Re: [PATCH] mm/swap_state: update zswap LRU's protection range with the folio locked

[Nhat Pham <nphamcs@gmail.com>]

On Tue, Feb 6, 2024 at 7:15 AM Johannes Weiner <hannes@cmpxchg.org> wrote:
>
> On Mon, Feb 05, 2024 at 03:24:42PM -0800, Nhat Pham wrote:
> > Move the zswap LRU protection range update above the swap_read_folio()
> > call, and only when a new page is allocated. This is the case where
> > (z)swapin could happen, which is a signal that the zswap shrinker should
> > be more conservative with its reclaiming action.
> >
> > It also prevents a race, in which folio migration can clear the
> > memcg_data of the now unlocked folio, resulting in a warning in the
> > inlined folio_lruvec() call.
>
> The warning is the most probable outcome, and it will cause the update
> to go against the root cgroup which is safe at least.
>
> But AFAICS there is no ordering guarantee to rule out a UAF if the
> lookup succeeds but the memcg and lruvec get freed before the update.

Ah nice. I didn't consider that. IIUC, having the folio locked should
prevent this too. Based on the documentation:

 * For a non-kmem folio any of the following ensures folio and memcg binding
 * stability:
 *
 * - the folio lock

I'll rework the commit log to include this, and make this more prominent :)

>
> I think that part should be more prominent in the changelog. It's more
> important than the first paragraph. Consider somebody scrolling
> through the git log and trying to decide whether to backport or not;
> it's helpful to describe the bug and its impact first thing, then put
> the explanation of the fix after.
>
> > Reported-by: syzbot+17a611d10af7d18a7092@syzkaller.appspotmail.com
> > Closes: https://lore.kernel.org/all/000000000000ae47f90610803260@google.com/
> > Fixes: b5ba474f3f51 ("zswap: shrink zswap pool based on memory pressure")
> > Signed-off-by: Nhat Pham <nphamcs@gmail.com>
>
> Would it make sense to add
>
>         VM_WARN_ON_ONCE(!folio_test_locked(folio));
>
> to zswap_folio_swapin() as well?