blob: e53871e4a097838f8eebdab828a3fd0ea23db5cb [file] [log] [blame]
Thomas Gleixner2874c5f2019-05-27 08:55:01 +02001// SPDX-License-Identifier: GPL-2.0-or-later
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * IPv4 Forwarding Information Base: semantics.
8 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 */
11
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080012#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/bitops.h>
14#include <linux/types.h>
15#include <linux/kernel.h>
16#include <linux/jiffies.h>
17#include <linux/mm.h>
18#include <linux/string.h>
19#include <linux/socket.h>
20#include <linux/sockios.h>
21#include <linux/errno.h>
22#include <linux/in.h>
23#include <linux/inet.h>
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020024#include <linux/inetdevice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/netdevice.h>
26#include <linux/if_arp.h>
27#include <linux/proc_fs.h>
28#include <linux/skbuff.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/init.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090030#include <linux/slab.h>
David Ahernc3ab2b42017-05-21 10:12:03 -060031#include <linux/netlink.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020033#include <net/arp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <net/ip.h>
35#include <net/protocol.h>
36#include <net/route.h>
37#include <net/tcp.h>
38#include <net/sock.h>
39#include <net/ip_fib.h>
David Ahern717a8f52019-04-05 16:30:32 -070040#include <net/ip6_fib.h>
David Ahern5481d732019-06-03 20:19:49 -070041#include <net/nexthop.h>
Thomas Graff21c7bc52006-08-15 00:34:17 -070042#include <net/netlink.h>
David Ahern3c618c12019-04-20 09:28:20 -070043#include <net/rtnh.h>
Roopa Prabhu571e7222015-07-21 10:43:47 +020044#include <net/lwtunnel.h>
Ido Schimmel04b1d4e2017-08-03 13:28:11 +020045#include <net/fib_notifier.h>
David Ahernc0a72072019-04-02 14:11:58 -070046#include <net/addrconf.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48#include "fib_lookup.h"
49
Stephen Hemminger832b4c52006-08-29 16:48:09 -070050static DEFINE_SPINLOCK(fib_info_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070051static struct hlist_head *fib_info_hash;
52static struct hlist_head *fib_info_laddrhash;
David S. Miller123b9732011-02-01 15:34:21 -080053static unsigned int fib_info_hash_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054static unsigned int fib_info_cnt;
55
56#define DEVINDEX_HASHBITS 8
57#define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
58static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
59
David Aherndcb1ecb2019-06-03 20:19:50 -070060/* for_nexthops and change_nexthops only used when nexthop object
61 * is not set in a fib_info. The logic within can reference fib_nh.
62 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070063#ifdef CONFIG_IP_ROUTE_MULTIPATH
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Eric Dumazet6a31d2a2010-10-04 20:00:18 +000065#define for_nexthops(fi) { \
66 int nhsel; const struct fib_nh *nh; \
67 for (nhsel = 0, nh = (fi)->fib_nh; \
David Ahern5481d732019-06-03 20:19:49 -070068 nhsel < fib_info_num_path((fi)); \
Eric Dumazet6a31d2a2010-10-04 20:00:18 +000069 nh++, nhsel++)
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
Eric Dumazet6a31d2a2010-10-04 20:00:18 +000071#define change_nexthops(fi) { \
72 int nhsel; struct fib_nh *nexthop_nh; \
73 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
David Ahern5481d732019-06-03 20:19:49 -070074 nhsel < fib_info_num_path((fi)); \
Eric Dumazet6a31d2a2010-10-04 20:00:18 +000075 nexthop_nh++, nhsel++)
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
77#else /* CONFIG_IP_ROUTE_MULTIPATH */
78
79/* Hope, that gcc will optimize it to get rid of dummy loop */
80
Eric Dumazet6a31d2a2010-10-04 20:00:18 +000081#define for_nexthops(fi) { \
82 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \
83 for (nhsel = 0; nhsel < 1; nhsel++)
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Eric Dumazet6a31d2a2010-10-04 20:00:18 +000085#define change_nexthops(fi) { \
86 int nhsel; \
87 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
88 for (nhsel = 0; nhsel < 1; nhsel++)
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
90#endif /* CONFIG_IP_ROUTE_MULTIPATH */
91
92#define endfor_nexthops(fi) }
93
94
David S. Miller3be06862011-03-07 15:01:10 -080095const struct fib_prop fib_props[RTN_MAX + 1] = {
Eric Dumazet6a31d2a2010-10-04 20:00:18 +000096 [RTN_UNSPEC] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 .error = 0,
98 .scope = RT_SCOPE_NOWHERE,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +000099 },
100 [RTN_UNICAST] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 .error = 0,
102 .scope = RT_SCOPE_UNIVERSE,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000103 },
104 [RTN_LOCAL] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 .error = 0,
106 .scope = RT_SCOPE_HOST,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000107 },
108 [RTN_BROADCAST] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 .error = 0,
110 .scope = RT_SCOPE_LINK,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000111 },
112 [RTN_ANYCAST] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 .error = 0,
114 .scope = RT_SCOPE_LINK,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000115 },
116 [RTN_MULTICAST] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 .error = 0,
118 .scope = RT_SCOPE_UNIVERSE,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000119 },
120 [RTN_BLACKHOLE] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 .error = -EINVAL,
122 .scope = RT_SCOPE_UNIVERSE,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000123 },
124 [RTN_UNREACHABLE] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 .error = -EHOSTUNREACH,
126 .scope = RT_SCOPE_UNIVERSE,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000127 },
128 [RTN_PROHIBIT] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 .error = -EACCES,
130 .scope = RT_SCOPE_UNIVERSE,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000131 },
132 [RTN_THROW] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 .error = -EAGAIN,
134 .scope = RT_SCOPE_UNIVERSE,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000135 },
136 [RTN_NAT] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 .error = -EINVAL,
138 .scope = RT_SCOPE_NOWHERE,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000139 },
140 [RTN_XRESOLVE] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 .error = -EINVAL,
142 .scope = RT_SCOPE_NOWHERE,
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000143 },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144};
145
David S. Millerc5038a82012-07-31 15:02:02 -0700146static void rt_fibinfo_free(struct rtable __rcu **rtp)
Eric Dumazet54764bb2012-07-31 01:08:23 +0000147{
148 struct rtable *rt = rcu_dereference_protected(*rtp, 1);
149
150 if (!rt)
151 return;
152
153 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
154 * because we waited an RCU grace period before calling
155 * free_fib_info_rcu()
156 */
157
Wei Wang95c47f92017-06-17 10:42:30 -0700158 dst_dev_put(&rt->dst);
Wei Wangb838d5e2017-06-17 10:42:32 -0700159 dst_release_immediate(&rt->dst);
Eric Dumazet54764bb2012-07-31 01:08:23 +0000160}
161
David Aherna5995e72019-04-30 07:45:50 -0700162static void free_nh_exceptions(struct fib_nh_common *nhc)
David S. Millerc5038a82012-07-31 15:02:02 -0700163{
Eric Dumazetcaa41522014-09-03 22:21:56 -0700164 struct fnhe_hash_bucket *hash;
David S. Millerc5038a82012-07-31 15:02:02 -0700165 int i;
166
David Aherna5995e72019-04-30 07:45:50 -0700167 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
Eric Dumazetcaa41522014-09-03 22:21:56 -0700168 if (!hash)
169 return;
David S. Millerc5038a82012-07-31 15:02:02 -0700170 for (i = 0; i < FNHE_HASH_SIZE; i++) {
171 struct fib_nh_exception *fnhe;
172
173 fnhe = rcu_dereference_protected(hash[i].chain, 1);
174 while (fnhe) {
175 struct fib_nh_exception *next;
Stephen Hemminger82695b32018-02-27 15:48:21 -0800176
David S. Millerc5038a82012-07-31 15:02:02 -0700177 next = rcu_dereference_protected(fnhe->fnhe_next, 1);
178
Timo Teräs2ffae992013-06-27 10:27:05 +0300179 rt_fibinfo_free(&fnhe->fnhe_rth_input);
180 rt_fibinfo_free(&fnhe->fnhe_rth_output);
David S. Millerc5038a82012-07-31 15:02:02 -0700181
182 kfree(fnhe);
183
184 fnhe = next;
185 }
186 }
187 kfree(hash);
188}
189
190static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
Eric Dumazetd26b3a72012-07-31 05:45:30 +0000191{
192 int cpu;
193
194 if (!rtp)
195 return;
196
197 for_each_possible_cpu(cpu) {
198 struct rtable *rt;
199
200 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
Wei Wang08301062017-06-17 10:42:29 -0700201 if (rt) {
Wei Wang95c47f92017-06-17 10:42:30 -0700202 dst_dev_put(&rt->dst);
Wei Wangb838d5e2017-06-17 10:42:32 -0700203 dst_release_immediate(&rt->dst);
Wei Wang08301062017-06-17 10:42:29 -0700204 }
Eric Dumazetd26b3a72012-07-31 05:45:30 +0000205 }
206 free_percpu(rtp);
207}
208
David Ahern979e2762019-03-27 20:53:58 -0700209void fib_nh_common_release(struct fib_nh_common *nhc)
210{
211 if (nhc->nhc_dev)
212 dev_put(nhc->nhc_dev);
213
214 lwtstate_put(nhc->nhc_lwtstate);
David Ahern0f457a32019-04-30 07:45:48 -0700215 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
216 rt_fibinfo_free(&nhc->nhc_rth_input);
David Aherna5995e72019-04-30 07:45:50 -0700217 free_nh_exceptions(nhc);
David Ahern979e2762019-03-27 20:53:58 -0700218}
219EXPORT_SYMBOL_GPL(fib_nh_common_release);
220
David Ahernfaa041a2019-03-27 20:53:49 -0700221void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
222{
223#ifdef CONFIG_IP_ROUTE_CLASSID
224 if (fib_nh->nh_tclassid)
225 net->ipv4.fib_num_tclassid_users--;
226#endif
David Ahern979e2762019-03-27 20:53:58 -0700227 fib_nh_common_release(&fib_nh->nh_common);
David Ahernfaa041a2019-03-27 20:53:49 -0700228}
229
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230/* Release a nexthop info record */
Yan, Zheng19c1ea12011-09-04 20:24:20 +0000231static void free_fib_info_rcu(struct rcu_head *head)
232{
233 struct fib_info *fi = container_of(head, struct fib_info, rcu);
234
David Ahern4c7e8082019-06-03 20:19:51 -0700235 if (fi->nh) {
236 nexthop_put(fi->nh);
237 } else {
238 change_nexthops(fi) {
239 fib_nh_release(fi->fib_net, nexthop_nh);
240 } endfor_nexthops(fi);
241 }
Yanmin Zhange49cc0d2012-05-23 15:39:45 +0000242
David Aherncc5f0eb2018-10-04 20:07:52 -0700243 ip_fib_metrics_put(fi->fib_metrics);
244
Yan, Zheng19c1ea12011-09-04 20:24:20 +0000245 kfree(fi);
246}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247
248void free_fib_info(struct fib_info *fi)
249{
250 if (fi->fib_dead == 0) {
Joe Perches058bd4d2012-03-11 18:36:11 +0000251 pr_warn("Freeing alive fib_info %p\n", fi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 return;
253 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 fib_info_cnt--;
David Ahernfaa041a2019-03-27 20:53:49 -0700255
Yan, Zheng19c1ea12011-09-04 20:24:20 +0000256 call_rcu(&fi->rcu, free_fib_info_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257}
Ido Schimmelb423cb12016-12-03 16:44:58 +0100258EXPORT_SYMBOL_GPL(free_fib_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
260void fib_release_info(struct fib_info *fi)
261{
Stephen Hemminger832b4c52006-08-29 16:48:09 -0700262 spin_lock_bh(&fib_info_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 if (fi && --fi->fib_treeref == 0) {
264 hlist_del(&fi->fib_hash);
265 if (fi->fib_prefsrc)
266 hlist_del(&fi->fib_lhash);
David Ahern4c7e8082019-06-03 20:19:51 -0700267 if (fi->nh) {
