Received-SPF: pass (google.com: domain of linux-kernel+bounces-156176-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45d1:ec00::1 as permitted sender) client-ip=2604:1380:45d1:ec00::1;
Precedence: bulk
MIME-Version: 1.0
References: <20240420-bpf_wq-v2-0-6c986a5a741f@kernel.org> <20240420-bpf_wq-v2-11-6c986a5a741f@kernel.org>
In-Reply-To: <20240420-bpf_wq-v2-11-6c986a5a741f@kernel.org>
From: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Date: Tue, 23 Apr 2024 19:55:52 -0700
Message-ID: <CAADnVQJ9Qw6Lr644xRTU-n16UkBCyHoTAQs7QqGiniOdsOVAJg@mail.gmail.com>
Subject: Re: [PATCH bpf-next v2 11/16] bpf: wq: add bpf_wq_init
To: Benjamin Tissoires <bentiss@kernel.org>
Cc: Alexei Starovoitov <ast@kernel.org>, Daniel Borkmann <daniel@iogearbox.net>, 
	Andrii Nakryiko <andrii@kernel.org>, Martin KaFai Lau <martin.lau@linux.dev>, 
	Eduard Zingerman <eddyz87@gmail.com>, Song Liu <song@kernel.org>, 
	Yonghong Song <yonghong.song@linux.dev>, 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>, Mykola Lysenko <mykolal@fb.com>, Shuah Khan <shuah@kernel.org>, 
	bpf <bpf@vger.kernel.org>, LKML <linux-kernel@vger.kernel.org>, 
	"open list:KERNEL SELFTEST FRAMEWORK" <linux-kselftest@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Sat, Apr 20, 2024 at 2:10=E2=80=AFAM Benjamin Tissoires <bentiss@kernel.=
org> wrote:
>
> We need to teach the verifier about the second argument which is declared
> as void * but which is of type KF_ARG_PTR_TO_MAP. We could have dropped
> this extra case if we declared the second argument as struct bpf_map *,
> but that means users will have to do extra casting to have their program
> compile.
>
> We also need to duplicate the timer code for the checking if the map
> argument is matching the provided workqueue.
>
> Signed-off-by: Benjamin Tissoires <bentiss@kernel.org>
>
> ---
>
> FWIW, I still have one concern with this implementation:
> - bpf_wq_work() access ->prog without protection, but I think this might
>   be racing with bpf_wq_set_callback(): if we have the following:
>
>   CPU 0                                     CPU 1
>   bpf_wq_set_callback()
>   bpf_start()
>                                             bpf_wq_work():
>                                               prog =3D cb->prog;
>
>   bpf_wq_set_callback()
>     cb->prog =3D prog;
>     bpf_prog_put(prev)
>     rcu_assign_ptr(cb->callback_fn,
>                    callback_fn);
>                                            callback =3D READ_ONCE(w->cb.c=
allback_fn);
>
>   As I understand callback_fn is fine, prog might be, but we clearly
>   have an inconstency between "prog" and "callback_fn" as they can come
>   from 2 different bpf_wq_set_callback() calls.
>
>   IMO we should protect this by the async->lock, but I'm not sure if
>   it's OK or not.

I see the concern, but I think it's overkill.
Here 'prog' is used to pass it into __bpf_prog_enter_sleepable_recur()
to keep the standard pattern of calling into sleepable prog.
But it won't recurse.
We can open code migrate_disable,etc from there except this_cpu_inc_return,
but it's an overkill.
The passed 'prog' is irrelevant.
If somebody tries really hard by having two progs sharing the same
map with bpf_wq and racing to set_callback... I can see how
prog won't match callback, but it won't make a difference.
prog is not going trigger recursion check (unless somebody
tries is obsessed) and not going to UAF.
I imagine it's possible to attach somewhere in core wq callback
invocation path with fentry, set_callback to the same prog,
and technically it's kinda sorta recursion, but different subprogs,
so not a safety issue.
The code as-is is fine. imo.