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 <asm/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
36#include <net/arp.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/netlink.h>
44#include <net/nexthop.h>
45#include <net/lwtunnel.h>
46
47#include "fib_lookup.h"
48
49static DEFINE_SPINLOCK(fib_info_lock);
50static struct hlist_head *fib_info_hash;
51static struct hlist_head *fib_info_laddrhash;
52static unsigned int fib_info_hash_size;
53static unsigned int fib_info_cnt;
54
55#define DEVINDEX_HASHBITS 8
56#define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
57static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
58
59#ifdef CONFIG_IP_ROUTE_MULTIPATH
60u32 fib_multipath_secret __read_mostly;
61
62#define for_nexthops(fi) { \
63 int nhsel; const struct fib_nh *nh; \
64 for (nhsel = 0, nh = (fi)->fib_nh; \
65 nhsel < (fi)->fib_nhs; \
66 nh++, nhsel++)
67
68#define change_nexthops(fi) { \
69 int nhsel; struct fib_nh *nexthop_nh; \
70 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
71 nhsel < (fi)->fib_nhs; \
72 nexthop_nh++, nhsel++)
73
74#else /* CONFIG_IP_ROUTE_MULTIPATH */
75
76/* Hope, that gcc will optimize it to get rid of dummy loop */
77
78#define for_nexthops(fi) { \
79 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \
80 for (nhsel = 0; nhsel < 1; nhsel++)
81
82#define change_nexthops(fi) { \
83 int nhsel; \
84 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
85 for (nhsel = 0; nhsel < 1; nhsel++)
86
87#endif /* CONFIG_IP_ROUTE_MULTIPATH */
88
89#define endfor_nexthops(fi) }
90
91
92const struct fib_prop fib_props[RTN_MAX + 1] = {
93 [RTN_UNSPEC] = {
94 .error = 0,
95 .scope = RT_SCOPE_NOWHERE,
96 },
97 [RTN_UNICAST] = {
98 .error = 0,
99 .scope = RT_SCOPE_UNIVERSE,
100 },
101 [RTN_LOCAL] = {
102 .error = 0,
103 .scope = RT_SCOPE_HOST,
104 },
105 [RTN_BROADCAST] = {
106 .error = 0,
107 .scope = RT_SCOPE_LINK,
108 },
109 [RTN_ANYCAST] = {
110 .error = 0,
111 .scope = RT_SCOPE_LINK,
112 },
113 [RTN_MULTICAST] = {
114 .error = 0,
115 .scope = RT_SCOPE_UNIVERSE,
116 },
117 [RTN_BLACKHOLE] = {
118 .error = -EINVAL,
119 .scope = RT_SCOPE_UNIVERSE,
120 },
121 [RTN_UNREACHABLE] = {
122 .error = -EHOSTUNREACH,
123 .scope = RT_SCOPE_UNIVERSE,
124 },
125 [RTN_PROHIBIT] = {
126 .error = -EACCES,
127 .scope = RT_SCOPE_UNIVERSE,
128 },
129 [RTN_THROW] = {
130 .error = -EAGAIN,
131 .scope = RT_SCOPE_UNIVERSE,
132 },
133 [RTN_NAT] = {
134 .error = -EINVAL,
135 .scope = RT_SCOPE_NOWHERE,
136 },
137 [RTN_XRESOLVE] = {
138 .error = -EINVAL,
139 .scope = RT_SCOPE_NOWHERE,
140 },
141};
142
143static void rt_fibinfo_free(struct rtable __rcu **rtp)
144{
145 struct rtable *rt = rcu_dereference_protected(*rtp, 1);
146
147 if (!rt)
148 return;
149
150 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
151 * because we waited an RCU grace period before calling
152 * free_fib_info_rcu()
153 */
154
155 dst_free(&rt->dst);
156}
157
158static void free_nh_exceptions(struct fib_nh *nh)
159{
160 struct fnhe_hash_bucket *hash;
161 int i;
162
163 hash = rcu_dereference_protected(nh->nh_exceptions, 1);
164 if (!hash)
165 return;
166 for (i = 0; i < FNHE_HASH_SIZE; i++) {
167 struct fib_nh_exception *fnhe;
168
169 fnhe = rcu_dereference_protected(hash[i].chain, 1);
170 while (fnhe) {
171 struct fib_nh_exception *next;
172
173 next = rcu_dereference_protected(fnhe->fnhe_next, 1);
174
175 rt_fibinfo_free(&fnhe->fnhe_rth_input);
176 rt_fibinfo_free(&fnhe->fnhe_rth_output);
177
178 kfree(fnhe);
179
180 fnhe = next;
181 }
182 }
183 kfree(hash);
184}
185
186static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
187{
188 int cpu;
189
190 if (!rtp)
191 return;
192
193 for_each_possible_cpu(cpu) {
194 struct rtable *rt;
195
196 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
197 if (rt)
198 dst_free(&rt->dst);
199 }
200 free_percpu(rtp);
201}
202
203/* Release a nexthop info record */
204static void free_fib_info_rcu(struct rcu_head *head)
205{
206 struct fib_info *fi = container_of(head, struct fib_info, rcu);
207
208 change_nexthops(fi) {
209 if (nexthop_nh->nh_dev)
210 dev_put(nexthop_nh->nh_dev);
211 lwtstate_put(nexthop_nh->nh_lwtstate);
212 free_nh_exceptions(nexthop_nh);
213 rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
214 rt_fibinfo_free(&nexthop_nh->nh_rth_input);
215 } endfor_nexthops(fi);
216
217 if (fi->fib_metrics != (u32 *) dst_default_metrics)
218 kfree(fi->fib_metrics);
219 kfree(fi);
220}
221
222void free_fib_info(struct fib_info *fi)
223{
224 if (fi->fib_dead == 0) {
225 pr_warn("Freeing alive fib_info %p\n", fi);
226 return;
227 }
228 fib_info_cnt--;
229#ifdef CONFIG_IP_ROUTE_CLASSID
230 change_nexthops(fi) {
231 if (nexthop_nh->nh_tclassid)
232 fi->fib_net->ipv4.fib_num_tclassid_users--;
233 } endfor_nexthops(fi);
234#endif
235 call_rcu(&fi->rcu, free_fib_info_rcu);
236}
237
238void fib_release_info(struct fib_info *fi)
239{
240 spin_lock_bh(&fib_info_lock);
241 if (fi && --fi->fib_treeref == 0) {
242 hlist_del(&fi->fib_hash);
243 if (fi->fib_prefsrc)
244 hlist_del(&fi->fib_lhash);
245 change_nexthops(fi) {
246 if (!nexthop_nh->nh_dev)
247 continue;
248 hlist_del(&nexthop_nh->nh_hash);
249 } endfor_nexthops(fi)
250 fi->fib_dead = 1;
251 fib_info_put(fi);
252 }
253 spin_unlock_bh(&fib_info_lock);
254}
255
256static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
257{
258 const struct fib_nh *onh = ofi->fib_nh;
259
260 for_nexthops(fi) {
261 if (nh->nh_oif != onh->nh_oif ||
262 nh->nh_gw != onh->nh_gw ||
263 nh->nh_scope != onh->nh_scope ||
264#ifdef CONFIG_IP_ROUTE_MULTIPATH
265 nh->nh_weight != onh->nh_weight ||
266#endif
267#ifdef CONFIG_IP_ROUTE_CLASSID
268 nh->nh_tclassid != onh->nh_tclassid ||
269#endif
270 lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) ||
271 ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
272 return -1;
273 onh++;
274 } endfor_nexthops(fi);
275 return 0;
276}
277
278static inline unsigned int fib_devindex_hashfn(unsigned int val)
279{
280 unsigned int mask = DEVINDEX_HASHSIZE - 1;
281
282 return (val ^
283 (val >> DEVINDEX_HASHBITS) ^
284 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
285}
286
287static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
288{
289 unsigned int mask = (fib_info_hash_size - 1);
290 unsigned int val = fi->fib_nhs;
291
292 val ^= (fi->fib_protocol << 8) | fi->fib_scope;
293 val ^= (__force u32)fi->fib_prefsrc;
294 val ^= fi->fib_priority;
295 for_nexthops(fi) {
296 val ^= fib_devindex_hashfn(nh->nh_oif);
297 } endfor_nexthops(fi)
298
299 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
300}
301
302static struct fib_info *fib_find_info(const struct fib_info *nfi)
303{
304 struct hlist_head *head;
305 struct fib_info *fi;
306 unsigned int hash;
307
308 hash = fib_info_hashfn(nfi);
309 head = &fib_info_hash[hash];
310
311 hlist_for_each_entry(fi, head, fib_hash) {
312 if (!net_eq(fi->fib_net, nfi->fib_net))
313 continue;
314 if (fi->fib_nhs != nfi->fib_nhs)
315 continue;
316 if (nfi->fib_protocol == fi->fib_protocol &&
317 nfi->fib_scope == fi->fib_scope &&
318 nfi->fib_prefsrc == fi->fib_prefsrc &&
319 nfi->fib_priority == fi->fib_priority &&
320 nfi->fib_type == fi->fib_type &&
321 memcmp(nfi->fib_metrics, fi->fib_metrics,
322 sizeof(u32) * RTAX_MAX) == 0 &&
323 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
324 (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
325 return fi;
326 }
327
328 return NULL;
329}
330
331/* Check, that the gateway is already configured.
332 * Used only by redirect accept routine.
333 */
334int ip_fib_check_default(__be32 gw, struct net_device *dev)
335{
336 struct hlist_head *head;
337 struct fib_nh *nh;
338 unsigned int hash;
339
340 spin_lock(&fib_info_lock);
341
342 hash = fib_devindex_hashfn(dev->ifindex);
343 head = &fib_info_devhash[hash];
344 hlist_for_each_entry(nh, head, nh_hash) {
345 if (nh->nh_dev == dev &&
346 nh->nh_gw == gw &&
347 !(nh->nh_flags & RTNH_F_DEAD)) {
348 spin_unlock(&fib_info_lock);
349 return 0;
350 }
351 }
352
353 spin_unlock(&fib_info_lock);
354
355 return -1;
356}
357
358static inline size_t fib_nlmsg_size(struct fib_info *fi)
359{
360 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
361 + nla_total_size(4) /* RTA_TABLE */
362 + nla_total_size(4) /* RTA_DST */
363 + nla_total_size(4) /* RTA_PRIORITY */
364 + nla_total_size(4) /* RTA_PREFSRC */
365 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
366
367 /* space for nested metrics */
368 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
369
370 if (fi->fib_nhs) {
371 size_t nh_encapsize = 0;
372 /* Also handles the special case fib_nhs == 1 */
373
374 /* each nexthop is packed in an attribute */
375 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
376
377 /* may contain flow and gateway attribute */
378 nhsize += 2 * nla_total_size(4);
379
380 /* grab encap info */
381 for_nexthops(fi) {
382 if (nh->nh_lwtstate) {
383 /* RTA_ENCAP_TYPE */
384 nh_encapsize += lwtunnel_get_encap_size(
385 nh->nh_lwtstate);
386 /* RTA_ENCAP */
387 nh_encapsize += nla_total_size(2);
388 }
389 } endfor_nexthops(fi);
390
391 /* all nexthops are packed in a nested attribute */
392 payload += nla_total_size((fi->fib_nhs * nhsize) +
393 nh_encapsize);
394
395 }
396
397 return payload;
398}
399
400void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
401 int dst_len, u32 tb_id, const struct nl_info *info,
402 unsigned int nlm_flags)
403{
404 struct sk_buff *skb;
405 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
406 int err = -ENOBUFS;
407
408 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
409 if (!skb)
410 goto errout;
411
412 err = fib_dump_info(skb, info->portid, seq, event, tb_id,
413 fa->fa_type, key, dst_len,
414 fa->fa_tos, fa->fa_info, nlm_flags);
415 if (err < 0) {
416 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
417 WARN_ON(err == -EMSGSIZE);
418 kfree_skb(skb);
419 goto errout;
420 }
421 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
422 info->nlh, GFP_KERNEL);
423 return;
424errout:
425 if (err < 0)
426 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
427}
428
429static int fib_detect_death(struct fib_info *fi, int order,
430 struct fib_info **last_resort, int *last_idx,
431 int dflt)
432{
433 struct neighbour *n;
434 int state = NUD_NONE;
435
436 n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
437 if (n) {
438 state = n->nud_state;
439 neigh_release(n);
440 } else {
441 return 0;
442 }
443 if (state == NUD_REACHABLE)
444 return 0;
445 if ((state & NUD_VALID) && order != dflt)
446 return 0;
447 if ((state & NUD_VALID) ||
448 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
449 *last_resort = fi;
450 *last_idx = order;
451 }
452 return 1;
453}
454
455#ifdef CONFIG_IP_ROUTE_MULTIPATH
456
457static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
458{
459 int nhs = 0;
460
461 while (rtnh_ok(rtnh, remaining)) {
462 nhs++;
463 rtnh = rtnh_next(rtnh, &remaining);
464 }
465
466 /* leftover implies invalid nexthop configuration, discard it */
467 return remaining > 0 ? 0 : nhs;
468}
469
470static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
471 int remaining, struct fib_config *cfg)
472{
473 struct net *net = cfg->fc_nlinfo.nl_net;
474 int ret;
475
476 change_nexthops(fi) {
477 int attrlen;
478
479 if (!rtnh_ok(rtnh, remaining))
480 return -EINVAL;
481
482 nexthop_nh->nh_flags =
483 (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
484 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
485 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
486
487 attrlen = rtnh_attrlen(rtnh);
488 if (attrlen > 0) {
489 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
490
491 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
492 nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
493#ifdef CONFIG_IP_ROUTE_CLASSID
494 nla = nla_find(attrs, attrlen, RTA_FLOW);
495 nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
496 if (nexthop_nh->nh_tclassid)
497 fi->fib_net->ipv4.fib_num_tclassid_users++;
498#endif
499 nla = nla_find(attrs, attrlen, RTA_ENCAP);
500 if (nla) {
501 struct lwtunnel_state *lwtstate;
502 struct net_device *dev = NULL;
503 struct nlattr *nla_entype;
504
505 nla_entype = nla_find(attrs, attrlen,
506 RTA_ENCAP_TYPE);
507 if (!nla_entype)
508 goto err_inval;
509 if (cfg->fc_oif)
510 dev = __dev_get_by_index(net, cfg->fc_oif);
511 ret = lwtunnel_build_state(dev, nla_get_u16(
512 nla_entype),
513 nla, AF_INET, cfg,
514 &lwtstate);
515 if (ret)
516 goto errout;
517 nexthop_nh->nh_lwtstate =
518 lwtstate_get(lwtstate);
519 }
520 }
521
522 rtnh = rtnh_next(rtnh, &remaining);
523 } endfor_nexthops(fi);
524
525 return 0;
526
527err_inval:
528 ret = -EINVAL;
529
530errout:
531 return ret;
532}
533
534static void fib_rebalance(struct fib_info *fi)
535{
536 int total;
537 int w;
538 struct in_device *in_dev;
539
540 if (fi->fib_nhs < 2)
541 return;
542
543 total = 0;
