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  * Authors:     see ip.c
   9  *
  10  * Fixes:
  11  *              Many            :       Split from ip.c , see ip_input.c for
  12  *                                      history.
  13  *              Dave Gregorich  :       NULL ip_rt_put fix for multicast
  14  *                                      routing.
  15  *              Jos Vos         :       Add call_out_firewall before sending,
  16  *                                      use output device for accounting.
  17  *              Jos Vos         :       Call forward firewall after routing
  18  *                                      (always use output device).
  19  *              Mike McLagan    :       Routing by source
  20  */
  21
  22 #include <linux/types.h>
  23 #include <linux/mm.h>
  24 #include <linux/skbuff.h>
  25 #include <linux/ip.h>
  26 #include <linux/icmp.h>
  27 #include <linux/netdevice.h>
  28 #include <linux/slab.h>
  29 #include <net/sock.h>
  30 #include <net/ip.h>
  31 #include <net/tcp.h>
  32 #include <net/udp.h>
  33 #include <net/icmp.h>
  34 #include <linux/tcp.h>
  35 #include <linux/udp.h>
  36 #include <linux/netfilter_ipv4.h>
  37 #include <net/checksum.h>
  38 #include <linux/route.h>
  39 #include <net/route.h>
  40 #include <net/xfrm.h>
  41
  42 static bool ip_exceeds_mtu(const struct sk_buff *skb, unsigned int mtu)
  43 {
  44         if (skb->len <= mtu)
  45                 return false;
  46
  47         if (unlikely((ip_hdr(skb)->frag_off & htons(IP_DF)) == 0))
  48                 return false;
  49
  50         /* original fragment exceeds mtu and DF is set */
  51         if (unlikely(IPCB(skb)->frag_max_size > mtu))
  52                 return true;
  53
  54         if (skb->ignore_df)
  55                 return false;
  56
  57         if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
  58                 return false;
  59
  60         return true;
  61 }
  62
  63
  64 static int ip_forward_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
  65 {
  66         struct ip_options *opt  = &(IPCB(skb)->opt);
  67
  68         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
  69         IP_ADD_STATS_BH(net, IPSTATS_MIB_OUTOCTETS, skb->len);
  70
  71         if (unlikely(opt->optlen))
  72                 ip_forward_options(skb);
  73
  74         skb_sender_cpu_clear(skb);
  75         return dst_output(net, sk, skb);
  76 }
  77
  78 int ip_forward(struct sk_buff *skb)
  79 {
  80         u32 mtu;
  81         struct iphdr *iph;      /* Our header */
  82         struct rtable *rt;      /* Route we use */
  83         struct ip_options *opt  = &(IPCB(skb)->opt);
  84         struct net *net;
  85
  86         /* that should never happen */
  87         if (skb->pkt_type != PACKET_HOST)
  88                 goto drop;
  89
  90         if (unlikely(skb->sk))
  91                 goto drop;
  92
  93         if (skb_warn_if_lro(skb))
  94                 goto drop;
  95
  96         if (!xfrm4_policy_check(NULL, XFRM_POLICY_FWD, skb))
  97                 goto drop;
  98
  99         if (IPCB(skb)->opt.router_alert && ip_call_ra_chain(skb))
 100                 return NET_RX_SUCCESS;
 101
 102         skb_forward_csum(skb);
 103         net = dev_net(skb->dev);
 104
 105         /*
 106          *      According to the RFC, we must first decrease the TTL field. If
 107          *      that reaches zero, we must reply an ICMP control message telling
 108          *      that the packet's lifetime expired.
 109          */
 110         if (ip_hdr(skb)->ttl <= 1)
 111                 goto too_many_hops;
 112
 113         if (!xfrm4_route_forward(skb))
 114                 goto drop;
 115
 116         rt = skb_rtable(skb);
 117
 118         if (opt->is_strictroute && rt->rt_uses_gateway)
 119                 goto sr_failed;
 120
 121         IPCB(skb)->flags |= IPSKB_FORWARDED;
 122         mtu = ip_dst_mtu_maybe_forward(&rt->dst, true);
 123         if (ip_exceeds_mtu(skb, mtu)) {
 124                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
 125                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
 126                           htonl(mtu));
 127                 goto drop;
 128         }
 129
 130         /* We are about to mangle packet. Copy it! */
 131         if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+rt->dst.header_len))
 132                 goto drop;
 133         iph = ip_hdr(skb);
 134
 135         /* Decrease ttl after skb cow done */
 136         ip_decrease_ttl(iph);
 137
 138         /*
 139          *      We now generate an ICMP HOST REDIRECT giving the route
 140          *      we calculated.
 141          */
 142         if (IPCB(skb)->flags & IPSKB_DOREDIRECT && !opt->srr &&
 143             !skb_sec_path(skb))
 144                 ip_rt_send_redirect(skb);
 145
 146         skb->priority = rt_tos2priority(iph->tos);
 147
 148         return NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
 149                        net, NULL, skb, skb->dev, rt->dst.dev,
 150                        ip_forward_finish);
 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         IP_INC_STATS_BH(net, IPSTATS_MIB_INHDRERRORS);
 162         icmp_send(skb, ICMP_TIME_EXCEEDED, ICMP_EXC_TTL, 0);
 163 drop:
 164         kfree_skb(skb);
 165         return NET_RX_DROP;
 166 }