On 11/19/24 9:27 AM, Bart Van Assche wrote:
> Fix a reference count leak in disk_zone_wplug_handle_error()
>
> Fixes: dd291d77cc90 ("block: Introduce zone write plugging")
> Signed-off-by: Bart Van Assche <bvanassche@xxxxxxx>
> ---
> block/blk-zoned.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/block/blk-zoned.c b/block/blk-zoned.c
> index 70211751df16..3346b8c53605 100644
> --- a/block/blk-zoned.c
> +++ b/block/blk-zoned.c
> @@ -1337,6 +1337,8 @@ static void disk_zone_wplug_handle_error(struct gendisk *disk,
>
> unlock:
> spin_unlock_irqrestore(&zwplug->lock, flags);
> +
> + disk_put_zone_wplug(zwplug);
The zone wplug put call is right after the single call site to
disk_zone_wplug_handle_error(). The reason it is *not* in that function is that
the reference on the wplug for handling an error is taken when the wplug is
added to the error list. disk_zone_wplug_handle_error() does not itself take a
reference on the wplug.
So how did you come up with this ? What workload/operation did you run to find
an issue ?
> }
>
> static void disk_zone_wplugs_work(struct work_struct *work)
--
Damien Le Moal
Western Digital Research
