[PATCH] f2fs: fix incorrect FI_NO_EXTENT handling in __destroy_extent_node()

[PATCH] f2fs: fix incorrect FI_NO_EXTENT handling in __destroy_extent_node()

[Yongpeng Yang]

From: yangyongpeng <yangyongpeng@xiaomi.com>

When __destroy_extent_node() sets the inode flag FI_NO_EXTENT, it does
not reset the length of the largest extent to 0 and update the inode
folio. Since modifications to the extent tree are disallowed afterward,
the cached largest extent may become stale. This can trigger the
following error in xfstests generic/388:

F2FS-fs (dm-0): sanity_check_extent_cache: inode (ino=1761) extent info [220057, 57, 6] is incorrect, run fsck to fix

In the f2fs_drop_inode path, __destroy_extent_node() does not need to
guarantee that et->node_cnt is 0, because concurrency with writeback
is expected in this path, and writeback may update the extent cache.

This patch updates __destroy_extent_node() to avoid setting the inode
flag FI_NO_EXTENT, and to remove the check zero of et->node_cnt.

Fixes: ed78aeebef05 ("f2fs: fix node_cnt race between extent node destroy and writeback")
Cc: stable@vger.kernel.org
Reported-by: Chao Yu <chao@kernel.org>
Suggested-by: Chao Yu <chao@kernel.org>
Signed-off-by: yangyongpeng <yangyongpeng@xiaomi.com>
---
 fs/f2fs/extent_cache.c | 4 ----
 1 file changed, 4 deletions(-)

diff --git a/fs/f2fs/extent_cache.c b/fs/f2fs/extent_cache.c
index 87169fd29d89..3adbead27953 100644
--- a/fs/f2fs/extent_cache.c
+++ b/fs/f2fs/extent_cache.c
@@ -645,14 +645,10 @@ static unsigned int __destroy_extent_node(struct inode *inode,
 
 	while (atomic_read(&et->node_cnt)) {
 		write_lock(&et->lock);
-		if (!is_inode_flag_set(inode, FI_NO_EXTENT))
-			set_inode_flag(inode, FI_NO_EXTENT);
 		node_cnt += __free_extent_tree(sbi, et, nr_shrink);
 		write_unlock(&et->lock);
 	}
 
-	f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
-
 	return node_cnt;
 }
 
-- 
2.43.0

Re: [PATCH] f2fs: fix incorrect FI_NO_EXTENT handling in __destroy_extent_node()

[Chao Yu]

On 4/22/2026 3:35 PM, Yongpeng Yang wrote:
> From: yangyongpeng <yangyongpeng@xiaomi.com>
> 
> When __destroy_extent_node() sets the inode flag FI_NO_EXTENT, it does
> not reset the length of the largest extent to 0 and update the inode
> folio. Since modifications to the extent tree are disallowed afterward,
> the cached largest extent may become stale. This can trigger the
> following error in xfstests generic/388:
> 
> F2FS-fs (dm-0): sanity_check_extent_cache: inode (ino=1761) extent info [220057, 57, 6] is incorrect, run fsck to fix
> 
> In the f2fs_drop_inode path, __destroy_extent_node() does not need to
> guarantee that et->node_cnt is 0, because concurrency with writeback
> is expected in this path, and writeback may update the extent cache.
> 
> This patch updates __destroy_extent_node() to avoid setting the inode
> flag FI_NO_EXTENT, and to remove the check zero of et->node_cnt.
> 
> Fixes: ed78aeebef05 ("f2fs: fix node_cnt race between extent node destroy and writeback")
> Cc: stable@vger.kernel.org
> Reported-by: Chao Yu <chao@kernel.org>
> Suggested-by: Chao Yu <chao@kernel.org>
> Signed-off-by: yangyongpeng <yangyongpeng@xiaomi.com>
> ---
>   fs/f2fs/extent_cache.c | 4 ----
>   1 file changed, 4 deletions(-)
> 
> diff --git a/fs/f2fs/extent_cache.c b/fs/f2fs/extent_cache.c
> index 87169fd29d89..3adbead27953 100644
> --- a/fs/f2fs/extent_cache.c
> +++ b/fs/f2fs/extent_cache.c
> @@ -645,14 +645,10 @@ static unsigned int __destroy_extent_node(struct inode *inode,
>   
>   	while (atomic_read(&et->node_cnt)) {
>   		write_lock(&et->lock);
> -		if (!is_inode_flag_set(inode, FI_NO_EXTENT))
> -			set_inode_flag(inode, FI_NO_EXTENT);

We'd better revert all change lines in "f2fs: fix node_cnt race between
extent node destroy and writeback"?

Thanks,

>   		node_cnt += __free_extent_tree(sbi, et, nr_shrink);
>   		write_unlock(&et->lock);
>   	}
>   
> -	f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
> -
>   	return node_cnt;
>   }
>