]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - net/ipv4/fib_semantics.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net: Fix up inet_addr_type checks
[mirror_ubuntu-bionic-kernel.git] / net / ipv4 / fib_semantics.c
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
49 static DEFINE_SPINLOCK(fib_info_lock);
50 static struct hlist_head *fib_info_hash;
51 static struct hlist_head *fib_info_laddrhash;
52 static unsigned int fib_info_hash_size;
53 static unsigned int fib_info_cnt;
54
55 #define DEVINDEX_HASHBITS 8
56 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
57 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
58
59 #ifdef CONFIG_IP_ROUTE_MULTIPATH
60
61 static DEFINE_SPINLOCK(fib_multipath_lock);
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
93 const 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
144 static 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_free(&rt->dst);
157 }
158
159 static void free_nh_exceptions(struct fib_nh *nh)
160 {
161 struct fnhe_hash_bucket *hash;
162 int i;
163
164 hash = rcu_dereference_protected(nh->nh_exceptions, 1);
165 if (!hash)
166 return;
167 for (i = 0; i < FNHE_HASH_SIZE; i++) {
168 struct fib_nh_exception *fnhe;
169
170 fnhe = rcu_dereference_protected(hash[i].chain, 1);
171 while (fnhe) {
172 struct fib_nh_exception *next;
173
174 next = rcu_dereference_protected(fnhe->fnhe_next, 1);
175
176 rt_fibinfo_free(&fnhe->fnhe_rth_input);
177 rt_fibinfo_free(&fnhe->fnhe_rth_output);
178
179 kfree(fnhe);
180
181 fnhe = next;
182 }
183 }
184 kfree(hash);
185 }
186
187 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
188 {
189 int cpu;
190
191 if (!rtp)
192 return;
193
194 for_each_possible_cpu(cpu) {
195 struct rtable *rt;
196
197 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
198 if (rt)
199 dst_free(&rt->dst);
200 }
201 free_percpu(rtp);
202 }
203
204 /* Release a nexthop info record */
205 static void free_fib_info_rcu(struct rcu_head *head)
206 {
207 struct fib_info *fi = container_of(head, struct fib_info, rcu);
208
209 change_nexthops(fi) {
210 if (nexthop_nh->nh_dev)
211 dev_put(nexthop_nh->nh_dev);
212 lwtstate_put(nexthop_nh->nh_lwtstate);
213 free_nh_exceptions(nexthop_nh);
214 rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
215 rt_fibinfo_free(&nexthop_nh->nh_rth_input);
216 } endfor_nexthops(fi);
217
218 if (fi->fib_metrics != (u32 *) dst_default_metrics)
219 kfree(fi->fib_metrics);
220 kfree(fi);
221 }
222
223 void free_fib_info(struct fib_info *fi)
224 {
225 if (fi->fib_dead == 0) {
226 pr_warn("Freeing alive fib_info %p\n", fi);
227 return;
228 }
229 fib_info_cnt--;
230 #ifdef CONFIG_IP_ROUTE_CLASSID
231 change_nexthops(fi) {
232 if (nexthop_nh->nh_tclassid)
233 fi->fib_net->ipv4.fib_num_tclassid_users--;
234 } endfor_nexthops(fi);
235 #endif
236 call_rcu(&fi->rcu, free_fib_info_rcu);
237 }
238
239 void 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
257 static 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
279 static 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
288 static 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
303 static 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 */
335 int 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
359 static 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
401 void 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;
425 errout:
426 if (err < 0)
427 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
428 }
429
430 static 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
458 static 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
471 static 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 nexthop_nh->nh_flags =
484 (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
485 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
486 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
487
488 attrlen = rtnh_attrlen(rtnh);
489 if (attrlen > 0) {
490 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
491
492 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
493 nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
494 #ifdef CONFIG_IP_ROUTE_CLASSID
495 nla = nla_find(attrs, attrlen, RTA_FLOW);
