Re: [PATCH hotfix] mm: fix crashes from deferred split racing folio migration

[David Hildenbrand <david@redhat.com>]

On 03.07.24 16:30, Zi Yan wrote:
> On 2 Jul 2024, at 3:40, Hugh Dickins wrote:
> 
>> Even on 6.10-rc6, I've been seeing elusive "Bad page state"s (often on
>> flags when freeing, yet the flags shown are not bad: PG_locked had been
>> set and cleared??), and VM_BUG_ON_PAGE(page_ref_count(page) == 0)s from
>> deferred_split_scan()'s folio_put(), and a variety of other BUG and WARN
>> symptoms implying double free by deferred split and large folio migration.
>>
>> 6.7 commit 9bcef5973e31 ("mm: memcg: fix split queue list crash when large
>> folio migration") was right to fix the memcg-dependent locking broken in
>> 85ce2c517ade ("memcontrol: only transfer the memcg data for migration"),
>> but missed a subtlety of deferred_split_scan(): it moves folios to its own
>> local list to work on them without split_queue_lock, during which time
>> folio->_deferred_list is not empty, but even the "right" lock does nothing
>> to secure the folio and the list it is on.
>>
>> Fortunately, deferred_split_scan() is careful to use folio_try_get(): so
>> folio_migrate_mapping() can avoid the race by folio_undo_large_rmappable()
>> while the old folio's reference count is temporarily frozen to 0 - adding
>> such a freeze in the !mapping case too (originally, folio lock and
>> unmapping and no swap cache left an anon folio unreachable, so no freezing
>> was needed there: but the deferred split queue offers a way to reach it).
>>
>> Fixes: 9bcef5973e31 ("mm: memcg: fix split queue list crash when large folio migration")
>> Signed-off-by: Hugh Dickins <hughd@google.com>
>> Cc: stable@vger.kernel.org
>> ---
>> This patch against 6.10-rc6: Kefeng has commits in the mm-tree which
>> which will need adjustment to go over this, but we can both check the
>> result.  I have wondered whether just reverting 85ce2c517ade and its
>> subsequent fixups would be better: but that would be a bigger job,
>> and probably not the right choice.
>>
>>   mm/memcontrol.c | 11 -----------
>>   mm/migrate.c    | 13 +++++++++++++
>>   2 files changed, 13 insertions(+), 11 deletions(-)
>>
>> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
>> index 71fe2a95b8bd..8f2f1bb18c9c 100644
>> --- a/mm/memcontrol.c
>> +++ b/mm/memcontrol.c
>> @@ -7823,17 +7823,6 @@ void mem_cgroup_migrate(struct folio *old, struct folio *new)
>>
>>   	/* Transfer the charge and the css ref */
>>   	commit_charge(new, memcg);
>> -	/*
>> -	 * If the old folio is a large folio and is in the split queue, it needs
>> -	 * to be removed from the split queue now, in case getting an incorrect
>> -	 * split queue in destroy_large_folio() after the memcg of the old folio
>> -	 * is cleared.
>> -	 *
>> -	 * In addition, the old folio is about to be freed after migration, so
>> -	 * removing from the split queue a bit earlier seems reasonable.
>> -	 */
>> -	if (folio_test_large(old) && folio_test_large_rmappable(old))
>> -		folio_undo_large_rmappable(old);
>>   	old->memcg_data = 0;
>>   }
>>
>> diff --git a/mm/migrate.c b/mm/migrate.c
>> index 20cb9f5f7446..a8c6f466e33a 100644
>> --- a/mm/migrate.c
>> +++ b/mm/migrate.c
>> @@ -415,6 +415,15 @@ int folio_migrate_mapping(struct address_space *mapping,
>>   		if (folio_ref_count(folio) != expected_count)
>>   			return -EAGAIN;
>>
>> +		/* Take off deferred split queue while frozen and memcg set */
>> +		if (folio_test_large(folio) &&
>> +		    folio_test_large_rmappable(folio)) {
>> +			if (!folio_ref_freeze(folio, expected_count))
>> +				return -EAGAIN;
>> +			folio_undo_large_rmappable(folio);
>> +			folio_ref_unfreeze(folio, expected_count);
>> +		}
>> +
> 
> I wonder if the patch below would make the code look better by using
> the same freeze/unfreeze pattern like file-backed path. After
> reading the emails between you and Baolin and checking the code,
> I think the patch looks good to me. Feel free to add
> Reviewed-by: Zi Yan <ziy@nvidia.com>
> 
> BTW, this subtlety is very error prone, as Matthew, Ryan, and I all
> encountered errors because of this[1][2]. Matthew has a good summary
> of the subtlety:
> 
> "the (undocumented) logic in deferred_split_scan() that a folio
> with a positive refcount will not be removed from the list."
> 
> [1] https://lore.kernel.org/linux-mm/Ze9EFdFLXQEUVtKl@casper.infradead.org/
> [2] https://lore.kernel.org/linux-mm/Ze_P6xagdTbcu1Kz@casper.infradead.org/
> 
> diff --git a/mm/migrate.c b/mm/migrate.c
> index a8c6f466e33a..afcc0653dcb7 100644
> --- a/mm/migrate.c
> +++ b/mm/migrate.c
> @@ -412,17 +412,15 @@ int folio_migrate_mapping(struct address_space *mapping,
> 
>          if (!mapping) {
>                  /* Anonymous page without mapping */
> -               if (folio_ref_count(folio) != expected_count)
> +               if (!folio_ref_freeze(folio, expected_count))
>                          return -EAGAIN;
> 
>                  /* Take off deferred split queue while frozen and memcg set */
>                  if (folio_test_large(folio) &&
> -                   folio_test_large_rmappable(folio)) {
> -                       if (!folio_ref_freeze(folio, expected_count))
> -                               return -EAGAIN;
> +                   folio_test_large_rmappable(folio))
>                          folio_undo_large_rmappable(folio);
> -                       folio_ref_unfreeze(folio, expected_count);
> -               }
> +
> +               folio_ref_unfreeze(folio, expected_count);
> 

The downside is freezing order-0, where we don't need to freeze, right?

-- 
Cheers,

David / dhildenb