Re: confusing mountinfo output when bind-mounting files

From: Austin S. Hemmelgarn
Date: Mon Mar 21 2016 - 13:51:54 EST


On 2016-03-21 13:29, Tycho Andersen wrote:
On Mon, Mar 21, 2016 at 11:22:06AM -0600, Chris Murphy wrote:
On Mon, Mar 21, 2016 at 9:21 AM, Tycho Andersen
<tycho.andersen@xxxxxxxxxxxxx> wrote:
Hi all,

I'm seeing some strange behavior when bind mounting files from a btrfs
subvolume. Consider the output below:

root@criu2:/tmp# mount -o loop /tmp/tester.btrfs /tmp/dir1
root@criu2:/tmp# touch dir1/file
root@criu2:/tmp# sudo mount --bind dir1/file dir2/file
root@criu2:/tmp# grep "/tmp/dir" /proc/self/mountinfo
128 24 0:45 / /tmp/dir1 rw,relatime shared:107 - btrfs /dev/loop0 rw,space_cache,subvolid=5,subvol=/
129 24 0:45 /file /tmp/dir2/file rw,relatime shared:107 - btrfs /dev/loop0 rw,space_cache,subvolid=5,subvol=/file
root@criu2:/tmp# btrfs --version
btrfs-progs v4.4
root@criu2:/tmp# uname -a
Linux criu2 4.4.0-8-generic #23-Ubuntu SMP Wed Feb 24 20:45:30 UTC 2016 x86_64 x86_64 x86_64 GNU/Linux

The issue here is that the "subvol=" mount option for the target of the bind
mount is "/file" when no such subvolume actually exists. Is this
intended? It's confusing to say the least, but seems like a bug to me.

Since btrfs mount subvol=<name> is a bind mount behind the scene, I'm
not sure the mount info code distinguishes between bind mounts.

At the moment, this is something of a secret decoder ring where if you
see subvolid=5 first, then anything after that other than / is just
not true (can't be). Hence probably why both subvolid and subvol are
listed for now; you kinda have to parse them both.

Ok, so if I'm trying to compare superblocks from userspace, I should
look for subvolid and see if those match, and if that isn't present
then look for subvol?
That depends on what you mean by 'compare superblocks'.

Internal to the kernel, there is exactly one super-block structure shared between all mounts of a given BTRFS filesystem, regardless of subvolume or bind-mount (the same is actually also true if you mount a XFS or ext4 filesystem more than once on the same system), so if you're trying to differentiate filesystems, you need to match on something like the block device or label.

If however, you're trying to differentiate where something is mounted from, you should be matching on both subvolid and subvol (I think, but I may be wrong), but that may not always work (I don't use bind mounts of BTRFS filesystems often outside of subvolume mounts, so I don't have much experience with this particular case).