~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/ipv4/fib_semantics.c

Version: ~ [ linux-5.15-rc1 ] ~ [ linux-5.14.5 ] ~ [ linux-5.13.18 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.66 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.147 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.206 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.246 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.282 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.283 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 
 46 #include "fib_lookup.h"
 47 
 48 static DEFINE_SPINLOCK(fib_info_lock);
 49 static struct hlist_head *fib_info_hash;
 50 static struct hlist_head *fib_info_laddrhash;
 51 static unsigned int fib_info_hash_size;
 52 static unsigned int fib_info_cnt;
 53 
 54 #define DEVINDEX_HASHBITS 8
 55 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
 56 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
 57 
 58 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 59 
 60 static DEFINE_SPINLOCK(fib_multipath_lock);
 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 
 92 const 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 
143 static 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 
158 static 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 
186 static 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 */
204 static 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                 free_nh_exceptions(nexthop_nh);
212                 rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
213                 rt_fibinfo_free(&nexthop_nh->nh_rth_input);
214         } endfor_nexthops(fi);
215 
216         if (fi->fib_metrics != (u32 *) dst_default_metrics)
217                 kfree(fi->fib_metrics);
218         kfree(fi);
219 }
220 
221 void free_fib_info(struct fib_info *fi)
222 {
223         if (fi->fib_dead == 0) {
224                 pr_warn("Freeing alive fib_info %p\n", fi);
225                 return;
226         }
227         fib_info_cnt--;
228 #ifdef CONFIG_IP_ROUTE_CLASSID
229         change_nexthops(fi) {
230                 if (nexthop_nh->nh_tclassid)
231                         fi->fib_net->ipv4.fib_num_tclassid_users--;
232         } endfor_nexthops(fi);
233 #endif
234         call_rcu(&fi->rcu, free_fib_info_rcu);
235 }
236 
237 void fib_release_info(struct fib_info *fi)
238 {
239         spin_lock_bh(&fib_info_lock);
240         if (fi && --fi->fib_treeref == 0) {
241                 hlist_del(&fi->fib_hash);
242                 if (fi->fib_prefsrc)
243                         hlist_del(&fi->fib_lhash);
244                 change_nexthops(fi) {
245                         if (!nexthop_nh->nh_dev)
246                                 continue;
247                         hlist_del(&nexthop_nh->nh_hash);
248                 } endfor_nexthops(fi)
249                 fi->fib_dead = 1;
250                 fib_info_put(fi);
251         }
252         spin_unlock_bh(&fib_info_lock);
253 }
254 
255 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
256 {
257         const struct fib_nh *onh = ofi->fib_nh;
258 
259         for_nexthops(fi) {
260                 if (nh->nh_oif != onh->nh_oif ||
261                     nh->nh_gw  != onh->nh_gw ||
262                     nh->nh_scope != onh->nh_scope ||
263 #ifdef CONFIG_IP_ROUTE_MULTIPATH
264                     nh->nh_weight != onh->nh_weight ||
265 #endif
266 #ifdef CONFIG_IP_ROUTE_CLASSID
267                     nh->nh_tclassid != onh->nh_tclassid ||
268 #endif
269                     ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
270                         return -1;
271                 onh++;
272         } endfor_nexthops(fi);
273         return 0;
274 }
275 
276 static inline unsigned int fib_devindex_hashfn(unsigned int val)
277 {
278         unsigned int mask = DEVINDEX_HASHSIZE - 1;
279 
280         return (val ^
281                 (val >> DEVINDEX_HASHBITS) ^
282                 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
283 }
284 
285 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
286 {
287         unsigned int mask = (fib_info_hash_size - 1);
288         unsigned int val = fi->fib_nhs;
289 
290         val ^= (fi->fib_protocol << 8) | fi->fib_scope;
291         val ^= (__force u32)fi->fib_prefsrc;
292         val ^= fi->fib_priority;
293         for_nexthops(fi) {
294                 val ^= fib_devindex_hashfn(nh->nh_oif);
295         } endfor_nexthops(fi)
296 
297         return (val ^ (val >> 7) ^ (val >> 12)) & mask;
298 }
299 
300 static struct fib_info *fib_find_info(const struct fib_info *nfi)
301 {
302         struct hlist_head *head;
303         struct fib_info *fi;
304         unsigned int hash;
305 
306         hash = fib_info_hashfn(nfi);
307         head = &fib_info_hash[hash];
308 
309         hlist_for_each_entry(fi, head, fib_hash) {
310                 if (!net_eq(fi->fib_net, nfi->fib_net))
311                         continue;
312                 if (fi->fib_nhs != nfi->fib_nhs)
313                         continue;
314                 if (nfi->fib_protocol == fi->fib_protocol &&
315                     nfi->fib_scope == fi->fib_scope &&
316                     nfi->fib_prefsrc == fi->fib_prefsrc &&
317                     nfi->fib_priority == fi->fib_priority &&
318                     nfi->fib_type == fi->fib_type &&
319                     memcmp(nfi->fib_metrics, fi->fib_metrics,
320                            sizeof(u32) * RTAX_MAX) == 0 &&
321                     !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
322                     (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
323                         return fi;
324         }
325 
326         return NULL;
327 }
328 
329 /* Check, that the gateway is already configured.
