blob: ee81307863d56329c097bb3ec30d06b07828b2d0 [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>
Peter Nørlund79a13152015-09-30 10:12:22 +020031#include <net/ip_fib.h>
David Ahern9478d122015-09-29 20:07:18 -070032#include <net/l3mdev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <linux/in_route.h>
34#include <linux/rtnetlink.h>
David S. Millerc6cffba2012-07-26 11:14:38 +000035#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/route.h>
37#include <linux/ip.h>
38#include <linux/cache.h>
Venkat Yekkiralabeb8d132006-08-04 23:12:42 -070039#include <linux/security.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +010041/* IPv4 datagram length is stored into 16bit field (tot_len) */
42#define IP_MAX_MTU 0xFFFFU
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#define RTO_ONLINK 0x01
45
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#define RT_CONN_FLAGS(sk) (RT_TOS(inet_sk(sk)->tos) | sock_flag(sk, SOCK_LOCALROUTE))
Francesco Fuscoaa661582013-09-24 15:43:09 +020047#define RT_CONN_FLAGS_TOS(sk,tos) (RT_TOS(tos) | sock_flag(sk, SOCK_LOCALROUTE))
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
49struct fib_nh;
David S. Miller62fa8a82011-01-26 20:51:05 -080050struct fib_info;
Eric Dumazet5055c372015-01-14 15:17:06 -080051struct uncached_list;
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +000052struct rtable {
Changli Gaod8d1f302010-06-10 23:31:35 -070053 struct dst_entry dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
Eric Dumazet29e75252008-01-31 17:05:09 -080055 int rt_genid;
Eric Dumazet95c96172012-04-15 05:58:06 +000056 unsigned int rt_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 __u16 rt_type;
Julian Anastasov155e8332012-10-08 11:41:18 +000058 __u8 rt_is_input;
59 __u8 rt_uses_gateway;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 int rt_iif;
62
63 /* Info on neighbour */
Al Virof2c3fe22006-09-26 21:26:42 -070064 __be32 rt_gateway;
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 /* Miscellaneous cached information */
David S. Miller59436342012-07-10 06:58:42 -070067 u32 rt_pmtu;
David S. Millercaacf052012-07-31 15:06:50 -070068
David Ahernb7503e02015-09-02 13:58:35 -070069 u32 rt_table_id;
70
David S. Millercaacf052012-07-31 15:06:50 -070071 struct list_head rt_uncached;
Eric Dumazet5055c372015-01-14 15:17:06 -080072 struct uncached_list *rt_uncached_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -070073};
74
Steffen Klassertb8400f32011-11-23 02:14:15 +000075static inline bool rt_is_input_route(const struct rtable *rt)
David S. Millerc7537962010-11-11 17:07:48 -080076{
David S. Miller9917e1e82012-07-17 14:44:26 -070077 return rt->rt_is_input != 0;
David S. Millerc7537962010-11-11 17:07:48 -080078}
79
Steffen Klassertb8400f32011-11-23 02:14:15 +000080static inline bool rt_is_output_route(const struct rtable *rt)
David S. Millerc7537962010-11-11 17:07:48 -080081{
David S. Miller9917e1e82012-07-17 14:44:26 -070082 return rt->rt_is_input == 0;
David S. Millerc7537962010-11-11 17:07:48 -080083}
84
David S. Millerf8126f12012-07-13 05:03:45 -070085static inline __be32 rt_nexthop(const struct rtable *rt, __be32 daddr)
86{
87 if (rt->rt_gateway)
88 return rt->rt_gateway;
89 return daddr;
90}
91
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +000092struct ip_rt_acct {
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 __u32 o_bytes;
94 __u32 o_packets;
95 __u32 i_bytes;
96 __u32 i_packets;
97};
98
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +000099struct rt_cache_stat {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 unsigned int in_slow_tot;
101 unsigned int in_slow_mc;
102 unsigned int in_no_route;
103 unsigned int in_brd;
104 unsigned int in_martian_dst;
