Skip to content

bpf: Allow access to perf sample data (v2)

Namhyung Kim bpf lkml 1 version In review

Get series with b4
b4 am [email protected]

Versions newest first

  1. v1 Latest2 patches 7 repliesIn review

Patches in v1 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/2bpf/perf: Call perf_prepare_sample() before bpf_prog_run()include/linux/bpf.h kernel/bpf/verifier.c kernel/events/core.c+5 -07 replies
2/2selftests/bpf: Add perf_event_read_sample test casestools/testing/selftests/bpf/prog_tests/perf_sample.c tools/testing/selftests/bpf/progs/test_perf_sample.c+200 -00 replies

Reviews trailers per patch

No Reviewed-by, Acked-by or Tested-by trailers for v1 yet.

Read the v1 thread

Cover letter [PATCH bpf-next 0/2] bpf: Allow access to perf sample data (v2)

Hello,

I'm working on perf event sample filtering using BPF. To do that BPF needs to access perf sample data and return 0 or 1 to drop or keep the samples.

Changes in v2)
- reuse perf_prepare_sample() instead of adding new bpf_prepare_sample()
- drop bpf_perf_event_read_helper() and access ctx->data directly using bpf_cast_to_kern_ctx().

v1) https://lore.kernel.org/r/[email protected]

Thanks to bpf_cast_to_kern_ctx() kfunc, it can easily access the sample data now. But the problem is that perf didn't populate the sample data at the time it calls bpf_prog_run(). I changed the code to simply call perf_prepare_sample function before calling the BPF program.

But it also checks if the BPF calls bpf_cast_to_kern_ctx() since calling perf_prepare_sample() is unnecessary if the BPF doesn't access to the sample. The perf_prepare_sample() was only called right before putting it to the perf ring buffer. I think I can add a little optimization not to fill already set fields as it can be called twice now. It can be a separate patch for perf.

Another issue is that perf sample data only has selected fields according to the sample_type flags in the perf_event_attr. Accessing other fields can result in uninitialized read. I'm not sure how much it's gonna be a problem but it seems there's no way to prevent it completely. So properly written programs should check the sample_type flags first when reading the sample data.

The code is available at 'bpf/perf-sample-v2' branch in

git://git.kernel.org/pub/scm/linux/kernel/git/namhyung/linux-perf.git

Thanks,
Namhyung

Namhyung Kim (2):
bpf/perf: Call perf_prepare_sample() before bpf_prog_run() selftests/bpf: Add perf_event_read_sample test cases

 include/linux/bpf.h                           |   1 +
 kernel/bpf/verifier.c                         |   1 +
 kernel/events/core.c                          |   3 +
 .../selftests/bpf/prog_tests/perf_sample.c    | 167 ++++++++++++++++++
 .../selftests/bpf/progs/test_perf_sample.c    |  33 ++++
 5 files changed, 205 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/perf_sample.c
 create mode 100644 tools/testing/selftests/bpf/progs/test_perf_sample.c
Signature · 1 line
2.39.0.314.g84b9a713c41-goog