268 list_del(&fi->nh_list);
269 } else {
270 change_nexthops(fi) {
271 if (!nexthop_nh->fib_nh_dev)
272 continue;
273 hlist_del(&nexthop_nh->nh_hash);
274 } endfor_nexthops(fi)
275 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 fi->fib_dead = 1;
277 fib_info_put(fi);
278 }
Stephen Hemminger832b4c52006-08-29 16:48:09 -0700279 spin_unlock_bh(&fib_info_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280}
281
David Ahern5481d732019-06-03 20:19:49 -0700282static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283{
David Ahern5481d732019-06-03 20:19:49 -0700284 const struct fib_nh *onh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285
David Ahern4c7e8082019-06-03 20:19:51 -0700286 if (fi->nh || ofi->nh)
287 return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
288
289 if (ofi->fib_nhs == 0)
290 return 0;
291
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 for_nexthops(fi) {
David Ahern5481d732019-06-03 20:19:49 -0700293 onh = fib_info_nh(ofi, nhsel);
294
David Ahernb75ed8b2019-03-27 20:53:55 -0700295 if (nh->fib_nh_oif != onh->fib_nh_oif ||
David Aherna4ea5d42019-04-05 16:30:30 -0700296 nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
David Ahernb75ed8b2019-03-27 20:53:55 -0700297 nh->fib_nh_scope != onh->fib_nh_scope ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298#ifdef CONFIG_IP_ROUTE_MULTIPATH
David Ahernb75ed8b2019-03-27 20:53:55 -0700299 nh->fib_nh_weight != onh->fib_nh_weight ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300#endif
Patrick McHardyc7066f72011-01-14 13:36:42 +0100301#ifdef CONFIG_IP_ROUTE_CLASSID
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 nh->nh_tclassid != onh->nh_tclassid ||
303#endif
David Ahernb75ed8b2019-03-27 20:53:55 -0700304 lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
305 ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 return -1;
David Aherna4ea5d42019-04-05 16:30:30 -0700307
308 if (nh->fib_nh_gw_family == AF_INET &&
309 nh->fib_nh_gw4 != onh->fib_nh_gw4)
310 return -1;
311
312 if (nh->fib_nh_gw_family == AF_INET6 &&
313 ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
314 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 } endfor_nexthops(fi);
316 return 0;
317}
318
David S. Miller88ebc722008-01-12 21:49:01 -0800319static inline unsigned int fib_devindex_hashfn(unsigned int val)
320{
321 unsigned int mask = DEVINDEX_HASHSIZE - 1;
322
323 return (val ^
324 (val >> DEVINDEX_HASHBITS) ^
325 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
326}
327
David Ahern6c48ea52019-06-08 14:53:33 -0700328static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
329 u32 prefsrc, u32 priority)
330{
331 unsigned int val = init_val;
332
333 val ^= (protocol << 8) | scope;
334 val ^= prefsrc;
335 val ^= priority;
336
337 return val;
338}
339
340static unsigned int fib_info_hashfn_result(unsigned int val)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341{
David S. Miller123b9732011-02-01 15:34:21 -0800342 unsigned int mask = (fib_info_hash_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343
David Ahern6c48ea52019-06-08 14:53:33 -0700344 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
345}
346
347static inline unsigned int fib_info_hashfn(struct fib_info *fi)
348{
349 unsigned int val;
350
351 val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
352 fi->fib_scope, (__force u32)fi->fib_prefsrc,
353 fi->fib_priority);
David Ahern4c7e8082019-06-03 20:19:51 -0700354
355 if (fi->nh) {
356 val ^= fib_devindex_hashfn(fi->nh->id);
357 } else {
358 for_nexthops(fi) {
359 val ^= fib_devindex_hashfn(nh->fib_nh_oif);
360 } endfor_nexthops(fi)
361 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
David Ahern6c48ea52019-06-08 14:53:33 -0700363 return fib_info_hashfn_result(val);
364}
365
366/* no metrics, only nexthop id */
367static struct fib_info *fib_find_info_nh(struct net *net,
368 const struct fib_config *cfg)
369{
370 struct hlist_head *head;
371 struct fib_info *fi;
372 unsigned int hash;
373
374 hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
375 cfg->fc_protocol, cfg->fc_scope,
376 (__force u32)cfg->fc_prefsrc,
377 cfg->fc_priority);
378 hash = fib_info_hashfn_result(hash);
379 head = &fib_info_hash[hash];
380
381 hlist_for_each_entry(fi, head, fib_hash) {
382 if (!net_eq(fi->fib_net, net))
383 continue;
384 if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
385 continue;
386 if (cfg->fc_protocol == fi->fib_protocol &&
387 cfg->fc_scope == fi->fib_scope &&
388 cfg->fc_prefsrc == fi->fib_prefsrc &&
389 cfg->fc_priority == fi->fib_priority &&
390 cfg->fc_type == fi->fib_type &&
391 cfg->fc_table == fi->fib_tb_id &&
392 !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
393 return fi;
394 }
395
396 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397}
398
David Ahern5481d732019-06-03 20:19:49 -0700399static struct fib_info *fib_find_info(struct fib_info *nfi)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400{
401 struct hlist_head *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 struct fib_info *fi;
403 unsigned int hash;
404
405 hash = fib_info_hashfn(nfi);
406 head = &fib_info_hash[hash];
407
Sasha Levinb67bfe02013-02-27 17:06:00 -0800408 hlist_for_each_entry(fi, head, fib_hash) {
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800409 if (!net_eq(fi->fib_net, nfi->fib_net))
Denis V. Lunev4814bdb2008-01-31 18:50:07 -0800410 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 if (fi->fib_nhs != nfi->fib_nhs)
412 continue;
413 if (nfi->fib_protocol == fi->fib_protocol &&
David S. Miller37e826c2011-03-24 18:06:47 -0700414 nfi->fib_scope == fi->fib_scope &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 nfi->fib_prefsrc == fi->fib_prefsrc &&
416 nfi->fib_priority == fi->fib_priority &&
Eric Dumazetf4ef85b2012-10-04 01:25:26 +0000417 nfi->fib_type == fi->fib_type &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 memcmp(nfi->fib_metrics, fi->fib_metrics,
Eric Dumazetfcd13f42011-03-24 07:01:24 +0000419 sizeof(u32) * RTAX_MAX) == 0 &&
Andy Gospodarek8a3d0312015-06-23 13:45:36 -0400420 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
David Ahern4c7e8082019-06-03 20:19:51 -0700421 nh_comp(fi, nfi) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 return fi;
423 }
424
425 return NULL;
426}
427
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428/* Check, that the gateway is already configured.
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000429 * Used only by redirect accept routine.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 */
Al Virod878e72e2006-09-26 22:18:13 -0700431int ip_fib_check_default(__be32 gw, struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432{
433 struct hlist_head *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 struct fib_nh *nh;
435 unsigned int hash;
436
Stephen Hemminger832b4c52006-08-29 16:48:09 -0700437 spin_lock(&fib_info_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
439 hash = fib_devindex_hashfn(dev->ifindex);
440 head = &fib_info_devhash[hash];
Sasha Levinb67bfe02013-02-27 17:06:00 -0800441 hlist_for_each_entry(nh, head, nh_hash) {
David Ahernb75ed8b2019-03-27 20:53:55 -0700442 if (nh->fib_nh_dev == dev &&
443 nh->fib_nh_gw4 == gw &&
444 !(nh->fib_nh_flags & RTNH_F_DEAD)) {
Stephen Hemminger832b4c52006-08-29 16:48:09 -0700445 spin_unlock(&fib_info_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 return 0;
447 }
448 }
449
Stephen Hemminger832b4c52006-08-29 16:48:09 -0700450 spin_unlock(&fib_info_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
452 return -1;
453}
454
Thomas Graf339bf982006-11-10 14:10:15 -0800455static inline size_t fib_nlmsg_size(struct fib_info *fi)
456{
457 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
458 + nla_total_size(4) /* RTA_TABLE */
459 + nla_total_size(4) /* RTA_DST */
460 + nla_total_size(4) /* RTA_PRIORITY */
Daniel Borkmannea697632015-01-05 23:57:47 +0100461 + nla_total_size(4) /* RTA_PREFSRC */
462 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
David Ahern5481d732019-06-03 20:19:49 -0700463 unsigned int nhs = fib_info_num_path(fi);
Thomas Graf339bf982006-11-10 14:10:15 -0800464
465 /* space for nested metrics */
466 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
467
David Ahern4c7e8082019-06-03 20:19:51 -0700468 if (fi->nh)
469 payload += nla_total_size(4); /* RTA_NH_ID */
470
David Ahern5481d732019-06-03 20:19:49 -0700471 if (nhs) {
Roopa Prabhu571e7222015-07-21 10:43:47 +0200472 size_t nh_encapsize = 0;
David Ahern5481d732019-06-03 20:19:49 -0700473 /* Also handles the special case nhs == 1 */
Thomas Graf339bf982006-11-10 14:10:15 -0800474
475 /* each nexthop is packed in an attribute */
476 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
David Aherndcb1ecb2019-06-03 20:19:50 -0700477 unsigned int i;
Thomas Graf339bf982006-11-10 14:10:15 -0800478
479 /* may contain flow and gateway attribute */
480 nhsize += 2 * nla_total_size(4);
481
Roopa Prabhu571e7222015-07-21 10:43:47 +0200482 /* grab encap info */
David Aherndcb1ecb2019-06-03 20:19:50 -0700483 for (i = 0; i < fib_info_num_path(fi); i++) {
484 struct fib_nh_common *nhc = fib_info_nhc(fi, i);
485
486 if (nhc->nhc_lwtstate) {
Roopa Prabhu571e7222015-07-21 10:43:47 +0200487 /* RTA_ENCAP_TYPE */
488 nh_encapsize += lwtunnel_get_encap_size(
David Aherndcb1ecb2019-06-03 20:19:50 -0700489 nhc->nhc_lwtstate);
Roopa Prabhu571e7222015-07-21 10:43:47 +0200490 /* RTA_ENCAP */
491 nh_encapsize += nla_total_size(2);
492 }
David Aherndcb1ecb2019-06-03 20:19:50 -0700493 }
Roopa Prabhu571e7222015-07-21 10:43:47 +0200494
Thomas Graf339bf982006-11-10 14:10:15 -0800495 /* all nexthops are packed in a nested attribute */
David Ahern5481d732019-06-03 20:19:49 -0700496 payload += nla_total_size((nhs * nhsize) + nh_encapsize);
Roopa Prabhu571e7222015-07-21 10:43:47 +0200497
Thomas Graf339bf982006-11-10 14:10:15 -0800498 }
499
500 return payload;
501}
502
Al Viro81f7bf62006-09-27 18:40:00 -0700503void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
Joe Perches9877b252013-10-17 13:34:11 -0700504 int dst_len, u32 tb_id, const struct nl_info *info,
Milan Kocianb8f55832007-05-23 14:55:06 -0700505 unsigned int nlm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506{
Ido Schimmel1e301fd2020-01-14 13:23:10 +0200507 struct fib_rt_info fri;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 struct sk_buff *skb;
Thomas Graf4e902c52006-08-17 18:14:52 -0700509 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
Thomas Graff21c7bc52006-08-15 00:34:17 -0700510 int err = -ENOBUFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511
Thomas Graf339bf982006-11-10 14:10:15 -0800512 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
Ian Morris51456b22015-04-03 09:17:26 +0100513 if (!skb)
Thomas Graff21c7bc52006-08-15 00:34:17 -0700514 goto errout;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515
Ido Schimmel1e301fd2020-01-14 13:23:10 +0200516 fri.fi = fa->fa_info;
517 fri.tb_id = tb_id;
518 fri.dst = key;
519 fri.dst_len = dst_len;
520 fri.tos = fa->fa_tos;
521 fri.type = fa->fa_type;
Ido Schimmel90b93f12020-01-14 13:23:11 +0200522 fri.offload = fa->offload;
523 fri.trap = fa->trap;
Ido Schimmel1e301fd2020-01-14 13:23:10 +0200524 err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
Patrick McHardy26932562007-01-31 23:16:40 -0800525 if (err < 0) {
526 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
527 WARN_ON(err == -EMSGSIZE);
528 kfree_skb(skb);
529 goto errout;
530 }
Eric W. Biederman15e47302012-09-07 20:12:54 +0000531 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
