On 2026/1/20 02:44, [email protected] wrote:
> > diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
> > index f73e08c223b5..64976009ba3c 100644
> > --- a/kernel/trace/bpf_trace.c
> > +++ b/kernel/trace/bpf_trace.c
> > @@ -1734,11 +1734,11 @@ tracing_prog_func_proto(enum bpf_func_id func_id,
> > const struct bpf_prog *prog)
> > case BPF_FUNC_d_path:
> > return &bpf_d_path_proto;
> > case BPF_FUNC_get_func_arg:
> > - return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_proto
> > : NULL;
> > + return &bpf_get_func_arg_proto;
>
> Removing the bpf_prog_has_trampoline() check here allows all TRACING
> program types to use bpf_get_func_arg(), not just RAW_TP and trampoline-
> based types. In particular, BPF_TRACE_ITER programs can now use this
> helper.
>
> However, looking at do_misc_fixups() in verifier.c, the inline code
> generation only handles BPF_TRACE_RAW_TP specially. For other non-
> trampoline types like BPF_TRACE_ITER, the else branch generates:
>
> BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -8)
>
> This loads nr_args from ctx-8, but ITER programs use a different context
> layout (e.g. struct bpf_iter__bpf_map_elem with meta, map, key, value
> pointers). There is no nr_args at ctx-8 for ITER programs.
>
> Would this cause ITER programs using bpf_get_func_arg() to read arbitrary
> memory at ctx-8, potentially leaking kernel data or causing incorrect
> bounds checks for subsequent argument access?
You are right. As Yonghong said, I missed the bpf_iter case, and it should
be skipped.
I'll fix it in the next version.
>
> > case BPF_FUNC_get_func_ret:
> > return bpf_prog_has_trampoline(prog) ? &bpf_get_func_ret_proto
> > : NULL;
> > case BPF_FUNC_get_func_arg_cnt:
> > - return bpf_prog_has_trampoline(prog) ?
> > &bpf_get_func_arg_cnt_proto : NULL;
> > + return &bpf_get_func_arg_cnt_proto;
>
> Same question applies here for bpf_get_func_arg_cnt() with ITER programs.
>
> > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> > index 9de0ec0c3ed9..0b281b7c41eb 100644
> > --- a/kernel/bpf/verifier.c
> > +++ b/kernel/bpf/verifier.c
> > @@ -23323,8 +23323,20 @@ static int do_misc_fixups(struct bpf_verifier_env
> > *env)
> > /* Implement bpf_get_func_arg inline. */
> > if (prog_type == BPF_PROG_TYPE_TRACING &&
> > insn->imm == BPF_FUNC_get_func_arg) {
> > + if (eatype == BPF_TRACE_RAW_TP) {
> > + int nr_args =
> > btf_type_vlen(prog->aux->attach_func_proto);
> > +
> > + /*
> > + * skip first 'void *__data' argument in
> > btf_trace_##name
> > + * typedef
> > + */
> > + nr_args--;
> > + /* Save nr_args to reg0 */
> > + insn_buf[0] = BPF_MOV64_IMM(BPF_REG_0, nr_args);
> > + } else {
> > + /* Load nr_args from ctx - 8 */
> > + insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0,
> > BPF_REG_1, -8);
> > + }
>
> This else branch now applies to BPF_TRACE_ITER as well, since the
> trampoline check was removed in tracing_prog_func_proto(). Should the
> condition be extended to also handle ITER, or should ITER be explicitly
> blocked from using these helpers?
>
> [ ... ]
>
>
> ---
> AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
> See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md
>
> CI run summary: https://github.com/kernel-patches/bpf/actions/runs/21147849826
>
