Loading...
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * IPv4 Forwarding Information Base: semantics.
7 *
8 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
14 */
15
16#include <linux/uaccess.h>
17#include <linux/bitops.h>
18#include <linux/types.h>
19#include <linux/kernel.h>
20#include <linux/jiffies.h>
21#include <linux/mm.h>
22#include <linux/string.h>
23#include <linux/socket.h>
24#include <linux/sockios.h>
25#include <linux/errno.h>
26#include <linux/in.h>
27#include <linux/inet.h>
28#include <linux/inetdevice.h>
29#include <linux/netdevice.h>
30#include <linux/if_arp.h>
31#include <linux/proc_fs.h>
32#include <linux/skbuff.h>
33#include <linux/init.h>
34#include <linux/slab.h>
35#include <linux/netlink.h>
36
37#include <net/arp.h>
38#include <net/ip.h>
39#include <net/protocol.h>
40#include <net/route.h>
41#include <net/tcp.h>
42#include <net/sock.h>
43#include <net/ip_fib.h>
44#include <net/netlink.h>
45#include <net/nexthop.h>
46#include <net/lwtunnel.h>
47#include <net/fib_notifier.h>
48
49#include "fib_lookup.h"
50
51static DEFINE_SPINLOCK(fib_info_lock);
52static struct hlist_head *fib_info_hash;
53static struct hlist_head *fib_info_laddrhash;
54static unsigned int fib_info_hash_size;
55static unsigned int fib_info_cnt;
56
57#define DEVINDEX_HASHBITS 8
58#define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
59static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
60
61#ifdef CONFIG_IP_ROUTE_MULTIPATH
62
63#define for_nexthops(fi) { \
64 int nhsel; const struct fib_nh *nh; \
65 for (nhsel = 0, nh = (fi)->fib_nh; \
66 nhsel < (fi)->fib_nhs; \
67 nh++, nhsel++)
68
69#define change_nexthops(fi) { \
70 int nhsel; struct fib_nh *nexthop_nh; \
71 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
72 nhsel < (fi)->fib_nhs; \
73 nexthop_nh++, nhsel++)
74
75#else /* CONFIG_IP_ROUTE_MULTIPATH */
76
77/* Hope, that gcc will optimize it to get rid of dummy loop */
78
79#define for_nexthops(fi) { \
80 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \
81 for (nhsel = 0; nhsel < 1; nhsel++)
82
83#define change_nexthops(fi) { \
84 int nhsel; \
85 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
86 for (nhsel = 0; nhsel < 1; nhsel++)
87
88#endif /* CONFIG_IP_ROUTE_MULTIPATH */
89
90#define endfor_nexthops(fi) }
91
92
93const struct fib_prop fib_props[RTN_MAX + 1] = {
94 [RTN_UNSPEC] = {
95 .error = 0,
96 .scope = RT_SCOPE_NOWHERE,
97 },
98 [RTN_UNICAST] = {
99 .error = 0,
100 .scope = RT_SCOPE_UNIVERSE,
101 },
102 [RTN_LOCAL] = {
103 .error = 0,
104 .scope = RT_SCOPE_HOST,
105 },
106 [RTN_BROADCAST] = {
107 .error = 0,
108 .scope = RT_SCOPE_LINK,
109 },
110 [RTN_ANYCAST] = {
111 .error = 0,
112 .scope = RT_SCOPE_LINK,
113 },
114 [RTN_MULTICAST] = {
115 .error = 0,
116 .scope = RT_SCOPE_UNIVERSE,
117 },
118 [RTN_BLACKHOLE] = {
119 .error = -EINVAL,
120 .scope = RT_SCOPE_UNIVERSE,
121 },
122 [RTN_UNREACHABLE] = {
123 .error = -EHOSTUNREACH,
124 .scope = RT_SCOPE_UNIVERSE,
125 },
126 [RTN_PROHIBIT] = {
127 .error = -EACCES,
128 .scope = RT_SCOPE_UNIVERSE,
129 },
130 [RTN_THROW] = {
131 .error = -EAGAIN,
132 .scope = RT_SCOPE_UNIVERSE,
133 },
134 [RTN_NAT] = {
135 .error = -EINVAL,
136 .scope = RT_SCOPE_NOWHERE,
137 },
138 [RTN_XRESOLVE] = {
139 .error = -EINVAL,
140 .scope = RT_SCOPE_NOWHERE,
141 },
142};
143
144static void rt_fibinfo_free(struct rtable __rcu **rtp)
145{
146 struct rtable *rt = rcu_dereference_protected(*rtp, 1);
147
148 if (!rt)
149 return;
150
151 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
152 * because we waited an RCU grace period before calling
153 * free_fib_info_rcu()
154 */
155
156 dst_dev_put(&rt->dst);
157 dst_release_immediate(&rt->dst);
158}
159
160static void free_nh_exceptions(struct fib_nh *nh)
161{
162 struct fnhe_hash_bucket *hash;
163 int i;
164
165 hash = rcu_dereference_protected(nh->nh_exceptions, 1);
166 if (!hash)
167 return;
168 for (i = 0; i < FNHE_HASH_SIZE; i++) {
169 struct fib_nh_exception *fnhe;
170
171 fnhe = rcu_dereference_protected(hash[i].chain, 1);
172 while (fnhe) {
173 struct fib_nh_exception *next;
174
175 next = rcu_dereference_protected(fnhe->fnhe_next, 1);
176
177 rt_fibinfo_free(&fnhe->fnhe_rth_input);
178 rt_fibinfo_free(&fnhe->fnhe_rth_output);
179
180 kfree(fnhe);
181
182 fnhe = next;
183 }
184 }
185 kfree(hash);
186}
187
188static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
189{
190 int cpu;
191
192 if (!rtp)
193 return;
194
195 for_each_possible_cpu(cpu) {
196 struct rtable *rt;
197
198 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
199 if (rt) {
200 dst_dev_put(&rt->dst);
201 dst_release_immediate(&rt->dst);
202 }
203 }
204 free_percpu(rtp);
205}
206
207/* Release a nexthop info record */
208static void free_fib_info_rcu(struct rcu_head *head)
209{
210 struct fib_info *fi = container_of(head, struct fib_info, rcu);
211 struct dst_metrics *m;
212
213 change_nexthops(fi) {
214 if (nexthop_nh->nh_dev)
215 dev_put(nexthop_nh->nh_dev);
216 lwtstate_put(nexthop_nh->nh_lwtstate);
217 free_nh_exceptions(nexthop_nh);
218 rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
219 rt_fibinfo_free(&nexthop_nh->nh_rth_input);
220 } endfor_nexthops(fi);
221
222 m = fi->fib_metrics;
223 if (m != &dst_default_metrics && refcount_dec_and_test(&m->refcnt))
224 kfree(m);
225 kfree(fi);
226}
227
228void free_fib_info(struct fib_info *fi)
229{
230 if (fi->fib_dead == 0) {
231 pr_warn("Freeing alive fib_info %p\n", fi);
232 return;
233 }
234 fib_info_cnt--;
235#ifdef CONFIG_IP_ROUTE_CLASSID
236 change_nexthops(fi) {
237 if (nexthop_nh->nh_tclassid)
238 fi->fib_net->ipv4.fib_num_tclassid_users--;
239 } endfor_nexthops(fi);
240#endif
241 call_rcu(&fi->rcu, free_fib_info_rcu);
242}
243EXPORT_SYMBOL_GPL(free_fib_info);
244
245void fib_release_info(struct fib_info *fi)
246{
247 spin_lock_bh(&fib_info_lock);
248 if (fi && --fi->fib_treeref == 0) {
249 hlist_del(&fi->fib_hash);
250 if (fi->fib_prefsrc)
251 hlist_del(&fi->fib_lhash);
252 change_nexthops(fi) {
253 if (!nexthop_nh->nh_dev)
254 continue;
255 hlist_del(&nexthop_nh->nh_hash);
256 } endfor_nexthops(fi)
257 fi->fib_dead = 1;
258 fib_info_put(fi);
259 }
260 spin_unlock_bh(&fib_info_lock);
261}
262
263static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
264{
265 const struct fib_nh *onh = ofi->fib_nh;
266
267 for_nexthops(fi) {
268 if (nh->nh_oif != onh->nh_oif ||
269 nh->nh_gw != onh->nh_gw ||
270 nh->nh_scope != onh->nh_scope ||
271#ifdef CONFIG_IP_ROUTE_MULTIPATH
272 nh->nh_weight != onh->nh_weight ||
273#endif
274#ifdef CONFIG_IP_ROUTE_CLASSID
275 nh->nh_tclassid != onh->nh_tclassid ||
276#endif
277 lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) ||
278 ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
279 return -1;
280 onh++;
281 } endfor_nexthops(fi);
282 return 0;
283}
284
285static inline unsigned int fib_devindex_hashfn(unsigned int val)
286{
287 unsigned int mask = DEVINDEX_HASHSIZE - 1;
288
289 return (val ^
290 (val >> DEVINDEX_HASHBITS) ^
291 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
292}
293
294static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
295{
296 unsigned int mask = (fib_info_hash_size - 1);
297 unsigned int val = fi->fib_nhs;
298
299 val ^= (fi->fib_protocol << 8) | fi->fib_scope;
300 val ^= (__force u32)fi->fib_prefsrc;
301 val ^= fi->fib_priority;
302 for_nexthops(fi) {
303 val ^= fib_devindex_hashfn(nh->nh_oif);
304 } endfor_nexthops(fi)
305
306 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
307}
308
309static struct fib_info *fib_find_info(const struct fib_info *nfi)
310{
311 struct hlist_head *head;
312 struct fib_info *fi;
313 unsigned int hash;
314
315 hash = fib_info_hashfn(nfi);
316 head = &fib_info_hash[hash];
317
318 hlist_for_each_entry(fi, head, fib_hash) {
319 if (!net_eq(fi->fib_net, nfi->fib_net))
320 continue;
321 if (fi->fib_nhs != nfi->fib_nhs)
322 continue;
323 if (nfi->fib_protocol == fi->fib_protocol &&
324 nfi->fib_scope == fi->fib_scope &&
325 nfi->fib_prefsrc == fi->fib_prefsrc &&
326 nfi->fib_priority == fi->fib_priority &&
327 nfi->fib_type == fi->fib_type &&
328 memcmp(nfi->fib_metrics, fi->fib_metrics,
329 sizeof(u32) * RTAX_MAX) == 0 &&
330 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
331 (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
332 return fi;
333 }
334
335 return NULL;
336}
337
338/* Check, that the gateway is already configured.
339 * Used only by redirect accept routine.
