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_may_fragment(const struct sk_buff *skb)
  43 {
  44         return unlikely((ip_hdr(skb)->frag_off & htons(IP_DF)) == 0) ||
  45                !skb->local_df;
  46 }
  47
  48 static bool ip_exceeds_mtu(const struct sk_buff *skb, unsigned int mtu)
  49 {
  50         if (skb->len <= mtu || skb->local_df)
  51                 return false;
  52
  53         if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
  54                 return false;
  55
  56         return true;
  57 }
  58
  59 static bool ip_gso_exceeds_dst_mtu(const struct sk_buff *skb)
  60 {
  61         unsigned int mtu;
  62
  63         if (skb->local_df || !skb_is_gso(skb))
  64                 return false;
  65
  66         mtu = ip_dst_mtu_maybe_forward(skb_dst(skb), true);
  67
  68         /* if seglen > mtu, do software segmentation for IP fragmentation on
  69          * output.  DF bit cannot be set since ip_forward would have sent
  70          * icmp error.
  71          */
  72         return skb_gso_network_seglen(skb) > mtu;
  73 }
  74
  75 /* called if GSO skb needs to be fragmented on forward */
  76 static int ip_forward_finish_gso(struct sk_buff *skb)
  77 {
  78         struct dst_entry *dst = skb_dst(skb);
  79         netdev_features_t features;
  80         struct sk_buff *segs;
  81         int ret = 0;
  82
  83         features = netif_skb_dev_features(skb, dst->dev);
  84         segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
  85         if (IS_ERR(segs)) {
  86                 kfree_skb(skb);
  87                 return -ENOMEM;
  88         }
  89
  90         consume_skb(skb);
  91
  92         do {
  93                 struct sk_buff *nskb = segs->next;
  94                 int err;
  95
  96                 segs->next = NULL;
  97                 err = dst_output(segs);
  98
  99                 if (err && ret == 0)
 100                         ret = err;
 101                 segs = nskb;
 102         } while (segs);
 103
 104         return ret;
 105 }
 106
 107 static int ip_forward_finish(struct sk_buff *skb)
 108 {
 109         struct ip_options *opt  = &(IPCB(skb)->opt);
 110
 111         IP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTFORWDATAGRAMS);
 112         IP_ADD_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTOCTETS, skb->len);
 113
 114         if (unlikely(opt->optlen))
 115                 ip_forward_options(skb);
 116
 117         if (ip_gso_exceeds_dst_mtu(skb))
 118                 return ip_forward_finish_gso(skb);
 119
 120         return dst_output(skb);
 121 }
 122
 123 int ip_forward(struct sk_buff *skb)
 124 {
 125         u32 mtu;
 126         struct iphdr *iph;      /* Our header */
 127         struct rtable *rt;      /* Route we use */
 128         struct ip_options *opt  = &(IPCB(skb)->opt);
 129
 130         if (skb_warn_if_lro(skb))
 131                 goto drop;
 132
 133         if (!xfrm4_policy_check(NULL, XFRM_POLICY_FWD, skb))
 134                 goto drop;
 135
 136         if (IPCB(skb)->opt.router_alert && ip_call_ra_chain(skb))
 137                 return NET_RX_SUCCESS;
 138
 139         if (skb->pkt_type != PACKET_HOST)
 140                 goto drop;
 141
 142         skb_forward_csum(skb);
 143
 144         /*
 145          *      According to the RFC, we must first decrease the TTL field. If
 146          *      that reaches zero, we must reply an ICMP control message telling
 147          *      that the packet's lifetime expired.
 148          */
 149         if (ip_hdr(skb)->ttl <= 1)
 150                 goto too_many_hops;
 151
 152         if (!xfrm4_route_forward(skb))
 153                 goto drop;
 154
 155         rt = skb_rtable(skb);
 156
 157         if (opt->is_strictroute && rt->rt_uses_gateway)
 158                 goto sr_failed;
 159
 160         IPCB(skb)->flags |= IPSKB_FORWARDED;
 161         mtu = ip_dst_mtu_maybe_forward(&rt->dst, true);
 162         if (!ip_may_fragment(skb) && ip_exceeds_mtu(skb, mtu)) {
 163                 IP_INC_STATS(dev_net(rt->dst.dev), IPSTATS_MIB_FRAGFAILS);
 164                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
 165                           htonl(mtu));
 166                 goto drop;
 167         }
 168
 169         /* We are about to mangle packet. Copy it! */
 170         if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+rt->dst.header_len))
 171                 goto drop;
 172         iph = ip_hdr(skb);
 173
 174         /* Decrease ttl after skb cow done */
 175         ip_decrease_ttl(iph);
 176
 177         /*
 178          *      We now generate an ICMP HOST REDIRECT giving the route
 179          *      we calculated.
 180          */
 181         if (rt->rt_flags&RTCF_DOREDIRECT && !opt->srr && !skb_sec_path(skb))
 182                 ip_rt_send_redirect(skb);
 183
 184         skb->priority = rt_tos2priority(iph->tos);
 185
 186         return NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD, skb, skb->dev,
 187                        rt->dst.dev, ip_forward_finish);
 188
 189 sr_failed:
 190         /*
 191          *      Strict routing permits no gatewaying
 192          */
 193          icmp_send(skb, ICMP_DEST_UNREACH, ICMP_SR_FAILED, 0);
 194          goto drop;
 195
 196 too_many_hops:
 197         /* Tell the sender its packet died... */
 198         IP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_INHDRERRORS);
 199         icmp_send(skb, ICMP_TIME_EXCEEDED, ICMP_EXC_TTL, 0);
 200 drop:
 201         kfree_skb(skb);
 202         return NET_RX_DROP;
 203 }