496 nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
497 if (nexthop_nh->nh_tclassid)
498 fi->fib_net->ipv4.fib_num_tclassid_users++;
499 #endif
500 nla = nla_find(attrs, attrlen, RTA_ENCAP);
501 if (nla) {
502 struct lwtunnel_state *lwtstate;
503 struct net_device *dev = NULL;
504 struct nlattr *nla_entype;
505
506 nla_entype = nla_find(attrs, attrlen,
507 RTA_ENCAP_TYPE);
508 if (!nla_entype)
509 goto err_inval;
510 if (cfg->fc_oif)
511 dev = __dev_get_by_index(net, cfg->fc_oif);
512 ret = lwtunnel_build_state(dev, nla_get_u16(
513 nla_entype),
514 nla, &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
527 err_inval:
528 ret = -EINVAL;
529
530 errout:
531 return ret;
532 }
533
534 #endif
535
536 int fib_encap_match(struct net *net, u16 encap_type,
537 struct nlattr *encap,
538 int oif, const struct fib_nh *nh)
539 {
540 struct lwtunnel_state *lwtstate;
541 struct net_device *dev = NULL;
542 int ret;
543
544 if (encap_type == LWTUNNEL_ENCAP_NONE)
545 return 0;
546
547 if (oif)
548 dev = __dev_get_by_index(net, oif);
549 ret = lwtunnel_build_state(dev, encap_type,
550 encap, &lwtstate);
551 if (!ret)
552 return lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
553
554 return 0;
555 }
556
557 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
558 {
559 struct net *net = cfg->fc_nlinfo.nl_net;
560 #ifdef CONFIG_IP_ROUTE_MULTIPATH
561 struct rtnexthop *rtnh;
562 int remaining;
563 #endif
564
565 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
566 return 1;
567
568 if (cfg->fc_oif || cfg->fc_gw) {
569 if (cfg->fc_encap) {
570 if (fib_encap_match(net, cfg->fc_encap_type,
571 cfg->fc_encap, cfg->fc_oif,
572 fi->fib_nh))
573 return 1;
574 }
575 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
576 (!cfg->fc_gw || cfg->fc_gw == fi->fib_nh->nh_gw))
577 return 0;
578 return 1;
579 }
580
581 #ifdef CONFIG_IP_ROUTE_MULTIPATH
582 if (!cfg->fc_mp)
583 return 0;
584
585 rtnh = cfg->fc_mp;
586 remaining = cfg->fc_mp_len;
587
588 for_nexthops(fi) {
589 int attrlen;
590
591 if (!rtnh_ok(rtnh, remaining))
592 return -EINVAL;
593
594 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
595 return 1;
596
597 attrlen = rtnh_attrlen(rtnh);
598 if (attrlen > 0) {
599 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
600
601 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
602 if (nla && nla_get_in_addr(nla) != nh->nh_gw)
603 return 1;
604 #ifdef CONFIG_IP_ROUTE_CLASSID
605 nla = nla_find(attrs, attrlen, RTA_FLOW);
606 if (nla && nla_get_u32(nla) != nh->nh_tclassid)
607 return 1;
608 #endif
609 }
610
611 rtnh = rtnh_next(rtnh, &remaining);
612 } endfor_nexthops(fi);
613 #endif
614 return 0;
615 }
616
617
618 /*
619 * Picture
620 * -------
621 *
622 * Semantics of nexthop is very messy by historical reasons.
623 * We have to take into account, that:
624 * a) gateway can be actually local interface address,
625 * so that gatewayed route is direct.
626 * b) gateway must be on-link address, possibly
627 * described not by an ifaddr, but also by a direct route.
628 * c) If both gateway and interface are specified, they should not
629 * contradict.
630 * d) If we use tunnel routes, gateway could be not on-link.
631 *
632 * Attempt to reconcile all of these (alas, self-contradictory) conditions
633 * results in pretty ugly and hairy code with obscure logic.
634 *
635 * I chose to generalized it instead, so that the size
636 * of code does not increase practically, but it becomes
637 * much more general.
638 * Every prefix is assigned a "scope" value: "host" is local address,
639 * "link" is direct route,
640 * [ ... "site" ... "interior" ... ]
641 * and "universe" is true gateway route with global meaning.
642 *
643 * Every prefix refers to a set of "nexthop"s (gw, oif),
644 * where gw must have narrower scope. This recursion stops
645 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
646 * which means that gw is forced to be on link.
647 *
648 * Code is still hairy, but now it is apparently logically
649 * consistent and very flexible. F.e. as by-product it allows
650 * to co-exists in peace independent exterior and interior
651 * routing processes.
652 *
653 * Normally it looks as following.
