Revert "netfilter: bridge: query conntrack about skb dnat"
authorFlorian Westphal <fw@strlen.de>
Wed, 20 May 2015 11:42:25 +0000 (13:42 +0200)
committerPablo Neira Ayuso <pablo@netfilter.org>
Wed, 20 May 2015 11:51:25 +0000 (13:51 +0200)
This reverts commit c055d5b03bb4cb69d349d787c9787c0383abd8b2.

There are two issues:
'dnat_took_place' made me think that this is related to
-j DNAT/MASQUERADE.

But thats only one part of the story.  This is also relevant for SNAT
when we undo snat translation in reverse/reply direction.

Furthermore, I originally wanted to do this mainly to avoid
storing ipv6 addresses once we make DNAT/REDIRECT work
for ipv6 on bridges.

However, I forgot about SNPT/DNPT which is stateless.

So we can't escape storing address for ipv6 anyway. Might as
well do it for ipv4 too.

Reported-and-tested-by: Bernhard Thaler <bernhard.thaler@wvnet.at>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
include/linux/skbuff.h
net/bridge/br_netfilter.c

index 66e374d..f15154a 100644 (file)
@@ -176,6 +176,7 @@ struct nf_bridge_info {
        struct net_device       *physindev;
        struct net_device       *physoutdev;
        char                    neigh_header[8];
+       __be32                  ipv4_daddr;
 };
 #endif
 
index ab55e24..60ddfbe 100644 (file)
 #include <net/route.h>
 #include <net/netfilter/br_netfilter.h>
 
-#if IS_ENABLED(CONFIG_NF_CONNTRACK)
-#include <net/netfilter/nf_conntrack.h>
-#endif
-
 #include <asm/uaccess.h>
 #include "br_private.h"
 #ifdef CONFIG_SYSCTL
@@ -350,24 +346,15 @@ free_skb:
        return 0;
 }
 
-static bool dnat_took_place(const struct sk_buff *skb)
+static bool daddr_was_changed(const struct sk_buff *skb,
+                             const struct nf_bridge_info *nf_bridge)
 {
-#if IS_ENABLED(CONFIG_NF_CONNTRACK)
-       enum ip_conntrack_info ctinfo;
-       struct nf_conn *ct;
-
-       ct = nf_ct_get(skb, &ctinfo);
-       if (!ct || nf_ct_is_untracked(ct))
-               return false;
-
-       return test_bit(IPS_DST_NAT_BIT, &ct->status);
-#else
-       return false;
-#endif
+       return ip_hdr(skb)->daddr != nf_bridge->ipv4_daddr;
 }
 
 /* This requires some explaining. If DNAT has taken place,
  * we will need to fix up the destination Ethernet address.
+ * This is also true when SNAT takes place (for the reply direction).
  *
  * There are two cases to consider:
  * 1. The packet was DNAT'ed to a device in the same bridge
@@ -421,7 +408,7 @@ static int br_nf_pre_routing_finish(struct sock *sk, struct sk_buff *skb)
                nf_bridge->pkt_otherhost = false;
        }
        nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
-       if (dnat_took_place(skb)) {
+       if (daddr_was_changed(skb, nf_bridge)) {
                if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
                        struct in_device *in_dev = __in_dev_get_rcu(dev);
 
@@ -632,6 +619,7 @@ static unsigned int br_nf_pre_routing(const struct nf_hook_ops *ops,
                                      struct sk_buff *skb,
                                      const struct nf_hook_state *state)
 {
+       struct nf_bridge_info *nf_bridge;
        struct net_bridge_port *p;
        struct net_bridge *br;
        __u32 len = nf_bridge_encap_header_len(skb);
@@ -669,6 +657,9 @@ static unsigned int br_nf_pre_routing(const struct nf_hook_ops *ops,
        if (!setup_pre_routing(skb))
                return NF_DROP;
 
+       nf_bridge = nf_bridge_info_get(skb);
+       nf_bridge->ipv4_daddr = ip_hdr(skb)->daddr;
+
        skb->protocol = htons(ETH_P_IP);
 
        NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, state->sk, skb,