TestStopwatch stopwatch;
+        HashRouter router(getSetup(5s, 5s), stopwatch);
+        uint256 const key(1);
+        HashRouter::PeerShortID peer1 = 1;
+        HashRouter::PeerShortID peer2 = 2;
+        auto const timeout = 2s;
+
+        BEAST_EXPECT(router.shouldProcessForPeer(key, peer1, timeout));
+        BEAST_EXPECT(!router.shouldProcessForPeer(key, peer1, timeout));
+        ++stopwatch;
+        BEAST_EXPECT(!router.shouldProcessForPeer(key, peer1, timeout));
+        BEAST_EXPECT(router.shouldProcessForPeer(key, peer2, timeout));
+        BEAST_EXPECT(!router.shouldProcessForPeer(key, peer2, timeout));
+        ++stopwatch;
+        BEAST_EXPECT(router.shouldProcessForPeer(key, peer1, timeout));
+        BEAST_EXPECT(!router.shouldProcessForPeer(key, peer2, timeout));
+        ++stopwatch;
+        BEAST_EXPECT(router.shouldProcessForPeer(key, peer2, timeout));
+        ++stopwatch;
+        BEAST_EXPECT(router.shouldProcessForPeer(key, peer1, timeout));
+        BEAST_EXPECT(!router.shouldProcessForPeer(key, peer2, timeout));
+    }
+
 public:
     void
     run() override
@@ -407,6 +434,7 @@ public:
         testProcess();
         testSetup();
         testFlagsOps();
+        testProcessPeer();
     }
 };
 

src/test/app/LedgerReplay_test.cpp

@@ -384,6 +384,11 @@ public:
     {
         return false;
     }
+    std::set<std::optional<uint64_t>>
+    releaseRequestCookies(uint256 const& requestHash) override
+    {
+        return {};
+    }
 
     [[nodiscard]] std::string const&
     fingerprint() const override

src/test/overlay/ProtocolVersion_test.cpp

@@ -63,8 +63,8 @@ public:
                 negotiateProtocolVersion("RTXP/1.2, XRPL/2.0, XRPL/2.1") == makeProtocol(2, 1));
             BEAST_EXPECT(negotiateProtocolVersion("XRPL/2.2") == makeProtocol(2, 2));
             BEAST_EXPECT(
-                negotiateProtocolVersion("RTXP/1.2, XRPL/2.2, XRPL/2.3, XRPL/999.999") ==
-                makeProtocol(2, 2));
+                negotiateProtocolVersion("RTXP/1.2, XRPL/2.2, XRPL/2.3, XRPL/2.4, XRPL/999.999") ==
+                makeProtocol(2, 3));
             BEAST_EXPECT(negotiateProtocolVersion("XRPL/999.999, WebSocket/1.0") == std::nullopt);
             BEAST_EXPECT(negotiateProtocolVersion("") == std::nullopt);
         }

src/test/overlay/reduce_relay_test.cpp

@@ -191,6 +191,11 @@ public:
     removeTxQueue(uint256 const&) override
     {
     }
+    std::set<std::optional<uint64_t>>
+    releaseRequestCookies(uint256 const& requestHash) override
+    {
+        return {};
+    }
 };
 
 /**

src/xrpld/app/consensus/RCLConsensus.cpp

@@ -1058,7 +1058,7 @@ void
 RCLConsensus::Adaptor::updateOperatingMode(std::size_t const positions) const
 {
     if ((positions == 0u) && app_.getOPs().isFull())
-        app_.getOPs().setMode(OperatingMode::CONNECTED);
+        app_.getOPs().setMode(OperatingMode::CONNECTED, "updateOperatingMode: no positions");
 }
 
 void

src/xrpld/app/ledger/InboundLedger.h

@@ -195,4 +195,22 @@ private:
     std::unique_ptr<PeerSet> peerSet_;
 };
 
+inline std::string
+to_string(InboundLedger::Reason reason)
+{
+    using enum InboundLedger::Reason;
+    switch (reason)
+    {
+        case HISTORY:
+            return "HISTORY";
+        case GENERIC:
+            return "GENERIC";
+        case CONSENSUS:
+            return "CONSENSUS";
+        default:
+            UNREACHABLE("ripple::to_string(InboundLedger::Reason) : unknown value");
+            return "unknown";
+    }
+}
+
 }  // namespace xrpl

src/xrpld/app/ledger/detail/InboundLedger.cpp

@@ -367,7 +367,14 @@ InboundLedger::onTimer(bool wasProgress, ScopedLockType&)
 
     if (!wasProgress)
     {
-        checkLocal();
+        if (checkLocal())
+        {
+            // Done. Something else (probably consensus) built the ledger
+            // locally while waiting for data (or possibly before requesting)
+            XRPL_ASSERT(isDone(), "ripple::InboundLedger::onTimer : done");
+            JLOG(journal_.info()) << "Finished while waiting " << hash_;
+            return;
+        }
 
         byHash_ = true;
 

src/xrpld/app/ledger/detail/InboundLedgers.cpp

@@ -6,6 +6,7 @@
 #include <xrpld/overlay/PeerSet.h>
 
 #include <xrpl/basics/Blob.h>
+#include <xrpl/basics/CanProcess.h>
 #include <xrpl/basics/DecayingSample.h>
 #include <xrpl/basics/Log.h>
 #include <xrpl/basics/Slice.h>
@@ -80,10 +81,78 @@ public:
             XRPL_ASSERT(
                 hash.isNonZero(), "xrpl::InboundLedgersImp::acquire::doAcquire : nonzero hash");
 
-            // probably not the right rule
-            if (app_.getOPs().isNeedNetworkLedger() && (reason != InboundLedger::Reason::GENERIC) &&
-                (reason != InboundLedger::Reason::CONSENSUS))
+            bool const needNetworkLedger = app_.getOPs().isNeedNetworkLedger();
+            bool const shouldAcquire = [&]() {
+                if (!needNetworkLedger)
+                    return true;
+                if (reason == InboundLedger::Reason::GENERIC)
+                    return true;
+                if (reason == InboundLedger::Reason::CONSENSUS)
+                    return true;
+                return false;
+            }();
+
+            std::stringstream ss;
+            ss << "InboundLedger::acquire: "
+               << "Request: " << to_string(hash) << ", " << seq
+               << " NeedNetworkLedger: " << (needNetworkLedger ? "yes" : "no")
+               << " Reason: " << to_string(reason)
+               << " Should acquire: " << (shouldAcquire ? "true." : "false.");
+
+            /*  Acquiring ledgers is somewhat expensive. It requires lots of
+             *  computation and network communication. Avoid it when it's not
+             *  appropriate. Every validation from a peer for a ledger that
+             *  we do not have locally results in a call to this function: even
+             *  if we are moments away from validating the same ledger.
+             */
+            bool const shouldBroadcast = [&]() {
+                // If the node is not in "full" state, it needs to sync to
+                // the network, and doesn't have the necessary tx's and
+                // ledger entries to build the ledger.
+                bool const isFull = app_.getOPs().isFull();
+                // If everything else is ok, don't try to acquire the ledger
+                // if the requested seq is in the near future relative to
+                // the validated ledger. If the requested ledger is between
+                // 1 and 19 inclusive ledgers ahead of the valid ledger this
+                // node has not built it yet, but it's possible/likely it
+                // has the tx's necessary to build it and get caught up.
+                // Plus it might not become validated. On the other hand, if
+                // it's more than 20 in the future, this node should request
+                // it so that it can jump ahead and get caught up.
+                LedgerIndex const validSeq = app_.getLedgerMaster().getValidLedgerIndex();
+                constexpr std::size_t lagLeeway = 20;
+                bool const nearFuture = (seq > validSeq) && (seq < validSeq + lagLeeway);
+                // If everything else is ok, don't try to acquire the ledger
+                // if the request is related to consensus. (Note that
+                // consensus calls usually pass a seq of 0, so nearFuture
+                // will be false other than on a brand new network.)
+                bool const consensus = reason == InboundLedger::Reason::CONSENSUS;
+                ss << " Evaluating whether to broadcast requests to peers"
+                   << ". full: " << (isFull ? "true" : "false") << ". ledger sequence " << seq
+                   << ". Valid sequence: " << validSeq << ". Lag leeway: " << lagLeeway
+                   << ". request for near future ledger: " << (nearFuture ? "true" : "false")
+                   << ". Consensus: " << (consensus ? "true" : "false");
+
+                // If the node is not synced, send requests.
+                if (!isFull)
+                    return true;
+                // If the ledger is in the near future, do NOT send requests.
+                // This node is probably about to build it.
+                if (nearFuture)
+                    return false;
+                // If the request is because of consensus, do NOT send requests.
+                // This node is probably about to build it.
+                if (consensus)
+                    return false;
+                return true;
+            }();
+            ss << ". Would broadcast to peers? " << (shouldBroadcast ? "true." : "false.");
+
+            if (!shouldAcquire)
+            {
+                JLOG(j_.debug()) << "Abort(rule): " << ss.str();
                 return {};
+            }
 
