GNU Linux-libre 4.4.289-gnu1
[releases.git] / net / ipv6 / ip6_fib.c
1 /*
2  *      Linux INET6 implementation
3  *      Forwarding Information Database
4  *
5  *      Authors:
6  *      Pedro Roque             <roque@di.fc.ul.pt>
7  *
8  *      This program is free software; you can redistribute it and/or
9  *      modify it under the terms of the GNU General Public License
10  *      as published by the Free Software Foundation; either version
11  *      2 of the License, or (at your option) any later version.
12  *
13  *      Changes:
14  *      Yuji SEKIYA @USAGI:     Support default route on router node;
15  *                              remove ip6_null_entry from the top of
16  *                              routing table.
17  *      Ville Nuorvala:         Fixed routing subtrees.
18  */
19
20 #define pr_fmt(fmt) "IPv6: " fmt
21
22 #include <linux/errno.h>
23 #include <linux/types.h>
24 #include <linux/net.h>
25 #include <linux/route.h>
26 #include <linux/netdevice.h>
27 #include <linux/in6.h>
28 #include <linux/init.h>
29 #include <linux/list.h>
30 #include <linux/slab.h>
31
32 #include <net/ipv6.h>
33 #include <net/ndisc.h>
34 #include <net/addrconf.h>
35 #include <net/lwtunnel.h>
36
37 #include <net/ip6_fib.h>
38 #include <net/ip6_route.h>
39
40 #define RT6_DEBUG 2
41
42 #if RT6_DEBUG >= 3
43 #define RT6_TRACE(x...) pr_debug(x)
44 #else
45 #define RT6_TRACE(x...) do { ; } while (0)
46 #endif
47
48 static struct kmem_cache *fib6_node_kmem __read_mostly;
49
50 struct fib6_cleaner {
51         struct fib6_walker w;
52         struct net *net;
53         int (*func)(struct rt6_info *, void *arg);
54         int sernum;
55         void *arg;
56 };
57
58 static DEFINE_RWLOCK(fib6_walker_lock);
59
60 #ifdef CONFIG_IPV6_SUBTREES
61 #define FWS_INIT FWS_S
62 #else
63 #define FWS_INIT FWS_L
64 #endif
65
66 static void fib6_prune_clones(struct net *net, struct fib6_node *fn);
67 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
68 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
69 static int fib6_walk(struct fib6_walker *w);
70 static int fib6_walk_continue(struct fib6_walker *w);
71
72 /*
73  *      A routing update causes an increase of the serial number on the
74  *      affected subtree. This allows for cached routes to be asynchronously
75  *      tested when modifications are made to the destination cache as a
76  *      result of redirects, path MTU changes, etc.
77  */
78
79 static void fib6_gc_timer_cb(unsigned long arg);
80
81 static LIST_HEAD(fib6_walkers);
82 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
83
84 static void fib6_walker_link(struct fib6_walker *w)
85 {
86         write_lock_bh(&fib6_walker_lock);
87         list_add(&w->lh, &fib6_walkers);
88         write_unlock_bh(&fib6_walker_lock);
89 }
90
91 static void fib6_walker_unlink(struct fib6_walker *w)
92 {
93         write_lock_bh(&fib6_walker_lock);
94         list_del(&w->lh);
95         write_unlock_bh(&fib6_walker_lock);
96 }
97
98 static int fib6_new_sernum(struct net *net)
99 {
100         int new, old;
101
102         do {
103                 old = atomic_read(&net->ipv6.fib6_sernum);
104                 new = old < INT_MAX ? old + 1 : 1;
105         } while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
106                                 old, new) != old);
107         return new;
108 }
109
110 enum {
111         FIB6_NO_SERNUM_CHANGE = 0,
112 };
113
114 /*
115  *      Auxiliary address test functions for the radix tree.
116  *
117  *      These assume a 32bit processor (although it will work on
118  *      64bit processors)
119  */
120
121 /*
122  *      test bit
123  */
124 #if defined(__LITTLE_ENDIAN)
125 # define BITOP_BE32_SWIZZLE     (0x1F & ~7)
126 #else
127 # define BITOP_BE32_SWIZZLE     0
128 #endif
129
130 static __be32 addr_bit_set(const void *token, int fn_bit)
131 {
132         const __be32 *addr = token;
133         /*
134          * Here,
135          *      1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
136          * is optimized version of
137          *      htonl(1 << ((~fn_bit)&0x1F))
138          * See include/asm-generic/bitops/le.h.
