Loading...
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * IPv4 Forwarding Information Base: semantics.
7 *
8 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
14 */
15
16#include <linux/uaccess.h>
17#include <linux/bitops.h>
18#include <linux/types.h>
19#include <linux/kernel.h>
20#include <linux/jiffies.h>
21#include <linux/mm.h>
22#include <linux/string.h>
23#include <linux/socket.h>
24#include <linux/sockios.h>
25#include <linux/errno.h>
26#include <linux/in.h>
27#include <linux/inet.h>
28#include <linux/inetdevice.h>
29#include <linux/netdevice.h>
30#include <linux/if_arp.h>
31#include <linux/proc_fs.h>
32#include <linux/skbuff.h>
33#include <linux/init.h>
34#include <linux/slab.h>
35#include <linux/netlink.h>
36
37#include <net/arp.h>
38#include <net/ip.h>
39#include <net/protocol.h>
40#include <net/route.h>
41#include <net/tcp.h>
42#include <net/sock.h>
43#include <net/ip_fib.h>
44#include <net/netlink.h>
45#include <net/nexthop.h>
46#include <net/lwtunnel.h>
47#include <net/fib_notifier.h>
48
49#include "fib_lookup.h"
50
51static DEFINE_SPINLOCK(fib_info_lock);
52static struct hlist_head *fib_info_hash;
53static struct hlist_head *fib_info_laddrhash;
54static unsigned int fib_info_hash_size;
55static unsigned int fib_info_cnt;
56
57#define DEVINDEX_HASHBITS 8
58#define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
59static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
60
61#ifdef CONFIG_IP_ROUTE_MULTIPATH
62
63#define for_nexthops(fi) { \
64 int nhsel; const struct fib_nh *nh; \
65 for (nhsel = 0, nh = (fi)->fib_nh; \
66 nhsel < (fi)->fib_nhs; \
67 nh++, nhsel++)
68
69#define change_nexthops(fi) { \
70 int nhsel; struct fib_nh *nexthop_nh; \
71 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
72 nhsel < (fi)->fib_nhs; \
73 nexthop_nh++, nhsel++)
74
75#else /* CONFIG_IP_ROUTE_MULTIPATH */
76
77/* Hope, that gcc will optimize it to get rid of dummy loop */
78
79#define for_nexthops(fi) { \
80 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \
81 for (nhsel = 0; nhsel < 1; nhsel++)
82
83#define change_nexthops(fi) { \
84 int nhsel; \
85 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
86 for (nhsel = 0; nhsel < 1; nhsel++)
87
88#endif /* CONFIG_IP_ROUTE_MULTIPATH */
89
90#define endfor_nexthops(fi) }
91
92
93const struct fib_prop fib_props[RTN_MAX + 1] = {
94 [RTN_UNSPEC] = {
95 .error = 0,
96 .scope = RT_SCOPE_NOWHERE,
97 },
98 [RTN_UNICAST] = {
99 .error = 0,
100 .scope = RT_SCOPE_UNIVERSE,
101 },
102 [RTN_LOCAL] = {
103 .error = 0,
104 .scope = RT_SCOPE_HOST,
105 },
106 [RTN_BROADCAST] = {
107 .error = 0,
108 .scope = RT_SCOPE_LINK,
109 },
110 [RTN_ANYCAST] = {
111 .error = 0,
112 .scope = RT_SCOPE_LINK,
113 },
114 [RTN_MULTICAST] = {
115 .error = 0,
116 .scope = RT_SCOPE_UNIVERSE,
117 },
118 [RTN_BLACKHOLE] = {
119 .error = -EINVAL,
120 .scope = RT_SCOPE_UNIVERSE,
121 },
122 [RTN_UNREACHABLE] = {
123 .error = -EHOSTUNREACH,
124 .scope = RT_SCOPE_UNIVERSE,
125 },
126 [RTN_PROHIBIT] = {
127 .error = -EACCES,
128 .scope = RT_SCOPE_UNIVERSE,
129 },
130 [RTN_THROW] = {
131 .error = -EAGAIN,
132 .scope = RT_SCOPE_UNIVERSE,
133 },
134 [RTN_NAT] = {
135 .error = -EINVAL,
136 .scope = RT_SCOPE_NOWHERE,
137 },
138 [RTN_XRESOLVE] = {
139 .error = -EINVAL,
140 .scope = RT_SCOPE_NOWHERE,
141 },
142};
143
144static void rt_fibinfo_free(struct rtable __rcu **rtp)
145{
146 struct rtable *rt = rcu_dereference_protected(*rtp, 1);
147
148 if (!rt)
149 return;
150
151 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
152 * because we waited an RCU grace period before calling
153 * free_fib_info_rcu()
154 */
155
156 dst_dev_put(&rt->dst);
157 dst_release_immediate(&rt->dst);
158}
159
160static void free_nh_exceptions(struct fib_nh *nh)
161{
162 struct fnhe_hash_bucket *hash;
163 int i;
164
165 hash = rcu_dereference_protected(nh->nh_exceptions, 1);
166 if (!hash)
167 return;
168 for (i = 0; i < FNHE_HASH_SIZE; i++) {
169 struct fib_nh_exception *fnhe;
170
171 fnhe = rcu_dereference_protected(hash[i].chain, 1);
172 while (fnhe) {
173 struct fib_nh_exception *next;
174
175 next = rcu_dereference_protected(fnhe->fnhe_next, 1);
176
177 rt_fibinfo_free(&fnhe->fnhe_rth_input);
178 rt_fibinfo_free(&fnhe->fnhe_rth_output);
179
180 kfree(fnhe);
181
182 fnhe = next;
183 }
184 }
185 kfree(hash);
186}
187
188static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
189{
190 int cpu;
191
192 if (!rtp)
193 return;
194
195 for_each_possible_cpu(cpu) {
196 struct rtable *rt;
197
198 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
199 if (rt) {
200 dst_dev_put(&rt->dst);
201 dst_release_immediate(&rt->dst);
202 }
203 }
204 free_percpu(rtp);
205}
206
207/* Release a nexthop info record */
208static void free_fib_info_rcu(struct rcu_head *head)
209{
210 struct fib_info *fi = container_of(head, struct fib_info, rcu);
211 struct dst_metrics *m;
212
213 change_nexthops(fi) {
214 if (nexthop_nh->nh_dev)
215 dev_put(nexthop_nh->nh_dev);
216 lwtstate_put(nexthop_nh->nh_lwtstate);
217 free_nh_exceptions(nexthop_nh);
218 rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
219 rt_fibinfo_free(&nexthop_nh->nh_rth_input);
220 } endfor_nexthops(fi);
221
222 m = fi->fib_metrics;
223 if (m != &dst_default_metrics && refcount_dec_and_test(&m->refcnt))
224 kfree(m);
225 kfree(fi);
226}
227
228void free_fib_info(struct fib_info *fi)
229{
230 if (fi->fib_dead == 0) {
231 pr_warn("Freeing alive fib_info %p\n", fi);
232 return;
233 }
234 fib_info_cnt--;
235#ifdef CONFIG_IP_ROUTE_CLASSID
236 change_nexthops(fi) {
237 if (nexthop_nh->nh_tclassid)
238 fi->fib_net->ipv4.fib_num_tclassid_users--;
239 } endfor_nexthops(fi);
240#endif
241 call_rcu(&fi->rcu, free_fib_info_rcu);
242}
243EXPORT_SYMBOL_GPL(free_fib_info);
244
245void fib_release_info(struct fib_info *fi)
246{
247 spin_lock_bh(&fib_info_lock);
248 if (fi && --fi->fib_treeref == 0) {
249 hlist_del(&fi->fib_hash);
250 if (fi->fib_prefsrc)
251 hlist_del(&fi->fib_lhash);
252 change_nexthops(fi) {
253 if (!nexthop_nh->nh_dev)
254 continue;
255 hlist_del(&nexthop_nh->nh_hash);
256 } endfor_nexthops(fi)
257 fi->fib_dead = 1;
258 fib_info_put(fi);
259 }
260 spin_unlock_bh(&fib_info_lock);
261}
262
263static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
264{
265 const struct fib_nh *onh = ofi->fib_nh;
266
267 for_nexthops(fi) {
268 if (nh->nh_oif != onh->nh_oif ||
269 nh->nh_gw != onh->nh_gw ||
270 nh->nh_scope != onh->nh_scope ||
271#ifdef CONFIG_IP_ROUTE_MULTIPATH
272 nh->nh_weight != onh->nh_weight ||
273#endif
274#ifdef CONFIG_IP_ROUTE_CLASSID
275 nh->nh_tclassid != onh->nh_tclassid ||
276#endif
277 lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) ||
278 ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
279 return -1;
280 onh++;
281 } endfor_nexthops(fi);
282 return 0;
283}
284
285static inline unsigned int fib_devindex_hashfn(unsigned int val)
286{
287 unsigned int mask = DEVINDEX_HASHSIZE - 1;
288
289 return (val ^
290 (val >> DEVINDEX_HASHBITS) ^
291 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
292}
293
294static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
295{
296 unsigned int mask = (fib_info_hash_size - 1);
297 unsigned int val = fi->fib_nhs;
298
299 val ^= (fi->fib_protocol << 8) | fi->fib_scope;
300 val ^= (__force u32)fi->fib_prefsrc;
301 val ^= fi->fib_priority;
302 for_nexthops(fi) {
303 val ^= fib_devindex_hashfn(nh->nh_oif);
304 } endfor_nexthops(fi)
305
306 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
307}
308
309static struct fib_info *fib_find_info(const struct fib_info *nfi)
310{
311 struct hlist_head *head;
312 struct fib_info *fi;
313 unsigned int hash;
314
315 hash = fib_info_hashfn(nfi);
316 head = &fib_info_hash[hash];
317
318 hlist_for_each_entry(fi, head, fib_hash) {
319 if (!net_eq(fi->fib_net, nfi->fib_net))
320 continue;
321 if (fi->fib_nhs != nfi->fib_nhs)
322 continue;
323 if (nfi->fib_protocol == fi->fib_protocol &&
324 nfi->fib_scope == fi->fib_scope &&
325 nfi->fib_prefsrc == fi->fib_prefsrc &&
326 nfi->fib_priority == fi->fib_priority &&
327 nfi->fib_type == fi->fib_type &&
328 memcmp(nfi->fib_metrics, fi->fib_metrics,
329 sizeof(u32) * RTAX_MAX) == 0 &&
330 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
331 (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
332 return fi;
333 }
334
335 return NULL;
336}
337
338/* Check, that the gateway is already configured.
339 * Used only by redirect accept routine.
340 */
341int ip_fib_check_default(__be32 gw, struct net_device *dev)
342{
343 struct hlist_head *head;
344 struct fib_nh *nh;
345 unsigned int hash;
346
347 spin_lock(&fib_info_lock);
348
349 hash = fib_devindex_hashfn(dev->ifindex);
350 head = &fib_info_devhash[hash];
351 hlist_for_each_entry(nh, head, nh_hash) {
352 if (nh->nh_dev == dev &&
353 nh->nh_gw == gw &&
354 !(nh->nh_flags & RTNH_F_DEAD)) {
355 spin_unlock(&fib_info_lock);
356 return 0;
357 }
358 }
359
360 spin_unlock(&fib_info_lock);
361
362 return -1;
363}
364
365static inline size_t fib_nlmsg_size(struct fib_info *fi)
366{
367 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
368 + nla_total_size(4) /* RTA_TABLE */
369 + nla_total_size(4) /* RTA_DST */
370 + nla_total_size(4) /* RTA_PRIORITY */
371 + nla_total_size(4) /* RTA_PREFSRC */
372 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
373
374 /* space for nested metrics */
375 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
376
377 if (fi->fib_nhs) {
378 size_t nh_encapsize = 0;
379 /* Also handles the special case fib_nhs == 1 */
380
381 /* each nexthop is packed in an attribute */
382 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
383
384 /* may contain flow and gateway attribute */
385 nhsize += 2 * nla_total_size(4);
386
387 /* grab encap info */
388 for_nexthops(fi) {
389 if (nh->nh_lwtstate) {
390 /* RTA_ENCAP_TYPE */
391 nh_encapsize += lwtunnel_get_encap_size(
392 nh->nh_lwtstate);
393 /* RTA_ENCAP */
394 nh_encapsize += nla_total_size(2);
395 }
396 } endfor_nexthops(fi);
397
398 /* all nexthops are packed in a nested attribute */
399 payload += nla_total_size((fi->fib_nhs * nhsize) +
400 nh_encapsize);
401
402 }
403
404 return payload;
405}
406
407void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
408 int dst_len, u32 tb_id, const struct nl_info *info,
409 unsigned int nlm_flags)
410{
411 struct sk_buff *skb;
412 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
413 int err = -ENOBUFS;
414
415 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
416 if (!skb)
417 goto errout;
418
419 err = fib_dump_info(skb, info->portid, seq, event, tb_id,
420 fa->fa_type, key, dst_len,
421 fa->fa_tos, fa->fa_info, nlm_flags);
422 if (err < 0) {
423 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
424 WARN_ON(err == -EMSGSIZE);
425 kfree_skb(skb);
426 goto errout;
427 }
428 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
429 info->nlh, GFP_KERNEL);
430 return;
431errout:
432 if (err < 0)
433 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
434}
435
436static int fib_detect_death(struct fib_info *fi, int order,
437 struct fib_info **last_resort, int *last_idx,
438 int dflt)
439{
440 struct neighbour *n;
441 int state = NUD_NONE;
442
443 n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
444 if (n) {
445 state = n->nud_state;
446 neigh_release(n);
447 } else {
448 return 0;
449 }
450 if (state == NUD_REACHABLE)
451 return 0;
452 if ((state & NUD_VALID) && order != dflt)
453 return 0;
454 if ((state & NUD_VALID) ||
455 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
456 *last_resort = fi;
457 *last_idx = order;
458 }
459 return 1;
460}
461
462#ifdef CONFIG_IP_ROUTE_MULTIPATH
463
464static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
465 struct netlink_ext_ack *extack)
466{
467 int nhs = 0;
468
469 while (rtnh_ok(rtnh, remaining)) {
470 nhs++;
471 rtnh = rtnh_next(rtnh, &remaining);
472 }
473
474 /* leftover implies invalid nexthop configuration, discard it */
475 if (remaining > 0) {
476 NL_SET_ERR_MSG(extack,
477 "Invalid nexthop configuration - extra data after nexthops");
478 nhs = 0;
479 }
480
481 return nhs;
482}
483
484static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
485 int remaining, struct fib_config *cfg,
486 struct netlink_ext_ack *extack)
487{
488 int ret;
489
490 change_nexthops(fi) {
491 int attrlen;
492
493 if (!rtnh_ok(rtnh, remaining)) {
494 NL_SET_ERR_MSG(extack,
495 "Invalid nexthop configuration - extra data after nexthop");
496 return -EINVAL;
497 }
498
499 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
500 NL_SET_ERR_MSG(extack,
501 "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
502 return -EINVAL;
503 }
504
505 nexthop_nh->nh_flags =
506 (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
