On Mon, Aug 01, 2022 at 11:23:56AM -0700, Stanislav Fomichev wrote:
> On Mon, Aug 1, 2022 at 10:45 AM Martin KaFai Lau <kafai@xxxxxx> wrote:
> >
> > On Fri, Jul 29, 2022 at 05:08:09PM -0700, Stanislav Fomichev wrote:
> > > Apparently, no existing selftest covers it. Add a new one where
> > > we load cgroup/bind4 program and attach fentry to it.
> > > Calling bpf_obj_get_info_by_fd on the fentry program
> > > should return non-zero btf_id/btf_obj_id instead of crashing the kernel.
> > >
> > > Signed-off-by: Stanislav Fomichev <sdf@xxxxxxxxxx>
> > > ---
> > > .../selftests/bpf/prog_tests/attach_to_bpf.c | 109 ++++++++++++++++++
> > > .../selftests/bpf/progs/attach_to_bpf.c | 12 ++
> > > 2 files changed, 121 insertions(+)
> > > create mode 100644 tools/testing/selftests/bpf/prog_tests/attach_to_bpf.c
> > > create mode 100644 tools/testing/selftests/bpf/progs/attach_to_bpf.c
> > >
> > > diff --git a/tools/testing/selftests/bpf/prog_tests/attach_to_bpf.c b/tools/testing/selftests/bpf/prog_tests/attach_to_bpf.c
> > > new file mode 100644
> > > index 000000000000..fcf726c5ff0f
> > > --- /dev/null
> > > +++ b/tools/testing/selftests/bpf/prog_tests/attach_to_bpf.c
> > > @@ -0,0 +1,109 @@
> > > +// SPDX-License-Identifier: GPL-2.0
> > > +#define _GNU_SOURCE
> > > +#include <stdlib.h>
> > > +#include <bpf/btf.h>
> > > +#include <test_progs.h>
> > > +#include <network_helpers.h>
> > > +#include "attach_to_bpf.skel.h"
> > > +
> > > +char bpf_log_buf[BPF_LOG_BUF_SIZE];
> > static
>
> Will remove bpf_log_buf. Sent v2 too soon :-(
>
> > > +
> > > +static int find_prog_btf_id(const char *name, __u32 attach_prog_fd)
> > > +{
> > > + struct bpf_prog_info info = {};
> > > + __u32 info_len = sizeof(info);
> > > + struct btf *btf;
> > > + int err;
> > > +
> > > + err = bpf_obj_get_info_by_fd(attach_prog_fd, &info, &info_len);
> > > + if (err)
> > > + return err;
> > > +
> > > + if (!info.btf_id)
> > > + return -EINVAL;
> > > +
> > > + btf = btf__load_from_kernel_by_id(info.btf_id);
> > > + err = libbpf_get_error(btf);
> > > + if (err)
> > > + return err;
> > > +
> > > + err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
> > > + btf__free(btf);
> > > + if (err <= 0)
> > > + return err;
> > > +
> > > + return err;
> > > +}
> > > +
> > > +int load_fentry(int attach_prog_fd, int attach_btf_id)
> > static
>
> Thx!
>
> > > +{
> > > + LIBBPF_OPTS(bpf_prog_load_opts, opts,
> > > + .expected_attach_type = BPF_TRACE_FENTRY,
> > > + .attach_prog_fd = attach_prog_fd,
> > > + .attach_btf_id = attach_btf_id,
> > > + .log_buf = bpf_log_buf,
> > > + .log_size = sizeof(bpf_log_buf),
> > > + );
> > > + struct bpf_insn insns[] = {
> > > + BPF_MOV64_IMM(BPF_REG_0, 0),
> > > + BPF_EXIT_INSN(),
> > > + };
> > > + int ret;
> > > +
> > > + ret = bpf_prog_load(BPF_PROG_TYPE_TRACING,
> > > + "bind4_fentry",
> > > + "GPL",
> > > + insns,
> > > + ARRAY_SIZE(insns),
> > > + &opts);
> > > + if (ret)
> > > + printf("verifier log: %s\n", bpf_log_buf);
> > If this fentry prog is in the attach_to_bpf.c and load by skel, this printf
> > and the bpf_log_buf can go away. I wonder if it can use the '?' like
> > SEC("?cgroup/bind4") and SEC("?fentry"). Then opens attach_to_bpf.skel.h
> > twice and use bpf_program__set_autoload() to load individual program .
>
> Good ideal, let me try to see if doing "?fentry" is easier..
> (unless we agree to keep load_fentry, see below)
>
> > Another option could be to reuse the progs/bind4_prog.c and directly
> > put the fentry program in the attach_to_bpf.c.
> >
> > btw, this test feels like something that could be a few line
> > addition to the test_fexit_bpf2bpf_common() in fexit_bpf2bpf.c.
> > Adding one to test fentry into a cgroup bpf prog is also good.
> > No strong opinion here also.
>
> I was trying to reuse fexit_bpf2bpf initially but I sank too much time
> into it and decided that it might be easier to write a
> simpler/separate reproducer instead :-(
> How about we reuse progs/bind4_prog.c and keep load_fentry() ? And put
> this new test in fexit_bpf2bpf.c ?
Reusing bind4_prog.c sounds good. Either use bind4_prog.c in the
new prog_tests/attach_to_bpf.c (likely need a better name) or the
fexit_bpf2bpf.c is fine. Whatever makes more sense.
I was mostly thinking to avoid the special verifier log buf handling and
the printf here. If an empty fentry prog may look weird, may be just
skip the bpf_log_buf and printf since it will never fail and keep the
load_fentry ?
