Skip to content

ShellStream.Dispose deadlocks when the server closes the channel at the same time #1851

Open

Description

@robsonfelix

Version

SSH.NET 2026.0.0 (tag 2026.0.0, 7b2fd3d). The same code is on develop today. .NET 10, Linux x64 and arm64.

Summary

Disposing a ShellStream while the server closes its channel can deadlock forever. The two threads take the same two locks in opposite order, and neither wait has a timeout:

Thread Holds Waits for
Caller: ShellStream.Dispose() ShellStream._sync Channel._messagingLock
Session listener: SSH_MSG_CHANNEL_CLOSE received Channel._messagingLock ShellStream._sync

The two paths

Caller thread, disposing the stream:

  1. ShellStream.Dispose(bool) takes lock (_sync) (src/Renci.SshNet/ShellStream.cs:970).
  2. Still inside that lock it calls _channel.Dispose() (ShellStream.cs:986).
  3. Channel.Dispose(bool) calls Close() (src/Renci.SshNet/Channels/Channel.cs:844), which takes lock (_messagingLock) (Channel.cs:534).

Listener thread, handling the server's SSH_MSG_CHANNEL_CLOSE:

  1. OnChannelClose calls OnClose() (Channel.cs:698), which calls Close() (Channel.cs:426) and takes lock (_messagingLock) (Channel.cs:534).
  2. Still inside that lock it raises Closed (Channel.cs:578).
  3. ShellStream.Channel_Closed handles it by calling Dispose() (ShellStream.cs:1006), which waits for lock (_sync) (ShellStream.cs:970).

If the caller enters step 1 of its path before the listener reaches step 3 of its own, each thread holds the lock the other one needs. A stack dump of a hung process shows exactly these two frames.

Repro

The race is timing-dependent. A simplified loop against a real OpenSSH server (a sketch of our test, not run as written here):

for (var i = 0; i < 100; i++)
{
    using var client = new SshClient(host, port, user, password);
    client.Connect();
    var stream = client.CreateShellStream("xterm", 80, 24, 0, 0, 4096);
    stream.WriteLine("exit");   // the server closes the channel shortly after
    stream.Dispose();           // races with the listener's Channel_Closed -> Dispose
}

In our test suite this hung 4 of 15 runs: a test class whose Dispose() disposed the ShellStream while the remote shell exited. The process never recovered; CI killed it after an hour.

Workaround we use

Call SshClient.Disconnect() before disposing the ShellStream. That stops the listener thread, so no Closed event can arrive while Dispose holds _sync. With it, 12 of 12 runs passed.

Possible fixes

  • In ShellStream.Dispose, detach the handlers and dispose _channel outside lock (_sync): take the lock only to set _disposed, then release it before calling _channel.Dispose().
  • Or, in Channel.Close, raise Closed after leaving lock (_messagingLock), so that no user code runs while that lock is held.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions