Re: [PATCH v2 2/3] md: Remove extra mddev_get() in md_seq_start()

[Paul Menzel]

Dear Logan,


Thank you for the updated series. One nit below, which could be 
corrected by the maintainer before applying.

Am 08.09.22 um 18:15 schrieb Logan Gunthorpe:
> A regression is seen where mddev devices stay permanently after they
> are stopped due to an elevated reference count.
>
> This was tracked down to an extra mddev_get() in md_seq_start().
>
> It only happened rarely because most of the time the md_seq_start()
> is called with a zero offset. The path with an extra mddev_get() only
> happens when it starts with a non-zero offset.
>
> The commit noted below changed an mddev_get() to check its success
> but inadevrtantly left the original call in. Remove the extra call.

inadvertently

> Fixes: 12a6caf27324 ("md: only delete entries from all_mddevs when the disk is freed")
> Signed-off-by: Logan Gunthorpe <logang@xxxxxxxxxxxx>
> Reviewed-by: Christoph Hellwig <hch@xxxxxx>
> Acked-by: Guoqing Jiang <Guoqing.jiang@xxxxxxxxx>
> ---
>   drivers/md/md.c | 1 -
>   1 file changed, 1 deletion(-)
>
> diff --git a/drivers/md/md.c b/drivers/md/md.c
> index afaf36b2f6ab..9dc0175280b4 100644
> --- a/drivers/md/md.c
> +++ b/drivers/md/md.c
> @@ -8154,7 +8154,6 @@ static void *md_seq_start(struct seq_file *seq, loff_t *pos)
>   	list_for_each(tmp,&all_mddevs)
>   		if (!l--) {
>   			mddev = list_entry(tmp, struct mddev, all_mddevs);
> -			mddev_get(mddev);
>   			if (!mddev_get(mddev))
>   				continue;
>   			spin_unlock(&all_mddevs_lock);


Kind regards,

Paul