105 unsigned int in_martian_src;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 unsigned int out_slow_tot;
107 unsigned int out_slow_mc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108};
109
Tejun Heo7d720c32010-02-16 15:20:26 +0000110extern struct ip_rt_acct __percpu *ip_rt_acct;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111
112struct in_device;
Joe Perches2702c4b2013-09-22 10:32:22 -0700113
114int ip_rt_init(void);
115void rt_cache_flush(struct net *net);
116void rt_flush_dev(struct net_device *dev);
Peter Nørlund79a13152015-09-30 10:12:22 +0200117struct rtable *__ip_route_output_key_hash(struct net *, struct flowi4 *flp,
118 int mp_hash);
119
120static inline struct rtable *__ip_route_output_key(struct net *net,
121 struct flowi4 *flp)
122{
123 return __ip_route_output_key_hash(net, flp, -1);
124}
125
Joe Perches2702c4b2013-09-22 10:32:22 -0700126struct rtable *ip_route_output_flow(struct net *, struct flowi4 *flp,
Eric Dumazet6f9c9612015-09-25 07:39:10 -0700127 const struct sock *sk);
Joe Perches2702c4b2013-09-22 10:32:22 -0700128struct dst_entry *ipv4_blackhole_route(struct net *net,
129 struct dst_entry *dst_orig);
Eric Dumazet407eadd2010-05-10 11:32:55 +0000130
David S. Miller9d6ec932011-03-12 01:12:47 -0500131static inline struct rtable *ip_route_output_key(struct net *net, struct flowi4 *flp)
David S. Miller5bfa7872011-03-02 14:56:30 -0800132{
133 return ip_route_output_flow(net, flp, NULL);
134}
135
David S. Miller78fbfd82011-03-12 00:00:52 -0500136static inline struct rtable *ip_route_output(struct net *net, __be32 daddr,
137 __be32 saddr, u8 tos, int oif)
138{
David S. Miller9d6ec932011-03-12 01:12:47 -0500139 struct flowi4 fl4 = {
140 .flowi4_oif = oif,
Roland Dreierc5d21c42012-06-10 20:05:24 +0000141 .flowi4_tos = tos,
David S. Miller9d6ec932011-03-12 01:12:47 -0500142 .daddr = daddr,
143 .saddr = saddr,
David S. Miller78fbfd82011-03-12 00:00:52 -0500144 };
David S. Miller9d6ec932011-03-12 01:12:47 -0500145 return ip_route_output_key(net, &fl4);
David S. Miller78fbfd82011-03-12 00:00:52 -0500146}
147
David S. Miller31e45432011-05-03 20:25:42 -0700148static inline struct rtable *ip_route_output_ports(struct net *net, struct flowi4 *fl4,
149 struct sock *sk,
David S. Miller78fbfd82011-03-12 00:00:52 -0500150 __be32 daddr, __be32 saddr,
151 __be16 dport, __be16 sport,
152 __u8 proto, __u8 tos, int oif)
153{
David S. Miller31e45432011-05-03 20:25:42 -0700154 flowi4_init_output(fl4, oif, sk ? sk->sk_mark : 0, tos,
David S. Miller94b92b82011-03-31 04:52:59 -0700155 RT_SCOPE_UNIVERSE, proto,
156 sk ? inet_sk_flowi_flags(sk) : 0,
157 daddr, saddr, dport, sport);
David S. Miller78fbfd82011-03-12 00:00:52 -0500158 if (sk)
David S. Miller31e45432011-05-03 20:25:42 -0700159 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
160 return ip_route_output_flow(net, fl4, sk);
David S. Miller78fbfd82011-03-12 00:00:52 -0500161}
162
David S. Millercbb1e852011-05-04 12:33:34 -0700163static inline struct rtable *ip_route_output_gre(struct net *net, struct flowi4 *fl4,
David S. Miller78fbfd82011-03-12 00:00:52 -0500164 __be32 daddr, __be32 saddr,
165 __be32 gre_key, __u8 tos, int oif)
166{
David S. Millercbb1e852011-05-04 12:33:34 -0700167 memset(fl4, 0, sizeof(*fl4));
168 fl4->flowi4_oif = oif;
169 fl4->daddr = daddr;
170 fl4->saddr = saddr;
171 fl4->flowi4_tos = tos;
172 fl4->flowi4_proto = IPPROTO_GRE;
173 fl4->fl4_gre_key = gre_key;
174 return ip_route_output_key(net, fl4);
David S. Miller78fbfd82011-03-12 00:00:52 -0500175}
176
Joe Perches2702c4b2013-09-22 10:32:22 -0700177int ip_route_input_noref(struct sk_buff *skb, __be32 dst, __be32 src,
178 u8 tos, struct net_device *devin);
David S. Millerc6cffba2012-07-26 11:14:38 +0000179