654 *
655 * {universe prefix} -> (gw, oif) [scope link]
656 * |
657 * |-> {link prefix} -> (gw, oif) [scope local]
658 * |
659 * |-> {local prefix} (terminal node)
660 */
661 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
662 struct fib_nh *nh)
663 {
664 int err;
665 struct net *net;
666 struct net_device *dev;
667
668 net = cfg->fc_nlinfo.nl_net;
669 if (nh->nh_gw) {
670 struct fib_result res;
671
672 if (nh->nh_flags & RTNH_F_ONLINK) {
673 unsigned int addr_type;
674
675 if (cfg->fc_scope >= RT_SCOPE_LINK)
676 return -EINVAL;
677 dev = __dev_get_by_index(net, nh->nh_oif);
678 if (!dev)
679 return -ENODEV;
680 if (!(dev->flags & IFF_UP))
681 return -ENETDOWN;
682 addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
683 if (addr_type != RTN_UNICAST)
684 return -EINVAL;
685 if (!netif_carrier_ok(dev))
686 nh->nh_flags |= RTNH_F_LINKDOWN;
687 nh->nh_dev = dev;
688 dev_hold(dev);
689 nh->nh_scope = RT_SCOPE_LINK;
690 return 0;
691 }
692 rcu_read_lock();
693 {
694 struct flowi4 fl4 = {
695 .daddr = nh->nh_gw,
696 .flowi4_scope = cfg->fc_scope + 1,
697 .flowi4_oif = nh->nh_oif,
698 .flowi4_iif = LOOPBACK_IFINDEX,
699 };
700
701 /* It is not necessary, but requires a bit of thinking */
702 if (fl4.flowi4_scope < RT_SCOPE_LINK)
703 fl4.flowi4_scope = RT_SCOPE_LINK;
704 err = fib_lookup(net, &fl4, &res,
705 FIB_LOOKUP_IGNORE_LINKSTATE);
706 if (err) {
707 rcu_read_unlock();
708 return err;
709 }
710 }
711 err = -EINVAL;
712 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
713 goto out;
714 nh->nh_scope = res.scope;
715 nh->nh_oif = FIB_RES_OIF(res);
716 nh->nh_dev = dev = FIB_RES_DEV(res);
717 if (!dev)
718 goto out;
719 dev_hold(dev);
720 if (!netif_carrier_ok(dev))
721 nh->nh_flags |= RTNH_F_LINKDOWN;
722 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
723 } else {
724 struct in_device *in_dev;
725
726 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
727 return -EINVAL;
728
729 rcu_read_lock();
730 err = -ENODEV;
731 in_dev = inetdev_by_index(net, nh->nh_oif);
732 if (!in_dev)
733 goto out;
734 err = -ENETDOWN;
735 if (!(in_dev->dev->flags & IFF_UP))
736 goto out;
737 nh->nh_dev = in_dev->dev;
738 dev_hold(nh->nh_dev);
739 nh->nh_scope = RT_SCOPE_HOST;
740 if (!netif_carrier_ok(nh->nh_dev))
741 nh->nh_flags |= RTNH_F_LINKDOWN;
742 err = 0;
743 }
744 out:
745 rcu_read_unlock();
746 return err;
747 }
748
749 static inline unsigned int fib_laddr_hashfn(__be32 val)
750 {
751 unsigned int mask = (fib_info_hash_size - 1);
752
753 return ((__force u32)val ^
754 ((__force u32)val >> 7) ^
755 ((__force u32)val >> 14)) & mask;
756 }
757
758 static struct hlist_head *fib_info_hash_alloc(int bytes)
759 {
760 if (bytes <= PAGE_SIZE)
761 return kzalloc(bytes, GFP_KERNEL);
762 else
763 return (struct hlist_head *)
764 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
765 get_order(bytes));
766 }
767
768 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
769 {
770 if (!hash)
771 return;
772
773 if (bytes <= PAGE_SIZE)
774 kfree(hash);
775 else
776 free_pages((unsigned long) hash, get_order(bytes));
777 }
778
779 static void fib_info_hash_move(struct hlist_head *new_info_hash,
780 struct hlist_head *new_laddrhash,
781 unsigned int new_size)
782 {
783 struct hlist_head *old_info_hash, *old_laddrhash;
784 unsigned int old_size = fib_info_hash_size;
785 unsigned int i, bytes;
786
787 spin_lock_bh(&fib_info_lock);
788 old_info_hash = fib_info_hash;
789 old_laddrhash = fib_info_laddrhash;
790 fib_info_hash_size = new_size;
791
792 for (i = 0; i < old_size; i++) {
793 struct hlist_head *head = &fib_info_hash[i];
794 struct hlist_node *n;
795 struct fib_info *fi;
796
797 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
798 struct hlist_head *dest;