139          */
140         return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
141                addr[fn_bit >> 5];
142 }
143
144 static struct fib6_node *node_alloc(void)
145 {
146         struct fib6_node *fn;
147
148         fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
149
150         return fn;
151 }
152
153 static void node_free_immediate(struct fib6_node *fn)
154 {
155         kmem_cache_free(fib6_node_kmem, fn);
156 }
157
158 static void node_free_rcu(struct rcu_head *head)
159 {
160         struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
161
162         kmem_cache_free(fib6_node_kmem, fn);
163 }
164
165 static void node_free(struct fib6_node *fn)
166 {
167         call_rcu(&fn->rcu, node_free_rcu);
168 }
169
170 static void rt6_rcu_free(struct rt6_info *rt)
171 {
172         call_rcu(&rt->dst.rcu_head, dst_rcu_free);
173 }
174
175 static void rt6_free_pcpu(struct rt6_info *non_pcpu_rt)
176 {
177         int cpu;
178
179         if (!non_pcpu_rt->rt6i_pcpu)
180                 return;
181
182         for_each_possible_cpu(cpu) {
183                 struct rt6_info **ppcpu_rt;
184                 struct rt6_info *pcpu_rt;
185
186                 ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
187                 pcpu_rt = *ppcpu_rt;
188                 if (pcpu_rt) {
189                         rt6_rcu_free(pcpu_rt);
190                         *ppcpu_rt = NULL;
191                 }
192         }
193
194         free_percpu(non_pcpu_rt->rt6i_pcpu);
195         non_pcpu_rt->rt6i_pcpu = NULL;
196 }
197
198 static void rt6_release(struct rt6_info *rt)
199 {
200         if (atomic_dec_and_test(&rt->rt6i_ref)) {
201                 rt6_free_pcpu(rt);
202                 rt6_rcu_free(rt);
203         }
204 }
205
206 static void fib6_free_table(struct fib6_table *table)
207 {
208         inetpeer_invalidate_tree(&table->tb6_peers);
209         kfree(table);
210 }
211
212 static void fib6_link_table(struct net *net, struct fib6_table *tb)
213 {
214         unsigned int h;
215
216         /*
217          * Initialize table lock at a single place to give lockdep a key,
218          * tables aren't visible prior to being linked to the list.
219          */
220         rwlock_init(&tb->tb6_lock);
221
222         h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
223
224         /*
225          * No protection necessary, this is the only list mutatation
226          * operation, tables never disappear once they exist.
227          */
228         hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
229 }
230
231 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
232
233 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
234 {
235         struct fib6_table *table;
236
237         table = kzalloc(sizeof(*table), GFP_ATOMIC);
238         if (table) {
239                 table->tb6_id = id;
240                 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
241                 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
242                 inet_peer_base_init(&table->tb6_peers);
243         }
244
245         return table;
246 }
247
248 struct fib6_table *fib6_new_table(struct net *net, u32 id)
249 {
250         struct fib6_table *tb;
251
252         if (id == 0)
253                 id = RT6_TABLE_MAIN;
254         tb = fib6_get_table(net, id);
255         if (tb)
256                 return tb;
257
258         tb = fib6_alloc_table(net, id);
259         if (tb)
260                 fib6_link_table(net, tb);
261
262         return tb;
263 }
264
265 struct fib6_table *fib6_get_table(struct net *net, u32 id)
266 {
267         struct fib6_table *tb;
268         struct hlist_head *head;
269         unsigned int h;
270
271         if (id == 0)
272                 id = RT6_TABLE_MAIN;
273         h = id & (FIB6_TABLE_HASHSZ - 1);
274         rcu_read_lock();
275         head = &net->ipv6.fib_table_hash[h];
276         hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
277                 if (tb->tb6_id == id) {
278                         rcu_read_unlock();
279                         return tb;
280                 }
281         }
282         rcu_read_unlock();
283
284         return NULL;
285 }
286 EXPORT_SYMBOL_GPL(fib6_get_table);
287
288 static void __net_init fib6_tables_init(struct net *net)
289 {
290         fib6_link_table(net, net->ipv6.fib6_main_tbl);
291         fib6_link_table(net, net->ipv6.fib6_local_tbl);
292 }
293 #else
294
295 struct fib6_table *fib6_new_table(struct net *net, u32 id)
296 {
297         return fib6_get_table(net, id);
298 }
299
300 struct fib6_table *fib6_get_table(struct net *net, u32 id)
301 {
302           return net->ipv6.fib6_main_tbl;
303 }
304
305 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
306                                    int flags, pol_lookup_t lookup)
307 {
308         struct rt6_info *rt;
309
310         rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
311         if (rt->dst.error == -EAGAIN) {
312                 ip6_rt_put(rt);
313                 rt = net->ipv6.ip6_null_entry;
314                 dst_hold(&rt->dst);
315         }
316
317         return &rt->dst;
318 }
319
320 static void __net_init fib6_tables_init(struct net *net)
321 {
322         fib6_link_table(net, net->ipv6.fib6_main_tbl);
323 }
324
325 #endif
326
327 static int fib6_dump_node(struct fib6_walker *w)
328 {
329         int res;
330         struct rt6_info *rt;
331
332         for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
333                 res = rt6_dump_route(rt, w->args);
334                 if (res < 0) {
335                         /* Frame is full, suspend walking */
336                         w->leaf = rt;
337                         return 1;
338                 }
339         }
340         w->leaf = NULL;
341         return 0;
342 }
343
344 static void fib6_dump_end(struct netlink_callback *cb)
345 {
346         struct fib6_walker *w = (void *)cb->args[2];
347
348         if (w) {
349                 if (cb->args[4]) {
350                         cb->args[4] = 0;
351                         fib6_walker_unlink(w);
352                 }
353                 cb->args[2] = 0;
354                 kfree(w);
355         }
356         cb->done = (void *)cb->args[3];
357         cb->args[1] = 3;
358 }
359
360 static int fib6_dump_done(struct netlink_callback *cb)
361 {
362         fib6_dump_end(cb);
363         return cb->done ? cb->done(cb) : 0;
364 }
365
366 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
367                            struct netlink_callback *cb)
368 {
369         struct fib6_walker *w;
370         int res;
371
372         w = (void *)cb->args[2];
373         w->root = &table->tb6_root;
374
375         if (cb->args[4] == 0) {
376                 w->count = 0;
377                 w->skip = 0;
378
379                 read_lock_bh(&table->tb6_lock);
380                 res = fib6_walk(w);
381                 read_unlock_bh(&table->tb6_lock);
382                 if (res > 0) {
383                         cb->args[4] = 1;
384                         cb->args[5] = w->root->fn_sernum;
385                 }
386         } else {
387                 if (cb->args[5] != w->root->fn_sernum) {
388                         /* Begin at the root if the tree changed */
389                         cb->args[5] = w->root->fn_sernum;
390                         w->state = FWS_INIT;
391                         w->node = w->root;
392                         w->skip = w->count;
393                 } else
394                         w->skip = 0;
395
396                 read_lock_bh(&table->tb6_lock);
397                 res = fib6_walk_continue(w);
398                 read_unlock_bh(&table->tb6_lock);
399                 if (res <= 0) {
400                         fib6_walker_unlink(w);
401                         cb->args[4] = 0;
402                 }
403         }
404
405         return res;
406 }
407
408 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
409 {
410         struct net *net = sock_net(skb->sk);
411         unsigned int h, s_h;
412         unsigned int e = 0, s_e;
413         struct rt6_rtnl_dump_arg arg;
414         struct fib6_walker *w;
415         struct fib6_table *tb;
416         struct hlist_head *head;
417         int res = 0;
418
419         s_h = cb->args[0];
420         s_e = cb->args[1];
421
422         w = (void *)cb->args[2];
423         if (!w) {
424                 /* New dump:
425                  *
426                  * 1. hook callback destructor.