             bool isNew = true;
             std::shared_ptr<InboundLedger> inbound;
@@ -91,6 +160,7 @@ public:
                 ScopedLockType sl(lock_);
                 if (stopping_)
                 {
+                    JLOG(j_.debug()) << "Abort(stopping): " << ss.str();
                     return {};
                 }
 
@@ -109,47 +179,51 @@ public:
                     ++counter_;
                 }
             }
+            ss << " IsNew: " << (isNew ? "true" : "false");
 
             if (inbound->isFailed())
+            {
+                JLOG(j_.debug()) << "Abort(failed): " << ss.str();
                 return {};
+            }
 
             if (!isNew)
                 inbound->update(seq);
 
             if (!inbound->isComplete())
+            {
+                JLOG(j_.debug()) << "InProgress: " << ss.str();
                 return {};
+            }
 
+            JLOG(j_.debug()) << "Complete: " << ss.str();
             return inbound->getLedger();
         };
         using namespace std::chrono_literals;
-        std::shared_ptr<Ledger const> ledger =
-            perf::measureDurationAndLog(doAcquire, "InboundLedgersImp::acquire", 500ms, j_);
-
-        return ledger;
+        return perf::measureDurationAndLog(doAcquire, "InboundLedgersImp::acquire", 500ms, j_);
     }
 
     void
     acquireAsync(uint256 const& hash, std::uint32_t seq, InboundLedger::Reason reason) override
     {
-        std::unique_lock lock(acquiresMutex_);
-        try
+        if (CanProcess const check{acquiresMutex_, pendingAcquires_, hash})
         {
-            if (pendingAcquires_.contains(hash))
-                return;
-            pendingAcquires_.insert(hash);
-            ScopeUnlock const unlock(lock);
-            acquire(hash, seq, reason);
+            try
+            {
+                acquire(hash, seq, reason);
+            }
+            catch (std::exception const& e)
+            {
+                JLOG(j_.warn()) << "Exception thrown for acquiring new inbound ledger " << hash
+                                << ": " << e.what();
+            }
+            catch (...)
+            {
+                JLOG(j_.warn()) << "Unknown exception thrown for acquiring new "
+                                   "inbound ledger "
+                                << hash;
+            }
         }
-        catch (std::exception const& e)
-        {
-            JLOG(j_.warn()) << "Exception thrown for acquiring new inbound ledger " << hash << ": "
-                            << e.what();
-        }
-        catch (...)
-        {
-            JLOG(j_.warn()) << "Unknown exception thrown for acquiring new inbound ledger " << hash;
-        }
-        pendingAcquires_.erase(hash);
     }
 
     std::shared_ptr<InboundLedger>

src/xrpld/app/ledger/detail/LedgerMaster.cpp

@@ -970,8 +970,9 @@ LedgerMaster::checkAccept(std::shared_ptr<Ledger const> const& ledger)
         return;
     }
 
-    JLOG(journal_.info()) << "Advancing accepted ledger to " << ledger->header().seq
-                          << " with >= " << minVal << " validations";
+    JLOG(journal_.info()) << "Advancing accepted ledger to " << ledger->header().seq << " ("
+                          << toShortString(ledger->header().hash) << ") with >= " << minVal
+                          << " validations";
 
     ledger->setValidated();
     ledger->setFull();

src/xrpld/app/ledger/detail/TimeoutCounter.cpp

@@ -25,7 +25,8 @@ TimeoutCounter::TimeoutCounter(
     QueueJobParameter&& jobParameter,
     beast::Journal journal)
     : app_(app)
-    , journal_(journal)
+    , sink_(journal, toShortString(hash) + " ")
+    , journal_(sink_)
     , hash_(hash)
     , timerInterval_(interval)
     , queueJobParameter_(std::move(jobParameter))
@@ -41,6 +42,7 @@ TimeoutCounter::setTimer(ScopedLockType& sl)
 {
     if (isDone())
         return;
+    JLOG(journal_.debug()) << "Setting timer for " << timerInterval_.count() << "ms";
     timer_.expires_after(timerInterval_);
     timer_.async_wait([wptr = pmDowncast()](boost::system::error_code const& ec) {
         if (ec == boost::asio::error::operation_aborted)
@@ -48,6 +50,10 @@ TimeoutCounter::setTimer(ScopedLockType& sl)
 
         if (auto ptr = wptr.lock())
         {
+            JLOG(ptr->journal_.debug())
+                << "timer: ec: " << ec
+                << " (operation_aborted: " << boost::asio::error::operation_aborted << " - "
+                << (ec == boost::asio::error::operation_aborted ? "aborted" : "other") << ")";
             ScopedLockType sl(ptr->mtx_);
             ptr->queueJob(sl);
         }

src/xrpld/app/ledger/detail/TimeoutCounter.h

@@ -4,6 +4,7 @@
 
 #include <xrpl/basics/base_uint.h>
 #include <xrpl/beast/utility/Journal.h>
+#include <xrpl/beast/utility/WrappedSink.h>
 #include <xrpl/core/Job.h>
 
 #include <boost/asio/basic_waitable_timer.hpp>
@@ -117,6 +118,7 @@ protected:
     // Used in this class for access to boost::asio::io_context and
     // xrpl::Overlay. Used in subtypes for the kitchen sink.
     Application& app_;
+    beast::WrappedSink sink_;
     beast::Journal journal_;
     mutable std::recursive_mutex mtx_;
 

src/xrpld/app/misc/NetworkOPs.cpp

@@ -34,6 +34,7 @@
 #include <xrpld/rpc/MPTokenIssuanceID.h>
 #include <xrpld/rpc/ServerHandler.h>
 
