net: Put flowi_* prefix on AF independent members of struct flowi
[linux-3.10.git] / net / netfilter / xt_TEE.c
index b3d7301..cb14ae2 100644 (file)
  */
 #include <linux/ip.h>
 #include <linux/module.h>
+#include <linux/percpu.h>
 #include <linux/route.h>
 #include <linux/skbuff.h>
+#include <linux/notifier.h>
 #include <net/checksum.h>
 #include <net/icmp.h>
 #include <net/ip.h>
 #      define WITH_IPV6 1
 #endif
 
+struct xt_tee_priv {
+       struct notifier_block   notifier;
+       struct xt_tee_tginfo    *tginfo;
+       int                     oif;
+};
+
 static const union nf_inet_addr tee_zero_address;
+static DEFINE_PER_CPU(bool, tee_active);
 
 static struct net *pick_net(struct sk_buff *skb)
 {
@@ -47,20 +56,6 @@ static struct net *pick_net(struct sk_buff *skb)
        return &init_net;
 }
 
-static bool tee_tg_route_oif(struct flowi *f, struct net *net,
-                            const struct xt_tee_tginfo *info)
-{
-       const struct net_device *dev;
-
-       if (*info->oif != '\0')
-               return true;
-       dev = dev_get_by_name(net, info->oif);
-       if (dev == NULL)
-               return false;
-       f->oif = dev->ifindex;
-       return true;
-}
-
 static bool
 tee_tg_route4(struct sk_buff *skb, const struct xt_tee_tginfo *info)
 {
@@ -70,27 +65,33 @@ tee_tg_route4(struct sk_buff *skb, const struct xt_tee_tginfo *info)
        struct flowi fl;
 
        memset(&fl, 0, sizeof(fl));
-       if (!tee_tg_route_oif(&fl, net, info))
-               return false;
-       fl.nl_u.ip4_u.daddr = info->gw.ip;
-       fl.nl_u.ip4_u.tos   = RT_TOS(iph->tos);
-       fl.nl_u.ip4_u.scope = RT_SCOPE_UNIVERSE;
-       if (ip_route_output_key(net, &rt, &fl) != 0)
+       if (info->priv) {
+               if (info->priv->oif == -1)
+                       return false;
+               fl.flowi_oif = info->priv->oif;
+       }
+       fl.fl4_dst = info->gw.ip;
+       fl.fl4_tos = RT_TOS(iph->tos);
+       fl.fl4_scope = RT_SCOPE_UNIVERSE;
+       rt = ip_route_output_key(net, &fl);
+       if (IS_ERR(rt))
                return false;
 
-       dst_release(skb_dst(skb));
-       skb_dst_set(skb, &rt->u.dst);
-       skb->dev      = rt->u.dst.dev;
+       skb_dst_drop(skb);
+       skb_dst_set(skb, &rt->dst);
+       skb->dev      = rt->dst.dev;
        skb->protocol = htons(ETH_P_IP);
        return true;
 }
 
 static unsigned int
-tee_tg4(struct sk_buff *skb, const struct xt_target_param *par)
+tee_tg4(struct sk_buff *skb, const struct xt_action_param *par)
 {
        const struct xt_tee_tginfo *info = par->targinfo;
        struct iphdr *iph;
 
+       if (percpu_read(tee_active))
+               return XT_CONTINUE;
        /*
         * Copy the skb, and route the copy. Will later return %XT_CONTINUE for
         * the original skb, which should continue on its way as if nothing has
@@ -104,7 +105,7 @@ tee_tg4(struct sk_buff *skb, const struct xt_target_param *par)
 #ifdef WITH_CONNTRACK
        /* Avoid counting cloned packets towards the original connection. */
        nf_conntrack_put(skb->nfct);
-       skb->nfct     = &nf_conntrack_untracked.ct_general;
+       skb->nfct     = &nf_ct_untracked_get()->ct_general;
        skb->nfctinfo = IP_CT_NEW;
        nf_conntrack_get(skb->nfct);
 #endif
@@ -125,24 +126,13 @@ tee_tg4(struct sk_buff *skb, const struct xt_target_param *par)
                --iph->ttl;
        ip_send_check(iph);
 
-       /*
-        * Xtables is not reentrant currently, so a choice has to be made:
-        * 1. return absolute verdict for the original and let the cloned
-        *    packet travel through the chains
-        * 2. let the original continue travelling and not pass the clone
-        *    to Xtables.
-        * #2 is chosen. Normally, we would use ip_local_out for the clone.
-        * Because iph->check is already correct and we don't pass it to
-        * Xtables anyway, a shortcut to dst_output [forwards to ip_output] can
-        * be taken. %IPSKB_REROUTED needs to be set so that ip_output does not
-        * invoke POSTROUTING on the cloned packet.
-        */
-       IPCB(skb)->flags |= IPSKB_REROUTED;
-       if (tee_tg_route4(skb, info))
-               ip_output(skb);
-       else
+       if (tee_tg_route4(skb, info)) {
+               percpu_write(tee_active, true);
+               ip_local_out(skb);
+               percpu_write(tee_active, false);
+       } else {
                kfree_skb(skb);
-
+       }
        return XT_CONTINUE;
 }
 