544 for_nexthops(fi) {
545 if (nh->nh_flags & RTNH_F_DEAD)
546 continue;
547
548 in_dev = __in_dev_get_rtnl(nh->nh_dev);
549
550 if (in_dev &&
551 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
552 nh->nh_flags & RTNH_F_LINKDOWN)
553 continue;
554
555 total += nh->nh_weight;
556 } endfor_nexthops(fi);
557
558 w = 0;
559 change_nexthops(fi) {
560 int upper_bound;
561
562 in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev);
563
564 if (nexthop_nh->nh_flags & RTNH_F_DEAD) {
565 upper_bound = -1;
566 } else if (in_dev &&
567 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
568 nexthop_nh->nh_flags & RTNH_F_LINKDOWN) {
569 upper_bound = -1;
570 } else {
571 w += nexthop_nh->nh_weight;
572 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
573 total) - 1;
574 }
575
576 atomic_set(&nexthop_nh->nh_upper_bound, upper_bound);
577 } endfor_nexthops(fi);
578
579 net_get_random_once(&fib_multipath_secret,
580 sizeof(fib_multipath_secret));
581}
582
583static inline void fib_add_weight(struct fib_info *fi,
584 const struct fib_nh *nh)
585{
586 fi->fib_weight += nh->nh_weight;
587}
588
589#else /* CONFIG_IP_ROUTE_MULTIPATH */
590
591#define fib_rebalance(fi) do { } while (0)
592#define fib_add_weight(fi, nh) do { } while (0)
593
594#endif /* CONFIG_IP_ROUTE_MULTIPATH */
595
596static int fib_encap_match(struct net *net, u16 encap_type,
597 struct nlattr *encap,
598 int oif, const struct fib_nh *nh,
599 const struct fib_config *cfg)
600{
601 struct lwtunnel_state *lwtstate;
602 struct net_device *dev = NULL;
603 int ret, result = 0;
604
605 if (encap_type == LWTUNNEL_ENCAP_NONE)
606 return 0;
607
608 if (oif)
609 dev = __dev_get_by_index(net, oif);
610 ret = lwtunnel_build_state(dev, encap_type, encap,
611 AF_INET, cfg, &lwtstate);
612 if (!ret) {
613 result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
614 lwtstate_free(lwtstate);
615 }
616
617 return result;
618}
619
620int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
621{
622 struct net *net = cfg->fc_nlinfo.nl_net;
623#ifdef CONFIG_IP_ROUTE_MULTIPATH
624 struct rtnexthop *rtnh;
625 int remaining;
626#endif
627
628 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
629 return 1;
630
631 if (cfg->fc_oif || cfg->fc_gw) {
632 if (cfg->fc_encap) {
633 if (fib_encap_match(net, cfg->fc_encap_type,
634 cfg->fc_encap, cfg->fc_oif,
635 fi->fib_nh, cfg))
636 return 1;
637 }
638 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
639 (!cfg->fc_gw || cfg->fc_gw == fi->fib_nh->nh_gw))
640 return 0;
641 return 1;
642 }
643
644#ifdef CONFIG_IP_ROUTE_MULTIPATH
645 if (!cfg->fc_mp)
646 return 0;
647
648 rtnh = cfg->fc_mp;
649 remaining = cfg->fc_mp_len;
650
651 for_nexthops(fi) {
652 int attrlen;
653
654 if (!rtnh_ok(rtnh, remaining))
655 return -EINVAL;
656
657 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
658 return 1;
659
660 attrlen = rtnh_attrlen(rtnh);
661 if (attrlen > 0) {
662 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
663
664 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
665 if (nla && nla_get_in_addr(nla) != nh->nh_gw)
666 return 1;
667#ifdef CONFIG_IP_ROUTE_CLASSID
668 nla = nla_find(attrs, attrlen, RTA_FLOW);
669 if (nla && nla_get_u32(nla) != nh->nh_tclassid)
670 return 1;
671#endif
672 }
673
674 rtnh = rtnh_next(rtnh, &remaining);
675 } endfor_nexthops(fi);
676#endif
677 return 0;
678}
679
680
681/*
682 * Picture
683 * -------
684 *
685 * Semantics of nexthop is very messy by historical reasons.
686 * We have to take into account, that:
687 * a) gateway can be actually local interface address,
688 * so that gatewayed route is direct.
689 * b) gateway must be on-link address, possibly
690 * described not by an ifaddr, but also by a direct route.
691 * c) If both gateway and interface are specified, they should not
692 * contradict.
693 * d) If we use tunnel routes, gateway could be not on-link.
694 *
695 * Attempt to reconcile all of these (alas, self-contradictory) conditions
696 * results in pretty ugly and hairy code with obscure logic.
697 *
698 * I chose to generalized it instead, so that the size
699 * of code does not increase practically, but it becomes
700 * much more general.
701 * Every prefix is assigned a "scope" value: "host" is local address,
702 * "link" is direct route,
703 * [ ... "site" ... "interior" ... ]
704 * and "universe" is true gateway route with global meaning.
705 *
706 * Every prefix refers to a set of "nexthop"s (gw, oif),
707 * where gw must have narrower scope. This recursion stops
708 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
709 * which means that gw is forced to be on link.
710 *
711 * Code is still hairy, but now it is apparently logically
712 * consistent and very flexible. F.e. as by-product it allows
713 * to co-exists in peace independent exterior and interior
714 * routing processes.
715 *
716 * Normally it looks as following.
717 *
718 * {universe prefix} -> (gw, oif) [scope link]
719 * |
720 * |-> {link prefix} -> (gw, oif) [scope local]
721 * |
722 * |-> {local prefix} (terminal node)
723 */
724static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
725 struct fib_nh *nh)
726{
727 int err = 0;
728 struct net *net;
729 struct net_device *dev;
730
731 net = cfg->fc_nlinfo.nl_net;
732 if (nh->nh_gw) {
733 struct fib_result res;
734
735 if (nh->nh_flags & RTNH_F_ONLINK) {
736 unsigned int addr_type;
737
738 if (cfg->fc_scope >= RT_SCOPE_LINK)
739 return -EINVAL;
740 dev = __dev_get_by_index(net, nh->nh_oif);
741 if (!dev)
742 return -ENODEV;
743 if (!(dev->flags & IFF_UP))
744 return -ENETDOWN;
745 addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
746 if (addr_type != RTN_UNICAST)
747 return -EINVAL;
748 if (!netif_carrier_ok(dev))
749 nh->nh_flags |= RTNH_F_LINKDOWN;
750 nh->nh_dev = dev;
751 dev_hold(dev);
752 nh->nh_scope = RT_SCOPE_LINK;
753 return 0;
754 }
755 rcu_read_lock();
756 {
757 struct fib_table *tbl = NULL;
758 struct flowi4 fl4 = {
759 .daddr = nh->nh_gw,
760 .flowi4_scope = cfg->fc_scope + 1,
761 .flowi4_oif = nh->nh_oif,
762 .flowi4_iif = LOOPBACK_IFINDEX,
763 };
764
765 /* It is not necessary, but requires a bit of thinking */
766 if (fl4.flowi4_scope < RT_SCOPE_LINK)
767 fl4.flowi4_scope = RT_SCOPE_LINK;
768
769 if (cfg->fc_table)
770 tbl = fib_get_table(net, cfg->fc_table);
771
772 if (tbl)
773 err = fib_table_lookup(tbl, &fl4, &res,
774 FIB_LOOKUP_IGNORE_LINKSTATE |
775 FIB_LOOKUP_NOREF);
776
777 /* on error or if no table given do full lookup. This
778 * is needed for example when nexthops are in the local
779 * table rather than the given table
780 */
781 if (!tbl || err) {
782 err = fib_lookup(net, &fl4, &res,
783 FIB_LOOKUP_IGNORE_LINKSTATE);
784 }
785
786 if (err) {
787 rcu_read_unlock();
788 return err;
789 }
790 }
791 err = -EINVAL;
792 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
793 goto out;
794 nh->nh_scope = res.scope;
795 nh->nh_oif = FIB_RES_OIF(res);
796 nh->nh_dev = dev = FIB_RES_DEV(res);
797 if (!dev)
798 goto out;
799 dev_hold(dev);
800 if (!netif_carrier_ok(dev))
801 nh->nh_flags |= RTNH_F_LINKDOWN;
802 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
803 } else {
804 struct in_device *in_dev;
805
806 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
807 return -EINVAL;
808
809 rcu_read_lock();
810 err = -ENODEV;
811 in_dev = inetdev_by_index(net, nh->nh_oif);
812 if (!in_dev)
813 goto out;
814 err = -ENETDOWN;
815 if (!(in_dev->dev->flags & IFF_UP))
816 goto out;
817 nh->nh_dev = in_dev->dev;
818 dev_hold(nh->nh_dev);
819 nh->nh_scope = RT_SCOPE_HOST;
820 if (!netif_carrier_ok(nh->nh_dev))
821 nh->nh_flags |= RTNH_F_LINKDOWN;
822 err = 0;
823 }
824out:
825 rcu_read_unlock();
826 return err;
827}
828
829static inline unsigned int fib_laddr_hashfn(__be32 val)
830{
831 unsigned int mask = (fib_info_hash_size - 1);
832
833 return ((__force u32)val ^
834 ((__force u32)val >> 7) ^
835 ((__force u32)val >> 14)) & mask;
836}
837
838static struct hlist_head *fib_info_hash_alloc(int bytes)
839{
840 if (bytes <= PAGE_SIZE)
841 return kzalloc(bytes, GFP_KERNEL);
842 else
843 return (struct hlist_head *)
844 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
845 get_order(bytes));
846}
847
848static void fib_info_hash_free(struct hlist_head *hash, int bytes)
849{
850 if (!hash)
851 return;
852
853 if (bytes <= PAGE_SIZE)
854 kfree(hash);
855 else
856 free_pages((unsigned long) hash, get_order(bytes));
857}
858
859static void fib_info_hash_move(struct hlist_head *new_info_hash,
860 struct hlist_head *new_laddrhash,
861 unsigned int new_size)
862{
863 struct hlist_head *old_info_hash, *old_laddrhash;
864 unsigned int old_size = fib_info_hash_size;
865 unsigned int i, bytes;
866
867 spin_lock_bh(&fib_info_lock);
868 old_info_hash = fib_info_hash;
869 old_laddrhash = fib_info_laddrhash;
870 fib_info_hash_size = new_size;
871
872 for (i = 0; i < old_size; i++) {
873 struct hlist_head *head = &fib_info_hash[i];
874 struct hlist_node *n;
875 struct fib_info *fi;
876
877 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
878 struct hlist_head *dest;
879 unsigned int new_hash;
880
881 new_hash = fib_info_hashfn(fi);
882 dest = &new_info_hash[new_hash];
883 hlist_add_head(&fi->fib_hash, dest);
884 }
885 }
886 fib_info_hash = new_info_hash;
887
888 for (i = 0; i < old_size; i++) {
889 struct hlist_head *lhead = &fib_info_laddrhash[i];
890 struct hlist_node *n;
891 struct fib_info *fi;
892
893 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
894 struct hlist_head *ldest;
895 unsigned int new_hash;
896
897 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
898 ldest = &new_laddrhash[new_hash];
899 hlist_add_head(&fi->fib_lhash, ldest);
900 }
901 }
902 fib_info_laddrhash = new_laddrhash;
903
904 spin_unlock_bh(&fib_info_lock);
905
906 bytes = old_size * sizeof(struct hlist_head *);
907 fib_info_hash_free(old_info_hash, bytes);
908 fib_info_hash_free(old_laddrhash, bytes);
909}
910
911__be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
912{
913 nh->nh_saddr = inet_select_addr(nh->nh_dev,
914 nh->nh_gw,
915 nh->nh_parent->fib_scope);
916 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
917
918 return nh->nh_saddr;
919}
920
921static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
922{
923 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
924 fib_prefsrc != cfg->fc_dst) {
925 u32 tb_id = cfg->fc_table;
926 int rc;
927
928 if (tb_id == RT_TABLE_MAIN)
929 tb_id = RT_TABLE_LOCAL;
930
931 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
932 fib_prefsrc, tb_id);
933
934 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
935 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
936 fib_prefsrc, RT_TABLE_LOCAL);
937 }
938
939 if (rc != RTN_LOCAL)
940 return false;
941 }
942 return true;
943}
944
945static int
946fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
947{
948 bool ecn_ca = false;
949 struct nlattr *nla;
950 int remaining;
951
952 if (!cfg->fc_mx)
953 return 0;
954
955 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
956 int type = nla_type(nla);
957 u32 val;
958
959 if (!type)
960 continue;
961 if (type > RTAX_MAX)
962 return -EINVAL;
963
964 if (type == RTAX_CC_ALGO) {
965 char tmp[TCP_CA_NAME_MAX];
966
967 nla_strlcpy(tmp, nla, sizeof(tmp));
968 val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
969 if (val == TCP_CA_UNSPEC)
970 return -EINVAL;
971 } else {
972 val = nla_get_u32(nla);
973 }
974 if (type == RTAX_ADVMSS && val > 65535 - 40)
975 val = 65535 - 40;
976 if (type == RTAX_MTU && val > 65535 - 15)
977 val = 65535 - 15;
978 if (type == RTAX_HOPLIMIT && val > 255)
979 val = 255;
980 if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
981 return -EINVAL;
982 fi->fib_metrics[type - 1] = val;
983 }
984
985 if (ecn_ca)
986 fi->fib_metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
987
988 return 0;
989}
990
991struct fib_info *fib_create_info(struct fib_config *cfg)
992{
993 int err;
994 struct fib_info *fi = NULL;
995 struct fib_info *ofi;
996 int nhs = 1;
997 struct net *net = cfg->fc_nlinfo.nl_net;
998
999 if (cfg->fc_type > RTN_MAX)
1000 goto err_inval;
1001
1002 /* Fast check to catch the most weird cases */
1003 if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
1004 goto err_inval;
1005
1006#ifdef CONFIG_IP_ROUTE_MULTIPATH
1007 if (cfg->fc_mp) {
1008 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
1009 if (nhs == 0)
1010 goto err_inval;
1011 }
1012#endif
1013
1014 err = -ENOBUFS;
1015 if (fib_info_cnt >= fib_info_hash_size) {
1016 unsigned int new_size = fib_info_hash_size << 1;
1017 struct hlist_head *new_info_hash;
1018 struct hlist_head *new_laddrhash;
1019 unsigned int bytes;
1020
1021 if (!new_size)
1022 new_size = 16;
1023 bytes = new_size * sizeof(struct hlist_head *);
1024 new_info_hash = fib_info_hash_alloc(bytes);
1025 new_laddrhash = fib_info_hash_alloc(bytes);
1026 if (!new_info_hash || !new_laddrhash) {
