Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Thu, 21 May 2020 14:14:32 -0700
In-Reply-To: <20200521125247.30178-1-fejes@inf.elte.hu>
Message-Id: <20200521211432.GC49942@google.com>
Mime-Version: 1.0
References: <20200521125247.30178-1-fejes@inf.elte.hu>
Subject: Re: [PATCH net-next] Extending bpf_setsockopt with SO_BINDTODEVICE sockopt
From:   sdf@google.com
To:     Ferenc Fejes <fejes@inf.elte.hu>
Cc:     netdev@vger.kernel.org, bpf@vger.kernel.org,
        linux-kernel@vger.kernel.org, Martin KaFai Lau <kafai@fb.com>,
        Lawrence Brakmo <brakmo@fb.com>,
        "David S . Miller" <davem@davemloft.net>
Content-Type: text/plain; charset="UTF-8"; format=flowed; delsp=yes
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 05/21, Ferenc Fejes wrote:
> This option makes possible to programatically bind sockets to netdevices.
> With the help of this option sockets of VRF unaware applications
> could be distributed between multiple VRFs with eBPF sock_ops program.
> This let the applications benefit from the multiple possible routes.

> Signed-off-by: Ferenc Fejes <fejes@inf.elte.hu>
> ---
>   net/core/filter.c | 39 ++++++++++++++++++++++++++++++++++++++-
>   1 file changed, 38 insertions(+), 1 deletion(-)

> diff --git a/net/core/filter.c b/net/core/filter.c
> index 822d662f97ef..25dac75bfc5d 100644
> --- a/net/core/filter.c
> +++ b/net/core/filter.c
> @@ -4248,6 +4248,9 @@ static const struct bpf_func_proto  
> bpf_get_socket_uid_proto = {
>   static int _bpf_setsockopt(struct sock *sk, int level, int optname,
>   			   char *optval, int optlen, u32 flags)
>   {
> +	char devname[IFNAMSIZ];
> +	struct net *net;
> +	int ifindex;
>   	int ret = 0;
>   	int val;

> @@ -4257,7 +4260,7 @@ static int _bpf_setsockopt(struct sock *sk, int  
> level, int optname,
>   	sock_owned_by_me(sk);

>   	if (level == SOL_SOCKET) {
> -		if (optlen != sizeof(int))
> +		if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
>   			return -EINVAL;
>   		val = *((int *)optval);

> @@ -4298,6 +4301,40 @@ static int _bpf_setsockopt(struct sock *sk, int  
> level, int optname,
>   				sk_dst_reset(sk);
>   			}
>   			break;
> +		case SO_BINDTODEVICE:
> +			ret = -ENOPROTOOPT;
> +#ifdef CONFIG_NETDEVICES
Any specific reason you're not reusing sock_setbindtodevice or at least
sock_setbindtodevice_locked here? I think, historically, we've
reimplemented some of the sockopts because they were 'easy' (i.e.
were just setting a flag in the socket), this one looks more involved.

I'd suggest, add an optional 'lock_sk' argument to sock_setbindtodevice,
call it with 'true' from real setsockopt, and call it with 'false'
here.

And, as Andrii pointed out, it would be nice to have a selftest
that exercises this new option.