330  * Used only by redirect accept routine.
331  */
332 int ip_fib_check_default(__be32 gw, struct net_device *dev)
333 {
334         struct hlist_head *head;
335         struct fib_nh *nh;
336         unsigned int hash;
337 
338         spin_lock(&fib_info_lock);
339 
340         hash = fib_devindex_hashfn(dev->ifindex);
341         head = &fib_info_devhash[hash];
342         hlist_for_each_entry(nh, head, nh_hash) {
343                 if (nh->nh_dev == dev &&
344                     nh->nh_gw == gw &&
345                     !(nh->nh_flags & RTNH_F_DEAD)) {
346                         spin_unlock(&fib_info_lock);
347                         return 0;
348                 }
349         }
350 
351         spin_unlock(&fib_info_lock);
352 
353         return -1;
354 }
355 
356 static inline size_t fib_nlmsg_size(struct fib_info *fi)
357 {
358         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
359                          + nla_total_size(4) /* RTA_TABLE */
360                          + nla_total_size(4) /* RTA_DST */
361                          + nla_total_size(4) /* RTA_PRIORITY */
362                          + nla_total_size(4) /* RTA_PREFSRC */
363                          + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
364 
365         /* space for nested metrics */
366         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
367 
368         if (fi->fib_nhs) {
369                 /* Also handles the special case fib_nhs == 1 */
370 
371                 /* each nexthop is packed in an attribute */
372                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
373 
374                 /* may contain flow and gateway attribute */
375                 nhsize += 2 * nla_total_size(4);
376 
377                 /* all nexthops are packed in a nested attribute */
378                 payload += nla_total_size(fi->fib_nhs * nhsize);
379         }
380 
381         return payload;
382 }
383 
384 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
385                int dst_len, u32 tb_id, const struct nl_info *info,
386                unsigned int nlm_flags)
387 {
388         struct sk_buff *skb;
389         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
390         int err = -ENOBUFS;
391 
392         skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
393         if (!skb)
394                 goto errout;
395 
396         err = fib_dump_info(skb, info->portid, seq, event, tb_id,
397                             fa->fa_type, key, dst_len,
398                             fa->fa_tos, fa->fa_info, nlm_flags);
399         if (err < 0) {
400                 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
401                 WARN_ON(err == -EMSGSIZE);
402                 kfree_skb(skb);
403                 goto errout;
404         }
405         rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
406                     info->nlh, GFP_KERNEL);
407         return;
408 errout:
409         if (err < 0)
410                 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
411 }
412 
413 static int fib_detect_death(struct fib_info *fi, int order,
414                             struct fib_info **last_resort, int *last_idx,
415                             int dflt)
416 {
417         struct neighbour *n;
418         int state = NUD_NONE;
419 
420         n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
421         if (n) {
422                 state = n->nud_state;
423                 neigh_release(n);
424         }
425         if (state == NUD_REACHABLE)
426                 return 0;
427         if ((state & NUD_VALID) && order != dflt)
428                 return 0;
429         if ((state & NUD_VALID) ||
430             (*last_idx < 0 && order > dflt)) {
431                 *last_resort = fi;
432                 *last_idx = order;
433         }
434         return 1;
435 }
436 
437 #ifdef CONFIG_IP_ROUTE_MULTIPATH
438 
439 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
440 {
441         int nhs = 0;
442 
443         while (rtnh_ok(rtnh, remaining)) {
444                 nhs++;
445                 rtnh = rtnh_next(rtnh, &remaining);
446         }
447 
448         /* leftover implies invalid nexthop configuration, discard it */
449         return remaining > 0 ? 0 : nhs;
450 }
451 
452 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
453                        int remaining, struct fib_config *cfg)
454 {
455         change_nexthops(fi) {
456                 int attrlen;
457 
458                 if (!rtnh_ok(rtnh, remaining))
459                         return -EINVAL;
460 
461                 nexthop_nh->nh_flags =
462                         (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
463                 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
464                 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
465 
466                 attrlen = rtnh_attrlen(rtnh);
467                 if (attrlen > 0) {
468                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
469 
470                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
471                         nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
472 #ifdef CONFIG_IP_ROUTE_CLASSID
473                         nla = nla_find(attrs, attrlen, RTA_FLOW);
474                         nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
475                         if (nexthop_nh->nh_tclassid)
476                                 fi->fib_net->ipv4.fib_num_tclassid_users++;
477 #endif
478                 }
479 
480                 rtnh = rtnh_next(rtnh, &remaining);
481         } endfor_nexthops(fi);
482 
483         return 0;
484 }
485 
486 #endif
487 
488 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
489 {
490 #ifdef CONFIG_IP_ROUTE_MULTIPATH
491         struct rtnexthop *rtnh;
492         int remaining;
493 #endif
494 
495         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
496                 return 1;
497 
498         if (cfg->fc_oif || cfg->fc_gw) {
499                 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
500                     (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
501                         return 0;
502                 return 1;
503         }
504 
505 #ifdef CONFIG_IP_ROUTE_MULTIPATH
506         if (!cfg->fc_mp)
507                 return 0;
508 
509         rtnh = cfg->fc_mp;
510         remaining = cfg->fc_mp_len;
511 
512         for_nexthops(fi) {
513                 int attrlen;
514 
515                 if (!rtnh_ok(rtnh, remaining))
516                         return -EINVAL;
517 
518                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
519                         return 1;
520 
521                 attrlen = rtnh_attrlen(rtnh);
522                 if (attrlen > 0) {
523                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
524 
525                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
526                         if (nla && nla_get_in_addr(nla) != nh->nh_gw)
527                                 return 1;
528 #ifdef CONFIG_IP_ROUTE_CLASSID
529                         nla = nla_find(attrs, attrlen, RTA_FLOW);
530                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
531                                 return 1;
532 #endif
533                 }
534 
535                 rtnh = rtnh_next(rtnh, &remaining);
536         } endfor_nexthops(fi);
537 #endif
538         return 0;
539 }
540 
541 
542 /*
543  * Picture
544  * -------
545  *
546  * Semantics of nexthop is very messy by historical reasons.