799 unsigned int new_hash;
800
801 new_hash = fib_info_hashfn(fi);
802 dest = &new_info_hash[new_hash];
803 hlist_add_head(&fi->fib_hash, dest);
804 }
805 }
806 fib_info_hash = new_info_hash;
807
808 for (i = 0; i < old_size; i++) {
809 struct hlist_head *lhead = &fib_info_laddrhash[i];
810 struct hlist_node *n;
811 struct fib_info *fi;
812
813 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
814 struct hlist_head *ldest;
815 unsigned int new_hash;
816
817 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
818 ldest = &new_laddrhash[new_hash];
819 hlist_add_head(&fi->fib_lhash, ldest);
820 }
821 }
822 fib_info_laddrhash = new_laddrhash;
823
824 spin_unlock_bh(&fib_info_lock);
825
826 bytes = old_size * sizeof(struct hlist_head *);
827 fib_info_hash_free(old_info_hash, bytes);
828 fib_info_hash_free(old_laddrhash, bytes);
829 }
830
831 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
832 {
833 nh->nh_saddr = inet_select_addr(nh->nh_dev,
834 nh->nh_gw,
835 nh->nh_parent->fib_scope);
836 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
837
838 return nh->nh_saddr;
839 }
840
841 struct fib_info *fib_create_info(struct fib_config *cfg)
842 {
843 int err;
844 struct fib_info *fi = NULL;
845 struct fib_info *ofi;
846 int nhs = 1;
847 struct net *net = cfg->fc_nlinfo.nl_net;
848
849 if (cfg->fc_type > RTN_MAX)
850 goto err_inval;
851
852 /* Fast check to catch the most weird cases */
853 if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
854 goto err_inval;
855
856 #ifdef CONFIG_IP_ROUTE_MULTIPATH
857 if (cfg->fc_mp) {
858 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
859 if (nhs == 0)
860 goto err_inval;
861 }
862 #endif
863
864 err = -ENOBUFS;
865 if (fib_info_cnt >= fib_info_hash_size) {
866 unsigned int new_size = fib_info_hash_size << 1;
867 struct hlist_head *new_info_hash;
868 struct hlist_head *new_laddrhash;
869 unsigned int bytes;
870
871 if (!new_size)
872 new_size = 16;
873 bytes = new_size * sizeof(struct hlist_head *);
874 new_info_hash = fib_info_hash_alloc(bytes);
875 new_laddrhash = fib_info_hash_alloc(bytes);
876 if (!new_info_hash || !new_laddrhash) {
877 fib_info_hash_free(new_info_hash, bytes);
878 fib_info_hash_free(new_laddrhash, bytes);
879 } else
880 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
881
882 if (!fib_info_hash_size)
883 goto failure;
884 }
885
886 fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
887 if (!fi)
888 goto failure;
889 fib_info_cnt++;
890 if (cfg->fc_mx) {
891 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
892 if (!fi->fib_metrics)
893 goto failure;
894 } else
895 fi->fib_metrics = (u32 *) dst_default_metrics;
896
897 fi->fib_net = net;
898 fi->fib_protocol = cfg->fc_protocol;
899 fi->fib_scope = cfg->fc_scope;
900 fi->fib_flags = cfg->fc_flags;
901 fi->fib_priority = cfg->fc_priority;
902 fi->fib_prefsrc = cfg->fc_prefsrc;
903 fi->fib_type = cfg->fc_type;
904
905 fi->fib_nhs = nhs;
906 change_nexthops(fi) {
907 nexthop_nh->nh_parent = fi;
908 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
909 if (!nexthop_nh->nh_pcpu_rth_output)
910 goto failure;
911 } endfor_nexthops(fi)
912
913 if (cfg->fc_mx) {
914 struct nlattr *nla;
915 int remaining;
916
917 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
918 int type = nla_type(nla);
919
920 if (type) {
921 u32 val;
922
923 if (type > RTAX_MAX)
924 goto err_inval;
925 if (type == RTAX_CC_ALGO) {
926 char tmp[TCP_CA_NAME_MAX];
927
928 nla_strlcpy(tmp, nla, sizeof(tmp));
929 val = tcp_ca_get_key_by_name(tmp);
930 if (val == TCP_CA_UNSPEC)
931 goto err_inval;
932 } else {
933 val = nla_get_u32(nla);
934 }
935 if (type == RTAX_ADVMSS && val > 65535 - 40)
936 val = 65535 - 40;
937 if (type == RTAX_MTU && val > 65535 - 15)