507 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
508 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
509
510 attrlen = rtnh_attrlen(rtnh);
511 if (attrlen > 0) {
512 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
513
514 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
515 nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
516#ifdef CONFIG_IP_ROUTE_CLASSID
517 nla = nla_find(attrs, attrlen, RTA_FLOW);
518 nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
519 if (nexthop_nh->nh_tclassid)
520 fi->fib_net->ipv4.fib_num_tclassid_users++;
521#endif
522 nla = nla_find(attrs, attrlen, RTA_ENCAP);
523 if (nla) {
524 struct lwtunnel_state *lwtstate;
525 struct nlattr *nla_entype;
526
527 nla_entype = nla_find(attrs, attrlen,
528 RTA_ENCAP_TYPE);
529 if (!nla_entype) {
530 NL_SET_BAD_ATTR(extack, nla);
531 NL_SET_ERR_MSG(extack,
532 "Encap type is missing");
533 goto err_inval;
534 }
535
536 ret = lwtunnel_build_state(nla_get_u16(
537 nla_entype),
538 nla, AF_INET, cfg,
539 &lwtstate, extack);
540 if (ret)
541 goto errout;
542 nexthop_nh->nh_lwtstate =
543 lwtstate_get(lwtstate);
544 }
545 }
546
547 rtnh = rtnh_next(rtnh, &remaining);
548 } endfor_nexthops(fi);
549
550 return 0;
551
552err_inval:
553 ret = -EINVAL;
554
555errout:
556 return ret;
557}
558
559static void fib_rebalance(struct fib_info *fi)
560{
561 int total;
562 int w;
563 struct in_device *in_dev;
564
565 if (fi->fib_nhs < 2)
566 return;
567
568 total = 0;
569 for_nexthops(fi) {
570 if (nh->nh_flags & RTNH_F_DEAD)
571 continue;
572
573 in_dev = __in_dev_get_rtnl(nh->nh_dev);
574
575 if (in_dev &&
576 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
577 nh->nh_flags & RTNH_F_LINKDOWN)
578 continue;
579
580 total += nh->nh_weight;
581 } endfor_nexthops(fi);
582
583 w = 0;
584 change_nexthops(fi) {
585 int upper_bound;
586
587 in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev);
588
589 if (nexthop_nh->nh_flags & RTNH_F_DEAD) {
590 upper_bound = -1;
591 } else if (in_dev &&
592 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
593 nexthop_nh->nh_flags & RTNH_F_LINKDOWN) {
594 upper_bound = -1;
595 } else {
596 w += nexthop_nh->nh_weight;
597 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
598 total) - 1;
599 }
600
601 atomic_set(&nexthop_nh->nh_upper_bound, upper_bound);
602 } endfor_nexthops(fi);
603}
604#else /* CONFIG_IP_ROUTE_MULTIPATH */
605
606#define fib_rebalance(fi) do { } while (0)
607
608#endif /* CONFIG_IP_ROUTE_MULTIPATH */
609
610static int fib_encap_match(u16 encap_type,
611 struct nlattr *encap,
612 const struct fib_nh *nh,
613 const struct fib_config *cfg,
614 struct netlink_ext_ack *extack)
615{
616 struct lwtunnel_state *lwtstate;
617 int ret, result = 0;
618
619 if (encap_type == LWTUNNEL_ENCAP_NONE)
620 return 0;
621
622 ret = lwtunnel_build_state(encap_type, encap, AF_INET,
623 cfg, &lwtstate, extack);
624 if (!ret) {
625 result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
626 lwtstate_free(lwtstate);
627 }
628
629 return result;
630}
631
632int fib_nh_match(struct fib_config *cfg, struct fib_info *fi,
633 struct netlink_ext_ack *extack)
634{
635#ifdef CONFIG_IP_ROUTE_MULTIPATH
636 struct rtnexthop *rtnh;
637 int remaining;
638#endif
639
640 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
641 return 1;
642
643 if (cfg->fc_oif || cfg->fc_gw) {
644 if (cfg->fc_encap) {
645 if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap,
646 fi->fib_nh, cfg, extack))
647 return 1;
648 }
649#ifdef CONFIG_IP_ROUTE_CLASSID
650 if (cfg->fc_flow &&
651 cfg->fc_flow != fi->fib_nh->nh_tclassid)
652 return 1;
653#endif
654 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
655 (!cfg->fc_gw || cfg->fc_gw == fi->fib_nh->nh_gw))
656 return 0;
657 return 1;
658 }
659
660#ifdef CONFIG_IP_ROUTE_MULTIPATH
661 if (!cfg->fc_mp)
662 return 0;
663
664 rtnh = cfg->fc_mp;
665 remaining = cfg->fc_mp_len;
666
667 for_nexthops(fi) {
668 int attrlen;
669
670 if (!rtnh_ok(rtnh, remaining))
671 return -EINVAL;
672
673 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
674 return 1;
675
676 attrlen = rtnh_attrlen(rtnh);
677 if (attrlen > 0) {
678 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
679
680 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
681 if (nla && nla_get_in_addr(nla) != nh->nh_gw)
682 return 1;
683#ifdef CONFIG_IP_ROUTE_CLASSID
684 nla = nla_find(attrs, attrlen, RTA_FLOW);
685 if (nla && nla_get_u32(nla) != nh->nh_tclassid)
686 return 1;
687#endif
688 }
689
690 rtnh = rtnh_next(rtnh, &remaining);
691 } endfor_nexthops(fi);
692#endif
693 return 0;
694}
695
696bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
697{
698 struct nlattr *nla;
699 int remaining;
700
701 if (!cfg->fc_mx)
702 return true;
703
704 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
705 int type = nla_type(nla);
706 u32 fi_val, val;
707
708 if (!type)
709 continue;
710 if (type > RTAX_MAX)
711 return false;
712
713 if (type == RTAX_CC_ALGO) {
714 char tmp[TCP_CA_NAME_MAX];
715 bool ecn_ca = false;
716
717 nla_strlcpy(tmp, nla, sizeof(tmp));
718 val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
719 } else {
720 val = nla_get_u32(nla);
721 }
722
723 fi_val = fi->fib_metrics->metrics[type - 1];
724 if (type == RTAX_FEATURES)
725 fi_val &= ~DST_FEATURE_ECN_CA;
726
727 if (fi_val != val)
728 return false;
729 }
730
731 return true;
732}
733
734
735/*
736 * Picture
737 * -------
738 *
739 * Semantics of nexthop is very messy by historical reasons.
740 * We have to take into account, that:
741 * a) gateway can be actually local interface address,
742 * so that gatewayed route is direct.
743 * b) gateway must be on-link address, possibly
744 * described not by an ifaddr, but also by a direct route.
745 * c) If both gateway and interface are specified, they should not
746 * contradict.
747 * d) If we use tunnel routes, gateway could be not on-link.
748 *
749 * Attempt to reconcile all of these (alas, self-contradictory) conditions
750 * results in pretty ugly and hairy code with obscure logic.
751 *
752 * I chose to generalized it instead, so that the size
753 * of code does not increase practically, but it becomes
754 * much more general.
755 * Every prefix is assigned a "scope" value: "host" is local address,
756 * "link" is direct route,
757 * [ ... "site" ... "interior" ... ]
758 * and "universe" is true gateway route with global meaning.
759 *
760 * Every prefix refers to a set of "nexthop"s (gw, oif),
761 * where gw must have narrower scope. This recursion stops
762 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
763 * which means that gw is forced to be on link.
764 *
765 * Code is still hairy, but now it is apparently logically
766 * consistent and very flexible. F.e. as by-product it allows
767 * to co-exists in peace independent exterior and interior
768 * routing processes.
769 *
770 * Normally it looks as following.
771 *
772 * {universe prefix} -> (gw, oif) [scope link]
773 * |
774 * |-> {link prefix} -> (gw, oif) [scope local]
775 * |
776 * |-> {local prefix} (terminal node)
777 */
778static int fib_check_nh(struct fib_config *cfg, struct fib_nh *nh,
779 struct netlink_ext_ack *extack)
780{
781 int err = 0;
782 struct net *net;
783 struct net_device *dev;
784
785 net = cfg->fc_nlinfo.nl_net;
786 if (nh->nh_gw) {
787 struct fib_result res;
788
789 if (nh->nh_flags & RTNH_F_ONLINK) {
790 unsigned int addr_type;
791
792 if (cfg->fc_scope >= RT_SCOPE_LINK) {
793 NL_SET_ERR_MSG(extack,
794 "Nexthop has invalid scope");
795 return -EINVAL;
796 }
797 dev = __dev_get_by_index(net, nh->nh_oif);
798 if (!dev)
799 return -ENODEV;
800 if (!(dev->flags & IFF_UP)) {
801 NL_SET_ERR_MSG(extack,
802 "Nexthop device is not up");
803 return -ENETDOWN;
804 }
805 addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
806 if (addr_type != RTN_UNICAST) {
807 NL_SET_ERR_MSG(extack,
808 "Nexthop has invalid gateway");
809 return -EINVAL;
810 }
811 if (!netif_carrier_ok(dev))
812 nh->nh_flags |= RTNH_F_LINKDOWN;
813 nh->nh_dev = dev;
814 dev_hold(dev);
815 nh->nh_scope = RT_SCOPE_LINK;
816 return 0;
817 }
818 rcu_read_lock();
819 {
820 struct fib_table *tbl = NULL;
821 struct flowi4 fl4 = {
822 .daddr = nh->nh_gw,
823 .flowi4_scope = cfg->fc_scope + 1,
824 .flowi4_oif = nh->nh_oif,
825 .flowi4_iif = LOOPBACK_IFINDEX,
826 };
827
828 /* It is not necessary, but requires a bit of thinking */
829 if (fl4.flowi4_scope < RT_SCOPE_LINK)
830 fl4.flowi4_scope = RT_SCOPE_LINK;
831
832 if (cfg->fc_table)
833 tbl = fib_get_table(net, cfg->fc_table);
834
835 if (tbl)
836 err = fib_table_lookup(tbl, &fl4, &res,
837 FIB_LOOKUP_IGNORE_LINKSTATE |
838 FIB_LOOKUP_NOREF);
839
840 /* on error or if no table given do full lookup. This
841 * is needed for example when nexthops are in the local
842 * table rather than the given table
843 */
844 if (!tbl || err) {
845 err = fib_lookup(net, &fl4, &res,
846 FIB_LOOKUP_IGNORE_LINKSTATE);
847 }
848
849 if (err) {
850 NL_SET_ERR_MSG(extack,
851 "Nexthop has invalid gateway");
852 rcu_read_unlock();
853 return err;
854 }
855 }
856 err = -EINVAL;
857 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
858 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
859 goto out;
860 }
861 nh->nh_scope = res.scope;
862 nh->nh_oif = FIB_RES_OIF(res);
863 nh->nh_dev = dev = FIB_RES_DEV(res);
864 if (!dev) {
865 NL_SET_ERR_MSG(extack,
866 "No egress device for nexthop gateway");
867 goto out;
868 }
869 dev_hold(dev);
870 if (!netif_carrier_ok(dev))
871 nh->nh_flags |= RTNH_F_LINKDOWN;
872 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
873 } else {
874 struct in_device *in_dev;
875
876 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
877 NL_SET_ERR_MSG(extack,
878 "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
879 return -EINVAL;
880 }
881 rcu_read_lock();
882 err = -ENODEV;
883 in_dev = inetdev_by_index(net, nh->nh_oif);
884 if (!in_dev)
885 goto out;
886 err = -ENETDOWN;
887 if (!(in_dev->dev->flags & IFF_UP)) {
888 NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
889 goto out;
890 }
891 nh->nh_dev = in_dev->dev;
892 dev_hold(nh->nh_dev);
893 nh->nh_scope = RT_SCOPE_HOST;
894 if (!netif_carrier_ok(nh->nh_dev))
895 nh->nh_flags |= RTNH_F_LINKDOWN;
896 err = 0;
897 }
898out:
899 rcu_read_unlock();
900 return err;
901}
902
903static inline unsigned int fib_laddr_hashfn(__be32 val)
904{
905 unsigned int mask = (fib_info_hash_size - 1);
906
907 return ((__force u32)val ^
908 ((__force u32)val >> 7) ^
909 ((__force u32)val >> 14)) & mask;
910}
911
912static struct hlist_head *fib_info_hash_alloc(int bytes)
913{
914 if (bytes <= PAGE_SIZE)
915 return kzalloc(bytes, GFP_KERNEL);
916 else
917 return (struct hlist_head *)
918 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
919 get_order(bytes));
920}
921
922static void fib_info_hash_free(struct hlist_head *hash, int bytes)
923{
924 if (!hash)
925 return;
926
927 if (bytes <= PAGE_SIZE)
928 kfree(hash);
929 else
930 free_pages((unsigned long) hash, get_order(bytes));
931}
932
933static void fib_info_hash_move(struct hlist_head *new_info_hash,
934 struct hlist_head *new_laddrhash,
935 unsigned int new_size)
936{
937 struct hlist_head *old_info_hash, *old_laddrhash;
938 unsigned int old_size = fib_info_hash_size;
939 unsigned int i, bytes;
940
941 spin_lock_bh(&fib_info_lock);
942 old_info_hash = fib_info_hash;
943 old_laddrhash = fib_info_laddrhash;
944 fib_info_hash_size = new_size;
945
946 for (i = 0; i < old_size; i++) {
947 struct hlist_head *head = &fib_info_hash[i];
948 struct hlist_node *n;
949 struct fib_info *fi;
950
951 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
952 struct hlist_head *dest;
953 unsigned int new_hash;
954
955 new_hash = fib_info_hashfn(fi);
956 dest = &new_info_hash[new_hash];
957 hlist_add_head(&fi->fib_hash, dest);
958 }
959 }
960 fib_info_hash = new_info_hash;
961
962 for (i = 0; i < old_size; i++) {
963 struct hlist_head *lhead = &fib_info_laddrhash[i];
964 struct hlist_node *n;
965 struct fib_info *fi;
966
967 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
968 struct hlist_head *ldest;
969 unsigned int new_hash;
970
971 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
972 ldest = &new_laddrhash[new_hash];
973 hlist_add_head(&fi->fib_lhash, ldest);
974 }
975 }
976 fib_info_laddrhash = new_laddrhash;
977
978 spin_unlock_bh(&fib_info_lock);
979
980 bytes = old_size * sizeof(struct hlist_head *);
981 fib_info_hash_free(old_info_hash, bytes);
982 fib_info_hash_free(old_laddrhash, bytes);
983}
984
985__be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
986{
987 nh->nh_saddr = inet_select_addr(nh->nh_dev,
988 nh->nh_gw,
989 nh->nh_parent->fib_scope);
990 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
991
992 return nh->nh_saddr;
993}
994
