Re: [PATCH 2/3] xfs_scrub: fix nr_dirs accounting problems

[Dave Chinner <david@fromorbit.com>]

On Mon, Aug 26, 2019 at 02:21:25PM -0700, Darrick J. Wong wrote:
> From: Darrick J. Wong <darrick.wong@oracle.com>
> 
> When we're scanning the directory tree, we bump nr_dirs every time we
> think we're going to queue a new directory to process, and we decrement
> it every time we're finished doing something with a directory
> (successful or not).  We forgot to undo a counter increment when
> workqueue_add fails, so refactor the code into helpers and call them
> as necessary for correct operation.
> 
> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> ---
>  scrub/vfs.c |   38 +++++++++++++++++++++++++++++---------
>  1 file changed, 29 insertions(+), 9 deletions(-)
> 
> 
> diff --git a/scrub/vfs.c b/scrub/vfs.c
> index ea2866d9..b358ab4a 100644
> --- a/scrub/vfs.c
> +++ b/scrub/vfs.c
> @@ -45,6 +45,32 @@ struct scan_fs_tree_dir {
>  
>  static void scan_fs_dir(struct workqueue *wq, xfs_agnumber_t agno, void *arg);
>  
> +/* Increment the number of directories that are queued for processing. */
> +static void
> +inc_nr_dirs(
> +	struct scan_fs_tree	*sft)
> +{
> +	pthread_mutex_lock(&sft->lock);
> +	sft->nr_dirs++;
> +	pthread_mutex_unlock(&sft->lock);
> +}
> +
> +/*
> + * Decrement the number of directories that are queued for processing and if
> + * we ran out of dirs to process, wake up anyone who was waiting for processing
> + * to finish.
> + */
> +static void
> +dec_nr_dirs(
> +	struct scan_fs_tree	*sft)
> +{
> +	pthread_mutex_lock(&sft->lock);
> +	sft->nr_dirs--;
> +	if (sft->nr_dirs == 0)
> +		pthread_cond_signal(&sft->wakeup);
> +	pthread_mutex_unlock(&sft->lock);
> +}
> +
>  /* Queue a directory for scanning. */
>  static bool
>  queue_subdir(
> @@ -73,11 +99,10 @@ queue_subdir(
>  	new_sftd->sft = sft;
>  	new_sftd->rootdir = is_rootdir;
>  
> -	pthread_mutex_lock(&sft->lock);
> -	sft->nr_dirs++;
> -	pthread_mutex_unlock(&sft->lock);
> +	inc_nr_dirs(sft);
>  	error = workqueue_add(wq, scan_fs_dir, 0, new_sftd);
>  	if (error) {
> +		dec_nr_dirs(sft);
>  		str_info(ctx, ctx->mntpoint,
>  _("Could not queue subdirectory scan work."));
>  		return false;

Ok, that's the bug fix for the previous patch. Potentially should be
a separate patch, but right now there is so much outstanding that I
don't think it's worthwhile to respin the series just to fix that.

Reviewed-by: Dave Chinner <dchinner@redhat.com>

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com