3e0da3ef6116df8b9b4fa6a70272ddeac8cac0d4
[linux-3.10.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 <asm/system.h>
18 #include <linux/bitops.h>
19 #include <linux/types.h>
20 #include <linux/kernel.h>
21 #include <linux/jiffies.h>
22 #include <linux/mm.h>
23 #include <linux/string.h>
24 #include <linux/socket.h>
25 #include <linux/sockios.h>
26 #include <linux/errno.h>
27 #include <linux/in.h>
28 #include <linux/inet.h>
29 #include <linux/inetdevice.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_arp.h>
32 #include <linux/proc_fs.h>
33 #include <linux/skbuff.h>
34 #include <linux/init.h>
35 #include <linux/slab.h>
36
37 #include <net/arp.h>
38 #include <net/ip.h>
39 #include <net/protocol.h>
40 #include <net/route.h>
41 #include <net/tcp.h>
42 #include <net/sock.h>
43 #include <net/ip_fib.h>
44 #include <net/netlink.h>
45 #include <net/nexthop.h>
46
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_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 static const struct
94 {
95         int     error;
96         u8      scope;
97 } fib_props[RTN_MAX + 1] = {
98         [RTN_UNSPEC] = {
99                 .error  = 0,
100                 .scope  = RT_SCOPE_NOWHERE,
101         },
102         [RTN_UNICAST] = {
103                 .error  = 0,
104                 .scope  = RT_SCOPE_UNIVERSE,
105         },
106         [RTN_LOCAL] = {
107                 .error  = 0,
108                 .scope  = RT_SCOPE_HOST,
109         },
110         [RTN_BROADCAST] = {
111                 .error  = 0,
112                 .scope  = RT_SCOPE_LINK,
113         },
114         [RTN_ANYCAST] = {
115                 .error  = 0,
116                 .scope  = RT_SCOPE_LINK,
117         },
118         [RTN_MULTICAST] = {
119                 .error  = 0,
120                 .scope  = RT_SCOPE_UNIVERSE,
121         },
122         [RTN_BLACKHOLE] = {
123                 .error  = -EINVAL,
124                 .scope  = RT_SCOPE_UNIVERSE,
125         },
126         [RTN_UNREACHABLE] = {
127                 .error  = -EHOSTUNREACH,
128                 .scope  = RT_SCOPE_UNIVERSE,
129         },
130         [RTN_PROHIBIT] = {
131                 .error  = -EACCES,
132                 .scope  = RT_SCOPE_UNIVERSE,
133         },
134         [RTN_THROW] = {
135                 .error  = -EAGAIN,
136                 .scope  = RT_SCOPE_UNIVERSE,
137         },
138         [RTN_NAT] = {
139                 .error  = -EINVAL,
140                 .scope  = RT_SCOPE_NOWHERE,
141         },
142         [RTN_XRESOLVE] = {
143                 .error  = -EINVAL,
144                 .scope  = RT_SCOPE_NOWHERE,
145         },
146 };
147
148
149 /* Release a nexthop info record */
150
151 static void free_fib_info_rcu(struct rcu_head *head)
152 {
153         struct fib_info *fi = container_of(head, struct fib_info, rcu);
154
155         kfree(fi);
156 }
157
158 void free_fib_info(struct fib_info *fi)
159 {
160         if (fi->fib_dead == 0) {
161                 pr_warning("Freeing alive fib_info %p\n", fi);
162                 return;
163         }
164         change_nexthops(fi) {
165                 if (nexthop_nh->nh_dev)
166                         dev_put(nexthop_nh->nh_dev);
167                 nexthop_nh->nh_dev = NULL;
168         } endfor_nexthops(fi);
169         fib_info_cnt--;
170         release_net(fi->fib_net);
171         call_rcu(&fi->rcu, free_fib_info_rcu);
172 }
173
174 void fib_release_info(struct fib_info *fi)
175 {
176         spin_lock_bh(&fib_info_lock);
177         if (fi && --fi->fib_treeref == 0) {
178                 hlist_del(&fi->fib_hash);
179                 if (fi->fib_prefsrc)
180                         hlist_del(&fi->fib_lhash);
181                 change_nexthops(fi) {
182                         if (!nexthop_nh->nh_dev)
183                                 continue;
184                         hlist_del(&nexthop_nh->nh_hash);
185                 } endfor_nexthops(fi)
186                 fi->fib_dead = 1;
187                 fib_info_put(fi);
188         }
189         spin_unlock_bh(&fib_info_lock);
190 }
191
192 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
193 {
194         const struct fib_nh *onh = ofi->fib_nh;
195
196         for_nexthops(fi) {
197                 if (nh->nh_oif != onh->nh_oif ||
198                     nh->nh_gw  != onh->nh_gw ||
199                     nh->nh_scope != onh->nh_scope ||
200 #ifdef CONFIG_IP_ROUTE_MULTIPATH
201                     nh->nh_weight != onh->nh_weight ||
202 #endif
203 #ifdef CONFIG_NET_CLS_ROUTE
204                     nh->nh_tclassid != onh->nh_tclassid ||
205 #endif
206                     ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_F_DEAD))
207                         return -1;
208                 onh++;
209         } endfor_nexthops(fi);
210         return 0;
211 }
212
213 static inline unsigned int fib_devindex_hashfn(unsigned int val)
214 {
215         unsigned int mask = DEVINDEX_HASHSIZE - 1;
216