995static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
996{
997 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
998 fib_prefsrc != cfg->fc_dst) {
999 u32 tb_id = cfg->fc_table;
1000 int rc;
1001
1002 if (tb_id == RT_TABLE_MAIN)
1003 tb_id = RT_TABLE_LOCAL;
1004
1005 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1006 fib_prefsrc, tb_id);
1007
1008 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1009 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1010 fib_prefsrc, RT_TABLE_LOCAL);
1011 }
1012
1013 if (rc != RTN_LOCAL)
1014 return false;
1015 }
1016 return true;
1017}
1018
1019static int
1020fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
1021{
1022 bool ecn_ca = false;
1023 struct nlattr *nla;
1024 int remaining;
1025
1026 if (!cfg->fc_mx)
1027 return 0;
1028
1029 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1030 int type = nla_type(nla);
1031 u32 val;
1032
1033 if (!type)
1034 continue;
1035 if (type > RTAX_MAX)
1036 return -EINVAL;
1037
1038 if (type == RTAX_CC_ALGO) {
1039 char tmp[TCP_CA_NAME_MAX];
1040
1041 nla_strlcpy(tmp, nla, sizeof(tmp));
1042 val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
1043 if (val == TCP_CA_UNSPEC)
1044 return -EINVAL;
1045 } else {
1046 val = nla_get_u32(nla);
1047 }
1048 if (type == RTAX_ADVMSS && val > 65535 - 40)
1049 val = 65535 - 40;
1050 if (type == RTAX_MTU && val > 65535 - 15)
1051 val = 65535 - 15;
1052 if (type == RTAX_HOPLIMIT && val > 255)
1053 val = 255;
1054 if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
1055 return -EINVAL;
1056 fi->fib_metrics->metrics[type - 1] = val;
1057 }
1058
1059 if (ecn_ca)
1060 fi->fib_metrics->metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
1061
1062 return 0;
1063}
1064
1065struct fib_info *fib_create_info(struct fib_config *cfg,
1066 struct netlink_ext_ack *extack)
1067{
1068 int err;
1069 struct fib_info *fi = NULL;
1070 struct fib_info *ofi;
1071 int nhs = 1;
1072 struct net *net = cfg->fc_nlinfo.nl_net;
1073
1074 if (cfg->fc_type > RTN_MAX)
1075 goto err_inval;
1076
1077 /* Fast check to catch the most weird cases */
1078 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1079 NL_SET_ERR_MSG(extack, "Invalid scope");
1080 goto err_inval;
1081 }
1082
1083 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1084 NL_SET_ERR_MSG(extack,
1085 "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1086 goto err_inval;
1087 }
1088
1089#ifdef CONFIG_IP_ROUTE_MULTIPATH
1090 if (cfg->fc_mp) {
1091 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1092 if (nhs == 0)
1093 goto err_inval;
1094 }
1095#endif
1096
1097 err = -ENOBUFS;
1098 if (fib_info_cnt >= fib_info_hash_size) {
1099 unsigned int new_size = fib_info_hash_size << 1;
1100 struct hlist_head *new_info_hash;
1101 struct hlist_head *new_laddrhash;
1102 unsigned int bytes;
1103
1104 if (!new_size)
1105 new_size = 16;
1106 bytes = new_size * sizeof(struct hlist_head *);
1107 new_info_hash = fib_info_hash_alloc(bytes);
1108 new_laddrhash = fib_info_hash_alloc(bytes);
1109 if (!new_info_hash || !new_laddrhash) {
1110 fib_info_hash_free(new_info_hash, bytes);
1111 fib_info_hash_free(new_laddrhash, bytes);
1112 } else
1113 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1114
1115 if (!fib_info_hash_size)
1116 goto failure;
1117 }
1118
1119 fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1120 if (!fi)
1121 goto failure;
1122 if (cfg->fc_mx) {
1123 fi->fib_metrics = kzalloc(sizeof(*fi->fib_metrics), GFP_KERNEL);
1124 if (unlikely(!fi->fib_metrics)) {
1125 kfree(fi);
1126 return ERR_PTR(err);
1127 }
1128 refcount_set(&fi->fib_metrics->refcnt, 1);
1129 } else {
1130 fi->fib_metrics = (struct dst_metrics *)&dst_default_metrics;
1131 }
1132 fib_info_cnt++;
1133 fi->fib_net = net;
1134 fi->fib_protocol = cfg->fc_protocol;
1135 fi->fib_scope = cfg->fc_scope;
1136 fi->fib_flags = cfg->fc_flags;
1137 fi->fib_priority = cfg->fc_priority;
1138 fi->fib_prefsrc = cfg->fc_prefsrc;
1139 fi->fib_type = cfg->fc_type;
1140 fi->fib_tb_id = cfg->fc_table;
1141
1142 fi->fib_nhs = nhs;
1143 change_nexthops(fi) {
1144 nexthop_nh->nh_parent = fi;
1145 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1146 if (!nexthop_nh->nh_pcpu_rth_output)
1147 goto failure;
1148 } endfor_nexthops(fi)
1149
1150 err = fib_convert_metrics(fi, cfg);
1151 if (err)
1152 goto failure;
1153
1154 if (cfg->fc_mp) {
1155#ifdef CONFIG_IP_ROUTE_MULTIPATH
1156 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, extack);
1157 if (err != 0)
1158 goto failure;
1159 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif) {
1160 NL_SET_ERR_MSG(extack,
1161 "Nexthop device index does not match RTA_OIF");
1162 goto err_inval;
1163 }
1164 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw) {
1165 NL_SET_ERR_MSG(extack,
1166 "Nexthop gateway does not match RTA_GATEWAY");
1167 goto err_inval;
1168 }
1169#ifdef CONFIG_IP_ROUTE_CLASSID
1170 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow) {
1171 NL_SET_ERR_MSG(extack,
1172 "Nexthop class id does not match RTA_FLOW");
1173 goto err_inval;
1174 }
1175#endif
1176#else
1177 NL_SET_ERR_MSG(extack,
1178 "Multipath support not enabled in kernel");
1179 goto err_inval;
1180#endif
1181 } else {
1182 struct fib_nh *nh = fi->fib_nh;
1183
1184 if (cfg->fc_encap) {
1185 struct lwtunnel_state *lwtstate;
1186
1187 if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE) {
1188 NL_SET_ERR_MSG(extack,
1189 "LWT encap type not specified");
1190 goto err_inval;
1191 }
1192 err = lwtunnel_build_state(cfg->fc_encap_type,
1193 cfg->fc_encap, AF_INET, cfg,
1194 &lwtstate, extack);
1195 if (err)
1196 goto failure;
1197
1198 nh->nh_lwtstate = lwtstate_get(lwtstate);
1199 }
1200 nh->nh_oif = cfg->fc_oif;
1201 nh->nh_gw = cfg->fc_gw;
1202 nh->nh_flags = cfg->fc_flags;
1203#ifdef CONFIG_IP_ROUTE_CLASSID
1204 nh->nh_tclassid = cfg->fc_flow;
1205 if (nh->nh_tclassid)
1206 fi->fib_net->ipv4.fib_num_tclassid_users++;
1207#endif
1208#ifdef CONFIG_IP_ROUTE_MULTIPATH
1209 nh->nh_weight = 1;
1210#endif
1211 }
1212
1213 if (fib_props[cfg->fc_type].error) {
1214 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp) {
1215 NL_SET_ERR_MSG(extack,
1216 "Gateway, device and multipath can not be specified for this route type");
1217 goto err_inval;
1218 }
1219 goto link_it;
1220 } else {
1221 switch (cfg->fc_type) {
1222 case RTN_UNICAST:
1223 case RTN_LOCAL:
1224 case RTN_BROADCAST:
1225 case RTN_ANYCAST:
1226 case RTN_MULTICAST:
1227 break;
1228 default:
1229 NL_SET_ERR_MSG(extack, "Invalid route type");
1230 goto err_inval;
1231 }
1232 }
1233
1234 if (cfg->fc_scope > RT_SCOPE_HOST) {
1235 NL_SET_ERR_MSG(extack, "Invalid scope");
1236 goto err_inval;
1237 }
1238
1239 if (cfg->fc_scope == RT_SCOPE_HOST) {
1240 struct fib_nh *nh = fi->fib_nh;
1241
1242 /* Local address is added. */
1243 if (nhs != 1) {
1244 NL_SET_ERR_MSG(extack,
1245 "Route with host scope can not have multiple nexthops");
1246 goto err_inval;
1247 }
1248 if (nh->nh_gw) {
1249 NL_SET_ERR_MSG(extack,
1250 "Route with host scope can not have a gateway");
1251 goto err_inval;
1252 }
1253 nh->nh_scope = RT_SCOPE_NOWHERE;
1254 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1255 err = -ENODEV;
1256 if (!nh->nh_dev)
1257 goto failure;
1258 } else {
1259 int linkdown = 0;
1260
1261 change_nexthops(fi) {
1262 err = fib_check_nh(cfg, nexthop_nh, extack);
1263 if (err != 0)
1264 goto failure;
1265 if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1266 linkdown++;
1267 } endfor_nexthops(fi)
1268 if (linkdown == fi->fib_nhs)
1269 fi->fib_flags |= RTNH_F_LINKDOWN;
1270 }
1271
1272 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1273 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1274 goto err_inval;
1275 }
1276
1277 change_nexthops(fi) {
1278 fib_info_update_nh_saddr(net, nexthop_nh);
1279 } endfor_nexthops(fi)
1280
1281 fib_rebalance(fi);
1282
1283link_it:
1284 ofi = fib_find_info(fi);
1285 if (ofi) {
1286 fi->fib_dead = 1;
1287 free_fib_info(fi);
1288 ofi->fib_treeref++;
1289 return ofi;
1290 }
1291
1292 fi->fib_treeref++;
1293 refcount_set(&fi->fib_clntref, 1);
1294 spin_lock_bh(&fib_info_lock);
1295 hlist_add_head(&fi->fib_hash,
1296 &fib_info_hash[fib_info_hashfn(fi)]);
1297 if (fi->fib_prefsrc) {
1298 struct hlist_head *head;
1299
1300 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1301 hlist_add_head(&fi->fib_lhash, head);
1302 }
1303 change_nexthops(fi) {
1304 struct hlist_head *head;
1305 unsigned int hash;
1306
1307 if (!nexthop_nh->nh_dev)
1308 continue;
1309 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1310 head = &fib_info_devhash[hash];
1311 hlist_add_head(&nexthop_nh->nh_hash, head);
1312 } endfor_nexthops(fi)
1313 spin_unlock_bh(&fib_info_lock);
1314 return fi;
1315
1316err_inval:
1317 err = -EINVAL;
1318
1319failure:
1320 if (fi) {
1321 fi->fib_dead = 1;
1322 free_fib_info(fi);
1323 }
1324
1325 return ERR_PTR(err);
1326}
1327
1328int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1329 u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1330 struct fib_info *fi, unsigned int flags)
1331{
1332 struct nlmsghdr *nlh;
1333 struct rtmsg *rtm;
1334
1335 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1336 if (!nlh)
1337 return -EMSGSIZE;
1338
1339 rtm = nlmsg_data(nlh);
1340 rtm->rtm_family = AF_INET;
1341 rtm->rtm_dst_len = dst_len;
1342 rtm->rtm_src_len = 0;
1343 rtm->rtm_tos = tos;
1344 if (tb_id < 256)
1345 rtm->rtm_table = tb_id;
1346 else
1347 rtm->rtm_table = RT_TABLE_COMPAT;
1348 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1349 goto nla_put_failure;
1350 rtm->rtm_type = type;
1351 rtm->rtm_flags = fi->fib_flags;
1352 rtm->rtm_scope = fi->fib_scope;
1353 rtm->rtm_protocol = fi->fib_protocol;
1354
1355 if (rtm->rtm_dst_len &&
1356 nla_put_in_addr(skb, RTA_DST, dst))
1357 goto nla_put_failure;
1358 if (fi->fib_priority &&
1359 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1360 goto nla_put_failure;
1361 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1362 goto nla_put_failure;
1363
1364 if (fi->fib_prefsrc &&
1365 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1366 goto nla_put_failure;
1367 if (fi->fib_nhs == 1) {
1368 if (fi->fib_nh->nh_gw &&
1369 nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1370 goto nla_put_failure;
1371 if (fi->fib_nh->nh_oif &&
1372 nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1373 goto nla_put_failure;
1374 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1375 struct in_device *in_dev;
1376
1377 rcu_read_lock();
1378 in_dev = __in_dev_get_rcu(fi->fib_nh->nh_dev);
1379 if (in_dev &&
1380 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1381 rtm->rtm_flags |= RTNH_F_DEAD;
1382 rcu_read_unlock();
1383 }
1384 if (fi->fib_nh->nh_flags & RTNH_F_OFFLOAD)
1385 rtm->rtm_flags |= RTNH_F_OFFLOAD;
1386#ifdef CONFIG_IP_ROUTE_CLASSID
1387 if (fi->fib_nh[0].nh_tclassid &&
1388 nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1389 goto nla_put_failure;
1390#endif
1391 if (fi->fib_nh->nh_lwtstate &&
1392 lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate) < 0)
1393 goto nla_put_failure;
1394 }
1395#ifdef CONFIG_IP_ROUTE_MULTIPATH
1396 if (fi->fib_nhs > 1) {
1397 struct rtnexthop *rtnh;
1398 struct nlattr *mp;
1399
1400 mp = nla_nest_start(skb, RTA_MULTIPATH);
1401 if (!mp)
1402 goto nla_put_failure;
1403
1404 for_nexthops(fi) {
1405 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1406 if (!rtnh)
1407 goto nla_put_failure;
1408
1409 rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1410 if (nh->nh_flags & RTNH_F_LINKDOWN) {
1411 struct in_device *in_dev;
1412
1413 rcu_read_lock();
1414 in_dev = __in_dev_get_rcu(nh->nh_dev);
1415 if (in_dev &&
1416 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1417 rtnh->rtnh_flags |= RTNH_F_DEAD;
1418 rcu_read_unlock();
1419 }
1420 rtnh->rtnh_hops = nh->nh_weight - 1;
1421 rtnh->rtnh_ifindex = nh->nh_oif;
1422
1423 if (nh->nh_gw &&
1424 nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1425 goto nla_put_failure;
1426#ifdef CONFIG_IP_ROUTE_CLASSID
1427 if (nh->nh_tclassid &&
1428 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1429 goto nla_put_failure;
1430#endif
1431 if (nh->nh_lwtstate &&
1432 lwtunnel_fill_encap(skb, nh->nh_lwtstate) < 0)
1433 goto nla_put_failure;
1434
1435 /* length of rtnetlink header + attributes */
1436 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1437 } endfor_nexthops(fi);
1438
1439 nla_nest_end(skb, mp);
1440 }
1441#endif
1442 nlmsg_end(skb, nlh);
1443 return 0;
1444
1445nla_put_failure:
1446 nlmsg_cancel(skb, nlh);
1447 return -EMSGSIZE;
1448}
1449
1450/*
1451 * Update FIB if:
1452 * - local address disappeared -> we must delete all the entries
1453 * referring to it.
1454 * - device went down -> we must shutdown all nexthops going via it.
