Merge bitcoin/bitcoin#25500: refactor: Move inbound eviction logic to its own translation unit

0101d2bc3c3bcf698d6cc2a237a680fc52395987 [net] Move eviction logic to its own file (dergoegge)
c741d748d4d9836940b99091cc7be09c65efcb79 [net] Move ConnectionType to its own file (Cory Fields)
a3c27070396ab8c2941c437e8099547e8fc9c110 [net] Add connection type to NodeEvictionCandidate (dergoegge)
42aa5d5b6269d27af525d5001907558442e96023 [net] Add NoBan status to NodeEvictionCandidate (dergoegge)

Pull request description:

  This PR splits of the first couple commits from #25268 that move the inbound eviction logic from `net.{h,cpp}` to `eviction.{h,cpp}`.

  Please look at #25268 for motivation and conceptual review.

ACKs for top commit:
  jnewbery:
    utACK 0101d2bc3c3bcf698d6cc2a237a680fc52395987
  theuni:
    utACK 0101d2bc3c3bcf698d6cc2a237a680fc52395987. I quickly verified with `git --color-moved` that the move-only changes are indeed move-only.

Tree-SHA512: e0c345a698030e049cb22fe281b44503c04403c5be5a3750ca14bfcc603a162ac6bac9a39552472feb57c460102b7ca91430b8ad6268f2efccc49b5e8959331b

4 files changed, 417 insertions, 0 deletions

diff --git a/src/node/connection_types.cpp b/src/node/connection_types.cpp
new file mode 100644
index 0000000000..904f4371aa
--- /dev/null
+++ b/src/node/connection_types.cpp
@@ -0,0 +1,26 @@
+// Copyright (c) 2022 The Bitcoin Core developers
+// Distributed under the MIT software license, see the accompanying
+// file COPYING or http://www.opensource.org/licenses/mit-license.php.
+
+#include <node/connection_types.h>
+#include <cassert>
+
+std::string ConnectionTypeAsString(ConnectionType conn_type)
+{
+    switch (conn_type) {
+    case ConnectionType::INBOUND:
+        return "inbound";
+    case ConnectionType::MANUAL:
+        return "manual";
+    case ConnectionType::FEELER:
+        return "feeler";
+    case ConnectionType::OUTBOUND_FULL_RELAY:
+        return "outbound-full-relay";
+    case ConnectionType::BLOCK_RELAY:
+        return "block-relay-only";
+    case ConnectionType::ADDR_FETCH:
+        return "addr-fetch";
+    } // no default case, so the compiler can warn about missing cases
+
+    assert(false);
+}
diff --git a/src/node/connection_types.h b/src/node/connection_types.h
new file mode 100644
index 0000000000..5e1abcace6
--- /dev/null
+++ b/src/node/connection_types.h
@@ -0,0 +1,82 @@
+// Copyright (c) 2022 The Bitcoin Core developers
+// Distributed under the MIT software license, see the accompanying
+// file COPYING or http://www.opensource.org/licenses/mit-license.php.
+
+#ifndef BITCOIN_NODE_CONNECTION_TYPES_H
+#define BITCOIN_NODE_CONNECTION_TYPES_H
+
+#include <string>
+
+/** Different types of connections to a peer. This enum encapsulates the
+ * information we have available at the time of opening or accepting the
+ * connection. Aside from INBOUND, all types are initiated by us.
+ *
+ * If adding or removing types, please update CONNECTION_TYPE_DOC in
+ * src/rpc/net.cpp and src/qt/rpcconsole.cpp, as well as the descriptions in
+ * src/qt/guiutil.cpp and src/bitcoin-cli.cpp::NetinfoRequestHandler. */
+enum class ConnectionType {
+    /**
+     * Inbound connections are those initiated by a peer. This is the only
+     * property we know at the time of connection, until P2P messages are
+     * exchanged.
+     */
+    INBOUND,
+
+    /**
+     * These are the default connections that we use to connect with the
+     * network. There is no restriction on what is relayed; by default we relay
+     * blocks, addresses & transactions. We automatically attempt to open
+     * MAX_OUTBOUND_FULL_RELAY_CONNECTIONS using addresses from our AddrMan.
+     */
+    OUTBOUND_FULL_RELAY,
+
+
+    /**
+     * We open manual connections to addresses that users explicitly requested
+     * via the addnode RPC or the -addnode/-connect configuration options. Even if a
+     * manual connection is misbehaving, we do not automatically disconnect or
+     * add it to our discouragement filter.
