Re: [PATCH v1 00/10] Move uid filtering to BPF filters

[Namhyung Kim <namhyung@kernel.org>]

On Wed, Feb 12, 2025 at 12:00:42PM -0800, Ian Rogers wrote:
> On Wed, Feb 12, 2025 at 10:46 AM Namhyung Kim <namhyung@kernel.org> wrote:
> > It's not completely broken and works sometimes.
> 
> No this is the definition of completely broken. If it only works
> sometimes then you can't use it, we can't put a test on it, there is
> no point in it. Even when it doesn't fail on perf_event_open, does it
> work for processes that start after /proc is scanned? No, it is
> completely broken.

Ok, we have a different definition for it.  Let's ignore the imaginary
users of the broken features.  Have you added a test for this change?

Anyway I've tested your change and found some issues:

1. It silently exited when BPF-skel is not built.  Better to put some
   messages at least.

  $ sudo ./perf record -u $(id -u) -- sleep 1

2. Even with BPF-skel, perf record doesn't work well.  It did something
   but failed to get sample data for some reason.

  $ sudo ./perf record -u $(id -u) -- sleep 1
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.045 MB perf.data ]
   
   Oh, I think you now need to pass -a because it now works in
   system-wide mode and drops samples for other users.

3. With BPF-skel, non-root users will see this.

  $ ./perf record -u $(id -u) -- sleep 1
  cannot get fd for 'filters' map
  failed to set filter "BPF" on event cycles:P with 13 (Permission denied)

   I think it's confusing and better to tell user that you need to run 
   'perf record --setup-filter pin' as root first.  But maybe due to the
   issue #2, you still need to run it as root.

Thanks,
Namhyung