ip6_tunnel: add optional fwmark inherit
authorEric Dumazet <eric.dumazet@gmail.com>
Tue, 20 Sep 2011 18:50:00 +0000 (14:50 -0400)
committerDavid S. Miller <davem@davemloft.net>
Tue, 20 Sep 2011 18:50:00 +0000 (14:50 -0400)
Add IP6_TNL_F_USE_ORIG_FWMARK to ip6_tunnel, so that ip6_tnl_xmit2()
makes a route lookup taking into account skb->fwmark and doesnt cache
lookup result.

This permits more flexibility in policies and firewall setups.

To setup such a tunnel, "fwmark inherit" option should be added to "ip
-f inet6 tunnel" command.

Reported-by: Anders Franzen <Anders.Franzen@ericsson.com>
CC: Hans Schillström <hans.schillstrom@ericsson.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/linux/ip6_tunnel.h
net/ipv6/ip6_tunnel.c

index acb9ad684d639167c6a2c1469e14eb1d6694e93b..bf22b03179022da13565642026213601e879fa88 100644 (file)
@@ -16,6 +16,8 @@
 #define IP6_TNL_F_MIP6_DEV 0x8
 /* copy DSCP from the outer packet */
 #define IP6_TNL_F_RCV_DSCP_COPY 0x10
+/* copy fwmark from inner packet */
+#define IP6_TNL_F_USE_ORIG_FWMARK 0x20
 
 struct ip6_tnl_parm {
        char name[IFNAMSIZ];    /* name of tunnel device */
index 694d70a8a0eef34074b31bc926a4956b4d5f2a31..bdc15c9003d781fbdb3d0288cf8b88aa84f71742 100644 (file)
@@ -889,7 +889,7 @@ static int ip6_tnl_xmit2(struct sk_buff *skb,
        struct net_device_stats *stats = &t->dev->stats;
        struct ipv6hdr *ipv6h = ipv6_hdr(skb);
        struct ipv6_tel_txoption opt;
-       struct dst_entry *dst, *ndst = NULL;
+       struct dst_entry *dst = NULL, *ndst = NULL;
        struct net_device *tdev;
        int mtu;
        unsigned int max_headroom = sizeof(struct ipv6hdr);
@@ -897,7 +897,8 @@ static int ip6_tnl_xmit2(struct sk_buff *skb,
        int err = -1;
        int pkt_len;
 
-       dst = ip6_tnl_dst_check(t);
+       if (!fl6->flowi6_mark)
+               dst = ip6_tnl_dst_check(t);
        if (!dst) {
                ndst = ip6_route_output(net, NULL, fl6);
 
@@ -955,8 +956,12 @@ static int ip6_tnl_xmit2(struct sk_buff *skb,
                skb = new_skb;
        }
        skb_dst_drop(skb);
-       skb_dst_set_noref(skb, dst);
-
+       if (fl6->flowi6_mark) {
+               skb_dst_set(skb, dst);
+               ndst = NULL;
+       } else {
+               skb_dst_set_noref(skb, dst);
+       }
        skb->transport_header = skb->network_header;
 
        proto = fl6->flowi6_proto;
@@ -1021,9 +1026,11 @@ ip4ip6_tnl_xmit(struct sk_buff *skb, struct net_device *dev)
 
        dsfield = ipv4_get_dsfield(iph);
 
-       if ((t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS))
+       if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
                fl6.flowlabel |= htonl((__u32)iph->tos << IPV6_TCLASS_SHIFT)
                                          & IPV6_TCLASS_MASK;
+       if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
+               fl6.flowi6_mark = skb->mark;
 
        err = ip6_tnl_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
        if (err != 0) {
@@ -1070,10 +1077,12 @@ ip6ip6_tnl_xmit(struct sk_buff *skb, struct net_device *dev)
        fl6.flowi6_proto = IPPROTO_IPV6;
 
        dsfield = ipv6_get_dsfield(ipv6h);
-       if ((t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS))
+       if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
                fl6.flowlabel |= (*(__be32 *) ipv6h & IPV6_TCLASS_MASK);
-       if ((t->parms.flags & IP6_TNL_F_USE_ORIG_FLOWLABEL))
+       if (t->parms.flags & IP6_TNL_F_USE_ORIG_FLOWLABEL)
                fl6.flowlabel |= (*(__be32 *) ipv6h & IPV6_FLOWLABEL_MASK);
+       if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
+               fl6.flowi6_mark = skb->mark;
 
        err = ip6_tnl_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
        if (err != 0) {