blob: 694de3b7aebfede6073433201e1200cf72008997 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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.
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09007 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * Authors: see ip.c
9 *
10 * Fixes:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090011 * Many : Split from ip.c , see ip_input.c for
Linus Torvalds1da177e2005-04-16 15:20:36 -070012 * history.
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090013 * Dave Gregorich : NULL ip_rt_put fix for multicast
Linus Torvalds1da177e2005-04-16 15:20:36 -070014 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/types.h>
23#include <linux/mm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/skbuff.h>
25#include <linux/ip.h>
26#include <linux/icmp.h>
27#include <linux/netdevice.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090028#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#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
Patrick McHardy861d0482007-10-15 01:48:39 -070042static int ip_forward_finish(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -070043{
Daniel Baluta5e73ea12012-04-15 01:34:41 +000044 struct ip_options *opt = &(IPCB(skb)->opt);
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Eric Dumazetadf30902009-06-02 05:19:30 +000046 IP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTFORWDATAGRAMS);
Vincent Bernat2d8dbb02012-06-05 03:41:42 +000047 IP_ADD_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTOCTETS, skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
49 if (unlikely(opt->optlen))
50 ip_forward_options(skb);
51
52 return dst_output(skb);
53}
54
55int ip_forward(struct sk_buff *skb)
56{
57 struct iphdr *iph; /* Our header */
58 struct rtable *rt; /* Route we use */
Daniel Baluta5e73ea12012-04-15 01:34:41 +000059 struct ip_options *opt = &(IPCB(skb)->opt);
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Ben Hutchings4497b072008-06-19 16:22:28 -070061 if (skb_warn_if_lro(skb))
62 goto drop;
63
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 if (!xfrm4_policy_check(NULL, XFRM_POLICY_FWD, skb))
65 goto drop;
66
67 if (IPCB(skb)->opt.router_alert && ip_call_ra_chain(skb))
68 return NET_RX_SUCCESS;
69
70 if (skb->pkt_type != PACKET_HOST)
71 goto drop;
72
Herbert Xu35fc92a2007-03-26 23:22:20 -070073 skb_forward_csum(skb);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090074
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 /*
76 * According to the RFC, we must first decrease the TTL field. If
77 * that reaches zero, we must reply an ICMP control message telling
78 * that the packet's lifetime expired.
79 */
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -070080 if (ip_hdr(skb)->ttl <= 1)
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090081 goto too_many_hops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
83 if (!xfrm4_route_forward(skb))
84 goto drop;
85
Eric Dumazet511c3f92009-06-02 05:14:27 +000086 rt = skb_rtable(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
Julian Anastasov155e8332012-10-08 11:41:18 +000088 if (opt->is_strictroute && rt->rt_uses_gateway)
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 goto sr_failed;
90
Changli Gaod8d1f302010-06-10 23:31:35 -070091 if (unlikely(skb->len > dst_mtu(&rt->dst) && !skb_is_gso(skb) &&
YOSHIFUJI Hideaki9c681b42007-07-19 10:43:47 +090092 (ip_hdr(skb)->frag_off & htons(IP_DF))) && !skb->local_df) {
Changli Gaod8d1f302010-06-10 23:31:35 -070093 IP_INC_STATS(dev_net(rt->dst.dev), IPSTATS_MIB_FRAGFAILS);
John Heffner9af39122007-03-25 23:32:29 -070094 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
Changli Gaod8d1f302010-06-10 23:31:35 -070095 htonl(dst_mtu(&rt->dst)));
John Heffner9af39122007-03-25 23:32:29 -070096 goto drop;
97 }
98
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 /* We are about to mangle packet. Copy it! */
Changli Gaod8d1f302010-06-10 23:31:35 -0700100 if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+rt->dst.header_len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 goto drop;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700102 iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
104 /* Decrease ttl after skb cow done */
105 ip_decrease_ttl(iph);
106
107 /*
108 * We now generate an ICMP HOST REDIRECT giving the route
109 * we calculated.
110 */
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -0700111 if (rt->rt_flags&RTCF_DOREDIRECT && !opt->srr && !skb_sec_path(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 ip_rt_send_redirect(skb);
113
114 skb->priority = rt_tos2priority(iph->tos);
115
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100116 return NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD, skb, skb->dev,
Changli Gaod8d1f302010-06-10 23:31:35 -0700117 rt->dst.dev, ip_forward_finish);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
119sr_failed:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900120 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 * Strict routing permits no gatewaying
122 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900123 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_SR_FAILED, 0);
124 goto drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125
126too_many_hops:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900127 /* Tell the sender its packet died... */
Eric Dumazetadf30902009-06-02 05:19:30 +0000128 IP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_INHDRERRORS);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900129 icmp_send(skb, ICMP_TIME_EXCEEDED, ICMP_EXC_TTL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130drop:
131 kfree_skb(skb);
132 return NET_RX_DROP;
133}