net/ipv4/ip_forward.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   4  *              operating system.  INET is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              The IP forwarding functionality.
   8  *
   9  * Authors:     see ip.c
  10  *
  11  * Fixes:
  12  *              Many            :       Split from ip.c , see ip_input.c for
  13  *                                      history.
  14  *              Dave Gregorich  :       NULL ip_rt_put fix for multicast
  15  *                                      routing.
  16  *              Jos Vos         :       Add call_out_firewall before sending,
  17  *                                      use output device for accounting.
  18  *              Jos Vos         :       Call forward firewall after routing
  19  *                                      (always use output device).
  20  *              Mike McLagan    :       Routing by source
  21  */
  22
  23 #include <linux/types.h>
  24 #include <linux/mm.h>
  25 #include <linux/skbuff.h>
  26 #include <linux/ip.h>
  27 #include <linux/icmp.h>
  28 #include <linux/netdevice.h>
  29 #include <linux/slab.h>
  30 #include <net/sock.h>
  31 #include <net/ip.h>
  32 #include <net/tcp.h>
  33 #include <net/udp.h>
  34 #include <net/icmp.h>
  35 #include <linux/tcp.h>
  36 #include <linux/udp.h>
  37 #include <linux/netfilter_ipv4.h>
  38 #include <net/checksum.h>
  39 #include <linux/route.h>
  40 #include <net/route.h>
  41 #include <net/xfrm.h>
  42
  43 static bool ip_exceeds_mtu(const struct sk_buff *skb, unsigned int mtu)
  44 {
  45         if (skb->len <= mtu)
  46                 return false;
  47
  48         if (unlikely((ip_hdr(skb)->frag_off & htons(IP_DF)) == 0))
  49                 return false;
  50
  51         /* original fragment exceeds mtu and DF is set */
  52         if (unlikely(IPCB(skb)->frag_max_size > mtu))
  53                 return true;
  54
  55         if (skb->ignore_df)
  56                 return false;
  57
  58         if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
  59                 return false;
  60
  61         return true;
  62 }
  63
  64
  65 static int ip_forward_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
  66 {
  67         struct ip_options *opt  = &(IPCB(skb)->opt);
  68
  69         __IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
  70         __IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
  71
  72 #ifdef CONFIG_NET_SWITCHDEV
  73         if (skb->offload_l3_fwd_mark) {
  74                 consume_skb(skb);
  75                 return 0;
  76         }
  77 #endif
  78
  79         if (unlikely(opt->optlen))
  80                 ip_forward_options(skb);
  81
  82         skb->tstamp = 0;
  83         return dst_output(net, sk, skb);
  84 }
  85
  86 int ip_forward(struct sk_buff *skb)
  87 {
  88         u32 mtu;
  89         struct iphdr *iph;      /* Our header */
  90         struct rtable *rt;      /* Route we use */
  91         struct ip_options *opt  = &(IPCB(skb)->opt);
  92         struct net *net;
  93
  94         /* that should never happen */
  95         if (skb->pkt_type != PACKET_HOST)
  96                 goto drop;
  97
  98         if (unlikely(skb->sk))
  99                 goto drop;
 100
 101         if (skb_warn_if_lro(skb))
 102                 goto drop;
 103
 104         if (!xfrm4_policy_check(NULL, XFRM_POLICY_FWD, skb))
 105                 goto drop;
 106
 107         if (IPCB(skb)->opt.router_alert && ip_call_ra_chain(skb))
 108                 return NET_RX_SUCCESS;
 109
 110         skb_forward_csum(skb);
 111         net = dev_net(skb->dev);
 112
 113         /*
 114          *      According to the RFC, we must first decrease the TTL field. If
 115          *      that reaches zero, we must reply an ICMP control message telling
 116          *      that the packet's lifetime expired.
 117          */
 118         if (ip_hdr(skb)->ttl <= 1)
 119                 goto too_many_hops;
 120
 121         if (!xfrm4_route_forward(skb))
 122                 goto drop;
 123
 124         rt = skb_rtable(skb);
 125
 126         if (opt->is_strictroute && rt->rt_uses_gateway)
 127                 goto sr_failed;
 128
 129         IPCB(skb)->flags |= IPSKB_FORWARDED;
 130         mtu = ip_dst_mtu_maybe_forward(&rt->dst, true);
 131         if (ip_exceeds_mtu(skb, mtu)) {
 132                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
 133                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
 134                           htonl(mtu));
 135                 goto drop;
 136         }
 137
 138         /* We are about to mangle packet. Copy it! */
 139         if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+rt->dst.header_len))
 140                 goto drop;
 141         iph = ip_hdr(skb);
 142
 143         /* Decrease ttl after skb cow done */
 144         ip_decrease_ttl(iph);
 145
 146         /*
 147          *      We now generate an ICMP HOST REDIRECT giving the route
 148          *      we calculated.
 149          */
 150         if (IPCB(skb)->flags & IPSKB_DOREDIRECT && !opt->srr &&
 151             !skb_sec_path(skb))
 152                 ip_rt_send_redirect(skb);
 153
 154         if (net->ipv4.sysctl_ip_fwd_update_priority)
 155                 skb->priority = rt_tos2priority(iph->tos);
 156
 157         return NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
 158                        net, NULL, skb, skb->dev, rt->dst.dev,
 159                        ip_forward_finish);
 160
 161 sr_failed:
 162         /*
 163          *      Strict routing permits no gatewaying
 164          */
 165          icmp_send(skb, ICMP_DEST_UNREACH, ICMP_SR_FAILED, 0);
 166          goto drop;
 167
 168 too_many_hops:
 169         /* Tell the sender its packet died... */
 170         __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
 171         icmp_send(skb, ICMP_TIME_EXCEEDED, ICMP_EXC_TTL, 0);
 172 drop:
 173         kfree_skb(skb);
 174         return NET_RX_DROP;
 175 }