perf: Disallow user-space callchains for function trace events
authorJiri Olsa <jolsa@redhat.com>
Sun, 2 Mar 2014 15:56:38 +0000 (16:56 +0100)
committerIngo Molnar <mingo@kernel.org>
Tue, 11 Mar 2014 10:57:57 +0000 (11:57 +0100)
Recent issues with user space callchains processing within
page fault handler tracing showed as Peter said 'there's
just too much fail surface'.

Related list discussions:

  http://marc.info/?t=139302086500001&r=1&w=2
  http://marc.info/?t=139301437300003&r=1&w=2

Suggested-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1393775800-13524-2-git-send-email-jolsa@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/trace/trace_event_perf.c

index e854f420e033eb65a2bca233bb8df2e42778faf7..d5e01c3f4e69f292e3345791e7a977d29e26fabe 100644 (file)
@@ -31,9 +31,18 @@ static int perf_trace_event_perm(struct ftrace_event_call *tp_event,
        }
 
        /* The ftrace function trace is allowed only for root. */
-       if (ftrace_event_is_function(tp_event) &&
-           perf_paranoid_tracepoint_raw() && !capable(CAP_SYS_ADMIN))
-               return -EPERM;
+       if (ftrace_event_is_function(tp_event)) {
+               if (perf_paranoid_tracepoint_raw() && !capable(CAP_SYS_ADMIN))
+                       return -EPERM;
+
+               /*
+                * We don't allow user space callchains for  function trace
+                * event, due to issues with page faults while tracing page
+                * fault handler and its overall trickiness nature.
+                */
+               if (!p_event->attr.exclude_callchain_user)
+                       return -EINVAL;
+       }
 
        /* No tracing, just counting, so no obvious leak */
        if (!(p_event->attr.sample_type & PERF_SAMPLE_RAW))