427                  */
428                 cb->args[3] = (long)cb->done;
429                 cb->done = fib6_dump_done;
430
431                 /*
432                  * 2. allocate and initialize walker.
433                  */
434                 w = kzalloc(sizeof(*w), GFP_ATOMIC);
435                 if (!w)
436                         return -ENOMEM;
437                 w->func = fib6_dump_node;
438                 cb->args[2] = (long)w;
439         }
440
441         arg.skb = skb;
442         arg.cb = cb;
443         arg.net = net;
444         w->args = &arg;
445
446         rcu_read_lock();
447         for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
448                 e = 0;
449                 head = &net->ipv6.fib_table_hash[h];
450                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
451                         if (e < s_e)
452                                 goto next;
453                         res = fib6_dump_table(tb, skb, cb);
454                         if (res != 0)
455                                 goto out;
456 next:
457                         e++;
458                 }
459         }
460 out:
461         rcu_read_unlock();
462         cb->args[1] = e;
463         cb->args[0] = h;
464
465         res = res < 0 ? res : skb->len;
466         if (res <= 0)
467                 fib6_dump_end(cb);
468         return res;
469 }
470
471 /*
472  *      Routing Table
473  *
474  *      return the appropriate node for a routing tree "add" operation
475  *      by either creating and inserting or by returning an existing
476  *      node.
477  */
478
479 static struct fib6_node *fib6_add_1(struct fib6_node *root,
480                                      struct in6_addr *addr, int plen,
481                                      int offset, int allow_create,
482                                      int replace_required, int sernum)
483 {
484         struct fib6_node *fn, *in, *ln;
485         struct fib6_node *pn = NULL;
486         struct rt6key *key;
487         int     bit;
488         __be32  dir = 0;
489
490         RT6_TRACE("fib6_add_1\n");
491
492         /* insert node in tree */
493
494         fn = root;
495
496         do {
497                 key = (struct rt6key *)((u8 *)fn->leaf + offset);
498
499                 /*
500                  *      Prefix match
501                  */
502                 if (plen < fn->fn_bit ||
503                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
504                         if (!allow_create) {
505                                 if (replace_required) {
506                                         pr_warn("Can't replace route, no match found\n");
507                                         return ERR_PTR(-ENOENT);
508                                 }
509                                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
510                         }
511                         goto insert_above;
512                 }
513
514                 /*
515                  *      Exact match ?
516                  */
517
518                 if (plen == fn->fn_bit) {
519                         /* clean up an intermediate node */
520                         if (!(fn->fn_flags & RTN_RTINFO)) {
521                                 rt6_release(fn->leaf);
522                                 fn->leaf = NULL;
523                         }
524
525                         fn->fn_sernum = sernum;
526
527                         return fn;
528                 }
529
530                 /*
531                  *      We have more bits to go
532                  */
533
534                 /* Try to walk down on tree. */
535                 fn->fn_sernum = sernum;
536                 dir = addr_bit_set(addr, fn->fn_bit);
537                 pn = fn;
538                 fn = dir ? fn->right : fn->left;
539         } while (fn);
540
541         if (!allow_create) {
542                 /* We should not create new node because
543                  * NLM_F_REPLACE was specified without NLM_F_CREATE
544                  * I assume it is safe to require NLM_F_CREATE when
545                  * REPLACE flag is used! Later we may want to remove the
546                  * check for replace_required, because according
547                  * to netlink specification, NLM_F_CREATE
548                  * MUST be specified if new route is created.
549                  * That would keep IPv6 consistent with IPv4
550                  */
551                 if (replace_required) {
552                         pr_warn("Can't replace route, no match found\n");
553                         return ERR_PTR(-ENOENT);
554                 }
555                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
556         }
557         /*
558          *      We walked to the bottom of tree.
559          *      Create new leaf node without children.
560          */
561
562         ln = node_alloc();
563
564         if (!ln)
565                 return ERR_PTR(-ENOMEM);
566         ln->fn_bit = plen;
567
568         ln->parent = pn;
569         ln->fn_sernum = sernum;
570
571         if (dir)
572                 pn->right = ln;
573         else
574                 pn->left  = ln;
575
576         return ln;
577
578
579 insert_above:
580         /*
581          * split since we don't have a common prefix anymore or
582          * we have a less significant route.
583          * we've to insert an intermediate node on the list
584          * this new node will point to the one we need to create
585          * and the current
586          */
587
588         pn = fn->parent;
589
590         /* find 1st bit in difference between the 2 addrs.
