blob: e5b7182fa099e683ee0484b3772c9aa58eba0d2f [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 * IPv4 Forwarding Information Base: semantics.
7 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
14 */
15
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/bitops.h>
18#include <linux/types.h>
19#include <linux/kernel.h>
20#include <linux/jiffies.h>
21#include <linux/mm.h>
22#include <linux/string.h>
23#include <linux/socket.h>
24#include <linux/sockios.h>
25#include <linux/errno.h>
26#include <linux/in.h>
27#include <linux/inet.h>
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020028#include <linux/inetdevice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/netdevice.h>
30#include <linux/if_arp.h>
31#include <linux/proc_fs.h>
32#include <linux/skbuff.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <linux/init.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090034#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020036#include <net/arp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <net/ip.h>
38#include <net/protocol.h>
39#include <net/route.h>
40#include <net/tcp.h>
41#include <net/sock.h>
42#include <net/ip_fib.h>
Thomas Graff21c7bc2006-08-15 00:34:17 -070043#include <net/netlink.h>
Thomas Graf4e902c52006-08-17 18:14:52 -070044#include <net/nexthop.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
46#include "fib_lookup.h"
47
Stephen Hemminger832b4c52006-08-29 16:48:09 -070048static DEFINE_SPINLOCK(fib_info_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070049static struct hlist_head *fib_info_hash;
50static struct hlist_head *fib_info_laddrhash;
David S. Miller123b9732011-02-01 15:34:21 -080051static unsigned int fib_info_hash_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052static unsigned int fib_info_cnt;
53
54#define DEVINDEX_HASHBITS 8
55#define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
56static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
57
58#ifdef CONFIG_IP_ROUTE_MULTIPATH
59
60static DEFINE_SPINLOCK(fib_multipath_lock);
61
Eric Dumazet6a31d2a2010-10-04 20:00:18 +000062#define for_nexthops(fi) { \
63 int nhsel; const struct fib_nh *nh; \
64 for (nhsel = 0, nh = (fi)->fib_nh; \
65 nhsel < (fi)->fib_nhs; \
66 nh++, nhsel++)
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
Eric Dumazet6a31d2a2010-10-04 20:00:18 +000068#define change_nexthops(fi) { \
69 int nhsel; struct fib_nh *nexthop_nh; \
70 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
71 nhsel < (fi)->fib_nhs; \
72 nexthop_nh++, nhsel++)
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
74#else /* CONFIG_IP_ROUTE_MULTIPATH */
75
76/* Hope, that gcc will optimize it to get rid of dummy loop */
77
Eric Dumazet6a31d2a2010-10-04 20:00:18 +000078#define for_nexthops(fi) { \
79 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \
80 for (nhsel = 0; nhsel < 1; nhsel++)
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
Eric Dumazet6a31d2a2010-10-04 20:00:18 +000082#define change_nexthops(fi) { \
83 int nhsel; \
84 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
85 for (nhsel = 0; nhsel < 1; nhsel++)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
87#endif /* CONFIG_IP_ROUTE_MULTIPATH */
88
89#define endfor_nexthops(fi) }
90
91
David S. Miller3be06862011-03-07 15:01:10 -080092const struct fib_prop fib_props[RTN_MAX + 1] = {
Eric Dumazet6a31d2a2010-10-04 20:00:18 +000093 [RTN_UNSPEC] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 .error = 0,
95 .scope = RT_SCOPE_NOWHERE,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +000096 },
97 [RTN_UNICAST] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 .error = 0,
99 .scope = RT_SCOPE_UNIVERSE,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000100 },
101 [RTN_LOCAL] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 .error = 0,
103 .scope = RT_SCOPE_HOST,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000104 },
105 [RTN_BROADCAST] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 .error = 0,
107 .scope = RT_SCOPE_LINK,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000108 },
109 [RTN_ANYCAST] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 .error = 0,
111 .scope = RT_SCOPE_LINK,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000112 },
113 [RTN_MULTICAST] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 .error = 0,
115 .scope = RT_SCOPE_UNIVERSE,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000116 },
117 [RTN_BLACKHOLE] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 .error = -EINVAL,
119 .scope = RT_SCOPE_UNIVERSE,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000120 },
121 [RTN_UNREACHABLE] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 .error = -EHOSTUNREACH,
123 .scope = RT_SCOPE_UNIVERSE,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000124 },
125 [RTN_PROHIBIT] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 .error = -EACCES,
127 .scope = RT_SCOPE_UNIVERSE,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000128 },
129 [RTN_THROW] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 .error = -EAGAIN,
131 .scope = RT_SCOPE_UNIVERSE,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000132 },
133 [RTN_NAT] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 .error = -EINVAL,
135 .scope = RT_SCOPE_NOWHERE,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000136 },
137 [RTN_XRESOLVE] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 .error = -EINVAL,
139 .scope = RT_SCOPE_NOWHERE,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000140 },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141};
142
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143/* Release a nexthop info record */
Yan, Zheng19c1ea12011-09-04 20:24:20 +0000144static void free_fib_info_rcu(struct rcu_head *head)
145{
146 struct fib_info *fi = container_of(head, struct fib_info, rcu);
147
Yanmin Zhange49cc0d2012-05-23 15:39:45 +0000148 change_nexthops(fi) {
149 if (nexthop_nh->nh_dev)
150 dev_put(nexthop_nh->nh_dev);
151 } endfor_nexthops(fi);
152
153 release_net(fi->fib_net);
Yan, Zheng19c1ea12011-09-04 20:24:20 +0000154 if (fi->fib_metrics != (u32 *) dst_default_metrics)
155 kfree(fi->fib_metrics);
156 kfree(fi);
157}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
159void free_fib_info(struct fib_info *fi)
160{
161 if (fi->fib_dead == 0) {
Joe Perches058bd4d2012-03-11 18:36:11 +0000162 pr_warn("Freeing alive fib_info %p\n", fi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 return;
164 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 fib_info_cnt--;
Yan, Zheng19c1ea12011-09-04 20:24:20 +0000166 call_rcu(&fi->rcu, free_fib_info_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167}
168
169void fib_release_info(struct fib_info *fi)
170{
Stephen Hemminger832b4c52006-08-29 16:48:09 -0700171 spin_lock_bh(&fib_info_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 if (fi && --fi->fib_treeref == 0) {
173 hlist_del(&fi->fib_hash);
174 if (fi->fib_prefsrc)
175 hlist_del(&fi->fib_lhash);
176 change_nexthops(fi) {
David S. Miller71fceff2010-01-15 01:16:40 -0800177 if (!nexthop_nh->nh_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 continue;
