Re: [PATCH bpf-next 2/2] bpf: Check link_create parameter for multi_uprobe

[Tao Chen]

在 2025/4/3 03:19, Jiri Olsa 写道:
> On Wed, Apr 02, 2025 at 11:01:48AM +0200, Jiri Olsa wrote:
> > On Tue, Apr 01, 2025 at 03:06:22PM -0700, Andrii Nakryiko wrote:
> > > On Tue, Apr 1, 2025 at 5:40 AM Tao Chen <chen.dylane@xxxxxxxxx> wrote:
> > > > 在 2025/4/1 19:03, Jiri Olsa 写道:
> > > > > On Mon, Mar 31, 2025 at 05:47:45PM +0800, Tao Chen wrote:
> > > > > > The target_fd and flags in link_create no used in multi_uprobe
> > > > > > , return -EINVAL if they assigned, keep it same as other link
> > > > > > attach apis.
> > > > > >
> > > > > > Fixes: 89ae89f53d20 ("bpf: Add multi uprobe link")
> > > > > > Signed-off-by: Tao Chen <chen.dylane@xxxxxxxxx>
> > > > > > ---
> > > > > >    kernel/trace/bpf_trace.c | 3 +++
> > > > > >    1 file changed, 3 insertions(+)
> > > > > >
> > > > > > diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
> > > > > > index 2f206a2a2..f7ebf17e3 100644
> > > > > > --- a/kernel/trace/bpf_trace.c
> > > > > > +++ b/kernel/trace/bpf_trace.c
> > > > > > @@ -3385,6 +3385,9 @@ int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr
> > > > > >       if (sizeof(u64) != sizeof(void *))
> > > > > >               return -EOPNOTSUPP;
> > > > > >
> > > > > > +    if (attr->link_create.target_fd || attr->link_create.flags)
> > > > > > +            return -EINVAL;
> > > > >
> > > > > I think the CI is failing because usdt code does uprobe multi detection
> > > > > with target_fd = -1 and it fails and perf-uprobe fallback will fail on
> > > > > not having enough file descriptors
> > > >
> > > > Hi jiri
> > > >
> > > > As you said, i found it, thanks.
> > > >
> > > > static int probe_uprobe_multi_link(int token_fd)
> > > > {
> > > >           LIBBPF_OPTS(bpf_prog_load_opts, load_opts,
> > > >                   .expected_attach_type = BPF_TRACE_UPROBE_MULTI,
> > > >                   .token_fd = token_fd,
> > > >                   .prog_flags = token_fd ? BPF_F_TOKEN_FD : 0,
> > > >           );
> > > >           LIBBPF_OPTS(bpf_link_create_opts, link_opts);
> > > >           struct bpf_insn insns[] = {
> > > >                   BPF_MOV64_IMM(BPF_REG_0, 0),
> > > >                   BPF_EXIT_INSN(),
> > > >           };
> > > >           int prog_fd, link_fd, err;
> > > >           unsigned long offset = 0;
> > > >
> > > >           prog_fd = bpf_prog_load(BPF_PROG_TYPE_KPROBE, NULL, "GPL",
> > > >                                   insns, ARRAY_SIZE(insns), &load_opts);
> > > >           if (prog_fd < 0)
> > > >                   return -errno;
> > > >
> > > >           /* Creating uprobe in '/' binary should fail with -EBADF. */
> > > >           link_opts.uprobe_multi.path = "/";
> > > >           link_opts.uprobe_multi.offsets = &offset;
> > > >           link_opts.uprobe_multi.cnt = 1;
> > > >
> > > >           link_fd = bpf_link_create(prog_fd, -1, BPF_TRACE_UPROBE_MULTI,
> > > > &link_opts);
> > > >
> > > > > but I think at this stage we will brake some user apps by introducing
> > > > > this check, link ebpf go library, which passes 0
> > > >
> > > > So is it ok just check the flags?
> > >
> > > good catch, Jiri! Yep, let's validate just flags?
> >
> > I think so.. I'll test that with ebpf/go to make sure we are safe
> > at least there ;-) I'll let you know
>
> sorry, got stuck.. link_create.flags are initialized to zero,
> so I think flags check should be fine (at least for ebpf/go)

Thank you very much for your detailed check. I will send it v2.

> jirka


--
Best Regards
Tao Chen