On Thu, Mar 7, 2019 at 4:33 AM Pavel Shilovsky <piastryyy@xxxxxxxxx> wrote:
>
> вт, 5 мар. 2019 г. в 21:45, Ronnie Sahlberg <lsahlber@xxxxxxxxxx>:
> >
> > A negative timeout is the same as the current behaviour, i.e. no timeout.
> >
> > Signed-off-by: Ronnie Sahlberg <lsahlber@xxxxxxxxxx>
> > ---
> > fs/cifs/transport.c | 40 ++++++++++++++++++++++++++++++----------
> > 1 file changed, 30 insertions(+), 10 deletions(-)
> >
> > diff --git a/fs/cifs/transport.c b/fs/cifs/transport.c
> > index f4ec9635dab2..3263e8b3a57d 100644
> > --- a/fs/cifs/transport.c
> > +++ b/fs/cifs/transport.c
> > @@ -478,11 +478,18 @@ smb_send(struct TCP_Server_Info *server, struct smb_hdr *smb_buffer,
> >
> > static int
> > wait_for_free_credits(struct TCP_Server_Info *server, const int num_credits,
> > - const int flags, unsigned int *instance)
> > + const int timeout, const int flags,
> > + unsigned int *instance)
> > {
> > int rc;
> > int *credits;
> > int optype;
> > + long int t;
> > +
> > + if (timeout < 0)
> > + t = MAX_JIFFY_OFFSET;
> > + else
> > + t = msecs_to_jiffies(timeout);
> >
> > optype = flags & CIFS_OP_MASK;
> >
> > @@ -507,11 +514,16 @@ wait_for_free_credits(struct TCP_Server_Info *server, const int num_credits,
> > if (*credits < num_credits) {
> > spin_unlock(&server->req_lock);
> > cifs_num_waiters_inc(server);
> > - rc = wait_event_killable(server->request_q,
> > - has_credits(server, credits, num_credits));
> > + rc = wait_event_killable_timeout(server->request_q,
> > + has_credits(server, credits, num_credits), t);
> > cifs_num_waiters_dec(server);
> > - if (rc)
> > - return rc;
> > + if (!rc) {
> > + cifs_dbg(VFS, "wait timed out after %d ms\n",
> > + timeout);
> > + return -EAGAIN;
>
> It would be hard to distinguish different EAGAIN reasons in the upper
> layers. As of now we use EAGAIN as retryable error indicating the
> upper layer to retry. This case is different because the retry most
> likely won't help and the sequential fallback is needed. In
> compound_send_recv() we return -ENOTSUPP if there is not enough
> credits to satisfy compounding and the number of requests in flight
> doesn't promise to get new credits in future.
>
> I suggest to return the same error code here (-ENOTSUPP). Probably
> this particular error code isn't ideal choice and suggestions to
> change it are welcome.
I will do that change. ENOTSUPP might not be ideal but should be fine for now.
If/when we need to take special action and start serializing these
requests we can
pick a better errno.
>
> > + }
> > + if (rc == -ERESTARTSYS)
> > + return -ERESTARTSYS;
> > spin_lock(&server->req_lock);
> > } else {
> > if (server->tcpStatus == CifsExiting) {
> > @@ -537,12 +549,19 @@ wait_for_free_credits(struct TCP_Server_Info *server, const int num_credits,
> > *credits <= MAX_COMPOUND) {
> > spin_unlock(&server->req_lock);
> > cifs_num_waiters_inc(server);
> > - rc = wait_event_killable(server->request_q,
> > + rc = wait_event_killable_timeout(
> > + server->request_q,
> > has_credits(server, credits,
> > - MAX_COMPOUND + 1));
> > + MAX_COMPOUND + 1),
> > + t);
> > cifs_num_waiters_dec(server);
> > - if (rc)
> > - return rc;
> > + if (!rc) {
> > + cifs_dbg(VFS, "wait timed out after %d ms\n",
> > + timeout);
> > + return -EAGAIN;
> > + }
> > + if (rc == -ERESTARTSYS)
> > + return -ERESTARTSYS;
> > spin_lock(&server->req_lock);
> > continue;
> > }
> > @@ -569,7 +588,8 @@ static int
> > wait_for_free_request(struct TCP_Server_Info *server, const int flags,
> > unsigned int *instance)
> > {
> > - return wait_for_free_credits(server, 1, flags, instance);
> > + return wait_for_free_credits(server, 1, -1, flags,
> > + instance);
> > }
> >
> > int
> > --
> > 2.13.6
> >
>
> --
> Best regards,
> Pavel Shilovsky
