Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20230720134123.13148-1-lhenriques@suse.de>
In-Reply-To: <20230720134123.13148-1-lhenriques@suse.de>
From:   Filipe Manana <fdmanana@kernel.org>
Date:   Thu, 20 Jul 2023 17:15:13 +0100
Message-ID: <CAL3q7H4uqXttKMCucHH=tJDYkxOFuNRGR04ZSBD7eBMj4BE1iA@mail.gmail.com>
Subject: Re: [PATCH v2] btrfs: propagate error from function unpin_extent_cache()
To:     =?UTF-8?Q?Lu=C3=ADs_Henriques?= <lhenriques@suse.de>
Cc:     Chris Mason <clm@fb.com>, Josef Bacik <josef@toxicpanda.com>,
        David Sterba <dsterba@suse.com>,
        Johannes Thumshirn <Johannes.Thumshirn@wdc.com>,
        linux-btrfs@vger.kernel.org, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Thu, Jul 20, 2023 at 5:05=E2=80=AFPM Lu=C3=ADs Henriques <lhenriques@sus=
e.de> wrote:
>
> Function unpin_extent_cache() doesn't propagate an error if the call to
> lookup_extent_mapping() fails.  This patch adds an error return (EINVAL)
> and simply logs it in the only caller.
>
> Signed-off-by: Lu=C3=ADs Henriques <lhenriques@suse.de>
> ---
> Hi!
>
> As per David and Johannes reviews, I'm now proposing a different approach=
.
> Note that I kept the WARN_ON() instead of replacing it by an ASSERT().  I=
n
> fact, I considered removing the WARN_ON() completely and simply return th=
e
> error if em->start !=3D start.  But I guess it may useful for debug.
>
> Changes since v1:
> Instead of changing unpin_extent_cache() into a void function, make it
> propage an error code instead.
>
>  fs/btrfs/extent_map.c | 4 +++-
>  fs/btrfs/inode.c      | 8 ++++++--
>  2 files changed, 9 insertions(+), 3 deletions(-)
>
> diff --git a/fs/btrfs/extent_map.c b/fs/btrfs/extent_map.c
> index 0cdb3e86f29b..f4e7956edc05 100644
> --- a/fs/btrfs/extent_map.c
> +++ b/fs/btrfs/extent_map.c
> @@ -304,8 +304,10 @@ int unpin_extent_cache(struct extent_map_tree *tree,=
 u64 start, u64 len,
>
>         WARN_ON(!em || em->start !=3D start);
>
> -       if (!em)
> +       if (!em) {
> +               ret =3D -EINVAL;
>                 goto out;
> +       }
>
>         em->generation =3D gen;
>         clear_bit(EXTENT_FLAG_PINNED, &em->flags);
> diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
> index dbbb67293e34..21eb66fcc0df 100644
> --- a/fs/btrfs/inode.c
> +++ b/fs/btrfs/inode.c
> @@ -3273,8 +3273,12 @@ int btrfs_finish_one_ordered(struct btrfs_ordered_=
extent *ordered_extent)
>                                                 ordered_extent->disk_num_=
bytes);
>                 }
>         }
> -       unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offs=
et,
> -                          ordered_extent->num_bytes, trans->transid);
> +
> +       /* Proceed even if we fail to unpin extent from cache */
> +       if (unpin_extent_cache(&inode->extent_tree, ordered_extent->file_=
offset,
> +                              ordered_extent->num_bytes, trans->transid)=
 < 0)
> +               btrfs_warn(fs_info, "failed to unpin extent from cache");

Well, this is not very useful. It doesn't provide any more useful
information than what we get from the WARN_ON() at
unpin_extent_cache(), making the patch not useful.

This warning has actually happened a few times when running fstests
that exercise relocation (not sure if it's gone and accidently fixed
by something recently).
But to make this more useful, I would place the message at
unpin_extent_cache() with useful information such as:

- inode number
- id of the root the inode belongs to
- the file offset (the start argument) and extent length (or end offset)
- why the warning triggered: we didn't find the extent map or we found
one with a different start offset
- if we found an unexpected extent map, dump its flags (so we can see
if it happens only with compressed or prealloc extents for e.g.) and
other details (length/end offset for e.g.)

Thanks.

> +
>         if (ret < 0) {
>                 btrfs_abort_transaction(trans, ret);
>                 goto out;