fork: unify and tighten up CLONE_NEWUSER/CLONE_NEWPID checks
authorOleg Nesterov <oleg@redhat.com>
Wed, 11 Sep 2013 21:19:41 +0000 (14:19 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 11 Sep 2013 22:56:20 +0000 (15:56 -0700)
do_fork() denies CLONE_THREAD | CLONE_PARENT if NEWUSER | NEWPID.

Then later copy_process() denies CLONE_SIGHAND if the new process will
be in a different pid namespace (task_active_pid_ns() doesn't match
current->nsproxy->pid_ns).

This looks confusing and inconsistent.  CLONE_NEWPID is very similar to
the case when ->pid_ns was already unshared, we want the same
restrictions so copy_process() should also nack CLONE_PARENT.

And it would be better to deny CLONE_NEWUSER && CLONE_SIGHAND as well
just for consistency.

Kill the "CLONE_NEWUSER | CLONE_NEWPID" check in do_fork() and change
copy_process() to do the same check along with ->pid_ns check we already
have.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Andy Lutomirski <luto@amacapital.net>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Colin Walters <walters@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
kernel/fork.c

index 68d508f2bfba46e2c8e9405613eb508da34bdf08..84703db06cf355b072a5fcf6a3cfafb832d49b67 100644 (file)
@@ -1173,13 +1173,16 @@ static struct task_struct *copy_process(unsigned long clone_flags,
                return ERR_PTR(-EINVAL);
 
        /*
-        * If the new process will be in a different pid namespace don't
-        * allow it to share a thread group or signal handlers with the
-        * forking task.
+        * If the new process will be in a different pid or user namespace
+        * do not allow it to share a thread group or signal handlers or
+        * parent with the forking task.
         */
-       if ((clone_flags & CLONE_SIGHAND) && (task_active_pid_ns(current) !=
-                                       current->nsproxy->pid_ns_for_children))
-               return ERR_PTR(-EINVAL);
+       if (clone_flags & (CLONE_SIGHAND | CLONE_PARENT)) {
+               if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
+                   (task_active_pid_ns(current) !=
+                               current->nsproxy->pid_ns_for_children))
+                       return ERR_PTR(-EINVAL);
+       }
 
        retval = security_task_create(clone_flags);
        if (retval)
@@ -1575,15 +1578,6 @@ long do_fork(unsigned long clone_flags,
        int trace = 0;
        long nr;
 
-       /*
-        * Do some preliminary argument and permissions checking before we
-        * actually start allocating stuff
-        */
-       if (clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) {
-               if (clone_flags & (CLONE_THREAD|CLONE_PARENT))
-                       return -EINVAL;
-       }
-
        /*
         * Determine whether and which event to report to ptracer.  When
         * called from kernel_thread or CLONE_UNTRACED is explicitly