591
592            See comment in __ipv6_addr_diff: bit may be an invalid value,
593            but if it is >= plen, the value is ignored in any case.
594          */
595
596         bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
597
598         /*
599          *              (intermediate)[in]
600          *                /        \
601          *      (new leaf node)[ln] (old node)[fn]
602          */
603         if (plen > bit) {
604                 in = node_alloc();
605                 ln = node_alloc();
606
607                 if (!in || !ln) {
608                         if (in)
609                                 node_free_immediate(in);
610                         if (ln)
611                                 node_free_immediate(ln);
612                         return ERR_PTR(-ENOMEM);
613                 }
614
615                 /*
616                  * new intermediate node.
617                  * RTN_RTINFO will
618                  * be off since that an address that chooses one of
619                  * the branches would not match less specific routes
620                  * in the other branch
621                  */
622
623                 in->fn_bit = bit;
624
625                 in->parent = pn;
626                 in->leaf = fn->leaf;
627                 atomic_inc(&in->leaf->rt6i_ref);
628
629                 in->fn_sernum = sernum;
630
631                 /* update parent pointer */
632                 if (dir)
633                         pn->right = in;
634                 else
635                         pn->left  = in;
636
637                 ln->fn_bit = plen;
638
639                 ln->parent = in;
640                 fn->parent = in;
641
642                 ln->fn_sernum = sernum;
643
644                 if (addr_bit_set(addr, bit)) {
645                         in->right = ln;
646                         in->left  = fn;
647                 } else {
648                         in->left  = ln;
649                         in->right = fn;
650                 }
651         } else { /* plen <= bit */
652
653                 /*
654                  *              (new leaf node)[ln]
655                  *                /        \
656                  *           (old node)[fn] NULL
657                  */
658
659                 ln = node_alloc();
660
661                 if (!ln)
662                         return ERR_PTR(-ENOMEM);
663
664                 ln->fn_bit = plen;
665
666                 ln->parent = pn;
667
668                 ln->fn_sernum = sernum;
669
670                 if (dir)
671                         pn->right = ln;
672                 else
673                         pn->left  = ln;
674
675                 if (addr_bit_set(&key->addr, plen))
676                         ln->right = fn;
677                 else
678                         ln->left  = fn;
679
680                 fn->parent = ln;
681         }
682         return ln;
683 }
684
685 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
686 {
687         return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
688                RTF_GATEWAY;
689 }
690
691 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
692 {
693         int i;
694
695         for (i = 0; i < RTAX_MAX; i++) {
696                 if (test_bit(i, mxc->mx_valid))
697                         mp[i] = mxc->mx[i];
698         }
699 }
700
701 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
702 {
703         if (!mxc->mx)
704                 return 0;
705
706         if (dst->flags & DST_HOST) {
707                 u32 *mp = dst_metrics_write_ptr(dst);
708
709                 if (unlikely(!mp))
710                         return -ENOMEM;
711
712                 fib6_copy_metrics(mp, mxc);
713         } else {
714                 dst_init_metrics(dst, mxc->mx, false);
715
716                 /* We've stolen mx now. */
717                 mxc->mx = NULL;
718         }
719
720         return 0;
721 }
722
723 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
724                           struct net *net)
725 {
726         if (atomic_read(&rt->rt6i_ref) != 1) {
727                 /* This route is used as dummy address holder in some split
728                  * nodes. It is not leaked, but it still holds other resources,
729                  * which must be released in time. So, scan ascendant nodes
730                  * and replace dummy references to this route with references
731                  * to still alive ones.
732                  */
733                 while (fn) {
734                         if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
735                                 fn->leaf = fib6_find_prefix(net, fn);
736                                 atomic_inc(&fn->leaf->rt6i_ref);
737                                 rt6_release(rt);
738                         }
739                         fn = fn->parent;
740                 }
741                 /* No more references are possible at this point. */
742                 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
743         }
744 }
745
746 /*
747  *      Insert routing information in a node.
748  */
749
750 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
751                             struct nl_info *info, struct mx6_config *mxc)
752 {
753         struct rt6_info *iter = NULL;
754         struct rt6_info **ins;
755         struct rt6_info **fallback_ins = NULL;
756         int replace = (info->nlh &&
757                        (info->nlh->nlmsg_flags & NLM_F_REPLACE));
758         int add = (!info->nlh ||
759                    (info->nlh->nlmsg_flags & NLM_F_CREATE));
760         int found = 0;
761         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
762         int err;
763
764         ins = &fn->leaf;
765
766         for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
767                 /*
768                  *      Search for duplicates
769                  */
770
771                 if (iter->rt6i_metric == rt->rt6i_metric) {
772                         /*
773                          *      Same priority level
774                          */
775                         if (info->nlh &&
776                             (info->nlh->nlmsg_flags & NLM_F_EXCL))
777                                 return -EEXIST;
778                         if (replace) {
779                                 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
780                                         found++;
781                                         break;
782                                 }
783                                 fallback_ins = fallback_ins ?: ins;
784                                 goto next_iter;
785                         }
786
787                         if (rt6_duplicate_nexthop(iter, rt)) {
788                                 if (rt->rt6i_nsiblings)
789                                         rt->rt6i_nsiblings = 0;
790                                 if (!(iter->rt6i_flags & RTF_EXPIRES))
791                                         return -EEXIST;
792                                 if (!(rt->rt6i_flags & RTF_EXPIRES))
793                                         rt6_clean_expires(iter);
794                                 else
795                                         rt6_set_expires(iter, rt->dst.expires);
796                                 iter->rt6i_pmtu = rt->rt6i_pmtu;
797                                 return -EEXIST;
798                         }
799                         /* If we have the same destination and the same metric,
800                          * but not the same gateway, then the route we try to
801                          * add is sibling to this route, increment our counter
802                          * of siblings, and later we will add our route to the
803                          * list.
