netdev - Re: [PATCH bpf-next] bpf: Add BPF-helper for accessing CLOCK

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: <c166aa47-e404-e6ee-0ec5-0ead1923f412@redhat.com>
Date:   Tue, 7 Jun 2022 21:35:05 +0200
From:   Jesper Dangaard Brouer <jbrouer@...hat.com>
To:     Thomas Gleixner <tglx@...utronix.de>,
        Alexei Starovoitov <alexei.starovoitov@...il.com>,
        Kurt Kanzenbach <kurt@...utronix.de>
Cc:     brouer@...hat.com, Alexei Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        Andrii Nakryiko <andrii@...nel.org>,
        Martin KaFai Lau <kafai@...com>,
        Song Liu <songliubraving@...com>, Yonghong Song <yhs@...com>,
        John Fastabend <john.fastabend@...il.com>,
        KP Singh <kpsingh@...nel.org>,
        Joanne Koong <joannelkoong@...il.com>,
        Jiri Olsa <jolsa@...nel.org>,
        Dave Marchevsky <davemarchevsky@...com>,
        Lorenzo Bianconi <lorenzo@...nel.org>,
        Geliang Tang <geliang.tang@...e.com>,
        Jakub Sitnicki <jakub@...udflare.com>,
        Network Development <netdev@...r.kernel.org>,
        bpf <bpf@...r.kernel.org>
Subject: Re: [PATCH bpf-next] bpf: Add BPF-helper for accessing CLOCK_TAI


On 07/06/2022 11.14, Thomas Gleixner wrote:
> Alexei,
> 
> On Mon, Jun 06 2022 at 08:57, Alexei Starovoitov wrote:
>> On Mon, Jun 6, 2022 at 3:38 AM Kurt Kanzenbach <kurt@...utronix.de> wrote:
>>>
>>> From: Jesper Dangaard Brouer <brouer@...hat.com>
>>>
>>> Commit 3dc6ffae2da2 ("timekeeping: Introduce fast accessor to clock tai")
>>> introduced a fast and NMI-safe accessor for CLOCK_TAI. Especially in time
>>> sensitive networks (TSN), where all nodes are synchronized by Precision Time
>>> Protocol (PTP), it's helpful to have the possibility to generate timestamps
>>> based on CLOCK_TAI instead of CLOCK_MONOTONIC. With a BPF helper for TAI in
>>> place, it becomes very convenient to correlate activity across different
>>> machines in the network.
>>
>> That's a fresh feature. It feels risky to bake it into uapi already.
> 
> What? That's just support for a different CLOCK. What's so risky about
> it?

I didn't think it was "risky" as this is already exported as:
  EXPORT_SYMBOL_GPL(ktime_get_tai_fast_ns);

Correct me if I'm wrong, but this simple gives BPF access to CLOCK_TAI
(see man clock_gettime(2)), right?
And CLOCK_TAI is not really a new/fresh type of CLOCK.

Especially for networking we need this CLOCK_TAI time as HW LaunchTime
need this (e.g. see qdisc's sch_etf.c and sch_taprio.c).

> 
>> imo it would be better to annotate tk_core variable in vmlinux BTF.
>> Then progs will be able to read all possible timekeeper offsets.
> 
> We are exposing APIs. APIs can be supported, but exposing implementation
> details creates ABIs of the worst sort because that prevents the kernel
> from changing the implementation. We've seen the fallout with the recent
> tracepoint changes already.

Hmm... annotate tk_core variable in vmlinux BTF and letting BPF progs
access this seems like an unsafe approach and we tempt BPF-developers to
think other parts are okay to access.

Accessing timekeeper->offs_tai might be okay as it is already "marked" 
with data_race(tk->offs_tai), but I'm not sure about other members, as 
I'm not expert in this area.

I assume that the include filename <linux/timekeeper_internal.h>
indicate that the maintainers don't want to open up access to struct
timekeeper...

--Jesper