+     */
+    MANUAL,
+
+    /**
+     * Feeler connections are short-lived connections made to check that a node
+     * is alive. They can be useful for:
+     * - test-before-evict: if one of the peers is considered for eviction from
+     *   our AddrMan because another peer is mapped to the same slot in the tried table,
+     *   evict only if this longer-known peer is offline.
+     * - move node addresses from New to Tried table, so that we have more
+     *   connectable addresses in our AddrMan.
+     * Note that in the literature ("Eclipse Attacks on Bitcoin’s Peer-to-Peer Network")
+     * only the latter feature is referred to as "feeler connections",
+     * although in our codebase feeler connections encompass test-before-evict as well.
+     * We make these connections approximately every FEELER_INTERVAL:
+     * first we resolve previously found collisions if they exist (test-before-evict),
+     * otherwise we connect to a node from the new table.
+     */
+    FEELER,
+
+    /**
+     * We use block-relay-only connections to help prevent against partition
+     * attacks. By not relaying transactions or addresses, these connections
+     * are harder to detect by a third party, thus helping obfuscate the
+     * network topology. We automatically attempt to open
+     * MAX_BLOCK_RELAY_ONLY_ANCHORS using addresses from our anchors.dat. Then
+     * addresses from our AddrMan if MAX_BLOCK_RELAY_ONLY_CONNECTIONS
+     * isn't reached yet.
+     */
+    BLOCK_RELAY,
+
+    /**
+     * AddrFetch connections are short lived connections used to solicit
+     * addresses from peers. These are initiated to addresses submitted via the
+     * -seednode command line argument, or under certain conditions when the
+     * AddrMan is empty.
+     */
+    ADDR_FETCH,
+};
+
+/** Convert ConnectionType enum to a string value */
+std::string ConnectionTypeAsString(ConnectionType conn_type);
+
+#endif // BITCOIN_NODE_CONNECTION_TYPES_H
diff --git a/src/node/eviction.cpp b/src/node/eviction.cpp
new file mode 100644
index 0000000000..33406931d4
--- /dev/null
+++ b/src/node/eviction.cpp
@@ -0,0 +1,240 @@
+// Copyright (c) 2022 The Bitcoin Core developers
+// Distributed under the MIT software license, see the accompanying
+// file COPYING or http://www.opensource.org/licenses/mit-license.php.
+
+#include <node/eviction.h>
+
+#include <algorithm>
+#include <array>
+#include <chrono>
+#include <cstdint>
+#include <functional>
+#include <map>
+#include <vector>
+
+
+static bool ReverseCompareNodeMinPingTime(const NodeEvictionCandidate &a, const NodeEvictionCandidate &b)
+{
+    return a.m_min_ping_time > b.m_min_ping_time;
+}
+
+static bool ReverseCompareNodeTimeConnected(const NodeEvictionCandidate &a, const NodeEvictionCandidate &b)
+{
+    return a.m_connected > b.m_connected;
+}
+
+static bool CompareNetGroupKeyed(const NodeEvictionCandidate &a, const NodeEvictionCandidate &b) {
+    return a.nKeyedNetGroup < b.nKeyedNetGroup;
+}
+
+static bool CompareNodeBlockTime(const NodeEvictionCandidate &a, const NodeEvictionCandidate &b)
+{
+    // There is a fall-through here because it is common for a node to have many peers which have not yet relayed a block.
+    if (a.m_last_block_time != b.m_last_block_time) return a.m_last_block_time < b.m_last_block_time;
+    if (a.fRelevantServices != b.fRelevantServices) return b.fRelevantServices;
+    return a.m_connected > b.m_connected;
+}
+
+static bool CompareNodeTXTime(const NodeEvictionCandidate &a, const NodeEvictionCandidate &b)
+{
+    // There is a fall-through here because it is common for a node to have more than a few peers that have not yet relayed txn.
+    if (a.m_last_tx_time != b.m_last_tx_time) return a.m_last_tx_time < b.m_last_tx_time;
+    if (a.m_relay_txs != b.m_relay_txs) return b.m_relay_txs;
+    if (a.fBloomFilter != b.fBloomFilter) return a.fBloomFilter;
+    return a.m_connected > b.m_connected;
+}
+
+// Pick out the potential block-relay only peers, and sort them by last block time.