+#include <xrpl/basics/CanProcess.h>
 #include <xrpl/basics/Log.h>
 #include <xrpl/basics/ToString.h>
 #include <xrpl/basics/UnorderedContainers.h>
@@ -488,7 +489,7 @@ public:
     isFull() override;
 
     void
-    setMode(OperatingMode om) override;
+    setMode(OperatingMode om, char const* reason) override;
 
     bool
     isBlocked() override;
@@ -975,7 +976,7 @@ NetworkOPsImp::strOperatingMode(bool const admin /* = false */) const
 inline void
 NetworkOPsImp::setStandAlone()
 {
-    setMode(OperatingMode::FULL);
+    setMode(OperatingMode::FULL, "setStandAlone");
 }
 
 inline void
@@ -1118,7 +1119,7 @@ NetworkOPsImp::processHeartbeatTimer()
         {
             if (mode_ != OperatingMode::DISCONNECTED)
             {
-                setMode(OperatingMode::DISCONNECTED);
+                setMode(OperatingMode::DISCONNECTED, "Heartbeat: insufficient peers");
                 std::stringstream ss;
                 ss << "Node count (" << numPeers << ") has fallen "
                    << "below required minimum (" << minPeerCount_ << ").";
@@ -1142,7 +1143,7 @@ NetworkOPsImp::processHeartbeatTimer()
 
         if (mode_ == OperatingMode::DISCONNECTED)
         {
-            setMode(OperatingMode::CONNECTED);
+            setMode(OperatingMode::CONNECTED, "Heartbeat: sufficient peers");
             JLOG(journal_.info()) << "Node count (" << numPeers << ") is sufficient.";
             CLOG(clog.ss()) << "setting mode to CONNECTED based on " << numPeers << " peers. ";
         }
@@ -1153,11 +1154,11 @@ NetworkOPsImp::processHeartbeatTimer()
         CLOG(clog.ss()) << "mode: " << strOperatingMode(origMode, true);
         if (mode_ == OperatingMode::SYNCING)
         {
-            setMode(OperatingMode::SYNCING);
+            setMode(OperatingMode::SYNCING, "Heartbeat: check syncing");
         }
         else if (mode_ == OperatingMode::CONNECTED)
         {
-            setMode(OperatingMode::CONNECTED);
+            setMode(OperatingMode::CONNECTED, "Heartbeat: check connected");
         }
         auto newMode = mode_.load();
         if (origMode != newMode)
@@ -1865,7 +1866,7 @@ void
 NetworkOPsImp::setAmendmentBlocked()
 {
     amendmentBlocked_ = true;
-    setMode(OperatingMode::CONNECTED);
+    setMode(OperatingMode::CONNECTED, "setAmendmentBlocked");
 }
 
 inline bool
@@ -1896,7 +1897,7 @@ void
 NetworkOPsImp::setUNLBlocked()
 {
     unlBlocked_ = true;
-    setMode(OperatingMode::CONNECTED);
+    setMode(OperatingMode::CONNECTED, "setUNLBlocked");
 }
 
 inline void
@@ -1996,7 +1997,7 @@ NetworkOPsImp::checkLastClosedLedger(Overlay::PeerSequence const& peerList, uint
 
     if ((mode_ == OperatingMode::TRACKING) || (mode_ == OperatingMode::FULL))
     {
-        setMode(OperatingMode::CONNECTED);
+        setMode(OperatingMode::CONNECTED, "check LCL: not on consensus ledger");
     }
 
     if (consensus)
@@ -2084,8 +2085,8 @@ NetworkOPsImp::beginConsensus(
         // this shouldn't happen unless we jump ledgers
         if (mode_ == OperatingMode::FULL)
         {
-            JLOG(journal_.warn()) << "Don't have LCL, going to tracking";
-            setMode(OperatingMode::TRACKING);
+            JLOG(journal_.warn()) << "beginConsensus Don't have LCL, going to tracking";
+            setMode(OperatingMode::TRACKING, "beginConsensus: No LCL");
             CLOG(clog) << "beginConsensus Don't have LCL, going to tracking. ";
         }
 
@@ -2213,7 +2214,7 @@ NetworkOPsImp::endConsensus(std::unique_ptr<std::stringstream> const& clog)
         // validations we have for LCL.  If the ledger is good enough, go to
         // TRACKING - TODO
         if (!needNetworkLedger_)
-            setMode(OperatingMode::TRACKING);
+            setMode(OperatingMode::TRACKING, "endConsensus: check tracking");
     }
 
     if (((mode_ == OperatingMode::CONNECTED) || (mode_ == OperatingMode::TRACKING)) &&
@@ -2226,7 +2227,7 @@ NetworkOPsImp::endConsensus(std::unique_ptr<std::stringstream> const& clog)
         if (registry_.get().getTimeKeeper().now() <
             (current->header().parentCloseTime + 2 * current->header().closeTimeResolution))
         {
-            setMode(OperatingMode::FULL);
+            setMode(OperatingMode::FULL, "endConsensus: check full");
         }
     }
 
@@ -2238,7 +2239,7 @@ NetworkOPsImp::consensusViewChange()
 {
     if ((mode_ == OperatingMode::FULL) || (mode_ == OperatingMode::TRACKING))
     {
-        setMode(OperatingMode::CONNECTED);
+        setMode(OperatingMode::CONNECTED, "consensusViewChange");
     }
 }
 
@@ -2542,7 +2543,7 @@ NetworkOPsImp::pubPeerStatus(std::function<json::Value(void)> const& func)
 }
 
 void
-NetworkOPsImp::setMode(OperatingMode om)
+NetworkOPsImp::setMode(OperatingMode om, char const* reason)
 {
     using namespace std::chrono_literals;
     if (om == OperatingMode::CONNECTED)
@@ -2562,11 +2563,12 @@ NetworkOPsImp::setMode(OperatingMode om)
     if (mode_ == om)
         return;
 
+    auto const sink = om < mode_ ? journal_.warn() : journal_.info();
     mode_ = om;
 
     accounting_.mode(om);
 
-    JLOG(journal_.info()) << "STATE->" << strOperatingMode();
+    JLOG(sink) << "STATE->" << strOperatingMode() << " - " << reason;
     pubServer();
 }
 
@@ -2575,36 +2577,24 @@ NetworkOPsImp::recvValidation(std::shared_ptr<STValidation> const& val, std::str
 {
     JLOG(journal_.trace()) << "recvValidation " << val->getLedgerHash() << " from " << source;
 