David S. Miller71fceff2010-01-15 01:16:40 -0800179 hlist_del(&nexthop_nh->nh_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 } endfor_nexthops(fi)
181 fi->fib_dead = 1;
182 fib_info_put(fi);
183 }
Stephen Hemminger832b4c52006-08-29 16:48:09 -0700184 spin_unlock_bh(&fib_info_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185}
186
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000187static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188{
189 const struct fib_nh *onh = ofi->fib_nh;
190
191 for_nexthops(fi) {
192 if (nh->nh_oif != onh->nh_oif ||
193 nh->nh_gw != onh->nh_gw ||
194 nh->nh_scope != onh->nh_scope ||
195#ifdef CONFIG_IP_ROUTE_MULTIPATH
196 nh->nh_weight != onh->nh_weight ||
197#endif
Patrick McHardyc7066f72011-01-14 13:36:42 +0100198#ifdef CONFIG_IP_ROUTE_CLASSID
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 nh->nh_tclassid != onh->nh_tclassid ||
200#endif
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000201 ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_F_DEAD))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 return -1;
203 onh++;
204 } endfor_nexthops(fi);
205 return 0;
206}
207
David S. Miller88ebc722008-01-12 21:49:01 -0800208static inline unsigned int fib_devindex_hashfn(unsigned int val)
209{
210 unsigned int mask = DEVINDEX_HASHSIZE - 1;
211
212 return (val ^
213 (val >> DEVINDEX_HASHBITS) ^
214 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
215}
216
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
218{
David S. Miller123b9732011-02-01 15:34:21 -0800219 unsigned int mask = (fib_info_hash_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 unsigned int val = fi->fib_nhs;
221
David S. Miller37e826c2011-03-24 18:06:47 -0700222 val ^= (fi->fib_protocol << 8) | fi->fib_scope;
Al Viro81f7bf62006-09-27 18:40:00 -0700223 val ^= (__force u32)fi->fib_prefsrc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 val ^= fi->fib_priority;
David S. Miller88ebc722008-01-12 21:49:01 -0800225 for_nexthops(fi) {
226 val ^= fib_devindex_hashfn(nh->nh_oif);
227 } endfor_nexthops(fi)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
229 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
230}
231
232static struct fib_info *fib_find_info(const struct fib_info *nfi)
233{
234 struct hlist_head *head;
235 struct hlist_node *node;
236 struct fib_info *fi;
237 unsigned int hash;
238
239 hash = fib_info_hashfn(nfi);
240 head = &fib_info_hash[hash];
241
242 hlist_for_each_entry(fi, node, head, fib_hash) {
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800243 if (!net_eq(fi->fib_net, nfi->fib_net))
Denis V. Lunev4814bdb2008-01-31 18:50:07 -0800244 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 if (fi->fib_nhs != nfi->fib_nhs)
246 continue;
247 if (nfi->fib_protocol == fi->fib_protocol &&
David S. Miller37e826c2011-03-24 18:06:47 -0700248 nfi->fib_scope == fi->fib_scope &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 nfi->fib_prefsrc == fi->fib_prefsrc &&
250 nfi->fib_priority == fi->fib_priority &&
251 memcmp(nfi->fib_metrics, fi->fib_metrics,
Eric Dumazetfcd13f42011-03-24 07:01:24 +0000252 sizeof(u32) * RTAX_MAX) == 0 &&
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000253 ((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_F_DEAD) == 0 &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
255 return fi;
256 }
257
258 return NULL;
259}
260
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261/* Check, that the gateway is already configured.
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000262 * Used only by redirect accept routine.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 */
Al Virod878e72e2006-09-26 22:18:13 -0700264int ip_fib_check_default(__be32 gw, struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265{
266 struct hlist_head *head;
267 struct hlist_node *node;
268 struct fib_nh *nh;
269 unsigned int hash;
270
Stephen Hemminger832b4c52006-08-29 16:48:09 -0700271 spin_lock(&fib_info_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
273 hash = fib_devindex_hashfn(dev->ifindex);
274 head = &fib_info_devhash[hash];
275 hlist_for_each_entry(nh, node, head, nh_hash) {
276 if (nh->nh_dev == dev &&
277 nh->nh_gw == gw &&
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000278 !(nh->nh_flags & RTNH_F_DEAD)) {
Stephen Hemminger832b4c52006-08-29 16:48:09 -0700279 spin_unlock(&fib_info_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 return 0;
281 }
282 }
283
Stephen Hemminger832b4c52006-08-29 16:48:09 -0700284 spin_unlock(&fib_info_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285
286 return -1;
287}
288
Thomas Graf339bf982006-11-10 14:10:15 -0800289static inline size_t fib_nlmsg_size(struct fib_info *fi)
290{
291 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
292 + nla_total_size(4) /* RTA_TABLE */
293 + nla_total_size(4) /* RTA_DST */
294 + nla_total_size(4) /* RTA_PRIORITY */
295 + nla_total_size(4); /* RTA_PREFSRC */
296
297 /* space for nested metrics */
298 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
299
300 if (fi->fib_nhs) {
301 /* Also handles the special case fib_nhs == 1 */
302
303 /* each nexthop is packed in an attribute */
304 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
305
306 /* may contain flow and gateway attribute */
307 nhsize += 2 * nla_total_size(4);
308
309 /* all nexthops are packed in a nested attribute */
310 payload += nla_total_size(fi->fib_nhs * nhsize);
311 }
312
313 return payload;
314}
315
Al Viro81f7bf62006-09-27 18:40:00 -0700316void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
Milan Kocianb8f55832007-05-23 14:55:06 -0700317 int dst_len, u32 tb_id, struct nl_info *info,
318 unsigned int nlm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
320 struct sk_buff *skb;
Thomas Graf4e902c52006-08-17 18:14:52 -0700321 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
Thomas Graff21c7bc2006-08-15 00:34:17 -0700322 int err = -ENOBUFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
Thomas Graf339bf982006-11-10 14:10:15 -0800324 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
Thomas Graff21c7bc2006-08-15 00:34:17 -0700325 if (skb == NULL)
326 goto errout;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327
Thomas Graf4e902c52006-08-17 18:14:52 -0700328 err = fib_dump_info(skb, info->pid, seq, event, tb_id,
David S. Miller37e826c2011-03-24 18:06:47 -0700329 fa->fa_type, key, dst_len,
Milan Kocianb8f55832007-05-23 14:55:06 -0700330 fa->fa_tos, fa->fa_info, nlm_flags);
Patrick McHardy26932562007-01-31 23:16:40 -0800331 if (err < 0) {
332 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
333 WARN_ON(err == -EMSGSIZE);
334 kfree_skb(skb);
335 goto errout;
336 }
Pablo Neira Ayuso1ce85fe2009-02-24 23:18:28 -0800337 rtnl_notify(skb, info->nl_net, info->pid, RTNLGRP_IPV4_ROUTE,
338 info->nlh, GFP_KERNEL);
339 return;
Thomas Graff21c7bc2006-08-15 00:34:17 -0700340errout:
341 if (err < 0)
Denis V. Lunev4d1169c2008-01-10 03:26:13 -0800342 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343}
344
345/* Return the first fib alias matching TOS with
346 * priority less than or equal to PRIO.