217         return (val ^
218                 (val >> DEVINDEX_HASHBITS) ^
219                 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
220 }
221
222 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
223 {
224         unsigned int mask = (fib_hash_size - 1);
225         unsigned int val = fi->fib_nhs;
226
227         val ^= fi->fib_protocol;
228         val ^= (__force u32)fi->fib_prefsrc;
229         val ^= fi->fib_priority;
230         for_nexthops(fi) {
231                 val ^= fib_devindex_hashfn(nh->nh_oif);
232         } endfor_nexthops(fi)
233
234         return (val ^ (val >> 7) ^ (val >> 12)) & mask;
235 }
236
237 static struct fib_info *fib_find_info(const struct fib_info *nfi)
238 {
239         struct hlist_head *head;
240         struct hlist_node *node;
241         struct fib_info *fi;
242         unsigned int hash;
243
244         hash = fib_info_hashfn(nfi);
245         head = &fib_info_hash[hash];
246
247         hlist_for_each_entry(fi, node, head, fib_hash) {
248                 if (!net_eq(fi->fib_net, nfi->fib_net))
249                         continue;
250                 if (fi->fib_nhs != nfi->fib_nhs)
251                         continue;
252                 if (nfi->fib_protocol == fi->fib_protocol &&
253                     nfi->fib_prefsrc == fi->fib_prefsrc &&
254                     nfi->fib_priority == fi->fib_priority &&
255                     memcmp(nfi->fib_metrics, fi->fib_metrics,
256                            sizeof(fi->fib_metrics)) == 0 &&
257                     ((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_F_DEAD) == 0 &&
258                     (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
259                         return fi;
260         }
261
262         return NULL;
263 }
264
265 /* Check, that the gateway is already configured.
266  * Used only by redirect accept routine.
267  */
268 int ip_fib_check_default(__be32 gw, struct net_device *dev)
269 {
270         struct hlist_head *head;
271         struct hlist_node *node;
272         struct fib_nh *nh;
273         unsigned int hash;
274
275         spin_lock(&fib_info_lock);
276
277         hash = fib_devindex_hashfn(dev->ifindex);
278         head = &fib_info_devhash[hash];
279         hlist_for_each_entry(nh, node, head, nh_hash) {
280                 if (nh->nh_dev == dev &&
281                     nh->nh_gw == gw &&
282                     !(nh->nh_flags & RTNH_F_DEAD)) {
283                         spin_unlock(&fib_info_lock);
284                         return 0;
285                 }
286         }
287
288         spin_unlock(&fib_info_lock);
289
290         return -1;
291 }
292
293 static inline size_t fib_nlmsg_size(struct fib_info *fi)
294 {
295         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
296                          + nla_total_size(4) /* RTA_TABLE */
297                          + nla_total_size(4) /* RTA_DST */
298                          + nla_total_size(4) /* RTA_PRIORITY */
299                          + nla_total_size(4); /* RTA_PREFSRC */
300
301         /* space for nested metrics */
302         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
303
304         if (fi->fib_nhs) {
305                 /* Also handles the special case fib_nhs == 1 */
306
307                 /* each nexthop is packed in an attribute */
308                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
309
310                 /* may contain flow and gateway attribute */
311                 nhsize += 2 * nla_total_size(4);
312
313                 /* all nexthops are packed in a nested attribute */
314                 payload += nla_total_size(fi->fib_nhs * nhsize);
315         }
316
317         return payload;
318 }
319
320 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
321                int dst_len, u32 tb_id, struct nl_info *info,
322                unsigned int nlm_flags)
323 {
324         struct sk_buff *skb;
325         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
326         int err = -ENOBUFS;
327
328         skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
329         if (skb == NULL)
330                 goto errout;
331
332         err = fib_dump_info(skb, info->pid, seq, event, tb_id,
333                             fa->fa_type, fa->fa_scope, key, dst_len,
334                             fa->fa_tos, fa->fa_info, nlm_flags);
335         if (err < 0) {
336                 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
337                 WARN_ON(err == -EMSGSIZE);
338                 kfree_skb(skb);
339                 goto errout;
340         }
341         rtnl_notify(skb, info->nl_net, info->pid, RTNLGRP_IPV4_ROUTE,
342                     info->nlh, GFP_KERNEL);
343         return;
344 errout:
345         if (err < 0)
346                 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
347 }
348
349 /* Return the first fib alias matching TOS with
350  * priority less than or equal to PRIO.