180static inline int ip_route_input(struct sk_buff *skb, __be32 dst, __be32 src,
181 u8 tos, struct net_device *devin)
182{
183 int err;
184
185 rcu_read_lock();
186 err = ip_route_input_noref(skb, dst, src, tos, devin);
187 if (!err)
188 skb_dst_force(skb);
189 rcu_read_unlock();
190
191 return err;
192}
Eric Dumazet407eadd2010-05-10 11:32:55 +0000193
Joe Perches2702c4b2013-09-22 10:32:22 -0700194void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu, int oif,
195 u32 mark, u8 protocol, int flow_flags);
196void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu);
197void ipv4_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
198 u8 protocol, int flow_flags);
199void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk);
200void ip_rt_send_redirect(struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
Joe Perches2702c4b2013-09-22 10:32:22 -0700202unsigned int inet_addr_type(struct net *net, __be32 addr);
David Ahern9b8ff512015-09-01 14:26:35 -0600203unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id);
Joe Perches2702c4b2013-09-22 10:32:22 -0700204unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
205 __be32 addr);
David Ahern30bbaa12015-08-13 14:59:05 -0600206unsigned int inet_addr_type_dev_table(struct net *net,
207 const struct net_device *dev,
208 __be32 addr);
Joe Perches2702c4b2013-09-22 10:32:22 -0700209void ip_rt_multicast_event(struct in_device *);
210int ip_rt_ioctl(struct net *, unsigned int cmd, void __user *arg);
211void ip_rt_get_source(u8 *src, struct sk_buff *skb, struct rtable *rt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
Jamal Hadi Salim0ff60a42005-11-22 14:47:37 -0800213struct in_ifaddr;
Joe Perches2702c4b2013-09-22 10:32:22 -0700214void fib_add_ifaddr(struct in_ifaddr *);
215void fib_del_ifaddr(struct in_ifaddr *, struct in_ifaddr *);
Jamal Hadi Salim0ff60a42005-11-22 14:47:37 -0800216
Eric Dumazet6da025f2012-10-28 22:33:23 +0000217static inline void ip_rt_put(struct rtable *rt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218{
Eric Dumazet6da025f2012-10-28 22:33:23 +0000219 /* dst_release() accepts a NULL parameter.
220 * We rely on dst being first structure in struct rtable
221 */
222 BUILD_BUG_ON(offsetof(struct rtable, dst) != 0);
223 dst_release(&rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224}
225
226#define IPTOS_RT_MASK (IPTOS_TOS_MASK & ~3)
227
Philippe De Muyter4839c522007-07-09 15:32:57 -0700228extern const __u8 ip_tos2prio[16];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
230static inline char rt_tos2priority(u8 tos)
231{
232 return ip_tos2prio[IPTOS_TOS(tos)>>1];
233}
234
David S. Miller2d7192d2011-04-26 13:28:44 -0700235/* ip_route_connect() and ip_route_newports() work in tandem whilst
236 * binding a socket for a new outgoing connection.
237 *
238 * In order to use IPSEC properly, we must, in the end, have a
239 * route that was looked up using all available keys including source
240 * and destination ports.
241 *
242 * However, if a source port needs to be allocated (the user specified
243 * a wildcard source port) we need to obtain addressing information
244 * in order to perform that allocation.
245 *
246 * So ip_route_connect() looks up a route using wildcarded source and
247 * destination ports in the key, simply so that we can get a pair of
248 * addresses to use for port allocation.
249 *
250 * Later, once the ports are allocated, ip_route_newports() will make
251 * another route lookup if needed to make sure we catch any IPSEC
252 * rules keyed on the port information.
253 *
254 * The callers allocate the flow key on their stack, and must pass in
255 * the same flowi4 object to both the ip_route_connect() and the
256 * ip_route_newports() calls.
