uprobes: Fix utask->depth accounting in handle_trampoline()
Oleg Nesterov [Wed, 11 Sep 2013 15:47:26 +0000 (17:47 +0200)]
commit 878b5a6efd38030c7a90895dc8346e8fb1e09b4c upstream.

Currently utask->depth is simply the number of allocated/pending
return_instance's in uprobe_task->return_instances list.

handle_trampoline() should decrement this counter every time we
handle/free an instance, but due to typo it does this only if
->chained == T. This means that in the likely case this counter
is never decremented and the probed task can't report more than
MAX_URETPROBE_DEPTH events.

Reported-by: Mikhail Kulemin <Mikhail.Kulemin@ru.ibm.com>
Reported-by: Hemant Kumar Shaw <hkshaw@linux.vnet.ibm.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Anton Arapov <anton@redhat.com>
Cc: masami.hiramatsu.pt@hitachi.com
Cc: srikar@linux.vnet.ibm.com
Cc: systemtap@sourceware.org
Link: http://lkml.kernel.org/r/20130911154726.GA8093@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

kernel/events/uprobes.c

index f356974..ad8e1bd 100644 (file)
@@ -1682,12 +1682,10 @@ static bool handle_trampoline(struct pt_regs *regs)
                tmp = ri;
                ri = ri->next;
                kfree(tmp);
+               utask->depth--;
 
                if (!chained)
                        break;
-
-               utask->depth--;
-
                BUG_ON(!ri);
        }