804                          * Only static routes (which don't have flag
805                          * RTF_EXPIRES) are used for ECMPv6.
806                          *
807                          * To avoid long list, we only had siblings if the
808                          * route have a gateway.
809                          */
810                         if (rt_can_ecmp &&
811                             rt6_qualify_for_ecmp(iter))
812                                 rt->rt6i_nsiblings++;
813                 }
814
815                 if (iter->rt6i_metric > rt->rt6i_metric)
816                         break;
817
818 next_iter:
819                 ins = &iter->dst.rt6_next;
820         }
821
822         if (fallback_ins && !found) {
823                 /* No matching route with same ecmp-able-ness found, replace
824                  * first matching route
825                  */
826                 ins = fallback_ins;
827                 iter = *ins;
828                 found++;
829         }
830
831         /* Reset round-robin state, if necessary */
832         if (ins == &fn->leaf)
833                 fn->rr_ptr = NULL;
834
835         /* Link this route to others same route. */
836         if (rt->rt6i_nsiblings) {
837                 unsigned int rt6i_nsiblings;
838                 struct rt6_info *sibling, *temp_sibling;
839
840                 /* Find the first route that have the same metric */
841                 sibling = fn->leaf;
842                 while (sibling) {
843                         if (sibling->rt6i_metric == rt->rt6i_metric &&
844                             rt6_qualify_for_ecmp(sibling)) {
845                                 list_add_tail(&rt->rt6i_siblings,
846                                               &sibling->rt6i_siblings);
847                                 break;
848                         }
849                         sibling = sibling->dst.rt6_next;
850                 }
851                 /* For each sibling in the list, increment the counter of
852                  * siblings. BUG() if counters does not match, list of siblings
853                  * is broken!
854                  */
855                 rt6i_nsiblings = 0;
856                 list_for_each_entry_safe(sibling, temp_sibling,
857                                          &rt->rt6i_siblings, rt6i_siblings) {
858                         sibling->rt6i_nsiblings++;
859                         BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
860                         rt6i_nsiblings++;
861                 }
862                 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
863         }
864
865         /*
866          *      insert node
867          */
868         if (!replace) {
869                 if (!add)
870                         pr_warn("NLM_F_CREATE should be set when creating new route\n");
871
872 add:
873                 err = fib6_commit_metrics(&rt->dst, mxc);
874                 if (err)
875                         return err;
876
877                 rt->dst.rt6_next = iter;
878                 *ins = rt;
879                 rcu_assign_pointer(rt->rt6i_node, fn);
880                 atomic_inc(&rt->rt6i_ref);
881                 inet6_rt_notify(RTM_NEWROUTE, rt, info, 0);
882                 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
883
884                 if (!(fn->fn_flags & RTN_RTINFO)) {
885                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
886                         fn->fn_flags |= RTN_RTINFO;
887                 }
888
889         } else {
890                 int nsiblings;
891
892                 if (!found) {
893                         if (add)
894                                 goto add;
895                         pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
896                         return -ENOENT;
897                 }
898
899                 err = fib6_commit_metrics(&rt->dst, mxc);
900                 if (err)
901                         return err;
902
903                 *ins = rt;
904                 rcu_assign_pointer(rt->rt6i_node, fn);
905                 rt->dst.rt6_next = iter->dst.rt6_next;
906                 atomic_inc(&rt->rt6i_ref);
907                 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
908                 if (!(fn->fn_flags & RTN_RTINFO)) {
909                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
910                         fn->fn_flags |= RTN_RTINFO;
911                 }
912                 nsiblings = iter->rt6i_nsiblings;
913                 fib6_purge_rt(iter, fn, info->nl_net);
914                 if (fn->rr_ptr == iter)
915                         fn->rr_ptr = NULL;
916                 rt6_release(iter);
917
918                 if (nsiblings) {
919                         /* Replacing an ECMP route, remove all siblings */
920                         ins = &rt->dst.rt6_next;
921                         iter = *ins;
922                         while (iter) {
923                                 if (iter->rt6i_metric > rt->rt6i_metric)
924                                         break;
925                                 if (rt6_qualify_for_ecmp(iter)) {
926                                         *ins = iter->dst.rt6_next;
927                                         fib6_purge_rt(iter, fn, info->nl_net);
928                                         if (fn->rr_ptr == iter)
929                                                 fn->rr_ptr = NULL;
930                                         rt6_release(iter);
931                                         nsiblings--;
932                                 } else {
933                                         ins = &iter->dst.rt6_next;
934                                 }
935                                 iter = *ins;
936                         }
937                         WARN_ON(nsiblings != 0);
938                 }
939         }
940
941         return 0;
942 }
943
944 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
945 {
946         if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
947             (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
948                 mod_timer(&net->ipv6.ip6_fib_timer,
949                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
950 }
951
952 void fib6_force_start_gc(struct net *net)
953 {
954         if (!timer_pending(&net->ipv6.ip6_fib_timer))
955                 mod_timer(&net->ipv6.ip6_fib_timer,
956                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
957 }
958
959 /*
960  *      Add routing information to the routing tree.
