bug#21226: FAIL: tests/containers.scm

[Thompson, David]

[ Forgot to "Reply All".  Re-sending to bug tracker ]

On Mon, Aug 10, 2015 at 4:23 AM, Jochem Raat <address@hidden> wrote:
> On 10-08-15 02:10, Thompson, David wrote:
>> Hello Jochem,
>>
>> Thanks for the report.  I've done some work to fix these in our
>> master, but perhaps you have a situation that I haven't addressed.
>> The container functionality requires a relatively recent version (3.8,
>> I think) of Linux in order to work.
>>
>> What version of Linux are you using?
>
> I think I am running linux 3.13 ('uname -r' returns: 3.13.0-30-generic).
>
>> What is the output of 'ls -l /proc/self/ns'?
>
> 'ls -l /proc/self/ns' returns:
> total 0
> lrwxrwxrwx 1 jm jm 0 aug 10 10:12 ipc -> ipc:[4026531839]
> lrwxrwxrwx 1 jm jm 0 aug 10 10:12 mnt -> mnt:[4026531840]
> lrwxrwxrwx 1 jm jm 0 aug 10 10:12 net -> net:[4026531956]
> lrwxrwxrwx 1 jm jm 0 aug 10 10:12 pid -> pid:[4026531836]
> lrwxrwxrwx 1 jm jm 0 aug 10 10:12 user -> user:[4026531837]
> lrwxrwxrwx 1 jm jm 0 aug 10 10:12 uts -> uts:[4026531838]
>

Thanks.  So, you have a new enough kernel for all 6 user namespaces to
work but the 'setgroups' interface is not present.  I did some reading
in the user_namespaces(7) man page and found that using setgroups
became a requirement in Linux 3.19 and only kernels may not have it. I
took a look at an Ubuntu 14.04 machine which also runs a 3.13 kernel
and /proc/self/setgroups exists, so indeed it is an optional thing.
The fix will be to test if /proc/self/setgroups exists before writing
to it.  I'll have this fixed next time I get a chance to hack.

Thanks again for reporting this issue!

- Dave