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#include <linux/hash.h>
33#include <linux/nospec.h>
34
35#include <net/arp.h>
36#include <net/inet_dscp.h>
37#include <net/ip.h>
38#include <net/protocol.h>
39#include <net/route.h>
40#include <net/tcp.h>
41#include <net/sock.h>
42#include <net/ip_fib.h>
43#include <net/ip6_fib.h>
44#include <net/nexthop.h>
45#include <net/netlink.h>
46#include <net/rtnh.h>
47#include <net/lwtunnel.h>
48#include <net/fib_notifier.h>
49#include <net/addrconf.h>
50
51#include "fib_lookup.h"
52
53static DEFINE_SPINLOCK(fib_info_lock);
54static struct hlist_head *fib_info_hash;
55static struct hlist_head *fib_info_laddrhash;
56static unsigned int fib_info_hash_size;
57static unsigned int fib_info_hash_bits;
58static unsigned int fib_info_cnt;
59
60#define DEVINDEX_HASHBITS 8
61#define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
62static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
63
64/* for_nexthops and change_nexthops only used when nexthop object
65 * is not set in a fib_info. The logic within can reference fib_nh.
66 */
67#ifdef CONFIG_IP_ROUTE_MULTIPATH
68
69#define for_nexthops(fi) { \
70 int nhsel; const struct fib_nh *nh; \
71 for (nhsel = 0, nh = (fi)->fib_nh; \
72 nhsel < fib_info_num_path((fi)); \
73 nh++, nhsel++)
74
75#define change_nexthops(fi) { \
76 int nhsel; struct fib_nh *nexthop_nh; \
77 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
78 nhsel < fib_info_num_path((fi)); \
79 nexthop_nh++, nhsel++)
80
81#else /* CONFIG_IP_ROUTE_MULTIPATH */
82
83/* Hope, that gcc will optimize it to get rid of dummy loop */
84
85#define for_nexthops(fi) { \
86 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \
87 for (nhsel = 0; nhsel < 1; nhsel++)
88
89#define change_nexthops(fi) { \
90 int nhsel; \
91 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
92 for (nhsel = 0; nhsel < 1; nhsel++)
93
94#endif /* CONFIG_IP_ROUTE_MULTIPATH */
95
96#define endfor_nexthops(fi) }
97
98
99const struct fib_prop fib_props[RTN_MAX + 1] = {
100 [RTN_UNSPEC] = {
101 .error = 0,
102 .scope = RT_SCOPE_NOWHERE,
103 },
104 [RTN_UNICAST] = {
105 .error = 0,
106 .scope = RT_SCOPE_UNIVERSE,
107 },
108 [RTN_LOCAL] = {
109 .error = 0,
110 .scope = RT_SCOPE_HOST,
111 },
112 [RTN_BROADCAST] = {
113 .error = 0,
114 .scope = RT_SCOPE_LINK,
115 },
116 [RTN_ANYCAST] = {
117 .error = 0,
118 .scope = RT_SCOPE_LINK,
119 },
120 [RTN_MULTICAST] = {
121 .error = 0,
122 .scope = RT_SCOPE_UNIVERSE,
123 },
124 [RTN_BLACKHOLE] = {
125 .error = -EINVAL,
126 .scope = RT_SCOPE_UNIVERSE,
127 },
128 [RTN_UNREACHABLE] = {
129 .error = -EHOSTUNREACH,
130 .scope = RT_SCOPE_UNIVERSE,
131 },
132 [RTN_PROHIBIT] = {
133 .error = -EACCES,
134 .scope = RT_SCOPE_UNIVERSE,
135 },
136 [RTN_THROW] = {
137 .error = -EAGAIN,
138 .scope = RT_SCOPE_UNIVERSE,
139 },
140 [RTN_NAT] = {
141 .error = -EINVAL,
142 .scope = RT_SCOPE_NOWHERE,
143 },
144 [RTN_XRESOLVE] = {
145 .error = -EINVAL,
146 .scope = RT_SCOPE_NOWHERE,
147 },
148};
149
150static void rt_fibinfo_free(struct rtable __rcu **rtp)
151{
152 struct rtable *rt = rcu_dereference_protected(*rtp, 1);
153
154 if (!rt)
155 return;
156
157 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
158 * because we waited an RCU grace period before calling
159 * free_fib_info_rcu()
160 */
161
162 dst_dev_put(&rt->dst);
163 dst_release_immediate(&rt->dst);
164}
165
166static void free_nh_exceptions(struct fib_nh_common *nhc)
167{
168 struct fnhe_hash_bucket *hash;
169 int i;
170
171 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
172 if (!hash)
173 return;
174 for (i = 0; i < FNHE_HASH_SIZE; i++) {
175 struct fib_nh_exception *fnhe;
176
177 fnhe = rcu_dereference_protected(hash[i].chain, 1);
178 while (fnhe) {
179 struct fib_nh_exception *next;
180
181 next = rcu_dereference_protected(fnhe->fnhe_next, 1);
182
183 rt_fibinfo_free(&fnhe->fnhe_rth_input);
184 rt_fibinfo_free(&fnhe->fnhe_rth_output);
185
186 kfree(fnhe);
187
188 fnhe = next;
189 }
190 }
191 kfree(hash);
192}
193
194static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
195{
196 int cpu;
197
198 if (!rtp)
199 return;
200
201 for_each_possible_cpu(cpu) {
202 struct rtable *rt;
203
204 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
205 if (rt) {
206 dst_dev_put(&rt->dst);
207 dst_release_immediate(&rt->dst);
208 }
209 }
210 free_percpu(rtp);
211}
212
213void fib_nh_common_release(struct fib_nh_common *nhc)
214{
215 netdev_put(nhc->nhc_dev, &nhc->nhc_dev_tracker);
216 lwtstate_put(nhc->nhc_lwtstate);
217 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
218 rt_fibinfo_free(&nhc->nhc_rth_input);
219 free_nh_exceptions(nhc);
220}
221EXPORT_SYMBOL_GPL(fib_nh_common_release);
222
223void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
224{
225#ifdef CONFIG_IP_ROUTE_CLASSID
226 if (fib_nh->nh_tclassid)
227 atomic_dec(&net->ipv4.fib_num_tclassid_users);
228#endif
229 fib_nh_common_release(&fib_nh->nh_common);
230}
231
232/* Release a nexthop info record */
233static void free_fib_info_rcu(struct rcu_head *head)
234{
235 struct fib_info *fi = container_of(head, struct fib_info, rcu);
236
237 if (fi->nh) {
238 nexthop_put(fi->nh);
239 } else {
240 change_nexthops(fi) {
241 fib_nh_release(fi->fib_net, nexthop_nh);
242 } endfor_nexthops(fi);
243 }
244
245 ip_fib_metrics_put(fi->fib_metrics);
246
247 kfree(fi);
248}
249
250void free_fib_info(struct fib_info *fi)
251{
252 if (fi->fib_dead == 0) {
253 pr_warn("Freeing alive fib_info %p\n", fi);
254 return;
255 }
256
257 call_rcu(&fi->rcu, free_fib_info_rcu);
258}
259EXPORT_SYMBOL_GPL(free_fib_info);
260
261void fib_release_info(struct fib_info *fi)
262{
263 spin_lock_bh(&fib_info_lock);
264 if (fi && refcount_dec_and_test(&fi->fib_treeref)) {
265 hlist_del(&fi->fib_hash);
266
267 /* Paired with READ_ONCE() in fib_create_info(). */
268 WRITE_ONCE(fib_info_cnt, fib_info_cnt - 1);
269
270 if (fi->fib_prefsrc)
271 hlist_del(&fi->fib_lhash);
272 if (fi->nh) {
273 list_del(&fi->nh_list);
274 } else {
275 change_nexthops(fi) {
276 if (!nexthop_nh->fib_nh_dev)
277 continue;
278 hlist_del(&nexthop_nh->nh_hash);
279 } endfor_nexthops(fi)
280 }
281 /* Paired with READ_ONCE() from fib_table_lookup() */
282 WRITE_ONCE(fi->fib_dead, 1);
283 fib_info_put(fi);
284 }
285 spin_unlock_bh(&fib_info_lock);
286}
287
288static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
289{
290 const struct fib_nh *onh;
291
292 if (fi->nh || ofi->nh)
293 return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
294
295 if (ofi->fib_nhs == 0)
296 return 0;
297
298 for_nexthops(fi) {
299 onh = fib_info_nh(ofi, nhsel);
300
301 if (nh->fib_nh_oif != onh->fib_nh_oif ||
302 nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
303 nh->fib_nh_scope != onh->fib_nh_scope ||
304#ifdef CONFIG_IP_ROUTE_MULTIPATH
305 nh->fib_nh_weight != onh->fib_nh_weight ||
306#endif
307#ifdef CONFIG_IP_ROUTE_CLASSID
308 nh->nh_tclassid != onh->nh_tclassid ||
309#endif
310 lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
311 ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