351  */
352 struct fib_alias *fib_find_alias(struct list_head *fah, u8 tos, u32 prio)
353 {
354         if (fah) {
355                 struct fib_alias *fa;
356                 list_for_each_entry(fa, fah, fa_list) {
357                         if (fa->fa_tos > tos)
358                                 continue;
359                         if (fa->fa_info->fib_priority >= prio ||
360                             fa->fa_tos < tos)
361                                 return fa;
362                 }
363         }
364         return NULL;
365 }
366
367 int fib_detect_death(struct fib_info *fi, int order,
368                      struct fib_info **last_resort, int *last_idx, int dflt)
369 {
370         struct neighbour *n;
371         int state = NUD_NONE;
372
373         n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
374         if (n) {
375                 state = n->nud_state;
376                 neigh_release(n);
377         }
378         if (state == NUD_REACHABLE)
379                 return 0;
380         if ((state & NUD_VALID) && order != dflt)
381                 return 0;
382         if ((state & NUD_VALID) ||
383             (*last_idx < 0 && order > dflt)) {
384                 *last_resort = fi;
385                 *last_idx = order;
386         }
387         return 1;
388 }
389
390 #ifdef CONFIG_IP_ROUTE_MULTIPATH
391
392 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
393 {
394         int nhs = 0;
395
396         while (rtnh_ok(rtnh, remaining)) {
397                 nhs++;
398                 rtnh = rtnh_next(rtnh, &remaining);
399         }
400
401         /* leftover implies invalid nexthop configuration, discard it */
402         return remaining > 0 ? 0 : nhs;
403 }
404
405 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
406                        int remaining, struct fib_config *cfg)
407 {
408         change_nexthops(fi) {
409                 int attrlen;
410
411                 if (!rtnh_ok(rtnh, remaining))
412                         return -EINVAL;
413
414                 nexthop_nh->nh_flags =
415                         (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
416                 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
417                 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
418
419                 attrlen = rtnh_attrlen(rtnh);
420                 if (attrlen > 0) {
421                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
422
423                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
424                         nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0;
425 #ifdef CONFIG_NET_CLS_ROUTE
426                         nla = nla_find(attrs, attrlen, RTA_FLOW);
427                         nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
428 #endif
429                 }
430
431                 rtnh = rtnh_next(rtnh, &remaining);
432         } endfor_nexthops(fi);
433
434         return 0;
435 }
436
437 #endif
438
439 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
440 {
441 #ifdef CONFIG_IP_ROUTE_MULTIPATH
442         struct rtnexthop *rtnh;
443         int remaining;
444 #endif
445
446         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
447                 return 1;
448
449         if (cfg->fc_oif || cfg->fc_gw) {
450                 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
451                     (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
452                         return 0;
453                 return 1;
454         }
455
456 #ifdef CONFIG_IP_ROUTE_MULTIPATH
457         if (cfg->fc_mp == NULL)
458                 return 0;
459
460         rtnh = cfg->fc_mp;
461         remaining = cfg->fc_mp_len;
462
463         for_nexthops(fi) {
464                 int attrlen;
465
466                 if (!rtnh_ok(rtnh, remaining))
467                         return -EINVAL;
468
469                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
470                         return 1;
471
472                 attrlen = rtnh_attrlen(rtnh);
473                 if (attrlen < 0) {
474                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
475
476                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
477                         if (nla && nla_get_be32(nla) != nh->nh_gw)
478                                 return 1;
479 #ifdef CONFIG_NET_CLS_ROUTE
480                         nla = nla_find(attrs, attrlen, RTA_FLOW);
481                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
482                                 return 1;
483 #endif
484                 }
485
486                 rtnh = rtnh_next(rtnh, &remaining);
487         } endfor_nexthops(fi);
488 #endif
489         return 0;
490 }
491
492
493 /*
494  * Picture
495  * -------
496  *
497  * Semantics of nexthop is very messy by historical reasons.
498  * We have to take into account, that:
499  * a) gateway can be actually local interface address,
500  *    so that gatewayed route is direct.
501  * b) gateway must be on-link address, possibly
502  *    described not by an ifaddr, but also by a direct route.
503  * c) If both gateway and interface are specified, they should not
504  *    contradict.
505  * d) If we use tunnel routes, gateway could be not on-link.
506  *
507  * Attempt to reconcile all of these (alas, self-contradictory) conditions
508  * results in pretty ugly and hairy code with obscure logic.
509  *
510  * I chose to generalized it instead, so that the size
511  * of code does not increase practically, but it becomes
512  * much more general.
513  * Every prefix is assigned a "scope" value: "host" is local address,
514  * "link" is direct route,
515  * [ ... "site" ... "interior" ... ]
516  * and "universe" is true gateway route with global meaning.
517  *
518  * Every prefix refers to a set of "nexthop"s (gw, oif),
519  * where gw must have narrower scope. This recursion stops
520  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
521  * which means that gw is forced to be on link.