938 val = 65535 - 15;
939 fi->fib_metrics[type - 1] = val;
940 }
941 }
942 }
943
944 if (cfg->fc_mp) {
945 #ifdef CONFIG_IP_ROUTE_MULTIPATH
946 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
947 if (err != 0)
948 goto failure;
949 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
950 goto err_inval;
951 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
952 goto err_inval;
953 #ifdef CONFIG_IP_ROUTE_CLASSID
954 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
955 goto err_inval;
956 #endif
957 #else
958 goto err_inval;
959 #endif
960 } else {
961 struct fib_nh *nh = fi->fib_nh;
962
963 if (cfg->fc_encap) {
964 struct lwtunnel_state *lwtstate;
965 struct net_device *dev = NULL;
966
967 if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
968 goto err_inval;
969 if (cfg->fc_oif)
970 dev = __dev_get_by_index(net, cfg->fc_oif);
971 err = lwtunnel_build_state(dev, cfg->fc_encap_type,
972 cfg->fc_encap, &lwtstate);
973 if (err)
974 goto failure;
975
976 nh->nh_lwtstate = lwtstate_get(lwtstate);
977 }
978 nh->nh_oif = cfg->fc_oif;
979 nh->nh_gw = cfg->fc_gw;
980 nh->nh_flags = cfg->fc_flags;
981 #ifdef CONFIG_IP_ROUTE_CLASSID
982 nh->nh_tclassid = cfg->fc_flow;
983 if (nh->nh_tclassid)
984 fi->fib_net->ipv4.fib_num_tclassid_users++;
985 #endif
986 #ifdef CONFIG_IP_ROUTE_MULTIPATH
987 nh->nh_weight = 1;
988 #endif
989 }
990
991 if (fib_props[cfg->fc_type].error) {
992 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
993 goto err_inval;
994 goto link_it;
995 } else {
996 switch (cfg->fc_type) {
997 case RTN_UNICAST:
998 case RTN_LOCAL:
999 case RTN_BROADCAST:
1000 case RTN_ANYCAST:
1001 case RTN_MULTICAST:
1002 break;
1003 default:
1004 goto err_inval;
1005 }
1006 }
1007
1008 if (cfg->fc_scope > RT_SCOPE_HOST)
1009 goto err_inval;
1010
1011 if (cfg->fc_scope == RT_SCOPE_HOST) {
1012 struct fib_nh *nh = fi->fib_nh;
1013
1014 /* Local address is added. */
1015 if (nhs != 1 || nh->nh_gw)
1016 goto err_inval;
1017 nh->nh_scope = RT_SCOPE_NOWHERE;
1018 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1019 err = -ENODEV;
1020 if (!nh->nh_dev)
1021 goto failure;
1022 } else {
1023 int linkdown = 0;
1024
1025 change_nexthops(fi) {
1026 err = fib_check_nh(cfg, fi, nexthop_nh);
1027 if (err != 0)
1028 goto failure;
1029 if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1030 linkdown++;
1031 } endfor_nexthops(fi)
1032 if (linkdown == fi->fib_nhs)
1033 fi->fib_flags |= RTNH_F_LINKDOWN;
1034 }
1035
1036 if (fi->fib_prefsrc) {
1037 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1038 fi->fib_prefsrc != cfg->fc_dst)
1039 if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
1040 goto err_inval;
1041 }
1042
1043 change_nexthops(fi) {
1044 fib_info_update_nh_saddr(net, nexthop_nh);
1045 } endfor_nexthops(fi)
1046
1047 link_it:
1048 ofi = fib_find_info(fi);
1049 if (ofi) {
1050 fi->fib_dead = 1;
1051 free_fib_info(fi);
1052 ofi->fib_treeref++;
1053 return ofi;
1054 }
1055
1056 fi->fib_treeref++;
1057 atomic_inc(&fi->fib_clntref);
1058 spin_lock_bh(&fib_info_lock);
1059 hlist_add_head(&fi->fib_hash,
1060 &fib_info_hash[fib_info_hashfn(fi)]);
1061 if (fi->fib_prefsrc) {
1062 struct hlist_head *head;
1063
1064 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1065 hlist_add_head(&fi->fib_lhash, head);
1066 }
1067 change_nexthops(fi) {
1068 struct hlist_head *head;
1069 unsigned int hash;
1070
1071 if (!nexthop_nh->nh_dev)
1072 continue;
1073 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1074 head = &fib_info_devhash[hash];
1075 hlist_add_head(&nexthop_nh->nh_hash, head);
1076 } endfor_nexthops(fi)
1077 spin_unlock_bh(&fib_info_lock);
1078 return fi;
1079
1080 err_inval:
1081 err = -EINVAL;
1082
1083 failure:
1084 if (fi) {
1085 fi->fib_dead = 1;