1027 fib_info_hash_free(new_info_hash, bytes);
1028 fib_info_hash_free(new_laddrhash, bytes);
1029 } else
1030 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1031
1032 if (!fib_info_hash_size)
1033 goto failure;
1034 }
1035
1036 fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1037 if (!fi)
1038 goto failure;
1039 fib_info_cnt++;
1040 if (cfg->fc_mx) {
1041 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
1042 if (!fi->fib_metrics)
1043 goto failure;
1044 } else
1045 fi->fib_metrics = (u32 *) dst_default_metrics;
1046
1047 fi->fib_net = net;
1048 fi->fib_protocol = cfg->fc_protocol;
1049 fi->fib_scope = cfg->fc_scope;
1050 fi->fib_flags = cfg->fc_flags;
1051 fi->fib_priority = cfg->fc_priority;
1052 fi->fib_prefsrc = cfg->fc_prefsrc;
1053 fi->fib_type = cfg->fc_type;
1054
1055 fi->fib_nhs = nhs;
1056 change_nexthops(fi) {
1057 nexthop_nh->nh_parent = fi;
1058 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1059 if (!nexthop_nh->nh_pcpu_rth_output)
1060 goto failure;
1061 } endfor_nexthops(fi)
1062
1063 err = fib_convert_metrics(fi, cfg);
1064 if (err)
1065 goto failure;
1066
1067 if (cfg->fc_mp) {
1068#ifdef CONFIG_IP_ROUTE_MULTIPATH
1069 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
1070 if (err != 0)
1071 goto failure;
1072 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
1073 goto err_inval;
1074 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
1075 goto err_inval;
1076#ifdef CONFIG_IP_ROUTE_CLASSID
1077 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
1078 goto err_inval;
1079#endif
1080#else
1081 goto err_inval;
1082#endif
1083 } else {
1084 struct fib_nh *nh = fi->fib_nh;
1085
1086 if (cfg->fc_encap) {
1087 struct lwtunnel_state *lwtstate;
1088 struct net_device *dev = NULL;
1089
1090 if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
1091 goto err_inval;
1092 if (cfg->fc_oif)
1093 dev = __dev_get_by_index(net, cfg->fc_oif);
1094 err = lwtunnel_build_state(dev, cfg->fc_encap_type,
1095 cfg->fc_encap, AF_INET, cfg,
1096 &lwtstate);
1097 if (err)
1098 goto failure;
1099
1100 nh->nh_lwtstate = lwtstate_get(lwtstate);
1101 }
1102 nh->nh_oif = cfg->fc_oif;
1103 nh->nh_gw = cfg->fc_gw;
1104 nh->nh_flags = cfg->fc_flags;
1105#ifdef CONFIG_IP_ROUTE_CLASSID
1106 nh->nh_tclassid = cfg->fc_flow;
1107 if (nh->nh_tclassid)
1108 fi->fib_net->ipv4.fib_num_tclassid_users++;
1109#endif
1110#ifdef CONFIG_IP_ROUTE_MULTIPATH
1111 nh->nh_weight = 1;
1112#endif
1113 }
1114
1115 if (fib_props[cfg->fc_type].error) {
1116 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
1117 goto err_inval;
1118 goto link_it;
1119 } else {
1120 switch (cfg->fc_type) {
1121 case RTN_UNICAST:
1122 case RTN_LOCAL:
1123 case RTN_BROADCAST:
1124 case RTN_ANYCAST:
1125 case RTN_MULTICAST:
1126 break;
1127 default:
1128 goto err_inval;
1129 }
1130 }
1131
1132 if (cfg->fc_scope > RT_SCOPE_HOST)
1133 goto err_inval;
1134
1135 if (cfg->fc_scope == RT_SCOPE_HOST) {
1136 struct fib_nh *nh = fi->fib_nh;
1137
1138 /* Local address is added. */
1139 if (nhs != 1 || nh->nh_gw)
1140 goto err_inval;
1141 nh->nh_scope = RT_SCOPE_NOWHERE;
1142 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1143 err = -ENODEV;
1144 if (!nh->nh_dev)
1145 goto failure;
1146 } else {
1147 int linkdown = 0;
1148
1149 change_nexthops(fi) {
1150 err = fib_check_nh(cfg, fi, nexthop_nh);
1151 if (err != 0)
1152 goto failure;
1153 if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1154 linkdown++;
1155 } endfor_nexthops(fi)
1156 if (linkdown == fi->fib_nhs)
1157 fi->fib_flags |= RTNH_F_LINKDOWN;
1158 }
1159
1160 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc))
1161 goto err_inval;
1162
1163 change_nexthops(fi) {
1164 fib_info_update_nh_saddr(net, nexthop_nh);
1165 fib_add_weight(fi, nexthop_nh);
1166 } endfor_nexthops(fi)
1167
1168 fib_rebalance(fi);
1169
1170link_it:
1171 ofi = fib_find_info(fi);
1172 if (ofi) {
1173 fi->fib_dead = 1;
1174 free_fib_info(fi);
1175 ofi->fib_treeref++;
1176 return ofi;
1177 }
1178
1179 fi->fib_treeref++;
1180 atomic_inc(&fi->fib_clntref);
1181 spin_lock_bh(&fib_info_lock);
1182 hlist_add_head(&fi->fib_hash,
1183 &fib_info_hash[fib_info_hashfn(fi)]);
1184 if (fi->fib_prefsrc) {
1185 struct hlist_head *head;
1186
1187 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1188 hlist_add_head(&fi->fib_lhash, head);
1189 }
1190 change_nexthops(fi) {
1191 struct hlist_head *head;
1192 unsigned int hash;
1193
1194 if (!nexthop_nh->nh_dev)
1195 continue;
1196 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1197 head = &fib_info_devhash[hash];
1198 hlist_add_head(&nexthop_nh->nh_hash, head);
1199 } endfor_nexthops(fi)
1200 spin_unlock_bh(&fib_info_lock);
1201 return fi;
1202
1203err_inval:
1204 err = -EINVAL;
1205
1206failure:
1207 if (fi) {
1208 fi->fib_dead = 1;
1209 free_fib_info(fi);
1210 }
1211
1212 return ERR_PTR(err);
1213}
1214
1215int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1216 u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1217 struct fib_info *fi, unsigned int flags)
1218{
1219 struct nlmsghdr *nlh;
1220 struct rtmsg *rtm;
1221
1222 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1223 if (!nlh)
1224 return -EMSGSIZE;
1225
1226 rtm = nlmsg_data(nlh);
1227 rtm->rtm_family = AF_INET;
1228 rtm->rtm_dst_len = dst_len;
1229 rtm->rtm_src_len = 0;
1230 rtm->rtm_tos = tos;
1231 if (tb_id < 256)
1232 rtm->rtm_table = tb_id;
1233 else
1234 rtm->rtm_table = RT_TABLE_COMPAT;
1235 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1236 goto nla_put_failure;
1237 rtm->rtm_type = type;
1238 rtm->rtm_flags = fi->fib_flags;
1239 rtm->rtm_scope = fi->fib_scope;
1240 rtm->rtm_protocol = fi->fib_protocol;
1241
1242 if (rtm->rtm_dst_len &&
1243 nla_put_in_addr(skb, RTA_DST, dst))
1244 goto nla_put_failure;
1245 if (fi->fib_priority &&
1246 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1247 goto nla_put_failure;
1248 if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1249 goto nla_put_failure;
1250
1251 if (fi->fib_prefsrc &&
1252 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1253 goto nla_put_failure;
1254 if (fi->fib_nhs == 1) {
1255 struct in_device *in_dev;
1256
1257 if (fi->fib_nh->nh_gw &&
1258 nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1259 goto nla_put_failure;
1260 if (fi->fib_nh->nh_oif &&
1261 nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1262 goto nla_put_failure;
1263 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1264 in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1265 if (in_dev &&
1266 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1267 rtm->rtm_flags |= RTNH_F_DEAD;
1268 }
1269#ifdef CONFIG_IP_ROUTE_CLASSID
1270 if (fi->fib_nh[0].nh_tclassid &&
1271 nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1272 goto nla_put_failure;
1273#endif
1274 if (fi->fib_nh->nh_lwtstate)
1275 lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate);
1276 }
1277#ifdef CONFIG_IP_ROUTE_MULTIPATH
1278 if (fi->fib_nhs > 1) {
1279 struct rtnexthop *rtnh;
1280 struct nlattr *mp;
1281
1282 mp = nla_nest_start(skb, RTA_MULTIPATH);
1283 if (!mp)
1284 goto nla_put_failure;
1285
1286 for_nexthops(fi) {
1287 struct in_device *in_dev;
1288
1289 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1290 if (!rtnh)
1291 goto nla_put_failure;
1292
1293 rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1294 if (nh->nh_flags & RTNH_F_LINKDOWN) {
1295 in_dev = __in_dev_get_rtnl(nh->nh_dev);
1296 if (in_dev &&
1297 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1298 rtnh->rtnh_flags |= RTNH_F_DEAD;
1299 }
1300 rtnh->rtnh_hops = nh->nh_weight - 1;
1301 rtnh->rtnh_ifindex = nh->nh_oif;
1302
1303 if (nh->nh_gw &&
1304 nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1305 goto nla_put_failure;
1306#ifdef CONFIG_IP_ROUTE_CLASSID
1307 if (nh->nh_tclassid &&
1308 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1309 goto nla_put_failure;
1310#endif
1311 if (nh->nh_lwtstate)
1312 lwtunnel_fill_encap(skb, nh->nh_lwtstate);
1313 /* length of rtnetlink header + attributes */
1314 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1315 } endfor_nexthops(fi);
1316
1317 nla_nest_end(skb, mp);
1318 }
1319#endif
1320 nlmsg_end(skb, nlh);
1321 return 0;
1322
1323nla_put_failure:
1324 nlmsg_cancel(skb, nlh);
1325 return -EMSGSIZE;
1326}
1327
1328/*
1329 * Update FIB if:
1330 * - local address disappeared -> we must delete all the entries
1331 * referring to it.
1332 * - device went down -> we must shutdown all nexthops going via it.
1333 */
1334int fib_sync_down_addr(struct net *net, __be32 local)
1335{
1336 int ret = 0;
1337 unsigned int hash = fib_laddr_hashfn(local);
1338 struct hlist_head *head = &fib_info_laddrhash[hash];
1339 struct fib_info *fi;
1340
1341 if (!fib_info_laddrhash || local == 0)
1342 return 0;
1343
1344 hlist_for_each_entry(fi, head, fib_lhash) {
1345 if (!net_eq(fi->fib_net, net))
1346 continue;
1347 if (fi->fib_prefsrc == local) {
1348 fi->fib_flags |= RTNH_F_DEAD;
1349 ret++;
1350 }
1351 }
1352 return ret;
1353}
1354
1355/* Event force Flags Description
1356 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host
1357 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host
1358 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed
1359 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed
1360 */
1361int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1362{
1363 int ret = 0;
1364 int scope = RT_SCOPE_NOWHERE;
1365 struct fib_info *prev_fi = NULL;
1366 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1367 struct hlist_head *head = &fib_info_devhash[hash];
1368 struct fib_nh *nh;
1369
1370 if (force)
1371 scope = -1;
1372
1373 hlist_for_each_entry(nh, head, nh_hash) {
1374 struct fib_info *fi = nh->nh_parent;
1375 int dead;
1376
1377 BUG_ON(!fi->fib_nhs);
1378 if (nh->nh_dev != dev || fi == prev_fi)
1379 continue;
1380 prev_fi = fi;
1381 dead = 0;
1382 change_nexthops(fi) {
1383 if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1384 dead++;
1385 else if (nexthop_nh->nh_dev == dev &&
1386 nexthop_nh->nh_scope != scope) {
1387 switch (event) {
1388 case NETDEV_DOWN:
1389 case NETDEV_UNREGISTER:
1390 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1391 /* fall through */
1392 case NETDEV_CHANGE:
1393 nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1394 break;
1395 }
1396 dead++;
1397 }
1398#ifdef CONFIG_IP_ROUTE_MULTIPATH
1399 if (event == NETDEV_UNREGISTER &&
1400 nexthop_nh->nh_dev == dev) {
1401 dead = fi->fib_nhs;
1402 break;
1403 }
1404#endif
1405 } endfor_nexthops(fi)
1406 if (dead == fi->fib_nhs) {
1407 switch (event) {
1408 case NETDEV_DOWN:
1409 case NETDEV_UNREGISTER:
1410 fi->fib_flags |= RTNH_F_DEAD;
1411 /* fall through */
1412 case NETDEV_CHANGE:
1413 fi->fib_flags |= RTNH_F_LINKDOWN;
1414 break;
1415 }
1416 ret++;
1417 }
1418
1419 fib_rebalance(fi);
1420 }
1421
1422 return ret;
1423}
1424
1425/* Must be invoked inside of an RCU protected region. */
1426void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1427{
1428 struct fib_info *fi = NULL, *last_resort = NULL;
1429 struct hlist_head *fa_head = res->fa_head;
1430 struct fib_table *tb = res->table;
1431 u8 slen = 32 - res->prefixlen;
1432 int order = -1, last_idx = -1;
1433 struct fib_alias *fa, *fa1 = NULL;
1434 u32 last_prio = res->fi->fib_priority;
1435 u8 last_tos = 0;
1436
1437 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1438 struct fib_info *next_fi = fa->fa_info;
1439
1440 if (fa->fa_slen != slen)
1441 continue;
1442 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1443 continue;
1444 if (fa->tb_id != tb->tb_id)
1445 continue;
1446 if (next_fi->fib_priority > last_prio &&
1447 fa->fa_tos == last_tos) {
1448 if (last_tos)
1449 continue;
1450 break;
1451 }
1452 if (next_fi->fib_flags & RTNH_F_DEAD)
1453 continue;
1454 last_tos = fa->fa_tos;
1455 last_prio = next_fi->fib_priority;
1456
1457 if (next_fi->fib_scope != res->scope ||
1458 fa->fa_type != RTN_UNICAST)
1459 continue;
1460 if (!next_fi->fib_nh[0].nh_gw ||
1461 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1462 continue;
1463
1464 fib_alias_accessed(fa);
1465
1466 if (!fi) {
1467 if (next_fi != res->fi)
1468 break;
1469 fa1 = fa;
1470 } else if (!fib_detect_death(fi, order, &last_resort,
1471 &last_idx, fa1->fa_default)) {
1472 fib_result_assign(res, fi);
1473 fa1->fa_default = order;
1474 goto out;
1475 }
1476 fi = next_fi;
1477 order++;
1478 }
1479
1480 if (order <= 0 || !fi) {
1481 if (fa1)
1482 fa1->fa_default = -1;
1483 goto out;
1484 }
1485
1486 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1487 fa1->fa_default)) {
1488 fib_result_assign(res, fi);
1489 fa1->fa_default = order;
1490 goto out;
1491 }
1492
1493 if (last_idx >= 0)
1494 fib_result_assign(res, last_resort);
1495 fa1->fa_default = last_idx;
1496out:
1497 return;
1498}
1499
1500/*
1501 * Dead device goes up. We wake up dead nexthops.
