Re: [PATCH] tracing/kprobe: Recover old array if fails to enable kprobe

[Oleg Nesterov <oleg@redhat.com>]

On 06/28, Steven Rostedt wrote:
>
> On Fri, 2013-06-28 at 16:27 +0200, Oleg Nesterov wrote:
> >
> > Ah, but this conflicts with the other changes I sent. They have
> > your acks, and iiuc Steven is going to apply them.
>
> I'll see if I can solve any conflicts. I need to get my -rt versions out
> and start on the new 3.6 stable today. Then after that, I plan on going
> though and getting all the tracing patches settled.

Thanks!

> > Besides, this fix is not complete afaics, we should also clear
> > TP_FLAG_TRACE/PROFILE if __enable_trace_probe() fails.

Yes.

And I forgot to mention, until we fix the races we discuss in another
thread, this WARN_ON() doesn't look right. So perhaps it would be
really better to delay this change a bit.

Oleg.

> > Perhaps you can do this later, on top of the pending changes?
> > 
> > Or. Given that this patch assumes that enable_kprobe() must succed,
> > can't we make a minimal change for now?
> > 
> > -------------------------------------------------------------------------------
> > [PATCH] tracing/kprobe: WARN() if enable_kprobe() fails.
> > 
> > enable_trace_probe() doesn't recover tp->files/flags if enable_kprobe()
> > fails, this looks confusing.
> > 
> > However, enable_kprobe() must not fail at this time except for unknown
> > bug or changing the implementation of enable_kprobe(), because usual
> > failure cases (not registered or gone) are already filtered.
> > 
> > So this patch simply adds WARN_ON(ret) to document this fact, even if
> > it makes sense to cleanup the logic anyway later.
> > 
> > Reported-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
> > Signed-off-by: Oleg Nesterov <oleg@redhat.com>
> > ---
> >  kernel/trace/trace_kprobe.c |    1 +
> >  1 files changed, 1 insertions(+), 0 deletions(-)
> > 
> > diff --git a/kernel/trace/trace_kprobe.c b/kernel/trace/trace_kprobe.c
> > index 5c070db..bb608b5 100644
> > --- a/kernel/trace/trace_kprobe.c
> > +++ b/kernel/trace/trace_kprobe.c
> > @@ -220,6 +220,7 @@ enable_trace_probe(struct trace_probe *tp, struct ftrace_event_file *file)
> >  			ret = enable_kretprobe(&tp->rp);
> >  		else
> >  			ret = enable_kprobe(&tp->rp.kp);
> > +		WARN_ON(ret);
> >  	}
> >   out:
> >  	return ret;
> 
>