340 */
341int ip_fib_check_default(__be32 gw, struct net_device *dev)
342{
343 struct hlist_head *head;
344 struct fib_nh *nh;
345 unsigned int hash;
346
347 spin_lock(&fib_info_lock);
348
349 hash = fib_devindex_hashfn(dev->ifindex);
350 head = &fib_info_devhash[hash];
351 hlist_for_each_entry(nh, head, nh_hash) {
352 if (nh->nh_dev == dev &&
353 nh->nh_gw == gw &&
354 !(nh->nh_flags & RTNH_F_DEAD)) {
355 spin_unlock(&fib_info_lock);
356 return 0;
357 }
358 }
359
360 spin_unlock(&fib_info_lock);
361
362 return -1;
363}
364
365static inline size_t fib_nlmsg_size(struct fib_info *fi)
366{
367 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
368 + nla_total_size(4) /* RTA_TABLE */
369 + nla_total_size(4) /* RTA_DST */
370 + nla_total_size(4) /* RTA_PRIORITY */
371 + nla_total_size(4) /* RTA_PREFSRC */
372 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
373
374 /* space for nested metrics */
375 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
376
377 if (fi->fib_nhs) {
378 size_t nh_encapsize = 0;
379 /* Also handles the special case fib_nhs == 1 */
380
381 /* each nexthop is packed in an attribute */
382 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
383
384 /* may contain flow and gateway attribute */
385 nhsize += 2 * nla_total_size(4);
386
387 /* grab encap info */
388 for_nexthops(fi) {
389 if (nh->nh_lwtstate) {
390 /* RTA_ENCAP_TYPE */
391 nh_encapsize += lwtunnel_get_encap_size(
392 nh->nh_lwtstate);
393 /* RTA_ENCAP */
394 nh_encapsize += nla_total_size(2);
395 }
396 } endfor_nexthops(fi);
397
398 /* all nexthops are packed in a nested attribute */
399 payload += nla_total_size((fi->fib_nhs * nhsize) +
400 nh_encapsize);
401
402 }
403
404 return payload;
405}
406
407void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
408 int dst_len, u32 tb_id, const struct nl_info *info,
409 unsigned int nlm_flags)
410{
411 struct sk_buff *skb;
412 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
413 int err = -ENOBUFS;
414
415 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
416 if (!skb)
417 goto errout;
418
419 err = fib_dump_info(skb, info->portid, seq, event, tb_id,
420 fa->fa_type, key, dst_len,
421 fa->fa_tos, fa->fa_info, nlm_flags);
422 if (err < 0) {
423 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
424 WARN_ON(err == -EMSGSIZE);
425 kfree_skb(skb);
426 goto errout;
427 }
428 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
429 info->nlh, GFP_KERNEL);
430 return;
431errout:
432 if (err < 0)
433 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
434}
435
436static int fib_detect_death(struct fib_info *fi, int order,
437 struct fib_info **last_resort, int *last_idx,
438 int dflt)
439{
440 struct neighbour *n;
441 int state = NUD_NONE;
442
443 n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
444 if (n) {
445 state = n->nud_state;
446 neigh_release(n);
447 } else {
448 return 0;
449 }
450 if (state == NUD_REACHABLE)
451 return 0;
452 if ((state & NUD_VALID) && order != dflt)
453 return 0;
454 if ((state & NUD_VALID) ||
455 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
456 *last_resort = fi;
457 *last_idx = order;
458 }
459 return 1;
460}
461
462#ifdef CONFIG_IP_ROUTE_MULTIPATH
463
464static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
465 struct netlink_ext_ack *extack)
466{
467 int nhs = 0;
468
469 while (rtnh_ok(rtnh, remaining)) {
470 nhs++;
471 rtnh = rtnh_next(rtnh, &remaining);
472 }
473
474 /* leftover implies invalid nexthop configuration, discard it */
475 if (remaining > 0) {
476 NL_SET_ERR_MSG(extack,
477 "Invalid nexthop configuration - extra data after nexthops");
478 nhs = 0;
479 }
480
481 return nhs;
482}
483
484static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
485 int remaining, struct fib_config *cfg,
486 struct netlink_ext_ack *extack)
487{
488 int ret;
489
490 change_nexthops(fi) {
491 int attrlen;
492
493 if (!rtnh_ok(rtnh, remaining)) {
494 NL_SET_ERR_MSG(extack,
495 "Invalid nexthop configuration - extra data after nexthop");
496 return -EINVAL;
497 }
498
499 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
500 NL_SET_ERR_MSG(extack,
501 "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
502 return -EINVAL;
503 }
504
505 nexthop_nh->nh_flags =
506 (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
507 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
508 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
509
510 attrlen = rtnh_attrlen(rtnh);
511 if (attrlen > 0) {
512 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
513
514 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
515 nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
516#ifdef CONFIG_IP_ROUTE_CLASSID
517 nla = nla_find(attrs, attrlen, RTA_FLOW);
518 nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
519 if (nexthop_nh->nh_tclassid)
520 fi->fib_net->ipv4.fib_num_tclassid_users++;
521#endif
522 nla = nla_find(attrs, attrlen, RTA_ENCAP);
523 if (nla) {
524 struct lwtunnel_state *lwtstate;
525 struct nlattr *nla_entype;
526
527 nla_entype = nla_find(attrs, attrlen,
528 RTA_ENCAP_TYPE);
529 if (!nla_entype) {
530 NL_SET_BAD_ATTR(extack, nla);
531 NL_SET_ERR_MSG(extack,
532 "Encap type is missing");
533 goto err_inval;
534 }
535
536 ret = lwtunnel_build_state(nla_get_u16(
537 nla_entype),
538 nla, AF_INET, cfg,
539 &lwtstate, extack);
540 if (ret)
541 goto errout;
542 nexthop_nh->nh_lwtstate =
543 lwtstate_get(lwtstate);
544 }
545 }
546
547 rtnh = rtnh_next(rtnh, &remaining);
548 } endfor_nexthops(fi);
549
550 return 0;
551
552err_inval:
553 ret = -EINVAL;
554
555errout:
556 return ret;
557}
558
559static void fib_rebalance(struct fib_info *fi)
560{
561 int total;
562 int w;
563 struct in_device *in_dev;
564
565 if (fi->fib_nhs < 2)
566 return;
567
568 total = 0;
569 for_nexthops(fi) {
570 if (nh->nh_flags & RTNH_F_DEAD)
571 continue;
572
573 in_dev = __in_dev_get_rtnl(nh->nh_dev);
574
575 if (in_dev &&
576 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
577 nh->nh_flags & RTNH_F_LINKDOWN)
578 continue;
579
580 total += nh->nh_weight;
581 } endfor_nexthops(fi);
582
583 w = 0;
584 change_nexthops(fi) {
585 int upper_bound;
586
587 in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev);
588
589 if (nexthop_nh->nh_flags & RTNH_F_DEAD) {
590 upper_bound = -1;
591 } else if (in_dev &&
592 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
593 nexthop_nh->nh_flags & RTNH_F_LINKDOWN) {
594 upper_bound = -1;
595 } else {
596 w += nexthop_nh->nh_weight;
597 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
598 total) - 1;
599 }
600
601 atomic_set(&nexthop_nh->nh_upper_bound, upper_bound);
602 } endfor_nexthops(fi);
603}
604#else /* CONFIG_IP_ROUTE_MULTIPATH */
605
606#define fib_rebalance(fi) do { } while (0)
607
608#endif /* CONFIG_IP_ROUTE_MULTIPATH */
609
610static int fib_encap_match(u16 encap_type,
611 struct nlattr *encap,
612 const struct fib_nh *nh,
613 const struct fib_config *cfg,
614 struct netlink_ext_ack *extack)
615{
616 struct lwtunnel_state *lwtstate;
617 int ret, result = 0;
618
619 if (encap_type == LWTUNNEL_ENCAP_NONE)
620 return 0;
621
622 ret = lwtunnel_build_state(encap_type, encap, AF_INET,
623 cfg, &lwtstate, extack);
624 if (!ret) {
625 result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
626 lwtstate_free(lwtstate);
627 }
628
629 return result;
630}
631
632int fib_nh_match(struct fib_config *cfg, struct fib_info *fi,
633 struct netlink_ext_ack *extack)
634{
635#ifdef CONFIG_IP_ROUTE_MULTIPATH
636 struct rtnexthop *rtnh;
637 int remaining;
638#endif
639
640 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
641 return 1;
642
643 if (cfg->fc_oif || cfg->fc_gw) {
644 if (cfg->fc_encap) {
645 if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap,
646 fi->fib_nh, cfg, extack))
647 return 1;
648 }
649#ifdef CONFIG_IP_ROUTE_CLASSID
650 if (cfg->fc_flow &&
651 cfg->fc_flow != fi->fib_nh->nh_tclassid)
652 return 1;
653#endif
654 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
655 (!cfg->fc_gw || cfg->fc_gw == fi->fib_nh->nh_gw))
656 return 0;
657 return 1;
658 }
659
660#ifdef CONFIG_IP_ROUTE_MULTIPATH
661 if (!cfg->fc_mp)
662 return 0;
663
664 rtnh = cfg->fc_mp;
665 remaining = cfg->fc_mp_len;
666
667 for_nexthops(fi) {
668 int attrlen;
669
670 if (!rtnh_ok(rtnh, remaining))
671 return -EINVAL;
672
673 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
674 return 1;
675
676 attrlen = rtnh_attrlen(rtnh);
677 if (attrlen > 0) {
678 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
679
680 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
681 if (nla && nla_get_in_addr(nla) != nh->nh_gw)
682 return 1;
683#ifdef CONFIG_IP_ROUTE_CLASSID
684 nla = nla_find(attrs, attrlen, RTA_FLOW);
685 if (nla && nla_get_u32(nla) != nh->nh_tclassid)
686 return 1;
687#endif
688 }
689
690 rtnh = rtnh_next(rtnh, &remaining);
691 } endfor_nexthops(fi);
692#endif
693 return 0;
694}
695
696bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
697{
698 struct nlattr *nla;
699 int remaining;
700
701 if (!cfg->fc_mx)
702 return true;
703
704 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
705 int type = nla_type(nla);
706 u32 fi_val, val;
707
708 if (!type)
709 continue;
710 if (type > RTAX_MAX)
711 return false;
712
713 if (type == RTAX_CC_ALGO) {
714 char tmp[TCP_CA_NAME_MAX];
715 bool ecn_ca = false;
716
717 nla_strlcpy(tmp, nla, sizeof(tmp));
718 val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
719 } else {
720 val = nla_get_u32(nla);
721 }
722
723 fi_val = fi->fib_metrics->metrics[type - 1];
724 if (type == RTAX_FEATURES)
725 fi_val &= ~DST_FEATURE_ECN_CA;
726
727 if (fi_val != val)
728 return false;
729 }
730
731 return true;
732}
733
734
735/*
736 * Picture
737 * -------
738 *
739 * Semantics of nexthop is very messy by historical reasons.
740 * We have to take into account, that:
741 * a) gateway can be actually local interface address,
742 * so that gatewayed route is direct.
743 * b) gateway must be on-link address, possibly
744 * described not by an ifaddr, but also by a direct route.
745 * c) If both gateway and interface are specified, they should not
746 * contradict.
747 * d) If we use tunnel routes, gateway could be not on-link.
748 *
749 * Attempt to reconcile all of these (alas, self-contradictory) conditions
750 * results in pretty ugly and hairy code with obscure logic.
751 *
752 * I chose to generalized it instead, so that the size
753 * of code does not increase practically, but it becomes
754 * much more general.
755 * Every prefix is assigned a "scope" value: "host" is local address,
756 * "link" is direct route,
757 * [ ... "site" ... "interior" ... ]
758 * and "universe" is true gateway route with global meaning.
759 *
760 * Every prefix refers to a set of "nexthop"s (gw, oif),
761 * where gw must have narrower scope. This recursion stops
762 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
763 * which means that gw is forced to be on link.
764 *
765 * Code is still hairy, but now it is apparently logically
766 * consistent and very flexible. F.e. as by-product it allows
767 * to co-exists in peace independent exterior and interior
768 * routing processes.
769 *
770 * Normally it looks as following.