347 */
348struct fib_alias *fib_find_alias(struct list_head *fah, u8 tos, u32 prio)
349{
350 if (fah) {
351 struct fib_alias *fa;
352 list_for_each_entry(fa, fah, fa_list) {
353 if (fa->fa_tos > tos)
354 continue;
355 if (fa->fa_info->fib_priority >= prio ||
356 fa->fa_tos < tos)
357 return fa;
358 }
359 }
360 return NULL;
361}
362
363int fib_detect_death(struct fib_info *fi, int order,
Denis V. Lunevc17860a2007-12-08 00:22:13 -0800364 struct fib_info **last_resort, int *last_idx, int dflt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365{
366 struct neighbour *n;
367 int state = NUD_NONE;
368
369 n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
370 if (n) {
371 state = n->nud_state;
372 neigh_release(n);
373 }
Jianjun Kongd93191002008-11-03 00:23:42 -0800374 if (state == NUD_REACHABLE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 return 0;
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000376 if ((state & NUD_VALID) && order != dflt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 return 0;
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000378 if ((state & NUD_VALID) ||
379 (*last_idx < 0 && order > dflt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 *last_resort = fi;
381 *last_idx = order;
382 }
383 return 1;
384}
385
386#ifdef CONFIG_IP_ROUTE_MULTIPATH
387
Thomas Graf4e902c52006-08-17 18:14:52 -0700388static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389{
390 int nhs = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
Thomas Graf4e902c52006-08-17 18:14:52 -0700392 while (rtnh_ok(rtnh, remaining)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 nhs++;
Thomas Graf4e902c52006-08-17 18:14:52 -0700394 rtnh = rtnh_next(rtnh, &remaining);
395 }
396
397 /* leftover implies invalid nexthop configuration, discard it */
398 return remaining > 0 ? 0 : nhs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399}
400
Thomas Graf4e902c52006-08-17 18:14:52 -0700401static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
402 int remaining, struct fib_config *cfg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 change_nexthops(fi) {
Thomas Graf4e902c52006-08-17 18:14:52 -0700405 int attrlen;
406
407 if (!rtnh_ok(rtnh, remaining))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 return -EINVAL;
Thomas Graf4e902c52006-08-17 18:14:52 -0700409
David S. Miller71fceff2010-01-15 01:16:40 -0800410 nexthop_nh->nh_flags =
411 (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
412 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
413 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
Thomas Graf4e902c52006-08-17 18:14:52 -0700414
415 attrlen = rtnh_attrlen(rtnh);
416 if (attrlen > 0) {
417 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
418
419 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
David S. Miller71fceff2010-01-15 01:16:40 -0800420 nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0;
Patrick McHardyc7066f72011-01-14 13:36:42 +0100421#ifdef CONFIG_IP_ROUTE_CLASSID
Thomas Graf4e902c52006-08-17 18:14:52 -0700422 nla = nla_find(attrs, attrlen, RTA_FLOW);
David S. Miller71fceff2010-01-15 01:16:40 -0800423 nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424#endif
425 }
Thomas Graf4e902c52006-08-17 18:14:52 -0700426
427 rtnh = rtnh_next(rtnh, &remaining);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 } endfor_nexthops(fi);
Thomas Graf4e902c52006-08-17 18:14:52 -0700429
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 return 0;
431}
432
433#endif
434
Thomas Graf4e902c52006-08-17 18:14:52 -0700435int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436{
437#ifdef CONFIG_IP_ROUTE_MULTIPATH
Thomas Graf4e902c52006-08-17 18:14:52 -0700438 struct rtnexthop *rtnh;
439 int remaining;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440#endif
441
Thomas Graf4e902c52006-08-17 18:14:52 -0700442 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 return 1;
444
Thomas Graf4e902c52006-08-17 18:14:52 -0700445 if (cfg->fc_oif || cfg->fc_gw) {
446 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
447 (!cfg->fc_gw || cfg->fc_gw == fi->fib_nh->nh_gw))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 return 0;
449 return 1;
450 }
451
452#ifdef CONFIG_IP_ROUTE_MULTIPATH
Thomas Graf4e902c52006-08-17 18:14:52 -0700453 if (cfg->fc_mp == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 return 0;
Thomas Graf4e902c52006-08-17 18:14:52 -0700455
456 rtnh = cfg->fc_mp;
457 remaining = cfg->fc_mp_len;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900458
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 for_nexthops(fi) {
Thomas Graf4e902c52006-08-17 18:14:52 -0700460 int attrlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
Thomas Graf4e902c52006-08-17 18:14:52 -0700462 if (!rtnh_ok(rtnh, remaining))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 return -EINVAL;
Thomas Graf4e902c52006-08-17 18:14:52 -0700464
465 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 return 1;
Thomas Graf4e902c52006-08-17 18:14:52 -0700467
468 attrlen = rtnh_attrlen(rtnh);
469 if (attrlen < 0) {
470 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
471
472 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
Al Viro17fb2c62006-09-26 22:15:25 -0700473 if (nla && nla_get_be32(nla) != nh->nh_gw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 return 1;
Patrick McHardyc7066f72011-01-14 13:36:42 +0100475#ifdef CONFIG_IP_ROUTE_CLASSID
Thomas Graf4e902c52006-08-17 18:14:52 -0700476 nla = nla_find(attrs, attrlen, RTA_FLOW);
477 if (nla && nla_get_u32(nla) != nh->nh_tclassid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 return 1;
479#endif
480 }
Thomas Graf4e902c52006-08-17 18:14:52 -0700481
482 rtnh = rtnh_next(rtnh, &remaining);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 } endfor_nexthops(fi);
484#endif
485 return 0;
486}
487
488
489/*
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000490 * Picture
491 * -------
492 *
493 * Semantics of nexthop is very messy by historical reasons.
494 * We have to take into account, that:
495 * a) gateway can be actually local interface address,
496 * so that gatewayed route is direct.
