вт, 19 янв. 2021 г. в 22:38, Steve French <smfrench@xxxxxxxxx>:
>
> The patch won't merge (also has some text corruptions in it). This
> line of code is different due to commit 6988a619f5b79
>
> 6988a619f5b79 (Paulo Alcantara 2020-11-28 15:57:06 -0300 342)
> cifs_dbg(FYI, "signal pending before send request\n");
> 6988a619f5b79 (Paulo Alcantara 2020-11-28 15:57:06 -0300 343)
> return -ERESTARTSYS;
>
> if (signal_pending(current)) {
> cifs_dbg(FYI, "signal pending before send request\n");
> return -ERESTARTSYS;
> }
>
> See:
>
> Author: Paulo Alcantara <pc@xxxxxx>
> Date: Sat Nov 28 15:57:06 2020 -0300
>
> cifs: allow syscalls to be restarted in __smb_send_rqst()
>
> A customer has reported that several files in their multi-threaded app
> were left with size of 0 because most of the read(2) calls returned
> -EINTR and they assumed no bytes were read. Obviously, they could
> have fixed it by simply retrying on -EINTR.
>
> We noticed that most of the -EINTR on read(2) were due to real-time
> signals sent by glibc to process wide credential changes (SIGRT_1),
> and its signal handler had been established with SA_RESTART, in which
> case those calls could have been automatically restarted by the
> kernel.
>
> Let the kernel decide to whether or not restart the syscalls when
> there is a signal pending in __smb_send_rqst() by returning
> -ERESTARTSYS. If it can't, it will return -EINTR anyway.
>
> Signed-off-by: Paulo Alcantara (SUSE) <pc@xxxxxx>
> CC: Stable <stable@xxxxxxxxxxxxxxx>
> Reviewed-by: Ronnie Sahlberg <lsahlber@xxxxxxxxxx>
> Reviewed-by: Pavel Shilovsky <pshilov@xxxxxxxxxxxxx>
>
> On Tue, Jan 19, 2021 at 10:32 PM Ronnie Sahlberg <lsahlber@xxxxxxxxxx> wrote:
> >
> > RHBZ 1848178
> >
> > There is no need to fail this function if non-fatal signals are
> > pending when we enter it.
> >
> > Signed-off-by: Ronnie Sahlberg <lsahlber@xxxxxxxxxx>
> > ---
> > fs/cifs/transport.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/fs/cifs/transport.c b/fs/cifs/transport.c
> > index c42bda5a5008..98752f7d2cd2 100644
> > --- a/fs/cifs/transport.c
> > +++ b/fs/cifs/transport.c
> > @@ -339,7 +339,7 @@ __smb_send_rqst(struct TCP_Server_Info *server, int num_rqst,
> > if (ssocket == NULL)
> > return -EAGAIN;
> >
> > - if (signal_pending(current)) {
> > + if (fatal_signal_pending(current)) {
> > cifs_dbg(FYI, "signal is pending before sending any data\n");
> > return -EINTR;
> > }
I have been thinking around the same lines. The original intent of
failing the function here was to avoid interrupting packet send in the
middle of the packet and not breaking an SMB connection.
That's also why signals are blocked around smb_send_kvec() calls. I
guess most of the time a socket buffer is not full, so, those
functions immediately return success without waiting internally and
checking for pending signals. With this change the code may break SMB
packets and cause reconnections if a non-fatal signal is pending
before we block signals but this is still better than failing the
function in the very beginning.
Reviewed-by: Pavel Shilovsky <pshilov@xxxxxxxxxxxxx>
I think this patch should go to stable too. Steve, Ronnie, thoughts?
--
Best regards,
Pavel Shilovsky
