You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
ConnectionInfo.SendTimeout — configurable socket send timeout to prevent indefinite hangs on TCP zero-window stalls #1793
socket.Send() is a blocking call and Socket.SendTimeout is never set, so it defaults to 0 (infinite). When the server's TCP receive window drops to zero (the server is alive but not consuming data), the OS
holds the call open indefinitely — no exception, no return. The standard TCP dead-peer timeout (several minutes) only applies when the server is completely unreachable, not in the zero-window scenario.
This is distinct from the channel-level window wait in Channel.cs, which already has a 30-second timeout via ConnectionInfo.Timeout. That protection only covers the SSH protocol layer; it never gets a chance
to fire because execution is stuck in socket.Send() first.
OperationTimeout (on SftpClient) similarly cannot help — it guards the wait for an SFTP protocol response after a packet has been sent, not the send itself.
Proposed fix
Add SendTimeout to ConnectionInfo (default Timeout.InfiniteTimeSpan to preserve existing behaviour) and apply it to the socket immediately after connection:
// ConnectionInfo.cs
private TimeSpan _sendTimeout = System.Threading.Timeout.InfiniteTimeSpan;
public TimeSpan SendTimeout
{
get => _sendTimeout;
set
{
value.EnsureValidTimeout(nameof(SendTimeout));
_sendTimeout = value;
}
}
// Session.cs — both Connect() and ConnectAsync() paths
_socket = _serviceFactory.CreateConnector(ConnectionInfo, _socketFactory)
.Connect(ConnectionInfo);
_socket.SendTimeout = ConnectionInfo.SendTimeout.AsTimeout();
When SendTimeout elapses, socket.Send() throws SocketException with SocketError.TimedOut, which propagates out of SendPacket() and terminates the session normally.
Notes
Default is infinite — no behaviour change for existing users
The timeout is a stall detector, not a total-transfer budget: it resets on every successful send call, so large files over slow-but-healthy connections are not affected
Applies to all SSH traffic (not just SFTP), which is correct — a stalled send on any packet type means the session is broken
Activity