547  * We have to take into account, that:
548  * a) gateway can be actually local interface address,
549  *    so that gatewayed route is direct.
550  * b) gateway must be on-link address, possibly
551  *    described not by an ifaddr, but also by a direct route.
552  * c) If both gateway and interface are specified, they should not
553  *    contradict.
554  * d) If we use tunnel routes, gateway could be not on-link.
555  *
556  * Attempt to reconcile all of these (alas, self-contradictory) conditions
557  * results in pretty ugly and hairy code with obscure logic.
558  *
559  * I chose to generalized it instead, so that the size
560  * of code does not increase practically, but it becomes
561  * much more general.
562  * Every prefix is assigned a "scope" value: "host" is local address,
563  * "link" is direct route,
564  * [ ... "site" ... "interior" ... ]
565  * and "universe" is true gateway route with global meaning.
566  *
567  * Every prefix refers to a set of "nexthop"s (gw, oif),
568  * where gw must have narrower scope. This recursion stops
569  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
570  * which means that gw is forced to be on link.
571  *
572  * Code is still hairy, but now it is apparently logically
573  * consistent and very flexible. F.e. as by-product it allows
574  * to co-exists in peace independent exterior and interior
575  * routing processes.
576  *
577  * Normally it looks as following.
578  *
579  * {universe prefix}  -> (gw, oif) [scope link]
580  *                |
581  *                |-> {link prefix} -> (gw, oif) [scope local]
582  *                                      |
583  *                                      |-> {local prefix} (terminal node)
584  */
585 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
586                         struct fib_nh *nh)
587 {
588         int err;
589         struct net *net;
590         struct net_device *dev;
591 
592         net = cfg->fc_nlinfo.nl_net;
593         if (nh->nh_gw) {
594                 struct fib_result res;
595 
596                 if (nh->nh_flags & RTNH_F_ONLINK) {
597 
598                         if (cfg->fc_scope >= RT_SCOPE_LINK)
599                                 return -EINVAL;
600                         if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
601                                 return -EINVAL;
602                         dev = __dev_get_by_index(net, nh->nh_oif);
603                         if (!dev)
604                                 return -ENODEV;
605                         if (!(dev->flags & IFF_UP))
606                                 return -ENETDOWN;
607                         if (!netif_carrier_ok(dev))
608                                 nh->nh_flags |= RTNH_F_LINKDOWN;
609                         nh->nh_dev = dev;
610                         dev_hold(dev);
611                         nh->nh_scope = RT_SCOPE_LINK;
612                         return 0;
613                 }
614                 rcu_read_lock();
615                 {
616                         struct flowi4 fl4 = {
617                                 .daddr = nh->nh_gw,
618                                 .flowi4_scope = cfg->fc_scope + 1,
619                                 .flowi4_oif = nh->nh_oif,
620                                 .flowi4_iif = LOOPBACK_IFINDEX,
621                         };
622 
623                         /* It is not necessary, but requires a bit of thinking */
624                         if (fl4.flowi4_scope < RT_SCOPE_LINK)
625                                 fl4.flowi4_scope = RT_SCOPE_LINK;
626                         err = fib_lookup(net, &fl4, &res,
627                                          FIB_LOOKUP_IGNORE_LINKSTATE);
628                         if (err) {
629                                 rcu_read_unlock();
630                                 return err;
631                         }
632                 }
633                 err = -EINVAL;
634                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
635                         goto out;
636                 nh->nh_scope = res.scope;
637                 nh->nh_oif = FIB_RES_OIF(res);
638                 nh->nh_dev = dev = FIB_RES_DEV(res);
639                 if (!dev)
640                         goto out;
641                 dev_hold(dev);
642                 if (!netif_carrier_ok(dev))
643                         nh->nh_flags |= RTNH_F_LINKDOWN;
644                 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
645         } else {
646                 struct in_device *in_dev;
647 
648                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
649                         return -EINVAL;
650 
651                 rcu_read_lock();
652                 err = -ENODEV;
653                 in_dev = inetdev_by_index(net, nh->nh_oif);
654                 if (!in_dev)
655                         goto out;
656                 err = -ENETDOWN;
657                 if (!(in_dev->dev->flags & IFF_UP))
658                         goto out;
659                 nh->nh_dev = in_dev->dev;
660                 dev_hold(nh->nh_dev);
661                 nh->nh_scope = RT_SCOPE_HOST;
662                 if (!netif_carrier_ok(nh->nh_dev))
663                         nh->nh_flags |= RTNH_F_LINKDOWN;
664                 err = 0;
665         }
666 out:
667         rcu_read_unlock();
668         return err;
669 }
670 
671 static inline unsigned int fib_laddr_hashfn(__be32 val)
672 {
673         unsigned int mask = (fib_info_hash_size - 1);
674 
675         return ((__force u32)val ^
676                 ((__force u32)val >> 7) ^
677                 ((__force u32)val >> 14)) & mask;
678 }