312 return -1;
313
314 if (nh->fib_nh_gw_family == AF_INET &&
315 nh->fib_nh_gw4 != onh->fib_nh_gw4)
316 return -1;
317
318 if (nh->fib_nh_gw_family == AF_INET6 &&
319 ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
320 return -1;
321 } endfor_nexthops(fi);
322 return 0;
323}
324
325static inline unsigned int fib_devindex_hashfn(unsigned int val)
326{
327 return hash_32(val, DEVINDEX_HASHBITS);
328}
329
330static struct hlist_head *
331fib_info_devhash_bucket(const struct net_device *dev)
332{
333 u32 val = net_hash_mix(dev_net(dev)) ^ dev->ifindex;
334
335 return &fib_info_devhash[fib_devindex_hashfn(val)];
336}
337
338static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
339 u32 prefsrc, u32 priority)
340{
341 unsigned int val = init_val;
342
343 val ^= (protocol << 8) | scope;
344 val ^= prefsrc;
345 val ^= priority;
346
347 return val;
348}
349
350static unsigned int fib_info_hashfn_result(unsigned int val)
351{
352 unsigned int mask = (fib_info_hash_size - 1);
353
354 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
355}
356
357static inline unsigned int fib_info_hashfn(struct fib_info *fi)
358{
359 unsigned int val;
360
361 val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
362 fi->fib_scope, (__force u32)fi->fib_prefsrc,
363 fi->fib_priority);
364
365 if (fi->nh) {
366 val ^= fib_devindex_hashfn(fi->nh->id);
367 } else {
368 for_nexthops(fi) {
369 val ^= fib_devindex_hashfn(nh->fib_nh_oif);
370 } endfor_nexthops(fi)
371 }
372
373 return fib_info_hashfn_result(val);
374}
375
376/* no metrics, only nexthop id */
377static struct fib_info *fib_find_info_nh(struct net *net,
378 const struct fib_config *cfg)
379{
380 struct hlist_head *head;
381 struct fib_info *fi;
382 unsigned int hash;
383
384 hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
385 cfg->fc_protocol, cfg->fc_scope,
386 (__force u32)cfg->fc_prefsrc,
387 cfg->fc_priority);
388 hash = fib_info_hashfn_result(hash);
389 head = &fib_info_hash[hash];
390
391 hlist_for_each_entry(fi, head, fib_hash) {
392 if (!net_eq(fi->fib_net, net))
393 continue;
394 if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
395 continue;
396 if (cfg->fc_protocol == fi->fib_protocol &&
397 cfg->fc_scope == fi->fib_scope &&
398 cfg->fc_prefsrc == fi->fib_prefsrc &&
399 cfg->fc_priority == fi->fib_priority &&
400 cfg->fc_type == fi->fib_type &&
401 cfg->fc_table == fi->fib_tb_id &&
402 !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
403 return fi;
404 }
405
406 return NULL;
407}
408
409static struct fib_info *fib_find_info(struct fib_info *nfi)
410{
411 struct hlist_head *head;
412 struct fib_info *fi;
413 unsigned int hash;
414
415 hash = fib_info_hashfn(nfi);
416 head = &fib_info_hash[hash];
417
418 hlist_for_each_entry(fi, head, fib_hash) {
419 if (!net_eq(fi->fib_net, nfi->fib_net))
420 continue;
421 if (fi->fib_nhs != nfi->fib_nhs)
422 continue;
423 if (nfi->fib_protocol == fi->fib_protocol &&
424 nfi->fib_scope == fi->fib_scope &&
425 nfi->fib_prefsrc == fi->fib_prefsrc &&
426 nfi->fib_priority == fi->fib_priority &&
427 nfi->fib_type == fi->fib_type &&
428 nfi->fib_tb_id == fi->fib_tb_id &&
429 memcmp(nfi->fib_metrics, fi->fib_metrics,
430 sizeof(u32) * RTAX_MAX) == 0 &&
431 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
432 nh_comp(fi, nfi) == 0)
433 return fi;
434 }
435
436 return NULL;
437}
438
439/* Check, that the gateway is already configured.
440 * Used only by redirect accept routine.
441 */
442int ip_fib_check_default(__be32 gw, struct net_device *dev)
443{
444 struct hlist_head *head;
445 struct fib_nh *nh;
446
447 spin_lock(&fib_info_lock);
448
449 head = fib_info_devhash_bucket(dev);
450
451 hlist_for_each_entry(nh, head, nh_hash) {
452 if (nh->fib_nh_dev == dev &&
453 nh->fib_nh_gw4 == gw &&
454 !(nh->fib_nh_flags & RTNH_F_DEAD)) {
455 spin_unlock(&fib_info_lock);
456 return 0;
457 }
458 }
459
460 spin_unlock(&fib_info_lock);
461
462 return -1;
463}
464
465size_t fib_nlmsg_size(struct fib_info *fi)
466{
467 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
468 + nla_total_size(4) /* RTA_TABLE */
469 + nla_total_size(4) /* RTA_DST */
470 + nla_total_size(4) /* RTA_PRIORITY */
471 + nla_total_size(4) /* RTA_PREFSRC */
472 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
473 unsigned int nhs = fib_info_num_path(fi);
474
475 /* space for nested metrics */
476 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
477
478 if (fi->nh)
479 payload += nla_total_size(4); /* RTA_NH_ID */
480
481 if (nhs) {
482 size_t nh_encapsize = 0;
483 /* Also handles the special case nhs == 1 */
484
485 /* each nexthop is packed in an attribute */
486 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
487 unsigned int i;
488
489 /* may contain flow and gateway attribute */
490 nhsize += 2 * nla_total_size(4);
491
492 /* grab encap info */
493 for (i = 0; i < fib_info_num_path(fi); i++) {
494 struct fib_nh_common *nhc = fib_info_nhc(fi, i);
495
496 if (nhc->nhc_lwtstate) {
497 /* RTA_ENCAP_TYPE */
498 nh_encapsize += lwtunnel_get_encap_size(
499 nhc->nhc_lwtstate);
500 /* RTA_ENCAP */
501 nh_encapsize += nla_total_size(2);
502 }
503 }
504
505 /* all nexthops are packed in a nested attribute */
506 payload += nla_total_size((nhs * nhsize) + nh_encapsize);
507
508 }
509
510 return payload;
511}
512
513void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
514 int dst_len, u32 tb_id, const struct nl_info *info,
515 unsigned int nlm_flags)
516{
517 struct fib_rt_info fri;
518 struct sk_buff *skb;
519 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
520 int err = -ENOBUFS;
521
522 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
523 if (!skb)
524 goto errout;
525
526 fri.fi = fa->fa_info;
527 fri.tb_id = tb_id;
528 fri.dst = key;
529 fri.dst_len = dst_len;
530 fri.dscp = fa->fa_dscp;
531 fri.type = fa->fa_type;
532 fri.offload = READ_ONCE(fa->offload);
533 fri.trap = READ_ONCE(fa->trap);
534 fri.offload_failed = READ_ONCE(fa->offload_failed);
535 err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
536 if (err < 0) {
537 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
538 WARN_ON(err == -EMSGSIZE);
539 kfree_skb(skb);
540 goto errout;
541 }
542 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
543 info->nlh, GFP_KERNEL);
544 return;
545errout:
546 if (err < 0)
547 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
548}
549
550static int fib_detect_death(struct fib_info *fi, int order,
551 struct fib_info **last_resort, int *last_idx,
552 int dflt)
553{
554 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
555 struct neighbour *n;
556 int state = NUD_NONE;
557
558 if (likely(nhc->nhc_gw_family == AF_INET))
559 n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
560 else if (nhc->nhc_gw_family == AF_INET6)
561 n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
562 nhc->nhc_dev);
563 else
564 n = NULL;
565
566 if (n) {
567 state = READ_ONCE(n->nud_state);
568 neigh_release(n);
569 } else {
570 return 0;
571 }
572 if (state == NUD_REACHABLE)
