Re: [PATCH v2 09/16] btrfs: release BG lock before calling btrfs_link_bg_list()

[Boris Burkov <boris@bur.io>]

On Wed, Aug 13, 2025 at 03:34:51PM +0100, Mark Harmstone wrote:
> Release block_group->lock before calling btrfs_link_bg_list() in
> btrfs_delete_unused_bgs(), as this was causing lockdep issues.
> 
> This lock isn't held in any other place that we call btrfs_link_bg_list(), as
> the block group lists are manipulated while holding fs_info->unused_bgs_lock.
> 

Please include the offending lockdep output you are fixing.

Is this a generic fix unrelated to your other changes? I think a
separate patch from the series is clearer in that case. And it would
need a Fixes: tag (probably my patch that added btrfs_link_bg_list, haha)

Thanks.

> Signed-off-by: Mark Harmstone <mark@harmstone.com>
> ---
>  fs/btrfs/block-group.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/fs/btrfs/block-group.c b/fs/btrfs/block-group.c
> index bed9c58b6cbc..8c28f829547e 100644
> --- a/fs/btrfs/block-group.c
> +++ b/fs/btrfs/block-group.c
> @@ -1620,6 +1620,8 @@ void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
>  		if ((space_info->total_bytes - block_group->length < used &&
>  		     block_group->zone_unusable < block_group->length) ||
>  		    has_unwritten_metadata(block_group)) {
> +			spin_unlock(&block_group->lock);
> +
>  			/*
>  			 * Add a reference for the list, compensate for the ref
>  			 * drop under the "next" label for the
> @@ -1628,7 +1630,6 @@ void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
>  			btrfs_link_bg_list(block_group, &retry_list);
>  
>  			trace_btrfs_skip_unused_block_group(block_group);
> -			spin_unlock(&block_group->lock);
>  			spin_unlock(&space_info->lock);
>  			up_write(&space_info->groups_sem);
>  			goto next;
> -- 
> 2.49.1
>