1455 */
1456int fib_sync_down_addr(struct net_device *dev, __be32 local)
1457{
1458 int ret = 0;
1459 unsigned int hash = fib_laddr_hashfn(local);
1460 struct hlist_head *head = &fib_info_laddrhash[hash];
1461 struct net *net = dev_net(dev);
1462 int tb_id = l3mdev_fib_table(dev);
1463 struct fib_info *fi;
1464
1465 if (!fib_info_laddrhash || local == 0)
1466 return 0;
1467
1468 hlist_for_each_entry(fi, head, fib_lhash) {
1469 if (!net_eq(fi->fib_net, net) ||
1470 fi->fib_tb_id != tb_id)
1471 continue;
1472 if (fi->fib_prefsrc == local) {
1473 fi->fib_flags |= RTNH_F_DEAD;
1474 ret++;
1475 }
1476 }
1477 return ret;
1478}
1479
1480static int call_fib_nh_notifiers(struct fib_nh *fib_nh,
1481 enum fib_event_type event_type)
1482{
1483 struct in_device *in_dev = __in_dev_get_rtnl(fib_nh->nh_dev);
1484 struct fib_nh_notifier_info info = {
1485 .fib_nh = fib_nh,
1486 };
1487
1488 switch (event_type) {
1489 case FIB_EVENT_NH_ADD:
1490 if (fib_nh->nh_flags & RTNH_F_DEAD)
1491 break;
1492 if (IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1493 fib_nh->nh_flags & RTNH_F_LINKDOWN)
1494 break;
1495 return call_fib4_notifiers(dev_net(fib_nh->nh_dev), event_type,
1496 &info.info);
1497 case FIB_EVENT_NH_DEL:
1498 if ((in_dev && IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1499 fib_nh->nh_flags & RTNH_F_LINKDOWN) ||
1500 (fib_nh->nh_flags & RTNH_F_DEAD))
1501 return call_fib4_notifiers(dev_net(fib_nh->nh_dev),
1502 event_type, &info.info);
1503 default:
1504 break;
1505 }
1506
1507 return NOTIFY_DONE;
1508}
1509
1510/* Event force Flags Description
1511 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host
1512 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host
1513 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed
1514 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed
1515 */
1516int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1517{
1518 int ret = 0;
1519 int scope = RT_SCOPE_NOWHERE;
1520 struct fib_info *prev_fi = NULL;
1521 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1522 struct hlist_head *head = &fib_info_devhash[hash];
1523 struct fib_nh *nh;
1524
1525 if (force)
1526 scope = -1;
1527
1528 hlist_for_each_entry(nh, head, nh_hash) {
1529 struct fib_info *fi = nh->nh_parent;
1530 int dead;
1531
1532 BUG_ON(!fi->fib_nhs);
1533 if (nh->nh_dev != dev || fi == prev_fi)
1534 continue;
1535 prev_fi = fi;
1536 dead = 0;
1537 change_nexthops(fi) {
1538 if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1539 dead++;
1540 else if (nexthop_nh->nh_dev == dev &&
1541 nexthop_nh->nh_scope != scope) {
1542 switch (event) {
1543 case NETDEV_DOWN:
1544 case NETDEV_UNREGISTER:
1545 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1546 /* fall through */
1547 case NETDEV_CHANGE:
1548 nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1549 break;
1550 }
1551 call_fib_nh_notifiers(nexthop_nh,
1552 FIB_EVENT_NH_DEL);
1553 dead++;
1554 }
1555#ifdef CONFIG_IP_ROUTE_MULTIPATH
1556 if (event == NETDEV_UNREGISTER &&
1557 nexthop_nh->nh_dev == dev) {
1558 dead = fi->fib_nhs;
1559 break;
1560 }
1561#endif
1562 } endfor_nexthops(fi)
1563 if (dead == fi->fib_nhs) {
1564 switch (event) {
1565 case NETDEV_DOWN:
1566 case NETDEV_UNREGISTER:
1567 fi->fib_flags |= RTNH_F_DEAD;
1568 /* fall through */
1569 case NETDEV_CHANGE:
1570 fi->fib_flags |= RTNH_F_LINKDOWN;
1571 break;
1572 }
1573 ret++;
1574 }
1575
1576 fib_rebalance(fi);
1577 }
1578
1579 return ret;
1580}
1581
1582/* Must be invoked inside of an RCU protected region. */
1583static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1584{
1585 struct fib_info *fi = NULL, *last_resort = NULL;
1586 struct hlist_head *fa_head = res->fa_head;
1587 struct fib_table *tb = res->table;
1588 u8 slen = 32 - res->prefixlen;
1589 int order = -1, last_idx = -1;
1590 struct fib_alias *fa, *fa1 = NULL;
1591 u32 last_prio = res->fi->fib_priority;
1592 u8 last_tos = 0;
1593
1594 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1595 struct fib_info *next_fi = fa->fa_info;
1596
1597 if (fa->fa_slen != slen)
1598 continue;
1599 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1600 continue;
1601 if (fa->tb_id != tb->tb_id)
1602 continue;
1603 if (next_fi->fib_priority > last_prio &&
1604 fa->fa_tos == last_tos) {
1605 if (last_tos)
1606 continue;
1607 break;
1608 }
1609 if (next_fi->fib_flags & RTNH_F_DEAD)
1610 continue;
1611 last_tos = fa->fa_tos;
1612 last_prio = next_fi->fib_priority;
1613
1614 if (next_fi->fib_scope != res->scope ||
1615 fa->fa_type != RTN_UNICAST)
1616 continue;
1617 if (!next_fi->fib_nh[0].nh_gw ||
1618 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1619 continue;
1620
1621 fib_alias_accessed(fa);
1622
1623 if (!fi) {
1624 if (next_fi != res->fi)
1625 break;
1626 fa1 = fa;
1627 } else if (!fib_detect_death(fi, order, &last_resort,
1628 &last_idx, fa1->fa_default)) {
1629 fib_result_assign(res, fi);
1630 fa1->fa_default = order;
1631 goto out;
1632 }
1633 fi = next_fi;
1634 order++;
1635 }
1636
1637 if (order <= 0 || !fi) {
1638 if (fa1)
1639 fa1->fa_default = -1;
1640 goto out;
1641 }
1642
1643 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1644 fa1->fa_default)) {
1645 fib_result_assign(res, fi);
1646 fa1->fa_default = order;
1647 goto out;
1648 }
1649
1650 if (last_idx >= 0)
1651 fib_result_assign(res, last_resort);
1652 fa1->fa_default = last_idx;
1653out:
1654 return;
1655}
1656
1657/*
1658 * Dead device goes up. We wake up dead nexthops.
1659 * It takes sense only on multipath routes.
1660 */
1661int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1662{
1663 struct fib_info *prev_fi;
1664 unsigned int hash;
1665 struct hlist_head *head;
1666 struct fib_nh *nh;
1667 int ret;
1668
1669 if (!(dev->flags & IFF_UP))
1670 return 0;
1671
1672 if (nh_flags & RTNH_F_DEAD) {
1673 unsigned int flags = dev_get_flags(dev);
1674
1675 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1676 nh_flags |= RTNH_F_LINKDOWN;
1677 }
1678
1679 prev_fi = NULL;
1680 hash = fib_devindex_hashfn(dev->ifindex);
1681 head = &fib_info_devhash[hash];
1682 ret = 0;
1683
1684 hlist_for_each_entry(nh, head, nh_hash) {
1685 struct fib_info *fi = nh->nh_parent;
1686 int alive;
1687
1688 BUG_ON(!fi->fib_nhs);
1689 if (nh->nh_dev != dev || fi == prev_fi)
1690 continue;
1691
1692 prev_fi = fi;
1693 alive = 0;
1694 change_nexthops(fi) {
1695 if (!(nexthop_nh->nh_flags & nh_flags)) {
1696 alive++;
1697 continue;
1698 }
1699 if (!nexthop_nh->nh_dev ||
1700 !(nexthop_nh->nh_dev->flags & IFF_UP))
1701 continue;
1702 if (nexthop_nh->nh_dev != dev ||
1703 !__in_dev_get_rtnl(dev))
1704 continue;
1705 alive++;
1706 nexthop_nh->nh_flags &= ~nh_flags;
1707 call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
1708 } endfor_nexthops(fi)
1709
1710 if (alive > 0) {
1711 fi->fib_flags &= ~nh_flags;
1712 ret++;
1713 }
1714
1715 fib_rebalance(fi);
1716 }
1717
1718 return ret;
1719}
1720
1721#ifdef CONFIG_IP_ROUTE_MULTIPATH
1722static bool fib_good_nh(const struct fib_nh *nh)
1723{
1724 int state = NUD_REACHABLE;
1725
1726 if (nh->nh_scope == RT_SCOPE_LINK) {
1727 struct neighbour *n;
1728
1729 rcu_read_lock_bh();
1730
1731 n = __ipv4_neigh_lookup_noref(nh->nh_dev,
1732 (__force u32)nh->nh_gw);
1733 if (n)
1734 state = n->nud_state;
1735
1736 rcu_read_unlock_bh();
1737 }
1738
1739 return !!(state & NUD_VALID);
1740}
1741
1742void fib_select_multipath(struct fib_result *res, int hash)
1743{
1744 struct fib_info *fi = res->fi;
1745 struct net *net = fi->fib_net;
1746 bool first = false;
1747
1748 for_nexthops(fi) {
1749 if (net->ipv4.sysctl_fib_multipath_use_neigh) {
1750 if (!fib_good_nh(nh))
1751 continue;
1752 if (!first) {
1753 res->nh_sel = nhsel;
1754 first = true;
1755 }
1756 }
1757
1758 if (hash > atomic_read(&nh->nh_upper_bound))
1759 continue;
1760
1761 res->nh_sel = nhsel;
1762 return;
1763 } endfor_nexthops(fi);
1764}
1765#endif
1766
1767void fib_select_path(struct net *net, struct fib_result *res,
1768 struct flowi4 *fl4, const struct sk_buff *skb)
1769{
1770 if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
1771 goto check_saddr;
1772
1773#ifdef CONFIG_IP_ROUTE_MULTIPATH
1774 if (res->fi->fib_nhs > 1) {
1775 int h = fib_multipath_hash(net, fl4, skb, NULL);
1776
1777 fib_select_multipath(res, h);
1778 }
1779 else
1780#endif
1781 if (!res->prefixlen &&
1782 res->table->tb_num_default > 1 &&
1783 res->type == RTN_UNICAST)
1784 fib_select_default(fl4, res);
1785
1786check_saddr:
1787 if (!fl4->saddr)
1788 fl4->saddr = FIB_RES_PREFSRC(net, *res);
1789}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
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 *
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 */
11
12#include <linux/uaccess.h>
13#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>
24#include <linux/inetdevice.h>
25#include <linux/netdevice.h>
26#include <linux/if_arp.h>
27#include <linux/proc_fs.h>
28#include <linux/skbuff.h>
29#include <linux/init.h>
30#include <linux/slab.h>
31#include <linux/netlink.h>
32#include <linux/hash.h>
33#include <linux/nospec.h>
34
35#include <net/arp.h>
36#include <net/inet_dscp.h>
37#include <net/ip.h>
38#include <net/protocol.h>
39#include <net/route.h>
40#include <net/tcp.h>
41#include <net/sock.h>
42#include <net/ip_fib.h>
43#include <net/ip6_fib.h>
44#include <net/nexthop.h>
45#include <net/netlink.h>
46#include <net/rtnh.h>
47#include <net/lwtunnel.h>
48#include <net/fib_notifier.h>
49#include <net/addrconf.h>
50
51#include "fib_lookup.h"
52
53static DEFINE_SPINLOCK(fib_info_lock);
54static struct hlist_head *fib_info_hash;
55static struct hlist_head *fib_info_laddrhash;
56static unsigned int fib_info_hash_size;
57static unsigned int fib_info_hash_bits;
58static unsigned int fib_info_cnt;
59
60#define DEVINDEX_HASHBITS 8
61#define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
62static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
63
64/* for_nexthops and change_nexthops only used when nexthop object
65 * is not set in a fib_info. The logic within can reference fib_nh.
