Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.36 as permitted sender) client-ip=23.128.96.36;
Message-ID: <1c769b1a72c5a7f6e19010dfccc78b2502484cf3.camel@redhat.com>
Subject: Re: [PATCH v2 6/7] net/tcp: Add sne_lock to access SNEs
From:   Paolo Abeni <pabeni@redhat.com>
To:     Dmitry Safonov <dima@arista.com>, David Ahern <dsahern@kernel.org>,
        Eric Dumazet <edumazet@google.com>,
        Jakub Kicinski <kuba@kernel.org>,
        "David S. Miller" <davem@davemloft.net>
Cc:     linux-kernel@vger.kernel.org,
        Dmitry Safonov <0x7f454c46@gmail.com>,
        Francesco Ruggeri <fruggeri05@gmail.com>,
        Salam Noureddine <noureddine@arista.com>,
        Simon Horman <horms@kernel.org>, netdev@vger.kernel.org
Date:   Mon, 27 Nov 2023 12:41:57 +0100
In-Reply-To: <20231124002720.102537-7-dima@arista.com>
References: <20231124002720.102537-1-dima@arista.com>
         <20231124002720.102537-7-dima@arista.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.46.4 (3.46.4-1.fc37) 
MIME-Version: 1.0
Precedence: bulk

On Fri, 2023-11-24 at 00:27 +0000, Dmitry Safonov wrote:
> RFC 5925 (6.2):
> > TCP-AO emulates a 64-bit sequence number space by inferring when to
> > increment the high-order 32-bit portion (the SNE) based on
> > transitions in the low-order portion (the TCP sequence number).
>=20
> snd_sne and rcv_sne are the upper 4 bytes of extended SEQ number.
> Unfortunately, reading two 4-bytes pointers can't be performed
> atomically (without synchronization).
>=20
> Let's keep it KISS and add an rwlock - that shouldn't create much
> contention as SNE are updated every 4Gb of traffic and the atomic region
> is quite small.
>=20
> Fixes: 64382c71a557 ("net/tcp: Add TCP-AO SNE support")
> Signed-off-by: Dmitry Safonov <dima@arista.com>
> ---
>  include/net/tcp_ao.h |  2 +-
>  net/ipv4/tcp_ao.c    | 34 +++++++++++++++++++++-------------
>  net/ipv4/tcp_input.c | 16 ++++++++++++++--
>  3 files changed, 36 insertions(+), 16 deletions(-)
>=20
> diff --git a/include/net/tcp_ao.h b/include/net/tcp_ao.h
> index 647781080613..beea3e6b39e2 100644
> --- a/include/net/tcp_ao.h
> +++ b/include/net/tcp_ao.h
> @@ -123,6 +123,7 @@ struct tcp_ao_info {
>  	 */
>  	u32			snd_sne;
>  	u32			rcv_sne;
> +	rwlock_t		sne_lock;

RW lock are problematic in the networking code, see commit
dbca1596bbb08318f5e3b3b99f8ca0a0d3830a65.

I think you can use a plain spinlock here, as both read and write
appears to be in the fastpath (?!?)

> @@ -781,8 +780,10 @@ int tcp_ao_prepare_reset(const struct sock *sk, stru=
ct sk_buff *skb,
>  		*traffic_key =3D snd_other_key(*key);
>  		rnext_key =3D READ_ONCE(ao_info->rnext_key);
>  		*keyid =3D rnext_key->rcvid;
> -		*sne =3D tcp_ao_compute_sne(READ_ONCE(ao_info->snd_sne),
> -					  snd_basis, seq);
> +		read_lock_irqsave(&ao_info->sne_lock, flags);
> +		*sne =3D tcp_ao_compute_sne(ao_info->snd_sne,
> +					  READ_ONCE(*snd_basis), seq);
> +		read_unlock_irqrestore(&ao_info->sne_lock, flags);

Why are you using the irqsave variant? bh should suffice.

Cheers,

Paolo