573 return 0;
574 if ((state & NUD_VALID) && order != dflt)
575 return 0;
576 if ((state & NUD_VALID) ||
577 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
578 *last_resort = fi;
579 *last_idx = order;
580 }
581 return 1;
582}
583
584int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
585 struct nlattr *encap, u16 encap_type,
586 void *cfg, gfp_t gfp_flags,
587 struct netlink_ext_ack *extack)
588{
589 int err;
590
591 nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
592 gfp_flags);
593 if (!nhc->nhc_pcpu_rth_output)
594 return -ENOMEM;
595
596 if (encap) {
597 struct lwtunnel_state *lwtstate;
598
599 if (encap_type == LWTUNNEL_ENCAP_NONE) {
600 NL_SET_ERR_MSG(extack, "LWT encap type not specified");
601 err = -EINVAL;
602 goto lwt_failure;
603 }
604 err = lwtunnel_build_state(net, encap_type, encap,
605 nhc->nhc_family, cfg, &lwtstate,
606 extack);
607 if (err)
608 goto lwt_failure;
609
610 nhc->nhc_lwtstate = lwtstate_get(lwtstate);
611 }
612
613 return 0;
614
615lwt_failure:
616 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
617 nhc->nhc_pcpu_rth_output = NULL;
618 return err;
619}
620EXPORT_SYMBOL_GPL(fib_nh_common_init);
621
622int fib_nh_init(struct net *net, struct fib_nh *nh,
623 struct fib_config *cfg, int nh_weight,
624 struct netlink_ext_ack *extack)
625{
626 int err;
627
628 nh->fib_nh_family = AF_INET;
629
630 err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
631 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
632 if (err)
633 return err;
634
635 nh->fib_nh_oif = cfg->fc_oif;
636 nh->fib_nh_gw_family = cfg->fc_gw_family;
637 if (cfg->fc_gw_family == AF_INET)
638 nh->fib_nh_gw4 = cfg->fc_gw4;
639 else if (cfg->fc_gw_family == AF_INET6)
640 nh->fib_nh_gw6 = cfg->fc_gw6;
641
642 nh->fib_nh_flags = cfg->fc_flags;
643
644#ifdef CONFIG_IP_ROUTE_CLASSID
645 nh->nh_tclassid = cfg->fc_flow;
646 if (nh->nh_tclassid)
647 atomic_inc(&net->ipv4.fib_num_tclassid_users);
648#endif
649#ifdef CONFIG_IP_ROUTE_MULTIPATH
650 nh->fib_nh_weight = nh_weight;
651#endif
652 return 0;
653}
654
655#ifdef CONFIG_IP_ROUTE_MULTIPATH
656
657static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
658 struct netlink_ext_ack *extack)
659{
660 int nhs = 0;
661
662 while (rtnh_ok(rtnh, remaining)) {
663 nhs++;
664 rtnh = rtnh_next(rtnh, &remaining);
665 }
666
667 /* leftover implies invalid nexthop configuration, discard it */
668 if (remaining > 0) {
669 NL_SET_ERR_MSG(extack,
670 "Invalid nexthop configuration - extra data after nexthops");
671 nhs = 0;
672 }
673
674 return nhs;
675}
676
677static int fib_gw_from_attr(__be32 *gw, struct nlattr *nla,
678 struct netlink_ext_ack *extack)
679{
680 if (nla_len(nla) < sizeof(*gw)) {
681 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_GATEWAY");
682 return -EINVAL;
683 }
684
685 *gw = nla_get_in_addr(nla);
686
687 return 0;
688}
689
690/* only called when fib_nh is integrated into fib_info */
691static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
692 int remaining, struct fib_config *cfg,
693 struct netlink_ext_ack *extack)
694{
695 struct net *net = fi->fib_net;
696 struct fib_config fib_cfg;
697 struct fib_nh *nh;
698 int ret;
699
700 change_nexthops(fi) {
701 int attrlen;
702
703 memset(&fib_cfg, 0, sizeof(fib_cfg));
704
705 if (!rtnh_ok(rtnh, remaining)) {
706 NL_SET_ERR_MSG(extack,
707 "Invalid nexthop configuration - extra data after nexthop");
708 return -EINVAL;
709 }
710
711 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
712 NL_SET_ERR_MSG(extack,
713 "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
714 return -EINVAL;
715 }
716
717 fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
718 fib_cfg.fc_oif = rtnh->rtnh_ifindex;
719
720 attrlen = rtnh_attrlen(rtnh);
721 if (attrlen > 0) {
722 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
723
724 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
725 nlav = nla_find(attrs, attrlen, RTA_VIA);
726 if (nla && nlav) {
727 NL_SET_ERR_MSG(extack,
728 "Nexthop configuration can not contain both GATEWAY and VIA");
729 return -EINVAL;
730 }
731 if (nla) {
732 ret = fib_gw_from_attr(&fib_cfg.fc_gw4, nla,
733 extack);
734 if (ret)
735 goto errout;
736
737 if (fib_cfg.fc_gw4)
738 fib_cfg.fc_gw_family = AF_INET;
739 } else if (nlav) {
740 ret = fib_gw_from_via(&fib_cfg, nlav, extack);
741 if (ret)
742 goto errout;
743 }
744
745 nla = nla_find(attrs, attrlen, RTA_FLOW);
746 if (nla) {
747 if (nla_len(nla) < sizeof(u32)) {
748 NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
749 return -EINVAL;
750 }
751 fib_cfg.fc_flow = nla_get_u32(nla);
752 }
753
754 fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
755 /* RTA_ENCAP_TYPE length checked in
756 * lwtunnel_valid_encap_type_attr
757 */
758 nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
759 if (nla)
760 fib_cfg.fc_encap_type = nla_get_u16(nla);
761 }
762
763 ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
764 rtnh->rtnh_hops + 1, extack);
765 if (ret)
766 goto errout;
767
768 rtnh = rtnh_next(rtnh, &remaining);
769 } endfor_nexthops(fi);
770
771 ret = -EINVAL;
772 nh = fib_info_nh(fi, 0);
773 if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
774 NL_SET_ERR_MSG(extack,
775 "Nexthop device index does not match RTA_OIF");
776 goto errout;
777 }
778 if (cfg->fc_gw_family) {
779 if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
780 (cfg->fc_gw_family == AF_INET &&
781 nh->fib_nh_gw4 != cfg->fc_gw4) ||
782 (cfg->fc_gw_family == AF_INET6 &&
783 ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
784 NL_SET_ERR_MSG(extack,
785 "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
786 goto errout;
787 }
788 }
789#ifdef CONFIG_IP_ROUTE_CLASSID
790 if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
791 NL_SET_ERR_MSG(extack,
792 "Nexthop class id does not match RTA_FLOW");
793 goto errout;
794 }
795#endif
796 ret = 0;
797errout:
798 return ret;
799}
800
801/* only called when fib_nh is integrated into fib_info */
802static void fib_rebalance(struct fib_info *fi)
803{
804 int total;
805 int w;
806
807 if (fib_info_num_path(fi) < 2)
808 return;
809
810 total = 0;
811 for_nexthops(fi) {
812 if (nh->fib_nh_flags & RTNH_F_DEAD)
813 continue;
814
815 if (ip_ignore_linkdown(nh->fib_nh_dev) &&
816 nh->fib_nh_flags & RTNH_F_LINKDOWN)
817 continue;
818
819 total += nh->fib_nh_weight;
820 } endfor_nexthops(fi);
821
822 w = 0;
823 change_nexthops(fi) {
824 int upper_bound;
825
826 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
827 upper_bound = -1;
828 } else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
829 nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
830 upper_bound = -1;
831 } else {
832 w += nexthop_nh->fib_nh_weight;
833 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
834 total) - 1;
835 }
836
837 atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
838 } endfor_nexthops(fi);
839}
840#else /* CONFIG_IP_ROUTE_MULTIPATH */
841
842static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
843 int remaining, struct fib_config *cfg,
