From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Darrick J. Wong" Subject: Re: [PATCH 11/12] xfs: drop s_umount over opening the log and RT devices Date: Wed, 2 Aug 2023 09:32:19 -0700 Message-ID: <20230802163219.GW11352@frogsfrogsfrogs> References: <20230802154131.2221419-1-hch@lst.de> <20230802154131.2221419-12-hch@lst.de> Mime-Version: 1.0 Return-path: DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1690993939; bh=GcVu5AKxQx3Xueu/J47RQ71hv6d/lK6wYIHHmRfGqaQ=; h=Date:From:To:Cc:Subject:References:In-Reply-To:From; b=Wtd/B0tT5WFKMlZ4FcUuBOY403mH8jUVkWi+G6Rc6yRr8xqjh9e6QJP4M5cVaQZjl RdSRk087/R+zJKvlBUY6iflSrQVduZHiM7vhrntHc8ykJ5MZ3CsTWNydd8G60nAjjB xsTI+0r8XJWzqubQJge2bfArwRHdY+O0Fl3cJpBpiRZiCeObWknsnSeiOIYUNNlQ5n gF1Wsh0ENOOBgf0aIBZa5zNVBIhSdH4vc5MXOCvbYSInOKhXU+6gtVD2ABSpP8k6sE GxKNKflCh1H01pQA/zivBFr3jCjg3b+DK9HkonEjcCq0iyGozH4qOncit0g9T89zgx DasFxOweGYc3Q== Content-Disposition: inline In-Reply-To: <20230802154131.2221419-12-hch-jcswGhMUV9g@public.gmane.org> List-ID: Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: Christoph Hellwig Cc: Al Viro , Christian Brauner , Jan Kara , Chris Mason , Josef Bacik , David Sterba , Theodore Ts'o , Andreas Dilger , Jaegeuk Kim , Chao Yu , Ryusuke Konishi , Jens Axboe , linux-btrfs-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, linux-ext4-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, linux-f2fs-devel-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org, linux-nilfs-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, linux-fsdevel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, linux-xfs-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, linux-block-u79uwXL29TY76Z2rM5mHXA@public.gmane.org On Wed, Aug 02, 2023 at 05:41:30PM +0200, Christoph Hellwig wrote: > Just like get_tree_bdev needs to drop s_umount when opening the main > device, we need to do the same for the xfs log and RT devices to avoid a > potential lock order reversal with s_unmount for the mark_dead path. > > It might be preferable to just drop s_umount over ->fill_super entirely, > but that will require a fairly massive audit first, so we'll do the easy > version here first. > > Signed-off-by: Christoph Hellwig > --- > fs/xfs/xfs_super.c | 15 +++++++++++---- > 1 file changed, 11 insertions(+), 4 deletions(-) > > diff --git a/fs/xfs/xfs_super.c b/fs/xfs/xfs_super.c > index 8185102431301d..d5042419ed9997 100644 > --- a/fs/xfs/xfs_super.c > +++ b/fs/xfs/xfs_super.c > @@ -448,17 +448,21 @@ STATIC int > xfs_open_devices( > struct xfs_mount *mp) > { > - struct block_device *ddev = mp->m_super->s_bdev; > + struct super_block *sb = mp->m_super; > + struct block_device *ddev = sb->s_bdev; > struct block_device *logdev = NULL, *rtdev = NULL; > int error; > > + /* see get_tree_bdev why this is needed and safe */ Which part of get_tree_bdev? Is it this? /* * s_umount nests inside open_mutex during * __invalidate_device(). blkdev_put() acquires * open_mutex and can't be called under s_umount. Drop * s_umount temporarily. This is safe as we're * holding an active reference. */ up_write(&s->s_umount); blkdev_put(bdev, fc->fs_type); down_write(&s->s_umount); > + up_write(&sb->s_umount); > + > /* > * Open real time and log devices - order is important. > */ > if (mp->m_logname) { > error = xfs_blkdev_get(mp, mp->m_logname, &logdev); > if (error) > - return error; > + goto out_unlock; > } > > if (mp->m_rtname) { > @@ -496,7 +500,10 @@ xfs_open_devices( > mp->m_logdev_targp = mp->m_ddev_targp; > } > > - return 0; > + error = 0; > +out_unlock: > + down_write(&sb->s_umount); Isn't down_write taking s_umount? I think the label should be out_relock or something less misleading. --D > + return error; > > out_free_rtdev_targ: > if (mp->m_rtdev_targp) > @@ -508,7 +515,7 @@ xfs_open_devices( > out_close_logdev: > if (logdev && logdev != ddev) > xfs_blkdev_put(mp, logdev); > - return error; > + goto out_unlock; > } > > /* > -- > 2.39.2 >