On Tue, Jun 23, 2020 at 10:00 AM Song Liu <songliubraving@xxxxxx> wrote:
>
>
>
> > On Jun 23, 2020, at 8:19 AM, Alexei Starovoitov <alexei.starovoitov@xxxxxxxxx> wrote:
> >
> > On Tue, Jun 23, 2020 at 12:08 AM Song Liu <songliubraving@xxxxxx> wrote:
> >>
>
> [...]
>
> >>
> >> +BPF_CALL_3(bpf_get_task_stack_trace, struct task_struct *, task,
> >> + void *, entries, u32, size)
> >> +{
> >> + return stack_trace_save_tsk(task, (unsigned long *)entries, size, 0);
> >> +}
> >> +
> >> +static int bpf_get_task_stack_trace_btf_ids[5];
> >> +static const struct bpf_func_proto bpf_get_task_stack_trace_proto = {
> >> + .func = bpf_get_task_stack_trace,
> >> + .gpl_only = true,
> >
> > why?
>
> Actually, I am not sure when we should use gpl_only = true.
>
> >
> >> + .ret_type = RET_INTEGER,
> >> + .arg1_type = ARG_PTR_TO_BTF_ID,
> >> + .arg2_type = ARG_PTR_TO_MEM,
> >> + .arg3_type = ARG_CONST_SIZE_OR_ZERO,
> >
> > OR_ZERO ? why?
>
> Will fix.
I actually think it's a good idea, because it makes writing code that
uses variable-sized buffers easier. Remember how we had
bpf_perf_event_output() forcing size > 0? That was a major PITA and
required unnecessary code gymnastics to prove verifier it's OK (even
if zero size was never possible). Yonghong eventually fixed that to be
_OR_ZERO.
So if this is not causing any problems, please leave it as _OR_ZERO.
Thank you from everyone who had to suffer through dealing with
anything variable-sized in BPF!
>
> >
> >> + .btf_id = bpf_get_task_stack_trace_btf_ids,
> >> +};
> >> +
> >> static const struct bpf_func_proto *
> >> raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
> >> {
> >> @@ -1521,6 +1538,10 @@ tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
> >> return prog->expected_attach_type == BPF_TRACE_ITER ?
> >> &bpf_seq_write_proto :
> >> NULL;
> >> + case BPF_FUNC_get_task_stack_trace:
> >> + return prog->expected_attach_type == BPF_TRACE_ITER ?
> >> + &bpf_get_task_stack_trace_proto :
> >
> > why limit to iter only?
>
> I guess it is also useful for other types. Maybe move to bpf_tracing_func_proto()?
>
> >
> >> + *
> >> + * int bpf_get_task_stack_trace(struct task_struct *task, void *entries, u32 size)
> >> + * Description
> >> + * Save a task stack trace into array *entries*. This is a wrapper
> >> + * over stack_trace_save_tsk().
> >
> > size is not documented and looks wrong.
> > the verifier checks it in bytes, but it's consumed as number of u32s.
>
> I am not 100% sure, but verifier seems check it correctly. And I think it is consumed
> as u64s?
>
> Thanks,
> Song
>