844 struct netlink_ext_ack *extack)
845{
846 NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
847
848 return -EINVAL;
849}
850
851#define fib_rebalance(fi) do { } while (0)
852
853#endif /* CONFIG_IP_ROUTE_MULTIPATH */
854
855static int fib_encap_match(struct net *net, u16 encap_type,
856 struct nlattr *encap,
857 const struct fib_nh *nh,
858 const struct fib_config *cfg,
859 struct netlink_ext_ack *extack)
860{
861 struct lwtunnel_state *lwtstate;
862 int ret, result = 0;
863
864 if (encap_type == LWTUNNEL_ENCAP_NONE)
865 return 0;
866
867 ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
868 cfg, &lwtstate, extack);
869 if (!ret) {
870 result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
871 lwtstate_free(lwtstate);
872 }
873
874 return result;
875}
876
877int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
878 struct netlink_ext_ack *extack)
879{
880#ifdef CONFIG_IP_ROUTE_MULTIPATH
881 struct rtnexthop *rtnh;
882 int remaining;
883#endif
884
885 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
886 return 1;
887
888 if (cfg->fc_nh_id) {
889 if (fi->nh && cfg->fc_nh_id == fi->nh->id)
890 return 0;
891 return 1;
892 }
893
894 if (fi->nh) {
895 if (cfg->fc_oif || cfg->fc_gw_family || cfg->fc_mp)
896 return 1;
897 return 0;
898 }
899
900 if (cfg->fc_oif || cfg->fc_gw_family) {
901 struct fib_nh *nh;
902
903 nh = fib_info_nh(fi, 0);
904 if (cfg->fc_encap) {
905 if (fib_encap_match(net, cfg->fc_encap_type,
906 cfg->fc_encap, nh, cfg, extack))
907 return 1;
908 }
909#ifdef CONFIG_IP_ROUTE_CLASSID
910 if (cfg->fc_flow &&
911 cfg->fc_flow != nh->nh_tclassid)
912 return 1;
913#endif
914 if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
915 (cfg->fc_gw_family &&
916 cfg->fc_gw_family != nh->fib_nh_gw_family))
917 return 1;
918
919 if (cfg->fc_gw_family == AF_INET &&
920 cfg->fc_gw4 != nh->fib_nh_gw4)
921 return 1;
922
923 if (cfg->fc_gw_family == AF_INET6 &&
924 ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
925 return 1;
926
927 return 0;
928 }
929
930#ifdef CONFIG_IP_ROUTE_MULTIPATH
931 if (!cfg->fc_mp)
932 return 0;
933
934 rtnh = cfg->fc_mp;
935 remaining = cfg->fc_mp_len;
936
937 for_nexthops(fi) {
938 int attrlen;
939
940 if (!rtnh_ok(rtnh, remaining))
941 return -EINVAL;
942
943 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
944 return 1;
945
946 attrlen = rtnh_attrlen(rtnh);
947 if (attrlen > 0) {
948 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
949 int err;
950
951 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
952 nlav = nla_find(attrs, attrlen, RTA_VIA);
953 if (nla && nlav) {
954 NL_SET_ERR_MSG(extack,
955 "Nexthop configuration can not contain both GATEWAY and VIA");
956 return -EINVAL;
957 }
958
959 if (nla) {
960 __be32 gw;
961
962 err = fib_gw_from_attr(&gw, nla, extack);
963 if (err)
964 return err;
965
966 if (nh->fib_nh_gw_family != AF_INET ||
967 gw != nh->fib_nh_gw4)
968 return 1;
969 } else if (nlav) {
970 struct fib_config cfg2;
971
972 err = fib_gw_from_via(&cfg2, nlav, extack);
973 if (err)
974 return err;
975
976 switch (nh->fib_nh_gw_family) {
977 case AF_INET:
978 if (cfg2.fc_gw_family != AF_INET ||
979 cfg2.fc_gw4 != nh->fib_nh_gw4)
980 return 1;
981 break;
982 case AF_INET6:
983 if (cfg2.fc_gw_family != AF_INET6 ||
984 ipv6_addr_cmp(&cfg2.fc_gw6,
985 &nh->fib_nh_gw6))
986 return 1;
987 break;
988 }
989 }
990
991#ifdef CONFIG_IP_ROUTE_CLASSID
992 nla = nla_find(attrs, attrlen, RTA_FLOW);
993 if (nla) {
994 if (nla_len(nla) < sizeof(u32)) {
995 NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
996 return -EINVAL;
997 }
998 if (nla_get_u32(nla) != nh->nh_tclassid)
999 return 1;
1000 }
1001#endif
1002 }
1003
1004 rtnh = rtnh_next(rtnh, &remaining);
1005 } endfor_nexthops(fi);
1006#endif
1007 return 0;
1008}
1009
1010bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
1011{
1012 struct nlattr *nla;
1013 int remaining;
1014
1015 if (!cfg->fc_mx)
1016 return true;
1017
1018 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1019 int type = nla_type(nla);
1020 u32 fi_val, val;
1021
1022 if (!type)
1023 continue;
1024 if (type > RTAX_MAX)
1025 return false;
1026
1027 type = array_index_nospec(type, RTAX_MAX + 1);
1028 if (type == RTAX_CC_ALGO) {
1029 char tmp[TCP_CA_NAME_MAX];
1030 bool ecn_ca = false;
1031
1032 nla_strscpy(tmp, nla, sizeof(tmp));
1033 val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
1034 } else {
1035 if (nla_len(nla) != sizeof(u32))
1036 return false;
1037 val = nla_get_u32(nla);
1038 }
1039
1040 fi_val = fi->fib_metrics->metrics[type - 1];
1041 if (type == RTAX_FEATURES)
1042 fi_val &= ~DST_FEATURE_ECN_CA;
1043
1044 if (fi_val != val)
1045 return false;
1046 }
1047
1048 return true;
1049}
1050
1051static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
1052 u32 table, struct netlink_ext_ack *extack)
1053{
1054 struct fib6_config cfg = {
1055 .fc_table = table,
1056 .fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1057 .fc_ifindex = nh->fib_nh_oif,
1058 .fc_gateway = nh->fib_nh_gw6,
1059 };
1060 struct fib6_nh fib6_nh = {};
1061 int err;
1062
1063 err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1064 if (!err) {
1065 nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1066 netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1067 GFP_KERNEL);
1068 nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1069 nh->fib_nh_scope = RT_SCOPE_LINK;
1070
1071 ipv6_stub->fib6_nh_release(&fib6_nh);
1072 }
1073
1074 return err;
1075}
1076
1077/*
1078 * Picture
1079 * -------
1080 *
1081 * Semantics of nexthop is very messy by historical reasons.
1082 * We have to take into account, that:
1083 * a) gateway can be actually local interface address,
1084 * so that gatewayed route is direct.
1085 * b) gateway must be on-link address, possibly
1086 * described not by an ifaddr, but also by a direct route.
1087 * c) If both gateway and interface are specified, they should not
1088 * contradict.
1089 * d) If we use tunnel routes, gateway could be not on-link.
1090 *
1091 * Attempt to reconcile all of these (alas, self-contradictory) conditions
1092 * results in pretty ugly and hairy code with obscure logic.
1093 *
1094 * I chose to generalized it instead, so that the size
1095 * of code does not increase practically, but it becomes
1096 * much more general.
1097 * Every prefix is assigned a "scope" value: "host" is local address,
1098 * "link" is direct route,
1099 * [ ... "site" ... "interior" ... ]
1100 * and "universe" is true gateway route with global meaning.
1101 *
1102 * Every prefix refers to a set of "nexthop"s (gw, oif),
1103 * where gw must have narrower scope. This recursion stops
1104 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1105 * which means that gw is forced to be on link.
1106 *
1107 * Code is still hairy, but now it is apparently logically
1108 * consistent and very flexible. F.e. as by-product it allows
1109 * to co-exists in peace independent exterior and interior
1110 * routing processes.
1111 *
1112 * Normally it looks as following.