771 *
772 * {universe prefix} -> (gw, oif) [scope link]
773 * |
774 * |-> {link prefix} -> (gw, oif) [scope local]
775 * |
776 * |-> {local prefix} (terminal node)
777 */
778static int fib_check_nh(struct fib_config *cfg, struct fib_nh *nh,
779 struct netlink_ext_ack *extack)
780{
781 int err = 0;
782 struct net *net;
783 struct net_device *dev;
784
785 net = cfg->fc_nlinfo.nl_net;
786 if (nh->nh_gw) {
787 struct fib_result res;
788
789 if (nh->nh_flags & RTNH_F_ONLINK) {
790 unsigned int addr_type;
791
792 if (cfg->fc_scope >= RT_SCOPE_LINK) {
793 NL_SET_ERR_MSG(extack,
794 "Nexthop has invalid scope");
795 return -EINVAL;
796 }
797 dev = __dev_get_by_index(net, nh->nh_oif);
798 if (!dev)
799 return -ENODEV;
800 if (!(dev->flags & IFF_UP)) {
801 NL_SET_ERR_MSG(extack,
802 "Nexthop device is not up");
803 return -ENETDOWN;
804 }
805 addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
806 if (addr_type != RTN_UNICAST) {
807 NL_SET_ERR_MSG(extack,
808 "Nexthop has invalid gateway");
809 return -EINVAL;
810 }
811 if (!netif_carrier_ok(dev))
812 nh->nh_flags |= RTNH_F_LINKDOWN;
813 nh->nh_dev = dev;
814 dev_hold(dev);
815 nh->nh_scope = RT_SCOPE_LINK;
816 return 0;
817 }
818 rcu_read_lock();
819 {
820 struct fib_table *tbl = NULL;
821 struct flowi4 fl4 = {
822 .daddr = nh->nh_gw,
823 .flowi4_scope = cfg->fc_scope + 1,
824 .flowi4_oif = nh->nh_oif,
825 .flowi4_iif = LOOPBACK_IFINDEX,
826 };
827
828 /* It is not necessary, but requires a bit of thinking */
829 if (fl4.flowi4_scope < RT_SCOPE_LINK)
830 fl4.flowi4_scope = RT_SCOPE_LINK;
831
832 if (cfg->fc_table)
833 tbl = fib_get_table(net, cfg->fc_table);
834
835 if (tbl)
836 err = fib_table_lookup(tbl, &fl4, &res,
837 FIB_LOOKUP_IGNORE_LINKSTATE |
838 FIB_LOOKUP_NOREF);
839
840 /* on error or if no table given do full lookup. This
841 * is needed for example when nexthops are in the local
842 * table rather than the given table
843 */
844 if (!tbl || err) {
845 err = fib_lookup(net, &fl4, &res,
846 FIB_LOOKUP_IGNORE_LINKSTATE);
847 }
848
849 if (err) {
850 NL_SET_ERR_MSG(extack,
851 "Nexthop has invalid gateway");
852 rcu_read_unlock();
853 return err;
854 }
855 }
856 err = -EINVAL;
857 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
858 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
859 goto out;
860 }
861 nh->nh_scope = res.scope;
862 nh->nh_oif = FIB_RES_OIF(res);
863 nh->nh_dev = dev = FIB_RES_DEV(res);
864 if (!dev) {
865 NL_SET_ERR_MSG(extack,
866 "No egress device for nexthop gateway");
867 goto out;
868 }
869 dev_hold(dev);
870 if (!netif_carrier_ok(dev))
871 nh->nh_flags |= RTNH_F_LINKDOWN;
872 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
873 } else {
874 struct in_device *in_dev;
875
876 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
877 NL_SET_ERR_MSG(extack,
878 "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
879 return -EINVAL;
880 }
881 rcu_read_lock();
882 err = -ENODEV;
883 in_dev = inetdev_by_index(net, nh->nh_oif);
884 if (!in_dev)
885 goto out;
886 err = -ENETDOWN;
887 if (!(in_dev->dev->flags & IFF_UP)) {
888 NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
889 goto out;
890 }
891 nh->nh_dev = in_dev->dev;
892 dev_hold(nh->nh_dev);
893 nh->nh_scope = RT_SCOPE_HOST;
894 if (!netif_carrier_ok(nh->nh_dev))
895 nh->nh_flags |= RTNH_F_LINKDOWN;
896 err = 0;
897 }
898out:
899 rcu_read_unlock();
900 return err;
901}
902
903static inline unsigned int fib_laddr_hashfn(__be32 val)
904{
905 unsigned int mask = (fib_info_hash_size - 1);
906
907 return ((__force u32)val ^
908 ((__force u32)val >> 7) ^
909 ((__force u32)val >> 14)) & mask;
910}
911
912static struct hlist_head *fib_info_hash_alloc(int bytes)
913{
914 if (bytes <= PAGE_SIZE)
915 return kzalloc(bytes, GFP_KERNEL);
916 else
917 return (struct hlist_head *)
918 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
919 get_order(bytes));
920}
921
922static void fib_info_hash_free(struct hlist_head *hash, int bytes)
923{
924 if (!hash)
925 return;
926
927 if (bytes <= PAGE_SIZE)
928 kfree(hash);
929 else
930 free_pages((unsigned long) hash, get_order(bytes));
931}
932
933static void fib_info_hash_move(struct hlist_head *new_info_hash,
934 struct hlist_head *new_laddrhash,
935 unsigned int new_size)
936{
937 struct hlist_head *old_info_hash, *old_laddrhash;
938 unsigned int old_size = fib_info_hash_size;
939 unsigned int i, bytes;
940
941 spin_lock_bh(&fib_info_lock);
942 old_info_hash = fib_info_hash;
943 old_laddrhash = fib_info_laddrhash;
944 fib_info_hash_size = new_size;
945
946 for (i = 0; i < old_size; i++) {
947 struct hlist_head *head = &fib_info_hash[i];
948 struct hlist_node *n;
949 struct fib_info *fi;
950
951 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
952 struct hlist_head *dest;
953 unsigned int new_hash;
954
955 new_hash = fib_info_hashfn(fi);
956 dest = &new_info_hash[new_hash];
957 hlist_add_head(&fi->fib_hash, dest);
958 }
959 }
960 fib_info_hash = new_info_hash;
961
962 for (i = 0; i < old_size; i++) {
963 struct hlist_head *lhead = &fib_info_laddrhash[i];
964 struct hlist_node *n;
965 struct fib_info *fi;
966
967 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
968 struct hlist_head *ldest;
969 unsigned int new_hash;
970
971 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
972 ldest = &new_laddrhash[new_hash];
973 hlist_add_head(&fi->fib_lhash, ldest);
974 }
975 }
976 fib_info_laddrhash = new_laddrhash;
977
978 spin_unlock_bh(&fib_info_lock);
979
980 bytes = old_size * sizeof(struct hlist_head *);
981 fib_info_hash_free(old_info_hash, bytes);
982 fib_info_hash_free(old_laddrhash, bytes);
983}
984
985__be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
986{
987 nh->nh_saddr = inet_select_addr(nh->nh_dev,
988 nh->nh_gw,
989 nh->nh_parent->fib_scope);
990 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
991
992 return nh->nh_saddr;
993}
994
995static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
996{
997 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
998 fib_prefsrc != cfg->fc_dst) {
999 u32 tb_id = cfg->fc_table;
1000 int rc;
1001
1002 if (tb_id == RT_TABLE_MAIN)
1003 tb_id = RT_TABLE_LOCAL;
1004
1005 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1006 fib_prefsrc, tb_id);
1007
1008 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1009 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1010 fib_prefsrc, RT_TABLE_LOCAL);
1011 }
1012
1013 if (rc != RTN_LOCAL)
1014 return false;
1015 }
1016 return true;
1017}
1018
1019static int
1020fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
1021{
1022 bool ecn_ca = false;
1023 struct nlattr *nla;
1024 int remaining;
1025
1026 if (!cfg->fc_mx)
1027 return 0;
1028
1029 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1030 int type = nla_type(nla);
1031 u32 val;
1032
1033 if (!type)
1034 continue;
1035 if (type > RTAX_MAX)
1036 return -EINVAL;
1037
1038 if (type == RTAX_CC_ALGO) {
1039 char tmp[TCP_CA_NAME_MAX];
1040
1041 nla_strlcpy(tmp, nla, sizeof(tmp));
1042 val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
1043 if (val == TCP_CA_UNSPEC)
1044 return -EINVAL;
1045 } else {
1046 val = nla_get_u32(nla);
1047 }
1048 if (type == RTAX_ADVMSS && val > 65535 - 40)
1049 val = 65535 - 40;
1050 if (type == RTAX_MTU && val > 65535 - 15)
1051 val = 65535 - 15;
1052 if (type == RTAX_HOPLIMIT && val > 255)
1053 val = 255;
1054 if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
1055 return -EINVAL;
1056 fi->fib_metrics->metrics[type - 1] = val;
1057 }
1058
1059 if (ecn_ca)
1060 fi->fib_metrics->metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
1061
1062 return 0;
1063}
1064
1065struct fib_info *fib_create_info(struct fib_config *cfg,
1066 struct netlink_ext_ack *extack)
1067{
1068 int err;
1069 struct fib_info *fi = NULL;
1070 struct fib_info *ofi;
1071 int nhs = 1;
1072 struct net *net = cfg->fc_nlinfo.nl_net;
1073
1074 if (cfg->fc_type > RTN_MAX)
1075 goto err_inval;
1076
1077 /* Fast check to catch the most weird cases */
1078 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1079 NL_SET_ERR_MSG(extack, "Invalid scope");
1080 goto err_inval;
1081 }
1082
1083 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1084 NL_SET_ERR_MSG(extack,
1085 "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1086 goto err_inval;
1087 }
1088
1089#ifdef CONFIG_IP_ROUTE_MULTIPATH
1090 if (cfg->fc_mp) {
1091 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1092 if (nhs == 0)
1093 goto err_inval;
1094 }
1095#endif
1096
1097 err = -ENOBUFS;
1098 if (fib_info_cnt >= fib_info_hash_size) {
1099 unsigned int new_size = fib_info_hash_size << 1;
1100 struct hlist_head *new_info_hash;
1101 struct hlist_head *new_laddrhash;
1102 unsigned int bytes;
1103
1104 if (!new_size)
1105 new_size = 16;
1106 bytes = new_size * sizeof(struct hlist_head *);
1107 new_info_hash = fib_info_hash_alloc(bytes);
1108 new_laddrhash = fib_info_hash_alloc(bytes);
1109 if (!new_info_hash || !new_laddrhash) {
1110 fib_info_hash_free(new_info_hash, bytes);
1111 fib_info_hash_free(new_laddrhash, bytes);
1112 } else
1113 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1114
1115 if (!fib_info_hash_size)
1116 goto failure;
1117 }
1118
1119 fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1120 if (!fi)
1121 goto failure;
1122 if (cfg->fc_mx) {
1123 fi->fib_metrics = kzalloc(sizeof(*fi->fib_metrics), GFP_KERNEL);
1124 if (unlikely(!fi->fib_metrics)) {
1125 kfree(fi);
1126 return ERR_PTR(err);
1127 }
1128 refcount_set(&fi->fib_metrics->refcnt, 1);
1129 } else {
1130 fi->fib_metrics = (struct dst_metrics *)&dst_default_metrics;
1131 }
1132 fib_info_cnt++;
1133 fi->fib_net = net;
1134 fi->fib_protocol = cfg->fc_protocol;
1135 fi->fib_scope = cfg->fc_scope;
1136 fi->fib_flags = cfg->fc_flags;
1137 fi->fib_priority = cfg->fc_priority;
1138 fi->fib_prefsrc = cfg->fc_prefsrc;
1139 fi->fib_type = cfg->fc_type;
1140 fi->fib_tb_id = cfg->fc_table;
1141
1142 fi->fib_nhs = nhs;
1143 change_nexthops(fi) {
1144 nexthop_nh->nh_parent = fi;
1145 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1146 if (!nexthop_nh->nh_pcpu_rth_output)
1147 goto failure;
1148 } endfor_nexthops(fi)
1149
1150 err = fib_convert_metrics(fi, cfg);
1151 if (err)
1152 goto failure;
1153
1154 if (cfg->fc_mp) {
1155#ifdef CONFIG_IP_ROUTE_MULTIPATH
1156 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, extack);
1157 if (err != 0)
1158 goto failure;
1159 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif) {
1160 NL_SET_ERR_MSG(extack,
1161 "Nexthop device index does not match RTA_OIF");
1162 goto err_inval;
1163 }
1164 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw) {
1165 NL_SET_ERR_MSG(extack,
1166 "Nexthop gateway does not match RTA_GATEWAY");
1167 goto err_inval;
1168 }
1169#ifdef CONFIG_IP_ROUTE_CLASSID
1170 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow) {
1171 NL_SET_ERR_MSG(extack,
1172 "Nexthop class id does not match RTA_FLOW");
1173 goto err_inval;
1174 }
1175#endif
1176#else
1177 NL_SET_ERR_MSG(extack,
1178 "Multipath support not enabled in kernel");
1179 goto err_inval;
1180#endif
1181 } else {
1182 struct fib_nh *nh = fi->fib_nh;
1183
1184 if (cfg->fc_encap) {
1185 struct lwtunnel_state *lwtstate;
1186
1187 if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE) {
1188 NL_SET_ERR_MSG(extack,
1189 "LWT encap type not specified");
1190 goto err_inval;
1191 }
1192 err = lwtunnel_build_state(cfg->fc_encap_type,
1193 cfg->fc_encap, AF_INET, cfg,
1194 &lwtstate, extack);
1195 if (err)
1196 goto failure;
1197
1198 nh->nh_lwtstate = lwtstate_get(lwtstate);
1199 }
1200 nh->nh_oif = cfg->fc_oif;
1201 nh->nh_gw = cfg->fc_gw;
1202 nh->nh_flags = cfg->fc_flags;
1203#ifdef CONFIG_IP_ROUTE_CLASSID
1204 nh->nh_tclassid = cfg->fc_flow;
1205 if (nh->nh_tclassid)
1206 fi->fib_net->ipv4.fib_num_tclassid_users++;
1207#endif
1208#ifdef CONFIG_IP_ROUTE_MULTIPATH
1209 nh->nh_weight = 1;
1210#endif
1211 }
1212
1213 if (fib_props[cfg->fc_type].error) {
1214 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp) {
1215 NL_SET_ERR_MSG(extack,