497 * b) gateway must be on-link address, possibly
498 * described not by an ifaddr, but also by a direct route.
499 * c) If both gateway and interface are specified, they should not
500 * contradict.
501 * d) If we use tunnel routes, gateway could be not on-link.
502 *
503 * Attempt to reconcile all of these (alas, self-contradictory) conditions
504 * results in pretty ugly and hairy code with obscure logic.
505 *
506 * I chose to generalized it instead, so that the size
507 * of code does not increase practically, but it becomes
508 * much more general.
509 * Every prefix is assigned a "scope" value: "host" is local address,
510 * "link" is direct route,
511 * [ ... "site" ... "interior" ... ]
512 * and "universe" is true gateway route with global meaning.
513 *
514 * Every prefix refers to a set of "nexthop"s (gw, oif),
515 * where gw must have narrower scope. This recursion stops
516 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
517 * which means that gw is forced to be on link.
518 *
519 * Code is still hairy, but now it is apparently logically
520 * consistent and very flexible. F.e. as by-product it allows
521 * to co-exists in peace independent exterior and interior
522 * routing processes.
523 *
524 * Normally it looks as following.
525 *
526 * {universe prefix} -> (gw, oif) [scope link]
527 * |
528 * |-> {link prefix} -> (gw, oif) [scope local]
529 * |
530 * |-> {local prefix} (terminal node)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 */
Thomas Graf4e902c52006-08-17 18:14:52 -0700532static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
533 struct fib_nh *nh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534{
535 int err;
Denis V. Lunev86167a32008-01-21 17:34:00 -0800536 struct net *net;
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000537 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538
Denis V. Lunev86167a32008-01-21 17:34:00 -0800539 net = cfg->fc_nlinfo.nl_net;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 if (nh->nh_gw) {
541 struct fib_result res;
542
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000543 if (nh->nh_flags & RTNH_F_ONLINK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544
Thomas Graf4e902c52006-08-17 18:14:52 -0700545 if (cfg->fc_scope >= RT_SCOPE_LINK)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 return -EINVAL;
Denis V. Lunev86167a32008-01-21 17:34:00 -0800547 if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 return -EINVAL;
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000549 dev = __dev_get_by_index(net, nh->nh_oif);
550 if (!dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 return -ENODEV;
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000552 if (!(dev->flags & IFF_UP))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 return -ENETDOWN;
554 nh->nh_dev = dev;
555 dev_hold(dev);
556 nh->nh_scope = RT_SCOPE_LINK;
557 return 0;
558 }
Eric Dumazetebc0ffa2010-10-05 10:41:36 +0000559 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 {
David S. Miller9ade2282011-03-12 02:02:42 -0500561 struct flowi4 fl4 = {
562 .daddr = nh->nh_gw,
563 .flowi4_scope = cfg->fc_scope + 1,
564 .flowi4_oif = nh->nh_oif,
Thomas Graf4e902c52006-08-17 18:14:52 -0700565 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
567 /* It is not necessary, but requires a bit of thinking */
David S. Miller9ade2282011-03-12 02:02:42 -0500568 if (fl4.flowi4_scope < RT_SCOPE_LINK)
569 fl4.flowi4_scope = RT_SCOPE_LINK;
570 err = fib_lookup(net, &fl4, &res);
Eric Dumazetebc0ffa2010-10-05 10:41:36 +0000571 if (err) {
572 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 return err;
Eric Dumazetebc0ffa2010-10-05 10:41:36 +0000574 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 }
576 err = -EINVAL;
577 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
578 goto out;
579 nh->nh_scope = res.scope;
580 nh->nh_oif = FIB_RES_OIF(res);
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000581 nh->nh_dev = dev = FIB_RES_DEV(res);
582 if (!dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 goto out;
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000584 dev_hold(dev);
Eric Dumazet8723e1b2010-10-19 00:39:26 +0000585 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 } else {
587 struct in_device *in_dev;
588
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000589 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 return -EINVAL;
591
Eric Dumazet8723e1b2010-10-19 00:39:26 +0000592 rcu_read_lock();
593 err = -ENODEV;
Denis V. Lunev86167a32008-01-21 17:34:00 -0800594 in_dev = inetdev_by_index(net, nh->nh_oif);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 if (in_dev == NULL)
Eric Dumazet8723e1b2010-10-19 00:39:26 +0000596 goto out;
597 err = -ENETDOWN;
598 if (!(in_dev->dev->flags & IFF_UP))
599 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 nh->nh_dev = in_dev->dev;
601 dev_hold(nh->nh_dev);
602 nh->nh_scope = RT_SCOPE_HOST;
Eric Dumazet8723e1b2010-10-19 00:39:26 +0000603 err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 }
Eric Dumazet8723e1b2010-10-19 00:39:26 +0000605out:
606 rcu_read_unlock();
607 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608}
609
Al Viro81f7bf62006-09-27 18:40:00 -0700610static inline unsigned int fib_laddr_hashfn(__be32 val)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611{
David S. Miller123b9732011-02-01 15:34:21 -0800612 unsigned int mask = (fib_info_hash_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000614 return ((__force u32)val ^
615 ((__force u32)val >> 7) ^
616 ((__force u32)val >> 14)) & mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617}
618
David S. Miller123b9732011-02-01 15:34:21 -0800619static struct hlist_head *fib_info_hash_alloc(int bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620{
621 if (bytes <= PAGE_SIZE)
Joonwoo Park88f83492007-11-26 23:29:32 +0800622 return kzalloc(bytes, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 else
624 return (struct hlist_head *)
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000625 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
626 get_order(bytes));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627}
628
David S. Miller123b9732011-02-01 15:34:21 -0800629static void fib_info_hash_free(struct hlist_head *hash, int bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630{
631 if (!hash)
632 return;
633