679 
680 static struct hlist_head *fib_info_hash_alloc(int bytes)
681 {
682         if (bytes <= PAGE_SIZE)
683                 return kzalloc(bytes, GFP_KERNEL);
684         else
685                 return (struct hlist_head *)
686                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
687                                          get_order(bytes));
688 }
689 
690 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
691 {
692         if (!hash)
693                 return;
694 
695         if (bytes <= PAGE_SIZE)
696                 kfree(hash);
697         else
698                 free_pages((unsigned long) hash, get_order(bytes));
699 }
700 
701 static void fib_info_hash_move(struct hlist_head *new_info_hash,
702                                struct hlist_head *new_laddrhash,
703                                unsigned int new_size)
704 {
705         struct hlist_head *old_info_hash, *old_laddrhash;
706         unsigned int old_size = fib_info_hash_size;
707         unsigned int i, bytes;
708 
709         spin_lock_bh(&fib_info_lock);
710         old_info_hash = fib_info_hash;
711         old_laddrhash = fib_info_laddrhash;
712         fib_info_hash_size = new_size;
713 
714         for (i = 0; i < old_size; i++) {
715                 struct hlist_head *head = &fib_info_hash[i];
716                 struct hlist_node *n;
717                 struct fib_info *fi;
718 
719                 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
720                         struct hlist_head *dest;
721                         unsigned int new_hash;
722 
723                         new_hash = fib_info_hashfn(fi);
724                         dest = &new_info_hash[new_hash];
725                         hlist_add_head(&fi->fib_hash, dest);
726                 }
727         }
728         fib_info_hash = new_info_hash;
729 
730         for (i = 0; i < old_size; i++) {
731                 struct hlist_head *lhead = &fib_info_laddrhash[i];
732                 struct hlist_node *n;
733                 struct fib_info *fi;
734 
735                 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
736                         struct hlist_head *ldest;
737                         unsigned int new_hash;
738 
739                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
740                         ldest = &new_laddrhash[new_hash];
741                         hlist_add_head(&fi->fib_lhash, ldest);
742                 }
743         }
744         fib_info_laddrhash = new_laddrhash;
745 
746         spin_unlock_bh(&fib_info_lock);
747 
748         bytes = old_size * sizeof(struct hlist_head *);
749         fib_info_hash_free(old_info_hash, bytes);
750         fib_info_hash_free(old_laddrhash, bytes);
751 }
752 
753 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
754 {
755         nh->nh_saddr = inet_select_addr(nh->nh_dev,
756                                         nh->nh_gw,
757                                         nh->nh_parent->fib_scope);
758         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
759 
760         return nh->nh_saddr;
761 }
762 
763 struct fib_info *fib_create_info(struct fib_config *cfg)
764 {
765         int err;
766         struct fib_info *fi = NULL;
767         struct fib_info *ofi;
768         int nhs = 1;
769         struct net *net = cfg->fc_nlinfo.nl_net;
770 
771         if (cfg->fc_type > RTN_MAX)
772                 goto err_inval;
773 
774         /* Fast check to catch the most weird cases */
775         if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
776                 goto err_inval;
777 
778 #ifdef CONFIG_IP_ROUTE_MULTIPATH
779         if (cfg->fc_mp) {
780                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
781                 if (nhs == 0)
782                         goto err_inval;
783         }
784 #endif
785 
786         err = -ENOBUFS;
787         if (fib_info_cnt >= fib_info_hash_size) {
788                 unsigned int new_size = fib_info_hash_size << 1;
789                 struct hlist_head *new_info_hash;
790                 struct hlist_head *new_laddrhash;
791                 unsigned int bytes;
792 
793                 if (!new_size)
794                         new_size = 16;
795                 bytes = new_size * sizeof(struct hlist_head *);
796                 new_info_hash = fib_info_hash_alloc(bytes);
797                 new_laddrhash = fib_info_hash_alloc(bytes);
798                 if (!new_info_hash || !new_laddrhash) {
799                         fib_info_hash_free(new_info_hash, bytes);
800                         fib_info_hash_free(new_laddrhash, bytes);
801                 } else
802                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
803 
804                 if (!fib_info_hash_size)
805                         goto failure;
806         }
807 
808         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
809         if (!fi)
810                 goto failure;
811         fib_info_cnt++;
812         if (cfg->fc_mx) {
813                 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
814                 if (!fi->fib_metrics)
815                         goto failure;
816         } else
817                 fi->fib_metrics = (u32 *) dst_default_metrics;
818 
819         fi->fib_net = net;
820         fi->fib_protocol = cfg->fc_protocol;
821         fi->fib_scope = cfg->fc_scope;
822         fi->fib_flags = cfg->fc_flags;
823         fi->fib_priority = cfg->fc_priority;
824         fi->fib_prefsrc = cfg->fc_prefsrc;
825         fi->fib_type = cfg->fc_type;
826 
827         fi->fib_nhs = nhs;
828         change_nexthops(fi) {
829                 nexthop_nh->nh_parent = fi;
