Re: fs: uninterruptible hang in handle_userfault

[Linus Torvalds]

On Tue, Mar 1, 2016 at 11:59 AM, Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
> On Tue, Mar 01, 2016 at 11:56:22AM -0800, Linus Torvalds wrote:
>> (a) special-case the PF_EXITING case for usefaultfd, something like
>>
>>     diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
>>     index 50311703135b..66cdb44616d5 100644
>>     --- a/fs/userfaultfd.c
>>     +++ b/fs/userfaultfd.c
>>     @@ -287,6 +287,12 @@ int handle_userfault(struct vm_area_struct
>> *vma, unsigned long address,
>>                     goto out;
>>
>>             /*
>>     +        * We don't do userfault handling for the final child pid update.
>>     +        */
>>     +       if (current->flags & PF_EXITING)
>>     +               goto out;
>
> Umm...  Probably a dumb question, but would that suffice when e.g. another
> thread is just starting to dump core?

So the only access we really care about is the child tid-pointer
clearing one, and that always happens after PF_EXITING has been set
afaik.

No other case really matters. If somebody accesses a userfault region
just as another thread is exiting, we don't care. I don't think it
would necessarily be wrong to ignore the fault, but I don't think it's
relevant either, since at that stage the normal "you can signal the
thread" still works. It's only the child tid access that comes *after*
we have stopped acceping signals, and that's marked by that
PF_EXITING.

Or maybe I misunderstood your worry entirely or missed something, and
my answer above is entirely beside your point. Did you have something
else in mind?

                 Linus