Date: Tue, 2 Jul 2024 17:25:13 +0800
From: Baolin Wang <baolin.wang@...ux.alibaba.com>
To: Hugh Dickins <hughd@...gle.com>, Andrew Morton <akpm@...ux-foundation.org>
Cc: Nhat Pham <nphamcs@...il.com>, Yang Shi <shy828301@...il.com>,
 Zi Yan <ziy@...dia.com>, Barry Song <baohua@...nel.org>,
 Kefeng Wang <wangkefeng.wang@...wei.com>,
 David Hildenbrand <david@...hat.com>, Matthew Wilcox <willy@...radead.org>,
 linux-kernel@...r.kernel.org, linux-mm@...ck.org
Subject: Re: [PATCH hotfix] mm: fix crashes from deferred split racing folio
 migration



On 2024/7/2 15:40, 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).

Thanks Hugh.

But after reading your analysis, I am concerned that the 
folio_undo_large_rmappable() and deferred_split_scan() may still 
encounter a race condition with the local list, even with your patch.

Suppose folio A has already been queued into the local list in 
deferred_split_scan() by thread A, but fails to 'folio_trylock' and then 
releases the reference count. At the same time, folio A can be frozen by 
another thread B in folio_migrate_mapping(). In such a case, 
folio_undo_large_rmappable() would remove folio A from the local list 
without *any* lock protection, creating a race condition with the local 
list iteration in deferred_split_scan().

Anyway, I think this patch can still fix some possible races. Feel free 
to add:
Reviewed-by: Baolin Wang <baolin.wang@...ux.alibaba.com>

> 
> Fixes: 9bcef5973e31 ("mm: memcg: fix split queue list crash when large folio migration")
> Signed-off-by: Hugh Dickins <hughd@...gle.com>
> Cc: stable@...r.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);
> +		}
> +
>   		/* No turning back from here */
>   		newfolio->index = folio->index;
>   		newfolio->mapping = folio->mapping;
> @@ -433,6 +442,10 @@ int folio_migrate_mapping(struct address_space *mapping,
>   		return -EAGAIN;
>   	}
>   
> +	/* Take off deferred split queue while frozen and memcg set */
> +	if (folio_test_large(folio) && folio_test_large_rmappable(folio))
> +		folio_undo_large_rmappable(folio);
> +
>   	/*
>   	 * Now we know that no one else is looking at the folio:
>   	 * no turning back from here.

Powered by blists - more mailing lists