Re: [PATCH] ipv4: add sysctl knob to control the discarding of skb from local in ip_forward

[Daniel Borkmann]

On 11/9/21 12:32 PM, Changliang Wu wrote:
> This change is meant to add a control for forwarding skb from local.
> By default, ip forward will not receive the pakcet to/from the local.
> But in some special cases, for example:
> -
> |  ovs-bridge  gw-port |  <---->   kube-proxy(iptables) |
> -
> Ovs sends the packet to the gateway, which requires iptables for nat,
> such as kube-proxy (iptables), and then sends it back to the gateway
> through routing for further processing in ovs.
>
> Signed-off-by: Changliang Wu <changliang.wu@xxxxxxxxxx>
> ---
>   include/net/netns/ipv4.h   | 1 +
>   net/ipv4/af_inet.c         | 1 +
>   net/ipv4/ip_forward.c      | 6 +++---
>   net/ipv4/sysctl_net_ipv4.c | 7 +++++++
>   4 files changed, 12 insertions(+), 3 deletions(-)
>
> diff --git a/include/net/netns/ipv4.h b/include/net/netns/ipv4.h
> index 2f65701..0dbe0d6 100644
> --- a/include/net/netns/ipv4.h
> +++ b/include/net/netns/ipv4.h
> @@ -94,6 +94,7 @@ struct netns_ipv4 {
>   	u8 sysctl_ip_no_pmtu_disc;
>   	u8 sysctl_ip_fwd_use_pmtu;
>   	u8 sysctl_ip_fwd_update_priority;
> +	u8 sysctl_ip_fwd_accept_local;
>   	u8 sysctl_ip_nonlocal_bind;
>   	u8 sysctl_ip_autobind_reuse;
>   	/* Shall we try to damage output packets if routing dev changes? */
> diff --git a/net/ipv4/af_inet.c b/net/ipv4/af_inet.c
> index 0189e3c..b5dc205 100644
> --- a/net/ipv4/af_inet.c
> +++ b/net/ipv4/af_inet.c
> @@ -1844,6 +1844,7 @@ static __net_init int inet_init_net(struct net *net)
>   	 */
>   	net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
>   	net->ipv4.sysctl_ip_fwd_update_priority = 1;
> +	net->ipv4.sysctl_ip_fwd_accept_local = 0;
>   	net->ipv4.sysctl_ip_dynaddr = 0;
>   	net->ipv4.sysctl_ip_early_demux = 1;
>   	net->ipv4.sysctl_udp_early_demux = 1;
> diff --git a/net/ipv4/ip_forward.c b/net/ipv4/ip_forward.c
> index 00ec819..06b7e00 100644
> --- a/net/ipv4/ip_forward.c
> +++ b/net/ipv4/ip_forward.c
> @@ -95,9 +95,6 @@ int ip_forward(struct sk_buff *skb)
>   	if (skb->pkt_type != PACKET_HOST)
>   		goto drop;
>   
> -	if (unlikely(skb->sk))
> -		goto drop;
> -
>   	if (skb_warn_if_lro(skb))
>   		goto drop;
>   
> @@ -110,6 +107,9 @@ int ip_forward(struct sk_buff *skb)
>   	skb_forward_csum(skb);
>   	net = dev_net(skb->dev);
>   
> +	if (unlikely(!net->ipv4.sysctl_ip_fwd_accept_local && skb->sk))
> +		goto drop;
> +

Why moving the check further down instead of initializing dev_net(skb->dev) earlier?

What about ip6_forward()? Same issue exists there too I presume?

>   	/*
>   	 *	According to the RFC, we must first decrease the TTL field. If
>   	 *	that reaches zero, we must reply an ICMP control message telling
> diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
> index 97eb547..d95e2e3 100644
> --- a/net/ipv4/sysctl_net_ipv4.c
> +++ b/net/ipv4/sysctl_net_ipv4.c
> @@ -756,6 +756,13 @@ static int proc_fib_multipath_hash_fields(struct ctl_table *table, int write,
>   		.extra2		= SYSCTL_ONE,
>   	},
>   	{
> +		.procname	= "ip_forward_accept_local",
> +		.data		= &init_net.ipv4.sysctl_ip_fwd_accept_local,
> +		.maxlen		= sizeof(u8),
> +		.mode		= 0644,
> +		.proc_handler	= proc_dou8vec_minmax,
> +	},
> +	{
>   		.procname	= "ip_nonlocal_bind",
>   		.data		= &init_net.ipv4.sysctl_ip_nonlocal_bind,
>   		.maxlen		= sizeof(u8),