fleet: count only admitted peers against the session limit so silent connections cannot lock the roster out

backend/fleet/src/Fleet.Link/Peer.cs

@@ -27,7 +27,15 @@ public sealed class PeerListenerOptions
 
     public int MaxFrameBytes { get; init; } = 8 << 20;
 
-    public int MaxSessions { get; init; } = 8;
+    /// <summary>
+    /// 同时在线的会话数上限，只数已经通过敲门的对端。每个拨号方每个地址族一个会话，所以要容得下「名册里的节点数 × 2」。
+    /// 之前这个上限连还没敲门的连接一起数：节点多起来之后，已建立的会话把名额占满，后到的拨号方（哪怕在名册里）一连上就被断开，
+    /// 而且一直连不上，重试只会得到同样的结果。
+    /// </summary>
+    public int MaxSessions { get; init; } = 64;
+
+    /// <summary>还没通过敲门的连接数上限（陌生人、扫描器、还在握手的）：敲门有读超时，占着名额的时间有限，但它们不能挤占已通过敲门的会话。</summary>
+    public int MaxPending { get; init; } = 64;
 
     /// <summary>敲门、TLS、挑战应答整个握手的上限。线路差的时候（RTT 好几秒、还丢包）握手本来就要很久，10 秒做不完就会一直连不上；敲门已经挡住了不认识的连接，放宽不会放大被占连接的风险。</summary>
     public TimeSpan HandshakeTimeout { get; init; } = PeerTimings.Handshake;
@@ -44,6 +52,7 @@ public sealed class PeerListener
     private readonly INodeLog _log;
     private readonly TimeProvider _time;
     private int _sessions;
+    private int _pending;
 
     public PeerListener(PeerListenerOptions options, Func<PeerSessionInfo, IRpcTransport, CancellationToken, Task> onSession, INodeLog log, TimeProvider time)
     {
@@ -74,9 +83,9 @@ public sealed class PeerListener
             while (!ct.IsCancellationRequested)
             {
                 TcpClient client = await listener.AcceptTcpClientAsync(ct);
-                if (Interlocked.Increment(ref _sessions) > _options.MaxSessions)
+                if (Interlocked.Increment(ref _pending) > _options.MaxPending)
                 {
-                    Interlocked.Decrement(ref _sessions);
+                    Interlocked.Decrement(ref _pending);
                     client.Dispose();
                     continue;
                 }
@@ -97,6 +106,7 @@ public sealed class PeerListener
     private async Task HandleAsync(TcpClient client, CancellationToken ct)
     {
         bool admitted = false;
+        bool counted = false;
         string phase = "knock";
         long started = _time.GetTimestamp();
         EndPoint? remote = client.Client.RemoteEndPoint;
@@ -113,6 +123,15 @@ public sealed class PeerListener
             }
 
             admitted = true;
+            Interlocked.Decrement(ref _pending);
+            if (Interlocked.Increment(ref _sessions) > _options.MaxSessions)
+            {
+                Interlocked.Decrement(ref _sessions);
+                _log.Warn("peer session limit reached", ("remote", remote), ("limit", _options.MaxSessions));
+                return;
+            }
+
+            counted = true;
             phase = "tls";
             _log.Info("peer knock accepted", ("remote", remote), ("phase", phase));
             var tls = new SslStream(stream, leaveInnerStreamOpen: false);
@@ -152,7 +171,15 @@ public sealed class PeerListener
         finally
         {
             client.Dispose();
-            Interlocked.Decrement(ref _sessions);
+            if (!admitted)
+            {
+                Interlocked.Decrement(ref _pending);
+            }
+
+            if (counted)
+            {
+                Interlocked.Decrement(ref _sessions);
+            }
         }
     }
 

backend/fleet/tests/Fleet.Tests/LinkTests.cs

@@ -303,7 +303,7 @@ public sealed class LinkTests
         private readonly CancellationTokenSource _stop = new();
         private readonly NodeIdentity _identity;
 