522  *
523  * Code is still hairy, but now it is apparently logically
524  * consistent and very flexible. F.e. as by-product it allows
525  * to co-exists in peace independent exterior and interior
526  * routing processes.
527  *
528  * Normally it looks as following.
529  *
530  * {universe prefix}  -> (gw, oif) [scope link]
531  *                |
532  *                |-> {link prefix} -> (gw, oif) [scope local]
533  *                                      |
534  *                                      |-> {local prefix} (terminal node)
535  */
536 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
537                         struct fib_nh *nh)
538 {
539         int err;
540         struct net *net;
541         struct net_device *dev;
542
543         net = cfg->fc_nlinfo.nl_net;
544         if (nh->nh_gw) {
545                 struct fib_result res;
546
547                 if (nh->nh_flags & RTNH_F_ONLINK) {
548
549                         if (cfg->fc_scope >= RT_SCOPE_LINK)
550                                 return -EINVAL;
551                         if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
552                                 return -EINVAL;
553                         dev = __dev_get_by_index(net, nh->nh_oif);
554                         if (!dev)
555                                 return -ENODEV;
556                         if (!(dev->flags & IFF_UP))
557                                 return -ENETDOWN;
558                         nh->nh_dev = dev;
559                         dev_hold(dev);
560                         nh->nh_scope = RT_SCOPE_LINK;
561                         return 0;
562                 }
563                 rcu_read_lock();
564                 {
565                         struct flowi fl = {
566                                 .nl_u = {
567                                         .ip4_u = {
568                                                 .daddr = nh->nh_gw,
569                                                 .scope = cfg->fc_scope + 1,
570                                         },
571                                 },
572                                 .oif = nh->nh_oif,
573                         };
574
575                         /* It is not necessary, but requires a bit of thinking */
576                         if (fl.fl4_scope < RT_SCOPE_LINK)
577                                 fl.fl4_scope = RT_SCOPE_LINK;
578                         err = fib_lookup(net, &fl, &res);
579                         if (err) {
580                                 rcu_read_unlock();
581                                 return err;
582                         }
583                 }
584                 err = -EINVAL;
585                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
586                         goto out;
587                 nh->nh_scope = res.scope;
588                 nh->nh_oif = FIB_RES_OIF(res);
589                 nh->nh_dev = dev = FIB_RES_DEV(res);
590                 if (!dev)
591                         goto out;
592                 dev_hold(dev);
593                 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
594         } else {
595                 struct in_device *in_dev;
596
597                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
598                         return -EINVAL;
599
600                 rcu_read_lock();
601                 err = -ENODEV;
602                 in_dev = inetdev_by_index(net, nh->nh_oif);
603                 if (in_dev == NULL)
604                         goto out;
605                 err = -ENETDOWN;
606                 if (!(in_dev->dev->flags & IFF_UP))
607                         goto out;
608                 nh->nh_dev = in_dev->dev;
609                 dev_hold(nh->nh_dev);
610                 nh->nh_scope = RT_SCOPE_HOST;
611                 err = 0;
612         }
613 out:
614         rcu_read_unlock();
615         return err;
616 }
617
618 static inline unsigned int fib_laddr_hashfn(__be32 val)
619 {
620         unsigned int mask = (fib_hash_size - 1);
621
622         return ((__force u32)val ^
623                 ((__force u32)val >> 7) ^
624                 ((__force u32)val >> 14)) & mask;
625 }
626
627 static struct hlist_head *fib_hash_alloc(int bytes)
628 {
629         if (bytes <= PAGE_SIZE)
630                 return kzalloc(bytes, GFP_KERNEL);
631         else
632                 return (struct hlist_head *)
633                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
634                                          get_order(bytes));
635 }
636
637 static void fib_hash_free(struct hlist_head *hash, int bytes)
638 {
639         if (!hash)
640                 return;
641
642         if (bytes <= PAGE_SIZE)
643                 kfree(hash);
644         else
645                 free_pages((unsigned long) hash, get_order(bytes));
646 }
647
648 static void fib_hash_move(struct hlist_head *new_info_hash,
649                           struct hlist_head *new_laddrhash,
650                           unsigned int new_size)
651 {
652         struct hlist_head *old_info_hash, *old_laddrhash;
653         unsigned int old_size = fib_hash_size;
654         unsigned int i, bytes;
655
656         spin_lock_bh(&fib_info_lock);
657         old_info_hash = fib_info_hash;
658         old_laddrhash = fib_info_laddrhash;
659         fib_hash_size = new_size;
660
661         for (i = 0; i < old_size; i++) {
662                 struct hlist_head *head = &fib_info_hash[i];
663                 struct hlist_node *node, *n;
664                 struct fib_info *fi;
665
666                 hlist_for_each_entry_safe(fi, node, n, head, fib_hash) {
667                         struct hlist_head *dest;
668                         unsigned int new_hash;
669
670                         hlist_del(&fi->fib_hash);
671