+static bool CompareNodeBlockRelayOnlyTime(const NodeEvictionCandidate &a, const NodeEvictionCandidate &b)
+{
+    if (a.m_relay_txs != b.m_relay_txs) return a.m_relay_txs;
+    if (a.m_last_block_time != b.m_last_block_time) return a.m_last_block_time < b.m_last_block_time;
+    if (a.fRelevantServices != b.fRelevantServices) return b.fRelevantServices;
+    return a.m_connected > b.m_connected;
+}
+
+/**
+ * Sort eviction candidates by network/localhost and connection uptime.
+ * Candidates near the beginning are more likely to be evicted, and those
+ * near the end are more likely to be protected, e.g. less likely to be evicted.
+ * - First, nodes that are not `is_local` and that do not belong to `network`,
+ *   sorted by increasing uptime (from most recently connected to connected longer).
+ * - Then, nodes that are `is_local` or belong to `network`, sorted by increasing uptime.
+ */
+struct CompareNodeNetworkTime {
+    const bool m_is_local;
+    const Network m_network;
+    CompareNodeNetworkTime(bool is_local, Network network) : m_is_local(is_local), m_network(network) {}
+    bool operator()(const NodeEvictionCandidate& a, const NodeEvictionCandidate& b) const
+    {
+        if (m_is_local && a.m_is_local != b.m_is_local) return b.m_is_local;
+        if ((a.m_network == m_network) != (b.m_network == m_network)) return b.m_network == m_network;
+        return a.m_connected > b.m_connected;
+    };
+};
+
+//! Sort an array by the specified comparator, then erase the last K elements where predicate is true.
+template <typename T, typename Comparator>
+static void EraseLastKElements(
+    std::vector<T>& elements, Comparator comparator, size_t k,
+    std::function<bool(const NodeEvictionCandidate&)> predicate = [](const NodeEvictionCandidate& n) { return true; })
+{
+    std::sort(elements.begin(), elements.end(), comparator);
+    size_t eraseSize = std::min(k, elements.size());
+    elements.erase(std::remove_if(elements.end() - eraseSize, elements.end(), predicate), elements.end());
+}
+
+void ProtectNoBanConnections(std::vector<NodeEvictionCandidate>& eviction_candidates)
+{
+    eviction_candidates.erase(std::remove_if(eviction_candidates.begin(), eviction_candidates.end(),
+                                             [](NodeEvictionCandidate const& n) {
+                                                 return n.m_noban;
+                                             }),
+                              eviction_candidates.end());
+}
+
+void ProtectOutboundConnections(std::vector<NodeEvictionCandidate>& eviction_candidates)
+{
+    eviction_candidates.erase(std::remove_if(eviction_candidates.begin(), eviction_candidates.end(),
+                                             [](NodeEvictionCandidate const& n) {
+                                                 return n.m_conn_type != ConnectionType::INBOUND;
+                                             }),
+                              eviction_candidates.end());
+}
+
+void ProtectEvictionCandidatesByRatio(std::vector<NodeEvictionCandidate>& eviction_candidates)
+{
+    // Protect the half of the remaining nodes which have been connected the longest.
+    // This replicates the non-eviction implicit behavior, and precludes attacks that start later.
+    // To favorise the diversity of our peer connections, reserve up to half of these protected
+    // spots for Tor/onion, localhost, I2P, and CJDNS peers, even if they're not longest uptime
+    // overall. This helps protect these higher-latency peers that tend to be otherwise
+    // disadvantaged under our eviction criteria.
+    const size_t initial_size = eviction_candidates.size();
+    const size_t total_protect_size{initial_size / 2};
+
+    // Disadvantaged networks to protect. In the case of equal counts, earlier array members
+    // have the first opportunity to recover unused slots from the previous iteration.
+    struct Net { bool is_local; Network id; size_t count; };
+    std::array<Net, 4> networks{
+        {{false, NET_CJDNS, 0}, {false, NET_I2P, 0}, {/*localhost=*/true, NET_MAX, 0}, {false, NET_ONION, 0}}};
+
+    // Count and store the number of eviction candidates per network.
+    for (Net& n : networks) {
+        n.count = std::count_if(eviction_candidates.cbegin(), eviction_candidates.cend(),
+                                [&n](const NodeEvictionCandidate& c) {
+                                    return n.is_local ? c.m_is_local : c.m_network == n.id;
+                                });
+    }
+    // Sort `networks` by ascending candidate count, to give networks having fewer candidates
+    // the first opportunity to recover unused protected slots from the previous iteration.