961  *      <destination addr>/<source addr>
962  *      with source addr info in sub-trees
963  */
964
965 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
966              struct nl_info *info, struct mx6_config *mxc)
967 {
968         struct fib6_node *fn, *pn = NULL;
969         int err = -ENOMEM;
970         int allow_create = 1;
971         int replace_required = 0;
972         int sernum = fib6_new_sernum(info->nl_net);
973
974         if (WARN_ON_ONCE((rt->dst.flags & DST_NOCACHE) &&
975                          !atomic_read(&rt->dst.__refcnt)))
976                 return -EINVAL;
977
978         if (info->nlh) {
979                 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
980                         allow_create = 0;
981                 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
982                         replace_required = 1;
983         }
984         if (!allow_create && !replace_required)
985                 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
986
987         fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
988                         offsetof(struct rt6_info, rt6i_dst), allow_create,
989                         replace_required, sernum);
990         if (IS_ERR(fn)) {
991                 err = PTR_ERR(fn);
992                 fn = NULL;
993                 goto out;
994         }
995
996         pn = fn;
997
998 #ifdef CONFIG_IPV6_SUBTREES
999         if (rt->rt6i_src.plen) {
1000                 struct fib6_node *sn;
1001
1002                 if (!fn->subtree) {
1003                         struct fib6_node *sfn;
1004
1005                         /*
1006                          * Create subtree.
1007                          *
1008                          *              fn[main tree]
1009                          *              |
1010                          *              sfn[subtree root]
1011                          *                 \
1012                          *                  sn[new leaf node]
1013                          */
1014
1015                         /* Create subtree root node */
1016                         sfn = node_alloc();
1017                         if (!sfn)
1018                                 goto failure;
1019
1020                         sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
1021                         atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
1022                         sfn->fn_flags = RTN_ROOT;
1023                         sfn->fn_sernum = sernum;
1024
1025                         /* Now add the first leaf node to new subtree */
1026
1027                         sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
1028                                         rt->rt6i_src.plen,
1029                                         offsetof(struct rt6_info, rt6i_src),
1030                                         allow_create, replace_required, sernum);
1031
1032                         if (IS_ERR(sn)) {
1033                                 /* If it is failed, discard just allocated
1034                                    root, and then (in failure) stale node
1035                                    in main tree.
1036                                  */
1037                                 node_free_immediate(sfn);
1038                                 err = PTR_ERR(sn);
1039                                 goto failure;
1040                         }
1041
1042                         /* Now link new subtree to main tree */
1043                         sfn->parent = fn;
1044                         fn->subtree = sfn;
1045                 } else {
1046                         sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
1047                                         rt->rt6i_src.plen,
1048                                         offsetof(struct rt6_info, rt6i_src),
1049                                         allow_create, replace_required, sernum);
1050
1051                         if (IS_ERR(sn)) {
1052                                 err = PTR_ERR(sn);
1053                                 goto failure;
1054                         }
1055                 }
1056
1057                 if (!fn->leaf) {
1058                         fn->leaf = rt;
1059                         atomic_inc(&rt->rt6i_ref);
1060                 }
1061                 fn = sn;
1062         }
1063 #endif
1064
1065         err = fib6_add_rt2node(fn, rt, info, mxc);
1066         if (!err) {
1067                 fib6_start_gc(info->nl_net, rt);
1068                 if (!(rt->rt6i_flags & RTF_CACHE))
1069                         fib6_prune_clones(info->nl_net, pn);
1070                 rt->dst.flags &= ~DST_NOCACHE;
1071         }
1072
1073 out:
1074         if (err) {
1075 #ifdef CONFIG_IPV6_SUBTREES
1076                 /*
1077                  * If fib6_add_1 has cleared the old leaf pointer in the
1078                  * super-tree leaf node we have to find a new one for it.
1079                  */
1080                 if (pn != fn && pn->leaf == rt) {
1081                         pn->leaf = NULL;
1082                         atomic_dec(&rt->rt6i_ref);
1083                 }
1084                 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
1085                         pn->leaf = fib6_find_prefix(info->nl_net, pn);
1086 #if RT6_DEBUG >= 2
1087                         if (!pn->leaf) {
1088                                 WARN_ON(pn->leaf == NULL);
1089                                 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
1090                         }
1091 #endif
1092                         atomic_inc(&pn->leaf->rt6i_ref);
1093                 }
1094 #endif
1095                 goto failure;
1096         }
1097         return err;
1098
1099 failure:
1100         /* fn->leaf could be NULL if fn is an intermediate node and we
1101          * failed to add the new route to it in both subtree creation
1102          * failure and fib6_add_rt2node() failure case.
1103          * In both cases, fib6_repair_tree() should be called to fix
1104          * fn->leaf.