1113 *
1114 * {universe prefix} -> (gw, oif) [scope link]
1115 * |
1116 * |-> {link prefix} -> (gw, oif) [scope local]
1117 * |
1118 * |-> {local prefix} (terminal node)
1119 */
1120static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1121 u8 scope, struct netlink_ext_ack *extack)
1122{
1123 struct net_device *dev;
1124 struct fib_result res;
1125 int err = 0;
1126
1127 if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1128 unsigned int addr_type;
1129
1130 if (scope >= RT_SCOPE_LINK) {
1131 NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1132 return -EINVAL;
1133 }
1134 dev = __dev_get_by_index(net, nh->fib_nh_oif);
1135 if (!dev) {
1136 NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1137 return -ENODEV;
1138 }
1139 if (!(dev->flags & IFF_UP)) {
1140 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1141 return -ENETDOWN;
1142 }
1143 addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1144 if (addr_type != RTN_UNICAST) {
1145 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1146 return -EINVAL;
1147 }
1148 if (!netif_carrier_ok(dev))
1149 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1150 nh->fib_nh_dev = dev;
1151 netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1152 nh->fib_nh_scope = RT_SCOPE_LINK;
1153 return 0;
1154 }
1155 rcu_read_lock();
1156 {
1157 struct fib_table *tbl = NULL;
1158 struct flowi4 fl4 = {
1159 .daddr = nh->fib_nh_gw4,
1160 .flowi4_scope = scope + 1,
1161 .flowi4_oif = nh->fib_nh_oif,
1162 .flowi4_iif = LOOPBACK_IFINDEX,
1163 };
1164
1165 /* It is not necessary, but requires a bit of thinking */
1166 if (fl4.flowi4_scope < RT_SCOPE_LINK)
1167 fl4.flowi4_scope = RT_SCOPE_LINK;
1168
1169 if (table && table != RT_TABLE_MAIN)
1170 tbl = fib_get_table(net, table);
1171
1172 if (tbl)
1173 err = fib_table_lookup(tbl, &fl4, &res,
1174 FIB_LOOKUP_IGNORE_LINKSTATE |
1175 FIB_LOOKUP_NOREF);
1176
1177 /* on error or if no table given do full lookup. This
1178 * is needed for example when nexthops are in the local
1179 * table rather than the given table
1180 */
1181 if (!tbl || err) {
1182 err = fib_lookup(net, &fl4, &res,
1183 FIB_LOOKUP_IGNORE_LINKSTATE);
1184 }
1185
1186 if (err) {
1187 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1188 goto out;
1189 }
1190 }
1191
1192 err = -EINVAL;
1193 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1194 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1195 goto out;
1196 }
1197 nh->fib_nh_scope = res.scope;
1198 nh->fib_nh_oif = FIB_RES_OIF(res);
1199 nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1200 if (!dev) {
1201 NL_SET_ERR_MSG(extack,
1202 "No egress device for nexthop gateway");
1203 goto out;
1204 }
1205 netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1206 if (!netif_carrier_ok(dev))
1207 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1208 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1209out:
1210 rcu_read_unlock();
1211 return err;
1212}
1213
1214static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1215 struct netlink_ext_ack *extack)
1216{
1217 struct in_device *in_dev;
1218 int err;
1219
1220 if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1221 NL_SET_ERR_MSG(extack,
1222 "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1223 return -EINVAL;
1224 }
1225
1226 rcu_read_lock();
1227
1228 err = -ENODEV;
1229 in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1230 if (!in_dev)
1231 goto out;
1232 err = -ENETDOWN;
1233 if (!(in_dev->dev->flags & IFF_UP)) {
1234 NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1235 goto out;
1236 }
1237
1238 nh->fib_nh_dev = in_dev->dev;
1239 netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1240 nh->fib_nh_scope = RT_SCOPE_HOST;
1241 if (!netif_carrier_ok(nh->fib_nh_dev))
1242 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1243 err = 0;
1244out:
1245 rcu_read_unlock();
1246 return err;
1247}
1248
1249int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1250 struct netlink_ext_ack *extack)
1251{
1252 int err;
1253
1254 if (nh->fib_nh_gw_family == AF_INET)
1255 err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1256 else if (nh->fib_nh_gw_family == AF_INET6)
1257 err = fib_check_nh_v6_gw(net, nh, table, extack);
1258 else
1259 err = fib_check_nh_nongw(net, nh, extack);
1260
1261 return err;
1262}
1263
1264static struct hlist_head *
1265fib_info_laddrhash_bucket(const struct net *net, __be32 val)
1266{
1267 u32 slot = hash_32(net_hash_mix(net) ^ (__force u32)val,
1268 fib_info_hash_bits);
1269
1270 return &fib_info_laddrhash[slot];
1271}
1272
1273static void fib_info_hash_move(struct hlist_head *new_info_hash,
1274 struct hlist_head *new_laddrhash,
1275 unsigned int new_size)
1276{
1277 struct hlist_head *old_info_hash, *old_laddrhash;
1278 unsigned int old_size = fib_info_hash_size;
1279 unsigned int i;
1280
1281 spin_lock_bh(&fib_info_lock);
1282 old_info_hash = fib_info_hash;
1283 old_laddrhash = fib_info_laddrhash;
1284 fib_info_hash_size = new_size;
1285 fib_info_hash_bits = ilog2(new_size);
1286
1287 for (i = 0; i < old_size; i++) {
1288 struct hlist_head *head = &fib_info_hash[i];
1289 struct hlist_node *n;
1290 struct fib_info *fi;
1291
1292 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1293 struct hlist_head *dest;
1294 unsigned int new_hash;
1295
1296 new_hash = fib_info_hashfn(fi);
1297 dest = &new_info_hash[new_hash];
1298 hlist_add_head(&fi->fib_hash, dest);
1299 }
1300 }
1301 fib_info_hash = new_info_hash;
1302
1303 fib_info_laddrhash = new_laddrhash;
1304 for (i = 0; i < old_size; i++) {
1305 struct hlist_head *lhead = &old_laddrhash[i];
1306 struct hlist_node *n;
1307 struct fib_info *fi;
1308
1309 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1310 struct hlist_head *ldest;
1311
1312 ldest = fib_info_laddrhash_bucket(fi->fib_net,
1313 fi->fib_prefsrc);
1314 hlist_add_head(&fi->fib_lhash, ldest);
1315 }
1316 }
1317
1318 spin_unlock_bh(&fib_info_lock);
1319
1320 kvfree(old_info_hash);
1321 kvfree(old_laddrhash);
1322}
1323
1324__be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1325 unsigned char scope)
1326{
1327 struct fib_nh *nh;
1328 __be32 saddr;
1329
1330 if (nhc->nhc_family != AF_INET)
1331 return inet_select_addr(nhc->nhc_dev, 0, scope);
1332
1333 nh = container_of(nhc, struct fib_nh, nh_common);
1334 saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1335
1336 WRITE_ONCE(nh->nh_saddr, saddr);
1337 WRITE_ONCE(nh->nh_saddr_genid, atomic_read(&net->ipv4.dev_addr_genid));
1338
1339 return saddr;
1340}
1341
1342__be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1343{
1344 struct fib_nh_common *nhc = res->nhc;
1345
1346 if (res->fi->fib_prefsrc)
1347 return res->fi->fib_prefsrc;
1348
1349 if (nhc->nhc_family == AF_INET) {
1350 struct fib_nh *nh;
1351
1352 nh = container_of(nhc, struct fib_nh, nh_common);
1353 if (READ_ONCE(nh->nh_saddr_genid) ==
1354 atomic_read(&net->ipv4.dev_addr_genid))
1355 return READ_ONCE(nh->nh_saddr);
1356 }
1357
1358 return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1359}
1360
1361static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1362{
1363 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1364 fib_prefsrc != cfg->fc_dst) {
1365 u32 tb_id = cfg->fc_table;
1366 int rc;
1367
1368 if (tb_id == RT_TABLE_MAIN)
1369 tb_id = RT_TABLE_LOCAL;
1370
1371 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1372 fib_prefsrc, tb_id);
1373
1374 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1375 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1376 fib_prefsrc, RT_TABLE_LOCAL);
1377 }
1378
1379 if (rc != RTN_LOCAL)
1380 return false;
1381 }
1382 return true;
1383}
1384
1385struct fib_info *fib_create_info(struct fib_config *cfg,
1386 struct netlink_ext_ack *extack)
1387{
1388 int err;
1389 struct fib_info *fi = NULL;
1390 struct nexthop *nh = NULL;