-    std::unique_lock lock(validationsMutex_);
-    BypassAccept bypassAccept = BypassAccept::No;
-    try
     {
-        if (pendingValidations_.contains(val->getLedgerHash()))
+        CanProcess const check(validationsMutex_, pendingValidations_, val->getLedgerHash());
+        try
         {
-            bypassAccept = BypassAccept::Yes;
+            BypassAccept bypassAccept = check ? BypassAccept::No : BypassAccept::Yes;
+            handleNewValidation(registry_.get().getApp(), val, source, bypassAccept, journal_);
         }
-        else
+        catch (std::exception const& e)
         {
-            pendingValidations_.insert(val->getLedgerHash());
+            JLOG(journal_.warn()) << "Exception thrown for handling new validation "
+                                  << val->getLedgerHash() << ": " << e.what();
+        }
+        catch (...)
+        {
+            JLOG(journal_.warn()) << "Unknown exception thrown for handling new validation "
+                                  << val->getLedgerHash();
         }
-        ScopeUnlock const unlock(lock);
-        handleNewValidation(registry_.get().getApp(), val, source, bypassAccept, journal_);
     }
-    catch (std::exception const& e)
-    {
-        JLOG(journal_.warn()) << "Exception thrown for handling new validation "
-                              << val->getLedgerHash() << ": " << e.what();
-    }
-    catch (...)
-    {
-        JLOG(journal_.warn()) << "Unknown exception thrown for handling new validation "
-                              << val->getLedgerHash();
-    }
-    if (bypassAccept == BypassAccept::No)
-    {
-        pendingValidations_.erase(val->getLedgerHash());
-    }
-    lock.unlock();
 
     pubValidation(val);
 

src/xrpld/app/misc/detail/AmendmentTable.cpp

@@ -202,11 +202,11 @@ public:
                                            iter->second.upVotes |
                                                boost::adaptors::transformed(to_string<256, void>),
                                            ", ");
-                    // TODO: Maybe transform using to_short_string once #5126 is
+                    // TODO: Maybe transform using toShortString once #5126 is
                     // merged
                     //
                     // iter->second.upVotes |
-                    // boost::adaptors::transformed(to_short_string<256, void>)
+                    // boost::adaptors::transformed(toShortString<256, void>)
                 }
                 else
                 {

src/xrpld/overlay/Peer.h

@@ -23,6 +23,7 @@ enum class ProtocolFeature {
     ValidatorListPropagation,
     ValidatorList2Propagation,
     LedgerReplay,
+    LedgerDataCookies
 };
 
 /**
@@ -135,6 +136,13 @@ public:
 
     [[nodiscard]] virtual bool
     txReduceRelayEnabled() const = 0;
+
+    //
+    // Messages
+    //
+
+    virtual std::set<std::optional<uint64_t>>
+    releaseRequestCookies(uint256 const& requestHash) = 0;
 };
 
 }  // namespace xrpl

src/xrpld/overlay/detail/PeerImp.cpp

@@ -115,6 +115,11 @@ constexpr std::chrono::milliseconds kPeerHighLatency{300};
  */
 constexpr std::chrono::seconds kPeerTimerInterval{60};
 
+/**
+ * How often we process duplicate incoming TMGetLedger messages
+ */
+std::chrono::seconds constexpr kGetLedgerInterval{15};
+
 }  // namespace
 
 // TODO: Remove this exclusion once unit tests are added after the hotfix
@@ -545,6 +550,8 @@ PeerImp::supportsFeature(ProtocolFeature f) const
             return protocol_ >= makeProtocol(2, 2);
         case ProtocolFeature::LedgerReplay:
             return ledgerReplayEnabled_;
+        case ProtocolFeature::LedgerDataCookies:
+            return protocol_ >= makeProtocol(2, 3);
     }
     return false;
 }
@@ -1414,8 +1421,9 @@ PeerImp::handleTransaction(
 void
 PeerImp::onMessage(std::shared_ptr<protocol::TMGetLedger> const& m)
 {
-    auto badData = [&](std::string const& msg) {
-        fee_.update(Resource::kFeeInvalidData, "get_ledger " + msg);
+    auto badData = [&](std::string const& msg, bool chargefee = true) {
+        if (chargefee)
+            fee_.update(Resource::kFeeInvalidData, "get_ledger " + msg);
         JLOG(pJournal_.warn()) << "TMGetLedger: " << msg;
     };
     auto const itype{m->itype()};
@@ -1513,11 +1521,68 @@ PeerImp::onMessage(std::shared_ptr<protocol::TMGetLedger> const& m)
         }
     }
 
+    // Drop duplicate requests from the same peer for at least
+    // `kGetLedgerInterval` seconds.
+    // Append a little junk to prevent the hash of an incoming messsage
+    // from matching the hash of the same outgoing message.
+    // `shouldProcessForPeer` does not distingish between incoming and
+    // outgoing, and some of the message relay logic checks the hash to see
+    // if the message has been relayed already. If the hashes are the same,
+    // a duplicate will be detected when sending the message is attempted,
+    // so it will fail.
+    auto const messageHash = sha512Half(*m, nullptr);
+    // Request cookies are not included in the hash. Track them here.
+    auto const requestCookie = [&m]() -> std::optional<uint64_t> {
+        if (m->has_requestcookie())
+            return m->requestcookie();
+        return std::nullopt;
+    }();
+    auto const [inserted, pending] = [&] {
+        std::lock_guard lock{cookieLock_};
+        auto& cookies = messageRequestCookies_[messageHash];
+        bool const pending = !cookies.empty();
+        return std::pair{cookies.emplace(requestCookie).second, pending};
+    }();
+    // Check if the request has been seen from this peer.
+    if (!app_.getHashRouter().shouldProcessForPeer(messageHash, id_, kGetLedgerInterval))
+    {
+        // This request has already been seen from this peer.
+        // Has it been seen with this request cookie (or lack thereof)?
+
+        if (inserted)
+        {
+            // This is a duplicate request, but with a new cookie. When a
+            // response is ready, one will be sent for each request cookie.
+            JLOG(pJournal_.debug())
+                << "TMGetLedger: duplicate request with new request cookie: "
+                << requestCookie.value_or(0) << ". Job pending: " << (pending ? "yes" : "no")
+                << ": " << messageHash;
+            if (pending)
+            {
+                // Don't bother queueing up a new job if other requests are
+                // already pending. This should limit entries in the job queue
+                // to one per peer per unique request.
+                JLOG(pJournal_.debug()) << "TMGetLedger: Suppressing recvGetLedger job, since one "
+                                           "is pending: "
+                                        << messageHash;
+                return;
+            }
+        }
+        else
+        {
+            // Don't punish nodes that don't know any better
+            return badData(
+                "duplicate request: " + to_string(messageHash),
+                supportsFeature(ProtocolFeature::LedgerDataCookies));
+        }
+    }
+
     // Queue a job to process the request
+    JLOG(pJournal_.debug()) << "TMGetLedger: Adding recvGetLedger job: " << messageHash;
     std::weak_ptr<PeerImp> const weak = shared_from_this();
-    app_.getJobQueue().addJob(JtLedgerReq, "RcvGetLedger", [weak, m]() {
+    app_.getJobQueue().addJob(JtLedgerReq, "RcvGetLedger", [weak, m, messageHash]() {
         if (auto peer = weak.lock())
-            peer->processLedgerRequest(m);
+            peer->processLedgerRequest(m, messageHash);
     });
 }
 
@@ -1624,8 +1689,9 @@ PeerImp::onMessage(std::shared_ptr<protocol::TMReplayDeltaResponse> const& m)
 void
 PeerImp::onMessage(std::shared_ptr<protocol::TMLedgerData> const& m)
 {
-    auto badData = [&](std::string const& msg) {
-        fee_.update(Resource::kFeeInvalidData, msg);
+    auto badData = [&](std::string const& msg, bool charge = true) {
+        if (charge)
+            fee_.update(Resource::kFeeInvalidData, msg);
         JLOG(pJournal_.warn()) << "TMLedgerData: " << msg;
     };
 