672                         new_hash = fib_info_hashfn(fi);
673                         dest = &new_info_hash[new_hash];
674                         hlist_add_head(&fi->fib_hash, dest);
675                 }
676         }
677         fib_info_hash = new_info_hash;
678
679         for (i = 0; i < old_size; i++) {
680                 struct hlist_head *lhead = &fib_info_laddrhash[i];
681                 struct hlist_node *node, *n;
682                 struct fib_info *fi;
683
684                 hlist_for_each_entry_safe(fi, node, n, lhead, fib_lhash) {
685                         struct hlist_head *ldest;
686                         unsigned int new_hash;
687
688                         hlist_del(&fi->fib_lhash);
689
690                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
691                         ldest = &new_laddrhash[new_hash];
692                         hlist_add_head(&fi->fib_lhash, ldest);
693                 }
694         }
695         fib_info_laddrhash = new_laddrhash;
696
697         spin_unlock_bh(&fib_info_lock);
698
699         bytes = old_size * sizeof(struct hlist_head *);
700         fib_hash_free(old_info_hash, bytes);
701         fib_hash_free(old_laddrhash, bytes);
702 }
703
704 struct fib_info *fib_create_info(struct fib_config *cfg)
705 {
706         int err;
707         struct fib_info *fi = NULL;
708         struct fib_info *ofi;
709         int nhs = 1;
710         struct net *net = cfg->fc_nlinfo.nl_net;
711
712         /* Fast check to catch the most weird cases */
713         if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
714                 goto err_inval;
715
716 #ifdef CONFIG_IP_ROUTE_MULTIPATH
717         if (cfg->fc_mp) {
718                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
719                 if (nhs == 0)
720                         goto err_inval;
721         }
722 #endif
723
724         err = -ENOBUFS;
725         if (fib_info_cnt >= fib_hash_size) {
726                 unsigned int new_size = fib_hash_size << 1;
727                 struct hlist_head *new_info_hash;
728                 struct hlist_head *new_laddrhash;
729                 unsigned int bytes;
730
731                 if (!new_size)
732                         new_size = 1;
733                 bytes = new_size * sizeof(struct hlist_head *);
734                 new_info_hash = fib_hash_alloc(bytes);
735                 new_laddrhash = fib_hash_alloc(bytes);
736                 if (!new_info_hash || !new_laddrhash) {
737                         fib_hash_free(new_info_hash, bytes);
738                         fib_hash_free(new_laddrhash, bytes);
739                 } else
740                         fib_hash_move(new_info_hash, new_laddrhash, new_size);
741
742                 if (!fib_hash_size)
743                         goto failure;
744         }
745
746         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
747         if (fi == NULL)
748                 goto failure;
749         fib_info_cnt++;
750
751         fi->fib_net = hold_net(net);
752         fi->fib_protocol = cfg->fc_protocol;
753         fi->fib_flags = cfg->fc_flags;
754         fi->fib_priority = cfg->fc_priority;
755         fi->fib_prefsrc = cfg->fc_prefsrc;
756
757         fi->fib_nhs = nhs;
758         change_nexthops(fi) {
759                 nexthop_nh->nh_parent = fi;
760         } endfor_nexthops(fi)
761
762         if (cfg->fc_mx) {
763                 struct nlattr *nla;
764                 int remaining;
765
766                 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
767                         int type = nla_type(nla);
768
769                         if (type) {
770                                 if (type > RTAX_MAX)
771                                         goto err_inval;
772                                 fi->fib_metrics[type - 1] = nla_get_u32(nla);
773                         }
774                 }
775         }
776
777         if (cfg->fc_mp) {
778 #ifdef CONFIG_IP_ROUTE_MULTIPATH
779                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
780                 if (err != 0)
781                         goto failure;
782                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
783                         goto err_inval;
784                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
785                         goto err_inval;
786 #ifdef CONFIG_NET_CLS_ROUTE
787                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
788                         goto err_inval;
789 #endif
790 #else
791                 goto err_inval;
792 #endif
793         } else {
794                 struct fib_nh *nh = fi->fib_nh;
795
796                 nh->nh_oif = cfg->fc_oif;
797                 nh->nh_gw = cfg->fc_gw;
798                 nh->nh_flags = cfg->fc_flags;
799 #ifdef CONFIG_NET_CLS_ROUTE
800                 nh->nh_tclassid = cfg->fc_flow;
801 #endif
802 #ifdef CONFIG_IP_ROUTE_MULTIPATH
803                 nh->nh_weight = 1;
804 #endif
805         }
806
807         if (fib_props[cfg->fc_type].error) {
808                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
809                         goto err_inval;
810                 goto link_it;
811         }
812
813         if (cfg->fc_scope > RT_SCOPE_HOST)
814                 goto err_inval;
815
816         if (cfg->fc_scope == RT_SCOPE_HOST) {
817                 struct fib_nh *nh = fi->fib_nh;
818
819                 /* Local address is added. */
820                 if (nhs != 1 || nh->nh_gw)
821                         goto err_inval;
822                 nh->nh_scope = RT_SCOPE_NOWHERE;
823                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
824                 err = -ENODEV;
825                 if (nh->nh_dev == NULL)
826                         goto failure;
827         } else {