Pablo Neira Ayuso1ce85fe2009-02-24 23:18:28 -0800532 info->nlh, GFP_KERNEL);
533 return;
Thomas Graff21c7bc52006-08-15 00:34:17 -0700534errout:
535 if (err < 0)
Denis V. Lunev4d1169c2008-01-10 03:26:13 -0800536 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537}
538
Stephen Hemmingerc9cb6b62013-12-28 11:05:36 -0800539static int fib_detect_death(struct fib_info *fi, int order,
540 struct fib_info **last_resort, int *last_idx,
541 int dflt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542{
David Ahern619d1822019-04-05 16:30:37 -0700543 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 struct neighbour *n;
545 int state = NUD_NONE;
546
David Ahern619d1822019-04-05 16:30:37 -0700547 if (likely(nhc->nhc_gw_family == AF_INET))
548 n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
549 else if (nhc->nhc_gw_family == AF_INET6)
550 n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
551 nhc->nhc_dev);
552 else
553 n = NULL;
554
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 if (n) {
556 state = n->nud_state;
557 neigh_release(n);
Julian Anastasov88f64322015-07-23 10:39:35 +0300558 } else {
559 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 }
Jianjun Kongd93191002008-11-03 00:23:42 -0800561 if (state == NUD_REACHABLE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 return 0;
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000563 if ((state & NUD_VALID) && order != dflt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 return 0;
Eric Dumazet6a31d2a2010-10-04 20:00:18 +0000565 if ((state & NUD_VALID) ||
Julian Anastasov88f64322015-07-23 10:39:35 +0300566 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 *last_resort = fi;
568 *last_idx = order;
569 }
570 return 1;
571}
572
Alexander Aringfaee6762020-03-27 18:00:21 -0400573int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
574 struct nlattr *encap, u16 encap_type,
575 void *cfg, gfp_t gfp_flags,
David Ahern979e2762019-03-27 20:53:58 -0700576 struct netlink_ext_ack *extack)
577{
David Ahern0f457a32019-04-30 07:45:48 -0700578 int err;
579
580 nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
581 gfp_flags);
582 if (!nhc->nhc_pcpu_rth_output)
583 return -ENOMEM;
584
David Ahern979e2762019-03-27 20:53:58 -0700585 if (encap) {
586 struct lwtunnel_state *lwtstate;
David Ahern979e2762019-03-27 20:53:58 -0700587
588 if (encap_type == LWTUNNEL_ENCAP_NONE) {
589 NL_SET_ERR_MSG(extack, "LWT encap type not specified");
David Ahern0f457a32019-04-30 07:45:48 -0700590 err = -EINVAL;
591 goto lwt_failure;
David Ahern979e2762019-03-27 20:53:58 -0700592 }
Alexander Aringfaee6762020-03-27 18:00:21 -0400593 err = lwtunnel_build_state(net, encap_type, encap,
594 nhc->nhc_family, cfg, &lwtstate,
595 extack);
David Ahern979e2762019-03-27 20:53:58 -0700596 if (err)
David Ahern0f457a32019-04-30 07:45:48 -0700597 goto lwt_failure;
David Ahern979e2762019-03-27 20:53:58 -0700598
599 nhc->nhc_lwtstate = lwtstate_get(lwtstate);
600 }
601
602 return 0;
David Ahern0f457a32019-04-30 07:45:48 -0700603
604lwt_failure:
605 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
606 nhc->nhc_pcpu_rth_output = NULL;
607 return err;
David Ahern979e2762019-03-27 20:53:58 -0700608}
609EXPORT_SYMBOL_GPL(fib_nh_common_init);
610
David Aherne4516ef2019-03-27 20:53:48 -0700611int fib_nh_init(struct net *net, struct fib_nh *nh,
612 struct fib_config *cfg, int nh_weight,
613 struct netlink_ext_ack *extack)
614{
David Ahern0f457a32019-04-30 07:45:48 -0700615 int err;
David Aherne4516ef2019-03-27 20:53:48 -0700616
David Ahernf1741732019-03-27 20:53:57 -0700617 nh->fib_nh_family = AF_INET;
618
Alexander Aringfaee6762020-03-27 18:00:21 -0400619 err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
David Ahern979e2762019-03-27 20:53:58 -0700620 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
621 if (err)
David Ahern0f457a32019-04-30 07:45:48 -0700622 return err;
David Aherne4516ef2019-03-27 20:53:48 -0700623
David Ahernb75ed8b2019-03-27 20:53:55 -0700624 nh->fib_nh_oif = cfg->fc_oif;
David Aherna4ea5d42019-04-05 16:30:30 -0700625 nh->fib_nh_gw_family = cfg->fc_gw_family;
626 if (cfg->fc_gw_family == AF_INET)
David Ahernf35b7942019-04-05 16:30:28 -0700627 nh->fib_nh_gw4 = cfg->fc_gw4;
David Aherna4ea5d42019-04-05 16:30:30 -0700628 else if (cfg->fc_gw_family == AF_INET6)
629 nh->fib_nh_gw6 = cfg->fc_gw6;
630
David Ahernb75ed8b2019-03-27 20:53:55 -0700631 nh->fib_nh_flags = cfg->fc_flags;
David Aherne4516ef2019-03-27 20:53:48 -0700632
633#ifdef CONFIG_IP_ROUTE_CLASSID
634 nh->nh_tclassid = cfg->fc_flow;
635 if (nh->nh_tclassid)
636 net->ipv4.fib_num_tclassid_users++;
637#endif
638#ifdef CONFIG_IP_ROUTE_MULTIPATH
David Ahernb75ed8b2019-03-27 20:53:55 -0700639 nh->fib_nh_weight = nh_weight;
David Aherne4516ef2019-03-27 20:53:48 -0700640#endif
641 return 0;
David Aherne4516ef2019-03-27 20:53:48 -0700642}
643
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644#ifdef CONFIG_IP_ROUTE_MULTIPATH
645
David Ahern6d8422a12017-05-21 10:12:02 -0600646static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
647 struct netlink_ext_ack *extack)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648{
649 int nhs = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
Thomas Graf4e902c52006-08-17 18:14:52 -0700651 while (rtnh_ok(rtnh, remaining)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 nhs++;
Thomas Graf4e902c52006-08-17 18:14:52 -0700653 rtnh = rtnh_next(rtnh, &remaining);
654 }
655
656 /* leftover implies invalid nexthop configuration, discard it */
David Ahernc3ab2b42017-05-21 10:12:03 -0600657 if (remaining > 0) {
658 NL_SET_ERR_MSG(extack,
659 "Invalid nexthop configuration - extra data after nexthops");
660 nhs = 0;
661 }
662
663 return nhs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664}
665
David Ahern4c7e8082019-06-03 20:19:51 -0700666/* only called when fib_nh is integrated into fib_info */
Thomas Graf4e902c52006-08-17 18:14:52 -0700667static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
David Ahern6d8422a12017-05-21 10:12:02 -0600668 int remaining, struct fib_config *cfg,
669 struct netlink_ext_ack *extack)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670{
David Aherne4516ef2019-03-27 20:53:48 -0700671 struct net *net = fi->fib_net;
672 struct fib_config fib_cfg;
David Ahern5481d732019-06-03 20:19:49 -0700673 struct fib_nh *nh;
Roopa Prabhu571e7222015-07-21 10:43:47 +0200674 int ret;
675
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 change_nexthops(fi) {
Thomas Graf4e902c52006-08-17 18:14:52 -0700677 int attrlen;
678
David Aherne4516ef2019-03-27 20:53:48 -0700679 memset(&fib_cfg, 0, sizeof(fib_cfg));
680
David Ahernc3ab2b42017-05-21 10:12:03 -0600681 if (!rtnh_ok(rtnh, remaining)) {
682 NL_SET_ERR_MSG(extack,
683 "Invalid nexthop configuration - extra data after nexthop");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 return -EINVAL;
David Ahernc3ab2b42017-05-21 10:12:03 -0600685 }
Thomas Graf4e902c52006-08-17 18:14:52 -0700686
David Ahernc3ab2b42017-05-21 10:12:03 -0600687 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
688 NL_SET_ERR_MSG(extack,
689 "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
Julian Anastasov80610222016-07-10 21:11:55 +0300690 return -EINVAL;
David Ahernc3ab2b42017-05-21 10:12:03 -0600691 }
Julian Anastasov80610222016-07-10 21:11:55 +0300692
David Aherne4516ef2019-03-27 20:53:48 -0700693 fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
694 fib_cfg.fc_oif = rtnh->rtnh_ifindex;
Thomas Graf4e902c52006-08-17 18:14:52 -0700695
696 attrlen = rtnh_attrlen(rtnh);
697 if (attrlen > 0) {
David Ahernd1566262019-04-05 16:30:40 -0700698 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
Thomas Graf4e902c52006-08-17 18:14:52 -0700699
700 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
David Ahernd1566262019-04-05 16:30:40 -0700701 nlav = nla_find(attrs, attrlen, RTA_VIA);
702 if (nla && nlav) {
703 NL_SET_ERR_MSG(extack,
704 "Nexthop configuration can not contain both GATEWAY and VIA");
705 return -EINVAL;
706 }
David Ahernf35b7942019-04-05 16:30:28 -0700707 if (nla) {
David Ahernf35b7942019-04-05 16:30:28 -0700708 fib_cfg.fc_gw4 = nla_get_in_addr(nla);
David Ahernd73f80f2019-04-10 10:05:51 -0700709 if (fib_cfg.fc_gw4)
710 fib_cfg.fc_gw_family = AF_INET;
David Ahernd1566262019-04-05 16:30:40 -0700711 } else if (nlav) {
712 ret = fib_gw_from_via(&fib_cfg, nlav, extack);
713 if (ret)
714 goto errout;
David Ahernf35b7942019-04-05 16:30:28 -0700715 }
David Aherne4516ef2019-03-27 20:53:48 -0700716
Thomas Graf4e902c52006-08-17 18:14:52 -0700717 nla = nla_find(attrs, attrlen, RTA_FLOW);
David Aherne4516ef2019-03-27 20:53:48 -0700718 if (nla)
719 fib_cfg.fc_flow = nla_get_u32(nla);
Roopa Prabhu571e7222015-07-21 10:43:47 +0200720
David Aherne4516ef2019-03-27 20:53:48 -0700721 fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
722 nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
723 if (nla)
724 fib_cfg.fc_encap_type = nla_get_u16(nla);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 }
Thomas Graf4e902c52006-08-17 18:14:52 -0700726
David Aherne4516ef2019-03-27 20:53:48 -0700727 ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
728 rtnh->rtnh_hops + 1, extack);
729 if (ret)
730 goto errout;
731
Thomas Graf4e902c52006-08-17 18:14:52 -0700732 rtnh = rtnh_next(rtnh, &remaining);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 } endfor_nexthops(fi);
Thomas Graf4e902c52006-08-17 18:14:52 -0700734
Roopa Prabhu571e7222015-07-21 10:43:47 +0200735 ret = -EINVAL;
David Ahern5481d732019-06-03 20:19:49 -0700736 nh = fib_info_nh(fi, 0);
737 if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
David Aherne4516ef2019-03-27 20:53:48 -0700738 NL_SET_ERR_MSG(extack,
739 "Nexthop device index does not match RTA_OIF");
740 goto errout;
741 }
David Ahernf35b7942019-04-05 16:30:28 -0700742 if (cfg->fc_gw_family) {
David Ahern5481d732019-06-03 20:19:49 -0700743 if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
David Ahernf35b7942019-04-05 16:30:28 -0700744 (cfg->fc_gw_family == AF_INET &&
David Ahern5481d732019-06-03 20:19:49 -0700745 nh->fib_nh_gw4 != cfg->fc_gw4) ||
David Aherna4ea5d42019-04-05 16:30:30 -0700746 (cfg->fc_gw_family == AF_INET6 &&
David Ahern5481d732019-06-03 20:19:49 -0700747 ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
David Ahernf35b7942019-04-05 16:30:28 -0700748 NL_SET_ERR_MSG(extack,
David Aherna4ea5d42019-04-05 16:30:30 -0700749 "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
David Ahernf35b7942019-04-05 16:30:28 -0700750 goto errout;
751 }
David Aherne4516ef2019-03-27 20:53:48 -0700752 }
753#ifdef CONFIG_IP_ROUTE_CLASSID
David Ahern5481d732019-06-03 20:19:49 -0700754 if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
David Aherne4516ef2019-03-27 20:53:48 -0700755 NL_SET_ERR_MSG(extack,
756 "Nexthop class id does not match RTA_FLOW");
757 goto errout;
758 }
759#endif
760 ret = 0;
Roopa Prabhu571e7222015-07-21 10:43:47 +0200761errout:
762 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763}
764
David Ahern4c7e8082019-06-03 20:19:51 -0700765/* only called when fib_nh is integrated into fib_info */
Peter Nørlund0e884c72015-09-30 10:12:21 +0200766static void fib_rebalance(struct fib_info *fi)
767{
768 int total;
769 int w;
Peter Nørlund0e884c72015-09-30 10:12:21 +0200770
David Ahern5481d732019-06-03 20:19:49 -0700771 if (fib_info_num_path(fi) < 2)
Peter Nørlund0e884c72015-09-30 10:12:21 +0200772 return;
773
774 total = 0;
775 for_nexthops(fi) {
David Ahernb75ed8b2019-03-27 20:53:55 -0700776 if (nh->fib_nh_flags & RTNH_F_DEAD)