66 */
67#ifdef CONFIG_IP_ROUTE_MULTIPATH
68
69#define for_nexthops(fi) { \
70 int nhsel; const struct fib_nh *nh; \
71 for (nhsel = 0, nh = (fi)->fib_nh; \
72 nhsel < fib_info_num_path((fi)); \
73 nh++, nhsel++)
74
75#define change_nexthops(fi) { \
76 int nhsel; struct fib_nh *nexthop_nh; \
77 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
78 nhsel < fib_info_num_path((fi)); \
79 nexthop_nh++, nhsel++)
80
81#else /* CONFIG_IP_ROUTE_MULTIPATH */
82
83/* Hope, that gcc will optimize it to get rid of dummy loop */
84
85#define for_nexthops(fi) { \
86 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \
87 for (nhsel = 0; nhsel < 1; nhsel++)
88
89#define change_nexthops(fi) { \
90 int nhsel; \
91 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
92 for (nhsel = 0; nhsel < 1; nhsel++)
93
94#endif /* CONFIG_IP_ROUTE_MULTIPATH */
95
96#define endfor_nexthops(fi) }
97
98
99const struct fib_prop fib_props[RTN_MAX + 1] = {
100 [RTN_UNSPEC] = {
101 .error = 0,
102 .scope = RT_SCOPE_NOWHERE,
103 },
104 [RTN_UNICAST] = {
105 .error = 0,
106 .scope = RT_SCOPE_UNIVERSE,
107 },
108 [RTN_LOCAL] = {
109 .error = 0,
110 .scope = RT_SCOPE_HOST,
111 },
112 [RTN_BROADCAST] = {
113 .error = 0,
114 .scope = RT_SCOPE_LINK,
115 },
116 [RTN_ANYCAST] = {
117 .error = 0,
118 .scope = RT_SCOPE_LINK,
119 },
120 [RTN_MULTICAST] = {
121 .error = 0,
122 .scope = RT_SCOPE_UNIVERSE,
123 },
124 [RTN_BLACKHOLE] = {
125 .error = -EINVAL,
126 .scope = RT_SCOPE_UNIVERSE,
127 },
128 [RTN_UNREACHABLE] = {
129 .error = -EHOSTUNREACH,
130 .scope = RT_SCOPE_UNIVERSE,
131 },
132 [RTN_PROHIBIT] = {
133 .error = -EACCES,
134 .scope = RT_SCOPE_UNIVERSE,
135 },
136 [RTN_THROW] = {
137 .error = -EAGAIN,
138 .scope = RT_SCOPE_UNIVERSE,
139 },
140 [RTN_NAT] = {
141 .error = -EINVAL,
142 .scope = RT_SCOPE_NOWHERE,
143 },
144 [RTN_XRESOLVE] = {
145 .error = -EINVAL,
146 .scope = RT_SCOPE_NOWHERE,
147 },
148};
149
150static void rt_fibinfo_free(struct rtable __rcu **rtp)
151{
152 struct rtable *rt = rcu_dereference_protected(*rtp, 1);
153
154 if (!rt)
155 return;
156
157 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
158 * because we waited an RCU grace period before calling
159 * free_fib_info_rcu()
160 */
161
162 dst_dev_put(&rt->dst);
163 dst_release_immediate(&rt->dst);
164}
165
166static void free_nh_exceptions(struct fib_nh_common *nhc)
167{
168 struct fnhe_hash_bucket *hash;
169 int i;
170
171 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
172 if (!hash)
173 return;
174 for (i = 0; i < FNHE_HASH_SIZE; i++) {
175 struct fib_nh_exception *fnhe;
176
177 fnhe = rcu_dereference_protected(hash[i].chain, 1);
178 while (fnhe) {
179 struct fib_nh_exception *next;
180
181 next = rcu_dereference_protected(fnhe->fnhe_next, 1);
182
183 rt_fibinfo_free(&fnhe->fnhe_rth_input);
184 rt_fibinfo_free(&fnhe->fnhe_rth_output);
185
186 kfree(fnhe);
187
188 fnhe = next;
189 }
190 }
191 kfree(hash);
192}
193
194static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
195{
196 int cpu;
197
198 if (!rtp)
199 return;
200
201 for_each_possible_cpu(cpu) {
202 struct rtable *rt;
203
204 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
205 if (rt) {
206 dst_dev_put(&rt->dst);
207 dst_release_immediate(&rt->dst);
208 }
209 }
210 free_percpu(rtp);
211}
212
213void fib_nh_common_release(struct fib_nh_common *nhc)
214{
215 netdev_put(nhc->nhc_dev, &nhc->nhc_dev_tracker);
216 lwtstate_put(nhc->nhc_lwtstate);
217 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
218 rt_fibinfo_free(&nhc->nhc_rth_input);
219 free_nh_exceptions(nhc);
220}
221EXPORT_SYMBOL_GPL(fib_nh_common_release);
222
223void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
224{
225#ifdef CONFIG_IP_ROUTE_CLASSID
226 if (fib_nh->nh_tclassid)
227 atomic_dec(&net->ipv4.fib_num_tclassid_users);
228#endif
229 fib_nh_common_release(&fib_nh->nh_common);
230}
231
232/* Release a nexthop info record */
233static void free_fib_info_rcu(struct rcu_head *head)
234{
235 struct fib_info *fi = container_of(head, struct fib_info, rcu);
236
237 if (fi->nh) {
238 nexthop_put(fi->nh);
239 } else {
240 change_nexthops(fi) {
241 fib_nh_release(fi->fib_net, nexthop_nh);
242 } endfor_nexthops(fi);
243 }
244
245 ip_fib_metrics_put(fi->fib_metrics);
246
247 kfree(fi);
248}
249
250void free_fib_info(struct fib_info *fi)
251{
252 if (fi->fib_dead == 0) {
253 pr_warn("Freeing alive fib_info %p\n", fi);
254 return;
255 }
256
257 call_rcu(&fi->rcu, free_fib_info_rcu);
258}
259EXPORT_SYMBOL_GPL(free_fib_info);
260
261void fib_release_info(struct fib_info *fi)
262{
263 spin_lock_bh(&fib_info_lock);
264 if (fi && refcount_dec_and_test(&fi->fib_treeref)) {
265 hlist_del(&fi->fib_hash);
266
267 /* Paired with READ_ONCE() in fib_create_info(). */
268 WRITE_ONCE(fib_info_cnt, fib_info_cnt - 1);
269
270 if (fi->fib_prefsrc)
271 hlist_del(&fi->fib_lhash);
272 if (fi->nh) {
273 list_del(&fi->nh_list);
274 } else {
275 change_nexthops(fi) {
276 if (!nexthop_nh->fib_nh_dev)
277 continue;
278 hlist_del(&nexthop_nh->nh_hash);
279 } endfor_nexthops(fi)
280 }
281 fi->fib_dead = 1;
282 fib_info_put(fi);
283 }
284 spin_unlock_bh(&fib_info_lock);
285}
286
287static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
288{
289 const struct fib_nh *onh;
290
291 if (fi->nh || ofi->nh)
292 return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
293
294 if (ofi->fib_nhs == 0)
295 return 0;
296
297 for_nexthops(fi) {
298 onh = fib_info_nh(ofi, nhsel);
299
300 if (nh->fib_nh_oif != onh->fib_nh_oif ||
301 nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
302 nh->fib_nh_scope != onh->fib_nh_scope ||
303#ifdef CONFIG_IP_ROUTE_MULTIPATH
304 nh->fib_nh_weight != onh->fib_nh_weight ||
305#endif
306#ifdef CONFIG_IP_ROUTE_CLASSID
307 nh->nh_tclassid != onh->nh_tclassid ||
308#endif
309 lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
310 ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
311 return -1;
312
313 if (nh->fib_nh_gw_family == AF_INET &&
314 nh->fib_nh_gw4 != onh->fib_nh_gw4)
315 return -1;
316
317 if (nh->fib_nh_gw_family == AF_INET6 &&
318 ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
319 return -1;
320 } endfor_nexthops(fi);
321 return 0;
322}
323
324static inline unsigned int fib_devindex_hashfn(unsigned int val)
325{
326 return hash_32(val, DEVINDEX_HASHBITS);
327}
328
329static struct hlist_head *
330fib_info_devhash_bucket(const struct net_device *dev)
331{
332 u32 val = net_hash_mix(dev_net(dev)) ^ dev->ifindex;
333
334 return &fib_info_devhash[fib_devindex_hashfn(val)];
335}
336
337static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
338 u32 prefsrc, u32 priority)
339{
340 unsigned int val = init_val;
341
342 val ^= (protocol << 8) | scope;
343 val ^= prefsrc;
344 val ^= priority;
345
346 return val;
347}
348
349static unsigned int fib_info_hashfn_result(unsigned int val)
350{
351 unsigned int mask = (fib_info_hash_size - 1);
352
353 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
354}
355
356static inline unsigned int fib_info_hashfn(struct fib_info *fi)
357{
358 unsigned int val;
359
360 val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
361 fi->fib_scope, (__force u32)fi->fib_prefsrc,
362 fi->fib_priority);
363
364 if (fi->nh) {
365 val ^= fib_devindex_hashfn(fi->nh->id);
366 } else {
367 for_nexthops(fi) {
368 val ^= fib_devindex_hashfn(nh->fib_nh_oif);
369 } endfor_nexthops(fi)
370 }
371
372 return fib_info_hashfn_result(val);
373}
374
375/* no metrics, only nexthop id */
376static struct fib_info *fib_find_info_nh(struct net *net,
377 const struct fib_config *cfg)
378{
379 struct hlist_head *head;
380 struct fib_info *fi;
381 unsigned int hash;
382
383 hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
384 cfg->fc_protocol, cfg->fc_scope,
385 (__force u32)cfg->fc_prefsrc,
386 cfg->fc_priority);
387 hash = fib_info_hashfn_result(hash);
388 head = &fib_info_hash[hash];
389
390 hlist_for_each_entry(fi, head, fib_hash) {
391 if (!net_eq(fi->fib_net, net))
392 continue;
393 if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
394 continue;
395 if (cfg->fc_protocol == fi->fib_protocol &&
396 cfg->fc_scope == fi->fib_scope &&
397 cfg->fc_prefsrc == fi->fib_prefsrc &&
398 cfg->fc_priority == fi->fib_priority &&
399 cfg->fc_type == fi->fib_type &&
400 cfg->fc_table == fi->fib_tb_id &&
401 !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
402 return fi;
403 }
404
405 return NULL;
406}
407
408static struct fib_info *fib_find_info(struct fib_info *nfi)
409{
410 struct hlist_head *head;
411 struct fib_info *fi;
412 unsigned int hash;
413
414 hash = fib_info_hashfn(nfi);
415 head = &fib_info_hash[hash];
416
417 hlist_for_each_entry(fi, head, fib_hash) {
418 if (!net_eq(fi->fib_net, nfi->fib_net))
419 continue;
420 if (fi->fib_nhs != nfi->fib_nhs)
421 continue;
422 if (nfi->fib_protocol == fi->fib_protocol &&
423 nfi->fib_scope == fi->fib_scope &&
424 nfi->fib_prefsrc == fi->fib_prefsrc &&
425 nfi->fib_priority == fi->fib_priority &&
426 nfi->fib_type == fi->fib_type &&
427 nfi->fib_tb_id == fi->fib_tb_id &&
428 memcmp(nfi->fib_metrics, fi->fib_metrics,
429 sizeof(u32) * RTAX_MAX) == 0 &&
430 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
431 nh_comp(fi, nfi) == 0)
432 return fi;
433 }
434
435 return NULL;
436}
437
438/* Check, that the gateway is already configured.
439 * Used only by redirect accept routine.
440 */
441int ip_fib_check_default(__be32 gw, struct net_device *dev)
442{
443 struct hlist_head *head;
444 struct fib_nh *nh;
445
446 spin_lock(&fib_info_lock);
447
448 head = fib_info_devhash_bucket(dev);
449
450 hlist_for_each_entry(nh, head, nh_hash) {
451 if (nh->fib_nh_dev == dev &&
452 nh->fib_nh_gw4 == gw &&
453 !(nh->fib_nh_flags & RTNH_F_DEAD)) {
454 spin_unlock(&fib_info_lock);
455 return 0;
456 }
457 }
458
459 spin_unlock(&fib_info_lock);
460
461 return -1;
462}
463
464size_t fib_nlmsg_size(struct fib_info *fi)
465{
466 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
467 + nla_total_size(4) /* RTA_TABLE */
468 + nla_total_size(4) /* RTA_DST */
469 + nla_total_size(4) /* RTA_PRIORITY */
470 + nla_total_size(4) /* RTA_PREFSRC */
471 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
472 unsigned int nhs = fib_info_num_path(fi);
473
474 /* space for nested metrics */
475 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
476
477 if (fi->nh)
478 payload += nla_total_size(4); /* RTA_NH_ID */
479
480 if (nhs) {
481 size_t nh_encapsize = 0;
482 /* Also handles the special case nhs == 1 */
483
484 /* each nexthop is packed in an attribute */
485 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
486 unsigned int i;
487
488 /* may contain flow and gateway attribute */
489 nhsize += 2 * nla_total_size(4);
490
491 /* grab encap info */
492 for (i = 0; i < fib_info_num_path(fi); i++) {
493 struct fib_nh_common *nhc = fib_info_nhc(fi, i);
494
495 if (nhc->nhc_lwtstate) {
496 /* RTA_ENCAP_TYPE */
497 nh_encapsize += lwtunnel_get_encap_size(
498 nhc->nhc_lwtstate);
499 /* RTA_ENCAP */
500 nh_encapsize += nla_total_size(2);
501 }
502 }
503
504 /* all nexthops are packed in a nested attribute */
505 payload += nla_total_size((nhs * nhsize) + nh_encapsize);
506
507 }
508
509 return payload;
510}
511
512void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
513 int dst_len, u32 tb_id, const struct nl_info *info,
514 unsigned int nlm_flags)
515{
516 struct fib_rt_info fri;
517 struct sk_buff *skb;
518 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
519 int err = -ENOBUFS;
520
521 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
522 if (!skb)
523 goto errout;
524
525 fri.fi = fa->fa_info;
526 fri.tb_id = tb_id;
527 fri.dst = key;
528 fri.dst_len = dst_len;
529 fri.dscp = fa->fa_dscp;
530 fri.type = fa->fa_type;
531 fri.offload = READ_ONCE(fa->offload);
532 fri.trap = READ_ONCE(fa->trap);
533 fri.offload_failed = READ_ONCE(fa->offload_failed);
534 err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
535 if (err < 0) {
536 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
537 WARN_ON(err == -EMSGSIZE);
538 kfree_skb(skb);
539 goto errout;
540 }
541 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
542 info->nlh, GFP_KERNEL);
543 return;
544errout:
545 if (err < 0)
546 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
547}
548
549static int fib_detect_death(struct fib_info *fi, int order,
550 struct fib_info **last_resort, int *last_idx,
551 int dflt)
552{
553 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
554 struct neighbour *n;
555 int state = NUD_NONE;
556
557 if (likely(nhc->nhc_gw_family == AF_INET))
558 n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
559 else if (nhc->nhc_gw_family == AF_INET6)
560 n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
561 nhc->nhc_dev);
562 else
563 n = NULL;
564
565 if (n) {
566 state = n->nud_state;
567 neigh_release(n);
568 } else {
569 return 0;
570 }
571 if (state == NUD_REACHABLE)
572 return 0;
573 if ((state & NUD_VALID) && order != dflt)
574 return 0;
575 if ((state & NUD_VALID) ||
576 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
577 *last_resort = fi;
578 *last_idx = order;
579 }
580 return 1;
581}
582
583int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
584 struct nlattr *encap, u16 encap_type,
585 void *cfg, gfp_t gfp_flags,
586 struct netlink_ext_ack *extack)
587{
588 int err;
589
590 nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
591 gfp_flags);
592 if (!nhc->nhc_pcpu_rth_output)
593 return -ENOMEM;
594
595 if (encap) {
