On Wed, 2021-07-14 at 11:50 -0400, Chuck Lever wrote:
> In some rare failure modes, the server is actually reading the
> transport, but then just dropping the requests on the floor.
> TCP_USER_TIMEOUT cannot detect that case.
>
> Prevent such a stuck server from pinning client resources
> indefinitely by ensuring that session and client ID clean-up can
> time out even if the connection is still operational.
>
> Signed-off-by: Chuck Lever <chuck.lever@xxxxxxxxxx>
> ---
> fs/nfs/nfs4client.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/fs/nfs/nfs4client.c b/fs/nfs/nfs4client.c
> index 28431acd1230..c5032f784ac0 100644
> --- a/fs/nfs/nfs4client.c
> +++ b/fs/nfs/nfs4client.c
> @@ -281,6 +281,7 @@ static void nfs4_destroy_callback(struct
> nfs_client *clp)
>
> static void nfs4_shutdown_client(struct nfs_client *clp)
> {
> + clp->cl_rpcclient->cl_noretranstimeo = 0;
> if (__test_and_clear_bit(NFS_CS_RENEWD, &clp->cl_res_state))
> nfs4_kill_renewd(clp);
> clp->cl_mvops->shutdown_client(clp);
>
>
I can't see how this will help. Again, I suggest we rather turn off the
retransmission default for the RPC calls where the server can drop
stuff on the floor. That's really only the RPCSEC_GSS control
messages.
Anything else is covered by the NFSv4 blanket ban on dropping RPC
calls.
--
Trond Myklebust
Linux NFS client maintainer, Hammerspace
trond.myklebust@xxxxxxxxxxxxxxx
