On Wed, 19 Oct 2011 13:57:55 +0400
Pavel Shilovsky <piastryyy@xxxxxxxxx> wrote:
> 2011/10/19 Pavel Shilovsky <piastryyy@xxxxxxxxx>:
> > 2011/10/19 Steve French <smfrench@xxxxxxxxx>:
> >> If the session could continue after an incorrect checksum (ie retry
> >> the operation
> >> that failed due to the network error/corruption)
> >> then there would be significant advantage in not taking the session down.
> >> In practice would that request time out and get killed and retried?
> >
> > As cifs_check_receive doesn't return any error if checksum doesn't
> > match the original, readpages continues the read-ahead process and
> > nothing else happens (I mean the request isn't time out, get killed
> > and retried).
> >
> > So, if we don't kill the session in this case, we can simply return
> > -EAGAIN and let the calling routine retry the operation itself.
> > Thoughts?
>
> I tried this patch and it works good to me:
>
> diff --git a/fs/cifs/transport.c b/fs/cifs/transport.c
> index 33a3fbf..b54a631 100644
> --- a/fs/cifs/transport.c
> +++ b/fs/cifs/transport.c
> @@ -508,8 +508,10 @@ cifs_check_receive(struct mid_q_entry *mid,
> struct TCP_Server_Info *server,
> iov.iov_len = len;
> /* FIXME: add code to kill session */
> if (cifs_verify_signature(&iov, 1, server,
> - mid->sequence_number + 1) != 0)
> + mid->sequence_number + 1) != 0) {
> cERROR(1, "Unexpected SMB signature");
> + return -EAGAIN;
> + }
> }
>
> /* BB special case reconnect tid and uid here? */
>
>
> I also applied almost the same patch (that includes the fix for
> cifs_readv_callback) to Jeff's async read patchset , fixed the
> problems with the error handling in cifs_readv_complete (I posted them
> to the list) and got the same good result.
>
My concern here would be that this could continuously loop if there
were a persistent problem with the signature. I'd suggest that we do
abort the connection after a certain number of signature check failures
on a socket (5? 10?).
--
Jeff Layton <jlayton@xxxxxxxxxx>
--
To unsubscribe from this list: send the line "unsubscribe linux-cifs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
