pid_ns: Fix race between setns'ed fork() and zap_pid_ns_processes()
authorKirill Tkhai <ktkhai@virtuozzo.com>
Fri, 12 May 2017 16:11:31 +0000 (19:11 +0300)
committerEric W. Biederman <ebiederm@xmission.com>
Sat, 13 May 2017 22:26:02 +0000 (17:26 -0500)
Imagine we have a pid namespace and a task from its parent's pid_ns,
which made setns() to the pid namespace. The task is doing fork(),
while the pid namespace's child reaper is dying. We have the race
between them:

Task from parent pid_ns             Child reaper
copy_process()                      ..
  alloc_pid()                       ..
  ..                                zap_pid_ns_processes()
  ..                                  disable_pid_allocation()
  ..                                  read_lock(&tasklist_lock)
  ..                                  iterate over pids in pid_ns
  ..                                    kill tasks linked to pids
  ..                                  read_unlock(&tasklist_lock)
  write_lock_irq(&tasklist_lock);   ..
  attach_pid(p, PIDTYPE_PID);       ..
  ..                                ..

So, just created task p won't receive SIGKILL signal,
and the pid namespace will be in contradictory state.
Only manual kill will help there, but does the userspace
care about this? I suppose, the most users just inject
a task into a pid namespace and wait a SIGCHLD from it.

The patch fixes the problem. It simply checks for
(pid_ns->nr_hashed & PIDNS_HASH_ADDING) in copy_process().
We do it under the tasklist_lock, and can't skip
PIDNS_HASH_ADDING as noted by Oleg:

"zap_pid_ns_processes() does disable_pid_allocation()
and then takes tasklist_lock to kill the whole namespace.
Given that copy_process() checks PIDNS_HASH_ADDING
under write_lock(tasklist) they can't race;
if copy_process() takes this lock first, the new child will
be killed, otherwise copy_process() can't miss
the change in ->nr_hashed."

If allocation is disabled, we just return -ENOMEM
like it's made for such cases in alloc_pid().

v2: Do not move disable_pid_allocation(), do not
introduce a new variable in copy_process() and simplify
the patch as suggested by Oleg Nesterov.
Account the problem with double irq enabling
found by Eric W. Biederman.

Fixes: c876ad768215 ("pidns: Stop pid allocation when init dies")
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
CC: Andrew Morton <akpm@linux-foundation.org>
CC: Ingo Molnar <mingo@kernel.org>
CC: Peter Zijlstra <peterz@infradead.org>
CC: Oleg Nesterov <oleg@redhat.com>
CC: Mike Rapoport <rppt@linux.vnet.ibm.com>
CC: Michal Hocko <mhocko@suse.com>
CC: Andy Lutomirski <luto@kernel.org>
CC: "Eric W. Biederman" <ebiederm@xmission.com>
CC: Andrei Vagin <avagin@openvz.org>
CC: Cyrill Gorcunov <gorcunov@openvz.org>
CC: Serge Hallyn <serge@hallyn.com>
Cc: stable@vger.kernel.org
Acked-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
kernel/fork.c

index 06d759ab4c62ff090ab51715eaf0bc4fcabc250e..aa1076c5e4a9f3a5d9e6f58fef1c6f34e332de8c 100644 (file)
@@ -1845,11 +1845,13 @@ static __latent_entropy struct task_struct *copy_process(
        */
        recalc_sigpending();
        if (signal_pending(current)) {
-               spin_unlock(&current->sighand->siglock);
-               write_unlock_irq(&tasklist_lock);
                retval = -ERESTARTNOINTR;
                goto bad_fork_cancel_cgroup;
        }
+       if (unlikely(!(ns_of_pid(pid)->nr_hashed & PIDNS_HASH_ADDING))) {
+               retval = -ENOMEM;
+               goto bad_fork_cancel_cgroup;
+       }
 
        if (likely(p->pid)) {
                ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
@@ -1907,6 +1909,8 @@ static __latent_entropy struct task_struct *copy_process(
        return p;
 
 bad_fork_cancel_cgroup:
+       spin_unlock(&current->sighand->siglock);
+       write_unlock_irq(&tasklist_lock);
        cgroup_cancel_fork(p);
 bad_fork_free_pid:
        cgroup_threadgroup_change_end(current);