Re: Strange issues with epoll since 5.0

[Eric Wong <e@80x24.org>]

Deepa Dinamani <deepa.kernel@gmail.com> wrote:
> So here is my analysis:

<snip everything I agree with>

> So the 854a6ed56839a40f6 seems to be better than the original code in
> that it detects the signal.

OTOH, does matter to anybody that a signal is detected slightly
sooner than it would've been, otherwise?

> But, the problem is that it doesn't
> communicate it to the userspace.

Yup, that's a big problem :)
 
> So a patch like below solves the problem. This is incomplete. I'll
> verify and send you a proper fix you can test soon. This is just for
> the sake of discussion:
> 
> diff --git a/fs/eventpoll.c b/fs/eventpoll.c
> index 4a0e98d87fcc..63a387329c3d 100644
> --- a/fs/eventpoll.c
> +++ b/fs/eventpoll.c
> @@ -2317,7 +2317,7 @@ SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct
> epoll_event __user *, events,
>                 int, maxevents, int, timeout, const sigset_t __user *, sigmask,
>                 size_t, sigsetsize)
>  {
> -       int error;
> +       int error, signal_detected;
>         sigset_t ksigmask, sigsaved;
> 
>         /*
> @@ -2330,7 +2330,10 @@ SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct
> epoll_event __user *, events,
> 
>         error = do_epoll_wait(epfd, events, maxevents, timeout);
> 
> -       restore_user_sigmask(sigmask, &sigsaved);
> +       signal_detected = restore_user_sigmask(sigmask, &sigsaved);
> +
> +       if (signal_detected && !error)
> +               return -EITNR;
> 
>         return error;

Looks like a reasonable API.

> @@ -2862,7 +2862,7 @@ void restore_user_sigmask(const void __user
> *usigmask, sigset_t *sigsaved)
>         if (signal_pending(current)) {
>                 current->saved_sigmask = *sigsaved;
>                 set_restore_sigmask();
> -               return;
> +               return 0;

Shouldn't that "return 1" if a signal is pending?