On Sat, Oct 29, 2016 at 11:52 AM, Al Viro <v...@zeniv.linux.org.uk> wrote:
>
> And that call can happen as soon as we return from __blockdev_direct_IO()
> (even earlier, actually). As soon as that happens, the reference to
> struct file we'd acquired in io_submit_one() is dropped. If descriptor
> table had been shared, another thread might have already closed that sucker,
> and fput() from aio_complete() would free struct file.
But that's the point. We don't *do* anything like that any more. We
now always do the final access from aio_complete(). So it doesn't
matter if that is called asynchronously (very early) or not.
That's the whole point of the patch. Exactly to do everything either
*before* we even submit it (at which point no completion can happen),
or doing it in aio_complete() which is guaranteed to be after the
submission. No races, no use-after-free.
What am I missing?
Linus
