blob: afdeeb5bec251ac773f03d8c7f0b830c9fa137e2 [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 * Definitions for the IP router.
7 *
8 * Version: @(#)route.h 1.0.4 05/27/93
9 *
Jesper Juhl02c30a82005-05-05 16:16:16 -070010 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Fixes:
13 * Alan Cox : Reformatted. Added ip_rt_local()
14 * Alan Cox : Support for TCP parameters.
15 * Alexey Kuznetsov: Major changes for new routing code.
16 * Mike McLagan : Routing by source
17 * Robert Olsson : Added rt_cache statistics
18 *
19 * This program is free software; you can redistribute it and/or
20 * modify it under the terms of the GNU General Public License
21 * as published by the Free Software Foundation; either version
22 * 2 of the License, or (at your option) any later version.
23 */
24#ifndef _ROUTE_H
25#define _ROUTE_H
26
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <net/dst.h>
28#include <net/inetpeer.h>
29#include <net/flow.h>
KOVACS Krisztian79876872008-10-01 07:35:39 -070030#include <net/inet_sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/in_route.h>
32#include <linux/rtnetlink.h>
David S. Millerc6cffba2012-07-26 11:14:38 +000033#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/route.h>
35#include <linux/ip.h>
36#include <linux/cache.h>
Venkat Yekkiralabeb8d132006-08-04 23:12:42 -070037#include <linux/security.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#define RTO_ONLINK 0x01
40
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#define RT_CONN_FLAGS(sk) (RT_TOS(inet_sk(sk)->tos) | sock_flag(sk, SOCK_LOCALROUTE))
42
43struct fib_nh;
David S. Miller62fa8a82011-01-26 20:51:05 -080044struct fib_info;
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +000045struct rtable {
Changli Gaod8d1f302010-06-10 23:31:35 -070046 struct dst_entry dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
Eric Dumazet29e75252008-01-31 17:05:09 -080048 int rt_genid;
Eric Dumazet95c96172012-04-15 05:58:06 +000049 unsigned int rt_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 __u16 rt_type;
Julian Anastasov155e8332012-10-08 11:41:18 +000051 __u8 rt_is_input;
52 __u8 rt_uses_gateway;
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
Linus Torvalds1da177e2005-04-16 15:20:36 -070054 int rt_iif;
55
56 /* Info on neighbour */
Al Virof2c3fe22006-09-26 21:26:42 -070057 __be32 rt_gateway;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 /* Miscellaneous cached information */
David S. Miller59436342012-07-10 06:58:42 -070060 u32 rt_pmtu;
David S. Millercaacf052012-07-31 15:06:50 -070061
62 struct list_head rt_uncached;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063};
64
Steffen Klassertb8400f32011-11-23 02:14:15 +000065static inline bool rt_is_input_route(const struct rtable *rt)
David S. Millerc7537962010-11-11 17:07:48 -080066{
David S. Miller9917e1e82012-07-17 14:44:26 -070067 return rt->rt_is_input != 0;
David S. Millerc7537962010-11-11 17:07:48 -080068}
69
Steffen Klassertb8400f32011-11-23 02:14:15 +000070static inline bool rt_is_output_route(const struct rtable *rt)
David S. Millerc7537962010-11-11 17:07:48 -080071{
David S. Miller9917e1e82012-07-17 14:44:26 -070072 return rt->rt_is_input == 0;
David S. Millerc7537962010-11-11 17:07:48 -080073}
74
David S. Millerf8126f12012-07-13 05:03:45 -070075static inline __be32 rt_nexthop(const struct rtable *rt, __be32 daddr)
76{
77 if (rt->rt_gateway)
78 return rt->rt_gateway;
79 return daddr;
80}
81
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +000082struct ip_rt_acct {
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 __u32 o_bytes;
84 __u32 o_packets;
85 __u32 i_bytes;
86 __u32 i_packets;
87};
88
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +000089struct rt_cache_stat {
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 unsigned int in_hit;