@@ -1682,19 +1748,93 @@ PeerImp::onMessage(std::shared_ptr<protocol::TMLedgerData> const& m)
         return;
     }
 
-    // If there is a request cookie, attempt to relay the message
-    if (m->has_requestcookie())
+    auto const messageHash = sha512Half(*m);
+    if (!app_.getHashRouter().addSuppressionPeer(messageHash, id_))
     {
-        if (auto peer = overlay_.findPeerByShortID(m->requestcookie()))
+        // Don't punish nodes that don't know any better
+        return badData(
+            "Duplicate message: " + to_string(messageHash),
+            supportsFeature(ProtocolFeature::LedgerDataCookies));
+    }
+
+    bool const routed =
+        m->has_directresponse() || m->responsecookies_size() || m->has_requestcookie();
+
+    {
+        // Check if this message needs to be forwarded to one or more peers.
+        // Maximum of one of the relevant fields should be populated.
+        XRPL_ASSERT(
+            !m->has_requestcookie() || !m->responsecookies_size(),
+            "ripple::PeerImp::onMessage(TMLedgerData) : valid cookie fields");
+
+        // Make a copy of the response cookies, then wipe the list so it can be
+        // forwarded cleanly
+        auto const responseCookies = m->responsecookies();
+        m->clear_responsecookies();
+        // Flag indicating if this response should be processed locally,
+        // possibly in addition to being forwarded.
+        bool const directResponse = m->has_directresponse() && m->directresponse();
+        m->clear_directresponse();
+
+        auto const relay = [this, m, &messageHash](auto const cookie) {
+            if (auto peer = overlay_.findPeerByShortID(cookie))
+            {
+                XRPL_ASSERT(
+                    !m->has_requestcookie() && !m->responsecookies_size(),
+                    "ripple::PeerImp::onMessage(TMLedgerData) relay : no "
+                    "cookies");
+                if (peer->supportsFeature(ProtocolFeature::LedgerDataCookies))
+                    // Setting this flag is not _strictly_ necessary for peers
+                    // that support it if there are no cookies included in the
+                    // message, but it is more accurate.
+                    m->set_directresponse(true);
+                else
+                    m->clear_directresponse();
+                peer->send(std::make_shared<Message>(*m, protocol::mtLEDGER_DATA));
+            }
+            else
+                JLOG(pJournal_.info()) << "Unable to route TX/ledger data reply to peer [" << cookie
+                                       << "]: " << messageHash;
+        };
+        // If there is a request cookie, attempt to relay the message
+        if (m->has_requestcookie())
         {
+            XRPL_ASSERT(
+                responseCookies.empty(),
+                "ripple::PeerImp::onMessage(TMLedgerData) : no response "
+                "cookies");
             m->clear_requestcookie();
-            peer->send(std::make_shared<Message>(*m, protocol::mtLEDGER_DATA));
+            relay(m->requestcookie());
+            if (!directResponse && responseCookies.empty())
+                return;
         }
-        else
+        // If there's a list of request cookies, attempt to relay the message to
+        // all of them.
+        if (responseCookies.size())
         {
-            JLOG(pJournal_.info()) << "Unable to route TX/ledger data reply";
+            for (auto const cookie : responseCookies)
+                relay(cookie);
+            if (!directResponse)
+                return;
+        }
+    }
+
+    // Now that any forwarding is done check the base message (data only, no
+    // routing info for duplicates)
+    if (routed)
+    {
+        m->clear_directresponse();
+        XRPL_ASSERT(
+            !m->has_requestcookie() && !m->responsecookies_size(),
+            "ripple::PeerImp::onMessage(TMLedgerData) : no cookies");
+        auto const baseMessageHash = sha512Half(*m);
+        if (!app_.getHashRouter().addSuppressionPeer(baseMessageHash, id_))
+        {
+            // Don't punish nodes that don't know any better
+            return badData(
+                "Duplicate message: " + to_string(baseMessageHash),
+                supportsFeature(ProtocolFeature::LedgerDataCookies));
         }
-        return;
     }
 
     uint256 const ledgerHash = uint256::fromRaw(m->ledgerhash());
@@ -3085,16 +3225,21 @@ PeerImp::checkValidation(
 // the TX tree with the specified root hash.
 //
 static std::shared_ptr<PeerImp>
-getPeerWithTree(OverlayImpl& ov, uint256 const& rootHash, PeerImp const* skip)
+getPeerWithTree(
+    OverlayImpl& ov,
+    uint256 const& rootHash,
+    PeerImp const* skip,
+    std::function<bool(Peer::id_t)> shouldProcessCallback)
 {
     std::shared_ptr<PeerImp> ret;
     int retScore = 0;
 
+    XRPL_ASSERT(shouldProcessCallback, "ripple::getPeerWithTree : callback provided");
     ov.forEach([&](std::shared_ptr<PeerImp>&& p) {
         if (p->hasTxSet(rootHash) && p.get() != skip)
         {
             auto score = p->getScore(true);
-            if (!ret || (score > retScore))
+            if (!ret || (score > retScore && shouldProcessCallback(p->id())))
             {
                 ret = std::move(p);
                 retScore = score;
@@ -3113,16 +3258,18 @@ getPeerWithLedger(
     OverlayImpl& ov,
     uint256 const& ledgerHash,
     LedgerIndex ledger,
-    PeerImp const* skip)
+    PeerImp const* skip,
+    std::function<bool(Peer::id_t)> shouldProcessCallback)
 {
     std::shared_ptr<PeerImp> ret;
     int retScore = 0;
 
+    XRPL_ASSERT(shouldProcessCallback, "ripple::getPeerWithLedger : callback provided");
     ov.forEach([&](std::shared_ptr<PeerImp>&& p) {
         if (p->hasLedger(ledgerHash, ledger) && p.get() != skip)
         {
             auto score = p->getScore(true);
-            if (!ret || (score > retScore))
+            if (!ret || (score > retScore && shouldProcessCallback(p->id())))
             {
                 ret = std::move(p);
                 retScore = score;
@@ -3136,7 +3283,8 @@ getPeerWithLedger(
 void
 PeerImp::sendLedgerBase(
     std::shared_ptr<Ledger const> const& ledger,
-    protocol::TMLedgerData& ledgerData)
+    protocol::TMLedgerData& ledgerData,
+    PeerCookieMap const& destinations)
 {
     JLOG(pJournal_.trace()) << "sendLedgerBase: Base data";
 