1391 struct fib_info *ofi;
1392 int nhs = 1;
1393 struct net *net = cfg->fc_nlinfo.nl_net;
1394
1395 if (cfg->fc_type > RTN_MAX)
1396 goto err_inval;
1397
1398 /* Fast check to catch the most weird cases */
1399 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1400 NL_SET_ERR_MSG(extack, "Invalid scope");
1401 goto err_inval;
1402 }
1403
1404 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1405 NL_SET_ERR_MSG(extack,
1406 "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1407 goto err_inval;
1408 }
1409
1410 if (cfg->fc_nh_id) {
1411 if (!cfg->fc_mx) {
1412 fi = fib_find_info_nh(net, cfg);
1413 if (fi) {
1414 refcount_inc(&fi->fib_treeref);
1415 return fi;
1416 }
1417 }
1418
1419 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1420 if (!nh) {
1421 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1422 goto err_inval;
1423 }
1424 nhs = 0;
1425 }
1426
1427#ifdef CONFIG_IP_ROUTE_MULTIPATH
1428 if (cfg->fc_mp) {
1429 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1430 if (nhs == 0)
1431 goto err_inval;
1432 }
1433#endif
1434
1435 err = -ENOBUFS;
1436
1437 /* Paired with WRITE_ONCE() in fib_release_info() */
1438 if (READ_ONCE(fib_info_cnt) >= fib_info_hash_size) {
1439 unsigned int new_size = fib_info_hash_size << 1;
1440 struct hlist_head *new_info_hash;
1441 struct hlist_head *new_laddrhash;
1442 size_t bytes;
1443
1444 if (!new_size)
1445 new_size = 16;
1446 bytes = (size_t)new_size * sizeof(struct hlist_head *);
1447 new_info_hash = kvzalloc(bytes, GFP_KERNEL);
1448 new_laddrhash = kvzalloc(bytes, GFP_KERNEL);
1449 if (!new_info_hash || !new_laddrhash) {
1450 kvfree(new_info_hash);
1451 kvfree(new_laddrhash);
1452 } else {
1453 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1454 }
1455 if (!fib_info_hash_size)
1456 goto failure;
1457 }
1458
1459 fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1460 if (!fi)
1461 goto failure;
1462 fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1463 cfg->fc_mx_len, extack);
1464 if (IS_ERR(fi->fib_metrics)) {
1465 err = PTR_ERR(fi->fib_metrics);
1466 kfree(fi);
1467 return ERR_PTR(err);
1468 }
1469
1470 fi->fib_net = net;
1471 fi->fib_protocol = cfg->fc_protocol;
1472 fi->fib_scope = cfg->fc_scope;
1473 fi->fib_flags = cfg->fc_flags;
1474 fi->fib_priority = cfg->fc_priority;
1475 fi->fib_prefsrc = cfg->fc_prefsrc;
1476 fi->fib_type = cfg->fc_type;
1477 fi->fib_tb_id = cfg->fc_table;
1478
1479 fi->fib_nhs = nhs;
1480 if (nh) {
1481 if (!nexthop_get(nh)) {
1482 NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1483 err = -EINVAL;
1484 } else {
1485 err = 0;
1486 fi->nh = nh;
1487 }
1488 } else {
1489 change_nexthops(fi) {
1490 nexthop_nh->nh_parent = fi;
1491 } endfor_nexthops(fi)
1492
1493 if (cfg->fc_mp)
1494 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1495 extack);
1496 else
1497 err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1498 }
1499
1500 if (err != 0)
1501 goto failure;
1502
1503 if (fib_props[cfg->fc_type].error) {
1504 if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1505 NL_SET_ERR_MSG(extack,
1506 "Gateway, device and multipath can not be specified for this route type");
1507 goto err_inval;
1508 }
1509 goto link_it;
1510 } else {
1511 switch (cfg->fc_type) {
1512 case RTN_UNICAST:
1513 case RTN_LOCAL:
1514 case RTN_BROADCAST:
1515 case RTN_ANYCAST:
1516 case RTN_MULTICAST:
1517 break;
1518 default:
1519 NL_SET_ERR_MSG(extack, "Invalid route type");
1520 goto err_inval;
1521 }
1522 }
1523
1524 if (cfg->fc_scope > RT_SCOPE_HOST) {
1525 NL_SET_ERR_MSG(extack, "Invalid scope");
1526 goto err_inval;
1527 }
1528
1529 if (fi->nh) {
1530 err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1531 if (err)
1532 goto failure;
1533 } else if (cfg->fc_scope == RT_SCOPE_HOST) {
1534 struct fib_nh *nh = fi->fib_nh;
1535
1536 /* Local address is added. */
1537 if (nhs != 1) {
1538 NL_SET_ERR_MSG(extack,
1539 "Route with host scope can not have multiple nexthops");
1540 goto err_inval;
1541 }
1542 if (nh->fib_nh_gw_family) {
1543 NL_SET_ERR_MSG(extack,
1544 "Route with host scope can not have a gateway");
1545 goto err_inval;
1546 }
1547 nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1548 nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1549 err = -ENODEV;
1550 if (!nh->fib_nh_dev)
1551 goto failure;
1552 netdev_tracker_alloc(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1553 GFP_KERNEL);
1554 } else {
1555 int linkdown = 0;
1556
1557 change_nexthops(fi) {
1558 err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1559 cfg->fc_table, cfg->fc_scope,
1560 extack);
1561 if (err != 0)
1562 goto failure;
1563 if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1564 linkdown++;
1565 } endfor_nexthops(fi)
1566 if (linkdown == fi->fib_nhs)
1567 fi->fib_flags |= RTNH_F_LINKDOWN;
1568 }
1569
1570 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1571 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1572 goto err_inval;
1573 }
1574
1575 if (!fi->nh) {
1576 change_nexthops(fi) {
1577 fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1578 fi->fib_scope);
1579 if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1580 fi->fib_nh_is_v6 = true;
1581 } endfor_nexthops(fi)
1582
1583 fib_rebalance(fi);
1584 }
1585
1586link_it:
1587 ofi = fib_find_info(fi);
1588 if (ofi) {
1589 /* fib_table_lookup() should not see @fi yet. */
1590 fi->fib_dead = 1;
1591 free_fib_info(fi);
1592 refcount_inc(&ofi->fib_treeref);
1593 return ofi;
1594 }
1595
1596 refcount_set(&fi->fib_treeref, 1);
1597 refcount_set(&fi->fib_clntref, 1);
1598 spin_lock_bh(&fib_info_lock);
1599 fib_info_cnt++;
1600 hlist_add_head(&fi->fib_hash,
1601 &fib_info_hash[fib_info_hashfn(fi)]);
1602 if (fi->fib_prefsrc) {
1603 struct hlist_head *head;
1604
1605 head = fib_info_laddrhash_bucket(net, fi->fib_prefsrc);
1606 hlist_add_head(&fi->fib_lhash, head);
1607 }
1608 if (fi->nh) {
1609 list_add(&fi->nh_list, &nh->fi_list);
1610 } else {
1611 change_nexthops(fi) {
1612 struct hlist_head *head;
1613
1614 if (!nexthop_nh->fib_nh_dev)
1615 continue;
1616 head = fib_info_devhash_bucket(nexthop_nh->fib_nh_dev);
1617 hlist_add_head(&nexthop_nh->nh_hash, head);
1618 } endfor_nexthops(fi)
1619 }
1620 spin_unlock_bh(&fib_info_lock);
1621 return fi;
1622
1623err_inval:
1624 err = -EINVAL;
1625
1626failure:
1627 if (fi) {
1628 /* fib_table_lookup() should not see @fi yet. */
1629 fi->fib_dead = 1;
1630 free_fib_info(fi);
1631 }
1632
1633 return ERR_PTR(err);
1634}
1635
1636int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1637 u8 rt_family, unsigned char *flags, bool skip_oif)
1638{
1639 if (nhc->nhc_flags & RTNH_F_DEAD)
1640 *flags |= RTNH_F_DEAD;
1641
1642 if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1643 *flags |= RTNH_F_LINKDOWN;
1644
1645 rcu_read_lock();
1646 switch (nhc->nhc_family) {
1647 case AF_INET:
1648 if (ip_ignore_linkdown(nhc->nhc_dev))
1649 *flags |= RTNH_F_DEAD;
1650 break;
1651 case AF_INET6:
1652 if (ip6_ignore_linkdown(nhc->nhc_dev))
1653 *flags |= RTNH_F_DEAD;
1654 break;
1655 }
1656 rcu_read_unlock();
1657 }
1658
1659 switch (nhc->nhc_gw_family) {
1660 case AF_INET:
1661 if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1662 goto nla_put_failure;
1663 break;
1664 case AF_INET6:
1665 /* if gateway family does not match nexthop family
1666 * gateway is encoded as RTA_VIA
1667 */
1668 if (rt_family != nhc->nhc_gw_family) {
1669 int alen = sizeof(struct in6_addr);
1670 struct nlattr *nla;
1671 struct rtvia *via;
1672
1673 nla = nla_reserve(skb, RTA_VIA, alen + 2);
1674 if (!nla)
1675 goto nla_put_failure;
1676
1677 via = nla_data(nla);
1678 via->rtvia_family = AF_INET6;
1679 memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1680 } else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1681 &nhc->nhc_gw.ipv6) < 0) {
1682 goto nla_put_failure;
1683 }
1684 break;
1685 }
1686
