[Devel] Re: [RFC][PATCH 7/7][v2] Define clone_with_pids syscall

[Serge E. Hallyn]

Quoting Sukadev Bhattiprolu (sukadev at linux.vnet.ibm.com):
> Serge E. Hallyn [serue at us.ibm.com] wrote:
> | Quoting Oren Laadan (orenl at cs.columbia.edu):
> | > In general, can a task figure out it's depth in the pid-ns hierarchy ?
> | 
> | In user-space?  No.
> 
> Not from any existing user interface. But hypothetically or in the
> long-term can/should sys_checkpoint() not be able to figure out if
> it has to C/R say a bash shell that has nested pid namespaces ?
> 
> I am guessing sys_checkpoint() can know if its crossing a pid-ns
> boundary by comparing or computing the max nesting level while
> walking a process tree. i.e if task_pid(task)->pid_ns->level is
> not the same for all process in the tree, then we have nested
> namespaces. How to checkpoint/restart such a tree is of course a
> bigger challenge.
> 
> But if sys_checkpoint() can find the max nesting level ? If that
> info is saved in checkpoint image, clone_with_pids() could
> use that info - no ?

If a container at pidns level 2, with 1 more pidns underneath it,
is checkpointed, then the checkpoint image should only reflect
the base and child pidns.  The two ancestor pidns levels should
not be there.

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