netdev - Re: [PATCH 0/3] perf/bpftool: Allow to link libbpf dynamically

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAEf4BzbqSKZyTyn5wTr3rt=-9W3bFZeupSiNr5YiTPp_Z8rOQw@mail.gmail.com>
Date:   Mon, 2 Dec 2019 11:54:27 -0800
From:   Andrii Nakryiko <andrii.nakryiko@...il.com>
To:     Jiri Olsa <jolsa@...hat.com>
Cc:     Toke Høiland-Jørgensen <toke@...hat.com>,
        Jiri Olsa <jolsa@...nel.org>,
        Arnaldo Carvalho de Melo <acme@...nel.org>,
        lkml <linux-kernel@...r.kernel.org>,
        Networking <netdev@...r.kernel.org>, bpf <bpf@...r.kernel.org>,
        Ingo Molnar <mingo@...nel.org>,
        Namhyung Kim <namhyung@...nel.org>,
        Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
        Peter Zijlstra <a.p.zijlstra@...llo.nl>,
        Michael Petlan <mpetlan@...hat.com>,
        Jesper Dangaard Brouer <brouer@...hat.com>,
        Daniel Borkmann <daniel@...earbox.net>,
        Alexei Starovoitov <alexei.starovoitov@...il.com>,
        Martin KaFai Lau <kafai@...com>,
        Song Liu <songliubraving@...com>, Yonghong Song <yhs@...com>,
        Andrii Nakryiko <andriin@...com>
Subject: Re: [PATCH 0/3] perf/bpftool: Allow to link libbpf dynamically

On Mon, Dec 2, 2019 at 11:21 AM Jiri Olsa <jolsa@...hat.com> wrote:
>
> On Mon, Dec 02, 2019 at 10:42:53AM -0800, Andrii Nakryiko wrote:
> > On Mon, Dec 2, 2019 at 10:09 AM Toke Høiland-Jørgensen <toke@...hat.com> wrote:
> > >
> > > Andrii Nakryiko <andrii.nakryiko@...il.com> writes:
> > >
> > > > On Wed, Nov 27, 2019 at 1:49 AM Jiri Olsa <jolsa@...nel.org> wrote:
> > > >>
> > > >> hi,
> > > >> adding support to link bpftool with libbpf dynamically,
> > > >> and config change for perf.
> > > >>
> > > >> It's now possible to use:
> > > >>   $ make -C tools/bpf/bpftool/ LIBBPF_DYNAMIC=1
> > > >
> > > > I wonder what's the motivation behind these changes, though? Why is
> > > > linking bpftool dynamically with libbpf is necessary and important?
> > > > They are both developed tightly within kernel repo, so I fail to see
> > > > what are the huge advantages one can get from linking them
> > > > dynamically.
> > >
> > > Well, all the regular reasons for using dynamic linking (memory usage,
> > > binary size, etc).
> >
> > bpftool is 327KB with statically linked libbpf. Hardly a huge problem
> > for either binary size or memory usage. CPU instruction cache usage is
> > also hardly a concern for bpftool specifically.
> >
> > > But in particular, the ability to update the libbpf
> > > package if there's a serious bug, and have that be picked up by all
> > > utilities making use of it.
> >
> > I agree, and that works only for utilities linking with libbpf
> > dynamically. For tools that build statically, you'd have to update
> > tools anyways. And if you can update libbpf, you can as well update
> > bpftool at the same time, so I don't think linking bpftool statically
> > with libbpf causes any new problems.
>
> it makes difference for us if we need to respin just one library
> instead of several applications (bpftool and perf at the moment),
> because of the bug in the library
>
> with the Toke's approach we compile some bits of libbpf statically into
> bpftool, but there's still the official API in the dynamic libbpf that
> we care about and that could carry on the fix without bpftool respin

See my replies on v4 of your patchset. I have doubts this actually
works as we hope it works.

I also don't see how that is going to work in general. Imagine
something like this:

static int some_state = 123;

LIBBPF_API void set_state(int x) { some_state = x; }

int get_state() { return some_state; }

If bpftool does:

set_state(42);
printf("%d\n", get_state());


How is this supposed to work with set_state() coming from libbpf.so,
while get_state() being statically linked? Who "owns" memory of `int
some_state` -- bpftool or libbpf.so? Can they magically share it? Or
rather maybe some_state will be actually two different variables in
two different memory regions? And set_state() would be setting one of
them, while get_state() would be reading another one?

It would be good to test this out. Do you mind checking?

>
> > > No reason why bpftool should be special in that respect.
> >
> > But I think bpftool is special and we actually want it to be special
> > and tightly coupled to libbpf with sometimes very intimate knowledge
> > of libbpf and access to "hidden" APIs. That allows us to experiment
> > with new stuff that requires use of bpftool (e.g., code generation for
> > BPF programs), without having to expose and seal public APIs. And I
> > don't think it's a problem from the point of code maintenance, because
> > both live in the same repository and are updated "atomically" when new
> > features are added or changed.
>
> I thought we solved this by Toke's approach, so there' no need
> to expose any new/experimental API .. also you guys will probably
> continue using static linking I guess
>
> jirka
>
> >
> > Beyond superficial binary size worries, I don't see any good reason
> > why we should add more complexity and variables to libbpf and bpftool
> > build processes just to have a "nice to have" option of linking
> > bpftool dynamically with libbpf.
>