Peter Nørlund0e884c72015-09-30 10:12:21 +0200777 continue;
778
David Ahernb75ed8b2019-03-27 20:53:55 -0700779 if (ip_ignore_linkdown(nh->fib_nh_dev) &&
780 nh->fib_nh_flags & RTNH_F_LINKDOWN)
Peter Nørlund0e884c72015-09-30 10:12:21 +0200781 continue;
782
David Ahernb75ed8b2019-03-27 20:53:55 -0700783 total += nh->fib_nh_weight;
Peter Nørlund0e884c72015-09-30 10:12:21 +0200784 } endfor_nexthops(fi);
785
786 w = 0;
787 change_nexthops(fi) {
788 int upper_bound;
789
David Ahernb75ed8b2019-03-27 20:53:55 -0700790 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
Peter Nørlund0e884c72015-09-30 10:12:21 +0200791 upper_bound = -1;
David Ahernb75ed8b2019-03-27 20:53:55 -0700792 } else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
793 nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
Peter Nørlund0e884c72015-09-30 10:12:21 +0200794 upper_bound = -1;
795 } else {
David Ahernb75ed8b2019-03-27 20:53:55 -0700796 w += nexthop_nh->fib_nh_weight;
Peter Nørlund0a837fe2015-10-06 07:24:47 +0200797 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
798 total) - 1;
Peter Nørlund0e884c72015-09-30 10:12:21 +0200799 }
800
David Ahernb75ed8b2019-03-27 20:53:55 -0700801 atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
Peter Nørlund0e884c72015-09-30 10:12:21 +0200802 } endfor_nexthops(fi);
Peter Nørlund0e884c72015-09-30 10:12:21 +0200803}
Peter Nørlund0e884c72015-09-30 10:12:21 +0200804#else /* CONFIG_IP_ROUTE_MULTIPATH */
805
David Ahern8373c6c2019-03-27 20:53:46 -0700806static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
807 int remaining, struct fib_config *cfg,
808 struct netlink_ext_ack *extack)
809{
810 NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
811
812 return -EINVAL;
813}
814
Peter Nørlund0e884c72015-09-30 10:12:21 +0200815#define fib_rebalance(fi) do { } while (0)
Peter Nørlund0e884c72015-09-30 10:12:21 +0200816
817#endif /* CONFIG_IP_ROUTE_MULTIPATH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
Alexander Aringfaee6762020-03-27 18:00:21 -0400819static int fib_encap_match(struct net *net, u16 encap_type,
Ying Xuee01286e2015-08-19 16:04:51 +0800820 struct nlattr *encap,
David Ahern30357d72017-01-30 12:07:37 -0800821 const struct fib_nh *nh,
David Ahern9ae28722017-05-27 16:19:28 -0600822 const struct fib_config *cfg,
823 struct netlink_ext_ack *extack)
Roopa Prabhu571e7222015-07-21 10:43:47 +0200824{
825 struct lwtunnel_state *lwtstate;
Jiri Bencdf383e62015-08-18 18:41:13 +0200826 int ret, result = 0;
Roopa Prabhu571e7222015-07-21 10:43:47 +0200827
828 if (encap_type == LWTUNNEL_ENCAP_NONE)
829 return 0;
830
Alexander Aringfaee6762020-03-27 18:00:21 -0400831 ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
David Ahern9ae28722017-05-27 16:19:28 -0600832 cfg, &lwtstate, extack);
Jiri Bencdf383e62015-08-18 18:41:13 +0200833 if (!ret) {
David Ahernb75ed8b2019-03-27 20:53:55 -0700834 result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
Jiri Bencdf383e62015-08-18 18:41:13 +0200835 lwtstate_free(lwtstate);
836 }
Roopa Prabhu571e7222015-07-21 10:43:47 +0200837
Jiri Bencdf383e62015-08-18 18:41:13 +0200838 return result;
Roopa Prabhu571e7222015-07-21 10:43:47 +0200839}
840
Alexander Aringfaee6762020-03-27 18:00:21 -0400841int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
David Ahern9ae28722017-05-27 16:19:28 -0600842 struct netlink_ext_ack *extack)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843{
844#ifdef CONFIG_IP_ROUTE_MULTIPATH
Thomas Graf4e902c52006-08-17 18:14:52 -0700845 struct rtnexthop *rtnh;
846 int remaining;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847#endif
848
Thomas Graf4e902c52006-08-17 18:14:52 -0700849 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 return 1;
851
David Ahern493ced12019-06-08 14:53:32 -0700852 if (cfg->fc_nh_id) {
853 if (fi->nh && cfg->fc_nh_id == fi->nh->id)
854 return 0;
855 return 1;
856 }
857
David Ahernf35b7942019-04-05 16:30:28 -0700858 if (cfg->fc_oif || cfg->fc_gw_family) {
David Ahern5481d732019-06-03 20:19:49 -0700859 struct fib_nh *nh = fib_info_nh(fi, 0);
860
Roopa Prabhu571e7222015-07-21 10:43:47 +0200861 if (cfg->fc_encap) {
Alexander Aringfaee6762020-03-27 18:00:21 -0400862 if (fib_encap_match(net, cfg->fc_encap_type,
863 cfg->fc_encap, nh, cfg, extack))
David Ahern9ae28722017-05-27 16:19:28 -0600864 return 1;
Roopa Prabhu571e7222015-07-21 10:43:47 +0200865 }
Stefano Brivioa8c6db12018-02-15 09:46:03 +0100866#ifdef CONFIG_IP_ROUTE_CLASSID
867 if (cfg->fc_flow &&
David Ahern5481d732019-06-03 20:19:49 -0700868 cfg->fc_flow != nh->nh_tclassid)
Stefano Brivioa8c6db12018-02-15 09:46:03 +0100869 return 1;
870#endif
David Ahern5481d732019-06-03 20:19:49 -0700871 if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
David Ahernf35b7942019-04-05 16:30:28 -0700872 (cfg->fc_gw_family &&
David Ahern5481d732019-06-03 20:19:49 -0700873 cfg->fc_gw_family != nh->fib_nh_gw_family))
David Ahernf35b7942019-04-05 16:30:28 -0700874 return 1;
875
876 if (cfg->fc_gw_family == AF_INET &&
David Ahern5481d732019-06-03 20:19:49 -0700877 cfg->fc_gw4 != nh->fib_nh_gw4)
David Ahernf35b7942019-04-05 16:30:28 -0700878 return 1;
879
David Aherna4ea5d42019-04-05 16:30:30 -0700880 if (cfg->fc_gw_family == AF_INET6 &&
David Ahern5481d732019-06-03 20:19:49 -0700881 ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
David Aherna4ea5d42019-04-05 16:30:30 -0700882 return 1;
883
David Ahernf35b7942019-04-05 16:30:28 -0700884 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 }
886
887#ifdef CONFIG_IP_ROUTE_MULTIPATH
Ian Morris51456b22015-04-03 09:17:26 +0100888 if (!cfg->fc_mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 return 0;
Thomas Graf4e902c52006-08-17 18:14:52 -0700890
891 rtnh = cfg->fc_mp;
892 remaining = cfg->fc_mp_len;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900893
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 for_nexthops(fi) {
Thomas Graf4e902c52006-08-17 18:14:52 -0700895 int attrlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
Thomas Graf4e902c52006-08-17 18:14:52 -0700897 if (!rtnh_ok(rtnh, remaining))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 return -EINVAL;
Thomas Graf4e902c52006-08-17 18:14:52 -0700899
David Ahernb75ed8b2019-03-27 20:53:55 -0700900 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 return 1;
Thomas Graf4e902c52006-08-17 18:14:52 -0700902
903 attrlen = rtnh_attrlen(rtnh);
Jiri Pirkof76936d2014-10-13 16:34:10 +0200904 if (attrlen > 0) {
David Ahernd1566262019-04-05 16:30:40 -0700905 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
Thomas Graf4e902c52006-08-17 18:14:52 -0700906
907 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
David Ahernd1566262019-04-05 16:30:40 -0700908 nlav = nla_find(attrs, attrlen, RTA_VIA);
909 if (nla && nlav) {
910 NL_SET_ERR_MSG(extack,
911 "Nexthop configuration can not contain both GATEWAY and VIA");
912 return -EINVAL;
913 }
914
915 if (nla) {
916 if (nh->fib_nh_gw_family != AF_INET ||
917 nla_get_in_addr(nla) != nh->fib_nh_gw4)
918 return 1;
919 } else if (nlav) {
920 struct fib_config cfg2;
921 int err;
922
923 err = fib_gw_from_via(&cfg2, nlav, extack);
924 if (err)
925 return err;
926
927 switch (nh->fib_nh_gw_family) {
928 case AF_INET:
929 if (cfg2.fc_gw_family != AF_INET ||
930 cfg2.fc_gw4 != nh->fib_nh_gw4)
931 return 1;
932 break;
933 case AF_INET6:
934 if (cfg2.fc_gw_family != AF_INET6 ||
935 ipv6_addr_cmp(&cfg2.fc_gw6,
936 &nh->fib_nh_gw6))
937 return 1;
938 break;
939 }
940 }
941
Patrick McHardyc7066f72011-01-14 13:36:42 +0100942#ifdef CONFIG_IP_ROUTE_CLASSID
Thomas Graf4e902c52006-08-17 18:14:52 -0700943 nla = nla_find(attrs, attrlen, RTA_FLOW);
944 if (nla && nla_get_u32(nla) != nh->nh_tclassid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 return 1;
946#endif
947 }
Thomas Graf4e902c52006-08-17 18:14:52 -0700948
949 rtnh = rtnh_next(rtnh, &remaining);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 } endfor_nexthops(fi);
951#endif
952 return 0;
953}
954
Xin Long5f9ae3d2017-08-23 10:07:26 +0800955bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
956{
957 struct nlattr *nla;
958 int remaining;
959
960 if (!cfg->fc_mx)
961 return true;
962
963 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
964 int type = nla_type(nla);
Phil Sutterd03a4552017-12-19 15:17:13 +0100965 u32 fi_val, val;
Xin Long5f9ae3d2017-08-23 10:07:26 +0800966
967 if (!type)
968 continue;
969 if (type > RTAX_MAX)
970 return false;
971
972 if (type == RTAX_CC_ALGO) {
973 char tmp[TCP_CA_NAME_MAX];
974 bool ecn_ca = false;
975
976 nla_strlcpy(tmp, nla, sizeof(tmp));
Stephen Hemminger6670e152017-11-14 08:25:49 -0800977 val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
Xin Long5f9ae3d2017-08-23 10:07:26 +0800978 } else {
Eric Dumazet5b5e7a02018-06-05 06:06:19 -0700979 if (nla_len(nla) != sizeof(u32))
980 return false;
Xin Long5f9ae3d2017-08-23 10:07:26 +0800981 val = nla_get_u32(nla);
982 }
983
Phil Sutterd03a4552017-12-19 15:17:13 +0100984 fi_val = fi->fib_metrics->metrics[type - 1];
985 if (type == RTAX_FEATURES)
986 fi_val &= ~DST_FEATURE_ECN_CA;
987
988 if (fi_val != val)
Xin Long5f9ae3d2017-08-23 10:07:26 +0800989 return false;
990 }
991
992 return true;
993}
994
David Ahern717a8f52019-04-05 16:30:32 -0700995static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
996 u32 table, struct netlink_ext_ack *extack)
997{
998 struct fib6_config cfg = {
999 .fc_table = table,
1000 .fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1001 .fc_ifindex = nh->fib_nh_oif,
1002 .fc_gateway = nh->fib_nh_gw6,
1003 };
1004 struct fib6_nh fib6_nh = {};
1005 int err;
1006
1007 err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1008 if (!err) {
1009 nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1010 dev_hold(nh->fib_nh_dev);
1011 nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1012 nh->fib_nh_scope = RT_SCOPE_LINK;
1013
1014 ipv6_stub->fib6_nh_release(&fib6_nh);
1015 }
1016
1017 return err;
1018}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019
1020/*
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00001021 * Picture
1022 * -------
1023 *
1024 * Semantics of nexthop is very messy by historical reasons.
1025 * We have to take into account, that:
1026 * a) gateway can be actually local interface address,
1027 * so that gatewayed route is direct.
1028 * b) gateway must be on-link address, possibly
1029 * described not by an ifaddr, but also by a direct route.
1030 * c) If both gateway and interface are specified, they should not
1031 * contradict.
1032 * d) If we use tunnel routes, gateway could be not on-link.
1033 *
1034 * Attempt to reconcile all of these (alas, self-contradictory) conditions
1035 * results in pretty ugly and hairy code with obscure logic.
1036 *
1037 * I chose to generalized it instead, so that the size
1038 * of code does not increase practically, but it becomes
1039 * much more general.
1040 * Every prefix is assigned a "scope" value: "host" is local address,
1041 * "link" is direct route,
1042 * [ ... "site" ... "interior" ... ]
1043 * and "universe" is true gateway route with global meaning.
1044 *
1045 * Every prefix refers to a set of "nexthop"s (gw, oif),
1046 * where gw must have narrower scope. This recursion stops
1047 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1048 * which means that gw is forced to be on link.
1049 *
1050 * Code is still hairy, but now it is apparently logically
1051 * consistent and very flexible. F.e. as by-product it allows
1052 * to co-exists in peace independent exterior and interior
1053 * routing processes.
1054 *
1055 * Normally it looks as following.
