Re: [PATCH] stat: annotate the lockless inode block count snapshot

[Jan Kara]

On Wed 11-11-26 14:47:30, Zhu Jun wrote:
> generic_fillattr() reads i_blocks without inode->i_lock, while concurrent
> space accounting updates it under that lock. KCSAN reports the read
> racing with inode_add_bytes(), for example during a tmpfs write.
> 
> The block count returned by stat is a snapshot taken while file contents
> may change. Annotate this intentional lockless read with data_race(),
> without marking or weakening the locked space-accounting updates.
> 
> Fixes: 1da177e4c3f41 ("Linux-2.6.12-rc2")
> Signed-off-by: Zhu Jun <zhujun2@xxxxxxxxxxxxxxxxxxxx>

OK. Feel free to add:

Reviewed-by: Jan Kara <jack@xxxxxxx>

								Honza

> ---
>  fs/stat.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/fs/stat.c b/fs/stat.c
> index a2e63c6c45b7..6e6f99aaa628 100644
> --- a/fs/stat.c
> +++ b/fs/stat.c
> @@ -103,7 +103,8 @@ void generic_fillattr(const struct mnt_idmap *idmap, u32 request_mask,
>  	}
>  
>  	stat->blksize = i_blocksize(inode);
> -	stat->blocks = inode->i_blocks;
> +	/* The block count is a snapshot and may race with space accounting. */
> +	stat->blocks = data_race(inode->i_blocks);
>  
>  	if ((request_mask & STATX_CHANGE_COOKIE) && IS_I_VERSION(inode)) {
>  		stat->result_mask |= STATX_CHANGE_COOKIE;
> -- 
> 2.51.0
> 
> 
-- 
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR