Re: [RFC][PATCH 3/4] tracing/kprobes: Fail to unregister if probeevent files are open

[Oleg Nesterov]

On 07/31, Steven Rostedt wrote:
>
> On Wed, 2013-07-03 at 23:33 -0400, Steven Rostedt wrote:
> > The above will corrupt the kprobe system, as the write to the enable
> > file will happen after the kprobe was deleted.
>
> Oleg,
>
> The above no longer triggers the bug due to your changes. The race is
> much tighter now

Yes, the changelog should be updated...

> and requires a process with the enable file opened and
> races with a write to enable it where the removal of the trace file
> checks the trace disabled, sees that it is, continues, but then the
> write enables it just as it gets deleted.

This should be fine. Either event_remove() path takes event_mutex
first and then ->write() fails, or ftrace_event_enable_disable()
actually disables this even successfully.

> Do you know of a way to trigger this bug?

I'll try to think more tomorrow, but most probably no. The race is
unlikely.

We need perf_trace_event_init() or ":enable_event:this-event" right
before trace_remove_event_call() takes the mutex.

And right after the caller (kprobes) checks "disabled".

> Hmm, what happens without this patch now? If it is active, and we delete
> it? It will call back into the kprobes and access a tracepoint that does
> not exist? Would this cause a crash?

I think yes, the crash is possible.

perf or FL_SOFT_MODE, this call/file has the external references, and
we are going to free it.

Oleg.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/