Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20230417154737.12740-1-laoar.shao@gmail.com> <20230417154737.12740-6-laoar.shao@gmail.com>
 <20230417201457.c43xfcukjzm4u6vx@dhcp-172-26-102-232.dhcp.thefacebook.com>
 <CALOAHbC4Bz_VX52zmv=sScBf0hzscMAC4+EwMCpnd1BcaSVJSw@mail.gmail.com>
 <CAADnVQJw9BCK2itE5bZWdQYz7D-8KdcH96E885zUakEDAOrC+Q@mail.gmail.com>
 <CALOAHbCtPR26it_Wdk7T_TETMTh2se6rgEbL_KC5XKtzvObjiA@mail.gmail.com>
 <CAADnVQ+FO-+1OALTtgVkcpH3Adc6xS9qjzORyq2vwVtwY2UoxQ@mail.gmail.com>
 <20230424174049.1c9e54dd@rorschach.local.home> <CALOAHbAx+W3-iBS6=FsPPShbEuSSZeyQWvLque+uF9Suwe3-HA@mail.gmail.com>
In-Reply-To: <CALOAHbAx+W3-iBS6=FsPPShbEuSSZeyQWvLque+uF9Suwe3-HA@mail.gmail.com>
From:   Yafang Shao <laoar.shao@gmail.com>
Date:   Thu, 27 Apr 2023 20:15:18 +0800
Message-ID: <CALOAHbAqsSq+gVg9xTYGAkrdZaFXc=PVoOYqej33dCEjWtHfFw@mail.gmail.com>
Subject: Re: [PATCH bpf-next 5/6] bpf: Improve tracing recursion prevention mechanism
To:     Steven Rostedt <rostedt@goodmis.org>,
        Alexei Starovoitov <alexei.starovoitov@gmail.com>
Cc:     Alexei Starovoitov <ast@kernel.org>,
        Daniel Borkmann <daniel@iogearbox.net>,
        Andrii Nakryiko <andrii@kernel.org>,
        Martin KaFai Lau <kafai@fb.com>,
        Song Liu <songliubraving@fb.com>, Yonghong Song <yhs@fb.com>,
        John Fastabend <john.fastabend@gmail.com>,
        KP Singh <kpsingh@kernel.org>,
        Stanislav Fomichev <sdf@google.com>,
        Hao Luo <haoluo@google.com>, Jiri Olsa <jolsa@kernel.org>,
        Masami Hiramatsu <mhiramat@kernel.org>,
        bpf <bpf@vger.kernel.org>, linux-trace-kernel@vger.kernel.org,
        LKML <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Thu, Apr 27, 2023 at 5:57=E2=80=AFPM Yafang Shao <laoar.shao@gmail.com> =
wrote:
>
> On Tue, Apr 25, 2023 at 5:40=E2=80=AFAM Steven Rostedt <rostedt@goodmis.o=
rg> wrote:
> >
> > On Wed, 19 Apr 2023 15:46:34 -0700
> > Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote:
> >
> > > No. Just one prog at entry into any of the kernel functions
> > > and another prog at entry of funcs that 1st bpf prog called indirectl=
y.
> > > Like one prog is tracing networking events while another
> > > is focusing on mm. They should not conflict.
> >
> > You mean that you have:
> >
> > function start:
> >   __bpf_prog_enter_recur()
> >     bpf_program1()
> >       __bpf_prog_enter_recur()
> >         bpf_program2();
> >       __bpf_prog_exit_recur()
> >   __bpf_prog_exit_recur()
> >
> >   rest of function
> >
> > That is, a bpf program can be called within another bpf pogram between
> > the prog_enter and prog_exit(), that is in the same context (normal,
> > softirq, irq, etc)?
> >
>
> Right, that can happen per my verification. Below is a simple bpf
> program to verify it.
>
> struct {
>     __uint(type, BPF_MAP_TYPE_LPM_TRIE);
>     __type(key, __u64);
>     __type(value, __u64);
>     __uint(max_entries, 1024);
>     __uint(map_flags, BPF_F_NO_PREALLOC);
> } write_map SEC(".maps");
>
> __u64 key;
>
> SEC("fentry/kernel_clone")
> int program1()
> {
>     __u64 value =3D 1;
>
>     bpf_printk("before update");
>     // It will call trie_update_elem and thus trigger program2.
>     bpf_map_update_elem(&write_map, &key, &value, BPF_ANY);
>     __sync_fetch_and_add(&key, 1);
>     bpf_printk("after update");
>     return 0;
> }
>
> SEC("fentry/trie_update_elem")
> int program2()
> {
>     bpf_printk("trie_update_elem");
>     return 0;
> }
>
> The result as follows,
>
>          kubelet-203203  [018] ....1  9579.862862:
> __bpf_prog_enter_recur: __bpf_prog_enter_recur
>          kubelet-203203  [018] ...11  9579.862869: bpf_trace_printk:
> before update
>          kubelet-203203  [018] ....2  9579.862869:
> __bpf_prog_enter_recur: __bpf_prog_enter_recur
>          kubelet-203203  [018] ...12  9579.862870: bpf_trace_printk:
> trie_update_elem
>          kubelet-203203  [018] ....2  9579.862870:
> __bpf_prog_exit_recur: __bpf_prog_exit_recur
>          kubelet-203203  [018] ...11  9579.862870: bpf_trace_printk:
> after update
>          kubelet-203203  [018] ....1  9579.862871:
> __bpf_prog_exit_recur: __bpf_prog_exit_recur
>
> Note that we can't trace __bpf_prog_enter_recur and
> __bpf_prog_exit_recur, so we have to modify the kernel to print them.
>

... However, surprisingly it still works even after this patchset is
applied, because the hardirq/softirq flag is set when the program2 is
running, see also the flags in the above trace_pipe output. Is that
expected ?!
I need  some time to figure it out, but maybe you have a quick answer...

> > The protection is on the trampoline where the bpf program is called.
> > Not sure how ftrace can stop BPF or BPF stop ftrace, unless bpf is
> > tracing a ftrace callback, or ftrace is tracing a bpf function.
> >
> > -- Steve
>


--=20
Regards
Yafang