Received-SPF: pass (google.com: domain of linux-kernel+bounces-10853-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
Precedence: bulk
MIME-Version: 1.0
References: <20231221071109.1562530-1-linan666@huaweicloud.com>
 <20231221071109.1562530-2-linan666@huaweicloud.com> <CAPhsuW5SPiy5bsSfagYrSLa3JTn2Gw0VJKUToS2PS9h4w2=zSA@mail.gmail.com>
 <d00e6729-6e13-b1ab-0991-ce79b1245cfd@huaweicloud.com>
In-Reply-To: <d00e6729-6e13-b1ab-0991-ce79b1245cfd@huaweicloud.com>
From: Song Liu <song@kernel.org>
Date: Sun, 24 Dec 2023 17:11:02 -0800
Message-ID: <CAPhsuW6CBdM=8kmZDwQ2Ko4rN3cAW6z0=LpAOc+_t--FJZO9KQ@mail.gmail.com>
Subject: Re: [PATCH 1/2] md: fix WARN_ON if create symlink fail in bind_rdev_to_array()
To: Li Nan <linan666@huaweicloud.com>
Cc: yukuai3@huawei.com, linux-raid@vger.kernel.org, 
	linux-kernel@vger.kernel.org, yi.zhang@huawei.com, houtao1@huawei.com, 
	yangerkun@huawei.com
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Thu, Dec 21, 2023 at 5:17=E2=80=AFPM Li Nan <linan666@huaweicloud.com> w=
rote:
>
>
>
> =E5=9C=A8 2023/12/22 2:58, Song Liu =E5=86=99=E9=81=93:
[...]
> > In general, I would like to avoid adding flags if possible.
> >
>
> This flag is mainly used to fix deadlock in next patch. Or should we
> export bd_find_holder_disk()? Link hodler if it return NULL.
> just like:
>
>    rdev_for_each_rcu
>      if (!bd_find_holder_disk)
>        bd_link_disk_holder

I was thinking we will not need the flag if we fail bind_rdev_to_array()
on error (below).

>
>
> >>   };
> >>
> >>   static inline int is_badblock(struct md_rdev *rdev, sector_t s, int =
sectors,
> >> diff --git a/drivers/md/md.c b/drivers/md/md.c
> >> index e05858653a41..d6612b922c76 100644
> >> --- a/drivers/md/md.c
> >> +++ b/drivers/md/md.c
> >> @@ -2526,7 +2526,8 @@ static int bind_rdev_to_array(struct md_rdev *rd=
ev, struct mddev *mddev)
> >>                  sysfs_get_dirent_safe(rdev->kobj.sd, "bad_blocks");
> >>
> >>          list_add_rcu(&rdev->same_set, &mddev->disks);
> >> -       bd_link_disk_holder(rdev->bdev, mddev->gendisk);
> >> +       if (!bd_link_disk_holder(rdev->bdev, mddev->gendisk))
> >> +               set_bit(SymlinkCreated, &rdev->flags);
> >
> > Shall we just fail bind_rdev_to_array() if bd_link_disk_holder()
> > returns non-zero?
> >
>
> I keep this action because of commit 00bcb4ac7ee7 ("md: reduce
> dependence on sysfs."). Fail bind_rdev_to_array is good to me.

I wonder whether the assumption in 00bcb4ac7ee7 is still true. If
bd_link_disk_holder() fails for valid reasons, we need to handle it
properly (set a flag, check the flag on unlink, etc.). If we only fail
bd_link_disk_holder() on extreme cases (ENOMEM, etc.), we can
just fail bind_rdev_to_array().

Thanks,
Song