1105          */
1106         if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1107                 fib6_repair_tree(info->nl_net, fn);
1108         if (!(rt->dst.flags & DST_NOCACHE))
1109                 dst_free(&rt->dst);
1110         return err;
1111 }
1112
1113 /*
1114  *      Routing tree lookup
1115  *
1116  */
1117
1118 struct lookup_args {
1119         int                     offset;         /* key offset on rt6_info       */
1120         const struct in6_addr   *addr;          /* search key                   */
1121 };
1122
1123 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1124                                        struct lookup_args *args)
1125 {
1126         struct fib6_node *fn;
1127         __be32 dir;
1128
1129         if (unlikely(args->offset == 0))
1130                 return NULL;
1131
1132         /*
1133          *      Descend on a tree
1134          */
1135
1136         fn = root;
1137
1138         for (;;) {
1139                 struct fib6_node *next;
1140
1141                 dir = addr_bit_set(args->addr, fn->fn_bit);
1142
1143                 next = dir ? fn->right : fn->left;
1144
1145                 if (next) {
1146                         fn = next;
1147                         continue;
1148                 }
1149                 break;
1150         }
1151
1152         while (fn) {
1153                 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1154                         struct rt6key *key;
1155
1156                         key = (struct rt6key *) ((u8 *) fn->leaf +
1157                                                  args->offset);
1158
1159                         if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1160 #ifdef CONFIG_IPV6_SUBTREES
1161                                 if (fn->subtree) {
1162                                         struct fib6_node *sfn;
1163                                         sfn = fib6_lookup_1(fn->subtree,
1164                                                             args + 1);
1165                                         if (!sfn)
1166                                                 goto backtrack;
1167                                         fn = sfn;
1168                                 }
1169 #endif
1170                                 if (fn->fn_flags & RTN_RTINFO)
1171                                         return fn;
1172                         }
1173                 }
1174 #ifdef CONFIG_IPV6_SUBTREES
1175 backtrack:
1176 #endif
1177                 if (fn->fn_flags & RTN_ROOT)
1178                         break;
1179
1180                 fn = fn->parent;
1181         }
1182
1183         return NULL;
1184 }
1185
1186 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1187                               const struct in6_addr *saddr)
1188 {
1189         struct fib6_node *fn;
1190         struct lookup_args args[] = {
1191                 {
1192                         .offset = offsetof(struct rt6_info, rt6i_dst),
1193                         .addr = daddr,
1194                 },
1195 #ifdef CONFIG_IPV6_SUBTREES
1196                 {
1197                         .offset = offsetof(struct rt6_info, rt6i_src),
1198                         .addr = saddr,
1199                 },
1200 #endif
1201                 {
1202                         .offset = 0,    /* sentinel */
1203                 }
1204         };
1205
1206         fn = fib6_lookup_1(root, daddr ? args : args + 1);
1207         if (!fn || fn->fn_flags & RTN_TL_ROOT)
1208                 fn = root;
1209
1210         return fn;
1211 }
1212
1213 /*
1214  *      Get node with specified destination prefix (and source prefix,
1215  *      if subtrees are used)
1216  */
1217
1218
1219 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1220                                        const struct in6_addr *addr,
1221                                        int plen, int offset)
1222 {
1223         struct fib6_node *fn;
1224
1225         for (fn = root; fn ; ) {
1226                 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1227
1228                 /*
1229                  *      Prefix match
1230                  */
1231                 if (plen < fn->fn_bit ||
1232                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1233                         return NULL;
1234
1235                 if (plen == fn->fn_bit)
1236                         return fn;
1237
1238                 /*
1239                  *      We have more bits to go
1240                  */
1241                 if (addr_bit_set(addr, fn->fn_bit))
1242                         fn = fn->right;
1243                 else
1244                         fn = fn->left;
1245         }
1246         return NULL;
1247 }
1248
1249 struct fib6_node *fib6_locate(struct fib6_node *root,
1250                               const struct in6_addr *daddr, int dst_len,
1251                               const struct in6_addr *saddr, int src_len)
1252 {
1253         struct fib6_node *fn;
1254
1255         fn = fib6_locate_1(root, daddr, dst_len,
1256                            offsetof(struct rt6_info, rt6i_dst));
1257
1258 #ifdef CONFIG_IPV6_SUBTREES
1259         if (src_len) {
1260                 WARN_ON(saddr == NULL);
1261                 if (fn && fn->subtree)
1262                         fn = fib6_locate_1(fn->subtree, saddr, src_len,
1263                                            offsetof(struct rt6_info, rt6i_src));
1264         }
1265 #endif
1266
1267         if (fn && fn->fn_flags & RTN_RTINFO)
1268                 return fn;
1269
1270         return NULL;
1271 }
1272
1273
1274 /*
1275  *      Deletion
1276  *
1277  */
1278
1279 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1280 {
1281         if (fn->fn_flags & RTN_ROOT)
1282                 return net->ipv6.ip6_null_entry;
1283
1284         while (fn) {
1285                 if (fn->left)
1286                         return fn->left->leaf;
1287                 if (fn->right)
1288                         return fn->right->leaf;
1289
1290                 fn = FIB6_SUBTREE(fn);
1291         }
1292         return NULL;
1293 }
1294
1295 /*
1296  *      Called to trim the tree of intermediate nodes when possible. "fn"
1297  *      is the node we want to try and remove.
