[lxc-devel] `top` within unprivileged container breaks host

[Mathias Gibbens]

On Fri, 2015-01-23 at 13:47 -0500, Stéphane Graber wrote:
> On Fri, Jan 23, 2015 at 06:19:46PM +0000, Serge Hallyn wrote:
> > Quoting Mathias Gibbens (mathias at calenhad.com):
> > > On Fri, 2015-01-23 at 00:48 +0000, Mathias Gibbens wrote:
> > > > Hello,
> > > > 
> > > >   I've been testing LXC 1.1-rc1 on a Debian jessie system. Specifically,
> > > > since all the pieces are now in place to support systemd in unprivileged
> > > > containers, I've been testing running Debian jessie within an
> > > > unprivileged container.
> > > > 
> > > >   The jessie container successfully starts (there are some errors along
> > > > the way, but that's another issue), and I can do a `lxc-attach`, add a
> > > > user and log in. However, if I run `top` within the container I only get
> > > > two lines of output:
> > > > 
> > > > top - 00:13:25 up 4 min,  0 users,  load average: 0.01, 0.05, 0.05
> > > > Tasks:  10 total,   1 running,   9 sleeping,   0 stopped,   0 zombie
> > > > 
> > > >   Furthermore, the host system breaks spectacularly:
> > > > 
> > > > lxc at lxc:~$ top
> > > > Error, do this: mount -t proc proc /proc
> > > > lxc at lxc:~$ /sbin/ifconfig
> > > > Warning: cannot open /proc/net/dev (No such file or directory). Limited
> > > > output.
> > > > lxc at lxc:~$ mount
> > > > mount: failed to read mtab: No such file or directory
> > > > lxc at lxc:~$ df
> > > > df: cannot read table of mounted file systems: No such file or directory
> > > > lxc at lxc:~$ sudo reboot
> > > > Running in chroot, ignoring request.
> > > > 
> > > >   It seems that /proc is being unmounted somehow on the host. I can
> > > > re-mount /proc on the host, and things seem to work again. (I haven't
> > > > tested too much after re-mounting, instead opting to just reboot the
> > > > host back to a known good state.)
> > > > 
> > > >   Any idea what is causing this to happen? I have also tried running
> > > > Ubuntu Vivid as an unprivileged container and see the same results. I
> > > > have not yet tried a different Linux distro for the host system.
> > > > 
> > > >   Host system details: Debian jessie with kernel 3.16.7-ckt2-1 x86_64,
> > > > systemd-215, LXC-1.1-rc1 and lxcfs built from current git checkout,
> > > > cgmanager-0.35 as packaged from the sid repository.
> > > > 
> > > > Thanks for any ideas,
> > > > Mathias
> > > 
> > >   This morning I setup a host running Ubuntu Vivid with cgmanager, lxc,
> > > and lxcfs packages installed from the daily PPA. I setup two
> > 
> > But did you install the systemd-sysv package?
> > 
> > > unprivileged containers, Debian jessie and Ubuntu Vivid, and ran `top`
> > > within each one. On the Vivid host, `top` runs properly and the host
> > > system does not break.
> > > 
> > >   So, it seems that the breakage is due to running jessie on the host
> > > system. Any ideas why this is happening, or where I could begin looking
> > > to debug this problem?
> > 
> > No, I've just reproduced it, but it requires systemd to do so.

  Ah, I had assumed that Vivid was using systemd by default, but didn't
check. pid1 in Vivid still defaults to upstart.

> > 
> > Note that you'll have to manually chown your name=systemd cgroup
> > to your user for the lxc-start to succeed.
> 
> 
> Hey,
> 
> So we've just tracked down that bug and it's a critical flaw in lxcfs
> which we've now fixed upstream. I tagged LXCFS 0.4 with the fix and am
> publishing it now on our website and uploading it to Ubuntu.
> 
> Thanks for reporting this!

  I can confirm that the fix in lxcfs resolves this issue and `top` now
operates as expected without trashing the host system.

Mathias
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part
URL: <http://lists.linuxcontainers.org/pipermail/lxc-devel/attachments/20150123/11540edf/attachment.sig>