Received-SPF: pass (google.com: domain of linux-ext4-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
From:   Andreas Dilger <adilger@dilger.ca>
Mime-Version: 1.0 (1.0)
Subject: Re: [RFCv1 01/72] e2fsck: Fix unbalanced mutex unlock for BOUNCE_MTX
Date:   Fri, 18 Nov 2022 07:20:55 -0600
Message-Id: <1BEDD834-2D4A-4E8E-936C-90DB5E322F9C@dilger.ca>
References: <20221118113711.qby7gtky5k36f7vd@riteshh-domain>
Cc:     Theodore Ts'o <tytso@mit.edu>, linux-ext4@vger.kernel.org,
        Harshad Shirwadkar <harshadshirwadkar@gmail.com>,
        Wang Shilong <wshilong@ddn.com>,
        Andreas Dilger <adilger.kernel@dilger.ca>, Li Xi <lixi@ddn.com>
In-Reply-To: <20221118113711.qby7gtky5k36f7vd@riteshh-domain>
To:     "Ritesh Harjani (IBM)" <ritesh.list@gmail.com>
Precedence: bulk

On Nov 18, 2022, at 05:37, Ritesh Harjani (IBM) <ritesh.list@gmail.com> wrot=
e:
>=20
> =EF=BB=BFOn 22/11/18 04:34AM, Andreas Dilger wrote:
>>> On Nov 7, 2022, at 06:22, Ritesh Harjani (IBM) <ritesh.list@gmail.com> w=
rote:
>>>=20
>>> f_crashdisk test failed with UNIX_IO_FORCE_BOUNCE=3Dyes due to unbalance=
d
>>> mutex unlock in below path.
>>>=20
>>> This patch fixes it.
>>>=20
>>> Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
>>> ---
>>> lib/ext2fs/unix_io.c | 1 -
>>> 1 file changed, 1 deletion(-)
>>>=20
>>> diff --git a/lib/ext2fs/unix_io.c b/lib/ext2fs/unix_io.c
>>> index e53db333..5b894826 100644
>>> --- a/lib/ext2fs/unix_io.c
>>> +++ b/lib/ext2fs/unix_io.c
>>> @@ -305,7 +305,6 @@ bounce_read:
>>>   while (size > 0) {
>>>       actual =3D read(data->dev, data->bounce, align_size);
>>>       if (actual !=3D align_size) {
>>> -            mutex_unlock(data, BOUNCE_MTX);
>>=20
>> This patch doesn't show enough context, but AFAIK this is jumping before m=
utex_down()
>> is called, so this *should* be correct as is?
>=20
> Thanks for the review, Andreas.
>=20
> Yeah, the patch diff above is not sufficient since it doesn't share enuf
> context.
> But essentially when "actual" is not equal to "align_size", then in this i=
f
> condition it goes to label "short_read:", which always goto error_unlock,
> where we anyways call mutex_unlock()
>=20
> Looking at a lot of labels in this function, this definitely looks like=20=

> something which can be cleaned up ("raw_read_blk()").=20
> I will add that to my list of todos.=20

You are correct, and it means this code is just not very clear to the reader=
. I think it
would make more sense to move the "short_read:" label to the end of the code=
:

                  actual =3D read(...);
                  if (actual !=3D size)
                          goto error_short_read;
                  goto success_unlock;
        :
                 actual =3D read(...);
                 if (actual !=3D align_size) {
                           actual =3D really_read;
                           buf -=3D really_read;
                           size +=3D really_read;
                           goto error_short_read;
                 }
        :
success_unlock:
        mutex_unlock(...);
        return 0;

error_short_read:
        if (actual < 0) {
                 retval =3D errno;
                 actual =3D 0;
        } else {
                 retval =3D EXT2_ET_SHORT_READ;
        }
error_unlock:
        mutex_unlock(...);

That way the code follows the normal error handling convention and is less l=
ikely to be
surprising to the reader.=20

Cheers, Andreas=