596 struct lwtunnel_state *lwtstate;
597
598 if (encap_type == LWTUNNEL_ENCAP_NONE) {
599 NL_SET_ERR_MSG(extack, "LWT encap type not specified");
600 err = -EINVAL;
601 goto lwt_failure;
602 }
603 err = lwtunnel_build_state(net, encap_type, encap,
604 nhc->nhc_family, cfg, &lwtstate,
605 extack);
606 if (err)
607 goto lwt_failure;
608
609 nhc->nhc_lwtstate = lwtstate_get(lwtstate);
610 }
611
612 return 0;
613
614lwt_failure:
615 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
616 nhc->nhc_pcpu_rth_output = NULL;
617 return err;
618}
619EXPORT_SYMBOL_GPL(fib_nh_common_init);
620
621int fib_nh_init(struct net *net, struct fib_nh *nh,
622 struct fib_config *cfg, int nh_weight,
623 struct netlink_ext_ack *extack)
624{
625 int err;
626
627 nh->fib_nh_family = AF_INET;
628
629 err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
630 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
631 if (err)
632 return err;
633
634 nh->fib_nh_oif = cfg->fc_oif;
635 nh->fib_nh_gw_family = cfg->fc_gw_family;
636 if (cfg->fc_gw_family == AF_INET)
637 nh->fib_nh_gw4 = cfg->fc_gw4;
638 else if (cfg->fc_gw_family == AF_INET6)
639 nh->fib_nh_gw6 = cfg->fc_gw6;
640
641 nh->fib_nh_flags = cfg->fc_flags;
642
643#ifdef CONFIG_IP_ROUTE_CLASSID
644 nh->nh_tclassid = cfg->fc_flow;
645 if (nh->nh_tclassid)
646 atomic_inc(&net->ipv4.fib_num_tclassid_users);
647#endif
648#ifdef CONFIG_IP_ROUTE_MULTIPATH
649 nh->fib_nh_weight = nh_weight;
650#endif
651 return 0;
652}
653
654#ifdef CONFIG_IP_ROUTE_MULTIPATH
655
656static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
657 struct netlink_ext_ack *extack)
658{
659 int nhs = 0;
660
661 while (rtnh_ok(rtnh, remaining)) {
662 nhs++;
663 rtnh = rtnh_next(rtnh, &remaining);
664 }
665
666 /* leftover implies invalid nexthop configuration, discard it */
667 if (remaining > 0) {
668 NL_SET_ERR_MSG(extack,
669 "Invalid nexthop configuration - extra data after nexthops");
670 nhs = 0;
671 }
672
673 return nhs;
674}
675
676static int fib_gw_from_attr(__be32 *gw, struct nlattr *nla,
677 struct netlink_ext_ack *extack)
678{
679 if (nla_len(nla) < sizeof(*gw)) {
680 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_GATEWAY");
681 return -EINVAL;
682 }
683
684 *gw = nla_get_in_addr(nla);
685
686 return 0;
687}
688
689/* only called when fib_nh is integrated into fib_info */
690static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
691 int remaining, struct fib_config *cfg,
692 struct netlink_ext_ack *extack)
693{
694 struct net *net = fi->fib_net;
695 struct fib_config fib_cfg;
696 struct fib_nh *nh;
697 int ret;
698
699 change_nexthops(fi) {
700 int attrlen;
701
702 memset(&fib_cfg, 0, sizeof(fib_cfg));
703
704 if (!rtnh_ok(rtnh, remaining)) {
705 NL_SET_ERR_MSG(extack,
706 "Invalid nexthop configuration - extra data after nexthop");
707 return -EINVAL;
708 }
709
710 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
711 NL_SET_ERR_MSG(extack,
712 "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
713 return -EINVAL;
714 }
715
716 fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
717 fib_cfg.fc_oif = rtnh->rtnh_ifindex;
718
719 attrlen = rtnh_attrlen(rtnh);
720 if (attrlen > 0) {
721 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
722
723 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
724 nlav = nla_find(attrs, attrlen, RTA_VIA);
725 if (nla && nlav) {
726 NL_SET_ERR_MSG(extack,
727 "Nexthop configuration can not contain both GATEWAY and VIA");
728 return -EINVAL;
729 }
730 if (nla) {
731 ret = fib_gw_from_attr(&fib_cfg.fc_gw4, nla,
732 extack);
733 if (ret)
734 goto errout;
735
736 if (fib_cfg.fc_gw4)
737 fib_cfg.fc_gw_family = AF_INET;
738 } else if (nlav) {
739 ret = fib_gw_from_via(&fib_cfg, nlav, extack);
740 if (ret)
741 goto errout;
742 }
743
744 nla = nla_find(attrs, attrlen, RTA_FLOW);
745 if (nla) {
746 if (nla_len(nla) < sizeof(u32)) {
747 NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
748 return -EINVAL;
749 }
750 fib_cfg.fc_flow = nla_get_u32(nla);
751 }
752
753 fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
754 /* RTA_ENCAP_TYPE length checked in
755 * lwtunnel_valid_encap_type_attr
756 */
757 nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
758 if (nla)
759 fib_cfg.fc_encap_type = nla_get_u16(nla);
760 }
761
762 ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
763 rtnh->rtnh_hops + 1, extack);
764 if (ret)
765 goto errout;
766
767 rtnh = rtnh_next(rtnh, &remaining);
768 } endfor_nexthops(fi);
769
770 ret = -EINVAL;
771 nh = fib_info_nh(fi, 0);
772 if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
773 NL_SET_ERR_MSG(extack,
774 "Nexthop device index does not match RTA_OIF");
775 goto errout;
776 }
777 if (cfg->fc_gw_family) {
778 if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
779 (cfg->fc_gw_family == AF_INET &&
780 nh->fib_nh_gw4 != cfg->fc_gw4) ||
781 (cfg->fc_gw_family == AF_INET6 &&
782 ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
783 NL_SET_ERR_MSG(extack,
784 "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
785 goto errout;
786 }
787 }
788#ifdef CONFIG_IP_ROUTE_CLASSID
789 if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
790 NL_SET_ERR_MSG(extack,
791 "Nexthop class id does not match RTA_FLOW");
792 goto errout;
793 }
794#endif
795 ret = 0;
796errout:
797 return ret;
798}
799
800/* only called when fib_nh is integrated into fib_info */
801static void fib_rebalance(struct fib_info *fi)
802{
803 int total;
804 int w;
805
806 if (fib_info_num_path(fi) < 2)
807 return;
808
809 total = 0;
810 for_nexthops(fi) {
811 if (nh->fib_nh_flags & RTNH_F_DEAD)
812 continue;
813
814 if (ip_ignore_linkdown(nh->fib_nh_dev) &&
815 nh->fib_nh_flags & RTNH_F_LINKDOWN)
816 continue;
817
818 total += nh->fib_nh_weight;
819 } endfor_nexthops(fi);
820
821 w = 0;
822 change_nexthops(fi) {
823 int upper_bound;
824
825 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
826 upper_bound = -1;
827 } else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
828 nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
829 upper_bound = -1;
830 } else {
831 w += nexthop_nh->fib_nh_weight;
832 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
833 total) - 1;
834 }
835
836 atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
837 } endfor_nexthops(fi);
838}
839#else /* CONFIG_IP_ROUTE_MULTIPATH */
840
841static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
842 int remaining, struct fib_config *cfg,
843 struct netlink_ext_ack *extack)
844{
845 NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
846
847 return -EINVAL;
848}
849
850#define fib_rebalance(fi) do { } while (0)
851
852#endif /* CONFIG_IP_ROUTE_MULTIPATH */
853
854static int fib_encap_match(struct net *net, u16 encap_type,
855 struct nlattr *encap,
856 const struct fib_nh *nh,
857 const struct fib_config *cfg,
858 struct netlink_ext_ack *extack)
859{
860 struct lwtunnel_state *lwtstate;
861 int ret, result = 0;
862
863 if (encap_type == LWTUNNEL_ENCAP_NONE)
864 return 0;
865
866 ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
867 cfg, &lwtstate, extack);
868 if (!ret) {
869 result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
870 lwtstate_free(lwtstate);
871 }
872
873 return result;
874}
875
876int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
877 struct netlink_ext_ack *extack)
878{
879#ifdef CONFIG_IP_ROUTE_MULTIPATH
880 struct rtnexthop *rtnh;
881 int remaining;
882#endif
883
884 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
885 return 1;
886
887 if (cfg->fc_nh_id) {
888 if (fi->nh && cfg->fc_nh_id == fi->nh->id)
889 return 0;
890 return 1;
891 }
892
893 if (fi->nh) {
894 if (cfg->fc_oif || cfg->fc_gw_family || cfg->fc_mp)
895 return 1;
896 return 0;
897 }
898
899 if (cfg->fc_oif || cfg->fc_gw_family) {
900 struct fib_nh *nh;
901
902 nh = fib_info_nh(fi, 0);
903 if (cfg->fc_encap) {
904 if (fib_encap_match(net, cfg->fc_encap_type,
905 cfg->fc_encap, nh, cfg, extack))
906 return 1;
907 }
908#ifdef CONFIG_IP_ROUTE_CLASSID
909 if (cfg->fc_flow &&
910 cfg->fc_flow != nh->nh_tclassid)
911 return 1;
912#endif
913 if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
914 (cfg->fc_gw_family &&
915 cfg->fc_gw_family != nh->fib_nh_gw_family))
916 return 1;
917
918 if (cfg->fc_gw_family == AF_INET &&
919 cfg->fc_gw4 != nh->fib_nh_gw4)
920 return 1;
921
922 if (cfg->fc_gw_family == AF_INET6 &&
923 ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
924 return 1;
925
926 return 0;
927 }
928
929#ifdef CONFIG_IP_ROUTE_MULTIPATH
930 if (!cfg->fc_mp)
931 return 0;
932
933 rtnh = cfg->fc_mp;
934 remaining = cfg->fc_mp_len;
935
936 for_nexthops(fi) {
937 int attrlen;
938
939 if (!rtnh_ok(rtnh, remaining))
940 return -EINVAL;
941
942 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
943 return 1;
944
945 attrlen = rtnh_attrlen(rtnh);
946 if (attrlen > 0) {
947 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
948 int err;
949
950 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
951 nlav = nla_find(attrs, attrlen, RTA_VIA);
952 if (nla && nlav) {
953 NL_SET_ERR_MSG(extack,
954 "Nexthop configuration can not contain both GATEWAY and VIA");
955 return -EINVAL;
956 }
957
958 if (nla) {
959 __be32 gw;
960
961 err = fib_gw_from_attr(&gw, nla, extack);
962 if (err)
963 return err;
964
965 if (nh->fib_nh_gw_family != AF_INET ||
966 gw != nh->fib_nh_gw4)
967 return 1;
968 } else if (nlav) {
969 struct fib_config cfg2;
970
971 err = fib_gw_from_via(&cfg2, nlav, extack);
972 if (err)
973 return err;
974
975 switch (nh->fib_nh_gw_family) {
976 case AF_INET:
977 if (cfg2.fc_gw_family != AF_INET ||
978 cfg2.fc_gw4 != nh->fib_nh_gw4)
979 return 1;
980 break;
981 case AF_INET6:
982 if (cfg2.fc_gw_family != AF_INET6 ||
983 ipv6_addr_cmp(&cfg2.fc_gw6,
984 &nh->fib_nh_gw6))
985 return 1;
986 break;
987 }
988 }
989
990#ifdef CONFIG_IP_ROUTE_CLASSID
991 nla = nla_find(attrs, attrlen, RTA_FLOW);
992 if (nla) {
993 if (nla_len(nla) < sizeof(u32)) {
994 NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
995 return -EINVAL;
996 }
997 if (nla_get_u32(nla) != nh->nh_tclassid)
998 return 1;
999 }
1000#endif
1001 }
1002
1003 rtnh = rtnh_next(rtnh, &remaining);
1004 } endfor_nexthops(fi);
1005#endif
1006 return 0;
1007}
1008
1009bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
1010{
1011 struct nlattr *nla;
1012 int remaining;
1013
1014 if (!cfg->fc_mx)
1015 return true;
1016
1017 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1018 int type = nla_type(nla);
1019 u32 fi_val, val;
1020
1021 if (!type)
1022 continue;
1023 if (type > RTAX_MAX)
1024 return false;
1025
1026 type = array_index_nospec(type, RTAX_MAX + 1);
1027 if (type == RTAX_CC_ALGO) {
1028 char tmp[TCP_CA_NAME_MAX];
1029 bool ecn_ca = false;
1030
1031 nla_strscpy(tmp, nla, sizeof(tmp));
1032 val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
1033 } else {
1034 if (nla_len(nla) != sizeof(u32))
1035 return false;
1036 val = nla_get_u32(nla);
1037 }
1038
1039 fi_val = fi->fib_metrics->metrics[type - 1];
1040 if (type == RTAX_FEATURES)
1041 fi_val &= ~DST_FEATURE_ECN_CA;
1042
1043 if (fi_val != val)
1044 return false;
1045 }
1046
1047 return true;
1048}
1049
1050static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
1051 u32 table, struct netlink_ext_ack *extack)
1052{
1053 struct fib6_config cfg = {
1054 .fc_table = table,
1055 .fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1056 .fc_ifindex = nh->fib_nh_oif,
1057 .fc_gateway = nh->fib_nh_gw6,
1058 };
1059 struct fib6_nh fib6_nh = {};
1060 int err;
1061
1062 err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1063 if (!err) {
1064 nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1065 netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1066 GFP_KERNEL);
1067 nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1068 nh->fib_nh_scope = RT_SCOPE_LINK;
1069
1070 ipv6_stub->fib6_nh_release(&fib6_nh);
1071 }
1072
1073 return err;
1074}
1075
1076/*
1077 * Picture
1078 * -------
1079 *
1080 * Semantics of nexthop is very messy by historical reasons.
1081 * We have to take into account, that:
1082 * a) gateway can be actually local interface address,
1083 * so that gatewayed route is direct.
1084 * b) gateway must be on-link address, possibly
1085 * described not by an ifaddr, but also by a direct route.
1086 * c) If both gateway and interface are specified, they should not
1087 * contradict.
1088 * d) If we use tunnel routes, gateway could be not on-link.
1089 *
1090 * Attempt to reconcile all of these (alas, self-contradictory) conditions
1091 * results in pretty ugly and hairy code with obscure logic.
1092 *
1093 * I chose to generalized it instead, so that the size
1094 * of code does not increase practically, but it becomes
1095 * much more general.
1096 * Every prefix is assigned a "scope" value: "host" is local address,
1097 * "link" is direct route,
1098 * [ ... "site" ... "interior" ... ]
1099 * and "universe" is true gateway route with global meaning.
1100 *
1101 * Every prefix refers to a set of "nexthop"s (gw, oif),
1102 * where gw must have narrower scope. This recursion stops
1103 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1104 * which means that gw is forced to be on link.
1105 *
1106 * Code is still hairy, but now it is apparently logically
1107 * consistent and very flexible. F.e. as by-product it allows
1108 * to co-exists in peace independent exterior and interior
1109 * routing processes.
1110 *
1111 * Normally it looks as following.
