Re: [PATCH 2/2] pidns: Expose task pid_ns_for_children to userspace

On 24.01.2017 00:49, Alban Crequy wrote:

On 14 January 2017 at 15:15, Kirill Tkhai [off-list ref] wrote:

quoted

For correct checkpointing/restoring of a task from userspace
it's need to know the task's pid_ns_for_children. Currently,
there is no a sane way to do that (the only possible trick
is to force the task create a new child and to analize the
child's /proc/[pid]/ns/pid link, that is performance-stupid).

The patch exposes pid_ns_for_children to ns directory
in standard way with the name "pid_for_children":

~# ls /proc/5531/ns -l | grep pid
lrwxrwxrwx 1 root root 0 Jan 14 16:38 pid -> pid:[4026531836]
lrwxrwxrwx 1 root root 0 Jan 14 16:38 pid_for_children -> pid:[4026532286]

Signed-off-by: Kirill Tkhai <redacted>

What's happening if a process, after unsharing CLONE_NEWPID, does not
fork but instead let another process open the new "pid_for_children"
and then setns()+fork()? Is that other process allowed to create the
"pid 1" in the new pid namespaces? Is that also allowed if the other
process lives in a sibling pid namespace? If so, that would break what
pid_namespaces(7) says:

    "the parental relationship between processes mirrors the parental
     relationship between PID namespaces: the parent of a process is
     either in the same namespace or resides in the immediate parent
     PID namespace."

You can setns() on a pid_ns only if your active pid_ns is a (grand)parent
for the target pid_ns. So, the situation you described is not possible.
See pidns_install() for the details.