On Thu, Mar 20, 2025 at 08:52:14AM +0100, Christoph Hellwig wrote:
> xfs_buf_free can call vunmap, which can sleep. The vunmap path is an
> unlikely one, so add might_sleep to ensure calling xfs_buf_free from
> atomic context gets caught more easily.
>
> Signed-off-by: Christoph Hellwig <hch@xxxxxx>
> ---
> fs/xfs/xfs_buf.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/fs/xfs/xfs_buf.c b/fs/xfs/xfs_buf.c
> index 8e7f1b324b3b..1a2b3f06fa71 100644
> --- a/fs/xfs/xfs_buf.c
> +++ b/fs/xfs/xfs_buf.c
> @@ -105,6 +105,7 @@ xfs_buf_free(
> {
> unsigned int size = BBTOB(bp->b_length);
>
> + might_sleep();
> trace_xfs_buf_free(bp, _RET_IP_);
>
> ASSERT(list_empty(&bp->b_lru));
If I followed it correct, vunmap can be caught via
xfs_buf_free_pages(). If that's the case, wouldn't make
more sense to put might_sleep() inside xfs_buf_free_pages()
giving it is not called only from xfs_buf_free()?
Otherwise,
Reviewed-by: Carlos Maiolino <cmaiolino@xxxxxxxxxx>
> --
> 2.45.2
>
