On Mon, 2025-08-04 at 12:08 -0700, Dai Ngo wrote:
> Currently, when an RPC connection times out during the connect phase,
> the task is retried by placing it back on the pending queue and
> waiting
> again. In some cases, the timeout occurs because TCP is unable to
> send
> the SYN packet. This situation most often arises on bare metal
> systems
> at boot time, when the NFS mount is attempted while the network link
> appears to be up but is not yet stable.
>
> This patch addresses the issue by updating call_connect_status to
> destroy
> the transport on ETIMEDOUT error before retrying the connection. This
> ensures that subsequent connection attempts use a fresh transport,
> reducing the likelihood of repeated failures due to lingering network
> issues.
>
> Signed-off-by: Dai Ngo <dai.ngo@xxxxxxxxxx>
> ---
> net/sunrpc/clnt.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
> index 21426c3049d3..701b742750c5 100644
> --- a/net/sunrpc/clnt.c
> +++ b/net/sunrpc/clnt.c
> @@ -2215,6 +2215,7 @@ call_connect_status(struct rpc_task *task)
> case -EHOSTUNREACH:
> case -EPIPE:
> case -EPROTO:
> + case -ETIMEDOUT:
> xprt_conditional_disconnect(task->tk_rqstp->rq_xprt,
> task->tk_rqstp-
> >rq_connect_cookie);
> if (RPC_IS_SOFTCONN(task))
> @@ -2225,7 +2226,6 @@ call_connect_status(struct rpc_task *task)
> case -EADDRINUSE:
> case -ENOTCONN:
> case -EAGAIN:
> - case -ETIMEDOUT:
> if (!(task->tk_flags & RPC_TASK_NO_ROUND_ROBIN) &&
> (task->tk_flags & RPC_TASK_MOVEABLE) &&
> test_bit(XPRT_REMOVE, &xprt->state)) {
Why is this needed? The ETIMEDOUT is supposed to be a task level error,
not a connection level thing.
Oh... Is this because of TLS? If so, then please fix that to use a more
appropriate error.
--
Trond Myklebust
Linux NFS client maintainer, Hammerspace
trondmy@xxxxxxxxxx, trond.myklebust@xxxxxxxxxxxxxxx
