Received-SPF: pass (google.com: domain of linux-kernel+bounces-82902-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
Message-ID: <aa5f1c11-4528-4d53-91f3-5ce8c02363ac@gmail.com>
Date: Tue, 27 Feb 2024 10:02:51 +0100
Precedence: bulk
MIME-Version: 1.0
Subject: Re: [PATCH net-next 2/2] net: geneve: enable local address bind for
 geneve sockets
To: Eyal Birger <eyal.birger@gmail.com>
Cc: davem@davemloft.net, edumazet@google.com, kuba@kernel.org,
 pabeni@redhat.com, idosch@nvidia.com, razor@blackwall.org,
 amcohen@nvidia.com, petrm@nvidia.com, jbenc@redhat.com, b.galvani@gmail.com,
 bpoirier@nvidia.com, gavinl@nvidia.com, martin.lau@kernel.org,
 daniel@iogearbox.net, herbert@gondor.apana.org.au, netdev@vger.kernel.org,
 linux-kernel@vger.kernel.org
References: <df300a49-7811-4126-a56a-a77100c8841b@gmail.com>
 <79a8ba83-86bf-4c22-845c-8f285c2d1396@gmail.com>
 <CAHsH6GvX7zYSoA7JVemRtunWWSaew1S11Y996WAGt6B9d8=cOA@mail.gmail.com>
From: Richard Gobert <richardbgobert@gmail.com>
In-Reply-To: <CAHsH6GvX7zYSoA7JVemRtunWWSaew1S11Y996WAGt6B9d8=cOA@mail.gmail.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Eyal Birger wrote:
> Hi,
> 
> On Thu, Feb 22, 2024 at 12:54 PM Richard Gobert
> <richardbgobert@gmail.com> wrote:
>>
>> This patch adds support for binding to a local address in geneve sockets.
> 
> Thanks for adding this.
> 
>> It achieves this by adding a geneve_addr union to represent local address
>> to bind to, and copying it to udp_port_cfg in geneve_create_sock.
> 
> AFICT in geneve_sock_add(), geneve_socket_create() is only called if there's
> no existing open socket with the GENEVE destination port. As such, wouldn't
> this bind work only for the first socket in the namespace?
> 
> If that is the case, then perhaps binding the socket isn't the right
> approach, and instead geneve_lookup() should search for the tunnel based on
> both the source and destination IPs.
> 
> Am I missing something?
> 
> Eyal

You are right, I missed it.
Binding the socket is the main reason for the patch, to prevent exposing
the geneve port on all interfaces.
I think it should be searched in geneve{6}_lookup and in geneve_find_sock:

static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
					    sa_family_t family,
					    union geneve_addr *saddr)
 {
 	struct geneve_sock *gs;

 	list_for_each_entry(gs, &gn->sock_list, list) {
		struct inet_sock *inet = inet_sk(gs->sock->sk);

		if (inet->inet_sport == dst_port && geneve_get_sk_family(gs) == family) {
			if (family == AF_INET && inet->inet_rcv_saddr == saddr->sin.sin_addr.s_addr)
				return gs;
        ...

This is also true for VXLAN
What do you think?
Thanks