+    std::stable_sort(networks.begin(), networks.end(), [](Net a, Net b) { return a.count < b.count; });
+
+    // Protect up to 25% of the eviction candidates by disadvantaged network.
+    const size_t max_protect_by_network{total_protect_size / 2};
+    size_t num_protected{0};
+
+    while (num_protected < max_protect_by_network) {
+        // Count the number of disadvantaged networks from which we have peers to protect.
+        auto num_networks = std::count_if(networks.begin(), networks.end(), [](const Net& n) { return n.count; });
+        if (num_networks == 0) {
+            break;
+        }
+        const size_t disadvantaged_to_protect{max_protect_by_network - num_protected};
+        const size_t protect_per_network{std::max(disadvantaged_to_protect / num_networks, static_cast<size_t>(1))};
+        // Early exit flag if there are no remaining candidates by disadvantaged network.
+        bool protected_at_least_one{false};
+
+        for (Net& n : networks) {
+            if (n.count == 0) continue;
+            const size_t before = eviction_candidates.size();
+            EraseLastKElements(eviction_candidates, CompareNodeNetworkTime(n.is_local, n.id),
+                               protect_per_network, [&n](const NodeEvictionCandidate& c) {
+                                   return n.is_local ? c.m_is_local : c.m_network == n.id;
+                               });
+            const size_t after = eviction_candidates.size();
+            if (before > after) {
+                protected_at_least_one = true;
+                const size_t delta{before - after};
+                num_protected += delta;
+                if (num_protected >= max_protect_by_network) {
+                    break;
+                }
+                n.count -= delta;
+            }
+        }
+        if (!protected_at_least_one) {
+            break;
+        }
+    }
+
+    // Calculate how many we removed, and update our total number of peers that
+    // we want to protect based on uptime accordingly.
+    assert(num_protected == initial_size - eviction_candidates.size());
+    const size_t remaining_to_protect{total_protect_size - num_protected};
+    EraseLastKElements(eviction_candidates, ReverseCompareNodeTimeConnected, remaining_to_protect);
+}
+
+[[nodiscard]] std::optional<NodeId> SelectNodeToEvict(std::vector<NodeEvictionCandidate>&& vEvictionCandidates)
+{
+    // Protect connections with certain characteristics
+
+    ProtectNoBanConnections(vEvictionCandidates);
+
+    ProtectOutboundConnections(vEvictionCandidates);
+
+    // Deterministically select 4 peers to protect by netgroup.
+    // An attacker cannot predict which netgroups will be protected
+    EraseLastKElements(vEvictionCandidates, CompareNetGroupKeyed, 4);
+    // Protect the 8 nodes with the lowest minimum ping time.
+    // An attacker cannot manipulate this metric without physically moving nodes closer to the target.
+    EraseLastKElements(vEvictionCandidates, ReverseCompareNodeMinPingTime, 8);
+    // Protect 4 nodes that most recently sent us novel transactions accepted into our mempool.
+    // An attacker cannot manipulate this metric without performing useful work.
+    EraseLastKElements(vEvictionCandidates, CompareNodeTXTime, 4);
+    // Protect up to 8 non-tx-relay peers that have sent us novel blocks.
+    EraseLastKElements(vEvictionCandidates, CompareNodeBlockRelayOnlyTime, 8,
+                       [](const NodeEvictionCandidate& n) { return !n.m_relay_txs && n.fRelevantServices; });
+
+    // Protect 4 nodes that most recently sent us novel blocks.
+    // An attacker cannot manipulate this metric without performing useful work.
+    EraseLastKElements(vEvictionCandidates, CompareNodeBlockTime, 4);
+
+    // Protect some of the remaining eviction candidates by ratios of desirable
+    // or disadvantaged characteristics.
+    ProtectEvictionCandidatesByRatio(vEvictionCandidates);
+
+    if (vEvictionCandidates.empty()) return std::nullopt;
+
+    // If any remaining peers are preferred for eviction consider only them.
+    // This happens after the other preferences since if a peer is really the best by other criteria (esp relaying blocks)
+    //  then we probably don't want to evict it no matter what.
