pidns: Deny strange cases when creating pid namespaces.
authorEric W. Biederman <ebiederm@xmission.com>
Thu, 2 Aug 2012 15:35:35 +0000 (08:35 -0700)
committerEric W. Biederman <ebiederm@xmission.com>
Mon, 19 Nov 2012 13:59:13 +0000 (05:59 -0800)
commit225778d68d98e7cfe2579f8d8b2d7b76f8541b8b
treed4bc4b78c3e19821eaaffacc5341d02328bcd3c9
parentaf4b8a83add95ef40716401395b44a1b579965f4
pidns: Deny strange cases when creating pid namespaces.

task_active_pid_ns(current) != current->ns_proxy->pid_ns will
soon be allowed to support unshare and setns.

The definition of creating a child pid namespace when
task_active_pid_ns(current) != current->ns_proxy->pid_ns could be that
we create a child pid namespace of current->ns_proxy->pid_ns.  However
that leads to strange cases like trying to have a single process be
init in multiple pid namespaces, which is racy and hard to think
about.

The definition of creating a child pid namespace when
task_active_pid_ns(current) != current->ns_proxy->pid_ns could be that
we create a child pid namespace of task_active_pid_ns(current).  While
that seems less racy it does not provide any utility.

Therefore define the semantics of creating a child pid namespace when
task_active_pid_ns(current) != current->ns_proxy->pid_ns to be that the
pid namespace creation fails.  That is easy to implement and easy
to think about.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
kernel/pid_namespace.c