Re: [PATCH 2/2] md: create symlink with disk holder after mddev resume

On Tue, Feb 6, 2024 at 6:46 AM Linux regression tracking (Thorsten
Leemhuis) [off-list ref] wrote:

Hi, Thorsten here, the Linux kernel's regression tracker.

On 21.12.23 09:49, Yu Kuai wrote:

quoted

在 2023/12/21 15:11, linan666@huaweicloud.com 写道:

quoted

From: Li Nan <redacted>

There is a risk of deadlock when a process gets disk->open_mutex after
suspending mddev, because other processes may hold open_mutex while
submitting io. For example:
[...]

Nice catch! This patch looks good except that the new flag
'SymlinkCreated' doesn't look accurate, perhaps 'HolderLinked'
will make more sense.

quoted

Fix it by getting disk->open_mutex after mddev resume, iterating each
mddev->disk to create symlink for rdev which has not been created yet.
and moving bd_unlink_disk_holder() to mddev_unlock(), rdev has been
deleted from mddev->disks here, which can avoid concurrent bind and
unbind,

Fixes: 1b0a2d950ee2 ("md: use new apis to suspend array for ioctls
involed array reconfiguration")

Hey, what happened to that patch? It looks a lot like things stalled
here. I'm asking, because there is a regression report that claims
1b0a2d950ee2 to be the culprit that might or might not be causes by the
problem this patch tries to fix:
https://bugzilla.kernel.org/show_bug.cgi?id=218459

Thanks for the heads-up. Replied to the thread.

Song