91 unsigned int in_slow_tot;
92 unsigned int in_slow_mc;
93 unsigned int in_no_route;
94 unsigned int in_brd;
95 unsigned int in_martian_dst;
96 unsigned int in_martian_src;
97 unsigned int out_hit;
98 unsigned int out_slow_tot;
99 unsigned int out_slow_mc;
100 unsigned int gc_total;
101 unsigned int gc_ignored;
102 unsigned int gc_goal_miss;
103 unsigned int gc_dst_overflow;
104 unsigned int in_hlist_search;
105 unsigned int out_hlist_search;
106};
107
Tejun Heo7d720c32010-02-16 15:20:26 +0000108extern struct ip_rt_acct __percpu *ip_rt_acct;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
110struct in_device;
111extern int ip_rt_init(void);
Nicolas Dichtelbafa6d92012-09-07 00:45:29 +0000112extern void rt_cache_flush(struct net *net);
David S. Millercaacf052012-07-31 15:06:50 -0700113extern void rt_flush_dev(struct net_device *dev);
David S. Miller813b3b52011-04-28 14:48:42 -0700114extern struct rtable *__ip_route_output_key(struct net *, struct flowi4 *flp);
David S. Miller9d6ec932011-03-12 01:12:47 -0500115extern struct rtable *ip_route_output_flow(struct net *, struct flowi4 *flp,
David S. Millerb23dd4f2011-03-02 14:31:35 -0800116 struct sock *sk);
David S. Miller2774c132011-03-01 14:59:04 -0800117extern struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig);
Eric Dumazet407eadd2010-05-10 11:32:55 +0000118
David S. Miller9d6ec932011-03-12 01:12:47 -0500119static inline struct rtable *ip_route_output_key(struct net *net, struct flowi4 *flp)
David S. Miller5bfa7872011-03-02 14:56:30 -0800120{
121 return ip_route_output_flow(net, flp, NULL);
122}
123
David S. Miller78fbfd82011-03-12 00:00:52 -0500124static inline struct rtable *ip_route_output(struct net *net, __be32 daddr,
125 __be32 saddr, u8 tos, int oif)
126{
David S. Miller9d6ec932011-03-12 01:12:47 -0500127 struct flowi4 fl4 = {
128 .flowi4_oif = oif,
Roland Dreierc5d21c42012-06-10 20:05:24 +0000129 .flowi4_tos = tos,
David S. Miller9d6ec932011-03-12 01:12:47 -0500130 .daddr = daddr,
131 .saddr = saddr,
David S. Miller78fbfd82011-03-12 00:00:52 -0500132 };
David S. Miller9d6ec932011-03-12 01:12:47 -0500133 return ip_route_output_key(net, &fl4);
David S. Miller78fbfd82011-03-12 00:00:52 -0500134}
135
David S. Miller31e45432011-05-03 20:25:42 -0700136static inline struct rtable *ip_route_output_ports(struct net *net, struct flowi4 *fl4,
137 struct sock *sk,
David S. Miller78fbfd82011-03-12 00:00:52 -0500138 __be32 daddr, __be32 saddr,
139 __be16 dport, __be16 sport,
140 __u8 proto, __u8 tos, int oif)
141{
David S. Miller31e45432011-05-03 20:25:42 -0700142 flowi4_init_output(fl4, oif, sk ? sk->sk_mark : 0, tos,
David S. Miller94b92b82011-03-31 04:52:59 -0700143 RT_SCOPE_UNIVERSE, proto,
144 sk ? inet_sk_flowi_flags(sk) : 0,
145 daddr, saddr, dport, sport);
David S. Miller78fbfd82011-03-12 00:00:52 -0500146 if (sk)
David S. Miller31e45432011-05-03 20:25:42 -0700147 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
148 return ip_route_output_flow(net, fl4, sk);
David S. Miller78fbfd82011-03-12 00:00:52 -0500149}
150
David S. Millercbb1e852011-05-04 12:33:34 -0700151static inline struct rtable *ip_route_output_gre(struct net *net, struct flowi4 *fl4,
David S. Miller78fbfd82011-03-12 00:00:52 -0500152 __be32 daddr, __be32 saddr,
153 __be32 gre_key, __u8 tos, int oif)
154{
David S. Millercbb1e852011-05-04 12:33:34 -0700155 memset(fl4, 0, sizeof(*fl4));
156 fl4->flowi4_oif = oif;
157 fl4->daddr = daddr;
158 fl4->saddr = saddr;
159 fl4->flowi4_tos = tos;
160 fl4->flowi4_proto = IPPROTO_GRE;
161 fl4->fl4_gre_key = gre_key;
162 return ip_route_output_key(net, fl4);
David S. Miller78fbfd82011-03-12 00:00:52 -0500163}
164
David S. Millerc6cffba2012-07-26 11:14:38 +0000165extern int ip_route_input_noref(struct sk_buff *skb, __be32 dst, __be32 src,
166 u8 tos, struct net_device *devin);