257 */
KOVACS Krisztian79876872008-10-01 07:35:39 -0700258
David S. Miller2d7192d2011-04-26 13:28:44 -0700259static inline void ip_route_connect_init(struct flowi4 *fl4, __be32 dst, __be32 src,
260 u32 tos, int oif, u8 protocol,
261 __be16 sport, __be16 dport,
Steffen Klassert0e0d44a2013-08-28 08:04:14 +0200262 struct sock *sk)
David S. Miller2d7192d2011-04-26 13:28:44 -0700263{
264 __u8 flow_flags = 0;
265
KOVACS Krisztian79876872008-10-01 07:35:39 -0700266 if (inet_sk(sk)->transparent)
David S. Miller94b92b82011-03-31 04:52:59 -0700267 flow_flags |= FLOWI_FLAG_ANYSRC;
David S. Miller94b92b82011-03-31 04:52:59 -0700268
David S. Miller2d7192d2011-04-26 13:28:44 -0700269 flowi4_init_output(fl4, oif, sk->sk_mark, tos, RT_SCOPE_UNIVERSE,
David S. Miller94b92b82011-03-31 04:52:59 -0700270 protocol, flow_flags, dst, src, dport, sport);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271}
272
David S. Miller2d7192d2011-04-26 13:28:44 -0700273static inline struct rtable *ip_route_connect(struct flowi4 *fl4,
274 __be32 dst, __be32 src, u32 tos,
275 int oif, u8 protocol,
276 __be16 sport, __be16 dport,
Steffen Klassert0e0d44a2013-08-28 08:04:14 +0200277 struct sock *sk)
David S. Miller2d7192d2011-04-26 13:28:44 -0700278{
279 struct net *net = sock_net(sk);
280 struct rtable *rt;
281
282 ip_route_connect_init(fl4, dst, src, tos, oif, protocol,
Steffen Klassert0e0d44a2013-08-28 08:04:14 +0200283 sport, dport, sk);
David S. Miller2d7192d2011-04-26 13:28:44 -0700284
David Ahern8cbb512c2015-10-05 08:51:26 -0700285 if (!src && oif) {
286 l3mdev_get_saddr(net, oif, fl4);
287 src = fl4->saddr;
288 }
David S. Miller2d7192d2011-04-26 13:28:44 -0700289 if (!dst || !src) {
290 rt = __ip_route_output_key(net, fl4);
291 if (IS_ERR(rt))
292 return rt;
David S. Miller2d7192d2011-04-26 13:28:44 -0700293 ip_rt_put(rt);
Julian Anastasove6b45242012-02-04 13:04:46 +0000294 flowi4_update_output(fl4, oif, tos, fl4->daddr, fl4->saddr);
David S. Miller2d7192d2011-04-26 13:28:44 -0700295 }
296 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
297 return ip_route_output_flow(net, fl4, sk);
298}
299
300static inline struct rtable *ip_route_newports(struct flowi4 *fl4, struct rtable *rt,
301 __be16 orig_sport, __be16 orig_dport,
302 __be16 sport, __be16 dport,
303 struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304{
David S. Millerdca8b082011-02-24 13:38:12 -0800305 if (sport != orig_sport || dport != orig_dport) {
David S. Miller2d7192d2011-04-26 13:28:44 -0700306 fl4->fl4_dport = dport;
307 fl4->fl4_sport = sport;
David S. Millerb23dd4f2011-03-02 14:31:35 -0800308 ip_rt_put(rt);
Julian Anastasove6b45242012-02-04 13:04:46 +0000309 flowi4_update_output(fl4, sk->sk_bound_dev_if,
310 RT_CONN_FLAGS(sk), fl4->daddr,
311 fl4->saddr);
David S. Miller2d7192d2011-04-26 13:28:44 -0700312 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
313 return ip_route_output_flow(sock_net(sk), fl4, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 }
David S. Millerb23dd4f2011-03-02 14:31:35 -0800315 return rt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316}
317
KOVACS Krisztian1668e012008-10-01 07:33:10 -0700318static inline int inet_iif(const struct sk_buff *skb)
319{
David S. Miller13378ca2012-07-23 13:57:45 -0700320 int iif = skb_rtable(skb)->rt_iif;
321
322 if (iif)
323 return iif;
324 return skb->skb_iif;
KOVACS Krisztian1668e012008-10-01 07:33:10 -0700325}
326
David S. Miller323e1262010-12-12 21:55:08 -0800327extern int sysctl_ip_default_ttl;
328
329static inline int ip4_dst_hoplimit(const struct dst_entry *dst)
330{
331 int hoplimit = dst_metric_raw(dst, RTAX_HOPLIMIT);
332
333 if (hoplimit == 0)
334 hoplimit = sysctl_ip_default_ttl;
335 return hoplimit;
336}
337
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338#endif /* _ROUTE_H */