1502 * It takes sense only on multipath routes.
1503 */
1504int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1505{
1506 struct fib_info *prev_fi;
1507 unsigned int hash;
1508 struct hlist_head *head;
1509 struct fib_nh *nh;
1510 int ret;
1511
1512 if (!(dev->flags & IFF_UP))
1513 return 0;
1514
1515 if (nh_flags & RTNH_F_DEAD) {
1516 unsigned int flags = dev_get_flags(dev);
1517
1518 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1519 nh_flags |= RTNH_F_LINKDOWN;
1520 }
1521
1522 prev_fi = NULL;
1523 hash = fib_devindex_hashfn(dev->ifindex);
1524 head = &fib_info_devhash[hash];
1525 ret = 0;
1526
1527 hlist_for_each_entry(nh, head, nh_hash) {
1528 struct fib_info *fi = nh->nh_parent;
1529 int alive;
1530
1531 BUG_ON(!fi->fib_nhs);
1532 if (nh->nh_dev != dev || fi == prev_fi)
1533 continue;
1534
1535 prev_fi = fi;
1536 alive = 0;
1537 change_nexthops(fi) {
1538 if (!(nexthop_nh->nh_flags & nh_flags)) {
1539 alive++;
1540 continue;
1541 }
1542 if (!nexthop_nh->nh_dev ||
1543 !(nexthop_nh->nh_dev->flags & IFF_UP))
1544 continue;
1545 if (nexthop_nh->nh_dev != dev ||
1546 !__in_dev_get_rtnl(dev))
1547 continue;
1548 alive++;
1549 nexthop_nh->nh_flags &= ~nh_flags;
1550 } endfor_nexthops(fi)
1551
1552 if (alive > 0) {
1553 fi->fib_flags &= ~nh_flags;
1554 ret++;
1555 }
1556
1557 fib_rebalance(fi);
1558 }
1559
1560 return ret;
1561}
1562
1563#ifdef CONFIG_IP_ROUTE_MULTIPATH
1564
1565void fib_select_multipath(struct fib_result *res, int hash)
1566{
1567 struct fib_info *fi = res->fi;
1568
1569 for_nexthops(fi) {
1570 if (hash > atomic_read(&nh->nh_upper_bound))
1571 continue;
1572
1573 res->nh_sel = nhsel;
1574 return;
1575 } endfor_nexthops(fi);
1576
1577 /* Race condition: route has just become dead. */
1578 res->nh_sel = 0;
1579}
1580#endif
1581
1582void fib_select_path(struct net *net, struct fib_result *res,
1583 struct flowi4 *fl4, int mp_hash)
1584{
1585#ifdef CONFIG_IP_ROUTE_MULTIPATH
1586 if (res->fi->fib_nhs > 1 && fl4->flowi4_oif == 0) {
1587 if (mp_hash < 0)
1588 mp_hash = get_hash_from_flowi4(fl4) >> 1;
1589
1590 fib_select_multipath(res, mp_hash);
1591 }
1592 else
1593#endif
1594 if (!res->prefixlen &&
1595 res->table->tb_num_default > 1 &&
1596 res->type == RTN_UNICAST && !fl4->flowi4_oif)
1597 fib_select_default(fl4, res);
1598
1599 if (!fl4->saddr)
1600 fl4->saddr = FIB_RES_PREFSRC(net, *res);
1601}
1602EXPORT_SYMBOL_GPL(fib_select_path);
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
33#include <net/arp.h>
34#include <net/ip.h>
35#include <net/protocol.h>
36#include <net/route.h>
37#include <net/tcp.h>
38#include <net/sock.h>
39#include <net/ip_fib.h>
40#include <net/ip6_fib.h>
41#include <net/nexthop.h>
42#include <net/netlink.h>
43#include <net/rtnh.h>
44#include <net/lwtunnel.h>
45#include <net/fib_notifier.h>
46#include <net/addrconf.h>
47
48#include "fib_lookup.h"
49
50static DEFINE_SPINLOCK(fib_info_lock);
51static struct hlist_head *fib_info_hash;
52static struct hlist_head *fib_info_laddrhash;
53static unsigned int fib_info_hash_size;
54static unsigned int fib_info_cnt;
55
56#define DEVINDEX_HASHBITS 8
57#define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
58static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
59
60/* for_nexthops and change_nexthops only used when nexthop object
61 * is not set in a fib_info. The logic within can reference fib_nh.
62 */
63#ifdef CONFIG_IP_ROUTE_MULTIPATH
64
65#define for_nexthops(fi) { \
66 int nhsel; const struct fib_nh *nh; \
67 for (nhsel = 0, nh = (fi)->fib_nh; \
68 nhsel < fib_info_num_path((fi)); \
69 nh++, nhsel++)
70
71#define change_nexthops(fi) { \
72 int nhsel; struct fib_nh *nexthop_nh; \
73 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
74 nhsel < fib_info_num_path((fi)); \
75 nexthop_nh++, nhsel++)
76
77#else /* CONFIG_IP_ROUTE_MULTIPATH */
78
79/* Hope, that gcc will optimize it to get rid of dummy loop */
80
81#define for_nexthops(fi) { \
82 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \
83 for (nhsel = 0; nhsel < 1; nhsel++)
84
85#define change_nexthops(fi) { \
86 int nhsel; \
87 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
88 for (nhsel = 0; nhsel < 1; nhsel++)
89
90#endif /* CONFIG_IP_ROUTE_MULTIPATH */
91
92#define endfor_nexthops(fi) }
93
94
95const struct fib_prop fib_props[RTN_MAX + 1] = {
96 [RTN_UNSPEC] = {
97 .error = 0,
98 .scope = RT_SCOPE_NOWHERE,
99 },
100 [RTN_UNICAST] = {
101 .error = 0,
102 .scope = RT_SCOPE_UNIVERSE,
103 },
104 [RTN_LOCAL] = {
105 .error = 0,
106 .scope = RT_SCOPE_HOST,
107 },
108 [RTN_BROADCAST] = {
109 .error = 0,
110 .scope = RT_SCOPE_LINK,
111 },
112 [RTN_ANYCAST] = {
113 .error = 0,
114 .scope = RT_SCOPE_LINK,
115 },
116 [RTN_MULTICAST] = {
117 .error = 0,
118 .scope = RT_SCOPE_UNIVERSE,
119 },
120 [RTN_BLACKHOLE] = {
121 .error = -EINVAL,
122 .scope = RT_SCOPE_UNIVERSE,
123 },
124 [RTN_UNREACHABLE] = {
125 .error = -EHOSTUNREACH,
126 .scope = RT_SCOPE_UNIVERSE,
127 },
128 [RTN_PROHIBIT] = {
129 .error = -EACCES,
130 .scope = RT_SCOPE_UNIVERSE,
131 },
132 [RTN_THROW] = {
133 .error = -EAGAIN,
134 .scope = RT_SCOPE_UNIVERSE,
135 },
136 [RTN_NAT] = {
137 .error = -EINVAL,
138 .scope = RT_SCOPE_NOWHERE,
139 },
140 [RTN_XRESOLVE] = {
141 .error = -EINVAL,
142 .scope = RT_SCOPE_NOWHERE,
143 },
144};
145
146static void rt_fibinfo_free(struct rtable __rcu **rtp)
147{
148 struct rtable *rt = rcu_dereference_protected(*rtp, 1);
149
150 if (!rt)
151 return;
152
153 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
154 * because we waited an RCU grace period before calling
155 * free_fib_info_rcu()
156 */
157
158 dst_dev_put(&rt->dst);
159 dst_release_immediate(&rt->dst);
160}
161
162static void free_nh_exceptions(struct fib_nh_common *nhc)
163{
164 struct fnhe_hash_bucket *hash;
165 int i;
166
167 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
168 if (!hash)
169 return;
170 for (i = 0; i < FNHE_HASH_SIZE; i++) {
171 struct fib_nh_exception *fnhe;
172
173 fnhe = rcu_dereference_protected(hash[i].chain, 1);
174 while (fnhe) {
175 struct fib_nh_exception *next;
176
177 next = rcu_dereference_protected(fnhe->fnhe_next, 1);
178
179 rt_fibinfo_free(&fnhe->fnhe_rth_input);
180 rt_fibinfo_free(&fnhe->fnhe_rth_output);
181
182 kfree(fnhe);
183
184 fnhe = next;
185 }
186 }
187 kfree(hash);
188}
189
190static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
191{
192 int cpu;
193
194 if (!rtp)
195 return;
196
197 for_each_possible_cpu(cpu) {
198 struct rtable *rt;
199
200 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
201 if (rt) {
202 dst_dev_put(&rt->dst);
203 dst_release_immediate(&rt->dst);
204 }
205 }
206 free_percpu(rtp);
207}
208
209void fib_nh_common_release(struct fib_nh_common *nhc)
210{
211 if (nhc->nhc_dev)
212 dev_put(nhc->nhc_dev);
213
214 lwtstate_put(nhc->nhc_lwtstate);
215 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
216 rt_fibinfo_free(&nhc->nhc_rth_input);
217 free_nh_exceptions(nhc);
218}
219EXPORT_SYMBOL_GPL(fib_nh_common_release);
220
221void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
222{
223#ifdef CONFIG_IP_ROUTE_CLASSID
224 if (fib_nh->nh_tclassid)
225 net->ipv4.fib_num_tclassid_users--;
226#endif
227 fib_nh_common_release(&fib_nh->nh_common);
228}
229
230/* Release a nexthop info record */
231static void free_fib_info_rcu(struct rcu_head *head)
232{
233 struct fib_info *fi = container_of(head, struct fib_info, rcu);
234
235 if (fi->nh) {
236 nexthop_put(fi->nh);
237 } else {
238 change_nexthops(fi) {
239 fib_nh_release(fi->fib_net, nexthop_nh);
240 } endfor_nexthops(fi);
241 }
242
243 ip_fib_metrics_put(fi->fib_metrics);
244
245 kfree(fi);
246}
247
248void free_fib_info(struct fib_info *fi)
249{
250 if (fi->fib_dead == 0) {
251 pr_warn("Freeing alive fib_info %p\n", fi);
252 return;
253 }
254 fib_info_cnt--;
255
256 call_rcu(&fi->rcu, free_fib_info_rcu);
257}
258EXPORT_SYMBOL_GPL(free_fib_info);
259
260void fib_release_info(struct fib_info *fi)
261{
262 spin_lock_bh(&fib_info_lock);
263 if (fi && --fi->fib_treeref == 0) {
264 hlist_del(&fi->fib_hash);
265 if (fi->fib_prefsrc)
266 hlist_del(&fi->fib_lhash);
267 if (fi->nh) {
268 list_del(&fi->nh_list);
269 } else {
270 change_nexthops(fi) {
271 if (!nexthop_nh->fib_nh_dev)
272 continue;
273 hlist_del(&nexthop_nh->nh_hash);
274 } endfor_nexthops(fi)
275 }
276 fi->fib_dead = 1;
277 fib_info_put(fi);
278 }
279 spin_unlock_bh(&fib_info_lock);
280}
281
282static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
283{
284 const struct fib_nh *onh;
285
286 if (fi->nh || ofi->nh)
287 return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
288
289 if (ofi->fib_nhs == 0)
290 return 0;
291
292 for_nexthops(fi) {
293 onh = fib_info_nh(ofi, nhsel);
294
295 if (nh->fib_nh_oif != onh->fib_nh_oif ||
296 nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
297 nh->fib_nh_scope != onh->fib_nh_scope ||
298#ifdef CONFIG_IP_ROUTE_MULTIPATH
299 nh->fib_nh_weight != onh->fib_nh_weight ||
300#endif
301#ifdef CONFIG_IP_ROUTE_CLASSID
302 nh->nh_tclassid != onh->nh_tclassid ||
303#endif
304 lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
305 ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
306 return -1;
307
308 if (nh->fib_nh_gw_family == AF_INET &&
309 nh->fib_nh_gw4 != onh->fib_nh_gw4)
310 return -1;
311
312 if (nh->fib_nh_gw_family == AF_INET6 &&
313 ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
314 return -1;
315 } endfor_nexthops(fi);
316 return 0;
317}
318
319static inline unsigned int fib_devindex_hashfn(unsigned int val)
320{
321 unsigned int mask = DEVINDEX_HASHSIZE - 1;
322
323 return (val ^
324 (val >> DEVINDEX_HASHBITS) ^
325 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
326}
327
328static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
329 u32 prefsrc, u32 priority)
330{
331 unsigned int val = init_val;
332
333 val ^= (protocol << 8) | scope;
334 val ^= prefsrc;
335 val ^= priority;
336
337 return val;
338}
339
340static unsigned int fib_info_hashfn_result(unsigned int val)
341{
342 unsigned int mask = (fib_info_hash_size - 1);
343
344 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
345}
346
347static inline unsigned int fib_info_hashfn(struct fib_info *fi)
348{
349 unsigned int val;
350
351 val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
352 fi->fib_scope, (__force u32)fi->fib_prefsrc,
353 fi->fib_priority);
354
355 if (fi->nh) {
356 val ^= fib_devindex_hashfn(fi->nh->id);
357 } else {
358 for_nexthops(fi) {
359 val ^= fib_devindex_hashfn(nh->fib_nh_oif);
360 } endfor_nexthops(fi)
361 }
362
363 return fib_info_hashfn_result(val);
364}
365
366/* no metrics, only nexthop id */
367static struct fib_info *fib_find_info_nh(struct net *net,
368 const struct fib_config *cfg)
369{
370 struct hlist_head *head;
371 struct fib_info *fi;
372 unsigned int hash;
373
374 hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
375 cfg->fc_protocol, cfg->fc_scope,
376 (__force u32)cfg->fc_prefsrc,
377 cfg->fc_priority);
378 hash = fib_info_hashfn_result(hash);
379 head = &fib_info_hash[hash];
380
381 hlist_for_each_entry(fi, head, fib_hash) {
382 if (!net_eq(fi->fib_net, net))
383 continue;
384 if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
385 continue;
386 if (cfg->fc_protocol == fi->fib_protocol &&
387 cfg->fc_scope == fi->fib_scope &&
388 cfg->fc_prefsrc == fi->fib_prefsrc &&
389 cfg->fc_priority == fi->fib_priority &&
390 cfg->fc_type == fi->fib_type &&
391 cfg->fc_table == fi->fib_tb_id &&
392 !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
393 return fi;
394 }
395
396 return NULL;
397}
398
399static struct fib_info *fib_find_info(struct fib_info *nfi)
400{
401 struct hlist_head *head;
402 struct fib_info *fi;
403 unsigned int hash;
404
405 hash = fib_info_hashfn(nfi);
406 head = &fib_info_hash[hash];
407
408 hlist_for_each_entry(fi, head, fib_hash) {
409 if (!net_eq(fi->fib_net, nfi->fib_net))
410 continue;
411 if (fi->fib_nhs != nfi->fib_nhs)
412 continue;
413 if (nfi->fib_protocol == fi->fib_protocol &&
414 nfi->fib_scope == fi->fib_scope &&
415 nfi->fib_prefsrc == fi->fib_prefsrc &&
416 nfi->fib_priority == fi->fib_priority &&
417 nfi->fib_type == fi->fib_type &&
418 memcmp(nfi->fib_metrics, fi->fib_metrics,
419 sizeof(u32) * RTAX_MAX) == 0 &&
420 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
421 nh_comp(fi, nfi) == 0)
422 return fi;
423 }
424
425 return NULL;
426}
427
428/* Check, that the gateway is already configured.
