Re: [PATCH RFC] btrfs: no need to transition to rdonly for ENOSPC error

From: Josef Bacik <josef@toxicpanda.com>
To: Anand Jain <anand.jain@oracle.com>, linux-btrfs@vger.kernel.org
Subject: Re: [PATCH RFC] btrfs: no need to transition to rdonly for ENOSPC error
Date: Tue, 12 Jan 2021 09:41:27 -0500	[thread overview]
Message-ID: <c5ccc5d5-c9e4-16ff-0b57-3b7dd2f3dfca@toxicpanda.com> (raw)
In-Reply-To: <169ba15cd316c181042bbe25e7d10c7963e3b7e8.1610444879.git.anand.jain@oracle.com>

On 1/12/21 5:34 AM, Anand Jain wrote:
> In the current kernel both scrub and balance fails due to ENOSPC,
> however there is no reason that it should be transitioned to the
> RDONLY and making free spaces difficult.
> 
> Signed-off-by: Anand Jain <anand.jain@oracle.com>
> ---
>   fs/btrfs/super.c | 3 +++
>   1 file changed, 3 insertions(+)
> 
> diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
> index 12d7d3be7cd4..8c1b858f55c4 100644
> --- a/fs/btrfs/super.c
> +++ b/fs/btrfs/super.c
> @@ -172,6 +172,9 @@ void __btrfs_handle_fs_error(struct btrfs_fs_info *fs_info, const char *function
>   	if (sb_rdonly(sb))
>   		return;
>   
> +	if (errno == -ENOSPC)
> +		return;
> +
>   	btrfs_discard_stop(fs_info);
>   
>   	/* btrfs handle error by forcing the filesystem readonly */
> 

Filipe and Qu are right, but there are cases where we probably shouldn't flip 
read only on ENOSPC, and we should address those at the specific spots where 
this happens.  Do you have an example?  If it's happening at btrfs_search_slot() 
time then we're not reserving enough space in whatever reserve we use, but if 
it's happening somewhere else, for example trying to mark a block group as read 
only and getting an ENOSPC there, we could probably skip flipping read only for 
that sort of case.  We need to address the specific scenarios you're hitting 
rather than blanket skipping the abort for ENOSPC, otherwise we'll get corrupt 
file systems.  Thanks,

Josef