Re: [PATCH 02/11] mm,migration: Do not try to migrate unmapped anonymous pages

[KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>]

On Mon, 15 Mar 2010 15:28:15 +0900
Minchan Kim <minchan.kim@gmail.com> wrote:

> On Mon, Mar 15, 2010 at 2:34 PM, KAMEZAWA Hiroyuki
> <kamezawa.hiroyu@jp.fujitsu.com> wrote:
> > On Mon, 15 Mar 2010 09:28:08 +0900
> > Minchan Kim <minchan.kim@gmail.com> wrote:
> >
> >> Hi, Mel.
> >> On Sat, Mar 13, 2010 at 1:41 AM, Mel Gorman <mel@csn.ul.ie> wrote:
> >> > rmap_walk_anon() was triggering errors in memory compaction that looks like
> >> > use-after-free errors in anon_vma. The problem appears to be that between
> >> > the page being isolated from the LRU and rcu_read_lock() being taken, the
> >> > mapcount of the page dropped to 0 and the anon_vma was freed. This patch
> >> > skips the migration of anon pages that are not mapped by anyone.
> >> >
> >> > Signed-off-by: Mel Gorman <mel@csn.ul.ie>
> >> > Acked-by: Rik van Riel <riel@redhat.com>
> >> > ---
> >> >  mm/migrate.c |   10 ++++++++++
> >> >  1 files changed, 10 insertions(+), 0 deletions(-)
> >> >
> >> > diff --git a/mm/migrate.c b/mm/migrate.c
> >> > index 98eaaf2..3c491e3 100644
> >> > --- a/mm/migrate.c
> >> > +++ b/mm/migrate.c
> >> > @@ -602,6 +602,16 @@ static int unmap_and_move(new_page_t get_new_page, unsigned long private,
> >> >         * just care Anon page here.
> >> >         */
> >> >        if (PageAnon(page)) {
> >> > +               /*
> >> > +                * If the page has no mappings any more, just bail. An
> >> > +                * unmapped anon page is likely to be freed soon but worse,
> >> > +                * it's possible its anon_vma disappeared between when
> >> > +                * the page was isolated and when we reached here while
> >> > +                * the RCU lock was not held
> >> > +                */
> >> > +               if (!page_mapcount(page))
> >>
> >> As looking code about mapcount of page, I got confused.
> >> I think mapcount of page is protected by pte lock.
> >> But I can't find pte lock in unmap_and_move.
> > There is no pte_lock.
> >
> >> If I am right, what protects race between this condition check and
> >> rcu_read_lock?
> >> This patch makes race window very small but It can't remove race totally.
> >>
> >> I think I am missing something.
> >> Pz, point me out. :)
> >>
> >
> > Hmm. This is my understanding of old story.
> >
> > At migration.
> >  1. we increase page_count().
> >  2. isolate it from LRU.
> >  3. call try_to_unmap() under rcu_read_lock(). Then,
> >  4. replace pte with swp_entry_t made by PFN. under pte_lock.
> >  5. do migarate
> >  6. remap new pages. under pte_lock()>
> >  7. release rcu_read_lock().
> >
> > Here, we don't care whether page->mapping holds valid anon_vma or not.
> >
> > Assume a racy threads which calls zap_pte_range() (or some other)
> >
> > a) When the thread finds valid pte under pte_lock and successfully call
> >   page_remove_rmap().
> >   In this case, migration thread finds try_to_unmap doesn't unmap any pte.
> >   Then, at 6, remap pte will not work.
> > b) When the thread finds migrateion PTE(as swap entry) in zap_page_range().
> >   In this case, migration doesn't find migrateion PTE and remap fails.
> >
> > Why rcu_read_lock() is necessary..
> >  - When page_mapcount() goes to 0, we shouldn't trust page->mapping is valid.
> >  - Possible cases are
> >        i) anon_vma (= page->mapping) is freed and used for other object.
> >        ii) anon_vma (= page->mapping) is freed
> >        iii) anon_vma (= page->mapping) is freed and used as anon_vma again.
> >
> > Here, anon_vma_cachep is created  by SLAB_DESTROY_BY_RCU. Then, possible cases
> > are only ii) and iii). While anon_vma is anon_vma, try_to_unmap and remap_page
> > can work well because of the list of vmas and address check. IOW, remap routine
> > just do nothing if anon_vma is freed.
> >
> > I'm not sure by what logic "use-after-free anon_vma" is caught. But yes,
> > there will be case, "anon_vma is touched after freed.", I think.
> >
> > Thanks,
> > -Kame
> >
> 
> Thanks for detail explanation, Kame.
> But it can't understand me enough, Sorry.
> 
> Mel said he met "use-after-free errors in anon_vma".
> So added the check in unmap_and_move.
> 
> if (PageAnon(page)) {
>  ....
>  if (!page_mapcount(page))
>    goto uncharge;
>  rcu_read_lock();
> 
> My concern what protects racy mapcount of the page?
> For example,
> 
> CPU A                                 CPU B
> unmap_and_move
> page_mapcount check pass    zap_pte_range
> <-- some stall -->                   pte_lock
> <-- some stall -->                   page_remove_rmap(map_count is zero!)
> <-- some stall -->                   pte_unlock
> <-- some stall -->                   anon_vma_unlink
> <-- some stall -->                   anon_vma free !!!!
> rcu_read_lock
> anon_vma has gone!!
> 
> I think above scenario make error "use-after-free", again.
> What prevent above scenario?
> 
I think this patch is not complete. 
I guess this patch in [1/11] is trigger for the race.
==
+
+	/* Drop an anon_vma reference if we took one */
+	if (anon_vma && atomic_dec_and_lock(&anon_vma->migrate_refcount, &anon_vma->lock)) {
+		int empty = list_empty(&anon_vma->head);
+		spin_unlock(&anon_vma->lock);
+		if (empty)
+			anon_vma_free(anon_vma);
+	}
==
If my understainding in above is correct, this "modify" freed anon_vma.
Then, use-after-free happens. (In old implementation, there are no refcnt,
so, there is no use-after-free ops.)


So, what I can think of now is a patch like following is necessary.

==
static inline struct anon_vma *anon_vma_alloc(void)
{
        struct anon_vma *anon_vma;
        anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL);
        atomic_set(&anon_vma->refcnt, 1);
}

void anon_vma_free(struct anon_vma *anon_vma)
{
        /*
         * This called when anon_vma is..
         * - anon_vma->vma_list becomes empty.
         * - incremetned refcnt while migration, ksm etc.. is dropped.
         * - allocated but unused.
         */
        if (atomic_dec_and_test(&anon_vma->refcnt))
                kmem_cache_free(anon_vma_cachep, anon_vma);
}
==
Then all things will go simple. 
Overhead is concern but list_empty() helps us much.

Thanks,
-Kame