429 * Used only by redirect accept routine.
430 */
431int ip_fib_check_default(__be32 gw, struct net_device *dev)
432{
433 struct hlist_head *head;
434 struct fib_nh *nh;
435 unsigned int hash;
436
437 spin_lock(&fib_info_lock);
438
439 hash = fib_devindex_hashfn(dev->ifindex);
440 head = &fib_info_devhash[hash];
441 hlist_for_each_entry(nh, head, nh_hash) {
442 if (nh->fib_nh_dev == dev &&
443 nh->fib_nh_gw4 == gw &&
444 !(nh->fib_nh_flags & RTNH_F_DEAD)) {
445 spin_unlock(&fib_info_lock);
446 return 0;
447 }
448 }
449
450 spin_unlock(&fib_info_lock);
451
452 return -1;
453}
454
455static inline size_t fib_nlmsg_size(struct fib_info *fi)
456{
457 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
458 + nla_total_size(4) /* RTA_TABLE */
459 + nla_total_size(4) /* RTA_DST */
460 + nla_total_size(4) /* RTA_PRIORITY */
461 + nla_total_size(4) /* RTA_PREFSRC */
462 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
463 unsigned int nhs = fib_info_num_path(fi);
464
465 /* space for nested metrics */
466 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
467
468 if (fi->nh)
469 payload += nla_total_size(4); /* RTA_NH_ID */
470
471 if (nhs) {
472 size_t nh_encapsize = 0;
473 /* Also handles the special case nhs == 1 */
474
475 /* each nexthop is packed in an attribute */
476 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
477 unsigned int i;
478
479 /* may contain flow and gateway attribute */
480 nhsize += 2 * nla_total_size(4);
481
482 /* grab encap info */
483 for (i = 0; i < fib_info_num_path(fi); i++) {
484 struct fib_nh_common *nhc = fib_info_nhc(fi, i);
485
486 if (nhc->nhc_lwtstate) {
487 /* RTA_ENCAP_TYPE */
488 nh_encapsize += lwtunnel_get_encap_size(
489 nhc->nhc_lwtstate);
490 /* RTA_ENCAP */
491 nh_encapsize += nla_total_size(2);
492 }
493 }
494
495 /* all nexthops are packed in a nested attribute */
496 payload += nla_total_size((nhs * nhsize) + nh_encapsize);
497
498 }
499
500 return payload;
501}
502
503void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
504 int dst_len, u32 tb_id, const struct nl_info *info,
505 unsigned int nlm_flags)
506{
507 struct fib_rt_info fri;
508 struct sk_buff *skb;
509 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
510 int err = -ENOBUFS;
511
512 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
513 if (!skb)
514 goto errout;
515
516 fri.fi = fa->fa_info;
517 fri.tb_id = tb_id;
518 fri.dst = key;
519 fri.dst_len = dst_len;
520 fri.tos = fa->fa_tos;
521 fri.type = fa->fa_type;
522 fri.offload = fa->offload;
523 fri.trap = fa->trap;
524 err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
525 if (err < 0) {
526 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
527 WARN_ON(err == -EMSGSIZE);
528 kfree_skb(skb);
529 goto errout;
530 }
531 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
532 info->nlh, GFP_KERNEL);
533 return;
534errout:
535 if (err < 0)
536 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
537}
538
539static int fib_detect_death(struct fib_info *fi, int order,
540 struct fib_info **last_resort, int *last_idx,
541 int dflt)
542{
543 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
544 struct neighbour *n;
545 int state = NUD_NONE;
546
547 if (likely(nhc->nhc_gw_family == AF_INET))
548 n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
549 else if (nhc->nhc_gw_family == AF_INET6)
550 n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
551 nhc->nhc_dev);
552 else
553 n = NULL;
554
555 if (n) {
556 state = n->nud_state;
557 neigh_release(n);
558 } else {
559 return 0;
560 }
561 if (state == NUD_REACHABLE)
562 return 0;
563 if ((state & NUD_VALID) && order != dflt)
564 return 0;
565 if ((state & NUD_VALID) ||
566 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
567 *last_resort = fi;
568 *last_idx = order;
569 }
570 return 1;
571}
572
573int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
574 struct nlattr *encap, u16 encap_type,
575 void *cfg, gfp_t gfp_flags,
576 struct netlink_ext_ack *extack)
577{
578 int err;
579
580 nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
581 gfp_flags);
582 if (!nhc->nhc_pcpu_rth_output)
583 return -ENOMEM;
584
585 if (encap) {
586 struct lwtunnel_state *lwtstate;
587
588 if (encap_type == LWTUNNEL_ENCAP_NONE) {
589 NL_SET_ERR_MSG(extack, "LWT encap type not specified");
590 err = -EINVAL;
591 goto lwt_failure;
592 }
593 err = lwtunnel_build_state(net, encap_type, encap,
594 nhc->nhc_family, cfg, &lwtstate,
595 extack);
596 if (err)
597 goto lwt_failure;
598
599 nhc->nhc_lwtstate = lwtstate_get(lwtstate);
600 }
601
602 return 0;
603
604lwt_failure:
605 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
606 nhc->nhc_pcpu_rth_output = NULL;
607 return err;
608}
609EXPORT_SYMBOL_GPL(fib_nh_common_init);
610
611int fib_nh_init(struct net *net, struct fib_nh *nh,
612 struct fib_config *cfg, int nh_weight,
613 struct netlink_ext_ack *extack)
614{
615 int err;
616
617 nh->fib_nh_family = AF_INET;
618
619 err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
620 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
621 if (err)
622 return err;
623
624 nh->fib_nh_oif = cfg->fc_oif;
625 nh->fib_nh_gw_family = cfg->fc_gw_family;
626 if (cfg->fc_gw_family == AF_INET)
627 nh->fib_nh_gw4 = cfg->fc_gw4;
628 else if (cfg->fc_gw_family == AF_INET6)
629 nh->fib_nh_gw6 = cfg->fc_gw6;
630
631 nh->fib_nh_flags = cfg->fc_flags;
632
633#ifdef CONFIG_IP_ROUTE_CLASSID
634 nh->nh_tclassid = cfg->fc_flow;
635 if (nh->nh_tclassid)
636 net->ipv4.fib_num_tclassid_users++;
637#endif
638#ifdef CONFIG_IP_ROUTE_MULTIPATH
639 nh->fib_nh_weight = nh_weight;
640#endif
641 return 0;
642}
643
644#ifdef CONFIG_IP_ROUTE_MULTIPATH
645
646static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
647 struct netlink_ext_ack *extack)
648{
649 int nhs = 0;
650
651 while (rtnh_ok(rtnh, remaining)) {
652 nhs++;
653 rtnh = rtnh_next(rtnh, &remaining);
654 }
655
656 /* leftover implies invalid nexthop configuration, discard it */
657 if (remaining > 0) {
658 NL_SET_ERR_MSG(extack,
659 "Invalid nexthop configuration - extra data after nexthops");
660 nhs = 0;
661 }
662
663 return nhs;
664}
665
666/* only called when fib_nh is integrated into fib_info */
667static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
668 int remaining, struct fib_config *cfg,
669 struct netlink_ext_ack *extack)
670{
671 struct net *net = fi->fib_net;
672 struct fib_config fib_cfg;
673 struct fib_nh *nh;
674 int ret;
675
676 change_nexthops(fi) {
677 int attrlen;
678
679 memset(&fib_cfg, 0, sizeof(fib_cfg));
680
681 if (!rtnh_ok(rtnh, remaining)) {
682 NL_SET_ERR_MSG(extack,
683 "Invalid nexthop configuration - extra data after nexthop");
684 return -EINVAL;
685 }
686
687 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
688 NL_SET_ERR_MSG(extack,
689 "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
690 return -EINVAL;
691 }
692
693 fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
694 fib_cfg.fc_oif = rtnh->rtnh_ifindex;
695
696 attrlen = rtnh_attrlen(rtnh);
697 if (attrlen > 0) {
698 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
699
700 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
701 nlav = nla_find(attrs, attrlen, RTA_VIA);
702 if (nla && nlav) {
703 NL_SET_ERR_MSG(extack,
704 "Nexthop configuration can not contain both GATEWAY and VIA");
705 return -EINVAL;
706 }
707 if (nla) {
708 fib_cfg.fc_gw4 = nla_get_in_addr(nla);
709 if (fib_cfg.fc_gw4)
710 fib_cfg.fc_gw_family = AF_INET;
711 } else if (nlav) {
712 ret = fib_gw_from_via(&fib_cfg, nlav, extack);
713 if (ret)
714 goto errout;
715 }
716
717 nla = nla_find(attrs, attrlen, RTA_FLOW);
718 if (nla)
719 fib_cfg.fc_flow = nla_get_u32(nla);
720
721 fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
722 nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
723 if (nla)
724 fib_cfg.fc_encap_type = nla_get_u16(nla);
725 }
726
727 ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
728 rtnh->rtnh_hops + 1, extack);
729 if (ret)
730 goto errout;
731
732 rtnh = rtnh_next(rtnh, &remaining);
733 } endfor_nexthops(fi);
734
735 ret = -EINVAL;
736 nh = fib_info_nh(fi, 0);
737 if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
738 NL_SET_ERR_MSG(extack,
739 "Nexthop device index does not match RTA_OIF");
740 goto errout;
741 }
742 if (cfg->fc_gw_family) {
743 if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
744 (cfg->fc_gw_family == AF_INET &&
745 nh->fib_nh_gw4 != cfg->fc_gw4) ||
746 (cfg->fc_gw_family == AF_INET6 &&
747 ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
748 NL_SET_ERR_MSG(extack,
749 "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
750 goto errout;
751 }
752 }
753#ifdef CONFIG_IP_ROUTE_CLASSID
754 if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
755 NL_SET_ERR_MSG(extack,
756 "Nexthop class id does not match RTA_FLOW");
757 goto errout;
758 }
759#endif
760 ret = 0;
761errout:
762 return ret;
763}
764
765/* only called when fib_nh is integrated into fib_info */
766static void fib_rebalance(struct fib_info *fi)
767{
768 int total;
769 int w;
770
771 if (fib_info_num_path(fi) < 2)
772 return;
773
774 total = 0;
775 for_nexthops(fi) {
776 if (nh->fib_nh_flags & RTNH_F_DEAD)
777 continue;
778
779 if (ip_ignore_linkdown(nh->fib_nh_dev) &&
780 nh->fib_nh_flags & RTNH_F_LINKDOWN)
781 continue;
782
783 total += nh->fib_nh_weight;
784 } endfor_nexthops(fi);
785
786 w = 0;
787 change_nexthops(fi) {
788 int upper_bound;
789
790 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
791 upper_bound = -1;
792 } else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
793 nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
794 upper_bound = -1;
795 } else {
796 w += nexthop_nh->fib_nh_weight;
797 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
798 total) - 1;
799 }
800
801 atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
802 } endfor_nexthops(fi);
803}
804#else /* CONFIG_IP_ROUTE_MULTIPATH */
805
806static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
807 int remaining, struct fib_config *cfg,
808 struct netlink_ext_ack *extack)
809{
810 NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
811
812 return -EINVAL;
813}
814
815#define fib_rebalance(fi) do { } while (0)
816
817#endif /* CONFIG_IP_ROUTE_MULTIPATH */
818
819static int fib_encap_match(struct net *net, u16 encap_type,
820 struct nlattr *encap,
821 const struct fib_nh *nh,
822 const struct fib_config *cfg,
823 struct netlink_ext_ack *extack)
824{
825 struct lwtunnel_state *lwtstate;
826 int ret, result = 0;
827
828 if (encap_type == LWTUNNEL_ENCAP_NONE)
829 return 0;
830
831 ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
832 cfg, &lwtstate, extack);
833 if (!ret) {
834 result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
835 lwtstate_free(lwtstate);
836 }
837
838 return result;
839}
840
841int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
842 struct netlink_ext_ack *extack)
843{
844#ifdef CONFIG_IP_ROUTE_MULTIPATH
845 struct rtnexthop *rtnh;
846 int remaining;
847#endif
848
849 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
850 return 1;
851
852 if (cfg->fc_nh_id) {
853 if (fi->nh && cfg->fc_nh_id == fi->nh->id)
854 return 0;
855 return 1;
856 }
857
858 if (cfg->fc_oif || cfg->fc_gw_family) {
859 struct fib_nh *nh = fib_info_nh(fi, 0);
860
861 if (cfg->fc_encap) {
862 if (fib_encap_match(net, cfg->fc_encap_type,
863 cfg->fc_encap, nh, cfg, extack))
864 return 1;
865 }
866#ifdef CONFIG_IP_ROUTE_CLASSID
867 if (cfg->fc_flow &&
868 cfg->fc_flow != nh->nh_tclassid)
869 return 1;
870#endif
871 if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
872 (cfg->fc_gw_family &&
873 cfg->fc_gw_family != nh->fib_nh_gw_family))
874 return 1;
875
876 if (cfg->fc_gw_family == AF_INET &&
877 cfg->fc_gw4 != nh->fib_nh_gw4)
878 return 1;
879
880 if (cfg->fc_gw_family == AF_INET6 &&
881 ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
882 return 1;
883
884 return 0;
885 }
886
887#ifdef CONFIG_IP_ROUTE_MULTIPATH
888 if (!cfg->fc_mp)
889 return 0;
890
891 rtnh = cfg->fc_mp;
892 remaining = cfg->fc_mp_len;
893
894 for_nexthops(fi) {
895 int attrlen;
896
897 if (!rtnh_ok(rtnh, remaining))
898 return -EINVAL;
899
900 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
901 return 1;
902
903 attrlen = rtnh_attrlen(rtnh);
904 if (attrlen > 0) {
905 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
906
907 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
