[Lxc-users] Can't start containers

[Serge Hallyn]

Quoting Milan Zamazal (pdm at zamazal.org):
> >>>>> "SH" == Serge Hallyn <serge.hallyn at canonical.com> writes:
> 
>     SH> Does /proc/self/mounts on the host change after you successfully
>     SH> start the container the first time?
> 
> I examined /proc/self/mounts at the beginning and at the end of
> /etc/init.d/lxc script.  The only difference is:
> 
>   < cgroup /cgroup cgroup rw,relatime,perf_event,blkio,net_cls,freezer,devices,cpuacct,cpu,cpuset 0 0
>   ---
>   > cgroup /cgroup cgroup rw,relatime,perf_event,blkio,net_cls,freezer,devices,cpuacct,cpu,cpuset,clone_children 0 0

Hmm.  My hope was that I'd see some remnants of the container mounts
indicating that you had some MS_SHARED settings on your host mounts
which could help explain this.

> Nothing gets changed in /proc/self/mounts thereafter.
>     
>     SH> Can you do 'lxc-start -n test -l DEBUG -o test.debug', twice,
>     SH> and send us the resulting test.debug file?
> 
> The two debug outputs are attached.  There is no significant difference
> in the outputs until the pivot_root call.  Please note that I commented
> out the lxc.rootfs.mount option so the mounts happen at default location
> rather than in /mnt/lxc reported in my previous message.  My Linux
> version is 3.2.0 from Debian testing.

(You've probably mentioned this before, but) can you tell us exactly what
host distro+release you're on?  Any customizations?

I'd like to get to the bottom of this, but do not have time *right* now.
If you need to work around this, you could do worse than that simply
replace the pivot_root(new_root, old_root) call with chroot(new_root)
and get rid of the subsequent umount(old_root).

-serge