Re: [PATCH v4 02/10] user_events: Add minimal support for trace_event into ftrace

[Beau Belgrave]

On Mon, Nov 08, 2021 at 05:30:53PM -0500, Steven Rostedt wrote:
> On Mon, 8 Nov 2021 14:09:45 -0800
> Beau Belgrave <beaub@xxxxxxxxxxxxxxxxxxx> wrote:
> > 
> > It seems like both histograms and filter both reference field flags to
> > determine how to get the data.
> > 
> > How would you feel about another FILTER_* flag on fields, like:
> > FILTER_DYN_STRING_SAFE
> > FILTER_PTR_STRING_SAFE
> 
> You mean "UNSAFE" ?
> 

Yes :) Unsafe data, safe filter method.

> > 
> > user_events when parsing would instead of leaving FILTER_OTHER for
> > __data_loc / __rel_loc switch to the above.
> > 
> > The predicate filter method would then switch based on those types to
> > safer versions.
> > 
> > That way other parts could take advantage of this if needed beyond
> > user_events.
> > 
> > If this is addressed at the filter/histogram level, would then the write
> > callsites still check bounds per-write? Or maybe only care about the
> > undersized data cases?
> 
> I'd have to look at the implementation of this. There's too many variables
> running around in my head right now.
> 
> -- Steve

Understood, thanks for thinking about this.

-Beau