830                 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
831                 if (!nexthop_nh->nh_pcpu_rth_output)
832                         goto failure;
833         } endfor_nexthops(fi)
834 
835         if (cfg->fc_mx) {
836                 struct nlattr *nla;
837                 int remaining;
838 
839                 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
840                         int type = nla_type(nla);
841 
842                         if (type) {
843                                 u32 val;
844 
845                                 if (type > RTAX_MAX)
846                                         goto err_inval;
847                                 if (type == RTAX_CC_ALGO) {
848                                         char tmp[TCP_CA_NAME_MAX];
849 
850                                         nla_strlcpy(tmp, nla, sizeof(tmp));
851                                         val = tcp_ca_get_key_by_name(tmp);
852                                         if (val == TCP_CA_UNSPEC)
853                                                 goto err_inval;
854                                 } else {
855                                         val = nla_get_u32(nla);
856                                 }
857                                 if (type == RTAX_ADVMSS && val > 65535 - 40)
858                                         val = 65535 - 40;
859                                 if (type == RTAX_MTU && val > 65535 - 15)
860                                         val = 65535 - 15;
861                                 fi->fib_metrics[type - 1] = val;
862                         }
863                 }
864         }
865 
866         if (cfg->fc_mp) {
867 #ifdef CONFIG_IP_ROUTE_MULTIPATH
868                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
869                 if (err != 0)
870                         goto failure;
871                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
872                         goto err_inval;
873                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
874                         goto err_inval;
875 #ifdef CONFIG_IP_ROUTE_CLASSID
876                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
877                         goto err_inval;
878 #endif
879 #else
880                 goto err_inval;
881 #endif
882         } else {
883                 struct fib_nh *nh = fi->fib_nh;
884 
885                 nh->nh_oif = cfg->fc_oif;
886                 nh->nh_gw = cfg->fc_gw;
887                 nh->nh_flags = cfg->fc_flags;
888 #ifdef CONFIG_IP_ROUTE_CLASSID
889                 nh->nh_tclassid = cfg->fc_flow;
890                 if (nh->nh_tclassid)
891                         fi->fib_net->ipv4.fib_num_tclassid_users++;
892 #endif
893 #ifdef CONFIG_IP_ROUTE_MULTIPATH
894                 nh->nh_weight = 1;
895 #endif
896         }
897 
898         if (fib_props[cfg->fc_type].error) {
899                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
900                         goto err_inval;
901                 goto link_it;
902         } else {
903                 switch (cfg->fc_type) {
904                 case RTN_UNICAST:
905                 case RTN_LOCAL:
906                 case RTN_BROADCAST:
907                 case RTN_ANYCAST:
908                 case RTN_MULTICAST:
909                         break;
910                 default:
911                         goto err_inval;
912                 }
913         }
914 
915         if (cfg->fc_scope > RT_SCOPE_HOST)
916                 goto err_inval;
917 
918         if (cfg->fc_scope == RT_SCOPE_HOST) {
919                 struct fib_nh *nh = fi->fib_nh;
920 
921                 /* Local address is added. */
922                 if (nhs != 1 || nh->nh_gw)
923                         goto err_inval;
924                 nh->nh_scope = RT_SCOPE_NOWHERE;
925                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
926                 err = -ENODEV;
927                 if (!nh->nh_dev)
928                         goto failure;
929         } else {
930                 int linkdown = 0;
931 
932                 change_nexthops(fi) {
933                         err = fib_check_nh(cfg, fi, nexthop_nh);
934                         if (err != 0)
935                                 goto failure;
936                         if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
937                                 linkdown++;
938                 } endfor_nexthops(fi)
939                 if (linkdown == fi->fib_nhs)
940                         fi->fib_flags |= RTNH_F_LINKDOWN;
941         }
942 
943         if (fi->fib_prefsrc) {
944                 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
945                     fi->fib_prefsrc != cfg->fc_dst)
946                         if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
947                                 goto err_inval;
948         }
949 
950         change_nexthops(fi) {
951                 fib_info_update_nh_saddr(net, nexthop_nh);
952         } endfor_nexthops(fi)
953 
954 link_it:
955         ofi = fib_find_info(fi);
956         if (ofi) {
957                 fi->fib_dead = 1;
958                 free_fib_info(fi);
959                 ofi->fib_treeref++;
960                 return ofi;
961         }
962 
963         fi->fib_treeref++;
964         atomic_inc(&fi->fib_clntref);
965         spin_lock_bh(&fib_info_lock);
966         hlist_add_head(&fi->fib_hash,
967                        &fib_info_hash[fib_info_hashfn(fi)]);
968         if (fi->fib_prefsrc) {
969                 struct hlist_head *head;
970 