1056 *
1057 * {universe prefix} -> (gw, oif) [scope link]
1058 * |
1059 * |-> {link prefix} -> (gw, oif) [scope local]
1060 * |
1061 * |-> {local prefix} (terminal node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 */
David Ahern448d7242019-04-05 16:30:31 -07001063static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1064 u8 scope, struct netlink_ext_ack *extack)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065{
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00001066 struct net_device *dev;
David Ahern448d7242019-04-05 16:30:31 -07001067 struct fib_result res;
Enrico Weigeltc3fee642019-06-06 16:43:17 +02001068 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069
David Ahern448d7242019-04-05 16:30:31 -07001070 if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1071 unsigned int addr_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072
David Ahern448d7242019-04-05 16:30:31 -07001073 if (scope >= RT_SCOPE_LINK) {
1074 NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1075 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 }
David Ahern448d7242019-04-05 16:30:31 -07001077 dev = __dev_get_by_index(net, nh->fib_nh_oif);
1078 if (!dev) {
1079 NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1080 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 }
David Ahern448d7242019-04-05 16:30:31 -07001082 if (!(dev->flags & IFF_UP)) {
1083 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1084 return -ENETDOWN;
1085 }
1086 addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1087 if (addr_type != RTN_UNICAST) {
1088 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1089 return -EINVAL;
1090 }
1091 if (!netif_carrier_ok(dev))
1092 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1093 nh->fib_nh_dev = dev;
1094 dev_hold(dev);
1095 nh->fib_nh_scope = RT_SCOPE_LINK;
1096 return 0;
1097 }
1098 rcu_read_lock();
1099 {
1100 struct fib_table *tbl = NULL;
1101 struct flowi4 fl4 = {
1102 .daddr = nh->fib_nh_gw4,
1103 .flowi4_scope = scope + 1,
1104 .flowi4_oif = nh->fib_nh_oif,
1105 .flowi4_iif = LOOPBACK_IFINDEX,
1106 };
1107
1108 /* It is not necessary, but requires a bit of thinking */
1109 if (fl4.flowi4_scope < RT_SCOPE_LINK)
1110 fl4.flowi4_scope = RT_SCOPE_LINK;
1111
1112 if (table)
1113 tbl = fib_get_table(net, table);
1114
1115 if (tbl)
1116 err = fib_table_lookup(tbl, &fl4, &res,
1117 FIB_LOOKUP_IGNORE_LINKSTATE |
1118 FIB_LOOKUP_NOREF);
1119
1120 /* on error or if no table given do full lookup. This
1121 * is needed for example when nexthops are in the local
1122 * table rather than the given table
1123 */
1124 if (!tbl || err) {
1125 err = fib_lookup(net, &fl4, &res,
1126 FIB_LOOKUP_IGNORE_LINKSTATE);
1127 }
1128
1129 if (err) {
David Ahernc3ab2b42017-05-21 10:12:03 -06001130 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 goto out;
David Ahernc3ab2b42017-05-21 10:12:03 -06001132 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 }
David Ahern448d7242019-04-05 16:30:31 -07001134
1135 err = -EINVAL;
1136 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1137 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1138 goto out;
1139 }
1140 nh->fib_nh_scope = res.scope;
1141 nh->fib_nh_oif = FIB_RES_OIF(res);
1142 nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1143 if (!dev) {
1144 NL_SET_ERR_MSG(extack,
1145 "No egress device for nexthop gateway");
1146 goto out;
1147 }
1148 dev_hold(dev);
1149 if (!netif_carrier_ok(dev))
1150 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1151 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
Eric Dumazet8723e1b2010-10-19 00:39:26 +00001152out:
1153 rcu_read_unlock();
1154 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155}
1156
David Ahern448d7242019-04-05 16:30:31 -07001157static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1158 struct netlink_ext_ack *extack)
1159{
1160 struct in_device *in_dev;
1161 int err;
1162
1163 if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1164 NL_SET_ERR_MSG(extack,
1165 "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1166 return -EINVAL;
1167 }
1168
1169 rcu_read_lock();
1170
1171 err = -ENODEV;
1172 in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1173 if (!in_dev)
1174 goto out;
1175 err = -ENETDOWN;
1176 if (!(in_dev->dev->flags & IFF_UP)) {
1177 NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1178 goto out;
1179 }
1180
1181 nh->fib_nh_dev = in_dev->dev;
1182 dev_hold(nh->fib_nh_dev);
1183 nh->fib_nh_scope = RT_SCOPE_HOST;
1184 if (!netif_carrier_ok(nh->fib_nh_dev))
1185 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1186 err = 0;
1187out:
1188 rcu_read_unlock();
1189 return err;
1190}
1191
David Ahernac1fab22019-05-22 12:04:43 -07001192int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1193 struct netlink_ext_ack *extack)
David Ahern448d7242019-04-05 16:30:31 -07001194{
David Ahern448d7242019-04-05 16:30:31 -07001195 int err;
1196
1197 if (nh->fib_nh_gw_family == AF_INET)
David Ahernac1fab22019-05-22 12:04:43 -07001198 err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
David Ahern717a8f52019-04-05 16:30:32 -07001199 else if (nh->fib_nh_gw_family == AF_INET6)
1200 err = fib_check_nh_v6_gw(net, nh, table, extack);
David Ahern448d7242019-04-05 16:30:31 -07001201 else
1202 err = fib_check_nh_nongw(net, nh, extack);
1203
1204 return err;
1205}
1206
Al Viro81f7bf62006-09-27 18:40:00 -07001207static inline unsigned int fib_laddr_hashfn(__be32 val)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208{
David S. Miller123b9732011-02-01 15:34:21 -08001209 unsigned int mask = (fib_info_hash_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00001211 return ((__force u32)val ^
1212 ((__force u32)val >> 7) ^
1213 ((__force u32)val >> 14)) & mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214}
1215
David S. Miller123b9732011-02-01 15:34:21 -08001216static struct hlist_head *fib_info_hash_alloc(int bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217{
1218 if (bytes <= PAGE_SIZE)
Joonwoo Park88f83492007-11-26 23:29:32 +08001219 return kzalloc(bytes, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 else
1221 return (struct hlist_head *)
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00001222 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
1223 get_order(bytes));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224}
1225
David S. Miller123b9732011-02-01 15:34:21 -08001226static void fib_info_hash_free(struct hlist_head *hash, int bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227{
1228 if (!hash)
1229 return;
1230
1231 if (bytes <= PAGE_SIZE)
1232 kfree(hash);
1233 else
1234 free_pages((unsigned long) hash, get_order(bytes));
1235}
1236
David S. Miller123b9732011-02-01 15:34:21 -08001237static void fib_info_hash_move(struct hlist_head *new_info_hash,
1238 struct hlist_head *new_laddrhash,
1239 unsigned int new_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240{
David S. Millerb7656e72005-08-05 04:12:48 -07001241 struct hlist_head *old_info_hash, *old_laddrhash;
David S. Miller123b9732011-02-01 15:34:21 -08001242 unsigned int old_size = fib_info_hash_size;
David S. Millerb7656e72005-08-05 04:12:48 -07001243 unsigned int i, bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
Stephen Hemminger832b4c52006-08-29 16:48:09 -07001245 spin_lock_bh(&fib_info_lock);
David S. Millerb7656e72005-08-05 04:12:48 -07001246 old_info_hash = fib_info_hash;
1247 old_laddrhash = fib_info_laddrhash;
David S. Miller123b9732011-02-01 15:34:21 -08001248 fib_info_hash_size = new_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249
1250 for (i = 0; i < old_size; i++) {
1251 struct hlist_head *head = &fib_info_hash[i];
Sasha Levinb67bfe02013-02-27 17:06:00 -08001252 struct hlist_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 struct fib_info *fi;
1254
Sasha Levinb67bfe02013-02-27 17:06:00 -08001255 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 struct hlist_head *dest;
1257 unsigned int new_hash;
1258
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 new_hash = fib_info_hashfn(fi);
1260 dest = &new_info_hash[new_hash];
1261 hlist_add_head(&fi->fib_hash, dest);
1262 }
1263 }
1264 fib_info_hash = new_info_hash;
1265
1266 for (i = 0; i < old_size; i++) {
1267 struct hlist_head *lhead = &fib_info_laddrhash[i];
Sasha Levinb67bfe02013-02-27 17:06:00 -08001268 struct hlist_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 struct fib_info *fi;
1270
Sasha Levinb67bfe02013-02-27 17:06:00 -08001271 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 struct hlist_head *ldest;
1273 unsigned int new_hash;
1274
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1276 ldest = &new_laddrhash[new_hash];
1277 hlist_add_head(&fi->fib_lhash, ldest);
1278 }
1279 }
1280 fib_info_laddrhash = new_laddrhash;
1281
Stephen Hemminger832b4c52006-08-29 16:48:09 -07001282 spin_unlock_bh(&fib_info_lock);
David S. Millerb7656e72005-08-05 04:12:48 -07001283
1284 bytes = old_size * sizeof(struct hlist_head *);
David S. Miller123b9732011-02-01 15:34:21 -08001285 fib_info_hash_free(old_info_hash, bytes);
1286 fib_info_hash_free(old_laddrhash, bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287}
1288
David Aherndcb1ecb2019-06-03 20:19:50 -07001289__be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1290 unsigned char scope)
David S. Miller436c3b62011-03-24 17:42:21 -07001291{
David Aherndcb1ecb2019-06-03 20:19:50 -07001292 struct fib_nh *nh;
1293
1294 if (nhc->nhc_family != AF_INET)
1295 return inet_select_addr(nhc->nhc_dev, 0, scope);
1296
1297 nh = container_of(nhc, struct fib_nh, nh_common);
David Ahernc3669482019-05-22 12:04:45 -07001298 nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
David S. Miller436c3b62011-03-24 17:42:21 -07001299 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1300
1301 return nh->nh_saddr;
1302}
1303
David Aherneba618a2019-04-02 14:11:55 -07001304__be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1305{
1306 struct fib_nh_common *nhc = res->nhc;
David Aherneba618a2019-04-02 14:11:55 -07001307
1308 if (res->fi->fib_prefsrc)
1309 return res->fi->fib_prefsrc;
1310
David Aherndcb1ecb2019-06-03 20:19:50 -07001311 if (nhc->nhc_family == AF_INET) {
1312 struct fib_nh *nh;
David Aherneba618a2019-04-02 14:11:55 -07001313
David Aherndcb1ecb2019-06-03 20:19:50 -07001314 nh = container_of(nhc, struct fib_nh, nh_common);
1315 if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1316 return nh->nh_saddr;
1317 }
1318
1319 return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
David Aherneba618a2019-04-02 14:11:55 -07001320}
1321
David Ahern021dd3b2015-08-13 14:59:06 -06001322static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1323{
1324 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1325 fib_prefsrc != cfg->fc_dst) {
David Ahern9b8ff512015-09-01 14:26:35 -06001326 u32 tb_id = cfg->fc_table;
David Aherne1b8d902015-11-03 15:59:28 -08001327 int rc;
David Ahern021dd3b2015-08-13 14:59:06 -06001328
1329 if (tb_id == RT_TABLE_MAIN)
1330 tb_id = RT_TABLE_LOCAL;
1331
David Aherne1b8d902015-11-03 15:59:28 -08001332 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1333 fib_prefsrc, tb_id);
1334
1335 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1336 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1337 fib_prefsrc, RT_TABLE_LOCAL);
David Ahern021dd3b2015-08-13 14:59:06 -06001338 }
David Aherne1b8d902015-11-03 15:59:28 -08001339
1340 if (rc != RTN_LOCAL)
1341 return false;
David Ahern021dd3b2015-08-13 14:59:06 -06001342 }
1343 return true;
1344}
1345
David Ahern6d8422a12017-05-21 10:12:02 -06001346struct fib_info *fib_create_info(struct fib_config *cfg,
1347 struct netlink_ext_ack *extack)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348{
1349 int err;
1350 struct fib_info *fi = NULL;
David Ahern4c7e8082019-06-03 20:19:51 -07001351 struct nexthop *nh = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 struct fib_info *ofi;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 int nhs = 1;
Denis V. Lunev7462bd742008-01-31 18:49:32 -08001354 struct net *net = cfg->fc_nlinfo.nl_net;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
David S. Miller4c8237c2011-03-07 14:27:38 -08001356 if (cfg->fc_type > RTN_MAX)
1357 goto err_inval;
1358
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 /* Fast check to catch the most weird cases */
David Ahernc3ab2b42017-05-21 10:12:03 -06001360 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1361 NL_SET_ERR_MSG(extack, "Invalid scope");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 goto err_inval;
David Ahernc3ab2b42017-05-21 10:12:03 -06001363 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
David Ahernc3ab2b42017-05-21 10:12:03 -06001365 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1366 NL_SET_ERR_MSG(extack,
1367 "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
Julian Anastasov80610222016-07-10 21:11:55 +03001368 goto err_inval;
David Ahernc3ab2b42017-05-21 10:12:03 -06001369 }
Julian Anastasov80610222016-07-10 21:11:55 +03001370
David Ahern493ced12019-06-08 14:53:32 -07001371 if (cfg->fc_nh_id) {
David Ahern6c48ea52019-06-08 14:53:33 -07001372 if (!cfg->fc_mx) {
1373 fi = fib_find_info_nh(net, cfg);
1374 if (fi) {
1375 fi->fib_treeref++;
1376 return fi;
1377 }
1378 }
1379
David Ahern493ced12019-06-08 14:53:32 -07001380 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1381 if (!nh) {