1086 free_fib_info(fi);
1087 }
1088
1089 return ERR_PTR(err);
1090 }
1091
1092 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1093 u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1094 struct fib_info *fi, unsigned int flags)
1095 {
1096 struct nlmsghdr *nlh;
1097 struct rtmsg *rtm;
1098
1099 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1100 if (!nlh)
1101 return -EMSGSIZE;
1102
1103 rtm = nlmsg_data(nlh);
1104 rtm->rtm_family = AF_INET;
1105 rtm->rtm_dst_len = dst_len;
1106 rtm->rtm_src_len = 0;
1107 rtm->rtm_tos = tos;
1108 if (tb_id < 256)
1109 rtm->rtm_table = tb_id;
1110 else
1111 rtm->rtm_table = RT_TABLE_COMPAT;
1112 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1113 goto nla_put_failure;
1114 rtm->rtm_type = type;
1115 rtm->rtm_flags = fi->fib_flags;
1116 rtm->rtm_scope = fi->fib_scope;
1117 rtm->rtm_protocol = fi->fib_protocol;
1118
1119 if (rtm->rtm_dst_len &&
1120 nla_put_in_addr(skb, RTA_DST, dst))
1121 goto nla_put_failure;
1122 if (fi->fib_priority &&
1123 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1124 goto nla_put_failure;
1125 if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1126 goto nla_put_failure;
1127
1128 if (fi->fib_prefsrc &&
1129 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1130 goto nla_put_failure;
1131 if (fi->fib_nhs == 1) {
1132 struct in_device *in_dev;
1133
1134 if (fi->fib_nh->nh_gw &&
1135 nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1136 goto nla_put_failure;
1137 if (fi->fib_nh->nh_oif &&
1138 nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1139 goto nla_put_failure;
1140 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1141 in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1142 if (in_dev &&
1143 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1144 rtm->rtm_flags |= RTNH_F_DEAD;
1145 }
1146 #ifdef CONFIG_IP_ROUTE_CLASSID
1147 if (fi->fib_nh[0].nh_tclassid &&
1148 nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1149 goto nla_put_failure;
1150 #endif
1151 if (fi->fib_nh->nh_lwtstate)
1152 lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate);
1153 }
1154 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1155 if (fi->fib_nhs > 1) {
1156 struct rtnexthop *rtnh;
1157 struct nlattr *mp;
1158
1159 mp = nla_nest_start(skb, RTA_MULTIPATH);
1160 if (!mp)
1161 goto nla_put_failure;
1162
1163 for_nexthops(fi) {
1164 struct in_device *in_dev;
1165
1166 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1167 if (!rtnh)
1168 goto nla_put_failure;
1169
1170 rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1171 if (nh->nh_flags & RTNH_F_LINKDOWN) {
1172 in_dev = __in_dev_get_rtnl(nh->nh_dev);
1173 if (in_dev &&
1174 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1175 rtnh->rtnh_flags |= RTNH_F_DEAD;
1176 }
1177 rtnh->rtnh_hops = nh->nh_weight - 1;
1178 rtnh->rtnh_ifindex = nh->nh_oif;
1179
1180 if (nh->nh_gw &&
1181 nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1182 goto nla_put_failure;
1183 #ifdef CONFIG_IP_ROUTE_CLASSID
1184 if (nh->nh_tclassid &&
1185 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1186 goto nla_put_failure;
1187 #endif
1188 if (nh->nh_lwtstate)
1189 lwtunnel_fill_encap(skb, nh->nh_lwtstate);
1190 /* length of rtnetlink header + attributes */
1191 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1192 } endfor_nexthops(fi);
1193
1194 nla_nest_end(skb, mp);
1195 }
1196 #endif
1197 nlmsg_end(skb, nlh);
1198 return 0;
1199
1200 nla_put_failure:
1201 nlmsg_cancel(skb, nlh);
1202 return -EMSGSIZE;
1203 }
1204
1205 /*
1206 * Update FIB if:
1207 * - local address disappeared -> we must delete all the entries
1208 * referring to it.
1209 * - device went down -> we must shutdown all nexthops going via it.