634 if (bytes <= PAGE_SIZE)
635 kfree(hash);
636 else
637 free_pages((unsigned long) hash, get_order(bytes));
638}
639
David S. Miller123b9732011-02-01 15:34:21 -0800640static void fib_info_hash_move(struct hlist_head *new_info_hash,
641 struct hlist_head *new_laddrhash,
642 unsigned int new_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643{
David S. Millerb7656e72005-08-05 04:12:48 -0700644 struct hlist_head *old_info_hash, *old_laddrhash;
David S. Miller123b9732011-02-01 15:34:21 -0800645 unsigned int old_size = fib_info_hash_size;
David S. Millerb7656e72005-08-05 04:12:48 -0700646 unsigned int i, bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647
Stephen Hemminger832b4c52006-08-29 16:48:09 -0700648 spin_lock_bh(&fib_info_lock);
David S. Millerb7656e72005-08-05 04:12:48 -0700649 old_info_hash = fib_info_hash;
650 old_laddrhash = fib_info_laddrhash;
David S. Miller123b9732011-02-01 15:34:21 -0800651 fib_info_hash_size = new_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
653 for (i = 0; i < old_size; i++) {
654 struct hlist_head *head = &fib_info_hash[i];
655 struct hlist_node *node, *n;
656 struct fib_info *fi;
657
658 hlist_for_each_entry_safe(fi, node, n, head, fib_hash) {
659 struct hlist_head *dest;
660 unsigned int new_hash;
661
662 hlist_del(&fi->fib_hash);
663
664 new_hash = fib_info_hashfn(fi);
665 dest = &new_info_hash[new_hash];
666 hlist_add_head(&fi->fib_hash, dest);
667 }
668 }
669 fib_info_hash = new_info_hash;
670
671 for (i = 0; i < old_size; i++) {
672 struct hlist_head *lhead = &fib_info_laddrhash[i];
673 struct hlist_node *node, *n;
674 struct fib_info *fi;
675
676 hlist_for_each_entry_safe(fi, node, n, lhead, fib_lhash) {
677 struct hlist_head *ldest;
678 unsigned int new_hash;
679
680 hlist_del(&fi->fib_lhash);
681
682 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
683 ldest = &new_laddrhash[new_hash];
684 hlist_add_head(&fi->fib_lhash, ldest);
685 }
686 }
687 fib_info_laddrhash = new_laddrhash;
688
Stephen Hemminger832b4c52006-08-29 16:48:09 -0700689 spin_unlock_bh(&fib_info_lock);
David S. Millerb7656e72005-08-05 04:12:48 -0700690
691 bytes = old_size * sizeof(struct hlist_head *);
David S. Miller123b9732011-02-01 15:34:21 -0800692 fib_info_hash_free(old_info_hash, bytes);
693 fib_info_hash_free(old_laddrhash, bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694}
695
David S. Miller436c3b62011-03-24 17:42:21 -0700696__be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
697{
698 nh->nh_saddr = inet_select_addr(nh->nh_dev,
699 nh->nh_gw,
David S. Miller37e826c2011-03-24 18:06:47 -0700700 nh->nh_parent->fib_scope);
David S. Miller436c3b62011-03-24 17:42:21 -0700701 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
702
703 return nh->nh_saddr;
704}
705
Thomas Graf4e902c52006-08-17 18:14:52 -0700706struct fib_info *fib_create_info(struct fib_config *cfg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707{
708 int err;
709 struct fib_info *fi = NULL;
710 struct fib_info *ofi;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 int nhs = 1;
Denis V. Lunev7462bd742008-01-31 18:49:32 -0800712 struct net *net = cfg->fc_nlinfo.nl_net;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
David S. Miller4c8237c2011-03-07 14:27:38 -0800714 if (cfg->fc_type > RTN_MAX)
715 goto err_inval;
716
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 /* Fast check to catch the most weird cases */
Thomas Graf4e902c52006-08-17 18:14:52 -0700718 if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 goto err_inval;
720
721#ifdef CONFIG_IP_ROUTE_MULTIPATH
Thomas Graf4e902c52006-08-17 18:14:52 -0700722 if (cfg->fc_mp) {
723 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 if (nhs == 0)
725 goto err_inval;
726 }
727#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
729 err = -ENOBUFS;
David S. Miller123b9732011-02-01 15:34:21 -0800730 if (fib_info_cnt >= fib_info_hash_size) {
731 unsigned int new_size = fib_info_hash_size << 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 struct hlist_head *new_info_hash;
733 struct hlist_head *new_laddrhash;
734 unsigned int bytes;
735
736 if (!new_size)
737 new_size = 1;
738 bytes = new_size * sizeof(struct hlist_head *);
David S. Miller123b9732011-02-01 15:34:21 -0800739 new_info_hash = fib_info_hash_alloc(bytes);
740 new_laddrhash = fib_info_hash_alloc(bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 if (!new_info_hash || !new_laddrhash) {
David S. Miller123b9732011-02-01 15:34:21 -0800742 fib_info_hash_free(new_info_hash, bytes);
743 fib_info_hash_free(new_laddrhash, bytes);
Joonwoo Park88f83492007-11-26 23:29:32 +0800744 } else
David S. Miller123b9732011-02-01 15:34:21 -0800745 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746
David S. Miller123b9732011-02-01 15:34:21 -0800747 if (!fib_info_hash_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 goto failure;
749 }
750
Panagiotis Issaris0da974f2006-07-21 14:51:30 -0700751 fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 if (fi == NULL)
753 goto failure;
David S. Miller725d1e12011-01-28 14:05:05 -0800754 if (cfg->fc_mx) {
755 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
756 if (!fi->fib_metrics)
757 goto failure;
758 } else
759 fi->fib_metrics = (u32 *) dst_default_metrics;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 fib_info_cnt++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761
Denis V. Lunev57d7a602008-04-16 02:00:50 -0700762 fi->fib_net = hold_net(net);
Thomas Graf4e902c52006-08-17 18:14:52 -0700763 fi->fib_protocol = cfg->fc_protocol;
David S. Miller37e826c2011-03-24 18:06:47 -0700764 fi->fib_scope = cfg->fc_scope;
Thomas Graf4e902c52006-08-17 18:14:52 -0700765 fi->fib_flags = cfg->fc_flags;
766 fi->fib_priority = cfg->fc_priority;
767 fi->fib_prefsrc = cfg->fc_prefsrc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768
769 fi->fib_nhs = nhs;
770 change_nexthops(fi) {
David S. Miller71fceff2010-01-15 01:16:40 -0800771 nexthop_nh->nh_parent = fi;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 } endfor_nexthops(fi)
773
Thomas Graf4e902c52006-08-17 18:14:52 -0700774 if (cfg->fc_mx) {
775 struct nlattr *nla;
776 int remaining;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
Thomas Graf4e902c52006-08-17 18:14:52 -0700778 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
Thomas Graf8f4c1f92007-09-12 14:44:36 +0200779 int type = nla_type(nla);