@@ -3166,14 +3314,92 @@ PeerImp::sendLedgerBase(
         }
     }
 
-    auto message{std::make_shared<Message>(ledgerData, protocol::mtLEDGER_DATA)};
-    send(message);
+    sendToMultiple(ledgerData, destinations);
+}
+
+void
+PeerImp::sendToMultiple(protocol::TMLedgerData& ledgerData, PeerCookieMap const& destinations)
+{
+    bool foundSelf = false;
+    for (auto const& [peer, cookies] : destinations)
+    {
+        if (peer.get() == this)
+            foundSelf = true;
+        bool const multipleCookies = peer->supportsFeature(ProtocolFeature::LedgerDataCookies);
+        std::vector<std::uint64_t> sendCookies;
+
+        bool directResponse = false;
+        if (!multipleCookies)
+        {
+            JLOG(pJournal_.debug()) << "sendToMultiple: Sending " << cookies.size()
+                                    << " TMLedgerData messages to peer [" << peer->id()
+                                    << "]: " << sha512Half(ledgerData);
+        }
+        for (auto const& cookie : cookies)
+        {
+            // Unfortunately, need a separate Message object for every
+            // combination
+            if (cookie)
+            {
+                if (multipleCookies)
+                {
+                    // Save this one for later to send a single message
+                    sendCookies.emplace_back(*cookie);
+                    continue;
+                }
+
+                // Feature not supported, so send a single message with a
+                // single cookie
+                ledgerData.set_requestcookie(*cookie);
+            }
+            else
+            {
+                if (multipleCookies)
+                {
+                    // Set this flag later on the single message
+                    directResponse = true;
+                    continue;
+                }
+
+                ledgerData.clear_requestcookie();
+            }
+            XRPL_ASSERT(
+                !multipleCookies,
+                "ripple::PeerImp::sendToMultiple : ledger data cookies "
+                "unsupported");
+            auto message{std::make_shared<Message>(ledgerData, protocol::mtLEDGER_DATA)};
+            peer->send(message);
+        }
+        if (multipleCookies)
+        {
+            // Send a single message with all the cookies and/or the direct
+            // response flag, so the receiver can farm out the single message to
+            // multiple peers and/or itself
+            XRPL_ASSERT(
+                sendCookies.size() || directResponse,
+                "ripple::PeerImp::sendToMultiple : valid response options");
+            ledgerData.clear_requestcookie();
+            ledgerData.clear_responsecookies();
+            ledgerData.set_directresponse(directResponse);
+            for (auto const& cookie : sendCookies)
+                ledgerData.add_responsecookies(cookie);
+            auto message{std::make_shared<Message>(ledgerData, protocol::mtLEDGER_DATA)};
+            peer->send(message);
+
+            JLOG(pJournal_.debug())
+                << "sendToMultiple: Sent 1 TMLedgerData message to peer [" << peer->id()
+                << "]: including " << (directResponse ? "the direct response flag and " : "")
+                << sendCookies.size() << " response cookies. "
+                << ": " << sha512Half(ledgerData);
+        }
+    }
+    XRPL_ASSERT(foundSelf, "ripple::PeerImp::sendToMultiple : current peer included");
 }
 
 std::shared_ptr<Ledger const>
-PeerImp::getLedger(std::shared_ptr<protocol::TMGetLedger> const& m)
+PeerImp::getLedger(std::shared_ptr<protocol::TMGetLedger> const& m, uint256 const& mHash)
 {
-    JLOG(pJournal_.trace()) << "getLedger: Ledger";
+    JLOG(pJournal_.trace()) << "getLedger: Ledger " << mHash;
 
     std::shared_ptr<Ledger const> ledger;
 
@@ -3189,16 +3415,30 @@ PeerImp::getLedger(std::shared_ptr<protocol::TMGetLedger> const& m)
             if (m->has_querytype() && !m->has_requestcookie())
             {
                 // Attempt to relay the request to a peer
+                // Note repeated messages will not relay to the same peer
+                // before `kGetLedgerInterval` seconds. This prevents one
+                // peer from getting flooded, and distributes the request
+                // load. If a request has been relayed to all eligible
+                // peers, then this message will not be relayed.
                 if (auto const peer = getPeerWithLedger(
-                        overlay_, ledgerHash, m->has_ledgerseq() ? m->ledgerseq() : 0, this))
+                        overlay_,
+                        ledgerHash,
+                        m->has_ledgerseq() ? m->ledgerseq() : 0,
+                        this,
+                        [&](Peer::id_t id) {
+                            return app_.getHashRouter().shouldProcessForPeer(
+                                mHash, id, kGetLedgerInterval);
+                        }))
                 {
                     m->set_requestcookie(id());
                     peer->send(std::make_shared<Message>(*m, protocol::mtGET_LEDGER));
-                    JLOG(pJournal_.debug()) << "getLedger: Request relayed to peer";
+                    JLOG(pJournal_.debug())
+                        << "getLedger: Request relayed to peer [" << peer->id() << "]: " << mHash;
                     return ledger;
                 }
 
-                JLOG(pJournal_.trace()) << "getLedger: Failed to find peer to relay request";
+                JLOG(pJournal_.trace())
+                    << "getLedger: Don't have ledger with hash " << ledgerHash << ": " << mHash;
             }
         }
     }
@@ -3207,15 +3447,15 @@ PeerImp::getLedger(std::shared_ptr<protocol::TMGetLedger> const& m)
         // Attempt to find ledger by sequence
         if (m->ledgerseq() < app_.getLedgerMaster().getEarliestFetch())
         {
-            JLOG(pJournal_.debug()) << "getLedger: Early ledger sequence request";
+            JLOG(pJournal_.debug()) << "getLedger: Early ledger sequence request " << mHash;
         }
         else
         {
             ledger = app_.getLedgerMaster().getLedgerBySeq(m->ledgerseq());
             if (!ledger)
             {
-                JLOG(pJournal_.debug())
-                    << "getLedger: Don't have ledger with sequence " << m->ledgerseq();
+                JLOG(pJournal_.debug()) << "getLedger: Don't have ledger with sequence "
+                                        << m->ledgerseq() << ": " << mHash;
             }
         }
     }
@@ -3237,27 +3477,29 @@ PeerImp::getLedger(std::shared_ptr<protocol::TMGetLedger> const& m)
                     charge(Resource::kFeeMalformedRequest, "get_ledger ledgerSeq");
 
                 ledger.reset();
-                JLOG(pJournal_.warn()) << "getLedger: Invalid ledger sequence " << ledgerSeq;
+                JLOG(pJournal_.warn())
+                    << "getLedger: Invalid ledger sequence " << ledgerSeq << ": " << mHash;
             }
         }
         else if (ledgerSeq < app_.getLedgerMaster().getEarliestFetch())
         {
             ledger.reset();
-            JLOG(pJournal_.debug()) << "getLedger: Early ledger sequence request " << ledgerSeq;
+            JLOG(pJournal_.debug())
+                << "getLedger: Early ledger sequence request " << ledgerSeq << ": " << mHash;
         }
     }
     else
     {
-        JLOG(pJournal_.debug()) << "getLedger: Unable to find ledger";
+        JLOG(pJournal_.debug()) << "getLedger: Unable to find ledger " << mHash;
     }
 
     return ledger;
 }
 
 std::shared_ptr<SHAMap const>
-PeerImp::getTxSet(std::shared_ptr<protocol::TMGetLedger> const& m) const
+PeerImp::getTxSet(std::shared_ptr<protocol::TMGetLedger> const& m, uint256 const& mHash) const
 {
-    JLOG(pJournal_.trace()) << "getTxSet: TX set";
+    JLOG(pJournal_.trace()) << "getTxSet: TX set " << mHash;
 