828                 change_nexthops(fi) {
829                         err = fib_check_nh(cfg, fi, nexthop_nh);
830                         if (err != 0)
831                                 goto failure;
832                 } endfor_nexthops(fi)
833         }
834
835         if (fi->fib_prefsrc) {
836                 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
837                     fi->fib_prefsrc != cfg->fc_dst)
838                         if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
839                                 goto err_inval;
840         }
841
842 link_it:
843         ofi = fib_find_info(fi);
844         if (ofi) {
845                 fi->fib_dead = 1;
846                 free_fib_info(fi);
847                 ofi->fib_treeref++;
848                 return ofi;
849         }
850
851         fi->fib_treeref++;
852         atomic_inc(&fi->fib_clntref);
853         spin_lock_bh(&fib_info_lock);
854         hlist_add_head(&fi->fib_hash,
855                        &fib_info_hash[fib_info_hashfn(fi)]);
856         if (fi->fib_prefsrc) {
857                 struct hlist_head *head;
858
859                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
860                 hlist_add_head(&fi->fib_lhash, head);
861         }
862         change_nexthops(fi) {
863                 struct hlist_head *head;
864                 unsigned int hash;
865
866                 if (!nexthop_nh->nh_dev)
867                         continue;
868                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
869                 head = &fib_info_devhash[hash];
870                 hlist_add_head(&nexthop_nh->nh_hash, head);
871         } endfor_nexthops(fi)
872         spin_unlock_bh(&fib_info_lock);
873         return fi;
874
875 err_inval:
876         err = -EINVAL;
877
878 failure:
879         if (fi) {
880                 fi->fib_dead = 1;
881                 free_fib_info(fi);
882         }
883
884         return ERR_PTR(err);
885 }
886
887 /* Note! fib_semantic_match intentionally uses  RCU list functions. */
888 int fib_semantic_match(struct list_head *head, const struct flowi *flp,
889                        struct fib_result *res, int prefixlen, int fib_flags)
890 {
891         struct fib_alias *fa;
892         int nh_sel = 0;
893
894         list_for_each_entry_rcu(fa, head, fa_list) {
895                 int err;
896
897                 if (fa->fa_tos &&
898                     fa->fa_tos != flp->fl4_tos)
899                         continue;
900
901                 if (fa->fa_scope < flp->fl4_scope)
902                         continue;
903
904                 fib_alias_accessed(fa);
905
906                 err = fib_props[fa->fa_type].error;
907                 if (err == 0) {
908                         struct fib_info *fi = fa->fa_info;
909
910                         if (fi->fib_flags & RTNH_F_DEAD)
911                                 continue;
912
913                         switch (fa->fa_type) {
914                         case RTN_UNICAST:
915                         case RTN_LOCAL:
916                         case RTN_BROADCAST:
917                         case RTN_ANYCAST:
918                         case RTN_MULTICAST:
919                                 for_nexthops(fi) {
920                                         if (nh->nh_flags & RTNH_F_DEAD)
921                                                 continue;
922                                         if (!flp->oif || flp->oif == nh->nh_oif)
923                                                 break;
924                                 }
925 #ifdef CONFIG_IP_ROUTE_MULTIPATH
926                                 if (nhsel < fi->fib_nhs) {
927                                         nh_sel = nhsel;
928                                         goto out_fill_res;
929                                 }
930 #else
931                                 if (nhsel < 1)
932                                         goto out_fill_res;
933 #endif
934                                 endfor_nexthops(fi);
935                                 continue;
936
937                         default:
938                                 pr_warning("fib_semantic_match bad type %#x\n",
939                                            fa->fa_type);
940                                 return -EINVAL;
941                         }
942                 }
943                 return err;
944         }
945         return 1;
946
947 out_fill_res:
948         res->prefixlen = prefixlen;
949         res->nh_sel = nh_sel;
950         res->type = fa->fa_type;
951         res->scope = fa->fa_scope;
952         res->fi = fa->fa_info;
953         if (!(fib_flags & FIB_LOOKUP_NOREF))
954                 atomic_inc(&res->fi->fib_clntref);
955         return 0;
956 }
957
958 /* Find appropriate source address to this destination */
959
960 __be32 __fib_res_prefsrc(struct fib_result *res)
961 {
962         return inet_select_addr(FIB_RES_DEV(*res), FIB_RES_GW(*res), res->scope);
963 }
964
965 int fib_dump_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
966                   u32 tb_id, u8 type, u8 scope, __be32 dst, int dst_len, u8 tos,
967                   struct fib_info *fi, unsigned int flags)
968 {
969         struct nlmsghdr *nlh;
970         struct rtmsg *rtm;
971
972         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*rtm), flags);
973         if (nlh == NULL)
974                 return -EMSGSIZE;
975
976         rtm = nlmsg_data(nlh);
977         rtm->rtm_family = AF_INET;
978         rtm->rtm_dst_len = dst_len;
979         rtm->rtm_src_len = 0;
980         rtm->rtm_tos = tos;
981         if (tb_id < 256)
982                 rtm->rtm_table = tb_id;
983         else
984                 rtm->rtm_table = RT_TABLE_COMPAT;
985         NLA_PUT_U32(skb, RTA_TABLE, tb_id);
986         rtm->rtm_type = type;
