Re: [PATCH] tracing: Fix the race between registering 'snapshot' event trigger and triggering 'snapshot' operation

[Xiao Yang]

On 2020/4/13 23:02, Steven Rostedt wrote:
> On Mon, 13 Apr 2020 15:12:52 +0800
> Xiao Yang<yangx.jy@xxxxxxxxxxxxxx>  wrote:
>
> > Traced event can trigger 'snapshot' operation(i.e. calls snapshot_trigger()
> > or snapshot_count_trigger()) when register_snapshot_trigger() has completed
> > registration but doesn't allocate spare buffer for 'snapshot' event trigger.
> > 'snapshot' operation always detects the lack of allocated buffer in the rare
> > case so make register_snapshot_trigger() allocate spare buffer first.
> >
> > trigger-snapshot.tc in kselftest reproduces the issue on slow vm:
> > -----------------------------------------------------------
> > cat trace
> > ...
> > ftracetest-3028  [002] ....   236.784290: sched_process_fork: comm=ftracetest pid=3028 child_comm=ftracetest child_pid=3036
> >       <...>-2875  [003] ....   240.460335: tracing_snapshot_instance_cond: *** SNAPSHOT NOT ALLOCATED ***
> >       <...>-2875  [003] ....   240.460338: tracing_snapshot_instance_cond: *** stopping trace here!   ***
> > -----------------------------------------------------------
> >
> > Signed-off-by: Xiao Yang<yangx.jy@xxxxxxxxxxxxxx>
> > ---
> >   kernel/trace/trace_events_trigger.c | 8 ++++++--
> >   1 file changed, 6 insertions(+), 2 deletions(-)
> >
> > diff --git a/kernel/trace/trace_events_trigger.c b/kernel/trace/trace_events_trigger.c
> > index dd34a1b46a86..00e54cdcef3e 100644
> > --- a/kernel/trace/trace_events_trigger.c
> > +++ b/kernel/trace/trace_events_trigger.c
> > @@ -1088,9 +1088,13 @@ register_snapshot_trigger(char *glob, struct event_trigger_ops *ops,
> >   			  struct event_trigger_data *data,
> >   			  struct trace_event_file *file)
> >   {
> > -	int ret = register_trigger(glob, ops, data, file);
> > +	int alloc_ret, ret;
> >
> > -	if (ret>  0&&  tracing_alloc_snapshot_instance(file->tr) != 0) {
> > +	alloc_ret = tracing_alloc_snapshot_instance(file->tr);
> > +
> > +	ret = register_trigger(glob, ops, data, file);
> > +
> > +	if (ret>  0&&  alloc_ret != 0) {
> >   		unregister_trigger(glob, ops, data, file);
> >   		ret = 0;
> >   	}
>
>
> Why register if the allocation failed? Just switch the logic:
>
> 	int ret = tracing_alloc_snapshot_instance(file->tr);
>
> 	if (ret != 0)
> 		return 0;
>
> 	return register_trigger(glob, ops, data, file);
Hi Steve,

It looks simpler and better.
I think we can drop the unnecessary ret variable as well.

Thanks,
Xiao Yang
> -- Steve
>
>
> .