Re: [PATCH 4/4] NFSv4: Treat ENETUNREACH errors as fatal for state recovery

[Trond Myklebust <trondmy@hammerspace.com>]

On Mon, 2025-03-24 at 20:46 -0400, trondmy@kernel.org wrote:
> From: Trond Myklebust <trond.myklebust@hammerspace.com>
> 
> If a containerised process is killed and causes an ENETUNREACH or
> ENETDOWN error to be propagated to the state manager, then mark the
> nfs_client as being dead so that we don't loop in functions that are
> expecting recovery to succeed.
> 
> Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
> ---
>  fs/nfs/nfs4state.c | 10 +++++++++-
>  1 file changed, 9 insertions(+), 1 deletion(-)
> 
> diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c
> index 738eb2789266..629578dd4a42 100644
> --- a/fs/nfs/nfs4state.c
> +++ b/fs/nfs/nfs4state.c
> @@ -2739,7 +2739,15 @@ static void nfs4_state_manager(struct
> nfs_client *clp)
>  	pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4
> server %s"
>  			" with error %d\n", section_sep, section,
>  			clp->cl_hostname, -status);
> -	ssleep(1);
> +	switch (status) {
> +	case -ENETDOWN:
> +	case -ENETUNREACH:
> +		nfs_mark_client_ready(clp, status);

Actually, that probably needs to be nfs_mark_client_ready(clp, -EIO) in
order to make the return value a fatal I/O error.

> +		break;
> +	default:
> +		ssleep(1);
> +		break;
> +	}
>  out_drain:
>  	memalloc_nofs_restore(memflags);
>  	nfs4_end_drain_session(clp);

-- 
Trond Myklebust
Linux NFS client maintainer, Hammerspace
trond.myklebust@hammerspace.com