1210 */
1211 int fib_sync_down_addr(struct net *net, __be32 local)
1212 {
1213 int ret = 0;
1214 unsigned int hash = fib_laddr_hashfn(local);
1215 struct hlist_head *head = &fib_info_laddrhash[hash];
1216 struct fib_info *fi;
1217
1218 if (!fib_info_laddrhash || local == 0)
1219 return 0;
1220
1221 hlist_for_each_entry(fi, head, fib_lhash) {
1222 if (!net_eq(fi->fib_net, net))
1223 continue;
1224 if (fi->fib_prefsrc == local) {
1225 fi->fib_flags |= RTNH_F_DEAD;
1226 ret++;
1227 }
1228 }
1229 return ret;
1230 }
1231
1232 int fib_sync_down_dev(struct net_device *dev, unsigned long event)
1233 {
1234 int ret = 0;
1235 int scope = RT_SCOPE_NOWHERE;
1236 struct fib_info *prev_fi = NULL;
1237 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1238 struct hlist_head *head = &fib_info_devhash[hash];
1239 struct fib_nh *nh;
1240
1241 if (event == NETDEV_UNREGISTER ||
1242 event == NETDEV_DOWN)
1243 scope = -1;
1244
1245 hlist_for_each_entry(nh, head, nh_hash) {
1246 struct fib_info *fi = nh->nh_parent;
1247 int dead;
1248
1249 BUG_ON(!fi->fib_nhs);
1250 if (nh->nh_dev != dev || fi == prev_fi)
1251 continue;
1252 prev_fi = fi;
1253 dead = 0;
1254 change_nexthops(fi) {
1255 if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1256 dead++;
1257 else if (nexthop_nh->nh_dev == dev &&
1258 nexthop_nh->nh_scope != scope) {
1259 switch (event) {
1260 case NETDEV_DOWN:
1261 case NETDEV_UNREGISTER:
1262 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1263 /* fall through */
1264 case NETDEV_CHANGE:
1265 nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1266 break;
1267 }
1268 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1269 spin_lock_bh(&fib_multipath_lock);
1270 fi->fib_power -= nexthop_nh->nh_power;
1271 nexthop_nh->nh_power = 0;
1272 spin_unlock_bh(&fib_multipath_lock);
1273 #endif
1274 dead++;
1275 }
1276 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1277 if (event == NETDEV_UNREGISTER &&
1278 nexthop_nh->nh_dev == dev) {
1279 dead = fi->fib_nhs;
1280 break;
1281 }
1282 #endif
1283 } endfor_nexthops(fi)
1284 if (dead == fi->fib_nhs) {
1285 switch (event) {
1286 case NETDEV_DOWN:
1287 case NETDEV_UNREGISTER:
1288 fi->fib_flags |= RTNH_F_DEAD;
1289 /* fall through */
1290 case NETDEV_CHANGE:
1291 fi->fib_flags |= RTNH_F_LINKDOWN;
1292 break;
1293 }
1294 ret++;
1295 }
1296 }
1297
1298 return ret;
1299 }
1300
1301 /* Must be invoked inside of an RCU protected region. */
1302 void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1303 {
1304 struct fib_info *fi = NULL, *last_resort = NULL;
1305 struct hlist_head *fa_head = res->fa_head;
1306 struct fib_table *tb = res->table;
1307 u8 slen = 32 - res->prefixlen;
1308 int order = -1, last_idx = -1;
1309 struct fib_alias *fa, *fa1 = NULL;
1310 u32 last_prio = res->fi->fib_priority;
1311 u8 last_tos = 0;
1312
1313 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1314 struct fib_info *next_fi = fa->fa_info;
1315
1316 if (fa->fa_slen != slen)
1317 continue;
1318 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1319 continue;
1320 if (fa->tb_id != tb->tb_id)
1321 continue;
1322 if (next_fi->fib_priority > last_prio &&
1323 fa->fa_tos == last_tos) {
1324 if (last_tos)
1325 continue;
1326 break;
1327 }
1328 if (next_fi->fib_flags & RTNH_F_DEAD)
1329 continue;
1330 last_tos = fa->fa_tos;
1331 last_prio = next_fi->fib_priority;
1332
1333 if (next_fi->fib_scope != res->scope ||
1334 fa->fa_type != RTN_UNICAST)
1335 continue;
1336 if (!next_fi->fib_nh[0].nh_gw ||
1337 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1338 continue;
1339
1340 fib_alias_accessed(fa);
1341
1342 if (!fi) {
1343 if (next_fi != res->fi)
1344 break;
1345 fa1 = fa;
1346 } else if (!fib_detect_death(fi, order, &last_resort,
1347 &last_idx, fa1->fa_default)) {
1348 fib_result_assign(res, fi);
1349 fa1->fa_default = order;
1350 goto out;
1351 }
1352 fi = next_fi;
1353 order++;
1354 }
1355
1356 if (order <= 0 || !fi) {
1357 if (fa1)
1358 fa1->fa_default = -1;
1359 goto out;
1360 }
1361
1362 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1363 fa1->fa_default)) {
1364 fib_result_assign(res, fi);
1365 fa1->fa_default = order;
1366 goto out;
1367 }
1368
1369 if (last_idx >= 0)
1370 fib_result_assign(res, last_resort);
1371 fa1->fa_default = last_idx;
1372 out:
1373 return;
1374 }
1375
1376 /*
1377 * Dead device goes up. We wake up dead nexthops.