Thomas Graf4e902c52006-08-17 18:14:52 -0700780
781 if (type) {
782 if (type > RTAX_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 goto err_inval;
Thomas Graf4e902c52006-08-17 18:14:52 -0700784 fi->fib_metrics[type - 1] = nla_get_u32(nla);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 }
787 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788
Thomas Graf4e902c52006-08-17 18:14:52 -0700789 if (cfg->fc_mp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790#ifdef CONFIG_IP_ROUTE_MULTIPATH
Thomas Graf4e902c52006-08-17 18:14:52 -0700791 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
792 if (err != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 goto failure;
Thomas Graf4e902c52006-08-17 18:14:52 -0700794 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 goto err_inval;
Thomas Graf4e902c52006-08-17 18:14:52 -0700796 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 goto err_inval;
Patrick McHardyc7066f72011-01-14 13:36:42 +0100798#ifdef CONFIG_IP_ROUTE_CLASSID
Thomas Graf4e902c52006-08-17 18:14:52 -0700799 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 goto err_inval;
801#endif
802#else
803 goto err_inval;
804#endif
805 } else {
806 struct fib_nh *nh = fi->fib_nh;
Thomas Graf4e902c52006-08-17 18:14:52 -0700807
808 nh->nh_oif = cfg->fc_oif;
809 nh->nh_gw = cfg->fc_gw;
810 nh->nh_flags = cfg->fc_flags;
Patrick McHardyc7066f72011-01-14 13:36:42 +0100811#ifdef CONFIG_IP_ROUTE_CLASSID
Thomas Graf4e902c52006-08-17 18:14:52 -0700812 nh->nh_tclassid = cfg->fc_flow;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814#ifdef CONFIG_IP_ROUTE_MULTIPATH
815 nh->nh_weight = 1;
816#endif
817 }
818
Thomas Graf4e902c52006-08-17 18:14:52 -0700819 if (fib_props[cfg->fc_type].error) {
820 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 goto err_inval;
822 goto link_it;
David S. Miller4c8237c2011-03-07 14:27:38 -0800823 } else {
824 switch (cfg->fc_type) {
825 case RTN_UNICAST:
826 case RTN_LOCAL:
827 case RTN_BROADCAST:
828 case RTN_ANYCAST:
829 case RTN_MULTICAST:
830 break;
831 default:
832 goto err_inval;
833 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 }
835
Thomas Graf4e902c52006-08-17 18:14:52 -0700836 if (cfg->fc_scope > RT_SCOPE_HOST)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 goto err_inval;
838
Thomas Graf4e902c52006-08-17 18:14:52 -0700839 if (cfg->fc_scope == RT_SCOPE_HOST) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 struct fib_nh *nh = fi->fib_nh;
841
842 /* Local address is added. */
843 if (nhs != 1 || nh->nh_gw)
844 goto err_inval;
845 nh->nh_scope = RT_SCOPE_NOWHERE;
Denis V. Lunev7462bd742008-01-31 18:49:32 -0800846 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 err = -ENODEV;
848 if (nh->nh_dev == NULL)
849 goto failure;
850 } else {
851 change_nexthops(fi) {
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000852 err = fib_check_nh(cfg, fi, nexthop_nh);
853 if (err != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 goto failure;
855 } endfor_nexthops(fi)
856 }
857
858 if (fi->fib_prefsrc) {
Thomas Graf4e902c52006-08-17 18:14:52 -0700859 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
860 fi->fib_prefsrc != cfg->fc_dst)
Denis V. Lunev7462bd742008-01-31 18:49:32 -0800861 if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 goto err_inval;
863 }
864
David S. Miller1fc050a2011-03-07 20:54:48 -0800865 change_nexthops(fi) {
David S. Miller436c3b62011-03-24 17:42:21 -0700866 fib_info_update_nh_saddr(net, nexthop_nh);
David S. Miller1fc050a2011-03-07 20:54:48 -0800867 } endfor_nexthops(fi)
868
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869link_it:
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000870 ofi = fib_find_info(fi);
871 if (ofi) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 fi->fib_dead = 1;
873 free_fib_info(fi);
874 ofi->fib_treeref++;
875 return ofi;
876 }
877
878 fi->fib_treeref++;
879 atomic_inc(&fi->fib_clntref);
Stephen Hemminger832b4c52006-08-29 16:48:09 -0700880 spin_lock_bh(&fib_info_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 hlist_add_head(&fi->fib_hash,
882 &fib_info_hash[fib_info_hashfn(fi)]);
883 if (fi->fib_prefsrc) {
884 struct hlist_head *head;
885
886 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
887 hlist_add_head(&fi->fib_lhash, head);
888 }
889 change_nexthops(fi) {
890 struct hlist_head *head;
891 unsigned int hash;
892
David S. Miller71fceff2010-01-15 01:16:40 -0800893 if (!nexthop_nh->nh_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 continue;
David S. Miller71fceff2010-01-15 01:16:40 -0800895 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 head = &fib_info_devhash[hash];
David S. Miller71fceff2010-01-15 01:16:40 -0800897 hlist_add_head(&nexthop_nh->nh_hash, head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 } endfor_nexthops(fi)
Stephen Hemminger832b4c52006-08-29 16:48:09 -0700899 spin_unlock_bh(&fib_info_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 return fi;
901
902err_inval:
903 err = -EINVAL;
904
905failure:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900906 if (fi) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 fi->fib_dead = 1;
908 free_fib_info(fi);
909 }
Thomas Graf4e902c52006-08-17 18:14:52 -0700910
911 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912}
913
Thomas Grafbe403ea2006-08-17 18:15:17 -0700914int fib_dump_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
David S. Miller37e826c2011-03-24 18:06:47 -0700915 u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
Thomas Grafbe403ea2006-08-17 18:15:17 -0700916 struct fib_info *fi, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917{
Thomas Grafbe403ea2006-08-17 18:15:17 -0700918 struct nlmsghdr *nlh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 struct rtmsg *rtm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
Thomas Grafbe403ea2006-08-17 18:15:17 -0700921 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*rtm), flags);
922 if (nlh == NULL)
Patrick McHardy26932562007-01-31 23:16:40 -0800923 return -EMSGSIZE;
Thomas Grafbe403ea2006-08-17 18:15:17 -0700924
925 rtm = nlmsg_data(nlh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 rtm->rtm_family = AF_INET;
927 rtm->rtm_dst_len = dst_len;
928 rtm->rtm_src_len = 0;
929 rtm->rtm_tos = tos;
Krzysztof Piotr Oledzki709772e2008-06-10 15:44:49 -0700930 if (tb_id < 256)
931 rtm->rtm_table = tb_id;
932 else
933 rtm->rtm_table = RT_TABLE_COMPAT;
