On Wed, Dec 3, 2025 at 8:34 PM Shuran Liu electronlsr@gmail.com wrote:
Hi Song,
Thanks for the review.
I don't get why we add this selftest here. It doesn't appear to be related to patch 1/2.
The regression that patch 1/2 fixes was originally hit by an LSM program calling bpf_d_path() from the bprm_check_security hook. The new subtest is a minimal reproducer for that scenario: without patch 1/2 the string comparison never matches due to verifier's faulty optimization, and with patch 1/2 it behaves correctly.
I somehow didn't reproduce this in my local tests. Thanks for the explanation.
Song