Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 31 May 2023 20:11:41 +0200
From:   Guillaume Nault <gnault@redhat.com>
To:     Mirsad Todorovac <mirsad.todorovac@alu.unizg.hr>
Cc:     netdev@vger.kernel.org, "David S. Miller" <davem@davemloft.net>,
        Eric Dumazet <edumazet@google.com>,
        Jakub Kicinski <kuba@kernel.org>,
        Paolo Abeni <pabeni@redhat.com>, Shuah Khan <shuah@kernel.org>,
        linux-kernel@vger.kernel.org, linux-kselftest@vger.kernel.org
Subject: Re: POSSIBLE BUG: selftests/net/fcnal-test.sh: [FAIL] in vrf "bind -
 ns-B IPv6 LLA" test
Message-ID: <ZHeN3bg28pGFFjJN@debian>
References: <b6191f90-ffca-dbca-7d06-88a9788def9c@alu.unizg.hr>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <b6191f90-ffca-dbca-7d06-88a9788def9c@alu.unizg.hr>
Precedence: bulk

On Wed, May 24, 2023 at 02:17:09PM +0200, Mirsad Todorovac wrote:
> Hi,

Hi Mirsad,

> The very recent 6.4-rc3 kernel build with AlmaLinux 8.7 on LENOVO 10TX000VCR
> desktop box fails one test:
> 
> [root@host net]# ./fcnal-test.sh
> [...]
> TEST: ping out, vrf device+address bind - ns-B loopback IPv6                  [ OK ]
> TEST: ping out, vrf device+address bind - ns-B IPv6 LLA                       [FAIL]
> TEST: ping in - ns-A IPv6                                                     [ OK ]
> [...]
> Tests passed: 887
> Tests failed:   1
> [root@host net]#

This test also fails on -net. The problem is specific to ping sockets
(same test passes with raw sockets). I believe this test has always
failed since fcnal-test.sh started using net.ipv4.ping_group_range
(commit e71b7f1f44d3 ("selftests: add ping test with ping_group_range
tuned")).

The executed command is:

ip netns exec ns-A ip vrf exec red /usr/bin/ping6 -c1 -w1 -I 2001:db8:3::1 fe80::a846:b5ff:fe4c:da4e%eth1

So ping6 is executed inside VRF 'red' and sets .sin6_scope_id to 'eth1'
(which is a slave device of VRF 'red'). Therefore, we have
sk->sk_bound_dev_if == 'red' and .sin6_scope_id == 'eth1'. This fails
because ping_v6_sendmsg() expects them to be equal:

static int ping_v6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
{
...
                if (__ipv6_addr_needs_scope_id(ipv6_addr_type(daddr)))
                        oif = u->sin6_scope_id;
...
        if ((__ipv6_addr_needs_scope_id(addr_type) && !oif) ||
            (addr_type & IPV6_ADDR_MAPPED) ||
            (oif && sk->sk_bound_dev_if && oif != sk->sk_bound_dev_if)) <-- oif='eth1', but ->sk_bound_dev_if='red'
                return -EINVAL;
...
}

I believe this condition should be relaxed to allow the case where
->sk_bound_dev_if is oif's master device (and maybe there are other
VRF cases to also consider).