David S. Millerf3756b72012-04-01 20:39:02 -0400934 if (nla_put_u32(skb, RTA_TABLE, tb_id))
935 goto nla_put_failure;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 rtm->rtm_type = type;
937 rtm->rtm_flags = fi->fib_flags;
David S. Miller37e826c2011-03-24 18:06:47 -0700938 rtm->rtm_scope = fi->fib_scope;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 rtm->rtm_protocol = fi->fib_protocol;
Thomas Grafbe403ea2006-08-17 18:15:17 -0700940
David S. Millerf3756b72012-04-01 20:39:02 -0400941 if (rtm->rtm_dst_len &&
942 nla_put_be32(skb, RTA_DST, dst))
943 goto nla_put_failure;
944 if (fi->fib_priority &&
945 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
946 goto nla_put_failure;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
Thomas Grafbe403ea2006-08-17 18:15:17 -0700948 goto nla_put_failure;
949
David S. Millerf3756b72012-04-01 20:39:02 -0400950 if (fi->fib_prefsrc &&
951 nla_put_be32(skb, RTA_PREFSRC, fi->fib_prefsrc))
952 goto nla_put_failure;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 if (fi->fib_nhs == 1) {
David S. Millerf3756b72012-04-01 20:39:02 -0400954 if (fi->fib_nh->nh_gw &&
955 nla_put_be32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
956 goto nla_put_failure;
957 if (fi->fib_nh->nh_oif &&
958 nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
959 goto nla_put_failure;
Patrick McHardyc7066f72011-01-14 13:36:42 +0100960#ifdef CONFIG_IP_ROUTE_CLASSID
David S. Millerf3756b72012-04-01 20:39:02 -0400961 if (fi->fib_nh[0].nh_tclassid &&
962 nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
963 goto nla_put_failure;
Patrick McHardy8265abc2006-07-21 15:09:55 -0700964#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 }
966#ifdef CONFIG_IP_ROUTE_MULTIPATH
967 if (fi->fib_nhs > 1) {
Thomas Grafbe403ea2006-08-17 18:15:17 -0700968 struct rtnexthop *rtnh;
969 struct nlattr *mp;
970
971 mp = nla_nest_start(skb, RTA_MULTIPATH);
972 if (mp == NULL)
973 goto nla_put_failure;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974
975 for_nexthops(fi) {
Thomas Grafbe403ea2006-08-17 18:15:17 -0700976 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
977 if (rtnh == NULL)
978 goto nla_put_failure;
979
980 rtnh->rtnh_flags = nh->nh_flags & 0xFF;
981 rtnh->rtnh_hops = nh->nh_weight - 1;
982 rtnh->rtnh_ifindex = nh->nh_oif;
983
David S. Millerf3756b72012-04-01 20:39:02 -0400984 if (nh->nh_gw &&
985 nla_put_be32(skb, RTA_GATEWAY, nh->nh_gw))
986 goto nla_put_failure;
Patrick McHardyc7066f72011-01-14 13:36:42 +0100987#ifdef CONFIG_IP_ROUTE_CLASSID
David S. Millerf3756b72012-04-01 20:39:02 -0400988 if (nh->nh_tclassid &&
989 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
990 goto nla_put_failure;
Patrick McHardy8265abc2006-07-21 15:09:55 -0700991#endif
Thomas Grafbe403ea2006-08-17 18:15:17 -0700992 /* length of rtnetlink header + attributes */
993 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 } endfor_nexthops(fi);
Thomas Grafbe403ea2006-08-17 18:15:17 -0700995
996 nla_nest_end(skb, mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 }
998#endif
Thomas Grafbe403ea2006-08-17 18:15:17 -0700999 return nlmsg_end(skb, nlh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
Thomas Grafbe403ea2006-08-17 18:15:17 -07001001nla_put_failure:
Patrick McHardy26932562007-01-31 23:16:40 -08001002 nlmsg_cancel(skb, nlh);
1003 return -EMSGSIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004}
1005
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006/*
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00001007 * Update FIB if:
1008 * - local address disappeared -> we must delete all the entries
1009 * referring to it.
1010 * - device went down -> we must shutdown all nexthops going via it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 */
Denis V. Lunev4814bdb2008-01-31 18:50:07 -08001012int fib_sync_down_addr(struct net *net, __be32 local)
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001013{
1014 int ret = 0;
1015 unsigned int hash = fib_laddr_hashfn(local);
1016 struct hlist_head *head = &fib_info_laddrhash[hash];
1017 struct hlist_node *node;
1018 struct fib_info *fi;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001020 if (fib_info_laddrhash == NULL || local == 0)
1021 return 0;
1022
1023 hlist_for_each_entry(fi, node, head, fib_lhash) {
Octavian Purdila09ad9bc2009-11-25 15:14:13 -08001024 if (!net_eq(fi->fib_net, net))
Denis V. Lunev4814bdb2008-01-31 18:50:07 -08001025 continue;
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001026 if (fi->fib_prefsrc == local) {
1027 fi->fib_flags |= RTNH_F_DEAD;
1028 ret++;
1029 }
1030 }
1031 return ret;
1032}
1033
1034int fib_sync_down_dev(struct net_device *dev, int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035{
1036 int ret = 0;
1037 int scope = RT_SCOPE_NOWHERE;
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001038 struct fib_info *prev_fi = NULL;
1039 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1040 struct hlist_head *head = &fib_info_devhash[hash];
1041 struct hlist_node *node;
1042 struct fib_nh *nh;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001043
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 if (force)
1045 scope = -1;
1046
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001047 hlist_for_each_entry(nh, node, head, nh_hash) {
1048 struct fib_info *fi = nh->nh_parent;
1049 int dead;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001051 BUG_ON(!fi->fib_nhs);
1052 if (nh->nh_dev != dev || fi == prev_fi)
1053 continue;
1054 prev_fi = fi;
1055 dead = 0;
1056 change_nexthops(fi) {
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00001057 if (nexthop_nh->nh_flags & RTNH_F_DEAD)
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001058 dead++;
David S. Miller71fceff2010-01-15 01:16:40 -08001059 else if (nexthop_nh->nh_dev == dev &&
1060 nexthop_nh->nh_scope != scope) {
1061 nexthop_nh->nh_flags |= RTNH_F_DEAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062#ifdef CONFIG_IP_ROUTE_MULTIPATH
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001063 spin_lock_bh(&fib_multipath_lock);
David S. Miller71fceff2010-01-15 01:16:40 -08001064 fi->fib_power -= nexthop_nh->nh_power;
1065 nexthop_nh->nh_power = 0;
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001066 spin_unlock_bh(&fib_multipath_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067#endif
