Re: [PATCH v2 2/2] mm: use pidfd_get_task()

[Christian Brauner]

On Tue, Oct 12, 2021 at 04:13:31PM +0200, David Hildenbrand wrote:
> On 11.10.21 15:32, Christian Brauner wrote:
> > From: Christian Brauner <christian.brauner@xxxxxxxxxx>
> > 
> > Instead of duplicating the same code in two places use the newly added
> > pidfd_get_task() helper. This fixes an (unimportant for now) bug where
> > PIDTYPE_PID is used whereas PIDTYPE_TGID should have been used.
> 
> What would have been the effect of the BUG? Is it worth Fixes: or better

Right now, there's no issue. I hope my "unimportant for now" gets that
across.
Retrieving it via PIDTYPE_PID or PIDTYPE_TGID doesn't matter right now
because at pidfd creation time we ensure that:
- the pid used with pidfd_open()
- the task created via clone{3}()'s CLONE_PIDFD
are used as PIDTYPE_TGID, i.e. the struct pid the pidfd references is
used as PIDTYPE_TGID, i.e. is a thread-group leader.
The concern is for the future were we may want to enable pidfds to refer
to individual threads. Once that happens the passed in pidfd to e.g.
process_mrelease() or process_madvise() can refer to a struct pid that
is only used as PIDTYPE_PID and not as PIDTYPE_TGID, i.e. it might be a
pidfd refering to a non-threadgroup leader. Once that happens we want to
make sure that all users of pidfds are ok working with non-threadgroup
leaders. If we have on central helper that becomes a relatively simple
exercise in grepping and we're sure that all current callers use
PIDTYPE_TGID as they're using the helper. If we let places use
PIDTYPE_PID or PIDTYPE_TGID interchangeably this becomes a more arduous
task. So in a sense it's a bug-in-the-making. It's arguably fixes the
addition of process_mrelease() since I mentioned this pretty early on
and requested the addition of a helper as part of the patchset. I think
it just got lost in the reviews though.

> even separating out the fix?
> 
> > 
> > Link: https://lore.kernel.org/r/20211004125050.1153693-3-christian.brauner@xxxxxxxxxx
> > Cc: Vlastimil Babka <vbabka@xxxxxxx>
> > Cc: Suren Baghdasaryan <surenb@xxxxxxxxxx>
> > Cc: Matthew Bobrowski <repnop@xxxxxxxxxx>
> > Cc: Alexander Duyck <alexander.h.duyck@xxxxxxxxxxxxxxx>
> > Cc: David Hildenbrand <david@xxxxxxxxxx>
> > Cc: Jan Kara <jack@xxxxxxx>
> > Cc: Minchan Kim <minchan@xxxxxxxxxx>
> > Reviewed-by: Matthew Bobrowski <repnop@xxxxxxxxxx>
> > Signed-off-by: Christian Brauner <christian.brauner@xxxxxxxxxx>
> > ---
> > /* v2 */
> > unchanged
> 
> Acked-by: David Hildenbrand <david@xxxxxxxxxx>

Thanks!
Christian