908 nlav = nla_find(attrs, attrlen, RTA_VIA);
909 if (nla && nlav) {
910 NL_SET_ERR_MSG(extack,
911 "Nexthop configuration can not contain both GATEWAY and VIA");
912 return -EINVAL;
913 }
914
915 if (nla) {
916 if (nh->fib_nh_gw_family != AF_INET ||
917 nla_get_in_addr(nla) != nh->fib_nh_gw4)
918 return 1;
919 } else if (nlav) {
920 struct fib_config cfg2;
921 int err;
922
923 err = fib_gw_from_via(&cfg2, nlav, extack);
924 if (err)
925 return err;
926
927 switch (nh->fib_nh_gw_family) {
928 case AF_INET:
929 if (cfg2.fc_gw_family != AF_INET ||
930 cfg2.fc_gw4 != nh->fib_nh_gw4)
931 return 1;
932 break;
933 case AF_INET6:
934 if (cfg2.fc_gw_family != AF_INET6 ||
935 ipv6_addr_cmp(&cfg2.fc_gw6,
936 &nh->fib_nh_gw6))
937 return 1;
938 break;
939 }
940 }
941
942#ifdef CONFIG_IP_ROUTE_CLASSID
943 nla = nla_find(attrs, attrlen, RTA_FLOW);
944 if (nla && nla_get_u32(nla) != nh->nh_tclassid)
945 return 1;
946#endif
947 }
948
949 rtnh = rtnh_next(rtnh, &remaining);
950 } endfor_nexthops(fi);
951#endif
952 return 0;
953}
954
955bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
956{
957 struct nlattr *nla;
958 int remaining;
959
960 if (!cfg->fc_mx)
961 return true;
962
963 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
964 int type = nla_type(nla);
965 u32 fi_val, val;
966
967 if (!type)
968 continue;
969 if (type > RTAX_MAX)
970 return false;
971
972 if (type == RTAX_CC_ALGO) {
973 char tmp[TCP_CA_NAME_MAX];
974 bool ecn_ca = false;
975
976 nla_strlcpy(tmp, nla, sizeof(tmp));
977 val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
978 } else {
979 if (nla_len(nla) != sizeof(u32))
980 return false;
981 val = nla_get_u32(nla);
982 }
983
984 fi_val = fi->fib_metrics->metrics[type - 1];
985 if (type == RTAX_FEATURES)
986 fi_val &= ~DST_FEATURE_ECN_CA;
987
988 if (fi_val != val)
989 return false;
990 }
991
992 return true;
993}
994
995static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
996 u32 table, struct netlink_ext_ack *extack)
997{
998 struct fib6_config cfg = {
999 .fc_table = table,
1000 .fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1001 .fc_ifindex = nh->fib_nh_oif,
1002 .fc_gateway = nh->fib_nh_gw6,
1003 };
1004 struct fib6_nh fib6_nh = {};
1005 int err;
1006
1007 err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1008 if (!err) {
1009 nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1010 dev_hold(nh->fib_nh_dev);
1011 nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1012 nh->fib_nh_scope = RT_SCOPE_LINK;
1013
1014 ipv6_stub->fib6_nh_release(&fib6_nh);
1015 }
1016
1017 return err;
1018}
1019
1020/*
1021 * Picture
1022 * -------
1023 *
1024 * Semantics of nexthop is very messy by historical reasons.
1025 * We have to take into account, that:
1026 * a) gateway can be actually local interface address,
1027 * so that gatewayed route is direct.
1028 * b) gateway must be on-link address, possibly
1029 * described not by an ifaddr, but also by a direct route.
1030 * c) If both gateway and interface are specified, they should not
1031 * contradict.
1032 * d) If we use tunnel routes, gateway could be not on-link.
1033 *
1034 * Attempt to reconcile all of these (alas, self-contradictory) conditions
1035 * results in pretty ugly and hairy code with obscure logic.
1036 *
1037 * I chose to generalized it instead, so that the size
1038 * of code does not increase practically, but it becomes
1039 * much more general.
1040 * Every prefix is assigned a "scope" value: "host" is local address,
1041 * "link" is direct route,
1042 * [ ... "site" ... "interior" ... ]
1043 * and "universe" is true gateway route with global meaning.
1044 *
1045 * Every prefix refers to a set of "nexthop"s (gw, oif),
1046 * where gw must have narrower scope. This recursion stops
1047 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1048 * which means that gw is forced to be on link.
1049 *
1050 * Code is still hairy, but now it is apparently logically
1051 * consistent and very flexible. F.e. as by-product it allows
1052 * to co-exists in peace independent exterior and interior
1053 * routing processes.
1054 *
1055 * Normally it looks as following.
1056 *
1057 * {universe prefix} -> (gw, oif) [scope link]
1058 * |
1059 * |-> {link prefix} -> (gw, oif) [scope local]
1060 * |
1061 * |-> {local prefix} (terminal node)
1062 */
1063static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1064 u8 scope, struct netlink_ext_ack *extack)
1065{
1066 struct net_device *dev;
1067 struct fib_result res;
1068 int err = 0;
1069
1070 if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1071 unsigned int addr_type;
1072
1073 if (scope >= RT_SCOPE_LINK) {
1074 NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1075 return -EINVAL;
1076 }
1077 dev = __dev_get_by_index(net, nh->fib_nh_oif);
1078 if (!dev) {
1079 NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1080 return -ENODEV;
1081 }
1082 if (!(dev->flags & IFF_UP)) {
1083 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1084 return -ENETDOWN;
1085 }
1086 addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1087 if (addr_type != RTN_UNICAST) {
1088 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1089 return -EINVAL;
1090 }
1091 if (!netif_carrier_ok(dev))
1092 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1093 nh->fib_nh_dev = dev;
1094 dev_hold(dev);
1095 nh->fib_nh_scope = RT_SCOPE_LINK;
1096 return 0;
1097 }
1098 rcu_read_lock();
1099 {
1100 struct fib_table *tbl = NULL;
1101 struct flowi4 fl4 = {
1102 .daddr = nh->fib_nh_gw4,
1103 .flowi4_scope = scope + 1,
1104 .flowi4_oif = nh->fib_nh_oif,
1105 .flowi4_iif = LOOPBACK_IFINDEX,
1106 };
1107
1108 /* It is not necessary, but requires a bit of thinking */
1109 if (fl4.flowi4_scope < RT_SCOPE_LINK)
1110 fl4.flowi4_scope = RT_SCOPE_LINK;
1111
1112 if (table && table != RT_TABLE_MAIN)
1113 tbl = fib_get_table(net, table);
1114
1115 if (tbl)
1116 err = fib_table_lookup(tbl, &fl4, &res,
1117 FIB_LOOKUP_IGNORE_LINKSTATE |
1118 FIB_LOOKUP_NOREF);
1119
1120 /* on error or if no table given do full lookup. This
1121 * is needed for example when nexthops are in the local
1122 * table rather than the given table
1123 */
1124 if (!tbl || err) {
1125 err = fib_lookup(net, &fl4, &res,
1126 FIB_LOOKUP_IGNORE_LINKSTATE);
1127 }
1128
1129 if (err) {
1130 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1131 goto out;
1132 }
1133 }
1134
1135 err = -EINVAL;
1136 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1137 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1138 goto out;
1139 }
1140 nh->fib_nh_scope = res.scope;
1141 nh->fib_nh_oif = FIB_RES_OIF(res);
1142 nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1143 if (!dev) {
1144 NL_SET_ERR_MSG(extack,
1145 "No egress device for nexthop gateway");
1146 goto out;
1147 }
1148 dev_hold(dev);
1149 if (!netif_carrier_ok(dev))
1150 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1151 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1152out:
1153 rcu_read_unlock();
1154 return err;
1155}
1156
1157static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1158 struct netlink_ext_ack *extack)
1159{
1160 struct in_device *in_dev;
1161 int err;
1162
1163 if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1164 NL_SET_ERR_MSG(extack,
1165 "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1166 return -EINVAL;
1167 }
1168
1169 rcu_read_lock();
1170
1171 err = -ENODEV;
1172 in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1173 if (!in_dev)
1174 goto out;
1175 err = -ENETDOWN;
1176 if (!(in_dev->dev->flags & IFF_UP)) {
1177 NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1178 goto out;
1179 }
1180
1181 nh->fib_nh_dev = in_dev->dev;
1182 dev_hold(nh->fib_nh_dev);
1183 nh->fib_nh_scope = RT_SCOPE_HOST;
1184 if (!netif_carrier_ok(nh->fib_nh_dev))
1185 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1186 err = 0;
1187out:
1188 rcu_read_unlock();
1189 return err;
1190}
1191
1192int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1193 struct netlink_ext_ack *extack)
1194{
1195 int err;
1196
1197 if (nh->fib_nh_gw_family == AF_INET)
1198 err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1199 else if (nh->fib_nh_gw_family == AF_INET6)
1200 err = fib_check_nh_v6_gw(net, nh, table, extack);
1201 else
1202 err = fib_check_nh_nongw(net, nh, extack);
1203
1204 return err;
1205}
1206
1207static inline unsigned int fib_laddr_hashfn(__be32 val)
1208{
1209 unsigned int mask = (fib_info_hash_size - 1);
1210
1211 return ((__force u32)val ^
1212 ((__force u32)val >> 7) ^
1213 ((__force u32)val >> 14)) & mask;
1214}
1215
1216static struct hlist_head *fib_info_hash_alloc(int bytes)
1217{
1218 if (bytes <= PAGE_SIZE)
1219 return kzalloc(bytes, GFP_KERNEL);
1220 else
1221 return (struct hlist_head *)
1222 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
1223 get_order(bytes));
1224}
1225
1226static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1227{
1228 if (!hash)
1229 return;
1230
1231 if (bytes <= PAGE_SIZE)
1232 kfree(hash);
1233 else
1234 free_pages((unsigned long) hash, get_order(bytes));
1235}
1236
1237static void fib_info_hash_move(struct hlist_head *new_info_hash,
1238 struct hlist_head *new_laddrhash,
1239 unsigned int new_size)
1240{
1241 struct hlist_head *old_info_hash, *old_laddrhash;
1242 unsigned int old_size = fib_info_hash_size;
1243 unsigned int i, bytes;
1244
1245 spin_lock_bh(&fib_info_lock);
1246 old_info_hash = fib_info_hash;
1247 old_laddrhash = fib_info_laddrhash;
1248 fib_info_hash_size = new_size;
1249
1250 for (i = 0; i < old_size; i++) {
1251 struct hlist_head *head = &fib_info_hash[i];
1252 struct hlist_node *n;
1253 struct fib_info *fi;
1254
1255 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1256 struct hlist_head *dest;
1257 unsigned int new_hash;
1258
1259 new_hash = fib_info_hashfn(fi);
1260 dest = &new_info_hash[new_hash];
1261 hlist_add_head(&fi->fib_hash, dest);
1262 }
1263 }
1264 fib_info_hash = new_info_hash;
1265
1266 for (i = 0; i < old_size; i++) {
1267 struct hlist_head *lhead = &fib_info_laddrhash[i];
1268 struct hlist_node *n;
1269 struct fib_info *fi;
1270
1271 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1272 struct hlist_head *ldest;
1273 unsigned int new_hash;
1274
1275 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1276 ldest = &new_laddrhash[new_hash];
1277 hlist_add_head(&fi->fib_lhash, ldest);
1278 }
1279 }
1280 fib_info_laddrhash = new_laddrhash;
1281
1282 spin_unlock_bh(&fib_info_lock);
1283
1284 bytes = old_size * sizeof(struct hlist_head *);
1285 fib_info_hash_free(old_info_hash, bytes);
1286 fib_info_hash_free(old_laddrhash, bytes);
1287}
1288
1289__be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1290 unsigned char scope)
1291{
1292 struct fib_nh *nh;
1293
1294 if (nhc->nhc_family != AF_INET)
1295 return inet_select_addr(nhc->nhc_dev, 0, scope);
1296
1297 nh = container_of(nhc, struct fib_nh, nh_common);
1298 nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1299 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1300
1301 return nh->nh_saddr;
1302}
1303
1304__be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1305{
1306 struct fib_nh_common *nhc = res->nhc;
1307
1308 if (res->fi->fib_prefsrc)
1309 return res->fi->fib_prefsrc;
1310
1311 if (nhc->nhc_family == AF_INET) {
1312 struct fib_nh *nh;
1313
1314 nh = container_of(nhc, struct fib_nh, nh_common);
1315 if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1316 return nh->nh_saddr;
1317 }
1318
1319 return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1320}
1321
1322static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1323{
1324 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1325 fib_prefsrc != cfg->fc_dst) {
1326 u32 tb_id = cfg->fc_table;
1327 int rc;
1328
1329 if (tb_id == RT_TABLE_MAIN)
1330 tb_id = RT_TABLE_LOCAL;
1331
1332 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1333 fib_prefsrc, tb_id);
1334
1335 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1336 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1337 fib_prefsrc, RT_TABLE_LOCAL);
1338 }
1339
1340 if (rc != RTN_LOCAL)
1341 return false;
1342 }
1343 return true;
1344}
1345
1346struct fib_info *fib_create_info(struct fib_config *cfg,
1347 struct netlink_ext_ack *extack)
1348{
1349 int err;
1350 struct fib_info *fi = NULL;
1351 struct nexthop *nh = NULL;
1352 struct fib_info *ofi;
1353 int nhs = 1;
1354 struct net *net = cfg->fc_nlinfo.nl_net;
1355