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001068 dead++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 }
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001070#ifdef CONFIG_IP_ROUTE_MULTIPATH
David S. Miller71fceff2010-01-15 01:16:40 -08001071 if (force > 1 && nexthop_nh->nh_dev == dev) {
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001072 dead = fi->fib_nhs;
1073 break;
1074 }
1075#endif
1076 } endfor_nexthops(fi)
1077 if (dead == fi->fib_nhs) {
1078 fi->fib_flags |= RTNH_F_DEAD;
1079 ret++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 }
1081 }
1082
1083 return ret;
1084}
1085
David S. Miller0c838ff2011-01-31 16:16:50 -08001086/* Must be invoked inside of an RCU protected region. */
1087void fib_select_default(struct fib_result *res)
1088{
1089 struct fib_info *fi = NULL, *last_resort = NULL;
1090 struct list_head *fa_head = res->fa_head;
1091 struct fib_table *tb = res->table;
1092 int order = -1, last_idx = -1;
1093 struct fib_alias *fa;
1094
1095 list_for_each_entry_rcu(fa, fa_head, fa_list) {
1096 struct fib_info *next_fi = fa->fa_info;
1097
David S. Miller37e826c2011-03-24 18:06:47 -07001098 if (next_fi->fib_scope != res->scope ||
David S. Miller0c838ff2011-01-31 16:16:50 -08001099 fa->fa_type != RTN_UNICAST)
1100 continue;
1101
1102 if (next_fi->fib_priority > res->fi->fib_priority)
1103 break;
1104 if (!next_fi->fib_nh[0].nh_gw ||
1105 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1106 continue;
1107
1108 fib_alias_accessed(fa);
1109
1110 if (fi == NULL) {
1111 if (next_fi != res->fi)
1112 break;
1113 } else if (!fib_detect_death(fi, order, &last_resort,
1114 &last_idx, tb->tb_default)) {
1115 fib_result_assign(res, fi);
1116 tb->tb_default = order;
1117 goto out;
1118 }
1119 fi = next_fi;
1120 order++;
1121 }
1122
1123 if (order <= 0 || fi == NULL) {
1124 tb->tb_default = -1;
1125 goto out;
1126 }
1127
1128 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1129 tb->tb_default)) {
1130 fib_result_assign(res, fi);
1131 tb->tb_default = order;
1132 goto out;
1133 }
1134
1135 if (last_idx >= 0)
1136 fib_result_assign(res, last_resort);
1137 tb->tb_default = last_idx;
1138out:
Eric Dumazet31d40932011-02-14 11:23:04 -08001139 return;
David S. Miller0c838ff2011-01-31 16:16:50 -08001140}
1141
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142#ifdef CONFIG_IP_ROUTE_MULTIPATH
1143
1144/*
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00001145 * Dead device goes up. We wake up dead nexthops.
1146 * It takes sense only on multipath routes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148int fib_sync_up(struct net_device *dev)
1149{
1150 struct fib_info *prev_fi;
1151 unsigned int hash;
1152 struct hlist_head *head;
1153 struct hlist_node *node;
1154 struct fib_nh *nh;
1155 int ret;
1156
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00001157 if (!(dev->flags & IFF_UP))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 return 0;
1159
1160 prev_fi = NULL;
1161 hash = fib_devindex_hashfn(dev->ifindex);
1162 head = &fib_info_devhash[hash];
1163 ret = 0;
1164
1165 hlist_for_each_entry(nh, node, head, nh_hash) {
1166 struct fib_info *fi = nh->nh_parent;
1167 int alive;
1168
1169 BUG_ON(!fi->fib_nhs);
1170 if (nh->nh_dev != dev || fi == prev_fi)
1171 continue;
1172
1173 prev_fi = fi;
1174 alive = 0;
1175 change_nexthops(fi) {
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00001176 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 alive++;
1178 continue;
1179 }
David S. Miller71fceff2010-01-15 01:16:40 -08001180 if (nexthop_nh->nh_dev == NULL ||
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00001181 !(nexthop_nh->nh_dev->flags & IFF_UP))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 continue;
David S. Miller71fceff2010-01-15 01:16:40 -08001183 if (nexthop_nh->nh_dev != dev ||
1184 !__in_dev_get_rtnl(dev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 continue;
1186 alive++;
1187 spin_lock_bh(&fib_multipath_lock);
David S. Miller71fceff2010-01-15 01:16:40 -08001188 nexthop_nh->nh_power = 0;
1189 nexthop_nh->nh_flags &= ~RTNH_F_DEAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 spin_unlock_bh(&fib_multipath_lock);
1191 } endfor_nexthops(fi)
1192
1193 if (alive > 0) {
1194 fi->fib_flags &= ~RTNH_F_DEAD;
1195 ret++;
1196 }
1197 }
1198
1199 return ret;
1200}
1201
1202/*
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00001203 * The algorithm is suboptimal, but it provides really
1204 * fair weighted route distribution.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 */
David S. Miller1b7fe5932011-03-10 17:01:16 -08001206void fib_select_multipath(struct fib_result *res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207{
1208 struct fib_info *fi = res->fi;
1209 int w;
1210
1211 spin_lock_bh(&fib_multipath_lock);
1212 if (fi->fib_power <= 0) {
1213 int power = 0;
1214 change_nexthops(fi) {
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00001215 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
David S. Miller71fceff2010-01-15 01:16:40 -08001216 power += nexthop_nh->nh_weight;
1217 nexthop_nh->nh_power = nexthop_nh->nh_weight;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 }
1219 } endfor_nexthops(fi);
1220 fi->fib_power = power;
1221 if (power <= 0) {
1222 spin_unlock_bh(&fib_multipath_lock);
1223 /* Race condition: route has just become dead. */
1224 res->nh_sel = 0;
1225 return;
1226 }
1227 }
1228
1229
1230 /* w should be random number [0..fi->fib_power-1],
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00001231 * it is pretty bad approximation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 */
1233
1234 w = jiffies % fi->fib_power;
1235
1236 change_nexthops(fi) {
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00001237 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
David S. Miller71fceff2010-01-15 01:16:40 -08001238 nexthop_nh->nh_power) {
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00001239 w -= nexthop_nh->nh_power;
1240 if (w <= 0) {
David S. Miller71fceff2010-01-15 01:16:40 -08001241 nexthop_nh->nh_power--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 fi->fib_power--;
1243 res->nh_sel = nhsel;
1244 spin_unlock_bh(&fib_multipath_lock);
1245 return;
1246 }
1247 }
1248 } endfor_nexthops(fi);
1249
1250 /* Race condition: route has just become dead. */
1251 res->nh_sel = 0;
1252 spin_unlock_bh(&fib_multipath_lock);
1253}
1254#endif