[PATCH] selinux: Fix error priority for bind with AF_UNSPEC on AF_INET6 socket

[Paul Moore]

On Thu, Dec 28, 2023 at 6:39 AM Mickaël Salaün <mic at digikod.net> wrote:
>
> The IPv6 network stack first checks the sockaddr length (-EINVAL error)
> before checking the family (-EAFNOSUPPORT error).
>
> This was discovered thanks to commit a549d055a22e ("selftests/landlock:
> Add network tests").
>
> Cc: Alexey Kodanev <alexey.kodanev at oracle.com>
> Cc: Eric Paris <eparis at parisplace.org>
> Cc: Konstantin Meskhidze <konstantin.meskhidze at huawei.com>
> Cc: Paul Moore <paul at paul-moore.com>
> Cc: Stephen Smalley <stephen.smalley.work at gmail.com>
> Reported-by: Muhammad Usama Anjum <usama.anjum at collabora.com>
> Closes: https://lore.kernel.org/r/0584f91c-537c-4188-9e4f-04f192565667@collabora.com
> Fixes: 0f8db8cc73df ("selinux: add AF_UNSPEC and INADDR_ANY checks to selinux_socket_bind()")
> Signed-off-by: Mickaël Salaün <mic at digikod.net>
> ---
>  security/selinux/hooks.c | 4 ++++
>  1 file changed, 4 insertions(+)
>
> diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c
> index feda711c6b7b..9fc55973d765 100644
> --- a/security/selinux/hooks.c
> +++ b/security/selinux/hooks.c
> @@ -4667,6 +4667,10 @@ static int selinux_socket_bind(struct socket *sock, struct sockaddr *address, in
>                                 return -EINVAL;
>                         addr4 = (struct sockaddr_in *)address;
>                         if (family_sa == AF_UNSPEC) {
> +                               if (sock->sk->__sk_common.skc_family ==
> +                                           AF_INET6 &&
> +                                   addrlen < SIN6_LEN_RFC2133)
> +                                       return -EINVAL;

Please use sock->sk_family to simplify the conditional above, or
better yet, use the local variable @family as it is set to the sock's
address family near the top of selinux_socket_bind() ... although, as
I'm looking at the existing code, is this patch necessary?

At the top of the AF_UNSPEC/AF_INET case there is an address length check:

  if (addrlen < sizeof(struct sockaddr_in))
    return -EINVAL;

... which I believe should be performing the required sockaddr length
check (and it is checking for IPv4 address lengths not IPv6 as in the
patch).  I see that we have a similar check for AF_INET6, so we should
be covered there as well.

I'm probably still in a bit of a holiday fog, can you help me see what
I'm missing here?

>                                 /* see __inet_bind(), we only want to allow
>                                  * AF_UNSPEC if the address is INADDR_ANY
>                                  */
> --
> 2.43.0

-- 
paul-moore.com