167
168static inline int ip_route_input(struct sk_buff *skb, __be32 dst, __be32 src,
169 u8 tos, struct net_device *devin)
170{
171 int err;
172
173 rcu_read_lock();
174 err = ip_route_input_noref(skb, dst, src, tos, devin);
175 if (!err)
176 skb_dst_force(skb);
177 rcu_read_unlock();
178
179 return err;
180}
Eric Dumazet407eadd2010-05-10 11:32:55 +0000181
David S. Miller36393392012-06-14 22:21:46 -0700182extern void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
183 int oif, u32 mark, u8 protocol, int flow_flags);
184extern void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu);
David S. Millerb42597e2012-07-11 21:25:45 -0700185extern void ipv4_redirect(struct sk_buff *skb, struct net *net,
186 int oif, u32 mark, u8 protocol, int flow_flags);
187extern void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk);
David S. Miller36393392012-06-14 22:21:46 -0700188extern void ip_rt_send_redirect(struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
Eric Dumazet95c96172012-04-15 05:58:06 +0000190extern unsigned int inet_addr_type(struct net *net, __be32 addr);
191extern unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev, __be32 addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192extern void ip_rt_multicast_event(struct in_device *);
Denis V. Lunev1bad1182008-01-10 03:29:53 -0800193extern int ip_rt_ioctl(struct net *, unsigned int cmd, void __user *arg);
David S. Miller8e363602011-05-13 17:29:41 -0400194extern void ip_rt_get_source(u8 *src, struct sk_buff *skb, struct rtable *rt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195extern int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb);
196
Jamal Hadi Salim0ff60a42005-11-22 14:47:37 -0800197struct in_ifaddr;
198extern void fib_add_ifaddr(struct in_ifaddr *);
Julian Anastasove6abbaa2011-03-19 12:13:49 +0000199extern void fib_del_ifaddr(struct in_ifaddr *, struct in_ifaddr *);
Jamal Hadi Salim0ff60a42005-11-22 14:47:37 -0800200
Eric Dumazet6da025f2012-10-28 22:33:23 +0000201static inline void ip_rt_put(struct rtable *rt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202{
Eric Dumazet6da025f2012-10-28 22:33:23 +0000203 /* dst_release() accepts a NULL parameter.
204 * We rely on dst being first structure in struct rtable
205 */
206 BUILD_BUG_ON(offsetof(struct rtable, dst) != 0);
207 dst_release(&rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208}
209
210#define IPTOS_RT_MASK (IPTOS_TOS_MASK & ~3)
211
Philippe De Muyter4839c522007-07-09 15:32:57 -0700212extern const __u8 ip_tos2prio[16];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213
214static inline char rt_tos2priority(u8 tos)
215{
216 return ip_tos2prio[IPTOS_TOS(tos)>>1];
217}
218
David S. Miller2d7192d2011-04-26 13:28:44 -0700219/* ip_route_connect() and ip_route_newports() work in tandem whilst
220 * binding a socket for a new outgoing connection.
221 *
222 * In order to use IPSEC properly, we must, in the end, have a
223 * route that was looked up using all available keys including source
224 * and destination ports.
225 *
226 * However, if a source port needs to be allocated (the user specified
227 * a wildcard source port) we need to obtain addressing information
228 * in order to perform that allocation.
229 *
230 * So ip_route_connect() looks up a route using wildcarded source and
231 * destination ports in the key, simply so that we can get a pair of
232 * addresses to use for port allocation.
233 *
234 * Later, once the ports are allocated, ip_route_newports() will make
235 * another route lookup if needed to make sure we catch any IPSEC
236 * rules keyed on the port information.
237 *
238 * The callers allocate the flow key on their stack, and must pass in
239 * the same flowi4 object to both the ip_route_connect() and the
240 * ip_route_newports() calls.