     uint256 const txSetHash = uint256::fromRaw(m->ledgerhash());
     std::shared_ptr<SHAMap> shaMap{app_.getInboundTransactions().getSet(txSetHash, false)};
@@ -3266,20 +3508,28 @@ PeerImp::getTxSet(std::shared_ptr<protocol::TMGetLedger> const& m) const
         if (m->has_querytype() && !m->has_requestcookie())
         {
             // Attempt to relay the request to a peer
-            if (auto const peer = getPeerWithTree(overlay_, txSetHash, this))
+            // Note repeated messages will not relay to the same peer
+            // before `kGetLedgerInterval` seconds. This prevents one
+            // peer from getting flooded, and distributes the request
+            // load. If a request has been relayed to all eligible
+            // peers, then this message will not be relayed.
+            if (auto const peer = getPeerWithTree(overlay_, txSetHash, this, [&](Peer::id_t id) {
+                    return app_.getHashRouter().shouldProcessForPeer(mHash, id, kGetLedgerInterval);
+                }))
             {
                 m->set_requestcookie(id());
                 peer->send(std::make_shared<Message>(*m, protocol::mtGET_LEDGER));
-                JLOG(pJournal_.debug()) << "getTxSet: Request relayed";
+                JLOG(pJournal_.debug())
+                    << "getTxSet: Request relayed to peer [" << peer->id() << "]: " << mHash;
             }
             else
             {
-                JLOG(pJournal_.debug()) << "getTxSet: Failed to find relay peer";
+                JLOG(pJournal_.debug()) << "getTxSet: Failed to find relay peer: " << mHash;
             }
         }
         else
         {
-            JLOG(pJournal_.debug()) << "getTxSet: Failed to find TX set";
+            JLOG(pJournal_.debug()) << "getTxSet: Failed to find TX set " << mHash;
         }
     }
 
@@ -3287,7 +3537,7 @@ PeerImp::getTxSet(std::shared_ptr<protocol::TMGetLedger> const& m) const
 }
 
 void
-PeerImp::processLedgerRequest(std::shared_ptr<protocol::TMGetLedger> const& m)
+PeerImp::processLedgerRequest(std::shared_ptr<protocol::TMGetLedger> const& m, uint256 const& mHash)
 {
     // Do not resource charge a peer responding to a relay
     if (!m->has_requestcookie())
@@ -3300,9 +3550,72 @@ PeerImp::processLedgerRequest(std::shared_ptr<protocol::TMGetLedger> const& m)
     bool fatLeaves{true};
     auto const itype{m->itype()};
 
+    auto getDestinations = [&] {
+        // If a ledger data message is generated, it's going to be sent to every
+        // peer that is waiting for it.
+
+        PeerCookieMap result;
+
+        std::size_t numCookies = 0;
+        {
+            // Don't do the work under this peer if this peer is not waiting for
+            // any replies
+            auto myCookies = releaseRequestCookies(mHash);
+            if (myCookies.empty())
+            {
+                JLOG(pJournal_.debug()) << "TMGetLedger: peer is no longer "
+                                           "waiting for response to request: "
+                                        << mHash;
+                return result;
+            }
+            numCookies += myCookies.size();
+            result[shared_from_this()] = myCookies;
+        }
+
+        std::set<HashRouter::PeerShortID> const peers = app_.getHashRouter().getPeers(mHash);
+        for (auto const peerID : peers)
+        {
+            // This loop does not need to be done under the HashRouter
+            // lock because findPeerByShortID and releaseRequestCookies
+            // are thread safe, and everything else is local
+            if (auto p = overlay_.findPeerByShortID(peerID))
+            {
+                auto cookies = p->releaseRequestCookies(mHash);
+                numCookies += cookies.size();
+                if (result.contains(p))
+                {
+                    // Unlikely, but if a request came in to this peer while
+                    // iterating, add the items instead of copying /
+                    // overwriting.
+                    XRPL_ASSERT(
+                        p.get() == this,
+                        "ripple::PeerImp::processLedgerRequest : found self in "
+                        "map");
+                    for (auto const& cookie : cookies)
+                        result[p].emplace(cookie);
+                }
+                else if (cookies.size())
+                    result[p] = cookies;
+            }
+        }
+
+        JLOG(pJournal_.debug()) << "TMGetLedger: Processing request for " << result.size()
+                                << " peers. Will send " << numCookies
+                                << " messages if successful: " << mHash;
+
+        return result;
+    };
+    // Will only populate this if we're going to do work.
+    PeerCookieMap destinations;
+
     if (itype == protocol::liTS_CANDIDATE)
     {
-        if (sharedMap = getTxSet(m); !sharedMap)
+        destinations = getDestinations();
+        if (destinations.empty())
+            // Nowhere to send the response!
+            return;
+
+        if (sharedMap = getTxSet(m, mHash); !sharedMap)
             return;
         map = sharedMap.get();
 
@@ -3310,8 +3623,6 @@ PeerImp::processLedgerRequest(std::shared_ptr<protocol::TMGetLedger> const& m)
         ledgerData.set_ledgerseq(0);
         ledgerData.set_ledgerhash(m->ledgerhash());
         ledgerData.set_type(protocol::liTS_CANDIDATE);
-        if (m->has_requestcookie())
-            ledgerData.set_requestcookie(m->requestcookie());
 
         // We'll already have most transactions
         fatLeaves = false;
@@ -3329,7 +3640,12 @@ PeerImp::processLedgerRequest(std::shared_ptr<protocol::TMGetLedger> const& m)
             return;
         }
 
-        if (ledger = getLedger(m); !ledger)
+        destinations = getDestinations();
+        if (destinations.empty())
+            // Nowhere to send the response!
+            return;
+
+        if (ledger = getLedger(m, mHash); !ledger)
             return;
 
         // Fill out the reply
@@ -3337,13 +3653,11 @@ PeerImp::processLedgerRequest(std::shared_ptr<protocol::TMGetLedger> const& m)
         ledgerData.set_ledgerhash(ledgerHash.begin(), ledgerHash.size());
         ledgerData.set_ledgerseq(ledger->header().seq);
         ledgerData.set_type(itype);
-        if (m->has_requestcookie())
-            ledgerData.set_requestcookie(m->requestcookie());
 
         switch (itype)
         {
             case protocol::liBASE:
-                sendLedgerBase(ledger, ledgerData);
+                sendLedgerBase(ledger, ledgerData, destinations);
                 return;
 
             case protocol::liTX_NODE:
@@ -3454,7 +3768,7 @@ PeerImp::processLedgerRequest(std::shared_ptr<protocol::TMGetLedger> const& m)
     if (ledgerData.nodes_size() == 0)
         return;
 
-    send(std::make_shared<Message>(ledgerData, protocol::mtLEDGER_DATA));
+    sendToMultiple(ledgerData, destinations);
 }
 
 // Differential pricing helper.  Returns only the *dynamic* component
@@ -3553,6 +3867,19 @@ PeerImp::isHighLatency() const
     return latency_ >= kPeerHighLatency;
 }
 
+std::set<std::optional<uint64_t>>
+PeerImp::releaseRequestCookies(uint256 const& requestHash)
+{
+    std::set<std::optional<uint64_t>> result;
+    std::lock_guard lock(cookieLock_);
+    if (messageRequestCookies_.contains(requestHash))
+    {
+        std::swap(result, messageRequestCookies_[requestHash]);
+        messageRequestCookies_.erase(requestHash);
+    }
+    return result;
+};
+
 void
 PeerImp::Metrics::addMessage(std::uint64_t bytes)
 {

src/xrpld/overlay/detail/PeerImp.h

@@ -211,6 +211,13 @@ private:
     bool ledgerReplayEnabled_ = false;
     LedgerReplayMsgHandler ledgerReplayMsgHandler_;
 