971                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
972                 hlist_add_head(&fi->fib_lhash, head);
973         }
974         change_nexthops(fi) {
975                 struct hlist_head *head;
976                 unsigned int hash;
977 
978                 if (!nexthop_nh->nh_dev)
979                         continue;
980                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
981                 head = &fib_info_devhash[hash];
982                 hlist_add_head(&nexthop_nh->nh_hash, head);
983         } endfor_nexthops(fi)
984         spin_unlock_bh(&fib_info_lock);
985         return fi;
986 
987 err_inval:
988         err = -EINVAL;
989 
990 failure:
991         if (fi) {
992                 fi->fib_dead = 1;
993                 free_fib_info(fi);
994         }
995 
996         return ERR_PTR(err);
997 }
998 
999 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1000                   u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1001                   struct fib_info *fi, unsigned int flags)
1002 {
1003         struct nlmsghdr *nlh;
1004         struct rtmsg *rtm;
1005 
1006         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1007         if (!nlh)
1008                 return -EMSGSIZE;
1009 
1010         rtm = nlmsg_data(nlh);
1011         rtm->rtm_family = AF_INET;
1012         rtm->rtm_dst_len = dst_len;
1013         rtm->rtm_src_len = 0;
1014         rtm->rtm_tos = tos;
1015         if (tb_id < 256)
1016                 rtm->rtm_table = tb_id;
1017         else
1018                 rtm->rtm_table = RT_TABLE_COMPAT;
1019         if (nla_put_u32(skb, RTA_TABLE, tb_id))
1020                 goto nla_put_failure;
1021         rtm->rtm_type = type;
1022         rtm->rtm_flags = fi->fib_flags;
1023         rtm->rtm_scope = fi->fib_scope;
1024         rtm->rtm_protocol = fi->fib_protocol;
1025 
1026         if (rtm->rtm_dst_len &&
1027             nla_put_in_addr(skb, RTA_DST, dst))
1028                 goto nla_put_failure;
1029         if (fi->fib_priority &&
1030             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1031                 goto nla_put_failure;
1032         if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1033                 goto nla_put_failure;
1034 
1035         if (fi->fib_prefsrc &&
1036             nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1037                 goto nla_put_failure;
1038         if (fi->fib_nhs == 1) {
1039                 struct in_device *in_dev;
1040 
1041                 if (fi->fib_nh->nh_gw &&
1042                     nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1043                         goto nla_put_failure;
1044                 if (fi->fib_nh->nh_oif &&
1045                     nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1046                         goto nla_put_failure;
1047                 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1048                         in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1049                         if (in_dev &&
1050                             IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1051                                 rtm->rtm_flags |= RTNH_F_DEAD;
1052                 }
1053 #ifdef CONFIG_IP_ROUTE_CLASSID
1054                 if (fi->fib_nh[0].nh_tclassid &&
1055                     nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1056                         goto nla_put_failure;
1057 #endif
1058         }
1059 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1060         if (fi->fib_nhs > 1) {
1061                 struct rtnexthop *rtnh;
1062                 struct nlattr *mp;
1063 
1064                 mp = nla_nest_start(skb, RTA_MULTIPATH);
1065                 if (!mp)
1066                         goto nla_put_failure;
1067 
1068                 for_nexthops(fi) {
1069                         struct in_device *in_dev;
1070 
1071                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1072                         if (!rtnh)
1073                                 goto nla_put_failure;
1074 
1075                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1076                         if (nh->nh_flags & RTNH_F_LINKDOWN) {
1077                                 in_dev = __in_dev_get_rtnl(nh->nh_dev);
1078                                 if (in_dev &&
1079                                     IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1080                                         rtnh->rtnh_flags |= RTNH_F_DEAD;
1081                         }
1082                         rtnh->rtnh_hops = nh->nh_weight - 1;
1083                         rtnh->rtnh_ifindex = nh->nh_oif;
1084 
1085                         if (nh->nh_gw &&
1086                             nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1087                                 goto nla_put_failure;
1088 #ifdef CONFIG_IP_ROUTE_CLASSID
1089                         if (nh->nh_tclassid &&
1090                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1091                                 goto nla_put_failure;
1092 #endif
1093                         /* length of rtnetlink header + attributes */
1094                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1095                 } endfor_nexthops(fi);
1096 
1097                 nla_nest_end(skb, mp);
1098         }
1099 #endif
1100         nlmsg_end(skb, nlh);
1101         return 0;
1102 
1103 nla_put_failure:
1104         nlmsg_cancel(skb, nlh);
1105         return -EMSGSIZE;
1106 }
1107 
1108 /*
1109  * Update FIB if:
1110  * - local address disappeared -> we must delete all the entries
1111  *   referring to it.