1216 "Gateway, device and multipath can not be specified for this route type");
1217 goto err_inval;
1218 }
1219 goto link_it;
1220 } else {
1221 switch (cfg->fc_type) {
1222 case RTN_UNICAST:
1223 case RTN_LOCAL:
1224 case RTN_BROADCAST:
1225 case RTN_ANYCAST:
1226 case RTN_MULTICAST:
1227 break;
1228 default:
1229 NL_SET_ERR_MSG(extack, "Invalid route type");
1230 goto err_inval;
1231 }
1232 }
1233
1234 if (cfg->fc_scope > RT_SCOPE_HOST) {
1235 NL_SET_ERR_MSG(extack, "Invalid scope");
1236 goto err_inval;
1237 }
1238
1239 if (cfg->fc_scope == RT_SCOPE_HOST) {
1240 struct fib_nh *nh = fi->fib_nh;
1241
1242 /* Local address is added. */
1243 if (nhs != 1) {
1244 NL_SET_ERR_MSG(extack,
1245 "Route with host scope can not have multiple nexthops");
1246 goto err_inval;
1247 }
1248 if (nh->nh_gw) {
1249 NL_SET_ERR_MSG(extack,
1250 "Route with host scope can not have a gateway");
1251 goto err_inval;
1252 }
1253 nh->nh_scope = RT_SCOPE_NOWHERE;
1254 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1255 err = -ENODEV;
1256 if (!nh->nh_dev)
1257 goto failure;
1258 } else {
1259 int linkdown = 0;
1260
1261 change_nexthops(fi) {
1262 err = fib_check_nh(cfg, nexthop_nh, extack);
1263 if (err != 0)
1264 goto failure;
1265 if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1266 linkdown++;
1267 } endfor_nexthops(fi)
1268 if (linkdown == fi->fib_nhs)
1269 fi->fib_flags |= RTNH_F_LINKDOWN;
1270 }
1271
1272 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1273 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1274 goto err_inval;
1275 }
1276
1277 change_nexthops(fi) {
1278 fib_info_update_nh_saddr(net, nexthop_nh);
1279 } endfor_nexthops(fi)
1280
1281 fib_rebalance(fi);
1282
1283link_it:
1284 ofi = fib_find_info(fi);
1285 if (ofi) {
1286 fi->fib_dead = 1;
1287 free_fib_info(fi);
1288 ofi->fib_treeref++;
1289 return ofi;
1290 }
1291
1292 fi->fib_treeref++;
1293 refcount_set(&fi->fib_clntref, 1);
1294 spin_lock_bh(&fib_info_lock);
1295 hlist_add_head(&fi->fib_hash,
1296 &fib_info_hash[fib_info_hashfn(fi)]);
1297 if (fi->fib_prefsrc) {
1298 struct hlist_head *head;
1299
1300 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1301 hlist_add_head(&fi->fib_lhash, head);
1302 }
1303 change_nexthops(fi) {
1304 struct hlist_head *head;
1305 unsigned int hash;
1306
1307 if (!nexthop_nh->nh_dev)
1308 continue;
1309 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1310 head = &fib_info_devhash[hash];
1311 hlist_add_head(&nexthop_nh->nh_hash, head);
1312 } endfor_nexthops(fi)
1313 spin_unlock_bh(&fib_info_lock);
1314 return fi;
1315
1316err_inval:
1317 err = -EINVAL;
1318
1319failure:
1320 if (fi) {
1321 fi->fib_dead = 1;
1322 free_fib_info(fi);
1323 }
1324
1325 return ERR_PTR(err);
1326}
1327
1328int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1329 u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1330 struct fib_info *fi, unsigned int flags)
1331{
1332 struct nlmsghdr *nlh;
1333 struct rtmsg *rtm;
1334
1335 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1336 if (!nlh)
1337 return -EMSGSIZE;
1338
1339 rtm = nlmsg_data(nlh);
1340 rtm->rtm_family = AF_INET;
1341 rtm->rtm_dst_len = dst_len;
1342 rtm->rtm_src_len = 0;
1343 rtm->rtm_tos = tos;
1344 if (tb_id < 256)
1345 rtm->rtm_table = tb_id;
1346 else
1347 rtm->rtm_table = RT_TABLE_COMPAT;
1348 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1349 goto nla_put_failure;
1350 rtm->rtm_type = type;
1351 rtm->rtm_flags = fi->fib_flags;
1352 rtm->rtm_scope = fi->fib_scope;
1353 rtm->rtm_protocol = fi->fib_protocol;
1354
1355 if (rtm->rtm_dst_len &&
1356 nla_put_in_addr(skb, RTA_DST, dst))
1357 goto nla_put_failure;
1358 if (fi->fib_priority &&
1359 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1360 goto nla_put_failure;
1361 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1362 goto nla_put_failure;
1363
1364 if (fi->fib_prefsrc &&
1365 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1366 goto nla_put_failure;
1367 if (fi->fib_nhs == 1) {
1368 if (fi->fib_nh->nh_gw &&
1369 nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1370 goto nla_put_failure;
1371 if (fi->fib_nh->nh_oif &&
1372 nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1373 goto nla_put_failure;
1374 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1375 struct in_device *in_dev;
1376
1377 rcu_read_lock();
1378 in_dev = __in_dev_get_rcu(fi->fib_nh->nh_dev);
1379 if (in_dev &&
1380 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1381 rtm->rtm_flags |= RTNH_F_DEAD;
1382 rcu_read_unlock();
1383 }
1384 if (fi->fib_nh->nh_flags & RTNH_F_OFFLOAD)
1385 rtm->rtm_flags |= RTNH_F_OFFLOAD;
1386#ifdef CONFIG_IP_ROUTE_CLASSID
1387 if (fi->fib_nh[0].nh_tclassid &&
1388 nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1389 goto nla_put_failure;
1390#endif
1391 if (fi->fib_nh->nh_lwtstate &&
1392 lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate) < 0)
1393 goto nla_put_failure;
1394 }
1395#ifdef CONFIG_IP_ROUTE_MULTIPATH
1396 if (fi->fib_nhs > 1) {
1397 struct rtnexthop *rtnh;
1398 struct nlattr *mp;
1399
1400 mp = nla_nest_start(skb, RTA_MULTIPATH);
1401 if (!mp)
1402 goto nla_put_failure;
1403
1404 for_nexthops(fi) {
1405 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1406 if (!rtnh)
1407 goto nla_put_failure;
1408
1409 rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1410 if (nh->nh_flags & RTNH_F_LINKDOWN) {
1411 struct in_device *in_dev;
1412
1413 rcu_read_lock();
1414 in_dev = __in_dev_get_rcu(nh->nh_dev);
1415 if (in_dev &&
1416 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1417 rtnh->rtnh_flags |= RTNH_F_DEAD;
1418 rcu_read_unlock();
1419 }
1420 rtnh->rtnh_hops = nh->nh_weight - 1;
1421 rtnh->rtnh_ifindex = nh->nh_oif;
1422
1423 if (nh->nh_gw &&
1424 nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1425 goto nla_put_failure;
1426#ifdef CONFIG_IP_ROUTE_CLASSID
1427 if (nh->nh_tclassid &&
1428 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1429 goto nla_put_failure;
1430#endif
1431 if (nh->nh_lwtstate &&
1432 lwtunnel_fill_encap(skb, nh->nh_lwtstate) < 0)
1433 goto nla_put_failure;
1434
1435 /* length of rtnetlink header + attributes */
1436 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1437 } endfor_nexthops(fi);
1438
1439 nla_nest_end(skb, mp);
1440 }
1441#endif
1442 nlmsg_end(skb, nlh);
1443 return 0;
1444
1445nla_put_failure:
1446 nlmsg_cancel(skb, nlh);
1447 return -EMSGSIZE;
1448}
1449
1450/*
1451 * Update FIB if:
1452 * - local address disappeared -> we must delete all the entries
1453 * referring to it.
1454 * - device went down -> we must shutdown all nexthops going via it.
1455 */
1456int fib_sync_down_addr(struct net_device *dev, __be32 local)
1457{
1458 int ret = 0;
1459 unsigned int hash = fib_laddr_hashfn(local);
1460 struct hlist_head *head = &fib_info_laddrhash[hash];
1461 struct net *net = dev_net(dev);
1462 int tb_id = l3mdev_fib_table(dev);
1463 struct fib_info *fi;
1464
1465 if (!fib_info_laddrhash || local == 0)
1466 return 0;
1467
1468 hlist_for_each_entry(fi, head, fib_lhash) {
1469 if (!net_eq(fi->fib_net, net) ||
1470 fi->fib_tb_id != tb_id)
1471 continue;
1472 if (fi->fib_prefsrc == local) {
1473 fi->fib_flags |= RTNH_F_DEAD;
1474 ret++;
1475 }
1476 }
1477 return ret;
1478}
1479
1480static int call_fib_nh_notifiers(struct fib_nh *fib_nh,
1481 enum fib_event_type event_type)
1482{
1483 struct in_device *in_dev = __in_dev_get_rtnl(fib_nh->nh_dev);
1484 struct fib_nh_notifier_info info = {
1485 .fib_nh = fib_nh,
1486 };
1487
1488 switch (event_type) {
1489 case FIB_EVENT_NH_ADD:
1490 if (fib_nh->nh_flags & RTNH_F_DEAD)
1491 break;
1492 if (IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1493 fib_nh->nh_flags & RTNH_F_LINKDOWN)
1494 break;
1495 return call_fib4_notifiers(dev_net(fib_nh->nh_dev), event_type,
1496 &info.info);
1497 case FIB_EVENT_NH_DEL:
1498 if ((in_dev && IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1499 fib_nh->nh_flags & RTNH_F_LINKDOWN) ||
1500 (fib_nh->nh_flags & RTNH_F_DEAD))
1501 return call_fib4_notifiers(dev_net(fib_nh->nh_dev),
1502 event_type, &info.info);
1503 default:
1504 break;
1505 }
1506
1507 return NOTIFY_DONE;
1508}
1509
1510/* Event force Flags Description
1511 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host
1512 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host
1513 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed
1514 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed
1515 */
1516int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1517{
1518 int ret = 0;
1519 int scope = RT_SCOPE_NOWHERE;
1520 struct fib_info *prev_fi = NULL;
1521 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1522 struct hlist_head *head = &fib_info_devhash[hash];
1523 struct fib_nh *nh;
1524
1525 if (force)
1526 scope = -1;
1527
1528 hlist_for_each_entry(nh, head, nh_hash) {
1529 struct fib_info *fi = nh->nh_parent;
1530 int dead;
1531
1532 BUG_ON(!fi->fib_nhs);
1533 if (nh->nh_dev != dev || fi == prev_fi)
1534 continue;
1535 prev_fi = fi;
1536 dead = 0;
1537 change_nexthops(fi) {
1538 if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1539 dead++;
1540 else if (nexthop_nh->nh_dev == dev &&
1541 nexthop_nh->nh_scope != scope) {
1542 switch (event) {
1543 case NETDEV_DOWN:
1544 case NETDEV_UNREGISTER:
1545 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1546 /* fall through */
1547 case NETDEV_CHANGE:
1548 nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1549 break;
1550 }
1551 call_fib_nh_notifiers(nexthop_nh,
1552 FIB_EVENT_NH_DEL);
1553 dead++;
1554 }
1555#ifdef CONFIG_IP_ROUTE_MULTIPATH
1556 if (event == NETDEV_UNREGISTER &&
1557 nexthop_nh->nh_dev == dev) {
1558 dead = fi->fib_nhs;
1559 break;
1560 }
1561#endif
1562 } endfor_nexthops(fi)
1563 if (dead == fi->fib_nhs) {
1564 switch (event) {
1565 case NETDEV_DOWN:
1566 case NETDEV_UNREGISTER:
1567 fi->fib_flags |= RTNH_F_DEAD;
1568 /* fall through */
1569 case NETDEV_CHANGE:
1570 fi->fib_flags |= RTNH_F_LINKDOWN;
1571 break;
1572 }
1573 ret++;
1574 }
1575
1576 fib_rebalance(fi);
1577 }
1578
1579 return ret;
1580}
1581
1582/* Must be invoked inside of an RCU protected region. */
1583static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1584{
1585 struct fib_info *fi = NULL, *last_resort = NULL;
1586 struct hlist_head *fa_head = res->fa_head;
1587 struct fib_table *tb = res->table;
1588 u8 slen = 32 - res->prefixlen;
1589 int order = -1, last_idx = -1;
1590 struct fib_alias *fa, *fa1 = NULL;
1591 u32 last_prio = res->fi->fib_priority;
1592 u8 last_tos = 0;
1593
1594 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1595 struct fib_info *next_fi = fa->fa_info;
1596
1597 if (fa->fa_slen != slen)
1598 continue;
1599 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1600 continue;
1601 if (fa->tb_id != tb->tb_id)
1602 continue;
1603 if (next_fi->fib_priority > last_prio &&
1604 fa->fa_tos == last_tos) {
1605 if (last_tos)
1606 continue;
1607 break;
1608 }
1609 if (next_fi->fib_flags & RTNH_F_DEAD)
1610 continue;
1611 last_tos = fa->fa_tos;
1612 last_prio = next_fi->fib_priority;
1613
1614 if (next_fi->fib_scope != res->scope ||
1615 fa->fa_type != RTN_UNICAST)
1616 continue;
1617 if (!next_fi->fib_nh[0].nh_gw ||
1618 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1619 continue;
1620
1621 fib_alias_accessed(fa);
1622
1623 if (!fi) {
1624 if (next_fi != res->fi)
1625 break;
1626 fa1 = fa;
1627 } else if (!fib_detect_death(fi, order, &last_resort,
1628 &last_idx, fa1->fa_default)) {
1629 fib_result_assign(res, fi);
1630 fa1->fa_default = order;
1631 goto out;
1632 }
1633 fi = next_fi;
1634 order++;
1635 }
1636
1637 if (order <= 0 || !fi) {
1638 if (fa1)
1639 fa1->fa_default = -1;
1640 goto out;
1641 }
1642
1643 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1644 fa1->fa_default)) {
1645 fib_result_assign(res, fi);
1646 fa1->fa_default = order;
1647 goto out;
1648 }
1649
1650 if (last_idx >= 0)
1651 fib_result_assign(res, last_resort);
1652 fa1->fa_default = last_idx;
1653out:
1654 return;
1655}
1656
1657/*
1658 * Dead device goes up. We wake up dead nexthops.
