On 2026/1/16 12:20 [email protected] write:
> > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> > index 7a375f608263..223e5b483bbd 100644
> > --- a/kernel/bpf/verifier.c
> > +++ b/kernel/bpf/verifier.c
> > @@ -23321,8 +23321,18 @@ 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;
> > +
> > + if (!prog->aux->attach_func_proto)
> > + return -EINVAL;
> > + nr_args =
> > btf_type_vlen(prog->aux->attach_func_proto);
>
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>
> The attach_func_proto for raw tracepoints points to the BTF func_proto of
> __bpf_trace_* or btf_trace_* which has an extra "void *__data" as the
> first parameter. For example, a tracepoint declared with TP_PROTO(int a)
> results in a btf_trace_* typedef of "void (*)(void *__data, int a)" which
> has 2 parameters.
>
> Does btf_type_vlen() return the count including the hidden __data param?
> If so, a tracepoint with 1 actual argument would report nr_args=2, but
> the BPF context only contains 1 argument. Calling bpf_get_func_arg(ctx, 1)
> would pass the bounds check (1 < 2) but read out of bounds from ctx.
>
> The bpf_trace_run* functions populate args[] with only the actual
> tracepoint arguments (see __bpf_trace_run() in kernel/trace/bpf_trace.c),
> not including __data. Should nr_args be adjusted by subtracting 1 to
> account for the hidden first parameter?
You are right about this point too. The wrong selftests made me
ignore this point :/
Thanks!
Menglong Dong
>
> > + /* 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);
> > + }
>
> [ ... ]
>
> > @@ -23374,8 +23384,18 @@ static int do_misc_fixups(struct bpf_verifier_env
> > *env)
> > /* Implement get_func_arg_cnt inline. */
> > if (prog_type == BPF_PROG_TYPE_TRACING &&
> > insn->imm == BPF_FUNC_get_func_arg_cnt) {
> > + if (eatype == BPF_TRACE_RAW_TP) {
> > + int nr_args;
> > +
> > + if (!prog->aux->attach_func_proto)
> > + return -EINVAL;
> > + nr_args =
> > btf_type_vlen(prog->aux->attach_func_proto);
>
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>
> Same question here for bpf_get_func_arg_cnt() - should this subtract 1 to
> exclude the __data parameter from the count returned to BPF programs?
>
> > + /* Save nr_args to reg0 */
> > + insn_buf[0] = BPF_MOV64_IMM(BPF_REG_0, nr_args);
>
> [ ... ]
>
>
> ---
> 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/21055137351
>