1112 *
1113 * {universe prefix} -> (gw, oif) [scope link]
1114 * |
1115 * |-> {link prefix} -> (gw, oif) [scope local]
1116 * |
1117 * |-> {local prefix} (terminal node)
1118 */
1119static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1120 u8 scope, struct netlink_ext_ack *extack)
1121{
1122 struct net_device *dev;
1123 struct fib_result res;
1124 int err = 0;
1125
1126 if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1127 unsigned int addr_type;
1128
1129 if (scope >= RT_SCOPE_LINK) {
1130 NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1131 return -EINVAL;
1132 }
1133 dev = __dev_get_by_index(net, nh->fib_nh_oif);
1134 if (!dev) {
1135 NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1136 return -ENODEV;
1137 }
1138 if (!(dev->flags & IFF_UP)) {
1139 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1140 return -ENETDOWN;
1141 }
1142 addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1143 if (addr_type != RTN_UNICAST) {
1144 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1145 return -EINVAL;
1146 }
1147 if (!netif_carrier_ok(dev))
1148 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1149 nh->fib_nh_dev = dev;
1150 netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1151 nh->fib_nh_scope = RT_SCOPE_LINK;
1152 return 0;
1153 }
1154 rcu_read_lock();
1155 {
1156 struct fib_table *tbl = NULL;
1157 struct flowi4 fl4 = {
1158 .daddr = nh->fib_nh_gw4,
1159 .flowi4_scope = scope + 1,
1160 .flowi4_oif = nh->fib_nh_oif,
1161 .flowi4_iif = LOOPBACK_IFINDEX,
1162 };
1163
1164 /* It is not necessary, but requires a bit of thinking */
1165 if (fl4.flowi4_scope < RT_SCOPE_LINK)
1166 fl4.flowi4_scope = RT_SCOPE_LINK;
1167
1168 if (table && table != RT_TABLE_MAIN)
1169 tbl = fib_get_table(net, table);
1170
1171 if (tbl)
1172 err = fib_table_lookup(tbl, &fl4, &res,
1173 FIB_LOOKUP_IGNORE_LINKSTATE |
1174 FIB_LOOKUP_NOREF);
1175
1176 /* on error or if no table given do full lookup. This
1177 * is needed for example when nexthops are in the local
1178 * table rather than the given table
1179 */
1180 if (!tbl || err) {
1181 err = fib_lookup(net, &fl4, &res,
1182 FIB_LOOKUP_IGNORE_LINKSTATE);
1183 }
1184
1185 if (err) {
1186 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1187 goto out;
1188 }
1189 }
1190
1191 err = -EINVAL;
1192 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1193 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1194 goto out;
1195 }
1196 nh->fib_nh_scope = res.scope;
1197 nh->fib_nh_oif = FIB_RES_OIF(res);
1198 nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1199 if (!dev) {
1200 NL_SET_ERR_MSG(extack,
1201 "No egress device for nexthop gateway");
1202 goto out;
1203 }
1204 netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1205 if (!netif_carrier_ok(dev))
1206 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1207 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1208out:
1209 rcu_read_unlock();
1210 return err;
1211}
1212
1213static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1214 struct netlink_ext_ack *extack)
1215{
1216 struct in_device *in_dev;
1217 int err;
1218
1219 if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1220 NL_SET_ERR_MSG(extack,
1221 "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1222 return -EINVAL;
1223 }
1224
1225 rcu_read_lock();
1226
1227 err = -ENODEV;
1228 in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1229 if (!in_dev)
1230 goto out;
1231 err = -ENETDOWN;
1232 if (!(in_dev->dev->flags & IFF_UP)) {
1233 NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1234 goto out;
1235 }
1236
1237 nh->fib_nh_dev = in_dev->dev;
1238 netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1239 nh->fib_nh_scope = RT_SCOPE_HOST;
1240 if (!netif_carrier_ok(nh->fib_nh_dev))
1241 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1242 err = 0;
1243out:
1244 rcu_read_unlock();
1245 return err;
1246}
1247
1248int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1249 struct netlink_ext_ack *extack)
1250{
1251 int err;
1252
1253 if (nh->fib_nh_gw_family == AF_INET)
1254 err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1255 else if (nh->fib_nh_gw_family == AF_INET6)
1256 err = fib_check_nh_v6_gw(net, nh, table, extack);
1257 else
1258 err = fib_check_nh_nongw(net, nh, extack);
1259
1260 return err;
1261}
1262
1263static struct hlist_head *
1264fib_info_laddrhash_bucket(const struct net *net, __be32 val)
1265{
1266 u32 slot = hash_32(net_hash_mix(net) ^ (__force u32)val,
1267 fib_info_hash_bits);
1268
1269 return &fib_info_laddrhash[slot];
1270}
1271
1272static void fib_info_hash_move(struct hlist_head *new_info_hash,
1273 struct hlist_head *new_laddrhash,
1274 unsigned int new_size)
1275{
1276 struct hlist_head *old_info_hash, *old_laddrhash;
1277 unsigned int old_size = fib_info_hash_size;
1278 unsigned int i;
1279
1280 spin_lock_bh(&fib_info_lock);
1281 old_info_hash = fib_info_hash;
1282 old_laddrhash = fib_info_laddrhash;
1283 fib_info_hash_size = new_size;
1284 fib_info_hash_bits = ilog2(new_size);
1285
1286 for (i = 0; i < old_size; i++) {
1287 struct hlist_head *head = &fib_info_hash[i];
1288 struct hlist_node *n;
1289 struct fib_info *fi;
1290
1291 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1292 struct hlist_head *dest;
1293 unsigned int new_hash;
1294
1295 new_hash = fib_info_hashfn(fi);
1296 dest = &new_info_hash[new_hash];
1297 hlist_add_head(&fi->fib_hash, dest);
1298 }
1299 }
1300 fib_info_hash = new_info_hash;
1301
1302 fib_info_laddrhash = new_laddrhash;
1303 for (i = 0; i < old_size; i++) {
1304 struct hlist_head *lhead = &old_laddrhash[i];
1305 struct hlist_node *n;
1306 struct fib_info *fi;
1307
1308 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1309 struct hlist_head *ldest;
1310
1311 ldest = fib_info_laddrhash_bucket(fi->fib_net,
1312 fi->fib_prefsrc);
1313 hlist_add_head(&fi->fib_lhash, ldest);
1314 }
1315 }
1316
1317 spin_unlock_bh(&fib_info_lock);
1318
1319 kvfree(old_info_hash);
1320 kvfree(old_laddrhash);
1321}
1322
1323__be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1324 unsigned char scope)
1325{
1326 struct fib_nh *nh;
1327
1328 if (nhc->nhc_family != AF_INET)
1329 return inet_select_addr(nhc->nhc_dev, 0, scope);
1330
1331 nh = container_of(nhc, struct fib_nh, nh_common);
1332 nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1333 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1334
1335 return nh->nh_saddr;
1336}
1337
1338__be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1339{
1340 struct fib_nh_common *nhc = res->nhc;
1341
1342 if (res->fi->fib_prefsrc)
1343 return res->fi->fib_prefsrc;
1344
1345 if (nhc->nhc_family == AF_INET) {
1346 struct fib_nh *nh;
1347
1348 nh = container_of(nhc, struct fib_nh, nh_common);
1349 if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1350 return nh->nh_saddr;
1351 }
1352
1353 return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1354}
1355
1356static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1357{
1358 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1359 fib_prefsrc != cfg->fc_dst) {
1360 u32 tb_id = cfg->fc_table;
1361 int rc;
1362
1363 if (tb_id == RT_TABLE_MAIN)
1364 tb_id = RT_TABLE_LOCAL;
1365
1366 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1367 fib_prefsrc, tb_id);
1368
1369 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1370 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1371 fib_prefsrc, RT_TABLE_LOCAL);
1372 }
1373
1374 if (rc != RTN_LOCAL)
1375 return false;
1376 }
1377 return true;
1378}
1379
1380struct fib_info *fib_create_info(struct fib_config *cfg,
1381 struct netlink_ext_ack *extack)
1382{
1383 int err;
1384 struct fib_info *fi = NULL;
1385 struct nexthop *nh = NULL;
1386 struct fib_info *ofi;
1387 int nhs = 1;
1388 struct net *net = cfg->fc_nlinfo.nl_net;
1389
1390 if (cfg->fc_type > RTN_MAX)
1391 goto err_inval;
1392
1393 /* Fast check to catch the most weird cases */
1394 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1395 NL_SET_ERR_MSG(extack, "Invalid scope");
1396 goto err_inval;
1397 }
1398
1399 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1400 NL_SET_ERR_MSG(extack,
1401 "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1402 goto err_inval;
1403 }
1404
1405 if (cfg->fc_nh_id) {
1406 if (!cfg->fc_mx) {
1407 fi = fib_find_info_nh(net, cfg);
1408 if (fi) {
1409 refcount_inc(&fi->fib_treeref);
1410 return fi;
1411 }
1412 }
1413
1414 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1415 if (!nh) {
1416 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1417 goto err_inval;
1418 }
1419 nhs = 0;
1420 }
1421
1422#ifdef CONFIG_IP_ROUTE_MULTIPATH
1423 if (cfg->fc_mp) {
1424 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1425 if (nhs == 0)
1426 goto err_inval;
1427 }
1428#endif
1429
1430 err = -ENOBUFS;
1431
1432 /* Paired with WRITE_ONCE() in fib_release_info() */
1433 if (READ_ONCE(fib_info_cnt) >= fib_info_hash_size) {
1434 unsigned int new_size = fib_info_hash_size << 1;
1435 struct hlist_head *new_info_hash;
1436 struct hlist_head *new_laddrhash;
1437 size_t bytes;
1438
1439 if (!new_size)
1440 new_size = 16;
1441 bytes = (size_t)new_size * sizeof(struct hlist_head *);
1442 new_info_hash = kvzalloc(bytes, GFP_KERNEL);
1443 new_laddrhash = kvzalloc(bytes, GFP_KERNEL);
1444 if (!new_info_hash || !new_laddrhash) {
1445 kvfree(new_info_hash);
1446 kvfree(new_laddrhash);
1447 } else {
1448 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1449 }
1450 if (!fib_info_hash_size)
1451 goto failure;
1452 }
1453
1454 fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1455 if (!fi)
1456 goto failure;
1457 fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1458 cfg->fc_mx_len, extack);
1459 if (IS_ERR(fi->fib_metrics)) {
1460 err = PTR_ERR(fi->fib_metrics);
1461 kfree(fi);
1462 return ERR_PTR(err);
1463 }
1464
1465 fi->fib_net = net;
1466 fi->fib_protocol = cfg->fc_protocol;
1467 fi->fib_scope = cfg->fc_scope;
1468 fi->fib_flags = cfg->fc_flags;
1469 fi->fib_priority = cfg->fc_priority;
1470 fi->fib_prefsrc = cfg->fc_prefsrc;
1471 fi->fib_type = cfg->fc_type;
1472 fi->fib_tb_id = cfg->fc_table;
1473
1474 fi->fib_nhs = nhs;
1475 if (nh) {
1476 if (!nexthop_get(nh)) {
1477 NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1478 err = -EINVAL;
1479 } else {
1480 err = 0;
1481 fi->nh = nh;
1482 }
1483 } else {
1484 change_nexthops(fi) {
1485 nexthop_nh->nh_parent = fi;
1486 } endfor_nexthops(fi)
1487
1488 if (cfg->fc_mp)
1489 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1490 extack);
1491 else
1492 err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1493 }
1494
1495 if (err != 0)
1496 goto failure;
1497
1498 if (fib_props[cfg->fc_type].error) {
1499 if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1500 NL_SET_ERR_MSG(extack,
1501 "Gateway, device and multipath can not be specified for this route type");
1502 goto err_inval;
1503 }
1504 goto link_it;
1505 } else {
1506 switch (cfg->fc_type) {
1507 case RTN_UNICAST:
1508 case RTN_LOCAL:
1509 case RTN_BROADCAST:
1510 case RTN_ANYCAST:
1511 case RTN_MULTICAST:
1512 break;
1513 default:
1514 NL_SET_ERR_MSG(extack, "Invalid route type");
1515 goto err_inval;
1516 }
1517 }
1518
1519 if (cfg->fc_scope > RT_SCOPE_HOST) {
1520 NL_SET_ERR_MSG(extack, "Invalid scope");
1521 goto err_inval;
1522 }
1523
1524 if (fi->nh) {
1525 err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1526 if (err)
1527 goto failure;
1528 } else if (cfg->fc_scope == RT_SCOPE_HOST) {
1529 struct fib_nh *nh = fi->fib_nh;
1530
1531 /* Local address is added. */
1532 if (nhs != 1) {
1533 NL_SET_ERR_MSG(extack,
1534 "Route with host scope can not have multiple nexthops");
1535 goto err_inval;
1536 }
1537 if (nh->fib_nh_gw_family) {
1538 NL_SET_ERR_MSG(extack,
1539 "Route with host scope can not have a gateway");
1540 goto err_inval;
1541 }
1542 nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1543 nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1544 err = -ENODEV;
1545 if (!nh->fib_nh_dev)
1546 goto failure;
1547 netdev_tracker_alloc(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1548 GFP_KERNEL);
1549 } else {
1550 int linkdown = 0;
1551
1552 change_nexthops(fi) {
1553 err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1554 cfg->fc_table, cfg->fc_scope,
1555 extack);
1556 if (err != 0)
1557 goto failure;
1558 if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1559 linkdown++;
1560 } endfor_nexthops(fi)
1561 if (linkdown == fi->fib_nhs)
1562 fi->fib_flags |= RTNH_F_LINKDOWN;
1563 }
1564
1565 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1566 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1567 goto err_inval;
1568 }
1569
1570 if (!fi->nh) {
1571 change_nexthops(fi) {
1572 fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1573 fi->fib_scope);
1574 if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1575 fi->fib_nh_is_v6 = true;
1576 } endfor_nexthops(fi)
1577
1578 fib_rebalance(fi);
1579 }
1580
1581link_it:
1582 ofi = fib_find_info(fi);
1583 if (ofi) {
1584 fi->fib_dead = 1;
1585 free_fib_info(fi);
1586 refcount_inc(&ofi->fib_treeref);
1587 return ofi;
1588 }
1589
1590 refcount_set(&fi->fib_treeref, 1);
1591 refcount_set(&fi->fib_clntref, 1);
1592 spin_lock_bh(&fib_info_lock);
1593 fib_info_cnt++;
1594 hlist_add_head(&fi->fib_hash,
1595 &fib_info_hash[fib_info_hashfn(fi)]);
1596 if (fi->fib_prefsrc) {
1597 struct hlist_head *head;
1598
1599 head = fib_info_laddrhash_bucket(net, fi->fib_prefsrc);
1600 hlist_add_head(&fi->fib_lhash, head);
1601 }
1602 if (fi->nh) {
1603 list_add(&fi->nh_list, &nh->fi_list);
1604 } else {
1605 change_nexthops(fi) {
1606 struct hlist_head *head;
1607
1608 if (!nexthop_nh->fib_nh_dev)
1609 continue;
1610 head = fib_info_devhash_bucket(nexthop_nh->fib_nh_dev);
1611 hlist_add_head(&nexthop_nh->nh_hash, head);
1612 } endfor_nexthops(fi)
1613 }
1614 spin_unlock_bh(&fib_info_lock);
1615 return fi;
1616
1617err_inval:
1618 err = -EINVAL;
1619
1620failure:
1621 if (fi) {
1622 fi->fib_dead = 1;
1623 free_fib_info(fi);
1624 }
1625
1626 return ERR_PTR(err);
1627}
1628
1629int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1630 u8 rt_family, unsigned char *flags, bool skip_oif)
1631{
1632 if (nhc->nhc_flags & RTNH_F_DEAD)
1633 *flags |= RTNH_F_DEAD;
1634
1635 if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1636 *flags |= RTNH_F_LINKDOWN;
1637
1638 rcu_read_lock();
1639 switch (nhc->nhc_family) {
1640 case AF_INET:
1641 if (ip_ignore_linkdown(nhc->nhc_dev))
1642 *flags |= RTNH_F_DEAD;
1643 break;
1644 case AF_INET6:
1645 if (ip6_ignore_linkdown(nhc->nhc_dev))
1646 *flags |= RTNH_F_DEAD;
1647 break;
1648 }
1649 rcu_read_unlock();
1650 }
1651
1652 switch (nhc->nhc_gw_family) {
1653 case AF_INET:
1654 if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1655 goto nla_put_failure;
1656 break;
1657 case AF_INET6:
1658 /* if gateway family does not match nexthop family
1659 * gateway is encoded as RTA_VIA
1660 */
1661 if (rt_family != nhc->nhc_gw_family) {
1662 int alen = sizeof(struct in6_addr);
1663 struct nlattr *nla;
1664 struct rtvia *via;
1665
1666 nla = nla_reserve(skb, RTA_VIA, alen + 2);
1667 if (!nla)
1668 goto nla_put_failure;
1669
1670 via = nla_data(nla);
1671 via->rtvia_family = AF_INET6;
1672 memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1673 } else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1674 &nhc->nhc_gw.ipv6) < 0) {
1675 goto nla_put_failure;
1676 }
1677 break;
1678 }
1679
1680 *flags |= (nhc->nhc_flags &
1681 (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP));
1682
1683 if (!skip_oif && nhc->nhc_dev &&
1684 nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1685 goto nla_put_failure;
1686
1687 if (nhc->nhc_lwtstate &&
1688 lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1689 RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1690 goto nla_put_failure;
1691
1692 return 0;
1693
1694nla_put_failure:
1695 return -EMSGSIZE;
1696}
1697EXPORT_SYMBOL_GPL(fib_nexthop_info);
1698
1699#if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1700int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1701 int nh_weight, u8 rt_family, u32 nh_tclassid)
1702{
1703 const struct net_device *dev = nhc->nhc_dev;
1704 struct rtnexthop *rtnh;
1705 unsigned char flags = 0;
1706
1707 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1708 if (!rtnh)
1709 goto nla_put_failure;
1710
1711 rtnh->rtnh_hops = nh_weight - 1;
1712 rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1713
1714 if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1715 goto nla_put_failure;
1716
1717 rtnh->rtnh_flags = flags;
1718
1719 if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1720 goto nla_put_failure;
1721
1722 /* length of rtnetlink header + attributes */
1723 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1724
1725 return 0;
1726
1727nla_put_failure:
1728 return -EMSGSIZE;
1729}
1730EXPORT_SYMBOL_GPL(fib_add_nexthop);
1731#endif
1732
1733#ifdef CONFIG_IP_ROUTE_MULTIPATH
1734static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1735{
1736 struct nlattr *mp;
1737
1738 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1739 if (!mp)
1740 goto nla_put_failure;
1741
1742 if (unlikely(fi->nh)) {
1743 if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1744 goto nla_put_failure;
1745 goto mp_end;
1746 }
1747
1748 for_nexthops(fi) {
1749 u32 nh_tclassid = 0;
1750#ifdef CONFIG_IP_ROUTE_CLASSID
1751 nh_tclassid = nh->nh_tclassid;
1752#endif
1753 if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1754 AF_INET, nh_tclassid) < 0)
1755 goto nla_put_failure;
1756 } endfor_nexthops(fi);
1757
1758mp_end:
1759 nla_nest_end(skb, mp);
1760
1761 return 0;
1762
1763nla_put_failure:
1764 return -EMSGSIZE;
1765}
1766#else
1767static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1768{
1769 return 0;
1770}
1771#endif
1772
1773int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1774 const struct fib_rt_info *fri, unsigned int flags)
1775{
1776 unsigned int nhs = fib_info_num_path(fri->fi);
1777 struct fib_info *fi = fri->fi;
1778 u32 tb_id = fri->tb_id;
1779 struct nlmsghdr *nlh;
1780 struct rtmsg *rtm;
1781
1782 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1783 if (!nlh)
1784 return -EMSGSIZE;
1785
1786 rtm = nlmsg_data(nlh);
1787 rtm->rtm_family = AF_INET;
1788 rtm->rtm_dst_len = fri->dst_len;
1789 rtm->rtm_src_len = 0;
1790 rtm->rtm_tos = inet_dscp_to_dsfield(fri->dscp);
1791 if (tb_id < 256)
1792 rtm->rtm_table = tb_id;
1793 else
1794 rtm->rtm_table = RT_TABLE_COMPAT;
1795 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1796 goto nla_put_failure;
1797 rtm->rtm_type = fri->type;
1798 rtm->rtm_flags = fi->fib_flags;
1799 rtm->rtm_scope = fi->fib_scope;
1800 rtm->rtm_protocol = fi->fib_protocol;
1801
1802 if (rtm->rtm_dst_len &&
1803 nla_put_in_addr(skb, RTA_DST, fri->dst))
1804 goto nla_put_failure;
1805 if (fi->fib_priority &&
1806 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1807 goto nla_put_failure;
1808 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1809 goto nla_put_failure;
1810
1811 if (fi->fib_prefsrc &&
1812 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1813 goto nla_put_failure;
1814
1815 if (fi->nh) {
1816 if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1817 goto nla_put_failure;
1818 if (nexthop_is_blackhole(fi->nh))
1819 rtm->rtm_type = RTN_BLACKHOLE;
1820 if (!READ_ONCE(fi->fib_net->ipv4.sysctl_nexthop_compat_mode))
1821 goto offload;
1822 }
1823
1824 if (nhs == 1) {
1825 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1826 unsigned char flags = 0;
1827
1828 if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1829 goto nla_put_failure;
1830
1831 rtm->rtm_flags = flags;
1832#ifdef CONFIG_IP_ROUTE_CLASSID
1833 if (nhc->nhc_family == AF_INET) {
1834 struct fib_nh *nh;
1835
1836 nh = container_of(nhc, struct fib_nh, nh_common);
1837 if (nh->nh_tclassid &&
1838 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1839 goto nla_put_failure;
1840 }
1841#endif
1842 } else {
1843 if (fib_add_multipath(skb, fi) < 0)
1844 goto nla_put_failure;
1845 }
1846
1847offload:
1848 if (fri->offload)
1849 rtm->rtm_flags |= RTM_F_OFFLOAD;
1850 if (fri->trap)
1851 rtm->rtm_flags |= RTM_F_TRAP;
1852 if (fri->offload_failed)
1853 rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
1854
1855 nlmsg_end(skb, nlh);
1856 return 0;
1857
1858nla_put_failure:
1859 nlmsg_cancel(skb, nlh);
1860 return -EMSGSIZE;
1861}
1862
1863/*
1864 * Update FIB if:
1865 * - local address disappeared -> we must delete all the entries
1866 * referring to it.
