net/ipv4/ip_forward.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              The IP forwarding functionality.
   7  *
   8  * Version:     $Id: ip_forward.c,v 1.47 2000/10/24 22:54:26 davem Exp $
   9  *
  10  * Authors:     see ip.c
  11  *
  12  * Fixes:
  13  *              Many            :       Split from ip.c , see ip_input.c for
  14  *                                      history.
  15  *              Dave Gregorich  :       NULL ip_rt_put fix for multicast
  16  *                                      routing.
  17  *              Jos Vos         :       Add call_out_firewall before sending,
  18  *                                      use output device for accounting.
  19  *              Jos Vos         :       Call forward firewall after routing
  20  *                                      (always use output device).
  21  *              Mike McLagan    :       Routing by source
  22  */
  23
  24 #include <linux/config.h>
  25 #include <linux/types.h>
  26 #include <linux/mm.h>
  27 #include <linux/sched.h>
  28 #include <linux/skbuff.h>
  29 #include <linux/ip.h>
  30 #include <linux/icmp.h>
  31 #include <linux/netdevice.h>
  32 #include <net/sock.h>
  33 #include <net/ip.h>
  34 #include <net/tcp.h>
  35 #include <net/udp.h>
  36 #include <net/icmp.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/netfilter_ipv4.h>
  40 #include <net/checksum.h>
  41 #include <linux/route.h>
  42 #include <net/route.h>
  43
  44 static inline int ip_forward_finish(struct sk_buff *skb)
  45 {
  46         struct ip_options * opt = &(IPCB(skb)->opt);
  47
  48         IP_INC_STATS_BH(IpForwDatagrams);
  49
  50         if (opt->optlen == 0) {
  51 #ifdef CONFIG_NET_FASTROUTE
  52                 struct rtable *rt = (struct rtable*)skb->dst;
  53
  54                 if (rt->rt_flags&RTCF_FAST && !netdev_fastroute_obstacles) {
  55                         struct dst_entry *old_dst;
  56                         unsigned h = ((*(u8*)&rt->key.dst)^(*(u8*)&rt->key.src))&NETDEV_FASTROUTE_HMASK;
  57
  58                         write_lock_irq(&skb->dev->fastpath_lock);
  59                         old_dst = skb->dev->fastpath[h];
  60                         skb->dev->fastpath[h] = dst_clone(&rt->u.dst);
  61                         write_unlock_irq(&skb->dev->fastpath_lock);
  62
  63                         dst_release(old_dst);
  64                 }
  65 #endif
  66                 return (ip_send(skb));
  67         }
  68
  69         ip_forward_options(skb);
  70         return (ip_send(skb));
  71 }
  72
  73 int ip_forward(struct sk_buff *skb)
  74 {
  75         struct net_device *dev2;        /* Output device */
  76         struct iphdr *iph;      /* Our header */
  77         struct rtable *rt;      /* Route we use */
  78         struct ip_options * opt = &(IPCB(skb)->opt);
  79         unsigned short mtu;
  80
  81         if (IPCB(skb)->opt.router_alert && ip_call_ra_chain(skb))
  82                 return NET_RX_SUCCESS;
  83
  84         if (skb->pkt_type != PACKET_HOST)
  85                 goto drop;
  86
  87         /*
  88          *      According to the RFC, we must first decrease the TTL field. If
  89          *      that reaches zero, we must reply an ICMP control message telling
  90          *      that the packet's lifetime expired.
  91          */
  92
  93         iph = skb->nh.iph;
  94         rt = (struct rtable*)skb->dst;
  95
  96         if (iph->ttl <= 1)
  97                 goto too_many_hops;
  98
  99         if (opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
 100                 goto sr_failed;
 101
 102         /*
 103          *      Having picked a route we can now send the frame out
 104          *      after asking the firewall permission to do so.
 105          */
 106
 107         skb->priority = rt_tos2priority(iph->tos);
 108         dev2 = rt->u.dst.dev;
 109         mtu = rt->u.dst.pmtu;
 110
 111         /*
 112          *      We now generate an ICMP HOST REDIRECT giving the route
 113          *      we calculated.
 114          */
 115         if (rt->rt_flags&RTCF_DOREDIRECT && !opt->srr)
 116                 ip_rt_send_redirect(skb);
 117
 118         /* We are about to mangle packet. Copy it! */
 119         if ((skb = skb_cow(skb, dev2->hard_header_len)) == NULL)
 120                 return NET_RX_DROP;
 121         iph = skb->nh.iph;
 122         opt = &(IPCB(skb)->opt);
 123
 124         /* Decrease ttl after skb cow done */
 125         ip_decrease_ttl(iph);
 126
 127         /*
 128          * We now may allocate a new buffer, and copy the datagram into it.
 129          * If the indicated interface is up and running, kick it.
 130          */
 131
 132         if (skb->len > mtu && (ntohs(iph->frag_off) & IP_DF))
 133                 goto frag_needed;
 134
 135 #ifdef CONFIG_IP_ROUTE_NAT
 136         if (rt->rt_flags & RTCF_NAT) {
 137                 if (ip_do_nat(skb)) {
 138                         kfree_skb(skb);
 139                         return NET_RX_BAD;
 140                 }
 141         }
 142 #endif
 143
 144         return NF_HOOK(PF_INET, NF_IP_FORWARD, skb, skb->dev, dev2,
 145                        ip_forward_finish);
 146
 147 frag_needed:
 148         IP_INC_STATS_BH(IpFragFails);
 149         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
 150         goto drop;
 151
 152 sr_failed:
 153         /*
 154          *      Strict routing permits no gatewaying
 155          */
 156          icmp_send(skb, ICMP_DEST_UNREACH, ICMP_SR_FAILED, 0);
 157          goto drop;
 158
 159 too_many_hops:
 160         /* Tell the sender its packet died... */
 161         icmp_send(skb, ICMP_TIME_EXCEEDED, ICMP_EXC_TTL, 0);
 162 drop:
 163         kfree_skb(skb);
 164         return NET_RX_DROP;
 165 }