+    // Track message requests and responses
+    // TODO: Use an expiring cache or something
+    using MessageCookieMap = std::map<uint256, std::set<std::optional<uint64_t>>>;
+    using PeerCookieMap = std::map<std::shared_ptr<Peer>, std::set<std::optional<uint64_t>>>;
+    std::mutex mutable cookieLock_;
+    MessageCookieMap messageRequestCookies_;
+
     friend class OverlayImpl;
 
     class Metrics
@@ -469,6 +476,13 @@ public:
         return txReduceRelayEnabled_;
     }
 
+    //
+    // Messages
+    //
+
+    std::set<std::optional<uint64_t>>
+    releaseRequestCookies(uint256 const& requestHash) override;
+
 private:
     void
     close();
@@ -670,16 +684,22 @@ private:
         std::shared_ptr<protocol::TMValidation> const& packet);
 
     void
-    sendLedgerBase(std::shared_ptr<Ledger const> const& ledger, protocol::TMLedgerData& ledgerData);
-
-    std::shared_ptr<Ledger const>
-    getLedger(std::shared_ptr<protocol::TMGetLedger> const& m);
-
-    std::shared_ptr<SHAMap const>
-    getTxSet(std::shared_ptr<protocol::TMGetLedger> const& m) const;
+    sendLedgerBase(
+        std::shared_ptr<Ledger const> const& ledger,
+        protocol::TMLedgerData& ledgerData,
+        PeerCookieMap const& destinations);
 
     void
-    processLedgerRequest(std::shared_ptr<protocol::TMGetLedger> const& m);
+    sendToMultiple(protocol::TMLedgerData& ledgerData, PeerCookieMap const& destinations);
+
+    std::shared_ptr<Ledger const>
+    getLedger(std::shared_ptr<protocol::TMGetLedger> const& m, uint256 const& mHash);
+
+    std::shared_ptr<SHAMap const>
+    getTxSet(std::shared_ptr<protocol::TMGetLedger> const& m, uint256 const& mHash) const;
+
+    void
+    processLedgerRequest(std::shared_ptr<protocol::TMGetLedger> const& m, uint256 const& mHash);
 
 protected:
     // Kept `protected` so test subclasses (see

src/xrpld/overlay/detail/PeerSet.cpp

@@ -7,6 +7,9 @@
 
 #include <xrpl/basics/Log.h>
 #include <xrpl/beast/utility/Journal.h>
+#include <xrpl/core/HashRouter.h>
+#include <xrpl/core/JobQueue.h>
+#include <xrpl/protocol/digest.h>
 
 #include <google/protobuf/message.h>
 
@@ -101,16 +104,45 @@ PeerSetImpl::sendRequest(
     std::shared_ptr<Peer> const& peer)
 {
     auto packet = std::make_shared<Message>(message, type);
+
+    auto const messageHash = [&]() {
+        auto const packetBuffer = packet->getBuffer(compression::Compressed::Off);
+        return sha512Half(Slice(packetBuffer.data(), packetBuffer.size()));
+    }();
+
+    // Allow messages to be re-sent to the same peer after a delay
+    using namespace std::chrono_literals;
+    constexpr std::chrono::seconds interval = 30s;
+
     if (peer)
     {
-        peer->send(packet);
+        if (app_.getHashRouter().shouldProcessForPeer(messageHash, peer->id(), interval))
+        {
+            JLOG(journal_.trace()) << "Sending " << protocolMessageName(type) << " message to ["
+                                   << peer->id() << "]: " << messageHash;
+            peer->send(packet);
+        }
+        else
+            JLOG(journal_.debug()) << "Suppressing sending duplicate " << protocolMessageName(type)
+                                   << " message to [" << peer->id() << "]: " << messageHash;
         return;
     }
 
     for (auto id : peers_)
     {
         if (auto p = app_.getOverlay().findPeerByShortID(id))
-            p->send(packet);
+        {
+            if (app_.getHashRouter().shouldProcessForPeer(messageHash, p->id(), interval))
+            {
+                JLOG(journal_.trace()) << "Sending " << protocolMessageName(type) << " message to ["
+                                       << p->id() << "]: " << messageHash;
+                p->send(packet);
+            }
+            else
+                JLOG(journal_.debug())
+                    << "Suppressing sending duplicate " << protocolMessageName(type)
+                    << " message to [" << p->id() << "]: " << messageHash;
+        }
     }
 }
 

src/xrpld/overlay/detail/ProtocolMessage.h

@@ -30,6 +30,12 @@ protocolMessageType(protocol::TMGetLedger const&)
     return protocol::mtGET_LEDGER;
 }
 
+inline protocol::MessageType
+protocolMessageType(protocol::TMLedgerData const&)
+{
+    return protocol::mtLEDGER_DATA;
+}
+
 inline protocol::MessageType
 protocolMessageType(protocol::TMReplayDeltaRequest const&)
 {
@@ -450,3 +456,63 @@ invokeProtocolMessage(Buffers const& buffers, Handler& handler, std::size_t& hin
 }
 
 }  // namespace xrpl
+
+namespace protocol {
+
+template <class Hasher>
+void
+hash_append(Hasher& h, TMGetLedger const& msg)
+{
+    using beast::hash_append;
+    using namespace xrpl;
+    hash_append(h, safeCast<int>(protocolMessageType(msg)));
+    hash_append(h, safeCast<int>(msg.itype()));
+    if (msg.has_ltype())
+        hash_append(h, safeCast<int>(msg.ltype()));
+
+    if (msg.has_ledgerhash())
+        hash_append(h, msg.ledgerhash());
+
+    if (msg.has_ledgerseq())
+        hash_append(h, msg.ledgerseq());
+
+    for (auto const& nodeId : msg.nodeids())
+        hash_append(h, nodeId);
+    hash_append(h, msg.nodeids_size());
+
+    // Do NOT include the request cookie. It does not affect the content of the
+    // request, but only where to route the results.
+    // if (msg.has_requestcookie())
+    //    hash_append(h, msg.requestcookie());
+
+    if (msg.has_querytype())
+        hash_append(h, safeCast<int>(msg.querytype()));
+
+    if (msg.has_querydepth())
+        hash_append(h, msg.querydepth());
+}
+
+template <class Hasher>
+void
+hash_append(Hasher& h, TMLedgerData const& msg)
+{
+    using beast::hash_append;
+    using namespace xrpl;
+    hash_append(h, safeCast<int>(protocolMessageType(msg)));
+    hash_append(h, msg.ledgerhash());
+    hash_append(h, msg.ledgerseq());
+    hash_append(h, safeCast<int>(msg.type()));
+    for (auto const& node : msg.nodes())
+    {
+        hash_append(h, node.nodedata());
+        if (node.has_nodeid())
+            hash_append(h, node.nodeid());
+    }
+    hash_append(h, msg.nodes_size());
+    if (msg.has_requestcookie())
+        hash_append(h, msg.requestcookie());
+    if (msg.has_error())
+        hash_append(h, safeCast<int>(msg.error()));
+}
+
+}  // namespace protocol

src/xrpld/overlay/detail/ProtocolVersion.cpp

@@ -29,6 +29,8 @@ namespace xrpl {
 constexpr ProtocolVersion const kSupportedProtocolList[]{
     {2, 1},
     {2, 2},
+    // Adds TMLedgerData::responseCookies and directResponse
+    {2, 3},
 };
 
 // This ugly construct ensures that supportedProtocolList is sorted in strictly