function-graph: Init curr_ret_stack with ret_stack
authorSteven Rostedt <srostedt@redhat.com>
Sat, 13 Mar 2010 00:41:23 +0000 (19:41 -0500)
committerSteven Rostedt <rostedt@goodmis.org>
Sat, 13 Mar 2010 01:28:02 +0000 (20:28 -0500)
If the graph tracer is active, and a task is forked but the allocating of
the processes graph stack fails, it can cause crash later on.

This is due to the temporary stack being NULL, but the curr_ret_stack
variable is copied from the parent. If it is not -1, then in
ftrace_graph_probe_sched_switch() the following:

for (index = next->curr_ret_stack; index >= 0; index--)
next->ret_stack[index].calltime += timestamp;

Will cause a kernel OOPS.

Found with Li Zefan's ftrace_stress_test.

Cc: stable@kernel.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
kernel/trace/ftrace.c

index d4d1238b096bb1c54b2e35a0d3c564ad0b00f1b3..bb53edbb5c8c7fc7a8f4a4f2db423aa1512b2a43 100644 (file)
@@ -3349,6 +3349,7 @@ void ftrace_graph_init_task(struct task_struct *t)
 {
        /* Make sure we do not use the parent ret_stack */
        t->ret_stack = NULL;
+       t->curr_ret_stack = -1;
 
        if (ftrace_graph_active) {
                struct ftrace_ret_stack *ret_stack;
@@ -3358,7 +3359,6 @@ void ftrace_graph_init_task(struct task_struct *t)
                                GFP_KERNEL);
                if (!ret_stack)
                        return;
-               t->curr_ret_stack = -1;
                atomic_set(&t->tracing_graph_pause, 0);
                atomic_set(&t->trace_overrun, 0);
                t->ftrace_timestamp = 0;