[google/syzkaller] a4ed0f: report/openbsd: also title KASAN faults raised fro...

[Greg Steuck]

Branch: refs/heads/gh-readonly-queue/master/pr-7603-2130f682322a1c2ef698e3c27a2b97e89e96142f
Home: https://github.com/google/syzkaller
Commit: a4ed0f4f0cf942f75e0d5bb54238189b2f5180d9
https://github.com/google/syzkaller/commit/a4ed0f4f0cf942f75e0d5bb54238189b2f5180d9
Author: Greg Steuck <gr...@nest.cx>
Date: 2026-07-15 (Wed, 15 Jul 2026)

Changed paths:
M pkg/report/openbsd.go
A pkg/report/testdata/openbsd/report/41
A pkg/report/testdata/openbsd/report/42

Log Message:
-----------
report/openbsd: also title KASAN faults raised from mem/str interceptors

The KASAN title-by-function format only skipped __asan_{load,store}N_noabort
frames, but the mem/str interceptors (kasan_memcpy/memmove/memset/memcmp in
subr_kasan.c) call kasan_shadow_check and panic DIRECTLY, with no __asan_
frame at all. Their frame (e.g. kasan_memcpy+0x1ec) was not in the skip list,
so the required helper-frame run failed to match. That had two bad outcomes,
both collapsing the live sysctl_sysvipc OOB write into a useless bucket:

- most reports fell back to the generic "Caught invalid memory access" title;
- a report that also captured an unrelated later backtrace (a second CPU
IPI-stopped mid page-fault, whose trace legitimately runs through
__asan_load4_noabort -> uvm_map_lookup_entry) matched THAT far-down __asan_
frame instead and mistitled the bug "in uvm_map_lookup_entry".

Add kasan_mem{cpy,move,set,cmp} to the skip list so the format matches the real
crash frame near the top of the report and names its caller. Testdata 41 is a
plain single-CPU kasan_memcpy report; testdata 42 is the two-backtrace report
above (now correctly titled sysctl_sysvipc, not uvm_map_lookup_entry).



To unsubscribe from these emails, change your notification settings at https://github.com/google/syzkaller/settings/notifications

[Greg Steuck]

Branch: refs/heads/master