Marat Khalili <[email protected]> wrote: > On 26/07/16 19:58, Stewart Brodie wrote: > > > > You won't be able to call those functions from a container not in the > > initial user namespace, even if you possess CAP_SYS_TIME, because of the > > way the kernel does its permission checks.
> I wonder if there's there really no workaround for ntpd? Special version > talking to the host through pipe probably? It is very convenient from > administration point of view to keep every network service in a separate > container. I am aware of two workarounds (apart from obviously just running ntpd in the host itself) As you suggest, you could try transporting the system calls through a pipe to something running on the host. It could work, but to me, that sounds awkward and introduces another level of stuff that has to be secured - and keeping things isolated and as secure as possible is presumably the reason why you want to run ntpd in a container in the first place! The workaround I use is to patch the kernel to permit the calls based only on whether the caller possesses CAP_SYS_TIME. I can only do that because I control all the software running on our embedded devices, though. I would strongly disrecommend doing that on anything where you don't control the entire machine. -- Stewart Brodie Senior Software Engineer Espial UK _______________________________________________ lxc-users mailing list [email protected] http://lists.linuxcontainers.org/listinfo/lxc-users
