[PATCH v5] f2fs: use post-decrement count for cp_wait wakeup

[PATCH v5] f2fs: use post-decrement count for cp_wait wakeup

[Wenjie Qi]

f2fs_write_end_io() decrements the writeback page counter and then
reads it again with get_pages() to decide whether the last
F2FS_WB_CP_DATA completion should wake cp_wait.

Use atomic_dec_return() for F2FS_WB_CP_DATA completions so the wakeup
decision is made from the value produced by the decrement itself. Keep
the existing dec_page_count() path for other writeback counters.

Fixes: e234088758fc ("f2fs: avoid wait if IO end up when do_checkpoint for better performance")
Fixes: ce2739e482bc ("f2fs: fix to avoid UAF in f2fs_write_end_io()")
Cc: stable@vger.kernel.org
Signed-off-by: Wenjie Qi <qiwenjie@xiaomi.com>
---
 fs/f2fs/data.c | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index d83a21998ec2..58d23eb74ec2 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -392,15 +392,17 @@ static void f2fs_write_end_io(struct bio *bio)
 		if (f2fs_in_warm_node_list(folio))
 			f2fs_del_fsync_node_entry(sbi, folio);
 
-		dec_page_count(sbi, type);
-
 		/*
 		 * we should access sbi before folio_end_writeback() to
 		 * avoid racing w/ kill_f2fs_super()
 		 */
-		if (type == F2FS_WB_CP_DATA && !get_pages(sbi, type) &&
-				wq_has_sleeper(&sbi->cp_wait))
-			wake_up(&sbi->cp_wait);
+		if (type == F2FS_WB_CP_DATA) {
+			if (!atomic_dec_return(&sbi->nr_pages[type]) &&
+			    wq_has_sleeper(&sbi->cp_wait))
+				wake_up(&sbi->cp_wait);
+		} else {
+			dec_page_count(sbi, type);
+		}
 
 		folio_clear_f2fs_gcing(folio);
 		folio_end_writeback(folio);

base-commit: c0b65f6129c7fbb526e921dd60261650f1b2bef9
-- 
2.43.0

Re: [f2fs-dev] [PATCH v5] f2fs: use post-decrement count for cp_wait wakeup

[Jaegeuk Kim]

On 06/16, Wenjie Qi wrote:
> f2fs_write_end_io() decrements the writeback page counter and then
> reads it again with get_pages() to decide whether the last
> F2FS_WB_CP_DATA completion should wake cp_wait.
> 
> Use atomic_dec_return() for F2FS_WB_CP_DATA completions so the wakeup
> decision is made from the value produced by the decrement itself. Keep
> the existing dec_page_count() path for other writeback counters.

Is there a race condition to do this? If so, can you describe? And, I think
we need a wrapper function instead of calling nr_pages directly.

> 
> Fixes: e234088758fc ("f2fs: avoid wait if IO end up when do_checkpoint for better performance")
> Fixes: ce2739e482bc ("f2fs: fix to avoid UAF in f2fs_write_end_io()")
> Cc: stable@vger.kernel.org
> Signed-off-by: Wenjie Qi <qiwenjie@xiaomi.com>
> ---
>  fs/f2fs/data.c | 12 +++++++-----
>  1 file changed, 7 insertions(+), 5 deletions(-)
> 
> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> index d83a21998ec2..58d23eb74ec2 100644
> --- a/fs/f2fs/data.c
> +++ b/fs/f2fs/data.c
> @@ -392,15 +392,17 @@ static void f2fs_write_end_io(struct bio *bio)
>  		if (f2fs_in_warm_node_list(folio))
>  			f2fs_del_fsync_node_entry(sbi, folio);
>  
> -		dec_page_count(sbi, type);
> -
>  		/*
>  		 * we should access sbi before folio_end_writeback() to
>  		 * avoid racing w/ kill_f2fs_super()
>  		 */
> -		if (type == F2FS_WB_CP_DATA && !get_pages(sbi, type) &&
> -				wq_has_sleeper(&sbi->cp_wait))
> -			wake_up(&sbi->cp_wait);
> +		if (type == F2FS_WB_CP_DATA) {
> +			if (!atomic_dec_return(&sbi->nr_pages[type]) &&
> +			    wq_has_sleeper(&sbi->cp_wait))
> +				wake_up(&sbi->cp_wait);
> +		} else {
> +			dec_page_count(sbi, type);
> +		}
>  
>  		folio_clear_f2fs_gcing(folio);
>  		folio_end_writeback(folio);
> 
> base-commit: c0b65f6129c7fbb526e921dd60261650f1b2bef9
> -- 
> 2.43.0
> 
> 
> 
> _______________________________________________
> Linux-f2fs-devel mailing list
> Linux-f2fs-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel