linux-kernel - Re: [PATCH block/for-4.5-fixes] writeback: keep superblock pinned during cgroup writeback association switches

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Wed, 17 Feb 2016 16:07:44 -0500
From:	Tejun Heo <tj@...nel.org>
To:	Jan Kara <jack@...e.cz>
Cc:	Jens Axboe <axboe@...nel.dk>, Tahsin Erdogan <tahsin@...gle.com>,
	cgroups@...r.kernel.org, Theodore Ts'o <tytso@....edu>,
	Nauman Rafique <nauman@...gle.com>,
	linux-kernel@...r.kernel.org, Jan Kara <jack@...e.com>
Subject: Re: [PATCH block/for-4.5-fixes] writeback: keep superblock pinned
 during cgroup writeback association switches

Hello, Jan.

On Wed, Feb 17, 2016 at 09:57:21PM +0100, Jan Kara wrote:
> Well, but this has the side-effect that trying to umount a filesystem while
> migrations are happening will result in EBUSY error. Without obvious reason
> why that happens. As an admin I would be rather upset when umount sometimes
> returns EBUSY without apparent reason and you have to basically implement a
> loop around umount to make it reliable. So a nack from me for this patch.

I see.  Can you please point me to the s_active check during umount?
I first tried s_umount but couldn't transfer its ownership to the
worker so ended up doing s_active.  I looked at how s_active is used
and couldn't find where it'd block umount.  may_umount() checks
mnt_count, not s_active, so it looked like holding s_active may delay
destruction of the superblock but not prevent umount.

> Traditionally, we have used sb->s_count and sb->s_umount semaphore to pin
> superblock while writeback code was working on it. That makes umount block
> until we can safely unmount the filesystem and thus doesn't result in these
> spurious EBUSY errors. But from a quick look this can be problematic for the
> cgroup setting.
> 
> Alternatively, you could either cancel all the switching work when
> unmounting filesystem or maybe just handle I_WB_SWITCH similarly to I_SYNC
> - don't grab inode reference when switching is going on, just make
> I_WB_SWITCH pin the inode and wait in evict() for it to be clear (similarly
> as we call inode_wait_for_writeback() there).

Yeah, this is an alternative but likely more involved.

Thanks.

-- 
tejun