1659 * It takes sense only on multipath routes.
1660 */
1661int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1662{
1663 struct fib_info *prev_fi;
1664 unsigned int hash;
1665 struct hlist_head *head;
1666 struct fib_nh *nh;
1667 int ret;
1668
1669 if (!(dev->flags & IFF_UP))
1670 return 0;
1671
1672 if (nh_flags & RTNH_F_DEAD) {
1673 unsigned int flags = dev_get_flags(dev);
1674
1675 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1676 nh_flags |= RTNH_F_LINKDOWN;
1677 }
1678
1679 prev_fi = NULL;
1680 hash = fib_devindex_hashfn(dev->ifindex);
1681 head = &fib_info_devhash[hash];
1682 ret = 0;
1683
1684 hlist_for_each_entry(nh, head, nh_hash) {
1685 struct fib_info *fi = nh->nh_parent;
1686 int alive;
1687
1688 BUG_ON(!fi->fib_nhs);
1689 if (nh->nh_dev != dev || fi == prev_fi)
1690 continue;
1691
1692 prev_fi = fi;
1693 alive = 0;
1694 change_nexthops(fi) {
1695 if (!(nexthop_nh->nh_flags & nh_flags)) {
1696 alive++;
1697 continue;
1698 }
1699 if (!nexthop_nh->nh_dev ||
1700 !(nexthop_nh->nh_dev->flags & IFF_UP))
1701 continue;
1702 if (nexthop_nh->nh_dev != dev ||
1703 !__in_dev_get_rtnl(dev))
1704 continue;
1705 alive++;
1706 nexthop_nh->nh_flags &= ~nh_flags;
1707 call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
1708 } endfor_nexthops(fi)
1709
1710 if (alive > 0) {
1711 fi->fib_flags &= ~nh_flags;
1712 ret++;
1713 }
1714
1715 fib_rebalance(fi);
1716 }
1717
1718 return ret;
1719}
1720
1721#ifdef CONFIG_IP_ROUTE_MULTIPATH
1722static bool fib_good_nh(const struct fib_nh *nh)
1723{
1724 int state = NUD_REACHABLE;
1725
1726 if (nh->nh_scope == RT_SCOPE_LINK) {
1727 struct neighbour *n;
1728
1729 rcu_read_lock_bh();
1730
1731 n = __ipv4_neigh_lookup_noref(nh->nh_dev,
1732 (__force u32)nh->nh_gw);
1733 if (n)
1734 state = n->nud_state;
1735
1736 rcu_read_unlock_bh();
1737 }
1738
1739 return !!(state & NUD_VALID);
1740}
1741
1742void fib_select_multipath(struct fib_result *res, int hash)
1743{
1744 struct fib_info *fi = res->fi;
1745 struct net *net = fi->fib_net;
1746 bool first = false;
1747
1748 for_nexthops(fi) {
1749 if (net->ipv4.sysctl_fib_multipath_use_neigh) {
1750 if (!fib_good_nh(nh))
1751 continue;
1752 if (!first) {
1753 res->nh_sel = nhsel;
1754 first = true;
1755 }
1756 }
1757
1758 if (hash > atomic_read(&nh->nh_upper_bound))
1759 continue;
1760
1761 res->nh_sel = nhsel;
1762 return;
1763 } endfor_nexthops(fi);
1764}
1765#endif
1766
1767void fib_select_path(struct net *net, struct fib_result *res,
1768 struct flowi4 *fl4, const struct sk_buff *skb)
1769{
1770 if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
1771 goto check_saddr;
1772
1773#ifdef CONFIG_IP_ROUTE_MULTIPATH
1774 if (res->fi->fib_nhs > 1) {
1775 int h = fib_multipath_hash(net, fl4, skb, NULL);
1776
1777 fib_select_multipath(res, h);
1778 }
1779 else
1780#endif
1781 if (!res->prefixlen &&
1782 res->table->tb_num_default > 1 &&
1783 res->type == RTN_UNICAST)
1784 fib_select_default(fl4, res);
1785
1786check_saddr:
1787 if (!fl4->saddr)
1788 fl4->saddr = FIB_RES_PREFSRC(net, *res);
1789}
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * IPv4 Forwarding Information Base: semantics.
7 *
8 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
14 */
15
16#include <linux/uaccess.h>
17#include <linux/bitops.h>
18#include <linux/types.h>
19#include <linux/kernel.h>
20#include <linux/jiffies.h>
21#include <linux/mm.h>
22#include <linux/string.h>
23#include <linux/socket.h>
24#include <linux/sockios.h>
25#include <linux/errno.h>
26#include <linux/in.h>
27#include <linux/inet.h>
28#include <linux/inetdevice.h>
29#include <linux/netdevice.h>
30#include <linux/if_arp.h>
31#include <linux/proc_fs.h>
32#include <linux/skbuff.h>
33#include <linux/init.h>
34#include <linux/slab.h>
35
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}
237EXPORT_SYMBOL_GPL(free_fib_info);
238
239void fib_release_info(struct fib_info *fi)
240{
241 spin_lock_bh(&fib_info_lock);
242 if (fi && --fi->fib_treeref == 0) {
243 hlist_del(&fi->fib_hash);
244 if (fi->fib_prefsrc)
245 hlist_del(&fi->fib_lhash);
246 change_nexthops(fi) {
247 if (!nexthop_nh->nh_dev)
248 continue;
249 hlist_del(&nexthop_nh->nh_hash);
250 } endfor_nexthops(fi)
251 fi->fib_dead = 1;
252 fib_info_put(fi);
253 }
254 spin_unlock_bh(&fib_info_lock);
255}
256
257static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
258{
259 const struct fib_nh *onh = ofi->fib_nh;
260
261 for_nexthops(fi) {
262 if (nh->nh_oif != onh->nh_oif ||
263 nh->nh_gw != onh->nh_gw ||
264 nh->nh_scope != onh->nh_scope ||
265#ifdef CONFIG_IP_ROUTE_MULTIPATH
266 nh->nh_weight != onh->nh_weight ||
267#endif
268#ifdef CONFIG_IP_ROUTE_CLASSID
269 nh->nh_tclassid != onh->nh_tclassid ||
270#endif
271 lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) ||
272 ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
273 return -1;
274 onh++;
275 } endfor_nexthops(fi);
276 return 0;
277}
278
279static inline unsigned int fib_devindex_hashfn(unsigned int val)
280{
281 unsigned int mask = DEVINDEX_HASHSIZE - 1;
282
283 return (val ^
284 (val >> DEVINDEX_HASHBITS) ^
285 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
286}
287
288static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
289{
290 unsigned int mask = (fib_info_hash_size - 1);
291 unsigned int val = fi->fib_nhs;
292
293 val ^= (fi->fib_protocol << 8) | fi->fib_scope;
294 val ^= (__force u32)fi->fib_prefsrc;
295 val ^= fi->fib_priority;
296 for_nexthops(fi) {
297 val ^= fib_devindex_hashfn(nh->nh_oif);
298 } endfor_nexthops(fi)
299
300 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
301}
302
303static struct fib_info *fib_find_info(const struct fib_info *nfi)
304{
305 struct hlist_head *head;
306 struct fib_info *fi;
307 unsigned int hash;
308
309 hash = fib_info_hashfn(nfi);
310 head = &fib_info_hash[hash];
311
312 hlist_for_each_entry(fi, head, fib_hash) {
313 if (!net_eq(fi->fib_net, nfi->fib_net))
314 continue;
315 if (fi->fib_nhs != nfi->fib_nhs)
316 continue;
317 if (nfi->fib_protocol == fi->fib_protocol &&
318 nfi->fib_scope == fi->fib_scope &&
319 nfi->fib_prefsrc == fi->fib_prefsrc &&
320 nfi->fib_priority == fi->fib_priority &&
321 nfi->fib_type == fi->fib_type &&
322 memcmp(nfi->fib_metrics, fi->fib_metrics,
323 sizeof(u32) * RTAX_MAX) == 0 &&
324 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
325 (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
326 return fi;
327 }
328
329 return NULL;
330}
331
332/* Check, that the gateway is already configured.
333 * Used only by redirect accept routine.