1356 if (cfg->fc_type > RTN_MAX)
1357 goto err_inval;
1358
1359 /* Fast check to catch the most weird cases */
1360 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1361 NL_SET_ERR_MSG(extack, "Invalid scope");
1362 goto err_inval;
1363 }
1364
1365 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1366 NL_SET_ERR_MSG(extack,
1367 "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1368 goto err_inval;
1369 }
1370
1371 if (cfg->fc_nh_id) {
1372 if (!cfg->fc_mx) {
1373 fi = fib_find_info_nh(net, cfg);
1374 if (fi) {
1375 fi->fib_treeref++;
1376 return fi;
1377 }
1378 }
1379
1380 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1381 if (!nh) {
1382 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1383 goto err_inval;
1384 }
1385 nhs = 0;
1386 }
1387
1388#ifdef CONFIG_IP_ROUTE_MULTIPATH
1389 if (cfg->fc_mp) {
1390 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1391 if (nhs == 0)
1392 goto err_inval;
1393 }
1394#endif
1395
1396 err = -ENOBUFS;
1397 if (fib_info_cnt >= fib_info_hash_size) {
1398 unsigned int new_size = fib_info_hash_size << 1;
1399 struct hlist_head *new_info_hash;
1400 struct hlist_head *new_laddrhash;
1401 unsigned int bytes;
1402
1403 if (!new_size)
1404 new_size = 16;
1405 bytes = new_size * sizeof(struct hlist_head *);
1406 new_info_hash = fib_info_hash_alloc(bytes);
1407 new_laddrhash = fib_info_hash_alloc(bytes);
1408 if (!new_info_hash || !new_laddrhash) {
1409 fib_info_hash_free(new_info_hash, bytes);
1410 fib_info_hash_free(new_laddrhash, bytes);
1411 } else
1412 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1413
1414 if (!fib_info_hash_size)
1415 goto failure;
1416 }
1417
1418 fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1419 if (!fi)
1420 goto failure;
1421 fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1422 cfg->fc_mx_len, extack);
1423 if (IS_ERR(fi->fib_metrics)) {
1424 err = PTR_ERR(fi->fib_metrics);
1425 kfree(fi);
1426 return ERR_PTR(err);
1427 }
1428
1429 fib_info_cnt++;
1430 fi->fib_net = net;
1431 fi->fib_protocol = cfg->fc_protocol;
1432 fi->fib_scope = cfg->fc_scope;
1433 fi->fib_flags = cfg->fc_flags;
1434 fi->fib_priority = cfg->fc_priority;
1435 fi->fib_prefsrc = cfg->fc_prefsrc;
1436 fi->fib_type = cfg->fc_type;
1437 fi->fib_tb_id = cfg->fc_table;
1438
1439 fi->fib_nhs = nhs;
1440 if (nh) {
1441 if (!nexthop_get(nh)) {
1442 NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1443 err = -EINVAL;
1444 } else {
1445 err = 0;
1446 fi->nh = nh;
1447 }
1448 } else {
1449 change_nexthops(fi) {
1450 nexthop_nh->nh_parent = fi;
1451 } endfor_nexthops(fi)
1452
1453 if (cfg->fc_mp)
1454 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1455 extack);
1456 else
1457 err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1458 }
1459
1460 if (err != 0)
1461 goto failure;
1462
1463 if (fib_props[cfg->fc_type].error) {
1464 if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1465 NL_SET_ERR_MSG(extack,
1466 "Gateway, device and multipath can not be specified for this route type");
1467 goto err_inval;
1468 }
1469 goto link_it;
1470 } else {
1471 switch (cfg->fc_type) {
1472 case RTN_UNICAST:
1473 case RTN_LOCAL:
1474 case RTN_BROADCAST:
1475 case RTN_ANYCAST:
1476 case RTN_MULTICAST:
1477 break;
1478 default:
1479 NL_SET_ERR_MSG(extack, "Invalid route type");
1480 goto err_inval;
1481 }
1482 }
1483
1484 if (cfg->fc_scope > RT_SCOPE_HOST) {
1485 NL_SET_ERR_MSG(extack, "Invalid scope");
1486 goto err_inval;
1487 }
1488
1489 if (fi->nh) {
1490 err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1491 if (err)
1492 goto failure;
1493 } else if (cfg->fc_scope == RT_SCOPE_HOST) {
1494 struct fib_nh *nh = fi->fib_nh;
1495
1496 /* Local address is added. */
1497 if (nhs != 1) {
1498 NL_SET_ERR_MSG(extack,
1499 "Route with host scope can not have multiple nexthops");
1500 goto err_inval;
1501 }
1502 if (nh->fib_nh_gw_family) {
1503 NL_SET_ERR_MSG(extack,
1504 "Route with host scope can not have a gateway");
1505 goto err_inval;
1506 }
1507 nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1508 nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1509 err = -ENODEV;
1510 if (!nh->fib_nh_dev)
1511 goto failure;
1512 } else {
1513 int linkdown = 0;
1514
1515 change_nexthops(fi) {
1516 err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1517 cfg->fc_table, cfg->fc_scope,
1518 extack);
1519 if (err != 0)
1520 goto failure;
1521 if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1522 linkdown++;
1523 } endfor_nexthops(fi)
1524 if (linkdown == fi->fib_nhs)
1525 fi->fib_flags |= RTNH_F_LINKDOWN;
1526 }
1527
1528 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1529 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1530 goto err_inval;
1531 }
1532
1533 if (!fi->nh) {
1534 change_nexthops(fi) {
1535 fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1536 fi->fib_scope);
1537 if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1538 fi->fib_nh_is_v6 = true;
1539 } endfor_nexthops(fi)
1540
1541 fib_rebalance(fi);
1542 }
1543
1544link_it:
1545 ofi = fib_find_info(fi);
1546 if (ofi) {
1547 fi->fib_dead = 1;
1548 free_fib_info(fi);
1549 ofi->fib_treeref++;
1550 return ofi;
1551 }
1552
1553 fi->fib_treeref++;
1554 refcount_set(&fi->fib_clntref, 1);
1555 spin_lock_bh(&fib_info_lock);
1556 hlist_add_head(&fi->fib_hash,
1557 &fib_info_hash[fib_info_hashfn(fi)]);
1558 if (fi->fib_prefsrc) {
1559 struct hlist_head *head;
1560
1561 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1562 hlist_add_head(&fi->fib_lhash, head);
1563 }
1564 if (fi->nh) {
1565 list_add(&fi->nh_list, &nh->fi_list);
1566 } else {
1567 change_nexthops(fi) {
1568 struct hlist_head *head;
1569 unsigned int hash;
1570
1571 if (!nexthop_nh->fib_nh_dev)
1572 continue;
1573 hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex);
1574 head = &fib_info_devhash[hash];
1575 hlist_add_head(&nexthop_nh->nh_hash, head);
1576 } endfor_nexthops(fi)
1577 }
1578 spin_unlock_bh(&fib_info_lock);
1579 return fi;
1580
1581err_inval:
1582 err = -EINVAL;
1583
1584failure:
1585 if (fi) {
1586 fi->fib_dead = 1;
1587 free_fib_info(fi);
1588 }
1589
1590 return ERR_PTR(err);
1591}
1592
1593int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1594 u8 rt_family, unsigned char *flags, bool skip_oif)
1595{
1596 if (nhc->nhc_flags & RTNH_F_DEAD)
1597 *flags |= RTNH_F_DEAD;
1598
1599 if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1600 *flags |= RTNH_F_LINKDOWN;
1601
1602 rcu_read_lock();
1603 switch (nhc->nhc_family) {
1604 case AF_INET:
1605 if (ip_ignore_linkdown(nhc->nhc_dev))
1606 *flags |= RTNH_F_DEAD;
1607 break;
1608 case AF_INET6:
1609 if (ip6_ignore_linkdown(nhc->nhc_dev))
1610 *flags |= RTNH_F_DEAD;
1611 break;
1612 }
1613 rcu_read_unlock();
1614 }
1615
1616 switch (nhc->nhc_gw_family) {
1617 case AF_INET:
1618 if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1619 goto nla_put_failure;
1620 break;
1621 case AF_INET6:
1622 /* if gateway family does not match nexthop family
1623 * gateway is encoded as RTA_VIA
1624 */
1625 if (rt_family != nhc->nhc_gw_family) {
1626 int alen = sizeof(struct in6_addr);
1627 struct nlattr *nla;
1628 struct rtvia *via;
1629
1630 nla = nla_reserve(skb, RTA_VIA, alen + 2);
1631 if (!nla)
1632 goto nla_put_failure;
1633
1634 via = nla_data(nla);
1635 via->rtvia_family = AF_INET6;
1636 memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1637 } else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1638 &nhc->nhc_gw.ipv6) < 0) {
1639 goto nla_put_failure;
1640 }
1641 break;
1642 }
1643
1644 *flags |= (nhc->nhc_flags & RTNH_F_ONLINK);
1645 if (nhc->nhc_flags & RTNH_F_OFFLOAD)
1646 *flags |= RTNH_F_OFFLOAD;
1647
1648 if (!skip_oif && nhc->nhc_dev &&
1649 nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1650 goto nla_put_failure;
1651
1652 if (nhc->nhc_lwtstate &&
1653 lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1654 RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1655 goto nla_put_failure;
1656
1657 return 0;
1658
1659nla_put_failure:
1660 return -EMSGSIZE;
1661}
1662EXPORT_SYMBOL_GPL(fib_nexthop_info);
1663
1664#if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1665int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1666 int nh_weight, u8 rt_family)
1667{
1668 const struct net_device *dev = nhc->nhc_dev;
1669 struct rtnexthop *rtnh;
1670 unsigned char flags = 0;
1671
1672 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1673 if (!rtnh)
1674 goto nla_put_failure;
1675
1676 rtnh->rtnh_hops = nh_weight - 1;
1677 rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1678
1679 if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1680 goto nla_put_failure;
1681
1682 rtnh->rtnh_flags = flags;
1683
1684 /* length of rtnetlink header + attributes */
1685 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1686
1687 return 0;
1688
1689nla_put_failure:
1690 return -EMSGSIZE;
1691}
1692EXPORT_SYMBOL_GPL(fib_add_nexthop);
1693#endif
1694
1695#ifdef CONFIG_IP_ROUTE_MULTIPATH
1696static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1697{
1698 struct nlattr *mp;
1699
1700 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1701 if (!mp)
1702 goto nla_put_failure;
1703
1704 if (unlikely(fi->nh)) {
1705 if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1706 goto nla_put_failure;
1707 goto mp_end;
1708 }
1709
1710 for_nexthops(fi) {
1711 if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1712 AF_INET) < 0)
1713 goto nla_put_failure;
1714#ifdef CONFIG_IP_ROUTE_CLASSID
1715 if (nh->nh_tclassid &&
1716 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1717 goto nla_put_failure;
1718#endif
1719 } endfor_nexthops(fi);
1720
1721mp_end:
1722 nla_nest_end(skb, mp);
1723
1724 return 0;
1725
1726nla_put_failure:
1727 return -EMSGSIZE;
1728}
1729#else
1730static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1731{
1732 return 0;
1733}
1734#endif
1735
1736int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1737 struct fib_rt_info *fri, unsigned int flags)
1738{
1739 unsigned int nhs = fib_info_num_path(fri->fi);
1740 struct fib_info *fi = fri->fi;
1741 u32 tb_id = fri->tb_id;
1742 struct nlmsghdr *nlh;
1743 struct rtmsg *rtm;
1744
1745 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1746 if (!nlh)
1747 return -EMSGSIZE;
1748
1749 rtm = nlmsg_data(nlh);
1750 rtm->rtm_family = AF_INET;
1751 rtm->rtm_dst_len = fri->dst_len;
1752 rtm->rtm_src_len = 0;
1753 rtm->rtm_tos = fri->tos;
1754 if (tb_id < 256)
1755 rtm->rtm_table = tb_id;
1756 else
1757 rtm->rtm_table = RT_TABLE_COMPAT;
1758 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1759 goto nla_put_failure;
1760 rtm->rtm_type = fri->type;
1761 rtm->rtm_flags = fi->fib_flags;
1762 rtm->rtm_scope = fi->fib_scope;
1763 rtm->rtm_protocol = fi->fib_protocol;
1764
1765 if (rtm->rtm_dst_len &&
1766 nla_put_in_addr(skb, RTA_DST, fri->dst))
1767 goto nla_put_failure;
1768 if (fi->fib_priority &&
1769 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1770 goto nla_put_failure;
1771 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1772 goto nla_put_failure;
1773
1774 if (fi->fib_prefsrc &&
1775 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1776 goto nla_put_failure;
1777
1778 if (fi->nh) {
1779 if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1780 goto nla_put_failure;
1781 if (nexthop_is_blackhole(fi->nh))
1782 rtm->rtm_type = RTN_BLACKHOLE;
1783 if (!fi->fib_net->ipv4.sysctl_nexthop_compat_mode)
1784 goto offload;
1785 }
1786
1787 if (nhs == 1) {
1788 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1789 unsigned char flags = 0;
1790
1791 if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1792 goto nla_put_failure;
1793
1794 rtm->rtm_flags = flags;
1795#ifdef CONFIG_IP_ROUTE_CLASSID
1796 if (nhc->nhc_family == AF_INET) {
1797 struct fib_nh *nh;
1798
1799 nh = container_of(nhc, struct fib_nh, nh_common);
1800 if (nh->nh_tclassid &&
1801 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1802 goto nla_put_failure;
1803 }
1804#endif
1805 } else {
1806 if (fib_add_multipath(skb, fi) < 0)
1807 goto nla_put_failure;
1808 }
1809
1810offload:
1811 if (fri->offload)
1812 rtm->rtm_flags |= RTM_F_OFFLOAD;
1813 if (fri->trap)
1814 rtm->rtm_flags |= RTM_F_TRAP;
1815
1816 nlmsg_end(skb, nlh);
1817 return 0;
1818
1819nla_put_failure:
1820 nlmsg_cancel(skb, nlh);
1821 return -EMSGSIZE;
1822}
1823
1824/*
1825 * Update FIB if:
1826 * - local address disappeared -> we must delete all the entries
1827 * referring to it.