1867 * - device went down -> we must shutdown all nexthops going via it.
1868 */
1869int fib_sync_down_addr(struct net_device *dev, __be32 local)
1870{
1871 int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1872 struct net *net = dev_net(dev);
1873 struct hlist_head *head;
1874 struct fib_info *fi;
1875 int ret = 0;
1876
1877 if (!fib_info_laddrhash || local == 0)
1878 return 0;
1879
1880 head = fib_info_laddrhash_bucket(net, local);
1881 hlist_for_each_entry(fi, head, fib_lhash) {
1882 if (!net_eq(fi->fib_net, net) ||
1883 fi->fib_tb_id != tb_id)
1884 continue;
1885 if (fi->fib_prefsrc == local) {
1886 fi->fib_flags |= RTNH_F_DEAD;
1887 ret++;
1888 }
1889 }
1890 return ret;
1891}
1892
1893static int call_fib_nh_notifiers(struct fib_nh *nh,
1894 enum fib_event_type event_type)
1895{
1896 bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1897 struct fib_nh_notifier_info info = {
1898 .fib_nh = nh,
1899 };
1900
1901 switch (event_type) {
1902 case FIB_EVENT_NH_ADD:
1903 if (nh->fib_nh_flags & RTNH_F_DEAD)
1904 break;
1905 if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1906 break;
1907 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1908 &info.info);
1909 case FIB_EVENT_NH_DEL:
1910 if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1911 (nh->fib_nh_flags & RTNH_F_DEAD))
1912 return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1913 event_type, &info.info);
1914 break;
1915 default:
1916 break;
1917 }
1918
1919 return NOTIFY_DONE;
1920}
1921
1922/* Update the PMTU of exceptions when:
1923 * - the new MTU of the first hop becomes smaller than the PMTU
1924 * - the old MTU was the same as the PMTU, and it limited discovery of
1925 * larger MTUs on the path. With that limit raised, we can now
1926 * discover larger MTUs
1927 * A special case is locked exceptions, for which the PMTU is smaller
1928 * than the minimal accepted PMTU:
1929 * - if the new MTU is greater than the PMTU, don't make any change
1930 * - otherwise, unlock and set PMTU
1931 */
1932void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1933{
1934 struct fnhe_hash_bucket *bucket;
1935 int i;
1936
1937 bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1938 if (!bucket)
1939 return;
1940
1941 for (i = 0; i < FNHE_HASH_SIZE; i++) {
1942 struct fib_nh_exception *fnhe;
1943
1944 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1945 fnhe;
1946 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1947 if (fnhe->fnhe_mtu_locked) {
1948 if (new <= fnhe->fnhe_pmtu) {
1949 fnhe->fnhe_pmtu = new;
1950 fnhe->fnhe_mtu_locked = false;
1951 }
1952 } else if (new < fnhe->fnhe_pmtu ||
1953 orig == fnhe->fnhe_pmtu) {
1954 fnhe->fnhe_pmtu = new;
1955 }
1956 }
1957 }
1958}
1959
1960void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1961{
1962 struct hlist_head *head = fib_info_devhash_bucket(dev);
1963 struct fib_nh *nh;
1964
1965 hlist_for_each_entry(nh, head, nh_hash) {
1966 if (nh->fib_nh_dev == dev)
1967 fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1968 }
1969}
1970
1971/* Event force Flags Description
1972 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host
1973 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host
1974 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed
1975 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed
1976 *
1977 * only used when fib_nh is built into fib_info
1978 */
1979int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1980{
1981 struct hlist_head *head = fib_info_devhash_bucket(dev);
1982 struct fib_info *prev_fi = NULL;
1983 int scope = RT_SCOPE_NOWHERE;
1984 struct fib_nh *nh;
1985 int ret = 0;
1986
1987 if (force)
1988 scope = -1;
1989
1990 hlist_for_each_entry(nh, head, nh_hash) {
1991 struct fib_info *fi = nh->nh_parent;
1992 int dead;
1993
1994 BUG_ON(!fi->fib_nhs);
1995 if (nh->fib_nh_dev != dev || fi == prev_fi)
1996 continue;
1997 prev_fi = fi;
1998 dead = 0;
1999 change_nexthops(fi) {
2000 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
2001 dead++;
2002 else if (nexthop_nh->fib_nh_dev == dev &&
2003 nexthop_nh->fib_nh_scope != scope) {
2004 switch (event) {
2005 case NETDEV_DOWN:
2006 case NETDEV_UNREGISTER:
2007 nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
2008 fallthrough;
2009 case NETDEV_CHANGE:
2010 nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
2011 break;
2012 }
2013 call_fib_nh_notifiers(nexthop_nh,
2014 FIB_EVENT_NH_DEL);
2015 dead++;
2016 }
2017#ifdef CONFIG_IP_ROUTE_MULTIPATH
2018 if (event == NETDEV_UNREGISTER &&
2019 nexthop_nh->fib_nh_dev == dev) {
2020 dead = fi->fib_nhs;
2021 break;
2022 }
2023#endif
2024 } endfor_nexthops(fi)
2025 if (dead == fi->fib_nhs) {
2026 switch (event) {
2027 case NETDEV_DOWN:
2028 case NETDEV_UNREGISTER:
2029 fi->fib_flags |= RTNH_F_DEAD;
2030 fallthrough;
2031 case NETDEV_CHANGE:
2032 fi->fib_flags |= RTNH_F_LINKDOWN;
2033 break;
2034 }
2035 ret++;
2036 }
2037
2038 fib_rebalance(fi);
2039 }
2040
2041 return ret;
2042}
2043
2044/* Must be invoked inside of an RCU protected region. */
2045static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2046{
2047 struct fib_info *fi = NULL, *last_resort = NULL;
2048 struct hlist_head *fa_head = res->fa_head;
2049 struct fib_table *tb = res->table;
2050 u8 slen = 32 - res->prefixlen;
2051 int order = -1, last_idx = -1;
2052 struct fib_alias *fa, *fa1 = NULL;
2053 u32 last_prio = res->fi->fib_priority;
2054 dscp_t last_dscp = 0;
2055
2056 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2057 struct fib_info *next_fi = fa->fa_info;
2058 struct fib_nh_common *nhc;
2059
2060 if (fa->fa_slen != slen)
2061 continue;
2062 if (fa->fa_dscp &&
2063 fa->fa_dscp != inet_dsfield_to_dscp(flp->flowi4_tos))
2064 continue;
2065 if (fa->tb_id != tb->tb_id)
2066 continue;
2067 if (next_fi->fib_priority > last_prio &&
2068 fa->fa_dscp == last_dscp) {
2069 if (last_dscp)
2070 continue;
2071 break;
2072 }
2073 if (next_fi->fib_flags & RTNH_F_DEAD)
2074 continue;
2075 last_dscp = fa->fa_dscp;
2076 last_prio = next_fi->fib_priority;
2077
2078 if (next_fi->fib_scope != res->scope ||
2079 fa->fa_type != RTN_UNICAST)
2080 continue;
2081
2082 nhc = fib_info_nhc(next_fi, 0);
2083 if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2084 continue;
2085
2086 fib_alias_accessed(fa);
2087
2088 if (!fi) {
2089 if (next_fi != res->fi)
2090 break;
2091 fa1 = fa;
2092 } else if (!fib_detect_death(fi, order, &last_resort,
2093 &last_idx, fa1->fa_default)) {
2094 fib_result_assign(res, fi);
2095 fa1->fa_default = order;
2096 goto out;
2097 }
2098 fi = next_fi;
2099 order++;
2100 }
2101
2102 if (order <= 0 || !fi) {
2103 if (fa1)
2104 fa1->fa_default = -1;
2105 goto out;
2106 }
2107
2108 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2109 fa1->fa_default)) {
2110 fib_result_assign(res, fi);
2111 fa1->fa_default = order;
2112 goto out;
2113 }
2114
2115 if (last_idx >= 0)
2116 fib_result_assign(res, last_resort);
2117 fa1->fa_default = last_idx;
2118out:
2119 return;
2120}
2121
2122/*
2123 * Dead device goes up. We wake up dead nexthops.
2124 * It takes sense only on multipath routes.
2125 *
2126 * only used when fib_nh is built into fib_info
2127 */
2128int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2129{
2130 struct fib_info *prev_fi;
2131 struct hlist_head *head;
2132 struct fib_nh *nh;
2133 int ret;
2134
2135 if (!(dev->flags & IFF_UP))
2136 return 0;
2137
2138 if (nh_flags & RTNH_F_DEAD) {
2139 unsigned int flags = dev_get_flags(dev);
2140
2141 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2142 nh_flags |= RTNH_F_LINKDOWN;
2143 }
2144
2145 prev_fi = NULL;
2146 head = fib_info_devhash_bucket(dev);
2147 ret = 0;
2148
2149 hlist_for_each_entry(nh, head, nh_hash) {
2150 struct fib_info *fi = nh->nh_parent;
2151 int alive;
2152
2153 BUG_ON(!fi->fib_nhs);
2154 if (nh->fib_nh_dev != dev || fi == prev_fi)
2155 continue;
2156
2157 prev_fi = fi;
2158 alive = 0;
2159 change_nexthops(fi) {
2160 if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2161 alive++;
2162 continue;
2163 }
2164 if (!nexthop_nh->fib_nh_dev ||
2165 !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2166 continue;
2167 if (nexthop_nh->fib_nh_dev != dev ||
2168 !__in_dev_get_rtnl(dev))
2169 continue;
2170 alive++;
2171 nexthop_nh->fib_nh_flags &= ~nh_flags;
2172 call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2173 } endfor_nexthops(fi)
2174
2175 if (alive > 0) {
2176 fi->fib_flags &= ~nh_flags;
2177 ret++;
2178 }
2179
2180 fib_rebalance(fi);
2181 }
2182
2183 return ret;
2184}
2185
2186#ifdef CONFIG_IP_ROUTE_MULTIPATH
2187static bool fib_good_nh(const struct fib_nh *nh)
2188{
2189 int state = NUD_REACHABLE;
2190
2191 if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2192 struct neighbour *n;
2193
2194 rcu_read_lock_bh();
2195
2196 if (likely(nh->fib_nh_gw_family == AF_INET))
2197 n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2198 (__force u32)nh->fib_nh_gw4);
2199 else if (nh->fib_nh_gw_family == AF_INET6)
2200 n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2201 &nh->fib_nh_gw6);
2202 else
2203 n = NULL;
2204 if (n)
2205 state = n->nud_state;
2206
2207 rcu_read_unlock_bh();
2208 }
2209
2210 return !!(state & NUD_VALID);
2211}
2212
2213void fib_select_multipath(struct fib_result *res, int hash)
2214{
2215 struct fib_info *fi = res->fi;
2216 struct net *net = fi->fib_net;
2217 bool first = false;
2218
2219 if (unlikely(res->fi->nh)) {
2220 nexthop_path_fib_result(res, hash);
2221 return;
2222 }
2223
2224 change_nexthops(fi) {
2225 if (READ_ONCE(net->ipv4.sysctl_fib_multipath_use_neigh)) {
2226 if (!fib_good_nh(nexthop_nh))
2227 continue;
2228 if (!first) {
2229 res->nh_sel = nhsel;
2230 res->nhc = &nexthop_nh->nh_common;
2231 first = true;
2232 }
2233 }
2234
2235 if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2236 continue;
2237
2238 res->nh_sel = nhsel;
2239 res->nhc = &nexthop_nh->nh_common;
2240 return;
2241 } endfor_nexthops(fi);
2242}
2243#endif
2244
2245void fib_select_path(struct net *net, struct fib_result *res,
2246 struct flowi4 *fl4, const struct sk_buff *skb)
2247{
2248 if (fl4->flowi4_oif)
2249 goto check_saddr;
2250
2251#ifdef CONFIG_IP_ROUTE_MULTIPATH
2252 if (fib_info_num_path(res->fi) > 1) {
2253 int h = fib_multipath_hash(net, fl4, skb, NULL);
2254
2255 fib_select_multipath(res, h);
2256 }
2257 else
2258#endif
2259 if (!res->prefixlen &&
2260 res->table->tb_num_default > 1 &&
2261 res->type == RTN_UNICAST)
2262 fib_select_default(fl4, res);
2263
2264check_saddr:
2265 if (!fl4->saddr)
2266 fl4->saddr = fib_result_prefsrc(net, res);
2267}