locking/futex: Check PF_KTHREAD rather than !p->mm to filter out kthreads

attach_to_pi_owner() checks p->mm to prevent attaching to kthreads and
this looks doubly wrong:

1. It should actually check PF_KTHREAD, kthread can do use_mm().

2. If this task is not kthread and it is actually the lock owner we can
   wrongly return -EPERM instead of -ESRCH or retry-if-EAGAIN.

   And note that this wrong EPERM is the likely case unless the exiting
   task is (auto)reaped quickly, we check ->mm before PF_EXITING.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Darren Hart <darren@dvhart.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Jerome Marchand <jmarchan@redhat.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mateusz Guzik <mguzik@redhat.com>
Link: http://lkml.kernel.org/r/20150202140536.GA26406@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

kernel/futex.c

diff --git a/kernel/futex.c b/kernel/futex.c
index 4eeb63d..1f6d646 100644
--- a/kernel/futex.c
+++ b/kernel/futex.c
@@ -900,7 +900,7 @@
 	if (!p)
 		return -ESRCH;
 
-	if (!p->mm) {
+	if (unlikely(p->flags & PF_KTHREAD)) {
 		put_task_struct(p);
 		return -EPERM;
 	}