Re: [PATCH bpf v2] selftests/bpf: Fix bpf-gcc CO-RE relocation in verifier_sockmap_mutate
From: Ihor Solodrai
Date: Thu Nov 19 2026 - 15:56:09 EST
On 11/17/26 5:24 AM, Jiayuan Chen wrote:
> bpf-gcc fails to load test_trace_iter_unix:
>
> Error: #766/6 verifier_sockmap_mutate/test_trace_iter_unix
> libbpf: elf: skipping section(3) .data (size 0)
> libbpf: elf: skipping unrecognized data section(18) .comment
> run_subtest:PASS:obj_open_mem 0 nsec
> libbpf: prog 'test_trace_iter_unix': relo #0: insn #0 (LDX/ST/STX) accesses field incorrectly. ...
> libbpf: prog 'test_trace_iter_unix': BPF program load failed: -EINVAL
> libbpf: prog 'test_trace_iter_unix': failed to load: -EINVAL
> libbpf: failed to load object 'verifier_sockmap_mutate'
> run_subtest:PASS:unexpected_load_success 0 nsec
> validate_msgs:FAIL:1014 expect_msg
> VERIFIER LOG:
> =============
> 0: R1=ctx() R10=fp0
> ; @ verifier_sockmap_mutate.c:124
> 0: <invalid CO-RE relocation>
> failed to resolve CO-RE relocation <byte_off> [4] struct bpf_iter__unix (0:0 @ offset 0)
> processed 1 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
> =============
> EXPECTED SUBSTR: 'cannot update sockmap in this context'
>
> GCC only applies preserve_access_index to types that carry the attribute
> themselves. The anonymous union holding the sock pointer doesn't, so the
> CO-RE access stops at the union ("0:0") instead of reaching the pointer
> ("0:0:0"), and libbpf rejects the relocation.
>
> Fixes: e16c83fa764f ("selftests/bpf: Add negative test for sockmap update from non-sockmap iter")
> Signed-off-by: Jiayuan Chen <jiayuan.chen@xxxxxxxxx>
Acked-by: Ihor Solodrai <ihor.solodrai@xxxxxxxxx>
CI confirms the patch works:
Failing: https://github.com/kernel-patches/bpf/actions/runs/44576883691/job/135760256786
Passing with the fix: https://github.com/kernel-patches/bpf/actions/runs/44083943280/job/134048457373
Thanks for the fix!
> [...]