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: <20220614084930.43276-1-nashuiliang@gmail.com> <62ad50fa9d42d_24b34208d6@john.notmuch>
In-Reply-To: <62ad50fa9d42d_24b34208d6@john.notmuch>
From:   chuang <nashuiliang@gmail.com>
Date:   Sat, 18 Jun 2022 13:31:01 +0800
Message-ID: <CACueBy7NqRszA3tCOvLhfi1OraUrL_GD9YZ9XOPNHzbR1=+z7g@mail.gmail.com>
Subject: Re: [PATCH] libbpf: Remove kprobe_event on failed kprobe_open_legacy
To:     John Fastabend <john.fastabend@gmail.com>
Cc:     Jingren Zhou <zhoujingren@didiglobal.com>,
        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>,
        KP Singh <kpsingh@kernel.org>, netdev@vger.kernel.org,
        bpf@vger.kernel.org, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi John,

On Sat, Jun 18, 2022 at 12:13 PM John Fastabend
<john.fastabend@gmail.com> wrote:
>
> Chuang W wrote:
> > In a scenario where livepatch and aggrprobe coexist, the creating
> > kprobe_event using tracefs API will succeed, a trace event (e.g.
> > /debugfs/tracing/events/kprobe/XX) will exist, but perf_event_open()
> > will return an error.
>
> This seems a bit strange from API side. I'm not really familiar with
> livepatch, but I guess this is UAPI now so fixing add_kprobe_event_legacy
> to fail is not an option?
>

The legacy kprobe API (i.e. tracefs API) has two steps:

1) register_kprobe
$ echo 'p:mykprobe XXX' > /sys/kernel/debug/tracing/kprobe_events
This will create a trace event of mykprobe and register a disable
kprobe that waits to be activated.

2) enable_kprobe
2.1) using syscall perf_event_open
as the following code, perf_event_kprobe_open_legacy (file:
tools/lib/bpf/libbpf.c):
---
attr.type = PERF_TYPE_TRACEPOINT;
pfd = syscall(__NR_perf_event_open, &attr,
              pid < 0 ? -1 : pid, /* pid */
              pid == -1 ? 0 : -1, /* cpu */
              -1 /* group_fd */,  PERF_FLAG_FD_CLOEXEC);
---
In the implementation code of perf_event_open, enable_kprobe() will be executed.
2.2) using shell
$ echo 1 > /sys/kernel/debug/tracing/events/kprobes/mykprobe/enable
As with perf_event_open, enable_kprobe() will also be executed.

When using the same function XXX, kprobe and livepatch cannot coexist,
that is, step 2) will return an error (ref: arm_kprobe_ftrace()),
however, step 1) is ok!
However, the new kprobe API (i.e. perf kprobe API) aggregates
register_kprobe and enable_kprobe, internally fixes the issue on
failed enable_kprobe.
But above all, for the legacy kprobe API, I think it should remove
kprobe_event on failed add_kprobe_event_legacy() in
perf_event_kprobe_open_legacy (file: tools/lib/bpf/libbpf.c).