1382 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1383 goto err_inval;
1384 }
1385 nhs = 0;
1386 }
1387
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388#ifdef CONFIG_IP_ROUTE_MULTIPATH
Thomas Graf4e902c52006-08-17 18:14:52 -07001389 if (cfg->fc_mp) {
David Ahern6d8422a12017-05-21 10:12:02 -06001390 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 if (nhs == 0)
1392 goto err_inval;
1393 }
1394#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395
1396 err = -ENOBUFS;
David S. Miller123b9732011-02-01 15:34:21 -08001397 if (fib_info_cnt >= fib_info_hash_size) {
1398 unsigned int new_size = fib_info_hash_size << 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 struct hlist_head *new_info_hash;
1400 struct hlist_head *new_laddrhash;
1401 unsigned int bytes;
1402
1403 if (!new_size)
Eric Dumazetd94ce9b2012-10-21 20:12:09 +00001404 new_size = 16;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 bytes = new_size * sizeof(struct hlist_head *);
David S. Miller123b9732011-02-01 15:34:21 -08001406 new_info_hash = fib_info_hash_alloc(bytes);
1407 new_laddrhash = fib_info_hash_alloc(bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 if (!new_info_hash || !new_laddrhash) {
David S. Miller123b9732011-02-01 15:34:21 -08001409 fib_info_hash_free(new_info_hash, bytes);
1410 fib_info_hash_free(new_laddrhash, bytes);
Joonwoo Park88f83492007-11-26 23:29:32 +08001411 } else
David S. Miller123b9732011-02-01 15:34:21 -08001412 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413
David S. Miller123b9732011-02-01 15:34:21 -08001414 if (!fib_info_hash_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 goto failure;
1416 }
1417
Gustavo A. R. Silva1f533ba2019-01-30 18:51:48 -06001418 fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
Ian Morris51456b22015-04-03 09:17:26 +01001419 if (!fi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 goto failure;
David Ahern767a2212018-10-04 20:07:51 -07001421 fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
David Ahernd7e774f2018-11-06 12:51:15 -08001422 cfg->fc_mx_len, extack);
Enrico Weigelt88e235b2019-06-05 23:09:05 +02001423 if (IS_ERR(fi->fib_metrics)) {
David Ahern767a2212018-10-04 20:07:51 -07001424 err = PTR_ERR(fi->fib_metrics);
1425 kfree(fi);
1426 return ERR_PTR(err);
Eric Dumazet187e5b32017-08-15 05:26:17 -07001427 }
David Ahern767a2212018-10-04 20:07:51 -07001428
Eric Dumazet187e5b32017-08-15 05:26:17 -07001429 fib_info_cnt++;
Eric W. Biedermanefd7ef12015-03-11 23:04:08 -05001430 fi->fib_net = net;
Thomas Graf4e902c52006-08-17 18:14:52 -07001431 fi->fib_protocol = cfg->fc_protocol;
David S. Miller37e826c2011-03-24 18:06:47 -07001432 fi->fib_scope = cfg->fc_scope;
Thomas Graf4e902c52006-08-17 18:14:52 -07001433 fi->fib_flags = cfg->fc_flags;
1434 fi->fib_priority = cfg->fc_priority;
1435 fi->fib_prefsrc = cfg->fc_prefsrc;
Eric Dumazetf4ef85b2012-10-04 01:25:26 +00001436 fi->fib_type = cfg->fc_type;
Mark Tomlinson5a56a0b2016-09-05 10:20:20 +12001437 fi->fib_tb_id = cfg->fc_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438
1439 fi->fib_nhs = nhs;
David Ahern4c7e8082019-06-03 20:19:51 -07001440 if (nh) {
1441 if (!nexthop_get(nh)) {
1442 NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1443 err = -EINVAL;
1444 } else {
1445 err = 0;
1446 fi->nh = nh;
1447 }
1448 } else {
1449 change_nexthops(fi) {
1450 nexthop_nh->nh_parent = fi;
1451 } endfor_nexthops(fi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
David Ahern4c7e8082019-06-03 20:19:51 -07001453 if (cfg->fc_mp)
1454 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1455 extack);
1456 else
1457 err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1458 }
Thomas Graf4e902c52006-08-17 18:14:52 -07001459
David Aherne4516ef2019-03-27 20:53:48 -07001460 if (err != 0)
1461 goto failure;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462
Thomas Graf4e902c52006-08-17 18:14:52 -07001463 if (fib_props[cfg->fc_type].error) {
David Ahernf35b7942019-04-05 16:30:28 -07001464 if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
David Ahernc3ab2b42017-05-21 10:12:03 -06001465 NL_SET_ERR_MSG(extack,
1466 "Gateway, device and multipath can not be specified for this route type");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 goto err_inval;
David Ahernc3ab2b42017-05-21 10:12:03 -06001468 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 goto link_it;
David S. Miller4c8237c2011-03-07 14:27:38 -08001470 } else {
1471 switch (cfg->fc_type) {
1472 case RTN_UNICAST:
1473 case RTN_LOCAL:
1474 case RTN_BROADCAST:
1475 case RTN_ANYCAST:
1476 case RTN_MULTICAST:
1477 break;
1478 default:
David Ahernc3ab2b42017-05-21 10:12:03 -06001479 NL_SET_ERR_MSG(extack, "Invalid route type");
David S. Miller4c8237c2011-03-07 14:27:38 -08001480 goto err_inval;
1481 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 }
1483
David Ahernc3ab2b42017-05-21 10:12:03 -06001484 if (cfg->fc_scope > RT_SCOPE_HOST) {
1485 NL_SET_ERR_MSG(extack, "Invalid scope");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 goto err_inval;
David Ahernc3ab2b42017-05-21 10:12:03 -06001487 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488
David Ahern4c7e8082019-06-03 20:19:51 -07001489 if (fi->nh) {
1490 err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1491 if (err)
1492 goto failure;
1493 } else if (cfg->fc_scope == RT_SCOPE_HOST) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 struct fib_nh *nh = fi->fib_nh;
1495
1496 /* Local address is added. */
David Ahernc3ab2b42017-05-21 10:12:03 -06001497 if (nhs != 1) {
1498 NL_SET_ERR_MSG(extack,
1499 "Route with host scope can not have multiple nexthops");
David Ahern6d8422a12017-05-21 10:12:02 -06001500 goto err_inval;
David Ahernc3ab2b42017-05-21 10:12:03 -06001501 }
David Ahernbdf00462019-04-05 16:30:26 -07001502 if (nh->fib_nh_gw_family) {
David Ahernc3ab2b42017-05-21 10:12:03 -06001503 NL_SET_ERR_MSG(extack,
1504 "Route with host scope can not have a gateway");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 goto err_inval;
David Ahernc3ab2b42017-05-21 10:12:03 -06001506 }
David Ahernb75ed8b2019-03-27 20:53:55 -07001507 nh->fib_nh_scope = RT_SCOPE_NOWHERE;
David Ahern5481d732019-06-03 20:19:49 -07001508 nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 err = -ENODEV;
David Ahernb75ed8b2019-03-27 20:53:55 -07001510 if (!nh->fib_nh_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 goto failure;
1512 } else {
Andy Gospodarek8a3d0312015-06-23 13:45:36 -04001513 int linkdown = 0;
1514
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 change_nexthops(fi) {
David Ahernac1fab22019-05-22 12:04:43 -07001516 err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1517 cfg->fc_table, cfg->fc_scope,
1518 extack);
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00001519 if (err != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 goto failure;
David Ahernb75ed8b2019-03-27 20:53:55 -07001521 if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
Andy Gospodarek8a3d0312015-06-23 13:45:36 -04001522 linkdown++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 } endfor_nexthops(fi)
Andy Gospodarek8a3d0312015-06-23 13:45:36 -04001524 if (linkdown == fi->fib_nhs)
1525 fi->fib_flags |= RTNH_F_LINKDOWN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 }
1527
David Ahernc3ab2b42017-05-21 10:12:03 -06001528 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1529 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
David Ahern021dd3b2015-08-13 14:59:06 -06001530 goto err_inval;
David Ahernc3ab2b42017-05-21 10:12:03 -06001531 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532
David Ahern4c7e8082019-06-03 20:19:51 -07001533 if (!fi->nh) {
1534 change_nexthops(fi) {
1535 fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1536 fi->fib_scope);
1537 if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1538 fi->fib_nh_is_v6 = true;
1539 } endfor_nexthops(fi)
David S. Miller1fc050a2011-03-07 20:54:48 -08001540
David Ahern4c7e8082019-06-03 20:19:51 -07001541 fib_rebalance(fi);
1542 }
Peter Nørlund0e884c72015-09-30 10:12:21 +02001543
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544link_it:
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00001545 ofi = fib_find_info(fi);
1546 if (ofi) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 fi->fib_dead = 1;
1548 free_fib_info(fi);
1549 ofi->fib_treeref++;
1550 return ofi;
1551 }
1552
1553 fi->fib_treeref++;
Reshetova, Elena0029c0d2017-07-04 09:35:02 +03001554 refcount_set(&fi->fib_clntref, 1);
Stephen Hemminger832b4c52006-08-29 16:48:09 -07001555 spin_lock_bh(&fib_info_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 hlist_add_head(&fi->fib_hash,
1557 &fib_info_hash[fib_info_hashfn(fi)]);
1558 if (fi->fib_prefsrc) {
1559 struct hlist_head *head;
1560
1561 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1562 hlist_add_head(&fi->fib_lhash, head);
1563 }
David Ahern4c7e8082019-06-03 20:19:51 -07001564 if (fi->nh) {
1565 list_add(&fi->nh_list, &nh->fi_list);
1566 } else {
1567 change_nexthops(fi) {
1568 struct hlist_head *head;
1569 unsigned int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570
David Ahern4c7e8082019-06-03 20:19:51 -07001571 if (!nexthop_nh->fib_nh_dev)
1572 continue;
1573 hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex);
1574 head = &fib_info_devhash[hash];
1575 hlist_add_head(&nexthop_nh->nh_hash, head);
1576 } endfor_nexthops(fi)
1577 }
Stephen Hemminger832b4c52006-08-29 16:48:09 -07001578 spin_unlock_bh(&fib_info_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 return fi;
1580
1581err_inval:
1582 err = -EINVAL;
1583
1584failure:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001585 if (fi) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 fi->fib_dead = 1;
1587 free_fib_info(fi);
1588 }
Thomas Graf4e902c52006-08-17 18:14:52 -07001589
1590 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591}
1592
David Ahernc0a72072019-04-02 14:11:58 -07001593int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
Donald Sharp7bdf4de2019-09-04 10:11:58 -04001594 u8 rt_family, unsigned char *flags, bool skip_oif)
David Ahernb0f60192019-04-02 14:11:56 -07001595{
David Ahernc2364192019-04-02 14:11:57 -07001596 if (nhc->nhc_flags & RTNH_F_DEAD)
David Ahernb0f60192019-04-02 14:11:56 -07001597 *flags |= RTNH_F_DEAD;
1598
David Ahernc2364192019-04-02 14:11:57 -07001599 if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
David Ahernb0f60192019-04-02 14:11:56 -07001600 *flags |= RTNH_F_LINKDOWN;
1601
1602 rcu_read_lock();
David Ahernc2364192019-04-02 14:11:57 -07001603 switch (nhc->nhc_family) {
1604 case AF_INET:
1605 if (ip_ignore_linkdown(nhc->nhc_dev))
1606 *flags |= RTNH_F_DEAD;
1607 break;
David Ahernc0a72072019-04-02 14:11:58 -07001608 case AF_INET6:
1609 if (ip6_ignore_linkdown(nhc->nhc_dev))
1610 *flags |= RTNH_F_DEAD;
1611 break;
David Ahernc2364192019-04-02 14:11:57 -07001612 }
David Ahernb0f60192019-04-02 14:11:56 -07001613 rcu_read_unlock();
1614 }
1615
David Ahernbdf00462019-04-05 16:30:26 -07001616 switch (nhc->nhc_gw_family) {
1617 case AF_INET:
1618 if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1619 goto nla_put_failure;
1620 break;
1621 case AF_INET6:
David Ahernd1566262019-04-05 16:30:40 -07001622 /* if gateway family does not match nexthop family
1623 * gateway is encoded as RTA_VIA
1624 */
Donald Sharp7bdf4de2019-09-04 10:11:58 -04001625 if (rt_family != nhc->nhc_gw_family) {
David Ahernd1566262019-04-05 16:30:40 -07001626 int alen = sizeof(struct in6_addr);
1627 struct nlattr *nla;
1628 struct rtvia *via;
1629
1630 nla = nla_reserve(skb, RTA_VIA, alen + 2);
1631 if (!nla)
1632 goto nla_put_failure;
1633
1634 via = nla_data(nla);
1635 via->rtvia_family = AF_INET6;
1636 memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1637 } else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1638 &nhc->nhc_gw.ipv6) < 0) {
David Ahernbdf00462019-04-05 16:30:26 -07001639 goto nla_put_failure;
David Ahernd1566262019-04-05 16:30:40 -07001640 }
David Ahernbdf00462019-04-05 16:30:26 -07001641 break;
David Ahernc2364192019-04-02 14:11:57 -07001642 }
David Ahernb0f60192019-04-02 14:11:56 -07001643
David Ahernc2364192019-04-02 14:11:57 -07001644 *flags |= (nhc->nhc_flags & RTNH_F_ONLINK);
1645 if (nhc->nhc_flags & RTNH_F_OFFLOAD)
David Ahernb0f60192019-04-02 14:11:56 -07001646 *flags |= RTNH_F_OFFLOAD;
1647
David Ahernc2364192019-04-02 14:11:57 -07001648 if (!skip_oif && nhc->nhc_dev &&
1649 nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
David Ahernb0f60192019-04-02 14:11:56 -07001650 goto nla_put_failure;
1651
David Ahernc2364192019-04-02 14:11:57 -07001652 if (nhc->nhc_lwtstate &&
David Ahernffa8ce52019-04-23 08:23:41 -07001653 lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1654 RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
David Ahernb0f60192019-04-02 14:11:56 -07001655 goto nla_put_failure;
1656
1657 return 0;
1658
1659nla_put_failure:
1660 return -EMSGSIZE;
1661}
David Ahernc0a72072019-04-02 14:11:58 -07001662EXPORT_SYMBOL_GPL(fib_nexthop_info);
David Ahernb0f60192019-04-02 14:11:56 -07001663
David Ahernc0a72072019-04-02 14:11:58 -07001664#if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1665int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
Donald Sharp7bdf4de2019-09-04 10:11:58 -04001666 int nh_weight, u8 rt_family)
David Ahernb0f60192019-04-02 14:11:56 -07001667{
