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 1.46 2000/01/09 02:19:37 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                 ip_send(skb);
  67                 return 0;
  68         }
  69
  70         ip_forward_options(skb);
  71         ip_send(skb);
  72         return 0;
  73 }
  74
  75 int ip_forward(struct sk_buff *skb)
  76 {
  77         struct net_device *dev2;        /* Output device */
  78         struct iphdr *iph;      /* Our header */
  79         struct rtable *rt;      /* Route we use */
  80         struct ip_options * opt = &(IPCB(skb)->opt);
  81         unsigned short mtu;
  82
  83         if (IPCB(skb)->opt.router_alert && ip_call_ra_chain(skb))
  84                 return 0;
  85
  86         if (skb->pkt_type != PACKET_HOST)
  87                 goto drop;
  88
  89         /*
  90          *      According to the RFC, we must first decrease the TTL field. If
  91          *      that reaches zero, we must reply an ICMP control message telling
  92          *      that the packet's lifetime expired.
  93          */
  94
  95         iph = skb->nh.iph;
  96         rt = (struct rtable*)skb->dst;
  97
  98         if (iph->ttl <= 1)
  99                 goto too_many_hops;
 100
 101         if (opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
 102                 goto sr_failed;
 103
 104         /*
 105          *      Having picked a route we can now send the frame out
 106          *      after asking the firewall permission to do so.
 107          */
 108
 109         skb->priority = rt_tos2priority(iph->tos);
 110         dev2 = rt->u.dst.dev;
 111         mtu = rt->u.dst.pmtu;
 112
 113         /*
 114          *      We now generate an ICMP HOST REDIRECT giving the route
 115          *      we calculated.
 116          */
 117         if (rt->rt_flags&RTCF_DOREDIRECT && !opt->srr)
 118                 ip_rt_send_redirect(skb);
 119
 120         /* We are about to mangle packet. Copy it! */
 121         if ((skb = skb_cow(skb, dev2->hard_header_len)) == NULL)
 122                 return -1;
 123         iph = skb->nh.iph;
 124         opt = &(IPCB(skb)->opt);
 125
 126         /* Decrease ttl after skb cow done */
 127         ip_decrease_ttl(iph);
 128
 129         /*
 130          * We now may allocate a new buffer, and copy the datagram into it.
 131          * If the indicated interface is up and running, kick it.
 132          */
 133
 134         if (skb->len > mtu && (ntohs(iph->frag_off) & IP_DF))
 135                 goto frag_needed;
 136
 137 #ifdef CONFIG_IP_ROUTE_NAT
 138         if (rt->rt_flags & RTCF_NAT) {
 139                 if (ip_do_nat(skb)) {
 140                         kfree_skb(skb);
 141                         return -1;
 142                 }
 143         }
 144 #endif
 145
 146         return NF_HOOK(PF_INET, NF_IP_FORWARD, skb, skb->dev, dev2,
 147                        ip_forward_finish);
 148
 149 frag_needed:
 150         IP_INC_STATS_BH(IpFragFails);
 151         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
 152         goto drop;
 153
 154 sr_failed:
 155         /*
 156          *      Strict routing permits no gatewaying
 157          */
 158          icmp_send(skb, ICMP_DEST_UNREACH, ICMP_SR_FAILED, 0);
 159          goto drop;
 160
 161 too_many_hops:
 162         /* Tell the sender its packet died... */
 163         icmp_send(skb, ICMP_TIME_EXCEEDED, ICMP_EXC_TTL, 0);
 164 drop:
 165         kfree_skb(skb);
 166         return -1;
 167 }