@@ -156,16 +146,19 @@ tee_tg_route6(struct sk_buff *skb, const struct xt_tee_tginfo *info)
        struct flowi fl;
 
        memset(&fl, 0, sizeof(fl));
-       if (!tee_tg_route_oif(&fl, net, info))
-               return false;
-       fl.nl_u.ip6_u.daddr = info->gw.in6;
-       fl.nl_u.ip6_u.flowlabel = ((iph->flow_lbl[0] & 0xF) << 16) |
-                                 (iph->flow_lbl[1] << 8) | iph->flow_lbl[2];
+       if (info->priv) {
+               if (info->priv->oif == -1)
+                       return false;
+               fl.flowi_oif = info->priv->oif;
+       }
+       fl.fl6_dst = info->gw.in6;
+       fl.fl6_flowlabel = ((iph->flow_lbl[0] & 0xF) << 16) |
+                          (iph->flow_lbl[1] << 8) | iph->flow_lbl[2];
        dst = ip6_route_output(net, NULL, &fl);
        if (dst == NULL)
                return false;
 
-       dst_release(skb_dst(skb));
+       skb_dst_drop(skb);
        skb_dst_set(skb, dst);
        skb->dev      = dst->dev;
        skb->protocol = htons(ETH_P_IPV6);
@@ -173,17 +166,19 @@ tee_tg_route6(struct sk_buff *skb, const struct xt_tee_tginfo *info)
 }
 
 static unsigned int
-tee_tg6(struct sk_buff *skb, const struct xt_target_param *par)
+tee_tg6(struct sk_buff *skb, const struct xt_action_param *par)
 {
        const struct xt_tee_tginfo *info = par->targinfo;
 
+       if (percpu_read(tee_active))
+               return XT_CONTINUE;
        skb = pskb_copy(skb, GFP_ATOMIC);
        if (skb == NULL)
                return XT_CONTINUE;
 
 #ifdef WITH_CONNTRACK
        nf_conntrack_put(skb->nfct);
-       skb->nfct     = &nf_conntrack_untracked.ct_general;
+       skb->nfct     = &nf_ct_untracked_get()->ct_general;
        skb->nfctinfo = IP_CT_NEW;
        nf_conntrack_get(skb->nfct);
 #endif
@@ -192,25 +187,82 @@ tee_tg6(struct sk_buff *skb, const struct xt_target_param *par)
                struct ipv6hdr *iph = ipv6_hdr(skb);
                --iph->hop_limit;
        }
-       IP6CB(skb)->flags |= IP6SKB_REROUTED;
-       if (tee_tg_route6(skb, info))
-               ip6_output(skb);
-       else
+       if (tee_tg_route6(skb, info)) {
+               percpu_write(tee_active, true);
+               ip6_local_out(skb);
+               percpu_write(tee_active, false);
+       } else {
                kfree_skb(skb);
-
+       }
        return XT_CONTINUE;
 }
 #endif /* WITH_IPV6 */
 
+static int tee_netdev_event(struct notifier_block *this, unsigned long event,
+                           void *ptr)
+{
+       struct net_device *dev = ptr;
+       struct xt_tee_priv *priv;
+
+       priv = container_of(this, struct xt_tee_priv, notifier);
+       switch (event) {
+       case NETDEV_REGISTER:
+               if (!strcmp(dev->name, priv->tginfo->oif))
+                       priv->oif = dev->ifindex;
+               break;
+       case NETDEV_UNREGISTER:
+               if (dev->ifindex == priv->oif)
+                       priv->oif = -1;
+               break;
+       case NETDEV_CHANGENAME:
+               if (!strcmp(dev->name, priv->tginfo->oif))
+                       priv->oif = dev->ifindex;
+               else if (dev->ifindex == priv->oif)
+                       priv->oif = -1;
+               break;
+       }
+
+       return NOTIFY_DONE;
+}
+
 static int tee_tg_check(const struct xt_tgchk_param *par)
 {
-       const struct xt_tee_tginfo *info = par->targinfo;
+       struct xt_tee_tginfo *info = par->targinfo;
+       struct xt_tee_priv *priv;
 
-       if (info->oif[sizeof(info->oif)-1] != '\0')
-               return -EINVAL;
        /* 0.0.0.0 and :: not allowed */
-       return (memcmp(&info->gw, &tee_zero_address,
-              sizeof(tee_zero_address)) == 0) ? -EINVAL : 0;
+       if (memcmp(&info->gw, &tee_zero_address,
+                  sizeof(tee_zero_address)) == 0)
+               return -EINVAL;
+
+       if (info->oif[0]) {
+               if (info->oif[sizeof(info->oif)-1] != '\0')
+                       return -EINVAL;
+
+               priv = kzalloc(sizeof(*priv), GFP_KERNEL);
+               if (priv == NULL)
+                       return -ENOMEM;
+
+               priv->tginfo  = info;
+               priv->oif     = -1;
+               priv->notifier.notifier_call = tee_netdev_event;
+               info->priv    = priv;
+
+               register_netdevice_notifier(&priv->notifier);
+       } else
+               info->priv = NULL;
+
+       return 0;
+}
+
+static void tee_tg_destroy(const struct xt_tgdtor_param *par)
+{
+       struct xt_tee_tginfo *info = par->targinfo;
+
+       if (info->priv) {
+               unregister_netdevice_notifier(&info->priv->notifier);
+               kfree(info->priv);
+       }
 }
 
 static struct xt_target tee_tg_reg[] __read_mostly = {
@@ -221,6 +273,7 @@ static struct xt_target tee_tg_reg[] __read_mostly = {
                .target     = tee_tg4,
                .targetsize = sizeof(struct xt_tee_tginfo),
                .checkentry = tee_tg_check,
+               .destroy    = tee_tg_destroy,
                .me         = THIS_MODULE,
        },
 #ifdef WITH_IPV6
@@ -231,6 +284,7 @@ static struct xt_target tee_tg_reg[] __read_mostly = {
                .target     = tee_tg6,
                .targetsize = sizeof(struct xt_tee_tginfo),
                .checkentry = tee_tg_check,
+               .destroy    = tee_tg_destroy,
                .me         = THIS_MODULE,
        },
 #endif