1828 * - device went down -> we must shutdown all nexthops going via it.
1829 */
1830int fib_sync_down_addr(struct net_device *dev, __be32 local)
1831{
1832 int ret = 0;
1833 unsigned int hash = fib_laddr_hashfn(local);
1834 struct hlist_head *head = &fib_info_laddrhash[hash];
1835 int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1836 struct net *net = dev_net(dev);
1837 struct fib_info *fi;
1838
1839 if (!fib_info_laddrhash || local == 0)
1840 return 0;
1841
1842 hlist_for_each_entry(fi, head, fib_lhash) {
1843 if (!net_eq(fi->fib_net, net) ||
1844 fi->fib_tb_id != tb_id)
1845 continue;
1846 if (fi->fib_prefsrc == local) {
1847 fi->fib_flags |= RTNH_F_DEAD;
1848 ret++;
1849 }
1850 }
1851 return ret;
1852}
1853
1854static int call_fib_nh_notifiers(struct fib_nh *nh,
1855 enum fib_event_type event_type)
1856{
1857 bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1858 struct fib_nh_notifier_info info = {
1859 .fib_nh = nh,
1860 };
1861
1862 switch (event_type) {
1863 case FIB_EVENT_NH_ADD:
1864 if (nh->fib_nh_flags & RTNH_F_DEAD)
1865 break;
1866 if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1867 break;
1868 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1869 &info.info);
1870 case FIB_EVENT_NH_DEL:
1871 if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1872 (nh->fib_nh_flags & RTNH_F_DEAD))
1873 return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1874 event_type, &info.info);
1875 default:
1876 break;
1877 }
1878
1879 return NOTIFY_DONE;
1880}
1881
1882/* Update the PMTU of exceptions when:
1883 * - the new MTU of the first hop becomes smaller than the PMTU
1884 * - the old MTU was the same as the PMTU, and it limited discovery of
1885 * larger MTUs on the path. With that limit raised, we can now
1886 * discover larger MTUs
1887 * A special case is locked exceptions, for which the PMTU is smaller
1888 * than the minimal accepted PMTU:
1889 * - if the new MTU is greater than the PMTU, don't make any change
1890 * - otherwise, unlock and set PMTU
1891 */
1892void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1893{
1894 struct fnhe_hash_bucket *bucket;
1895 int i;
1896
1897 bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1898 if (!bucket)
1899 return;
1900
1901 for (i = 0; i < FNHE_HASH_SIZE; i++) {
1902 struct fib_nh_exception *fnhe;
1903
1904 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1905 fnhe;
1906 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1907 if (fnhe->fnhe_mtu_locked) {
1908 if (new <= fnhe->fnhe_pmtu) {
1909 fnhe->fnhe_pmtu = new;
1910 fnhe->fnhe_mtu_locked = false;
1911 }
1912 } else if (new < fnhe->fnhe_pmtu ||
1913 orig == fnhe->fnhe_pmtu) {
1914 fnhe->fnhe_pmtu = new;
1915 }
1916 }
1917 }
1918}
1919
1920void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1921{
1922 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1923 struct hlist_head *head = &fib_info_devhash[hash];
1924 struct fib_nh *nh;
1925
1926 hlist_for_each_entry(nh, head, nh_hash) {
1927 if (nh->fib_nh_dev == dev)
1928 fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1929 }
1930}
1931
1932/* Event force Flags Description
1933 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host
1934 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host
1935 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed
1936 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed
1937 *
1938 * only used when fib_nh is built into fib_info
1939 */
1940int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1941{
1942 int ret = 0;
1943 int scope = RT_SCOPE_NOWHERE;
1944 struct fib_info *prev_fi = NULL;
1945 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1946 struct hlist_head *head = &fib_info_devhash[hash];
1947 struct fib_nh *nh;
1948
1949 if (force)
1950 scope = -1;
1951
1952 hlist_for_each_entry(nh, head, nh_hash) {
1953 struct fib_info *fi = nh->nh_parent;
1954 int dead;
1955
1956 BUG_ON(!fi->fib_nhs);
1957 if (nh->fib_nh_dev != dev || fi == prev_fi)
1958 continue;
1959 prev_fi = fi;
1960 dead = 0;
1961 change_nexthops(fi) {
1962 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1963 dead++;
1964 else if (nexthop_nh->fib_nh_dev == dev &&
1965 nexthop_nh->fib_nh_scope != scope) {
1966 switch (event) {
1967 case NETDEV_DOWN:
1968 case NETDEV_UNREGISTER:
1969 nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
1970 fallthrough;
1971 case NETDEV_CHANGE:
1972 nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1973 break;
1974 }
1975 call_fib_nh_notifiers(nexthop_nh,
1976 FIB_EVENT_NH_DEL);
1977 dead++;
1978 }
1979#ifdef CONFIG_IP_ROUTE_MULTIPATH
1980 if (event == NETDEV_UNREGISTER &&
1981 nexthop_nh->fib_nh_dev == dev) {
1982 dead = fi->fib_nhs;
1983 break;
1984 }
1985#endif
1986 } endfor_nexthops(fi)
1987 if (dead == fi->fib_nhs) {
1988 switch (event) {
1989 case NETDEV_DOWN:
1990 case NETDEV_UNREGISTER:
1991 fi->fib_flags |= RTNH_F_DEAD;
1992 fallthrough;
1993 case NETDEV_CHANGE:
1994 fi->fib_flags |= RTNH_F_LINKDOWN;
1995 break;
1996 }
1997 ret++;
1998 }
1999
2000 fib_rebalance(fi);
2001 }
2002
2003 return ret;
2004}
2005
2006/* Must be invoked inside of an RCU protected region. */
2007static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2008{
2009 struct fib_info *fi = NULL, *last_resort = NULL;
2010 struct hlist_head *fa_head = res->fa_head;
2011 struct fib_table *tb = res->table;
2012 u8 slen = 32 - res->prefixlen;
2013 int order = -1, last_idx = -1;
2014 struct fib_alias *fa, *fa1 = NULL;
2015 u32 last_prio = res->fi->fib_priority;
2016 u8 last_tos = 0;
2017
2018 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2019 struct fib_info *next_fi = fa->fa_info;
2020 struct fib_nh_common *nhc;
2021
2022 if (fa->fa_slen != slen)
2023 continue;
2024 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
2025 continue;
2026 if (fa->tb_id != tb->tb_id)
2027 continue;
2028 if (next_fi->fib_priority > last_prio &&
2029 fa->fa_tos == last_tos) {
2030 if (last_tos)
2031 continue;
2032 break;
2033 }
2034 if (next_fi->fib_flags & RTNH_F_DEAD)
2035 continue;
2036 last_tos = fa->fa_tos;
2037 last_prio = next_fi->fib_priority;
2038
2039 if (next_fi->fib_scope != res->scope ||
2040 fa->fa_type != RTN_UNICAST)
2041 continue;
2042
2043 nhc = fib_info_nhc(next_fi, 0);
2044 if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2045 continue;
2046
2047 fib_alias_accessed(fa);
2048
2049 if (!fi) {
2050 if (next_fi != res->fi)
2051 break;
2052 fa1 = fa;
2053 } else if (!fib_detect_death(fi, order, &last_resort,
2054 &last_idx, fa1->fa_default)) {
2055 fib_result_assign(res, fi);
2056 fa1->fa_default = order;
2057 goto out;
2058 }
2059 fi = next_fi;
2060 order++;
2061 }
2062
2063 if (order <= 0 || !fi) {
2064 if (fa1)
2065 fa1->fa_default = -1;
2066 goto out;
2067 }
2068
2069 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2070 fa1->fa_default)) {
2071 fib_result_assign(res, fi);
2072 fa1->fa_default = order;
2073 goto out;
2074 }
2075
2076 if (last_idx >= 0)
2077 fib_result_assign(res, last_resort);
2078 fa1->fa_default = last_idx;
2079out:
2080 return;
2081}
2082
2083/*
2084 * Dead device goes up. We wake up dead nexthops.
2085 * It takes sense only on multipath routes.
2086 *
2087 * only used when fib_nh is built into fib_info
2088 */
2089int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2090{
2091 struct fib_info *prev_fi;
2092 unsigned int hash;
2093 struct hlist_head *head;
2094 struct fib_nh *nh;
2095 int ret;
2096
2097 if (!(dev->flags & IFF_UP))
2098 return 0;
2099
2100 if (nh_flags & RTNH_F_DEAD) {
2101 unsigned int flags = dev_get_flags(dev);
2102
2103 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2104 nh_flags |= RTNH_F_LINKDOWN;
2105 }
2106
2107 prev_fi = NULL;
2108 hash = fib_devindex_hashfn(dev->ifindex);
2109 head = &fib_info_devhash[hash];
2110 ret = 0;
2111
2112 hlist_for_each_entry(nh, head, nh_hash) {
2113 struct fib_info *fi = nh->nh_parent;
2114 int alive;
2115
2116 BUG_ON(!fi->fib_nhs);
2117 if (nh->fib_nh_dev != dev || fi == prev_fi)
2118 continue;
2119
2120 prev_fi = fi;
2121 alive = 0;
2122 change_nexthops(fi) {
2123 if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2124 alive++;
2125 continue;
2126 }
2127 if (!nexthop_nh->fib_nh_dev ||
2128 !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2129 continue;
2130 if (nexthop_nh->fib_nh_dev != dev ||
2131 !__in_dev_get_rtnl(dev))
2132 continue;
2133 alive++;
2134 nexthop_nh->fib_nh_flags &= ~nh_flags;
2135 call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2136 } endfor_nexthops(fi)
2137
2138 if (alive > 0) {
2139 fi->fib_flags &= ~nh_flags;
2140 ret++;
2141 }
2142
2143 fib_rebalance(fi);
2144 }
2145
2146 return ret;
2147}
2148
2149#ifdef CONFIG_IP_ROUTE_MULTIPATH
2150static bool fib_good_nh(const struct fib_nh *nh)
2151{
2152 int state = NUD_REACHABLE;
2153
2154 if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2155 struct neighbour *n;
2156
2157 rcu_read_lock_bh();
2158
2159 if (likely(nh->fib_nh_gw_family == AF_INET))
2160 n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2161 (__force u32)nh->fib_nh_gw4);
2162 else if (nh->fib_nh_gw_family == AF_INET6)
2163 n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2164 &nh->fib_nh_gw6);
2165 else
2166 n = NULL;
2167 if (n)
2168 state = n->nud_state;
2169
2170 rcu_read_unlock_bh();
2171 }
2172
2173 return !!(state & NUD_VALID);
2174}
2175
2176void fib_select_multipath(struct fib_result *res, int hash)
2177{
2178 struct fib_info *fi = res->fi;
2179 struct net *net = fi->fib_net;
2180 bool first = false;
2181
2182 if (unlikely(res->fi->nh)) {
2183 nexthop_path_fib_result(res, hash);
2184 return;
2185 }
2186
2187 change_nexthops(fi) {
2188 if (net->ipv4.sysctl_fib_multipath_use_neigh) {
2189 if (!fib_good_nh(nexthop_nh))
2190 continue;
2191 if (!first) {
2192 res->nh_sel = nhsel;
2193 res->nhc = &nexthop_nh->nh_common;
2194 first = true;
2195 }
2196 }
2197
2198 if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2199 continue;
2200
2201 res->nh_sel = nhsel;
2202 res->nhc = &nexthop_nh->nh_common;
2203 return;
2204 } endfor_nexthops(fi);
2205}
2206#endif
2207
2208void fib_select_path(struct net *net, struct fib_result *res,
2209 struct flowi4 *fl4, const struct sk_buff *skb)
2210{
2211 if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2212 goto check_saddr;
2213
2214#ifdef CONFIG_IP_ROUTE_MULTIPATH
2215 if (fib_info_num_path(res->fi) > 1) {
2216 int h = fib_multipath_hash(net, fl4, skb, NULL);
2217
2218 fib_select_multipath(res, h);
2219 }
2220 else
2221#endif
2222 if (!res->prefixlen &&
2223 res->table->tb_num_default > 1 &&
2224 res->type == RTN_UNICAST)
2225 fib_select_default(fl4, res);
2226
2227check_saddr:
2228 if (!fl4->saddr)
2229 fl4->saddr = fib_result_prefsrc(net, res);
2230}