987         rtm->rtm_flags = fi->fib_flags;
988         rtm->rtm_scope = scope;
989         rtm->rtm_protocol = fi->fib_protocol;
990
991         if (rtm->rtm_dst_len)
992                 NLA_PUT_BE32(skb, RTA_DST, dst);
993
994         if (fi->fib_priority)
995                 NLA_PUT_U32(skb, RTA_PRIORITY, fi->fib_priority);
996
997         if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
998                 goto nla_put_failure;
999
1000         if (fi->fib_prefsrc)
1001                 NLA_PUT_BE32(skb, RTA_PREFSRC, fi->fib_prefsrc);
1002
1003         if (fi->fib_nhs == 1) {
1004                 if (fi->fib_nh->nh_gw)
1005                         NLA_PUT_BE32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw);
1006
1007                 if (fi->fib_nh->nh_oif)
1008                         NLA_PUT_U32(skb, RTA_OIF, fi->fib_nh->nh_oif);
1009 #ifdef CONFIG_NET_CLS_ROUTE
1010                 if (fi->fib_nh[0].nh_tclassid)
1011                         NLA_PUT_U32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid);
1012 #endif
1013         }
1014 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1015         if (fi->fib_nhs > 1) {
1016                 struct rtnexthop *rtnh;
1017                 struct nlattr *mp;
1018
1019                 mp = nla_nest_start(skb, RTA_MULTIPATH);
1020                 if (mp == NULL)
1021                         goto nla_put_failure;
1022
1023                 for_nexthops(fi) {
1024                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1025                         if (rtnh == NULL)
1026                                 goto nla_put_failure;
1027
1028                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1029                         rtnh->rtnh_hops = nh->nh_weight - 1;
1030                         rtnh->rtnh_ifindex = nh->nh_oif;
1031
1032                         if (nh->nh_gw)
1033                                 NLA_PUT_BE32(skb, RTA_GATEWAY, nh->nh_gw);
1034 #ifdef CONFIG_NET_CLS_ROUTE
1035                         if (nh->nh_tclassid)
1036                                 NLA_PUT_U32(skb, RTA_FLOW, nh->nh_tclassid);
1037 #endif
1038                         /* length of rtnetlink header + attributes */
1039                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1040                 } endfor_nexthops(fi);
1041
1042                 nla_nest_end(skb, mp);
1043         }
1044 #endif
1045         return nlmsg_end(skb, nlh);
1046
1047 nla_put_failure:
1048         nlmsg_cancel(skb, nlh);
1049         return -EMSGSIZE;
1050 }
1051
1052 /*
1053  * Update FIB if:
1054  * - local address disappeared -> we must delete all the entries
1055  *   referring to it.
1056  * - device went down -> we must shutdown all nexthops going via it.
1057  */
1058 int fib_sync_down_addr(struct net *net, __be32 local)
1059 {
1060         int ret = 0;
1061         unsigned int hash = fib_laddr_hashfn(local);
1062         struct hlist_head *head = &fib_info_laddrhash[hash];
1063         struct hlist_node *node;
1064         struct fib_info *fi;
1065
1066         if (fib_info_laddrhash == NULL || local == 0)
1067                 return 0;
1068
1069         hlist_for_each_entry(fi, node, head, fib_lhash) {
1070                 if (!net_eq(fi->fib_net, net))
1071                         continue;
1072                 if (fi->fib_prefsrc == local) {
1073                         fi->fib_flags |= RTNH_F_DEAD;
1074                         ret++;
1075                 }
1076         }
1077         return ret;
1078 }
1079
1080 int fib_sync_down_dev(struct net_device *dev, int force)
1081 {
1082         int ret = 0;
1083         int scope = RT_SCOPE_NOWHERE;
1084         struct fib_info *prev_fi = NULL;
1085         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1086         struct hlist_head *head = &fib_info_devhash[hash];
1087         struct hlist_node *node;
1088         struct fib_nh *nh;
1089
1090         if (force)
1091                 scope = -1;
1092
1093         hlist_for_each_entry(nh, node, head, nh_hash) {
1094                 struct fib_info *fi = nh->nh_parent;
1095                 int dead;
1096
1097                 BUG_ON(!fi->fib_nhs);
1098                 if (nh->nh_dev != dev || fi == prev_fi)
1099                         continue;
1100                 prev_fi = fi;
1101                 dead = 0;
1102                 change_nexthops(fi) {
1103                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1104                                 dead++;
1105                         else if (nexthop_nh->nh_dev == dev &&
1106                                  nexthop_nh->nh_scope != scope) {
1107                                 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1108 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1109                                 spin_lock_bh(&fib_multipath_lock);
1110                                 fi->fib_power -= nexthop_nh->nh_power;
1111                                 nexthop_nh->nh_power = 0;
1112                                 spin_unlock_bh(&fib_multipath_lock);
1113 #endif
1114                                 dead++;
1115                         }
1116 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1117                         if (force > 1 && nexthop_nh->nh_dev == dev) {
1118                                 dead = fi->fib_nhs;
1119                                 break;
1120                         }
1121 #endif
1122                 } endfor_nexthops(fi)
1123                 if (dead == fi->fib_nhs) {
1124                         fi->fib_flags |= RTNH_F_DEAD;
1125                         ret++;
1126                 }
1127         }
1128
1129         return ret;
1130 }
1131
1132 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1133
1134 /*
1135  * Dead device goes up. We wake up dead nexthops.
