xref: /linux/net/ipv4/ip_forward.c (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
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->ignore_df;
46 }
47 
48 static bool ip_exceeds_mtu(const struct sk_buff *skb, unsigned int mtu)
49 {
50 	if (skb->len <= mtu)
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 
60 static int ip_forward_finish(struct sk_buff *skb)
61 {
62 	struct ip_options *opt	= &(IPCB(skb)->opt);
63 
64 	IP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTFORWDATAGRAMS);
65 	IP_ADD_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTOCTETS, skb->len);
66 
67 	if (unlikely(opt->optlen))
68 		ip_forward_options(skb);
69 
70 	return dst_output(skb);
71 }
72 
73 int ip_forward(struct sk_buff *skb)
74 {
75 	u32 mtu;
76 	struct iphdr *iph;	/* Our header */
77 	struct rtable *rt;	/* Route we use */
78 	struct ip_options *opt	= &(IPCB(skb)->opt);
79 
80 	/* that should never happen */
81 	if (skb->pkt_type != PACKET_HOST)
82 		goto drop;
83 
84 	if (skb_warn_if_lro(skb))
85 		goto drop;
86 
87 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_FWD, skb))
88 		goto drop;
89 
90 	if (IPCB(skb)->opt.router_alert && ip_call_ra_chain(skb))
91 		return NET_RX_SUCCESS;
92 
93 	skb_forward_csum(skb);
94 
95 	/*
96 	 *	According to the RFC, we must first decrease the TTL field. If
97 	 *	that reaches zero, we must reply an ICMP control message telling
98 	 *	that the packet's lifetime expired.
99 	 */
100 	if (ip_hdr(skb)->ttl <= 1)
101 		goto too_many_hops;
102 
103 	if (!xfrm4_route_forward(skb))
104 		goto drop;
105 
106 	rt = skb_rtable(skb);
107 
108 	if (opt->is_strictroute && rt->rt_uses_gateway)
109 		goto sr_failed;
110 
111 	IPCB(skb)->flags |= IPSKB_FORWARDED;
112 	mtu = ip_dst_mtu_maybe_forward(&rt->dst, true);
113 	if (!ip_may_fragment(skb) && ip_exceeds_mtu(skb, mtu)) {
114 		IP_INC_STATS(dev_net(rt->dst.dev), IPSTATS_MIB_FRAGFAILS);
115 		icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
116 			  htonl(mtu));
117 		goto drop;
118 	}
119 
120 	/* We are about to mangle packet. Copy it! */
121 	if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+rt->dst.header_len))
122 		goto drop;
123 	iph = ip_hdr(skb);
124 
125 	/* Decrease ttl after skb cow done */
126 	ip_decrease_ttl(iph);
127 
128 	/*
129 	 *	We now generate an ICMP HOST REDIRECT giving the route
130 	 *	we calculated.
131 	 */
132 	if (rt->rt_flags&RTCF_DOREDIRECT && !opt->srr && !skb_sec_path(skb))
133 		ip_rt_send_redirect(skb);
134 
135 	skb->priority = rt_tos2priority(iph->tos);
136 
137 	return NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD, skb, skb->dev,
138 		       rt->dst.dev, ip_forward_finish);
139 
140 sr_failed:
141 	/*
142 	 *	Strict routing permits no gatewaying
143 	 */
144 	 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_SR_FAILED, 0);
145 	 goto drop;
146 
147 too_many_hops:
148 	/* Tell the sender its packet died... */
149 	IP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_INHDRERRORS);
150 	icmp_send(skb, ICMP_TIME_EXCEEDED, ICMP_EXC_TTL, 0);
151 drop:
152 	kfree_skb(skb);
153 	return NET_RX_DROP;
154 }
155