[Devel] Re: user-cr: Extra unshare() calls ?

[Serge E. Hallyn]

Quoting Sukadev Bhattiprolu (sukadev at linux.vnet.ibm.com):
> Sukadev Bhattiprolu [sukadev at linux.vnet.ibm.com] wrote:
> | Serge E. Hallyn [serue at us.ibm.com] wrote:
> | | Quoting Sukadev Bhattiprolu (sukadev at linux.vnet.ibm.com):
> | | > 
> | | > Came across this while testing LXC.
> | | > 
> | | > 
> | | > 1. Does ckpt_remount_proc() need to unshare() ? Or can we have the
> | | >    clone() that calls __ckpt_coordinator() clone with CLONE_NEWNS|CLONE_FS
> | | >    instead ?
> | | > 
> | | >    The problem with the unshare() in ckpt_remount_proc() is that it
> | | >    creates an extra level in cgroup hierarchy (see below) after restart.
> | | >    So applications expecting the cgroup hierarchy before chckpoint will
> | | >    be surprised.
> | | > 
> | | > 2. When --mount-pty (or --mntns) is specified, do we need to unshare() 
> | | >    in the parent process ? Considering only the full-container restart
> | | >    for now (ignore self-restart and subtree restart), can we just
> | | >    specify (CLONE_NEWNS|CLONE_FS) at the time of creating the first
> | | >    restarted process ?
> | | 
> | | And then move remounting of devpts into ckpt_remount_proc() called
> | | from __ckpt_coordinator() as well?
> | 
> | Yes, I missed that in the hack.
> 
> Here is a more involved fix, but not sure if there is a more efficient
> way to solve.

I don't know what you mean by more efficient - this looks good to me.
(Apart from probably also pulling the self-restart stuff off into a helper)

-serge
_______________________________________________
Containers mailing list
Containers at lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers