Re: [PATCH] tracing: Check for NULL field_name in __synth_event_add_val()

[Tom Zanussi]

Hi Steve,

On Sat, 2023-02-18 at 10:59 -0500, Steven Rostedt wrote:
> From: "Steven Rostedt (Google)" <rostedt@xxxxxxxxxxx>
> 
> It is possible that the field_name passed into
> __synth_event_add_val() can
> be NULL with the trace_state set to add_name (possibly set from a
> previous
> call), in which case it needs to be checked.

Hmm, I don't think this really is possible, see below...

> 
> Cc: stable@xxxxxxxxxxxxxxx
> Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=217053
> Fixes: 8dcc53ad956d2 ("tracing: Add synth_event_trace() and related
> functions")
> Signed-off-by: Steven Rostedt (Google) <rostedt@xxxxxxxxxxx>
> ---
> 
> Tom, can you review this. Is there a legitimate case where you can
> have a
> previous call set "add_name" but the next call not require it? This
> patch
> assumes that it can't.
> 

No, because this code just above it makes sure you can't mix add_name
with add_next.  Once add_name is set it will return -EINVAL if
field_name is ever null after that, and add_name will never be changed
once set:

       /* can't mix add_next_synth_val() with add_synth_val() */
        if (field_name) {
                if (trace_state->add_next) {
                        ret = -EINVAL;
                        goto out;
                }
                trace_state->add_name = true;
        } else {
                if (trace_state->add_name) {
                        ret = -EINVAL;
                        goto out;
                }
                trace_state->add_next = true;
        }


>  kernel/trace/trace_events_synth.c | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/kernel/trace/trace_events_synth.c
> b/kernel/trace/trace_events_synth.c
> index 70bddb25d9c0..fa28c1da06d2 100644
> --- a/kernel/trace/trace_events_synth.c
> +++ b/kernel/trace/trace_events_synth.c
> @@ -1982,6 +1982,10 @@ static int __synth_event_add_val(const char
> *field_name, u64 val,
>  
>         event = trace_state->event;
>         if (trace_state->add_name) {
> +               if (!field_name) {
> +                       ret = -EINVAL;
> +                       goto out;
> +               }

So if add_name is set here, it must also mean that field_name can't be
null, because of the above.

>                 for (i = 0; i < event->n_fields; i++) {
>                         field = event->fields[i];
>                         if (strcmp(field->name, field_name) == 0)

And if field_name can't be null, then I don't see how this strcmp could
fail due to a null field_name.

So I don't see the need for this patch.  The bugzilla shows a compiler
warning when using -Wnonnull - could this just be a spurious gcc
warning?

Tom