David Ahernc2364192019-04-02 14:11:57 -07001668 const struct net_device *dev = nhc->nhc_dev;
David Ahernb0f60192019-04-02 14:11:56 -07001669 struct rtnexthop *rtnh;
David Ahernecc56632019-04-23 08:48:09 -07001670 unsigned char flags = 0;
David Ahernb0f60192019-04-02 14:11:56 -07001671
1672 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1673 if (!rtnh)
1674 goto nla_put_failure;
1675
David Ahernc2364192019-04-02 14:11:57 -07001676 rtnh->rtnh_hops = nh_weight - 1;
David Ahernb0f60192019-04-02 14:11:56 -07001677 rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1678
Donald Sharp7bdf4de2019-09-04 10:11:58 -04001679 if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
David Ahernb0f60192019-04-02 14:11:56 -07001680 goto nla_put_failure;
1681
1682 rtnh->rtnh_flags = flags;
1683
1684 /* length of rtnetlink header + attributes */
1685 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1686
1687 return 0;
1688
1689nla_put_failure:
1690 return -EMSGSIZE;
1691}
David Ahernc0a72072019-04-02 14:11:58 -07001692EXPORT_SYMBOL_GPL(fib_add_nexthop);
David Ahernc2364192019-04-02 14:11:57 -07001693#endif
David Ahernb0f60192019-04-02 14:11:56 -07001694
David Ahernc2364192019-04-02 14:11:57 -07001695#ifdef CONFIG_IP_ROUTE_MULTIPATH
David Ahernb0f60192019-04-02 14:11:56 -07001696static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1697{
1698 struct nlattr *mp;
1699
Michal Kubecekae0be8d2019-04-26 11:13:06 +02001700 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
David Ahernb0f60192019-04-02 14:11:56 -07001701 if (!mp)
1702 goto nla_put_failure;
1703
David Ahern4c7e8082019-06-03 20:19:51 -07001704 if (unlikely(fi->nh)) {
Donald Sharp7bdf4de2019-09-04 10:11:58 -04001705 if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
David Ahern4c7e8082019-06-03 20:19:51 -07001706 goto nla_put_failure;
1707 goto mp_end;
1708 }
1709
David Ahernb0f60192019-04-02 14:11:56 -07001710 for_nexthops(fi) {
Donald Sharp7bdf4de2019-09-04 10:11:58 -04001711 if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1712 AF_INET) < 0)
David Ahernb0f60192019-04-02 14:11:56 -07001713 goto nla_put_failure;
1714#ifdef CONFIG_IP_ROUTE_CLASSID
1715 if (nh->nh_tclassid &&
1716 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1717 goto nla_put_failure;
1718#endif
1719 } endfor_nexthops(fi);
1720
David Ahern4c7e8082019-06-03 20:19:51 -07001721mp_end:
David Ahernb0f60192019-04-02 14:11:56 -07001722 nla_nest_end(skb, mp);
1723
1724 return 0;
1725
1726nla_put_failure:
1727 return -EMSGSIZE;
1728}
1729#else
1730static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1731{
1732 return 0;
1733}
1734#endif
1735
Eric W. Biederman15e47302012-09-07 20:12:54 +00001736int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
Ido Schimmel1e301fd2020-01-14 13:23:10 +02001737 struct fib_rt_info *fri, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738{
Ido Schimmel1e301fd2020-01-14 13:23:10 +02001739 unsigned int nhs = fib_info_num_path(fri->fi);
1740 struct fib_info *fi = fri->fi;
1741 u32 tb_id = fri->tb_id;
Thomas Grafbe403ea2006-08-17 18:15:17 -07001742 struct nlmsghdr *nlh;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 struct rtmsg *rtm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744
Eric W. Biederman15e47302012-09-07 20:12:54 +00001745 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
Ian Morris51456b22015-04-03 09:17:26 +01001746 if (!nlh)
Patrick McHardy26932562007-01-31 23:16:40 -08001747 return -EMSGSIZE;
Thomas Grafbe403ea2006-08-17 18:15:17 -07001748
1749 rtm = nlmsg_data(nlh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 rtm->rtm_family = AF_INET;
Ido Schimmel1e301fd2020-01-14 13:23:10 +02001751 rtm->rtm_dst_len = fri->dst_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 rtm->rtm_src_len = 0;
Ido Schimmel1e301fd2020-01-14 13:23:10 +02001753 rtm->rtm_tos = fri->tos;
Krzysztof Piotr Oledzki709772e2008-06-10 15:44:49 -07001754 if (tb_id < 256)
1755 rtm->rtm_table = tb_id;
1756 else
1757 rtm->rtm_table = RT_TABLE_COMPAT;
David S. Millerf3756b72012-04-01 20:39:02 -04001758 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1759 goto nla_put_failure;
Ido Schimmel1e301fd2020-01-14 13:23:10 +02001760 rtm->rtm_type = fri->type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 rtm->rtm_flags = fi->fib_flags;
David S. Miller37e826c2011-03-24 18:06:47 -07001762 rtm->rtm_scope = fi->fib_scope;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 rtm->rtm_protocol = fi->fib_protocol;
Thomas Grafbe403ea2006-08-17 18:15:17 -07001764
David S. Millerf3756b72012-04-01 20:39:02 -04001765 if (rtm->rtm_dst_len &&
Ido Schimmel1e301fd2020-01-14 13:23:10 +02001766 nla_put_in_addr(skb, RTA_DST, fri->dst))
David S. Millerf3756b72012-04-01 20:39:02 -04001767 goto nla_put_failure;
1768 if (fi->fib_priority &&
1769 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1770 goto nla_put_failure;
Eric Dumazet3fb07da2017-05-25 14:27:35 -07001771 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
Thomas Grafbe403ea2006-08-17 18:15:17 -07001772 goto nla_put_failure;
1773
David S. Millerf3756b72012-04-01 20:39:02 -04001774 if (fi->fib_prefsrc &&
Jiri Benc930345e2015-03-29 16:59:25 +02001775 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
David S. Millerf3756b72012-04-01 20:39:02 -04001776 goto nla_put_failure;
David Ahern4c7e8082019-06-03 20:19:51 -07001777
1778 if (fi->nh) {
1779 if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1780 goto nla_put_failure;
1781 if (nexthop_is_blackhole(fi->nh))
1782 rtm->rtm_type = RTN_BLACKHOLE;
Roopa Prabhu4f801162020-04-27 13:56:46 -07001783 if (!fi->fib_net->ipv4.sysctl_nexthop_compat_mode)
1784 goto offload;
David Ahern4c7e8082019-06-03 20:19:51 -07001785 }
1786
David Ahern5481d732019-06-03 20:19:49 -07001787 if (nhs == 1) {
David Aherndcb1ecb2019-06-03 20:19:50 -07001788 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
David Ahernecc56632019-04-23 08:48:09 -07001789 unsigned char flags = 0;
David Ahernb0f60192019-04-02 14:11:56 -07001790
Donald Sharp7bdf4de2019-09-04 10:11:58 -04001791 if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
David S. Millerf3756b72012-04-01 20:39:02 -04001792 goto nla_put_failure;
David Ahernb0f60192019-04-02 14:11:56 -07001793
1794 rtm->rtm_flags = flags;
Patrick McHardyc7066f72011-01-14 13:36:42 +01001795#ifdef CONFIG_IP_ROUTE_CLASSID
David Aherndcb1ecb2019-06-03 20:19:50 -07001796 if (nhc->nhc_family == AF_INET) {
1797 struct fib_nh *nh;
1798
1799 nh = container_of(nhc, struct fib_nh, nh_common);
1800 if (nh->nh_tclassid &&
1801 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1802 goto nla_put_failure;
1803 }
Patrick McHardy8265abc2006-07-21 15:09:55 -07001804#endif
David Ahernb0f60192019-04-02 14:11:56 -07001805 } else {
1806 if (fib_add_multipath(skb, fi) < 0)
David Ahernea7a8082017-01-11 14:29:54 -08001807 goto nla_put_failure;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 }
Thomas Grafbe403ea2006-08-17 18:15:17 -07001809
Roopa Prabhu4f801162020-04-27 13:56:46 -07001810offload:
Ido Schimmel90b93f12020-01-14 13:23:11 +02001811 if (fri->offload)
1812 rtm->rtm_flags |= RTM_F_OFFLOAD;
1813 if (fri->trap)
1814 rtm->rtm_flags |= RTM_F_TRAP;
1815
Johannes Berg053c0952015-01-16 22:09:00 +01001816 nlmsg_end(skb, nlh);
1817 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818
Thomas Grafbe403ea2006-08-17 18:15:17 -07001819nla_put_failure:
Patrick McHardy26932562007-01-31 23:16:40 -08001820 nlmsg_cancel(skb, nlh);
1821 return -EMSGSIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822}
1823
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824/*
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00001825 * Update FIB if:
1826 * - local address disappeared -> we must delete all the entries
1827 * referring to it.
1828 * - device went down -> we must shutdown all nexthops going via it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 */
Mark Tomlinson5a56a0b2016-09-05 10:20:20 +12001830int fib_sync_down_addr(struct net_device *dev, __be32 local)
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001831{
1832 int ret = 0;
1833 unsigned int hash = fib_laddr_hashfn(local);
1834 struct hlist_head *head = &fib_info_laddrhash[hash];
David Aherne0a31262019-11-07 18:29:52 +00001835 int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
Mark Tomlinson5a56a0b2016-09-05 10:20:20 +12001836 struct net *net = dev_net(dev);
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001837 struct fib_info *fi;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838
Ian Morris51456b22015-04-03 09:17:26 +01001839 if (!fib_info_laddrhash || local == 0)
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001840 return 0;
1841
Sasha Levinb67bfe02013-02-27 17:06:00 -08001842 hlist_for_each_entry(fi, head, fib_lhash) {
Mark Tomlinson5a56a0b2016-09-05 10:20:20 +12001843 if (!net_eq(fi->fib_net, net) ||
1844 fi->fib_tb_id != tb_id)
Denis V. Lunev4814bdb2008-01-31 18:50:07 -08001845 continue;
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001846 if (fi->fib_prefsrc == local) {
1847 fi->fib_flags |= RTNH_F_DEAD;
1848 ret++;
1849 }
1850 }
1851 return ret;
1852}
1853
David Ahernb75ed8b2019-03-27 20:53:55 -07001854static int call_fib_nh_notifiers(struct fib_nh *nh,
Ido Schimmel982acb92017-02-08 11:16:39 +01001855 enum fib_event_type event_type)
1856{
David Ahernb75ed8b2019-03-27 20:53:55 -07001857 bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
Ido Schimmel982acb92017-02-08 11:16:39 +01001858 struct fib_nh_notifier_info info = {
David Ahernb75ed8b2019-03-27 20:53:55 -07001859 .fib_nh = nh,
Ido Schimmel982acb92017-02-08 11:16:39 +01001860 };
1861
1862 switch (event_type) {
1863 case FIB_EVENT_NH_ADD:
David Ahernb75ed8b2019-03-27 20:53:55 -07001864 if (nh->fib_nh_flags & RTNH_F_DEAD)
Ido Schimmel982acb92017-02-08 11:16:39 +01001865 break;
David Ahernb75ed8b2019-03-27 20:53:55 -07001866 if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
Ido Schimmel982acb92017-02-08 11:16:39 +01001867 break;
David Ahernb75ed8b2019-03-27 20:53:55 -07001868 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
Ido Schimmel04b1d4e2017-08-03 13:28:11 +02001869 &info.info);
Ido Schimmel982acb92017-02-08 11:16:39 +01001870 case FIB_EVENT_NH_DEL:
David Ahernb75ed8b2019-03-27 20:53:55 -07001871 if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1872 (nh->fib_nh_flags & RTNH_F_DEAD))
1873 return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
Ido Schimmel04b1d4e2017-08-03 13:28:11 +02001874 event_type, &info.info);
Ido Schimmel982acb92017-02-08 11:16:39 +01001875 default:
1876 break;
1877 }
1878
1879 return NOTIFY_DONE;
1880}
1881
Sabrina Dubrocaaf7d6cc2018-10-09 17:48:14 +02001882/* Update the PMTU of exceptions when:
1883 * - the new MTU of the first hop becomes smaller than the PMTU
1884 * - the old MTU was the same as the PMTU, and it limited discovery of
1885 * larger MTUs on the path. With that limit raised, we can now
1886 * discover larger MTUs
1887 * A special case is locked exceptions, for which the PMTU is smaller
1888 * than the minimal accepted PMTU:
1889 * - if the new MTU is greater than the PMTU, don't make any change
1890 * - otherwise, unlock and set PMTU
1891 */
David Ahern06c77c32019-05-22 12:04:46 -07001892void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
Sabrina Dubrocaaf7d6cc2018-10-09 17:48:14 +02001893{
1894 struct fnhe_hash_bucket *bucket;
1895 int i;
1896
David Aherna5995e72019-04-30 07:45:50 -07001897 bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
Sabrina Dubrocaaf7d6cc2018-10-09 17:48:14 +02001898 if (!bucket)
1899 return;
1900
1901 for (i = 0; i < FNHE_HASH_SIZE; i++) {
1902 struct fib_nh_exception *fnhe;
1903
1904 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1905 fnhe;
1906 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1907 if (fnhe->fnhe_mtu_locked) {
1908 if (new <= fnhe->fnhe_pmtu) {
1909 fnhe->fnhe_pmtu = new;
1910 fnhe->fnhe_mtu_locked = false;
1911 }
1912 } else if (new < fnhe->fnhe_pmtu ||
1913 orig == fnhe->fnhe_pmtu) {
1914 fnhe->fnhe_pmtu = new;
1915 }
1916 }
1917 }
1918}
1919
1920void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1921{
1922 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1923 struct hlist_head *head = &fib_info_devhash[hash];
1924 struct fib_nh *nh;
1925
1926 hlist_for_each_entry(nh, head, nh_hash) {
David Ahernb75ed8b2019-03-27 20:53:55 -07001927 if (nh->fib_nh_dev == dev)
David Ahern06c77c32019-05-22 12:04:46 -07001928 fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
Sabrina Dubrocaaf7d6cc2018-10-09 17:48:14 +02001929 }
1930}
1931
Julian Anastasov4f823de2015-10-30 10:23:33 +02001932/* Event force Flags Description
1933 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host
1934 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host
1935 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed
1936 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed
David Ahern4c7e8082019-06-03 20:19:51 -07001937 *
1938 * only used when fib_nh is built into fib_info
Julian Anastasov4f823de2015-10-30 10:23:33 +02001939 */
1940int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941{
1942 int ret = 0;
1943 int scope = RT_SCOPE_NOWHERE;
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001944 struct fib_info *prev_fi = NULL;
1945 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1946 struct hlist_head *head = &fib_info_devhash[hash];
