Re: [PATCH v2 0/4] trace-cmd reset: Add option to preserve specific events

On Tue, 15 Oct 2024 14:30:58 +0100
Metin Kaya <[email protected]> wrote:

> On 14/10/2024 10:59 pm, Steven Rostedt wrote:
> > On Mon, 14 Oct 2024 13:31:32 +0100
> > Metin Kaya <[email protected]> wrote:
> >  
> >> Changes since v1:
> >> - Update man page for -k command line parameter of "trace-cmd reset".
> >> - Add tab completion support for "trace-cmd reset".  
> >
> > Thanks, I'll go ahead and apply this. But can you do me a favor, and add a
> > test case to utest/tracecmd-utest.c that tests this. It should create a
> > each of the types, start tracing, do the reset, and make sure that the -k
> > keeps what is expected. Doesn't need to be too fancy.  
> 
> I wrote the patch below, but noticed I'm unable to preserve uprobes
> during reset. I verified trace-cmd tries to prevent
> TRACEFS_DYNEVENT_UPROBE from being destroyed, but I cannot see them
> after running "trace-cmd reset -k uprobe" command.
> The same problem exists for uretprobe, too.
> I confirmed utest uses correct trace-cmd executable.
> 
> Any ideas?

Ug, that's because uprobes are defined in dynamic_events with a "p:", just
like kprobes. :-p  And the tracefs dynamic event parser flags uprobes as
kprobes. Thus, if we are deleting kprobes, we will delete uprobes too.

I need to fix that. We can't change the kernel as it's ABI unfortunately,
but there's other ways libtracefs can differentiate the two.

-- Steve