Re: [PATCH net] raw: annotate data-races in raw_v6_match()

[Eric Dumazet]

On Mon, Jun 1, 2026 at 12:42 AM Runyu Xiao <runyu.xiao@xxxxxxxxxx> wrote:
>
> raw_v6_match() is a lockless match helper under sk_for_each_rcu()
> and still reads inet_sk(sk)->inet_num, sk->sk_bound_dev_if,
> sk->sk_v6_daddr and sk->sk_v6_rcv_saddr with plain loads.
>
> Add READ_ONCE() annotations for these fields.
>
> Signed-off-by: Runyu Xiao <runyu.xiao@xxxxxxxxxx>
> ---
>  net/ipv6/raw.c | 29 +++++++++++++++++++++++------
>  1 file changed, 23 insertions(+), 6 deletions(-)
>
> diff --git a/net/ipv6/raw.c b/net/ipv6/raw.c
> index 3cc58698cbbd..7160d5513742 100644
> --- a/net/ipv6/raw.c
> +++ b/net/ipv6/raw.c
> @@ -64,20 +64,37 @@
>  struct raw_hashinfo raw_v6_hashinfo;
>  EXPORT_SYMBOL_GPL(raw_v6_hashinfo);
>
> +static void raw_v6_addr_snapshot(struct in6_addr *dst,
> +                                const struct in6_addr *src)
> +{
> +       dst->s6_addr32[0] = READ_ONCE(src->s6_addr32[0]);
> +       dst->s6_addr32[1] = READ_ONCE(src->s6_addr32[1]);
> +       dst->s6_addr32[2] = READ_ONCE(src->s6_addr32[2]);
> +       dst->s6_addr32[3] = READ_ONCE(src->s6_addr32[3]);
> +}

I do not think we want to incur an extra cost for such a minor race.

Note that your patch does not solve the race, since writers do not
update these fields atomically anyway.