1378 * It takes sense only on multipath routes.
1379 */
1380 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1381 {
1382 struct fib_info *prev_fi;
1383 unsigned int hash;
1384 struct hlist_head *head;
1385 struct fib_nh *nh;
1386 int ret;
1387
1388 if (!(dev->flags & IFF_UP))
1389 return 0;
1390
1391 prev_fi = NULL;
1392 hash = fib_devindex_hashfn(dev->ifindex);
1393 head = &fib_info_devhash[hash];
1394 ret = 0;
1395
1396 hlist_for_each_entry(nh, head, nh_hash) {
1397 struct fib_info *fi = nh->nh_parent;
1398 int alive;
1399
1400 BUG_ON(!fi->fib_nhs);
1401 if (nh->nh_dev != dev || fi == prev_fi)
1402 continue;
1403
1404 prev_fi = fi;
1405 alive = 0;
1406 change_nexthops(fi) {
1407 if (!(nexthop_nh->nh_flags & nh_flags)) {
1408 alive++;
1409 continue;
1410 }
1411 if (!nexthop_nh->nh_dev ||
1412 !(nexthop_nh->nh_dev->flags & IFF_UP))
1413 continue;
1414 if (nexthop_nh->nh_dev != dev ||
1415 !__in_dev_get_rtnl(dev))
1416 continue;
1417 alive++;
1418 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1419 spin_lock_bh(&fib_multipath_lock);
1420 nexthop_nh->nh_power = 0;
1421 nexthop_nh->nh_flags &= ~nh_flags;
1422 spin_unlock_bh(&fib_multipath_lock);
1423 #else
1424 nexthop_nh->nh_flags &= ~nh_flags;
1425 #endif
1426 } endfor_nexthops(fi)
1427
1428 if (alive > 0) {
1429 fi->fib_flags &= ~nh_flags;
1430 ret++;
1431 }
1432 }
1433
1434 return ret;
1435 }
1436
1437 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1438
1439 /*
1440 * The algorithm is suboptimal, but it provides really
1441 * fair weighted route distribution.
1442 */
1443 void fib_select_multipath(struct fib_result *res)
1444 {
1445 struct fib_info *fi = res->fi;
1446 struct in_device *in_dev;
1447 int w;
1448
1449 spin_lock_bh(&fib_multipath_lock);
1450 if (fi->fib_power <= 0) {
1451 int power = 0;
1452 change_nexthops(fi) {
1453 in_dev = __in_dev_get_rcu(nexthop_nh->nh_dev);
1454 if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1455 continue;
1456 if (in_dev &&
1457 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1458 nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1459 continue;
1460 power += nexthop_nh->nh_weight;
1461 nexthop_nh->nh_power = nexthop_nh->nh_weight;
1462 } endfor_nexthops(fi);
1463 fi->fib_power = power;
1464 if (power <= 0) {
1465 spin_unlock_bh(&fib_multipath_lock);
1466 /* Race condition: route has just become dead. */
1467 res->nh_sel = 0;
1468 return;
1469 }
1470 }
1471
1472
1473 /* w should be random number [0..fi->fib_power-1],
1474 * it is pretty bad approximation.
1475 */
1476
1477 w = jiffies % fi->fib_power;
1478
1479 change_nexthops(fi) {
1480 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1481 nexthop_nh->nh_power) {
1482 w -= nexthop_nh->nh_power;
1483 if (w <= 0) {
1484 nexthop_nh->nh_power--;
1485 fi->fib_power--;
1486 res->nh_sel = nhsel;
1487 spin_unlock_bh(&fib_multipath_lock);
1488 return;
1489 }
1490 }
1491 } endfor_nexthops(fi);
1492
1493 /* Race condition: route has just become dead. */
1494 res->nh_sel = 0;
1495 spin_unlock_bh(&fib_multipath_lock);
1496 }
1497 #endif
ubuntu-bionic-kernel mirror