Re: [CRIU] [PATCH 1/4] signalfd: add ability to return siginfo in a raw format

[Oleg Nesterov <oleg@redhat.com>]

On 12/27, Andrey Wagin wrote:
>
> 2012/12/27 Oleg Nesterov <oleg@redhat.com>:
> > On 12/27, Andrey Wagin wrote:
> >>
> >> On Wed, Dec 26, 2012 at 05:31:12PM +0100, Oleg Nesterov wrote:
> >> >
> >> > So I think we should not use llseek. But, probably we can rely on pread() ?
> >> > This way we can avoid the problem above, and this looks even simpler.
> >>
> >> Yes. It is a good idea. A new patch is attached to this email. I
> >> implemented pread for signalfd and fixed all your previous comments.
> >>
> ...
> > I think it would be better to pass ppos, not *ppos, because ...
> >
> >> +             if (*ppos)
> >> +                     (*ppos)++;
> >
> > in this case we can update *ppos in signalfd_peek() and simplify the
> > code a bit.
> >
> > Compared to the previous version it is "safe" to change *ppos even if
> > copy_to_user() fails, *ppos will be "lost" anyway after we return.
>
> Yes. But ppos is updated, because pread() reads all siginfo-s from a queue,
> which fit in a buffer starting with the initial ppos.

Can't understand. And I guess you misunderstood.

I meant, we can update *ppos in signalfd_peek() _and_ we can do this
unconditionally even if copy_to_user() failed.

> >> @@ -321,6 +372,7 @@ SYSCALL_DEFINE4(signalfd4, int, ufd, sigset_t __user *, user_mask,
> >>               }
> >>
> >>               file->f_flags |= flags & SFD_RAW;
> >> +             file->f_mode |= FMODE_PREAD;
> >>
> >>               fd_install(ufd, file);
> >
> > Hmm. Looks like it is based on other patches I didnt see...
> >
> > But I don't understand how FMODE_PREAD connects to this patch, we need
> > this flag set even for regular sys_read() ???
>
> It doesn't need for sys_read(), but this patch is about pread() and
> sys_pread() checks it.
>
> SYSCALL_DEFINE(pread64)(unsigned int fd, char __user *buf,
>                         size_t count, loff_t pos)
> {
> ....
>         if (f.file) {
>                 ret = -ESPIPE;
>                 if (f.file->f_mode & FMODE_PREAD)
>                         ret = vfs_read(f.file, buf, count, &pos);

And sys_read() checks it too, that was my point.

So either this code was already broken before this patch, or FMODE_PREAD
is already set and you do not need this chunk.

Oleg.