Re: [GIT PULL] tracing: final fixes for events and some

[Dave Jones <davej@redhat.com>]

On Tue, Aug 13, 2013 at 01:06:18PM +0200, Ingo Molnar wrote:
 > 
 > * Steven Rostedt <rostedt@goodmis.org> wrote:
 > 
 > > On Mon, 12 Aug 2013 22:39:22 -0400
 > > Dave Jones <davej@redhat.com> wrote:
 > > 
 > > 
 > > > I'm not sure why I got cc'd on this, but for my part, the perf/tracing 
 > > > related bugs I've found recently may have been there for ages, but 
 > > > they were triggerable as non-root users.
 > > 
 > > Which bug was able to trigger with non-root? The hash one that you 
 > > reported? That is something with the function tracer. Are you able to 
 > > enable function tracing as non-root?
 > > 
 > > Or is it because you gave more permissions for non-root to use extra 
 > > perf commands. Because no one but root should be able to enable function 
 > > tracing, as that can add a large overhead to the system.
 > 
 > Hm, also, tracepoints should in general only be accessible to root - all 
 > sorts of random privileged info leaks through tracepoints, a thorough 
 > review and sanitizing is needed to expose that to users. (and that does 
 > not consider the complication caused by exposing timing info.)

With the benefit of sleep, I might take that back, though I'm working
from memory of a bug that I've not seen in over a month.

I may have been chasing a user-triggerable bug, and used tracing (as root)
to diagnose and then walked into one of these.  

But the recent fix where you had a test-case that did module unloads
didn't really seem to fit the profile of what I was seeing.
It's feasible that my fuzzer can trigger module _loads_, but I
don't think there's any way we can trigger an rmmod.

	Dave