Re: [RFC PATCH bpf-next 7/8] bpf: Support ->fill_link_info for perf_event

[Yafang Shao <laoar.shao@gmail.com>]

On Wed, May 31, 2023 at 8:37 AM Alexei Starovoitov
<alexei.starovoitov@gmail.com> wrote:
>
> On Sun, May 28, 2023 at 02:20:26PM +0000, Yafang Shao wrote:
> > By adding support for ->fill_link_info to the perf_event link, users will
> > be able to inspect it using `bpftool link show`. While users can currently
> > access this information via `bpftool perf show`, consolidating the link
> > information for all link types in one place would be more convenient.
> >
> > Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
> > ---
> >  include/uapi/linux/bpf.h       |  6 ++++++
> >  kernel/bpf/syscall.c           | 46 ++++++++++++++++++++++++++++++++++++++++++
> >  tools/include/uapi/linux/bpf.h |  6 ++++++
> >  3 files changed, 58 insertions(+)
> >
> > diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> > index 6be9b1d..1f2be1d 100644
> > --- a/include/uapi/linux/bpf.h
> > +++ b/include/uapi/linux/bpf.h
> > @@ -6438,6 +6438,12 @@ struct bpf_link_info {
> >                       __aligned_u64 addrs;
> >                       __u32 count;
> >               } kprobe_multi;
> > +             struct {
> > +                     __aligned_u64 name;
> > +                     __aligned_u64 addr;
>
> __aligned_u64 ? what is the reason?

It is because of the copy-and-paste.  Will use _u64 instead.

>
> > +                     __u32 name_len;
> > +                     __u32 offset;
> > +             } perf_event;
> >       };
> >  } __attribute__((aligned(8)));
> >
> > diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
> > index 33a72ec..b12707e 100644
> > --- a/kernel/bpf/syscall.c
> > +++ b/kernel/bpf/syscall.c
> > @@ -3329,10 +3329,56 @@ static void bpf_perf_link_show_fdinfo(const struct bpf_link *link,
> >       seq_printf(seq, "offset:\t%llu\n", probe_offset);
> >  }
> >
> > +static int bpf_perf_link_fill_link_info(const struct bpf_link *link,
> > +                                     struct bpf_link_info *info)
> > +{
> > +     struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
> > +     char __user *ubuf = u64_to_user_ptr(info->perf_event.name);
> > +     u32 ulen = info->perf_event.name_len;
> > +     const struct perf_event *event;
> > +     u64 probe_offset, probe_addr;
> > +     u32 prog_id, fd_type;
> > +     const char *buf;
> > +     size_t len;
> > +     int err;
> > +
> > +     if (!ulen ^ !ubuf)
> > +             return -EINVAL;
> > +     if (!ubuf)
> > +             return 0;
> > +
> > +     event = perf_get_event(perf_link->perf_file);
> > +     if (IS_ERR(event))
> > +             return PTR_ERR(event);
> > +
> > +     err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
> > +                                   &buf, &probe_offset,
> > +                                   &probe_addr);
> > +     if (err)
> > +             return err;
> > +
> > +     len = strlen(buf);
> > +     info->perf_event.name_len = len + 1;
>
> this use of name_len contradicts with patch 8.
> Is it 'in' or 'out' field?

My mistake. I should remove this sentence.
The reason I didn't do it the same with
bpf_raw_tp_link_fill_link_info() is that if we return the buf length
to the userspace when the ubuf is NULL, we have to call
bpf_get_perf_event_info() multiple times.

-- 
Regards
Yafang