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001947 struct fib_nh *nh;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001948
Julian Anastasov4f823de2015-10-30 10:23:33 +02001949 if (force)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 scope = -1;
1951
Sasha Levinb67bfe02013-02-27 17:06:00 -08001952 hlist_for_each_entry(nh, head, nh_hash) {
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001953 struct fib_info *fi = nh->nh_parent;
1954 int dead;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001956 BUG_ON(!fi->fib_nhs);
David Ahernb75ed8b2019-03-27 20:53:55 -07001957 if (nh->fib_nh_dev != dev || fi == prev_fi)
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001958 continue;
1959 prev_fi = fi;
1960 dead = 0;
1961 change_nexthops(fi) {
David Ahernb75ed8b2019-03-27 20:53:55 -07001962 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001963 dead++;
David Ahernb75ed8b2019-03-27 20:53:55 -07001964 else if (nexthop_nh->fib_nh_dev == dev &&
1965 nexthop_nh->fib_nh_scope != scope) {
Andy Gospodarek8a3d0312015-06-23 13:45:36 -04001966 switch (event) {
1967 case NETDEV_DOWN:
1968 case NETDEV_UNREGISTER:
David Ahernb75ed8b2019-03-27 20:53:55 -07001969 nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
Joe Perchesa8eceea2020-03-12 15:50:22 -07001970 fallthrough;
Andy Gospodarek8a3d0312015-06-23 13:45:36 -04001971 case NETDEV_CHANGE:
David Ahernb75ed8b2019-03-27 20:53:55 -07001972 nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
Andy Gospodarek8a3d0312015-06-23 13:45:36 -04001973 break;
1974 }
Ido Schimmel982acb92017-02-08 11:16:39 +01001975 call_fib_nh_notifiers(nexthop_nh,
1976 FIB_EVENT_NH_DEL);
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001977 dead++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 }
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001979#ifdef CONFIG_IP_ROUTE_MULTIPATH
Andy Gospodarek8a3d0312015-06-23 13:45:36 -04001980 if (event == NETDEV_UNREGISTER &&
David Ahernb75ed8b2019-03-27 20:53:55 -07001981 nexthop_nh->fib_nh_dev == dev) {
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001982 dead = fi->fib_nhs;
1983 break;
1984 }
1985#endif
1986 } endfor_nexthops(fi)
1987 if (dead == fi->fib_nhs) {
Andy Gospodarek8a3d0312015-06-23 13:45:36 -04001988 switch (event) {
1989 case NETDEV_DOWN:
1990 case NETDEV_UNREGISTER:
1991 fi->fib_flags |= RTNH_F_DEAD;
Joe Perchesa8eceea2020-03-12 15:50:22 -07001992 fallthrough;
Andy Gospodarek8a3d0312015-06-23 13:45:36 -04001993 case NETDEV_CHANGE:
1994 fi->fib_flags |= RTNH_F_LINKDOWN;
1995 break;
1996 }
Denis V. Lunev85326fa2008-01-31 18:48:47 -08001997 ret++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 }
Peter Nørlund0e884c72015-09-30 10:12:21 +02001999
2000 fib_rebalance(fi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 }
2002
2003 return ret;
2004}
2005
David S. Miller0c838ff2011-01-31 16:16:50 -08002006/* Must be invoked inside of an RCU protected region. */
David Ahernc7b371e2017-01-05 19:33:59 -08002007static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
David S. Miller0c838ff2011-01-31 16:16:50 -08002008{
2009 struct fib_info *fi = NULL, *last_resort = NULL;
Alexander Duyck56315f92015-02-25 15:31:31 -08002010 struct hlist_head *fa_head = res->fa_head;
David S. Miller0c838ff2011-01-31 16:16:50 -08002011 struct fib_table *tb = res->table;
Julian Anastasov18a912e92015-07-22 10:43:22 +03002012 u8 slen = 32 - res->prefixlen;
David S. Miller0c838ff2011-01-31 16:16:50 -08002013 int order = -1, last_idx = -1;
Julian Anastasov2392deb2015-07-22 10:43:23 +03002014 struct fib_alias *fa, *fa1 = NULL;
2015 u32 last_prio = res->fi->fib_priority;
2016 u8 last_tos = 0;
David S. Miller0c838ff2011-01-31 16:16:50 -08002017
Alexander Duyck56315f92015-02-25 15:31:31 -08002018 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
David S. Miller0c838ff2011-01-31 16:16:50 -08002019 struct fib_info *next_fi = fa->fa_info;
David Ahern7c74b0b2020-04-22 15:40:20 -06002020 struct fib_nh_common *nhc;
David S. Miller0c838ff2011-01-31 16:16:50 -08002021
Julian Anastasov18a912e92015-07-22 10:43:22 +03002022 if (fa->fa_slen != slen)
2023 continue;
Julian Anastasov2392deb2015-07-22 10:43:23 +03002024 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
2025 continue;
Julian Anastasov18a912e92015-07-22 10:43:22 +03002026 if (fa->tb_id != tb->tb_id)
2027 continue;
Julian Anastasov2392deb2015-07-22 10:43:23 +03002028 if (next_fi->fib_priority > last_prio &&
2029 fa->fa_tos == last_tos) {
2030 if (last_tos)
2031 continue;
2032 break;
2033 }
2034 if (next_fi->fib_flags & RTNH_F_DEAD)
2035 continue;
2036 last_tos = fa->fa_tos;
2037 last_prio = next_fi->fib_priority;
2038
David S. Miller37e826c2011-03-24 18:06:47 -07002039 if (next_fi->fib_scope != res->scope ||
David S. Miller0c838ff2011-01-31 16:16:50 -08002040 fa->fa_type != RTN_UNICAST)
2041 continue;
David Ahern5481d732019-06-03 20:19:49 -07002042
David Ahern7c74b0b2020-04-22 15:40:20 -06002043 nhc = fib_info_nhc(next_fi, 0);
2044 if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
David S. Miller0c838ff2011-01-31 16:16:50 -08002045 continue;
2046
2047 fib_alias_accessed(fa);
2048
Ian Morris51456b22015-04-03 09:17:26 +01002049 if (!fi) {
David S. Miller0c838ff2011-01-31 16:16:50 -08002050 if (next_fi != res->fi)
2051 break;
Julian Anastasov2392deb2015-07-22 10:43:23 +03002052 fa1 = fa;
David S. Miller0c838ff2011-01-31 16:16:50 -08002053 } else if (!fib_detect_death(fi, order, &last_resort,
Julian Anastasov2392deb2015-07-22 10:43:23 +03002054 &last_idx, fa1->fa_default)) {
David S. Miller0c838ff2011-01-31 16:16:50 -08002055 fib_result_assign(res, fi);
Julian Anastasov2392deb2015-07-22 10:43:23 +03002056 fa1->fa_default = order;
David S. Miller0c838ff2011-01-31 16:16:50 -08002057 goto out;
2058 }
2059 fi = next_fi;
2060 order++;
2061 }
2062
Ian Morris51456b22015-04-03 09:17:26 +01002063 if (order <= 0 || !fi) {
Julian Anastasov2392deb2015-07-22 10:43:23 +03002064 if (fa1)
2065 fa1->fa_default = -1;
David S. Miller0c838ff2011-01-31 16:16:50 -08002066 goto out;
2067 }
2068
2069 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
Julian Anastasov2392deb2015-07-22 10:43:23 +03002070 fa1->fa_default)) {
David S. Miller0c838ff2011-01-31 16:16:50 -08002071 fib_result_assign(res, fi);
Julian Anastasov2392deb2015-07-22 10:43:23 +03002072 fa1->fa_default = order;
David S. Miller0c838ff2011-01-31 16:16:50 -08002073 goto out;
2074 }
2075
2076 if (last_idx >= 0)
2077 fib_result_assign(res, last_resort);
Julian Anastasov2392deb2015-07-22 10:43:23 +03002078 fa1->fa_default = last_idx;
David S. Miller0c838ff2011-01-31 16:16:50 -08002079out:
Eric Dumazet31d40932011-02-14 11:23:04 -08002080 return;
David S. Miller0c838ff2011-01-31 16:16:50 -08002081}
2082
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083/*
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00002084 * Dead device goes up. We wake up dead nexthops.
2085 * It takes sense only on multipath routes.
David Ahern4c7e8082019-06-03 20:19:51 -07002086 *
2087 * only used when fib_nh is built into fib_info
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 */
David Ahernecc56632019-04-23 08:48:09 -07002089int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090{
2091 struct fib_info *prev_fi;
2092 unsigned int hash;
2093 struct hlist_head *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 struct fib_nh *nh;
2095 int ret;
2096
Eric Dumazet6a31d2a2010-10-04 20:00:18 +00002097 if (!(dev->flags & IFF_UP))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 return 0;
2099
Julian Anastasovc9b32922015-10-30 10:23:34 +02002100 if (nh_flags & RTNH_F_DEAD) {
2101 unsigned int flags = dev_get_flags(dev);
2102
2103 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2104 nh_flags |= RTNH_F_LINKDOWN;
2105 }
2106
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 prev_fi = NULL;
2108 hash = fib_devindex_hashfn(dev->ifindex);
2109 head = &fib_info_devhash[hash];
2110 ret = 0;
2111
Sasha Levinb67bfe02013-02-27 17:06:00 -08002112 hlist_for_each_entry(nh, head, nh_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 struct fib_info *fi = nh->nh_parent;
2114 int alive;
2115
2116 BUG_ON(!fi->fib_nhs);
David Ahernb75ed8b2019-03-27 20:53:55 -07002117 if (nh->fib_nh_dev != dev || fi == prev_fi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 continue;
2119
2120 prev_fi = fi;
2121 alive = 0;
2122 change_nexthops(fi) {
David Ahernb75ed8b2019-03-27 20:53:55 -07002123 if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 alive++;
2125 continue;
2126 }
David Ahernb75ed8b2019-03-27 20:53:55 -07002127 if (!nexthop_nh->fib_nh_dev ||
2128 !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 continue;
David Ahernb75ed8b2019-03-27 20:53:55 -07002130 if (nexthop_nh->fib_nh_dev != dev ||
David S. Miller71fceff2010-01-15 01:16:40 -08002131 !__in_dev_get_rtnl(dev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 continue;
2133 alive++;
David Ahernb75ed8b2019-03-27 20:53:55 -07002134 nexthop_nh->fib_nh_flags &= ~nh_flags;
Ido Schimmel982acb92017-02-08 11:16:39 +01002135 call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 } endfor_nexthops(fi)
2137
2138 if (alive > 0) {
Andy Gospodarek8a3d0312015-06-23 13:45:36 -04002139 fi->fib_flags &= ~nh_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 ret++;
2141 }
Peter Nørlund0e884c72015-09-30 10:12:21 +02002142
2143 fib_rebalance(fi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 }
2145
2146 return ret;
2147}
2148
Andy Gospodarek8a3d0312015-06-23 13:45:36 -04002149#ifdef CONFIG_IP_ROUTE_MULTIPATH
David Aherna6db4492016-04-07 07:21:00 -07002150static bool fib_good_nh(const struct fib_nh *nh)
2151{
2152 int state = NUD_REACHABLE;
2153
David Ahernb75ed8b2019-03-27 20:53:55 -07002154 if (nh->fib_nh_scope == RT_SCOPE_LINK) {
David Aherna6db4492016-04-07 07:21:00 -07002155 struct neighbour *n;
2156
2157 rcu_read_lock_bh();
2158
David Ahern1a38c432019-04-05 16:30:38 -07002159 if (likely(nh->fib_nh_gw_family == AF_INET))
2160 n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2161 (__force u32)nh->fib_nh_gw4);
2162 else if (nh->fib_nh_gw_family == AF_INET6)
2163 n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2164 &nh->fib_nh_gw6);
2165 else
2166 n = NULL;
David Aherna6db4492016-04-07 07:21:00 -07002167 if (n)
2168 state = n->nud_state;
2169
2170 rcu_read_unlock_bh();
2171 }
2172
2173 return !!(state & NUD_VALID);
2174}
Andy Gospodarek8a3d0312015-06-23 13:45:36 -04002175
Peter Nørlund0e884c72015-09-30 10:12:21 +02002176void fib_select_multipath(struct fib_result *res, int hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177{
2178 struct fib_info *fi = res->fi;
David Aherna6db4492016-04-07 07:21:00 -07002179 struct net *net = fi->fib_net;
2180 bool first = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181
David Ahern4c7e8082019-06-03 20:19:51 -07002182 if (unlikely(res->fi->nh)) {
2183 nexthop_path_fib_result(res, hash);
2184 return;
2185 }
2186
David Aherneba618a2019-04-02 14:11:55 -07002187 change_nexthops(fi) {
Xin Long6174a302018-04-01 22:40:35 +08002188 if (net->ipv4.sysctl_fib_multipath_use_neigh) {
David Aherneba618a2019-04-02 14:11:55 -07002189 if (!fib_good_nh(nexthop_nh))
Xin Long6174a302018-04-01 22:40:35 +08002190 continue;
2191 if (!first) {
2192 res->nh_sel = nhsel;
David Aherneba618a2019-04-02 14:11:55 -07002193 res->nhc = &nexthop_nh->nh_common;
Xin Long6174a302018-04-01 22:40:35 +08002194 first = true;
2195 }
2196 }
2197
David Aherneba618a2019-04-02 14:11:55 -07002198 if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
Peter Nørlund0e884c72015-09-30 10:12:21 +02002199 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200
Xin Long6174a302018-04-01 22:40:35 +08002201 res->nh_sel = nhsel;
David Aherneba618a2019-04-02 14:11:55 -07002202 res->nhc = &nexthop_nh->nh_common;
Xin Long6174a302018-04-01 22:40:35 +08002203 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 } endfor_nexthops(fi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205}
2206#endif
David Ahern3ce58d82015-10-05 08:51:25 -07002207
2208void fib_select_path(struct net *net, struct fib_result *res,
Nikolay Aleksandrovbf4e0a32017-03-16 15:28:00 +02002209 struct flowi4 *fl4, const struct sk_buff *skb)
David Ahern3ce58d82015-10-05 08:51:25 -07002210{
David Ahern0d876f22018-02-13 08:11:34 -08002211 if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2212 goto check_saddr;
David Ahern7a18c5b2017-01-10 14:37:35 -08002213
David Ahern3ce58d82015-10-05 08:51:25 -07002214#ifdef CONFIG_IP_ROUTE_MULTIPATH
David Ahern5481d732019-06-03 20:19:49 -07002215 if (fib_info_num_path(res->fi) > 1) {
David Ahern7efc0b62018-03-02 08:32:12 -08002216 int h = fib_multipath_hash(net, fl4, skb, NULL);
Paolo Abeni9920e482015-10-29 22:20:40 +01002217
Nikolay Aleksandrovbf4e0a32017-03-16 15:28:00 +02002218 fib_select_multipath(res, h);
David Ahern3ce58d82015-10-05 08:51:25 -07002219 }
2220 else
2221#endif
2222 if (!res->prefixlen &&
2223 res->table->tb_num_default > 1 &&
David Ahern0d876f22018-02-13 08:11:34 -08002224 res->type == RTN_UNICAST)
David Ahern3ce58d82015-10-05 08:51:25 -07002225 fib_select_default(fl4, res);
2226
David Ahern0d876f22018-02-13 08:11:34 -08002227check_saddr:
David Ahern3ce58d82015-10-05 08:51:25 -07002228 if (!fl4->saddr)
David Aherneba618a2019-04-02 14:11:55 -07002229 fl4->saddr = fib_result_prefsrc(net, res);
David Ahern3ce58d82015-10-05 08:51:25 -07002230}