-        public PeerFixture(Func<string, byte[]?> lookup, Func<PeerSessionInfo, IRpcTransport, CancellationToken, Task>? onSession = null)
+        public PeerFixture(Func<string, byte[]?> lookup, Func<PeerSessionInfo, IRpcTransport, CancellationToken, Task>? onSession = null, int maxSessions = 64)
         {
             Certificate = NodeCertificate.LoadPfx(NodeCertificate.CreateSelfSignedPfx("agent", TimeSpan.FromDays(30), TimeProvider.System));
             _identity = NodeIdentity.FromPkcs8(Certificate.GetECDsaPrivateKey()!.ExportPkcs8PrivateKey());
@@ -314,6 +314,7 @@ public sealed class LinkTests
                     Certificate = Certificate,
                     LookupPeerKey = lookup,
                     Knock = new KnockGate(_identity, () => KnockPeers, TimeProvider.System),
+                    MaxSessions = maxSessions,
                 },
                 onSession ?? ((_, transport, ct) => KeepOpenAsync(transport, ct)),
                 NullNodeLog.Instance, TimeProvider.System);
@@ -462,6 +463,65 @@ public sealed class LinkTests
         await ok.DisposeAsync();
     }
 
+    /// <summary>名额只数通过敲门的会话：一堆还没敲门的连接占着，名册里的拨号方照样连得上（之前它们共用一个上限，节点多了之后后到的拨号方被永远拒绝）。</summary>
+    [Fact]
+    public async Task PeerListener_SessionLimitCountsOnlyAdmittedPeers_SoSilentConnectionsCannotLockOutTheRoster()
+    {
+        using NodeIdentity first = NodeIdentity.Generate();
+        using NodeIdentity second = NodeIdentity.Generate();
+        using NodeIdentity third = NodeIdentity.Generate();
+        await using var agent = new PeerFixture(id => id == "dlr_1" ? first.PublicKeySpki : id == "dlr_2" ? second.PublicKeySpki : third.PublicKeySpki, maxSessions: 2);
+        agent.KnockPeers.AddRange([first.PublicKeySpki, second.PublicKeySpki, third.PublicKeySpki]);
+        await agent.Listener.Started.Task;
+
+        // 20 个什么都不说的连接：占着待敲门的名额，但不占会话名额
+        var silent = new List<TcpClient>();
+        try
+        {
+            for (int i = 0; i < 20; i++)
+            {
+                var client = new TcpClient();
+                await client.ConnectAsync(IPAddress.Loopback, agent.Listener.Port);
+                silent.Add(client);
+            }
+
+            (_, TlsFrameTransport a) = await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, first, "dlr_1", null, 1 << 20, TimeProvider.System, default);
+            (_, TlsFrameTransport b) = await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, second, "dlr_2", null, 1 << 20, TimeProvider.System, default);
+
+            // 会话名额（2）满了：第三个在名册里的拨号方被拒绝，一个会话结束之后又能连上
+            await Assert.ThrowsAnyAsync<Exception>(async () =>
+            {
+                (_, TlsFrameTransport refused) = await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, third, "dlr_3", null, 1 << 20, TimeProvider.System, default);
+                await refused.DisposeAsync();
+            });
+            await a.DisposeAsync();
+            bool reconnected = false;
+            for (int attempt = 0; attempt < 50 && !reconnected; attempt++)
+            {
+                try
+                {
+                    (_, TlsFrameTransport again) = await PeerConnector.ConnectAsync("127.0.0.1", agent.Listener.Port, agent.PublicKey, third, "dlr_3", null, 1 << 20, TimeProvider.System, default);
+                    await again.DisposeAsync();
+                    reconnected = true;
+                }
+                catch (Exception ex) when (ex is IOException or System.Security.Authentication.AuthenticationException or InvalidOperationException)
+                {
+                    await Task.Delay(100);
+                }
+            }
+
+            Assert.True(reconnected);
+            await b.DisposeAsync();
+        }
+        finally
+        {
+            foreach (TcpClient client in silent)
+            {
+                client.Dispose();
+            }
+        }
+    }
+
     [Fact]
     public async Task PeerListener_WritesNothingToStrangers_NotEvenATlsServerHello()
     {