334 */
335int ip_fib_check_default(__be32 gw, struct net_device *dev)
336{
337 struct hlist_head *head;
338 struct fib_nh *nh;
339 unsigned int hash;
340
341 spin_lock(&fib_info_lock);
342
343 hash = fib_devindex_hashfn(dev->ifindex);
344 head = &fib_info_devhash[hash];
345 hlist_for_each_entry(nh, head, nh_hash) {
346 if (nh->nh_dev == dev &&
347 nh->nh_gw == gw &&
348 !(nh->nh_flags & RTNH_F_DEAD)) {
349 spin_unlock(&fib_info_lock);
350 return 0;
351 }
352 }
353
354 spin_unlock(&fib_info_lock);
355
356 return -1;
357}
358
359static inline size_t fib_nlmsg_size(struct fib_info *fi)
360{
361 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
362 + nla_total_size(4) /* RTA_TABLE */
363 + nla_total_size(4) /* RTA_DST */
364 + nla_total_size(4) /* RTA_PRIORITY */
365 + nla_total_size(4) /* RTA_PREFSRC */
366 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
367
368 /* space for nested metrics */
369 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
370
371 if (fi->fib_nhs) {
372 size_t nh_encapsize = 0;
373 /* Also handles the special case fib_nhs == 1 */
374
375 /* each nexthop is packed in an attribute */
376 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
377
378 /* may contain flow and gateway attribute */
379 nhsize += 2 * nla_total_size(4);
380
381 /* grab encap info */
382 for_nexthops(fi) {
383 if (nh->nh_lwtstate) {
384 /* RTA_ENCAP_TYPE */
385 nh_encapsize += lwtunnel_get_encap_size(
386 nh->nh_lwtstate);
387 /* RTA_ENCAP */
388 nh_encapsize += nla_total_size(2);
389 }
390 } endfor_nexthops(fi);
391
392 /* all nexthops are packed in a nested attribute */
393 payload += nla_total_size((fi->fib_nhs * nhsize) +
394 nh_encapsize);
395
396 }
397
398 return payload;
399}
400
401void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
402 int dst_len, u32 tb_id, const struct nl_info *info,
403 unsigned int nlm_flags)
404{
405 struct sk_buff *skb;
406 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
407 int err = -ENOBUFS;
408
409 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
410 if (!skb)
411 goto errout;
412
413 err = fib_dump_info(skb, info->portid, seq, event, tb_id,
414 fa->fa_type, key, dst_len,
415 fa->fa_tos, fa->fa_info, nlm_flags);
416 if (err < 0) {
417 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
418 WARN_ON(err == -EMSGSIZE);
419 kfree_skb(skb);
420 goto errout;
421 }
422 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
423 info->nlh, GFP_KERNEL);
424 return;
425errout:
426 if (err < 0)
427 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
428}
429
430static int fib_detect_death(struct fib_info *fi, int order,
431 struct fib_info **last_resort, int *last_idx,
432 int dflt)
433{
434 struct neighbour *n;
435 int state = NUD_NONE;
436
437 n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
438 if (n) {
439 state = n->nud_state;
440 neigh_release(n);
441 } else {
442 return 0;
443 }
444 if (state == NUD_REACHABLE)
445 return 0;
446 if ((state & NUD_VALID) && order != dflt)
447 return 0;
448 if ((state & NUD_VALID) ||
449 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
450 *last_resort = fi;
451 *last_idx = order;
452 }
453 return 1;
454}
455
456#ifdef CONFIG_IP_ROUTE_MULTIPATH
457
458static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
459{
460 int nhs = 0;
461
462 while (rtnh_ok(rtnh, remaining)) {
463 nhs++;
464 rtnh = rtnh_next(rtnh, &remaining);
465 }
466
467 /* leftover implies invalid nexthop configuration, discard it */
468 return remaining > 0 ? 0 : nhs;
469}
470
471static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
472 int remaining, struct fib_config *cfg)
473{
474 struct net *net = cfg->fc_nlinfo.nl_net;
475 int ret;
476
477 change_nexthops(fi) {
478 int attrlen;
479
480 if (!rtnh_ok(rtnh, remaining))
481 return -EINVAL;
482
483 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
484 return -EINVAL;
485
486 nexthop_nh->nh_flags =
487 (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
488 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
489 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
490
491 attrlen = rtnh_attrlen(rtnh);
492 if (attrlen > 0) {
493 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
494
495 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
496 nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
497#ifdef CONFIG_IP_ROUTE_CLASSID
498 nla = nla_find(attrs, attrlen, RTA_FLOW);
499 nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
500 if (nexthop_nh->nh_tclassid)
501 fi->fib_net->ipv4.fib_num_tclassid_users++;
502#endif
503 nla = nla_find(attrs, attrlen, RTA_ENCAP);
504 if (nla) {
505 struct lwtunnel_state *lwtstate;
506 struct net_device *dev = NULL;
507 struct nlattr *nla_entype;
508
509 nla_entype = nla_find(attrs, attrlen,
510 RTA_ENCAP_TYPE);
511 if (!nla_entype)
512 goto err_inval;
513 if (cfg->fc_oif)
514 dev = __dev_get_by_index(net, cfg->fc_oif);
515 ret = lwtunnel_build_state(dev, nla_get_u16(
516 nla_entype),
517 nla, AF_INET, cfg,
518 &lwtstate);
519 if (ret)
520 goto errout;
521 nexthop_nh->nh_lwtstate =
522 lwtstate_get(lwtstate);
523 }
524 }
525
526 rtnh = rtnh_next(rtnh, &remaining);
527 } endfor_nexthops(fi);
528
529 return 0;
530
531err_inval:
532 ret = -EINVAL;
533
534errout:
535 return ret;
536}
537
538static void fib_rebalance(struct fib_info *fi)
539{
540 int total;
541 int w;
542 struct in_device *in_dev;
543
544 if (fi->fib_nhs < 2)
545 return;
546
547 total = 0;
548 for_nexthops(fi) {
549 if (nh->nh_flags & RTNH_F_DEAD)
550 continue;
551
552 in_dev = __in_dev_get_rtnl(nh->nh_dev);
553
554 if (in_dev &&
555 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
556 nh->nh_flags & RTNH_F_LINKDOWN)
557 continue;
558
559 total += nh->nh_weight;
560 } endfor_nexthops(fi);
561
562 w = 0;
563 change_nexthops(fi) {
564 int upper_bound;
565
566 in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev);
567
568 if (nexthop_nh->nh_flags & RTNH_F_DEAD) {
569 upper_bound = -1;
570 } else if (in_dev &&
571 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
572 nexthop_nh->nh_flags & RTNH_F_LINKDOWN) {
573 upper_bound = -1;
574 } else {
575 w += nexthop_nh->nh_weight;
576 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
577 total) - 1;
578 }
579
580 atomic_set(&nexthop_nh->nh_upper_bound, upper_bound);
581 } endfor_nexthops(fi);
582
583 net_get_random_once(&fib_multipath_secret,
584 sizeof(fib_multipath_secret));
585}
586
587static inline void fib_add_weight(struct fib_info *fi,
588 const struct fib_nh *nh)
589{
590 fi->fib_weight += nh->nh_weight;
591}
592
593#else /* CONFIG_IP_ROUTE_MULTIPATH */
594
595#define fib_rebalance(fi) do { } while (0)
596#define fib_add_weight(fi, nh) do { } while (0)
597
598#endif /* CONFIG_IP_ROUTE_MULTIPATH */
599
600static int fib_encap_match(struct net *net, u16 encap_type,
601 struct nlattr *encap,
602 int oif, const struct fib_nh *nh,
603 const struct fib_config *cfg)
604{
605 struct lwtunnel_state *lwtstate;
606 struct net_device *dev = NULL;
607 int ret, result = 0;
608
609 if (encap_type == LWTUNNEL_ENCAP_NONE)
610 return 0;
611
612 if (oif)
613 dev = __dev_get_by_index(net, oif);
614 ret = lwtunnel_build_state(dev, encap_type, encap,
615 AF_INET, cfg, &lwtstate);
616 if (!ret) {
617 result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
618 lwtstate_free(lwtstate);
619 }
620
621 return result;
622}
623
624int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
625{
626 struct net *net = cfg->fc_nlinfo.nl_net;
627#ifdef CONFIG_IP_ROUTE_MULTIPATH
628 struct rtnexthop *rtnh;
629 int remaining;
630#endif
631
632 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
633 return 1;
634
635 if (cfg->fc_oif || cfg->fc_gw) {
636 if (cfg->fc_encap) {
637 if (fib_encap_match(net, cfg->fc_encap_type,
638 cfg->fc_encap, cfg->fc_oif,
639 fi->fib_nh, cfg))
640 return 1;
641 }
642 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
643 (!cfg->fc_gw || cfg->fc_gw == fi->fib_nh->nh_gw))
644 return 0;
645 return 1;
646 }
647
648#ifdef CONFIG_IP_ROUTE_MULTIPATH
649 if (!cfg->fc_mp)
650 return 0;
651
652 rtnh = cfg->fc_mp;
653 remaining = cfg->fc_mp_len;
654
655 for_nexthops(fi) {
656 int attrlen;
657
658 if (!rtnh_ok(rtnh, remaining))
659 return -EINVAL;
660
661 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
662 return 1;
663
664 attrlen = rtnh_attrlen(rtnh);
665 if (attrlen > 0) {
666 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
667
668 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
669 if (nla && nla_get_in_addr(nla) != nh->nh_gw)
670 return 1;
671#ifdef CONFIG_IP_ROUTE_CLASSID
672 nla = nla_find(attrs, attrlen, RTA_FLOW);
673 if (nla && nla_get_u32(nla) != nh->nh_tclassid)
674 return 1;
675#endif
676 }
677
678 rtnh = rtnh_next(rtnh, &remaining);
679 } endfor_nexthops(fi);
680#endif
681 return 0;
682}
683
684
685/*
686 * Picture
687 * -------
688 *
689 * Semantics of nexthop is very messy by historical reasons.
690 * We have to take into account, that:
691 * a) gateway can be actually local interface address,
692 * so that gatewayed route is direct.
693 * b) gateway must be on-link address, possibly
694 * described not by an ifaddr, but also by a direct route.
695 * c) If both gateway and interface are specified, they should not
696 * contradict.
697 * d) If we use tunnel routes, gateway could be not on-link.
698 *
699 * Attempt to reconcile all of these (alas, self-contradictory) conditions
700 * results in pretty ugly and hairy code with obscure logic.
701 *
702 * I chose to generalized it instead, so that the size
703 * of code does not increase practically, but it becomes
704 * much more general.
705 * Every prefix is assigned a "scope" value: "host" is local address,
706 * "link" is direct route,
707 * [ ... "site" ... "interior" ... ]
708 * and "universe" is true gateway route with global meaning.
709 *
710 * Every prefix refers to a set of "nexthop"s (gw, oif),
711 * where gw must have narrower scope. This recursion stops
712 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
713 * which means that gw is forced to be on link.
714 *
715 * Code is still hairy, but now it is apparently logically
716 * consistent and very flexible. F.e. as by-product it allows
717 * to co-exists in peace independent exterior and interior
718 * routing processes.
719 *
720 * Normally it looks as following.
721 *
722 * {universe prefix} -> (gw, oif) [scope link]
723 * |
724 * |-> {link prefix} -> (gw, oif) [scope local]
725 * |
726 * |-> {local prefix} (terminal node)
727 */
728static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
729 struct fib_nh *nh)
730{
731 int err = 0;
732 struct net *net;
733 struct net_device *dev;
734
735 net = cfg->fc_nlinfo.nl_net;
736 if (nh->nh_gw) {
737 struct fib_result res;
738
739 if (nh->nh_flags & RTNH_F_ONLINK) {
740 unsigned int addr_type;
741
742 if (cfg->fc_scope >= RT_SCOPE_LINK)
743 return -EINVAL;
744 dev = __dev_get_by_index(net, nh->nh_oif);
745 if (!dev)
746 return -ENODEV;
747 if (!(dev->flags & IFF_UP))
748 return -ENETDOWN;
749 addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
750 if (addr_type != RTN_UNICAST)
751 return -EINVAL;
752 if (!netif_carrier_ok(dev))
753 nh->nh_flags |= RTNH_F_LINKDOWN;
754 nh->nh_dev = dev;
755 dev_hold(dev);
756 nh->nh_scope = RT_SCOPE_LINK;
757 return 0;
758 }
759 rcu_read_lock();
760 {
761 struct fib_table *tbl = NULL;
762 struct flowi4 fl4 = {
763 .daddr = nh->nh_gw,
764 .flowi4_scope = cfg->fc_scope + 1,
765 .flowi4_oif = nh->nh_oif,
766 .flowi4_iif = LOOPBACK_IFINDEX,
767 };
768
769 /* It is not necessary, but requires a bit of thinking */
770 if (fl4.flowi4_scope < RT_SCOPE_LINK)
771 fl4.flowi4_scope = RT_SCOPE_LINK;
772
773 if (cfg->fc_table)
774 tbl = fib_get_table(net, cfg->fc_table);
775
776 if (tbl)
777 err = fib_table_lookup(tbl, &fl4, &res,
778 FIB_LOOKUP_IGNORE_LINKSTATE |
779 FIB_LOOKUP_NOREF);
780
781 /* on error or if no table given do full lookup. This
782 * is needed for example when nexthops are in the local
783 * table rather than the given table
784 */
785 if (!tbl || err) {
786 err = fib_lookup(net, &fl4, &res,
787 FIB_LOOKUP_IGNORE_LINKSTATE);
788 }
789
790 if (err) {
791 rcu_read_unlock();
792 return err;
793 }
794 }
795 err = -EINVAL;
796 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
797 goto out;
798 nh->nh_scope = res.scope;
799 nh->nh_oif = FIB_RES_OIF(res);
800 nh->nh_dev = dev = FIB_RES_DEV(res);
801 if (!dev)
802 goto out;
803 dev_hold(dev);
804 if (!netif_carrier_ok(dev))
805 nh->nh_flags |= RTNH_F_LINKDOWN;
806 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
807 } else {
808 struct in_device *in_dev;
809
810 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
811 return -EINVAL;
812
813 rcu_read_lock();
814 err = -ENODEV;
815 in_dev = inetdev_by_index(net, nh->nh_oif);
816 if (!in_dev)
817 goto out;
818 err = -ENETDOWN;
819 if (!(in_dev->dev->flags & IFF_UP))
820 goto out;
821 nh->nh_dev = in_dev->dev;
822 dev_hold(nh->nh_dev);
823 nh->nh_scope = RT_SCOPE_HOST;
824 if (!netif_carrier_ok(nh->nh_dev))
825 nh->nh_flags |= RTNH_F_LINKDOWN;
826 err = 0;
827 }
828out:
829 rcu_read_unlock();
830 return err;
831}
832
833static inline unsigned int fib_laddr_hashfn(__be32 val)
834{
835 unsigned int mask = (fib_info_hash_size - 1);
836
837 return ((__force u32)val ^
838 ((__force u32)val >> 7) ^
839 ((__force u32)val >> 14)) & mask;
840}
841
842static struct hlist_head *fib_info_hash_alloc(int bytes)
843{
844 if (bytes <= PAGE_SIZE)
845 return kzalloc(bytes, GFP_KERNEL);