1136  * It takes sense only on multipath routes.
1137  */
1138 int fib_sync_up(struct net_device *dev)
1139 {
1140         struct fib_info *prev_fi;
1141         unsigned int hash;
1142         struct hlist_head *head;
1143         struct hlist_node *node;
1144         struct fib_nh *nh;
1145         int ret;
1146
1147         if (!(dev->flags & IFF_UP))
1148                 return 0;
1149
1150         prev_fi = NULL;
1151         hash = fib_devindex_hashfn(dev->ifindex);
1152         head = &fib_info_devhash[hash];
1153         ret = 0;
1154
1155         hlist_for_each_entry(nh, node, head, nh_hash) {
1156                 struct fib_info *fi = nh->nh_parent;
1157                 int alive;
1158
1159                 BUG_ON(!fi->fib_nhs);
1160                 if (nh->nh_dev != dev || fi == prev_fi)
1161                         continue;
1162
1163                 prev_fi = fi;
1164                 alive = 0;
1165                 change_nexthops(fi) {
1166                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1167                                 alive++;
1168                                 continue;
1169                         }
1170                         if (nexthop_nh->nh_dev == NULL ||
1171                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1172                                 continue;
1173                         if (nexthop_nh->nh_dev != dev ||
1174                             !__in_dev_get_rtnl(dev))
1175                                 continue;
1176                         alive++;
1177                         spin_lock_bh(&fib_multipath_lock);
1178                         nexthop_nh->nh_power = 0;
1179                         nexthop_nh->nh_flags &= ~RTNH_F_DEAD;
1180                         spin_unlock_bh(&fib_multipath_lock);
1181                 } endfor_nexthops(fi)
1182
1183                 if (alive > 0) {
1184                         fi->fib_flags &= ~RTNH_F_DEAD;
1185                         ret++;
1186                 }
1187         }
1188
1189         return ret;
1190 }
1191
1192 /*
1193  * The algorithm is suboptimal, but it provides really
1194  * fair weighted route distribution.
1195  */
1196 void fib_select_multipath(const struct flowi *flp, struct fib_result *res)
1197 {
1198         struct fib_info *fi = res->fi;
1199         int w;
1200
1201         spin_lock_bh(&fib_multipath_lock);
1202         if (fi->fib_power <= 0) {
1203                 int power = 0;
1204                 change_nexthops(fi) {
1205                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1206                                 power += nexthop_nh->nh_weight;
1207                                 nexthop_nh->nh_power = nexthop_nh->nh_weight;
1208                         }
1209                 } endfor_nexthops(fi);
1210                 fi->fib_power = power;
1211                 if (power <= 0) {
1212                         spin_unlock_bh(&fib_multipath_lock);
1213                         /* Race condition: route has just become dead. */
1214                         res->nh_sel = 0;
1215                         return;
1216                 }
1217         }
1218
1219
1220         /* w should be random number [0..fi->fib_power-1],
1221          * it is pretty bad approximation.
1222          */
1223
1224         w = jiffies % fi->fib_power;
1225
1226         change_nexthops(fi) {
1227                 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1228                     nexthop_nh->nh_power) {
1229                         w -= nexthop_nh->nh_power;
1230                         if (w <= 0) {
1231                                 nexthop_nh->nh_power--;
1232                                 fi->fib_power--;
1233                                 res->nh_sel = nhsel;
1234                                 spin_unlock_bh(&fib_multipath_lock);
1235                                 return;
1236                         }
1237                 }
1238         } endfor_nexthops(fi);
1239
1240         /* Race condition: route has just become dead. */
1241         res->nh_sel = 0;
1242         spin_unlock_bh(&fib_multipath_lock);
1243 }
1244 #endif