1112  * - device went down -> we must shutdown all nexthops going via it.
1113  */
1114 int fib_sync_down_addr(struct net *net, __be32 local)
1115 {
1116         int ret = 0;
1117         unsigned int hash = fib_laddr_hashfn(local);
1118         struct hlist_head *head = &fib_info_laddrhash[hash];
1119         struct fib_info *fi;
1120 
1121         if (!fib_info_laddrhash || local == 0)
1122                 return 0;
1123 
1124         hlist_for_each_entry(fi, head, fib_lhash) {
1125                 if (!net_eq(fi->fib_net, net))
1126                         continue;
1127                 if (fi->fib_prefsrc == local) {
1128                         fi->fib_flags |= RTNH_F_DEAD;
1129                         ret++;
1130                 }
1131         }
1132         return ret;
1133 }
1134 
1135 /* Event              force Flags           Description
1136  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1137  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1138  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1139  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1140  */
1141 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1142 {
1143         int ret = 0;
1144         int scope = RT_SCOPE_NOWHERE;
1145         struct fib_info *prev_fi = NULL;
1146         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1147         struct hlist_head *head = &fib_info_devhash[hash];
1148         struct fib_nh *nh;
1149 
1150         if (force)
1151                 scope = -1;
1152 
1153         hlist_for_each_entry(nh, head, nh_hash) {
1154                 struct fib_info *fi = nh->nh_parent;
1155                 int dead;
1156 
1157                 BUG_ON(!fi->fib_nhs);
1158                 if (nh->nh_dev != dev || fi == prev_fi)
1159                         continue;
1160                 prev_fi = fi;
1161                 dead = 0;
1162                 change_nexthops(fi) {
1163                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1164                                 dead++;
1165                         else if (nexthop_nh->nh_dev == dev &&
1166                                  nexthop_nh->nh_scope != scope) {
1167                                 switch (event) {
1168                                 case NETDEV_DOWN:
1169                                 case NETDEV_UNREGISTER:
1170                                         nexthop_nh->nh_flags |= RTNH_F_DEAD;
1171                                         /* fall through */
1172                                 case NETDEV_CHANGE:
1173                                         nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1174                                         break;
1175                                 }
1176 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1177                                 spin_lock_bh(&fib_multipath_lock);
1178                                 fi->fib_power -= nexthop_nh->nh_power;
1179                                 nexthop_nh->nh_power = 0;
1180                                 spin_unlock_bh(&fib_multipath_lock);
1181 #endif
1182                                 dead++;
1183                         }
1184 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1185                         if (event == NETDEV_UNREGISTER &&
1186                             nexthop_nh->nh_dev == dev) {
1187                                 dead = fi->fib_nhs;
1188                                 break;
1189                         }
1190 #endif
1191                 } endfor_nexthops(fi)
1192                 if (dead == fi->fib_nhs) {
1193                         switch (event) {
1194                         case NETDEV_DOWN:
1195                         case NETDEV_UNREGISTER:
1196                                 fi->fib_flags |= RTNH_F_DEAD;
1197                                 /* fall through */
1198                         case NETDEV_CHANGE:
1199                                 fi->fib_flags |= RTNH_F_LINKDOWN;
1200                                 break;
1201                         }
1202                         ret++;
1203                 }
1204         }
1205 
1206         return ret;
1207 }
1208 
1209 /* Must be invoked inside of an RCU protected region.  */
1210 void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1211 {
1212         struct fib_info *fi = NULL, *last_resort = NULL;
1213         struct hlist_head *fa_head = res->fa_head;
1214         struct fib_table *tb = res->table;
1215         u8 slen = 32 - res->prefixlen;
1216         int order = -1, last_idx = -1;
1217         struct fib_alias *fa, *fa1 = NULL;
1218         u32 last_prio = res->fi->fib_priority;
1219         u8 last_tos = 0;
1220 
1221         hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1222                 struct fib_info *next_fi = fa->fa_info;
1223 
1224                 if (fa->fa_slen != slen)
1225                         continue;
1226                 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1227                         continue;
1228                 if (fa->tb_id != tb->tb_id)
1229                         continue;
1230                 if (next_fi->fib_priority > last_prio &&
1231                     fa->fa_tos == last_tos) {
1232                         if (last_tos)
1233                                 continue;
1234                         break;
1235                 }
1236                 if (next_fi->fib_flags & RTNH_F_DEAD)
1237                         continue;
1238                 last_tos = fa->fa_tos;
1239                 last_prio = next_fi->fib_priority;
1240 
1241                 if (next_fi->fib_scope != res->scope ||
1242                     fa->fa_type != RTN_UNICAST)
1243                         continue;
1244                 if (!next_fi->fib_nh[0].nh_gw ||
1245                     next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1246                         continue;
1247 
1248                 fib_alias_accessed(fa);
1249 
1250                 if (!fi) {
1251                         if (next_fi != res->fi)
1252                                 break;
1253                         fa1 = fa;
1254                 } else if (!fib_detect_death(fi, order, &last_resort,
1255                                              &last_idx, fa1->fa_default)) {
1256                         fib_result_assign(res, fi);
1257                         fa1->fa_default = order;
1258                         goto out;
1259                 }
1260                 fi = next_fi;
1261                 order++;
1262         }
1263 
1264         if (order <= 0 || !fi) {
1265                 if (fa1)
1266                         fa1->fa_default = -1;
1267                 goto out;
1268         }
1269 
1270         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1271                               fa1->fa_default)) {
1272                 fib_result_assign(res, fi);
1273                 fa1->fa_default = order;
1274                 goto out;
1275         }
1276 
1277         if (last_idx >= 0)
1278                 fib_result_assign(res, last_resort);
1279         fa1->fa_default = last_idx;
1280 out:
1281         return;
1282 }
1283 
1284 /*
1285  * Dead device goes up. We wake up dead nexthops.