846 else
847 return (struct hlist_head *)
848 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
849 get_order(bytes));
850}
851
852static void fib_info_hash_free(struct hlist_head *hash, int bytes)
853{
854 if (!hash)
855 return;
856
857 if (bytes <= PAGE_SIZE)
858 kfree(hash);
859 else
860 free_pages((unsigned long) hash, get_order(bytes));
861}
862
863static void fib_info_hash_move(struct hlist_head *new_info_hash,
864 struct hlist_head *new_laddrhash,
865 unsigned int new_size)
866{
867 struct hlist_head *old_info_hash, *old_laddrhash;
868 unsigned int old_size = fib_info_hash_size;
869 unsigned int i, bytes;
870
871 spin_lock_bh(&fib_info_lock);
872 old_info_hash = fib_info_hash;
873 old_laddrhash = fib_info_laddrhash;
874 fib_info_hash_size = new_size;
875
876 for (i = 0; i < old_size; i++) {
877 struct hlist_head *head = &fib_info_hash[i];
878 struct hlist_node *n;
879 struct fib_info *fi;
880
881 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
882 struct hlist_head *dest;
883 unsigned int new_hash;
884
885 new_hash = fib_info_hashfn(fi);
886 dest = &new_info_hash[new_hash];
887 hlist_add_head(&fi->fib_hash, dest);
888 }
889 }
890 fib_info_hash = new_info_hash;
891
892 for (i = 0; i < old_size; i++) {
893 struct hlist_head *lhead = &fib_info_laddrhash[i];
894 struct hlist_node *n;
895 struct fib_info *fi;
896
897 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
898 struct hlist_head *ldest;
899 unsigned int new_hash;
900
901 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
902 ldest = &new_laddrhash[new_hash];
903 hlist_add_head(&fi->fib_lhash, ldest);
904 }
905 }
906 fib_info_laddrhash = new_laddrhash;
907
908 spin_unlock_bh(&fib_info_lock);
909
910 bytes = old_size * sizeof(struct hlist_head *);
911 fib_info_hash_free(old_info_hash, bytes);
912 fib_info_hash_free(old_laddrhash, bytes);
913}
914
915__be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
916{
917 nh->nh_saddr = inet_select_addr(nh->nh_dev,
918 nh->nh_gw,
919 nh->nh_parent->fib_scope);
920 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
921
922 return nh->nh_saddr;
923}
924
925static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
926{
927 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
928 fib_prefsrc != cfg->fc_dst) {
929 u32 tb_id = cfg->fc_table;
930 int rc;
931
932 if (tb_id == RT_TABLE_MAIN)
933 tb_id = RT_TABLE_LOCAL;
934
935 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
936 fib_prefsrc, tb_id);
937
938 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
939 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
940 fib_prefsrc, RT_TABLE_LOCAL);
941 }
942
943 if (rc != RTN_LOCAL)
944 return false;
945 }
946 return true;
947}
948
949static int
950fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
951{
952 bool ecn_ca = false;
953 struct nlattr *nla;
954 int remaining;
955
956 if (!cfg->fc_mx)
957 return 0;
958
959 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
960 int type = nla_type(nla);
961 u32 val;
962
963 if (!type)
964 continue;
965 if (type > RTAX_MAX)
966 return -EINVAL;
967
968 if (type == RTAX_CC_ALGO) {
969 char tmp[TCP_CA_NAME_MAX];
970
971 nla_strlcpy(tmp, nla, sizeof(tmp));
972 val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
973 if (val == TCP_CA_UNSPEC)
974 return -EINVAL;
975 } else {
976 val = nla_get_u32(nla);
977 }
978 if (type == RTAX_ADVMSS && val > 65535 - 40)
979 val = 65535 - 40;
980 if (type == RTAX_MTU && val > 65535 - 15)
981 val = 65535 - 15;
982 if (type == RTAX_HOPLIMIT && val > 255)
983 val = 255;
984 if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
985 return -EINVAL;
986 fi->fib_metrics[type - 1] = val;
987 }
988
989 if (ecn_ca)
990 fi->fib_metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
991
992 return 0;
993}
994
995struct fib_info *fib_create_info(struct fib_config *cfg)
996{
997 int err;
998 struct fib_info *fi = NULL;
999 struct fib_info *ofi;
1000 int nhs = 1;
1001 struct net *net = cfg->fc_nlinfo.nl_net;
1002
1003 if (cfg->fc_type > RTN_MAX)
1004 goto err_inval;
1005
1006 /* Fast check to catch the most weird cases */
1007 if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
1008 goto err_inval;
1009
1010 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
1011 goto err_inval;
1012
1013#ifdef CONFIG_IP_ROUTE_MULTIPATH
1014 if (cfg->fc_mp) {
1015 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
1016 if (nhs == 0)
1017 goto err_inval;
1018 }
1019#endif
1020
1021 err = -ENOBUFS;
1022 if (fib_info_cnt >= fib_info_hash_size) {
1023 unsigned int new_size = fib_info_hash_size << 1;
1024 struct hlist_head *new_info_hash;
1025 struct hlist_head *new_laddrhash;
1026 unsigned int bytes;
1027
1028 if (!new_size)
1029 new_size = 16;
1030 bytes = new_size * sizeof(struct hlist_head *);
1031 new_info_hash = fib_info_hash_alloc(bytes);
1032 new_laddrhash = fib_info_hash_alloc(bytes);
1033 if (!new_info_hash || !new_laddrhash) {
1034 fib_info_hash_free(new_info_hash, bytes);
1035 fib_info_hash_free(new_laddrhash, bytes);
1036 } else
1037 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1038
1039 if (!fib_info_hash_size)
1040 goto failure;
1041 }
1042
1043 fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1044 if (!fi)
1045 goto failure;
1046 fib_info_cnt++;
1047 if (cfg->fc_mx) {
1048 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
1049 if (!fi->fib_metrics)
1050 goto failure;
1051 } else
1052 fi->fib_metrics = (u32 *) dst_default_metrics;
1053
1054 fi->fib_net = net;
1055 fi->fib_protocol = cfg->fc_protocol;
1056 fi->fib_scope = cfg->fc_scope;
1057 fi->fib_flags = cfg->fc_flags;
1058 fi->fib_priority = cfg->fc_priority;
1059 fi->fib_prefsrc = cfg->fc_prefsrc;
1060 fi->fib_type = cfg->fc_type;
1061 fi->fib_tb_id = cfg->fc_table;
1062
1063 fi->fib_nhs = nhs;
1064 change_nexthops(fi) {
1065 nexthop_nh->nh_parent = fi;
1066 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1067 if (!nexthop_nh->nh_pcpu_rth_output)
1068 goto failure;
1069 } endfor_nexthops(fi)
1070
1071 err = fib_convert_metrics(fi, cfg);
1072 if (err)
1073 goto failure;
1074
1075 if (cfg->fc_mp) {
1076#ifdef CONFIG_IP_ROUTE_MULTIPATH
1077 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
1078 if (err != 0)
1079 goto failure;
1080 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
1081 goto err_inval;
1082 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
1083 goto err_inval;
1084#ifdef CONFIG_IP_ROUTE_CLASSID
1085 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
1086 goto err_inval;
1087#endif
1088#else
1089 goto err_inval;
1090#endif
1091 } else {
1092 struct fib_nh *nh = fi->fib_nh;
1093
1094 if (cfg->fc_encap) {
1095 struct lwtunnel_state *lwtstate;
1096 struct net_device *dev = NULL;
1097
1098 if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
1099 goto err_inval;
1100 if (cfg->fc_oif)
1101 dev = __dev_get_by_index(net, cfg->fc_oif);
1102 err = lwtunnel_build_state(dev, cfg->fc_encap_type,
1103 cfg->fc_encap, AF_INET, cfg,
1104 &lwtstate);
1105 if (err)
1106 goto failure;
1107
1108 nh->nh_lwtstate = lwtstate_get(lwtstate);
1109 }
1110 nh->nh_oif = cfg->fc_oif;
1111 nh->nh_gw = cfg->fc_gw;
1112 nh->nh_flags = cfg->fc_flags;
1113#ifdef CONFIG_IP_ROUTE_CLASSID
1114 nh->nh_tclassid = cfg->fc_flow;
1115 if (nh->nh_tclassid)
1116 fi->fib_net->ipv4.fib_num_tclassid_users++;
1117#endif
1118#ifdef CONFIG_IP_ROUTE_MULTIPATH
1119 nh->nh_weight = 1;
1120#endif
1121 }
1122
1123 if (fib_props[cfg->fc_type].error) {
1124 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
1125 goto err_inval;
1126 goto link_it;
1127 } else {
1128 switch (cfg->fc_type) {
1129 case RTN_UNICAST:
1130 case RTN_LOCAL:
1131 case RTN_BROADCAST:
1132 case RTN_ANYCAST:
1133 case RTN_MULTICAST:
1134 break;
1135 default:
1136 goto err_inval;
1137 }
1138 }
1139
1140 if (cfg->fc_scope > RT_SCOPE_HOST)
1141 goto err_inval;
1142
1143 if (cfg->fc_scope == RT_SCOPE_HOST) {
1144 struct fib_nh *nh = fi->fib_nh;
1145
1146 /* Local address is added. */
1147 if (nhs != 1 || nh->nh_gw)
1148 goto err_inval;
1149 nh->nh_scope = RT_SCOPE_NOWHERE;
1150 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1151 err = -ENODEV;
1152 if (!nh->nh_dev)
1153 goto failure;
1154 } else {
1155 int linkdown = 0;
1156
1157 change_nexthops(fi) {
1158 err = fib_check_nh(cfg, fi, nexthop_nh);
1159 if (err != 0)
1160 goto failure;
1161 if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1162 linkdown++;
1163 } endfor_nexthops(fi)
1164 if (linkdown == fi->fib_nhs)
1165 fi->fib_flags |= RTNH_F_LINKDOWN;
1166 }
1167
1168 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc))
1169 goto err_inval;
1170
1171 change_nexthops(fi) {
1172 fib_info_update_nh_saddr(net, nexthop_nh);
1173 fib_add_weight(fi, nexthop_nh);
1174 } endfor_nexthops(fi)
1175
1176 fib_rebalance(fi);
1177
1178link_it:
1179 ofi = fib_find_info(fi);
1180 if (ofi) {
1181 fi->fib_dead = 1;
1182 free_fib_info(fi);
1183 ofi->fib_treeref++;
1184 return ofi;
1185 }
1186
1187 fi->fib_treeref++;
1188 atomic_inc(&fi->fib_clntref);
1189 spin_lock_bh(&fib_info_lock);
1190 hlist_add_head(&fi->fib_hash,
1191 &fib_info_hash[fib_info_hashfn(fi)]);
1192 if (fi->fib_prefsrc) {
1193 struct hlist_head *head;
1194
1195 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1196 hlist_add_head(&fi->fib_lhash, head);
1197 }
1198 change_nexthops(fi) {
1199 struct hlist_head *head;
1200 unsigned int hash;
1201
1202 if (!nexthop_nh->nh_dev)
1203 continue;
1204 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1205 head = &fib_info_devhash[hash];
1206 hlist_add_head(&nexthop_nh->nh_hash, head);
1207 } endfor_nexthops(fi)
1208 spin_unlock_bh(&fib_info_lock);
1209 return fi;
1210
1211err_inval:
1212 err = -EINVAL;
1213
1214failure:
1215 if (fi) {
1216 fi->fib_dead = 1;
1217 free_fib_info(fi);
1218 }
1219
1220 return ERR_PTR(err);
1221}
1222
1223int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1224 u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1225 struct fib_info *fi, unsigned int flags)
1226{
1227 struct nlmsghdr *nlh;
1228 struct rtmsg *rtm;
1229
1230 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1231 if (!nlh)
1232 return -EMSGSIZE;
1233
1234 rtm = nlmsg_data(nlh);
1235 rtm->rtm_family = AF_INET;
1236 rtm->rtm_dst_len = dst_len;
1237 rtm->rtm_src_len = 0;
1238 rtm->rtm_tos = tos;
1239 if (tb_id < 256)
1240 rtm->rtm_table = tb_id;
1241 else
1242 rtm->rtm_table = RT_TABLE_COMPAT;
1243 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1244 goto nla_put_failure;
1245 rtm->rtm_type = type;
1246 rtm->rtm_flags = fi->fib_flags;
1247 rtm->rtm_scope = fi->fib_scope;
1248 rtm->rtm_protocol = fi->fib_protocol;
1249
1250 if (rtm->rtm_dst_len &&
1251 nla_put_in_addr(skb, RTA_DST, dst))
1252 goto nla_put_failure;
1253 if (fi->fib_priority &&
1254 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1255 goto nla_put_failure;
1256 if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1257 goto nla_put_failure;
1258
1259 if (fi->fib_prefsrc &&
1260 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1261 goto nla_put_failure;
1262 if (fi->fib_nhs == 1) {
1263 struct in_device *in_dev;
1264
1265 if (fi->fib_nh->nh_gw &&
1266 nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1267 goto nla_put_failure;
1268 if (fi->fib_nh->nh_oif &&
1269 nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1270 goto nla_put_failure;
1271 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1272 in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1273 if (in_dev &&
1274 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1275 rtm->rtm_flags |= RTNH_F_DEAD;
1276 }
1277#ifdef CONFIG_IP_ROUTE_CLASSID
1278 if (fi->fib_nh[0].nh_tclassid &&
1279 nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1280 goto nla_put_failure;
1281#endif
1282 if (fi->fib_nh->nh_lwtstate &&
1283 lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate) < 0)
1284 goto nla_put_failure;
1285 }
1286#ifdef CONFIG_IP_ROUTE_MULTIPATH
1287 if (fi->fib_nhs > 1) {
1288 struct rtnexthop *rtnh;
1289 struct nlattr *mp;
1290
1291 mp = nla_nest_start(skb, RTA_MULTIPATH);
1292 if (!mp)
1293 goto nla_put_failure;
1294
1295 for_nexthops(fi) {
1296 struct in_device *in_dev;
1297
1298 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1299 if (!rtnh)
1300 goto nla_put_failure;
1301
1302 rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1303 if (nh->nh_flags & RTNH_F_LINKDOWN) {
1304 in_dev = __in_dev_get_rtnl(nh->nh_dev);
1305 if (in_dev &&
1306 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1307 rtnh->rtnh_flags |= RTNH_F_DEAD;
1308 }
1309 rtnh->rtnh_hops = nh->nh_weight - 1;
1310 rtnh->rtnh_ifindex = nh->nh_oif;
1311
1312 if (nh->nh_gw &&
1313 nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1314 goto nla_put_failure;
1315#ifdef CONFIG_IP_ROUTE_CLASSID
1316 if (nh->nh_tclassid &&
1317 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1318 goto nla_put_failure;
1319#endif
1320 if (nh->nh_lwtstate &&
1321 lwtunnel_fill_encap(skb, nh->nh_lwtstate) < 0)
1322 goto nla_put_failure;
1323
1324 /* length of rtnetlink header + attributes */
1325 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1326 } endfor_nexthops(fi);
1327
1328 nla_nest_end(skb, mp);
1329 }
1330#endif
1331 nlmsg_end(skb, nlh);
1332 return 0;
1333
1334nla_put_failure:
1335 nlmsg_cancel(skb, nlh);
1336 return -EMSGSIZE;
1337}
1338
1339/*
1340 * Update FIB if:
1341 * - local address disappeared -> we must delete all the entries
1342 * referring to it.
1343 * - device went down -> we must shutdown all nexthops going via it.
1344 */
1345int fib_sync_down_addr(struct net_device *dev, __be32 local)
1346{
1347 int ret = 0;
1348 unsigned int hash = fib_laddr_hashfn(local);
1349 struct hlist_head *head = &fib_info_laddrhash[hash];
1350 struct net *net = dev_net(dev);
1351 int tb_id = l3mdev_fib_table(dev);
1352 struct fib_info *fi;
1353
1354 if (!fib_info_laddrhash || local == 0)
1355 return 0;
1356
1357 hlist_for_each_entry(fi, head, fib_lhash) {
1358 if (!net_eq(fi->fib_net, net) ||
1359 fi->fib_tb_id != tb_id)
1360 continue;
1361 if (fi->fib_prefsrc == local) {
1362 fi->fib_flags |= RTNH_F_DEAD;
1363 ret++;
1364 }
1365 }
1366 return ret;
1367}
1368
1369/* Event force Flags Description
1370 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host
1371 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host
1372 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed
1373 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed
1374 */
1375int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1376{
1377 int ret = 0;
1378 int scope = RT_SCOPE_NOWHERE;
1379 struct fib_info *prev_fi = NULL;
1380 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1381 struct hlist_head *head = &fib_info_devhash[hash];
1382 struct fib_nh *nh;
1383
1384 if (force)
1385 scope = -1;
1386
1387 hlist_for_each_entry(nh, head, nh_hash) {
1388 struct fib_info *fi = nh->nh_parent;
1389 int dead;
1390
1391 BUG_ON(!fi->fib_nhs);
1392 if (nh->nh_dev != dev || fi == prev_fi)
1393 continue;
1394 prev_fi = fi;
1395 dead = 0;
1396 change_nexthops(fi) {
1397 if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1398 dead++;
1399 else if (nexthop_nh->nh_dev == dev &&
1400 nexthop_nh->nh_scope != scope) {
1401 switch (event) {
1402 case NETDEV_DOWN:
1403 case NETDEV_UNREGISTER:
1404 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1405 /* fall through */
1406 case NETDEV_CHANGE:
1407 nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1408 break;
1409 }
1410 dead++;
1411 }
1412#ifdef CONFIG_IP_ROUTE_MULTIPATH
1413 if (event == NETDEV_UNREGISTER &&
1414 nexthop_nh->nh_dev == dev) {
1415 dead = fi->fib_nhs;
1416 break;
1417 }
1418#endif
1419 } endfor_nexthops(fi)
1420 if (dead == fi->fib_nhs) {
1421 switch (event) {
1422 case NETDEV_DOWN:
1423 case NETDEV_UNREGISTER:
1424 fi->fib_flags |= RTNH_F_DEAD;
1425 /* fall through */
1426 case NETDEV_CHANGE:
1427 fi->fib_flags |= RTNH_F_LINKDOWN;
1428 break;
1429 }
1430 ret++;
1431 }
1432
1433 fib_rebalance(fi);
1434 }
1435
1436 return ret;
1437}
1438
1439/* Must be invoked inside of an RCU protected region. */
1440void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1441{
1442 struct fib_info *fi = NULL, *last_resort = NULL;
1443 struct hlist_head *fa_head = res->fa_head;
1444 struct fib_table *tb = res->table;
1445 u8 slen = 32 - res->prefixlen;
1446 int order = -1, last_idx = -1;
1447 struct fib_alias *fa, *fa1 = NULL;
1448 u32 last_prio = res->fi->fib_priority;
1449 u8 last_tos = 0;
1450
1451 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1452 struct fib_info *next_fi = fa->fa_info;
1453
1454 if (fa->fa_slen != slen)
1455 continue;
1456 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1457 continue;
1458 if (fa->tb_id != tb->tb_id)
1459 continue;
1460 if (next_fi->fib_priority > last_prio &&
1461 fa->fa_tos == last_tos) {
1462 if (last_tos)
1463 continue;
1464 break;
1465 }
1466 if (next_fi->fib_flags & RTNH_F_DEAD)
1467 continue;
1468 last_tos = fa->fa_tos;
1469 last_prio = next_fi->fib_priority;
1470
1471 if (next_fi->fib_scope != res->scope ||
1472 fa->fa_type != RTN_UNICAST)
1473 continue;
1474 if (!next_fi->fib_nh[0].nh_gw ||
1475 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1476 continue;
1477
1478 fib_alias_accessed(fa);
1479
1480 if (!fi) {
1481 if (next_fi != res->fi)
1482 break;
1483 fa1 = fa;
1484 } else if (!fib_detect_death(fi, order, &last_resort,
1485 &last_idx, fa1->fa_default)) {
1486 fib_result_assign(res, fi);
1487 fa1->fa_default = order;
1488 goto out;
1489 }
1490 fi = next_fi;
1491 order++;
1492 }
1493
1494 if (order <= 0 || !fi) {
1495 if (fa1)
1496 fa1->fa_default = -1;
1497 goto out;
1498 }
1499
1500 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1501 fa1->fa_default)) {
1502 fib_result_assign(res, fi);
1503 fa1->fa_default = order;
1504 goto out;
1505 }
1506
1507 if (last_idx >= 0)
1508 fib_result_assign(res, last_resort);
1509 fa1->fa_default = last_idx;
1510out:
1511 return;
1512}
1513
1514/*
1515 * Dead device goes up. We wake up dead nexthops.
1516 * It takes sense only on multipath routes.
1517 */
1518int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1519{
1520 struct fib_info *prev_fi;
1521 unsigned int hash;
1522 struct hlist_head *head;
1523 struct fib_nh *nh;
1524 int ret;
1525
1526 if (!(dev->flags & IFF_UP))
1527 return 0;
1528
1529 if (nh_flags & RTNH_F_DEAD) {
1530 unsigned int flags = dev_get_flags(dev);
1531
1532 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1533 nh_flags |= RTNH_F_LINKDOWN;
1534 }
1535
1536 prev_fi = NULL;
1537 hash = fib_devindex_hashfn(dev->ifindex);
1538 head = &fib_info_devhash[hash];
1539 ret = 0;
1540
1541 hlist_for_each_entry(nh, head, nh_hash) {
1542 struct fib_info *fi = nh->nh_parent;
1543 int alive;
1544
1545 BUG_ON(!fi->fib_nhs);
1546 if (nh->nh_dev != dev || fi == prev_fi)
1547 continue;
1548
1549 prev_fi = fi;
1550 alive = 0;
1551 change_nexthops(fi) {
1552 if (!(nexthop_nh->nh_flags & nh_flags)) {
1553 alive++;
1554 continue;
1555 }
1556 if (!nexthop_nh->nh_dev ||
1557 !(nexthop_nh->nh_dev->flags & IFF_UP))
1558 continue;
1559 if (nexthop_nh->nh_dev != dev ||
1560 !__in_dev_get_rtnl(dev))
1561 continue;
1562 alive++;
1563 nexthop_nh->nh_flags &= ~nh_flags;
1564 } endfor_nexthops(fi)
1565
1566 if (alive > 0) {
1567 fi->fib_flags &= ~nh_flags;
1568 ret++;
1569 }
1570
1571 fib_rebalance(fi);
1572 }
1573
1574 return ret;
1575}
1576
1577#ifdef CONFIG_IP_ROUTE_MULTIPATH
1578static bool fib_good_nh(const struct fib_nh *nh)
1579{
1580 int state = NUD_REACHABLE;
1581
1582 if (nh->nh_scope == RT_SCOPE_LINK) {
1583 struct neighbour *n;
1584
1585 rcu_read_lock_bh();
1586
1587 n = __ipv4_neigh_lookup_noref(nh->nh_dev,
1588 (__force u32)nh->nh_gw);
1589 if (n)
1590 state = n->nud_state;
1591
1592 rcu_read_unlock_bh();
1593 }
1594
1595 return !!(state & NUD_VALID);
1596}
1597
1598void fib_select_multipath(struct fib_result *res, int hash)
1599{
1600 struct fib_info *fi = res->fi;
1601 struct net *net = fi->fib_net;
1602 bool first = false;
1603
1604 for_nexthops(fi) {
1605 if (hash > atomic_read(&nh->nh_upper_bound))
1606 continue;
1607
1608 if (!net->ipv4.sysctl_fib_multipath_use_neigh ||
1609 fib_good_nh(nh)) {
1610 res->nh_sel = nhsel;
1611 return;
1612 }
1613 if (!first) {
1614 res->nh_sel = nhsel;
1615 first = true;
1616 }
1617 } endfor_nexthops(fi);
1618}
1619#endif
1620
1621void fib_select_path(struct net *net, struct fib_result *res,
1622 struct flowi4 *fl4, int mp_hash)
1623{
1624 bool oif_check;
1625
1626 oif_check = (fl4->flowi4_oif == 0 ||
1627 fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF);
1628
1629#ifdef CONFIG_IP_ROUTE_MULTIPATH
1630 if (res->fi->fib_nhs > 1 && oif_check) {
1631 if (mp_hash < 0)
1632 mp_hash = get_hash_from_flowi4(fl4) >> 1;
1633
1634 fib_select_multipath(res, mp_hash);
1635 }
1636 else
1637#endif
1638 if (!res->prefixlen &&
1639 res->table->tb_num_default > 1 &&
1640 res->type == RTN_UNICAST && oif_check)
1641 fib_select_default(fl4, res);
1642
1643 if (!fl4->saddr)
1644 fl4->saddr = FIB_RES_PREFSRC(net, *res);
1645}
1646EXPORT_SYMBOL_GPL(fib_select_path);