241 */
KOVACS Krisztian79876872008-10-01 07:35:39 -0700242
David S. Miller2d7192d2011-04-26 13:28:44 -0700243static inline void ip_route_connect_init(struct flowi4 *fl4, __be32 dst, __be32 src,
244 u32 tos, int oif, u8 protocol,
245 __be16 sport, __be16 dport,
246 struct sock *sk, bool can_sleep)
247{
248 __u8 flow_flags = 0;
249
KOVACS Krisztian79876872008-10-01 07:35:39 -0700250 if (inet_sk(sk)->transparent)
David S. Miller94b92b82011-03-31 04:52:59 -0700251 flow_flags |= FLOWI_FLAG_ANYSRC;
David S. Miller5df65e52011-03-01 14:22:19 -0800252 if (can_sleep)
David S. Miller94b92b82011-03-31 04:52:59 -0700253 flow_flags |= FLOWI_FLAG_CAN_SLEEP;
254
David S. Miller2d7192d2011-04-26 13:28:44 -0700255 flowi4_init_output(fl4, oif, sk->sk_mark, tos, RT_SCOPE_UNIVERSE,
David S. Miller94b92b82011-03-31 04:52:59 -0700256 protocol, flow_flags, dst, src, dport, sport);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257}
258
David S. Miller2d7192d2011-04-26 13:28:44 -0700259static inline struct rtable *ip_route_connect(struct flowi4 *fl4,
260 __be32 dst, __be32 src, u32 tos,
261 int oif, u8 protocol,
262 __be16 sport, __be16 dport,
263 struct sock *sk, bool can_sleep)
264{
265 struct net *net = sock_net(sk);
266 struct rtable *rt;
267
268 ip_route_connect_init(fl4, dst, src, tos, oif, protocol,
269 sport, dport, sk, can_sleep);
270
271 if (!dst || !src) {
272 rt = __ip_route_output_key(net, fl4);
273 if (IS_ERR(rt))
274 return rt;
David S. Miller2d7192d2011-04-26 13:28:44 -0700275 ip_rt_put(rt);
Julian Anastasove6b45242012-02-04 13:04:46 +0000276 flowi4_update_output(fl4, oif, tos, fl4->daddr, fl4->saddr);
David S. Miller2d7192d2011-04-26 13:28:44 -0700277 }
278 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
279 return ip_route_output_flow(net, fl4, sk);
280}
281
282static inline struct rtable *ip_route_newports(struct flowi4 *fl4, struct rtable *rt,
283 __be16 orig_sport, __be16 orig_dport,
284 __be16 sport, __be16 dport,
285 struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286{
David S. Millerdca8b082011-02-24 13:38:12 -0800287 if (sport != orig_sport || dport != orig_dport) {
David S. Miller2d7192d2011-04-26 13:28:44 -0700288 fl4->fl4_dport = dport;
289 fl4->fl4_sport = sport;
David S. Millerb23dd4f2011-03-02 14:31:35 -0800290 ip_rt_put(rt);
Julian Anastasove6b45242012-02-04 13:04:46 +0000291 flowi4_update_output(fl4, sk->sk_bound_dev_if,
292 RT_CONN_FLAGS(sk), fl4->daddr,
293 fl4->saddr);
David S. Miller2d7192d2011-04-26 13:28:44 -0700294 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
295 return ip_route_output_flow(sock_net(sk), fl4, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 }
David S. Millerb23dd4f2011-03-02 14:31:35 -0800297 return rt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298}
299
KOVACS Krisztian1668e012008-10-01 07:33:10 -0700300static inline int inet_iif(const struct sk_buff *skb)
301{
David S. Miller13378ca2012-07-23 13:57:45 -0700302 int iif = skb_rtable(skb)->rt_iif;
303
304 if (iif)
305 return iif;
306 return skb->skb_iif;
KOVACS Krisztian1668e012008-10-01 07:33:10 -0700307}
308
David S. Miller323e1262010-12-12 21:55:08 -0800309extern int sysctl_ip_default_ttl;
310
311static inline int ip4_dst_hoplimit(const struct dst_entry *dst)
312{
313 int hoplimit = dst_metric_raw(dst, RTAX_HOPLIMIT);
314
315 if (hoplimit == 0)
316 hoplimit = sysctl_ip_default_ttl;
317 return hoplimit;
318}
319
Hannes Frederic Sowa0ea9d5e2013-08-13 04:35:58 +0200320static inline int ip_skb_dst_mtu(struct sk_buff *skb)
321{
322 struct inet_sock *inet = skb->sk ? inet_sk(skb->sk) : NULL;
323
324 return (inet && inet->pmtudisc == IP_PMTUDISC_PROBE) ?
325 skb_dst(skb)->dev->mtu : dst_mtu(skb_dst(skb));
326}
327
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328#endif /* _ROUTE_H */