1286  * It takes sense only on multipath routes.
1287  */
1288 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1289 {
1290         struct fib_info *prev_fi;
1291         unsigned int hash;
1292         struct hlist_head *head;
1293         struct fib_nh *nh;
1294         int ret;
1295 
1296         if (!(dev->flags & IFF_UP))
1297                 return 0;
1298 
1299         if (nh_flags & RTNH_F_DEAD) {
1300                 unsigned int flags = dev_get_flags(dev);
1301 
1302                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1303                         nh_flags |= RTNH_F_LINKDOWN;
1304         }
1305 
1306         prev_fi = NULL;
1307         hash = fib_devindex_hashfn(dev->ifindex);
1308         head = &fib_info_devhash[hash];
1309         ret = 0;
1310 
1311         hlist_for_each_entry(nh, head, nh_hash) {
1312                 struct fib_info *fi = nh->nh_parent;
1313                 int alive;
1314 
1315                 BUG_ON(!fi->fib_nhs);
1316                 if (nh->nh_dev != dev || fi == prev_fi)
1317                         continue;
1318 
1319                 prev_fi = fi;
1320                 alive = 0;
1321                 change_nexthops(fi) {
1322                         if (!(nexthop_nh->nh_flags & nh_flags)) {
1323                                 alive++;
1324                                 continue;
1325                         }
1326                         if (!nexthop_nh->nh_dev ||
1327                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1328                                 continue;
1329                         if (nexthop_nh->nh_dev != dev ||
1330                             !__in_dev_get_rtnl(dev))
1331                                 continue;
1332                         alive++;
1333 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1334                         spin_lock_bh(&fib_multipath_lock);
1335                         nexthop_nh->nh_power = 0;
1336                         nexthop_nh->nh_flags &= ~nh_flags;
1337                         spin_unlock_bh(&fib_multipath_lock);
1338 #else
1339                         nexthop_nh->nh_flags &= ~nh_flags;
1340 #endif
1341                 } endfor_nexthops(fi)
1342 
1343                 if (alive > 0) {
1344                         fi->fib_flags &= ~nh_flags;
1345                         ret++;
1346                 }
1347         }
1348 
1349         return ret;
1350 }
1351 
1352 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1353 
1354 /*
1355  * The algorithm is suboptimal, but it provides really
1356  * fair weighted route distribution.
1357  */
1358 void fib_select_multipath(struct fib_result *res)
1359 {
1360         struct fib_info *fi = res->fi;
1361         struct in_device *in_dev;
1362         int w;
1363 
1364         spin_lock_bh(&fib_multipath_lock);
1365         if (fi->fib_power <= 0) {
1366                 int power = 0;
1367                 change_nexthops(fi) {
1368                         in_dev = __in_dev_get_rcu(nexthop_nh->nh_dev);
1369                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1370                                 continue;
1371                         if (in_dev &&
1372                             IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1373                             nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1374                                 continue;
1375                         power += nexthop_nh->nh_weight;
1376                         nexthop_nh->nh_power = nexthop_nh->nh_weight;
1377                 } endfor_nexthops(fi);
1378                 fi->fib_power = power;
1379                 if (power <= 0) {
1380                         spin_unlock_bh(&fib_multipath_lock);
1381                         /* Race condition: route has just become dead. */
1382                         res->nh_sel = 0;
1383                         return;
1384                 }
1385         }
1386 
1387 
1388         /* w should be random number [0..fi->fib_power-1],
1389          * it is pretty bad approximation.
1390          */
1391 
1392         w = jiffies % fi->fib_power;
1393 
1394         change_nexthops(fi) {
1395                 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1396                     nexthop_nh->nh_power) {
1397                         w -= nexthop_nh->nh_power;
1398                         if (w <= 0) {
1399                                 nexthop_nh->nh_power--;
1400                                 fi->fib_power--;
1401                                 res->nh_sel = nhsel;
1402                                 spin_unlock_bh(&fib_multipath_lock);
1403                                 return;
1404                         }
1405                 }
1406         } endfor_nexthops(fi);
1407 
1408         /* Race condition: route has just become dead. */
1409         res->nh_sel = 0;
1410         spin_unlock_bh(&fib_multipath_lock);
1411 }
1412 #endif
1413 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp