Re: [PATCH mptcp-next 0/3] mptcp: avoid data-races around the sysctls

[Matthieu Baerts <matttbe@kernel.org>]

Hi Gang Yan,

On 17/08/2026 03:24, Gang Yan wrote:
> From: Gang Yan <yangang@kylinos.cn>
> 
> Hi,
> 
> As decribed in [1], this series is for addressing the KCSAN issues in
> sysctls.
> 
> Following the tcp_congestion_control() model, patch 1 and 2 now store
> pointers to the immutable ops and swap them atomically via xchg().
> 
> Patch 3 (from Matthieu) covers the remaining scalar sysctls with READ_ONCE().
> 
> [1] https://github.com/multipath-tcp/mptcp_net-next/issues/626

Thank you for looking at that with Tao Cui.

Globally, it looks good to me, but I have a few small requests:

- Do you mind resending this series, please: for some unknown reasons,
  Sashiko didn't get it. But before: ↓

- Can you not add Eric in Cc of the 3rd patch? Better not to spam him
for dev work → b4 send --no-trailer-to-cc ; or: git send-email
--suppress-cc=misc-by

- Could you split the code around bpf module support? I guess these
patches will be sent before adding BPF PM/sched support. In this case,
you would have squash-to patches for the BPF parts at the end of the series.

- I think you can move the WRITE_ONCE(pernet->pm_type, pm_type); from
  current patch 2 to 3, with Tao Cui's Co-dev.

Cheers,
Matt
-- 
Sponsored by the NGI0 Core fund.