1687 *flags |= (nhc->nhc_flags &
1688 (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP));
1689
1690 if (!skip_oif && nhc->nhc_dev &&
1691 nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1692 goto nla_put_failure;
1693
1694 if (nhc->nhc_lwtstate &&
1695 lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1696 RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1697 goto nla_put_failure;
1698
1699 return 0;
1700
1701nla_put_failure:
1702 return -EMSGSIZE;
1703}
1704EXPORT_SYMBOL_GPL(fib_nexthop_info);
1705
1706#if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1707int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1708 int nh_weight, u8 rt_family, u32 nh_tclassid)
1709{
1710 const struct net_device *dev = nhc->nhc_dev;
1711 struct rtnexthop *rtnh;
1712 unsigned char flags = 0;
1713
1714 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1715 if (!rtnh)
1716 goto nla_put_failure;
1717
1718 rtnh->rtnh_hops = nh_weight - 1;
1719 rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1720
1721 if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1722 goto nla_put_failure;
1723
1724 rtnh->rtnh_flags = flags;
1725
1726 if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1727 goto nla_put_failure;
1728
1729 /* length of rtnetlink header + attributes */
1730 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1731
1732 return 0;
1733
1734nla_put_failure:
1735 return -EMSGSIZE;
1736}
1737EXPORT_SYMBOL_GPL(fib_add_nexthop);
1738#endif
1739
1740#ifdef CONFIG_IP_ROUTE_MULTIPATH
1741static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1742{
1743 struct nlattr *mp;
1744
1745 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1746 if (!mp)
1747 goto nla_put_failure;
1748
1749 if (unlikely(fi->nh)) {
1750 if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1751 goto nla_put_failure;
1752 goto mp_end;
1753 }
1754
1755 for_nexthops(fi) {
1756 u32 nh_tclassid = 0;
1757#ifdef CONFIG_IP_ROUTE_CLASSID
1758 nh_tclassid = nh->nh_tclassid;
1759#endif
1760 if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1761 AF_INET, nh_tclassid) < 0)
1762 goto nla_put_failure;
1763 } endfor_nexthops(fi);
1764
1765mp_end:
1766 nla_nest_end(skb, mp);
1767
1768 return 0;
1769
1770nla_put_failure:
1771 return -EMSGSIZE;
1772}
1773#else
1774static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1775{
1776 return 0;
1777}
1778#endif
1779
1780int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1781 const struct fib_rt_info *fri, unsigned int flags)
1782{
1783 unsigned int nhs = fib_info_num_path(fri->fi);
1784 struct fib_info *fi = fri->fi;
1785 u32 tb_id = fri->tb_id;
1786 struct nlmsghdr *nlh;
1787 struct rtmsg *rtm;
1788
1789 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1790 if (!nlh)
1791 return -EMSGSIZE;
1792
1793 rtm = nlmsg_data(nlh);
1794 rtm->rtm_family = AF_INET;
1795 rtm->rtm_dst_len = fri->dst_len;
1796 rtm->rtm_src_len = 0;
1797 rtm->rtm_tos = inet_dscp_to_dsfield(fri->dscp);
1798 if (tb_id < 256)
1799 rtm->rtm_table = tb_id;
1800 else
1801 rtm->rtm_table = RT_TABLE_COMPAT;
1802 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1803 goto nla_put_failure;
1804 rtm->rtm_type = fri->type;
1805 rtm->rtm_flags = fi->fib_flags;
1806 rtm->rtm_scope = fi->fib_scope;
1807 rtm->rtm_protocol = fi->fib_protocol;
1808
1809 if (rtm->rtm_dst_len &&
1810 nla_put_in_addr(skb, RTA_DST, fri->dst))
1811 goto nla_put_failure;
1812 if (fi->fib_priority &&
1813 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1814 goto nla_put_failure;
1815 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1816 goto nla_put_failure;
1817
1818 if (fi->fib_prefsrc &&
1819 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1820 goto nla_put_failure;
1821
1822 if (fi->nh) {
1823 if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1824 goto nla_put_failure;
1825 if (nexthop_is_blackhole(fi->nh))
1826 rtm->rtm_type = RTN_BLACKHOLE;
1827 if (!READ_ONCE(fi->fib_net->ipv4.sysctl_nexthop_compat_mode))
1828 goto offload;
1829 }
1830
1831 if (nhs == 1) {
1832 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1833 unsigned char flags = 0;
1834
1835 if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1836 goto nla_put_failure;
1837
1838 rtm->rtm_flags = flags;
1839#ifdef CONFIG_IP_ROUTE_CLASSID
1840 if (nhc->nhc_family == AF_INET) {
1841 struct fib_nh *nh;
1842
1843 nh = container_of(nhc, struct fib_nh, nh_common);
1844 if (nh->nh_tclassid &&
1845 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1846 goto nla_put_failure;
1847 }
1848#endif
1849 } else {
1850 if (fib_add_multipath(skb, fi) < 0)
1851 goto nla_put_failure;
1852 }
1853
1854offload:
1855 if (fri->offload)
1856 rtm->rtm_flags |= RTM_F_OFFLOAD;
1857 if (fri->trap)
1858 rtm->rtm_flags |= RTM_F_TRAP;
1859 if (fri->offload_failed)
1860 rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
1861
1862 nlmsg_end(skb, nlh);
1863 return 0;
1864
1865nla_put_failure:
1866 nlmsg_cancel(skb, nlh);
1867 return -EMSGSIZE;
1868}
1869
1870/*
1871 * Update FIB if:
1872 * - local address disappeared -> we must delete all the entries
1873 * referring to it.
1874 * - device went down -> we must shutdown all nexthops going via it.
1875 */
1876int fib_sync_down_addr(struct net_device *dev, __be32 local)
1877{
1878 int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1879 struct net *net = dev_net(dev);
1880 struct hlist_head *head;
1881 struct fib_info *fi;
1882 int ret = 0;
1883
1884 if (!fib_info_laddrhash || local == 0)
1885 return 0;
1886
1887 head = fib_info_laddrhash_bucket(net, local);
1888 hlist_for_each_entry(fi, head, fib_lhash) {
1889 if (!net_eq(fi->fib_net, net) ||
1890 fi->fib_tb_id != tb_id)
1891 continue;
1892 if (fi->fib_prefsrc == local) {
1893 fi->fib_flags |= RTNH_F_DEAD;
1894 fi->pfsrc_removed = true;
1895 ret++;
1896 }
1897 }
1898 return ret;
1899}
1900
1901static int call_fib_nh_notifiers(struct fib_nh *nh,
1902 enum fib_event_type event_type)
1903{
1904 bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1905 struct fib_nh_notifier_info info = {
1906 .fib_nh = nh,
1907 };
1908
1909 switch (event_type) {
1910 case FIB_EVENT_NH_ADD:
1911 if (nh->fib_nh_flags & RTNH_F_DEAD)
1912 break;
1913 if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1914 break;
1915 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1916 &info.info);
1917 case FIB_EVENT_NH_DEL:
1918 if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1919 (nh->fib_nh_flags & RTNH_F_DEAD))
1920 return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1921 event_type, &info.info);
1922 break;
1923 default:
1924 break;
1925 }
1926
1927 return NOTIFY_DONE;
1928}
1929
1930/* Update the PMTU of exceptions when:
1931 * - the new MTU of the first hop becomes smaller than the PMTU
1932 * - the old MTU was the same as the PMTU, and it limited discovery of
1933 * larger MTUs on the path. With that limit raised, we can now
1934 * discover larger MTUs
1935 * A special case is locked exceptions, for which the PMTU is smaller
1936 * than the minimal accepted PMTU:
1937 * - if the new MTU is greater than the PMTU, don't make any change
1938 * - otherwise, unlock and set PMTU
1939 */
1940void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1941{
1942 struct fnhe_hash_bucket *bucket;
1943 int i;
1944
1945 bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1946 if (!bucket)
1947 return;
1948
1949 for (i = 0; i < FNHE_HASH_SIZE; i++) {
1950 struct fib_nh_exception *fnhe;
1951
1952 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1953 fnhe;
1954 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1955 if (fnhe->fnhe_mtu_locked) {
1956 if (new <= fnhe->fnhe_pmtu) {
1957 fnhe->fnhe_pmtu = new;
1958 fnhe->fnhe_mtu_locked = false;
1959 }
1960 } else if (new < fnhe->fnhe_pmtu ||
1961 orig == fnhe->fnhe_pmtu) {
1962 fnhe->fnhe_pmtu = new;
1963 }
1964 }
1965 }