+    if (std::any_of(vEvictionCandidates.begin(),vEvictionCandidates.end(),[](NodeEvictionCandidate const &n){return n.prefer_evict;})) {
+        vEvictionCandidates.erase(std::remove_if(vEvictionCandidates.begin(),vEvictionCandidates.end(),
+                                  [](NodeEvictionCandidate const &n){return !n.prefer_evict;}),vEvictionCandidates.end());
+    }
+
+    // Identify the network group with the most connections and youngest member.
+    // (vEvictionCandidates is already sorted by reverse connect time)
+    uint64_t naMostConnections;
+    unsigned int nMostConnections = 0;
+    std::chrono::seconds nMostConnectionsTime{0};
+    std::map<uint64_t, std::vector<NodeEvictionCandidate> > mapNetGroupNodes;
+    for (const NodeEvictionCandidate &node : vEvictionCandidates) {
+        std::vector<NodeEvictionCandidate> &group = mapNetGroupNodes[node.nKeyedNetGroup];
+        group.push_back(node);
+        const auto grouptime{group[0].m_connected};
+
+        if (group.size() > nMostConnections || (group.size() == nMostConnections && grouptime > nMostConnectionsTime)) {
+            nMostConnections = group.size();
+            nMostConnectionsTime = grouptime;
+            naMostConnections = node.nKeyedNetGroup;
+        }
+    }
+
+    // Reduce to the network group with the most connections
+    vEvictionCandidates = std::move(mapNetGroupNodes[naMostConnections]);
+
+    // Disconnect from the network group with the most connections
+    return vEvictionCandidates.front().id;
+}
diff --git a/src/node/eviction.h b/src/node/eviction.h
new file mode 100644
index 0000000000..1bb32e5327
--- /dev/null
+++ b/src/node/eviction.h
@@ -0,0 +1,69 @@
+// Copyright (c) 2022 The Bitcoin Core developers
+// Distributed under the MIT software license, see the accompanying
+// file COPYING or http://www.opensource.org/licenses/mit-license.php.
+
+#ifndef BITCOIN_NODE_EVICTION_H
+#define BITCOIN_NODE_EVICTION_H
+
+#include <node/connection_types.h>
+#include <net_permissions.h>
+
+#include <chrono>
+#include <cstdint>
+#include <optional>
+#include <vector>
+
+typedef int64_t NodeId;
+
+struct NodeEvictionCandidate {
+    NodeId id;
+    std::chrono::seconds m_connected;
+    std::chrono::microseconds m_min_ping_time;
+    std::chrono::seconds m_last_block_time;
+    std::chrono::seconds m_last_tx_time;
+    bool fRelevantServices;
+    bool m_relay_txs;
+    bool fBloomFilter;
+    uint64_t nKeyedNetGroup;
+    bool prefer_evict;
+    bool m_is_local;
+    Network m_network;
+    bool m_noban;
+    ConnectionType m_conn_type;
+};
+
+/**
+ * Select an inbound peer to evict after filtering out (protecting) peers having
+ * distinct, difficult-to-forge characteristics. The protection logic picks out
+ * fixed numbers of desirable peers per various criteria, followed by (mostly)
+ * ratios of desirable or disadvantaged peers. If any eviction candidates
+ * remain, the selection logic chooses a peer to evict.
+ */
+[[nodiscard]] std::optional<NodeId> SelectNodeToEvict(std::vector<NodeEvictionCandidate>&& vEvictionCandidates);
+
+/** Protect desirable or disadvantaged inbound peers from eviction by ratio.
+ *
+ * This function protects half of the peers which have been connected the
+ * longest, to replicate the non-eviction implicit behavior and preclude attacks
+ * that start later.
+ *
+ * Half of these protected spots (1/4 of the total) are reserved for the
+ * following categories of peers, sorted by longest uptime, even if they're not
+ * longest uptime overall:
+ *
+ * - onion peers connected via our tor control service
+ *
+ * - localhost peers, as manually configured hidden services not using
+ *   `-bind=addr[:port]=onion` will not be detected as inbound onion connections
+ *
+ * - I2P peers
+ *
+ * - CJDNS peers
+ *
+ * This helps protect these privacy network peers, which tend to be otherwise
+ * disadvantaged under our eviction criteria for their higher min ping times
+ * relative to IPv4/IPv6 peers, and favorise the diversity of peer connections.
+ */
+void ProtectEvictionCandidatesByRatio(std::vector<NodeEvictionCandidate>& vEvictionCandidates);
+
+#endif // BITCOIN_NODE_EVICTION_H