Re: [PATCH bpf-next 07/10] bpftool: expose attach_type-to-string array to non-cgroup code

From: Andrii Nakryiko <andrii.nakryiko@gmail.com>
To: Quentin Monnet <quentin@isovalent.com>
Cc: Andrii Nakryiko <andriin@fb.com>, bpf <bpf@vger.kernel.org>,
	Networking <netdev@vger.kernel.org>,
	Alexei Starovoitov <ast@fb.com>,
	Daniel Borkmann <daniel@iogearbox.net>,
	Kernel Team <kernel-team@fb.com>
Subject: Re: [PATCH bpf-next 07/10] bpftool: expose attach_type-to-string array to non-cgroup code
Date: Fri, 24 Apr 2020 09:27:22 -0700	[thread overview]
Message-ID: <CAEf4BzY9tjQm1f8eTyjYjthTF9n6tZ59r1mpUsYWL4+bFuch2Q@mail.gmail.com> (raw)
In-Reply-To: <34110254-6384-153f-af39-d5f9f3a50acb@isovalent.com>

On Fri, Apr 24, 2020 at 3:32 AM Quentin Monnet <quentin@isovalent.com> wrote:
>
> 2020-04-23 22:35 UTC-0700 ~ Andrii Nakryiko <andriin@fb.com>
> > Move attach_type_strings into main.h for access in non-cgroup code.
> > bpf_attach_type is used for non-cgroup attach types quite widely now. So also
> > complete missing string translations for non-cgroup attach types.
> >
> > Signed-off-by: Andrii Nakryiko <andriin@fb.com>
> > ---
> >  tools/bpf/bpftool/cgroup.c | 28 +++-------------------------
> >  tools/bpf/bpftool/main.h   | 32 ++++++++++++++++++++++++++++++++
> >  2 files changed, 35 insertions(+), 25 deletions(-)
> >
> > diff --git a/tools/bpf/bpftool/cgroup.c b/tools/bpf/bpftool/cgroup.c
> > index 62c6a1d7cd18..d1fd9c9f2690 100644
> > --- a/tools/bpf/bpftool/cgroup.c
> > +++ b/tools/bpf/bpftool/cgroup.c
> > @@ -31,35 +31,13 @@
> >
> >  static unsigned int query_flags;
> >
> > -static const char * const attach_type_strings[] = {
> > -     [BPF_CGROUP_INET_INGRESS] = "ingress",
> > -     [BPF_CGROUP_INET_EGRESS] = "egress",
> > -     [BPF_CGROUP_INET_SOCK_CREATE] = "sock_create",
> > -     [BPF_CGROUP_SOCK_OPS] = "sock_ops",
> > -     [BPF_CGROUP_DEVICE] = "device",
> > -     [BPF_CGROUP_INET4_BIND] = "bind4",
> > -     [BPF_CGROUP_INET6_BIND] = "bind6",
> > -     [BPF_CGROUP_INET4_CONNECT] = "connect4",
> > -     [BPF_CGROUP_INET6_CONNECT] = "connect6",
> > -     [BPF_CGROUP_INET4_POST_BIND] = "post_bind4",
> > -     [BPF_CGROUP_INET6_POST_BIND] = "post_bind6",
> > -     [BPF_CGROUP_UDP4_SENDMSG] = "sendmsg4",
> > -     [BPF_CGROUP_UDP6_SENDMSG] = "sendmsg6",
> > -     [BPF_CGROUP_SYSCTL] = "sysctl",
> > -     [BPF_CGROUP_UDP4_RECVMSG] = "recvmsg4",
> > -     [BPF_CGROUP_UDP6_RECVMSG] = "recvmsg6",
> > -     [BPF_CGROUP_GETSOCKOPT] = "getsockopt",
> > -     [BPF_CGROUP_SETSOCKOPT] = "setsockopt",
> > -     [__MAX_BPF_ATTACH_TYPE] = NULL,
>
> So you removed the "[__MAX_BPF_ATTACH_TYPE] = NULL" from the new array,
> if I understand correctly this is because all attach type enum members
> are now in the new attach_type_name[] so we're safe by looping until we
> reach __MAX_BPF_ATTACH_TYPE. Sounds good in theory but...
>

Well, NULL is default value, so having [__MAX_BPF_ATTACH_TYPE] = NULL
just increases ARRAY_SIZE(attach_type_names) by one. Which is
generally not needed, because we do proper < ARRAY_SIZE() checks
everywhere... except for one place. show_bpf_prog in cgroup.c looks up
name directly and can pass NULL into jsonw_string_field which will
crash.

I can fix that by setting [__MAX_BPF_ATTACH_TYPE] to "unknown" or
adding extra check in show_bpf_prog() code? Any preferences?

> > -};
> > -
> >  static enum bpf_attach_type parse_attach_type(const char *str)
> >  {
> >       enum bpf_attach_type type;
> >
> >       for (type = 0; type < __MAX_BPF_ATTACH_TYPE; type++) {
> > -             if (attach_type_strings[type] &&
> > -                 is_prefix(str, attach_type_strings[type]))
> > +             if (attach_type_name[type] &&
> > +                 is_prefix(str, attach_type_name[type]))
> >                       return type;
> >       }
>
> ... I'm concerned the "attach_type_name[type]" here could segfault if we
> add a new attach type to the kernel, but don't report it immediately to
> bpftool's array.

I don't think so. Here we'll iterate over all possible bpf_attach_type
(as far as our copy of UAPI header is concerned, of course). If some
of the values don't have entries in attach_type_name array, we'll get
back NULL (same as with explicit [__MAX_BPF_ATTACH_TYPE] = NULL, btw),
which will get handled properly in the loop. And caller will get back
__MAX_BPF_ATTACH_TYPE as bpf_attach_type value. So unless I'm still
missing something, it seems to be working exactly the same as before?

>
> Is there any drawback with keeping the "[__MAX_BPF_ATTACH_TYPE] = NULL"?
> Or change here to loop on ARRAY_SIZE(), as you do in your own patch for
> link?

ARRAY_SIZE() == __MAX_BPF_ATTACH_TYPE, isn't it? Previously ARRAY_SIZE
was (__MAX_BPF_ATTACH_TYPE + 1), but I don't think it's necessary?

The only difference is show_bpf_prog() which now is going to do out of
array reads, while previously it would get NULL. But both cases are
bad and needs fixing.