1966}
1967
1968void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1969{
1970 struct hlist_head *head = fib_info_devhash_bucket(dev);
1971 struct fib_nh *nh;
1972
1973 hlist_for_each_entry(nh, head, nh_hash) {
1974 if (nh->fib_nh_dev == dev)
1975 fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1976 }
1977}
1978
1979/* Event force Flags Description
1980 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host
1981 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host
1982 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed
1983 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed
1984 *
1985 * only used when fib_nh is built into fib_info
1986 */
1987int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1988{
1989 struct hlist_head *head = fib_info_devhash_bucket(dev);
1990 struct fib_info *prev_fi = NULL;
1991 int scope = RT_SCOPE_NOWHERE;
1992 struct fib_nh *nh;
1993 int ret = 0;
1994
1995 if (force)
1996 scope = -1;
1997
1998 hlist_for_each_entry(nh, head, nh_hash) {
1999 struct fib_info *fi = nh->nh_parent;
2000 int dead;
2001
2002 BUG_ON(!fi->fib_nhs);
2003 if (nh->fib_nh_dev != dev || fi == prev_fi)
2004 continue;
2005 prev_fi = fi;
2006 dead = 0;
2007 change_nexthops(fi) {
2008 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
2009 dead++;
2010 else if (nexthop_nh->fib_nh_dev == dev &&
2011 nexthop_nh->fib_nh_scope != scope) {
2012 switch (event) {
2013 case NETDEV_DOWN:
2014 case NETDEV_UNREGISTER:
2015 nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
2016 fallthrough;
2017 case NETDEV_CHANGE:
2018 nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
2019 break;
2020 }
2021 call_fib_nh_notifiers(nexthop_nh,
2022 FIB_EVENT_NH_DEL);
2023 dead++;
2024 }
2025#ifdef CONFIG_IP_ROUTE_MULTIPATH
2026 if (event == NETDEV_UNREGISTER &&
2027 nexthop_nh->fib_nh_dev == dev) {
2028 dead = fi->fib_nhs;
2029 break;
2030 }
2031#endif
2032 } endfor_nexthops(fi)
2033 if (dead == fi->fib_nhs) {
2034 switch (event) {
2035 case NETDEV_DOWN:
2036 case NETDEV_UNREGISTER:
2037 fi->fib_flags |= RTNH_F_DEAD;
2038 fallthrough;
2039 case NETDEV_CHANGE:
2040 fi->fib_flags |= RTNH_F_LINKDOWN;
2041 break;
2042 }
2043 ret++;
2044 }
2045
2046 fib_rebalance(fi);
2047 }
2048
2049 return ret;
2050}
2051
2052/* Must be invoked inside of an RCU protected region. */
2053static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2054{
2055 struct fib_info *fi = NULL, *last_resort = NULL;
2056 struct hlist_head *fa_head = res->fa_head;
2057 struct fib_table *tb = res->table;
2058 u8 slen = 32 - res->prefixlen;
2059 int order = -1, last_idx = -1;
2060 struct fib_alias *fa, *fa1 = NULL;
2061 u32 last_prio = res->fi->fib_priority;
2062 dscp_t last_dscp = 0;
2063
2064 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2065 struct fib_info *next_fi = fa->fa_info;
2066 struct fib_nh_common *nhc;
2067
2068 if (fa->fa_slen != slen)
2069 continue;
2070 if (fa->fa_dscp &&
2071 fa->fa_dscp != inet_dsfield_to_dscp(flp->flowi4_tos))
2072 continue;
2073 if (fa->tb_id != tb->tb_id)
2074 continue;
2075 if (next_fi->fib_priority > last_prio &&
2076 fa->fa_dscp == last_dscp) {
2077 if (last_dscp)
2078 continue;
2079 break;
2080 }
2081 if (next_fi->fib_flags & RTNH_F_DEAD)
2082 continue;
2083 last_dscp = fa->fa_dscp;
2084 last_prio = next_fi->fib_priority;
2085
2086 if (next_fi->fib_scope != res->scope ||
2087 fa->fa_type != RTN_UNICAST)
2088 continue;
2089
2090 nhc = fib_info_nhc(next_fi, 0);
2091 if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2092 continue;
2093
2094 fib_alias_accessed(fa);
2095
2096 if (!fi) {
2097 if (next_fi != res->fi)
2098 break;
2099 fa1 = fa;
2100 } else if (!fib_detect_death(fi, order, &last_resort,
2101 &last_idx, fa1->fa_default)) {
2102 fib_result_assign(res, fi);
2103 fa1->fa_default = order;
2104 goto out;
2105 }
2106 fi = next_fi;
2107 order++;
2108 }
2109
2110 if (order <= 0 || !fi) {
2111 if (fa1)
2112 fa1->fa_default = -1;
2113 goto out;
2114 }
2115
2116 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2117 fa1->fa_default)) {
2118 fib_result_assign(res, fi);
2119 fa1->fa_default = order;
2120 goto out;
2121 }
2122
2123 if (last_idx >= 0)
2124 fib_result_assign(res, last_resort);
2125 fa1->fa_default = last_idx;
2126out:
2127 return;
2128}
2129
2130/*
2131 * Dead device goes up. We wake up dead nexthops.
2132 * It takes sense only on multipath routes.
2133 *
2134 * only used when fib_nh is built into fib_info
2135 */
2136int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2137{
2138 struct fib_info *prev_fi;
2139 struct hlist_head *head;
2140 struct fib_nh *nh;
2141 int ret;
2142
2143 if (!(dev->flags & IFF_UP))
2144 return 0;
2145
2146 if (nh_flags & RTNH_F_DEAD) {
2147 unsigned int flags = dev_get_flags(dev);
2148
2149 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2150 nh_flags |= RTNH_F_LINKDOWN;
2151 }
2152
2153 prev_fi = NULL;
2154 head = fib_info_devhash_bucket(dev);
2155 ret = 0;
2156
2157 hlist_for_each_entry(nh, head, nh_hash) {
2158 struct fib_info *fi = nh->nh_parent;
2159 int alive;
2160
2161 BUG_ON(!fi->fib_nhs);
2162 if (nh->fib_nh_dev != dev || fi == prev_fi)
2163 continue;
2164
2165 prev_fi = fi;
2166 alive = 0;
2167 change_nexthops(fi) {
2168 if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2169 alive++;
2170 continue;
2171 }
2172 if (!nexthop_nh->fib_nh_dev ||
2173 !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2174 continue;
2175 if (nexthop_nh->fib_nh_dev != dev ||
2176 !__in_dev_get_rtnl(dev))
2177 continue;
2178 alive++;
2179 nexthop_nh->fib_nh_flags &= ~nh_flags;
2180 call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2181 } endfor_nexthops(fi)
2182
2183 if (alive > 0) {
2184 fi->fib_flags &= ~nh_flags;
2185 ret++;
2186 }
2187
2188 fib_rebalance(fi);
2189 }
2190
2191 return ret;
2192}
2193
2194#ifdef CONFIG_IP_ROUTE_MULTIPATH
2195static bool fib_good_nh(const struct fib_nh *nh)
2196{
2197 int state = NUD_REACHABLE;
2198
2199 if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2200 struct neighbour *n;
2201
2202 rcu_read_lock();
2203
2204 if (likely(nh->fib_nh_gw_family == AF_INET))
2205 n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2206 (__force u32)nh->fib_nh_gw4);
2207 else if (nh->fib_nh_gw_family == AF_INET6)
2208 n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2209 &nh->fib_nh_gw6);
2210 else
2211 n = NULL;
2212 if (n)
2213 state = READ_ONCE(n->nud_state);
2214
2215 rcu_read_unlock();
2216 }
2217
2218 return !!(state & NUD_VALID);
2219}
2220
2221void fib_select_multipath(struct fib_result *res, int hash)
2222{
2223 struct fib_info *fi = res->fi;
2224 struct net *net = fi->fib_net;
2225 bool first = false;
2226
2227 if (unlikely(res->fi->nh)) {
2228 nexthop_path_fib_result(res, hash);
2229 return;
2230 }
2231
2232 change_nexthops(fi) {
2233 if (READ_ONCE(net->ipv4.sysctl_fib_multipath_use_neigh)) {
2234 if (!fib_good_nh(nexthop_nh))
2235 continue;
2236 if (!first) {
2237 res->nh_sel = nhsel;
2238 res->nhc = &nexthop_nh->nh_common;
2239 first = true;
2240 }
2241 }
2242
2243 if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2244 continue;
2245
2246 res->nh_sel = nhsel;
2247 res->nhc = &nexthop_nh->nh_common;
2248 return;
2249 } endfor_nexthops(fi);
2250}
2251#endif
2252
2253void fib_select_path(struct net *net, struct fib_result *res,
2254 struct flowi4 *fl4, const struct sk_buff *skb)
2255{
2256 if (fl4->flowi4_oif)
2257 goto check_saddr;
2258
2259#ifdef CONFIG_IP_ROUTE_MULTIPATH
2260 if (fib_info_num_path(res->fi) > 1) {
2261 int h = fib_multipath_hash(net, fl4, skb, NULL);
2262
2263 fib_select_multipath(res, h);
2264 }
2265 else
2266#endif
2267 if (!res->prefixlen &&
2268 res->table->tb_num_default > 1 &&
2269 res->type == RTN_UNICAST)
2270 fib_select_default(fl4, res);
2271
2272check_saddr:
2273 if (!fl4->saddr)
2274 fl4->saddr = fib_result_prefsrc(net, res);
2275}