1298  */
1299
1300 static struct fib6_node *fib6_repair_tree(struct net *net,
1301                                            struct fib6_node *fn)
1302 {
1303         int children;
1304         int nstate;
1305         struct fib6_node *child, *pn;
1306         struct fib6_walker *w;
1307         int iter = 0;
1308
1309         for (;;) {
1310                 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1311                 iter++;
1312
1313                 WARN_ON(fn->fn_flags & RTN_RTINFO);
1314                 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1315                 WARN_ON(fn->leaf);
1316
1317                 children = 0;
1318                 child = NULL;
1319                 if (fn->right)
1320                         child = fn->right, children |= 1;
1321                 if (fn->left)
1322                         child = fn->left, children |= 2;
1323
1324                 if (children == 3 || FIB6_SUBTREE(fn)
1325 #ifdef CONFIG_IPV6_SUBTREES
1326                     /* Subtree root (i.e. fn) may have one child */
1327                     || (children && fn->fn_flags & RTN_ROOT)
1328 #endif
1329                     ) {
1330                         fn->leaf = fib6_find_prefix(net, fn);
1331 #if RT6_DEBUG >= 2
1332                         if (!fn->leaf) {
1333                                 WARN_ON(!fn->leaf);
1334                                 fn->leaf = net->ipv6.ip6_null_entry;
1335                         }
1336 #endif
1337                         atomic_inc(&fn->leaf->rt6i_ref);
1338                         return fn->parent;
1339                 }
1340
1341                 pn = fn->parent;
1342 #ifdef CONFIG_IPV6_SUBTREES
1343                 if (FIB6_SUBTREE(pn) == fn) {
1344                         WARN_ON(!(fn->fn_flags & RTN_ROOT));
1345                         FIB6_SUBTREE(pn) = NULL;
1346                         nstate = FWS_L;
1347                 } else {
1348                         WARN_ON(fn->fn_flags & RTN_ROOT);
1349 #endif
1350                         if (pn->right == fn)
1351                                 pn->right = child;
1352                         else if (pn->left == fn)
1353                                 pn->left = child;
1354 #if RT6_DEBUG >= 2
1355                         else
1356                                 WARN_ON(1);
1357 #endif
1358                         if (child)
1359                                 child->parent = pn;
1360                         nstate = FWS_R;
1361 #ifdef CONFIG_IPV6_SUBTREES
1362                 }
1363 #endif
1364
1365                 read_lock(&fib6_walker_lock);
1366                 FOR_WALKERS(w) {
1367                         if (!child) {
1368                                 if (w->root == fn) {
1369                                         w->root = w->node = NULL;
1370                                         RT6_TRACE("W %p adjusted by delroot 1\n", w);
1371                                 } else if (w->node == fn) {
1372                                         RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1373                                         w->node = pn;
1374                                         w->state = nstate;
1375                                 }
1376                         } else {
1377                                 if (w->root == fn) {
1378                                         w->root = child;
1379                                         RT6_TRACE("W %p adjusted by delroot 2\n", w);
1380                                 }
1381                                 if (w->node == fn) {
1382                                         w->node = child;
1383                                         if (children&2) {
1384                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1385                                                 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1386                                         } else {
1387                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1388                                                 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1389                                         }
1390                                 }
1391                         }
1392                 }
1393                 read_unlock(&fib6_walker_lock);
1394
1395                 node_free(fn);
1396                 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1397                         return pn;
1398
1399                 rt6_release(pn->leaf);
1400                 pn->leaf = NULL;
1401                 fn = pn;
1402         }
1403 }
1404
1405 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1406                            struct nl_info *info)
1407 {
1408         struct fib6_walker *w;
1409         struct rt6_info *rt = *rtp;
1410         struct net *net = info->nl_net;
1411
1412         RT6_TRACE("fib6_del_route\n");
1413
1414         /* Unlink it */
1415         *rtp = rt->dst.rt6_next;
1416         rt->rt6i_node = NULL;
1417         net->ipv6.rt6_stats->fib_rt_entries--;
1418         net->ipv6.rt6_stats->fib_discarded_routes++;
1419
1420         /* Reset round-robin state, if necessary */
1421         if (fn->rr_ptr == rt)
1422                 fn->rr_ptr = NULL;
1423
1424         /* Remove this entry from other siblings */
1425         if (rt->rt6i_nsiblings) {
1426                 struct rt6_info *sibling, *next_sibling;
1427
1428                 list_for_each_entry_safe(sibling, next_sibling,
1429                                          &rt->rt6i_siblings, rt6i_siblings)
1430                         sibling->rt6i_nsiblings--;
1431                 rt->rt6i_nsiblings = 0;
1432                 list_del_init(&rt->rt6i_siblings);
1433         }
1434
1435         /* Adjust walkers */
1436         read_lock(&fib6_walker_lock);
1437         FOR_WALKERS(w) {
1438                 if (w->state == FWS_C && w->leaf == rt) {
1439                         RT6_TRACE("walker %p adjusted by delroute\n", w);
1440                         w->leaf = rt->dst.rt6_next;
1441                         if (!w->leaf)
1442                                 w->state = FWS_U;
1443                 }
1444         }
1445         read_unlock(&fib6_walker_lock);
1446
1447         rt->dst.rt6_next = NULL;
1448
1449         /* If it was last route, expunge its radix tree node */
1450         if (!fn->leaf) {
1451                 fn->fn_flags &= ~RTN_RTINFO;
1452                 net->ipv6.rt6_stats->fib_route_nodes--;
1453                 fn = fib6_repair_tree(net, fn);
1454         }
1455
1456         fib6_purge_rt(rt, fn, net);
1457
1458         inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1459         rt6_release(rt);
1460 }
1461
1462 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1463 {
1464         struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1465                                     lockdep_is_held(&rt->rt6i_table->tb6_lock));
1466         struct net *net = info->nl_net;
1467         struct rt6_info **rtp;
1468
1469 #if RT6_DEBUG >= 2
1470         if (rt->dst.obsolete > 0) {
1471                 WARN_ON(fn);
1472                 return -ENOENT;
1473         }
1474 #endif
1475         if (!fn || rt == net->ipv6.ip6_null_entry)
1476                 return -ENOENT;
1477
1478         WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1479
1480         if (!(rt->rt6i_flags & RTF_CACHE)) {
1481                 struct fib6_node *pn = fn;
1482 #ifdef CONFIG_IPV6_SUBTREES
1483                 /* clones of this route might be in another subtree */
1484                 if (rt->rt6i_src.plen) {
1485                         while (!(pn->fn_flags & RTN_ROOT))
1486                                 pn = pn->parent;
1487                         pn = pn->parent;
1488                 }
1489 #endif
1490                 fib6_prune_clones(info->nl_net, pn);
1491         }
1492
1493         /*
1494          *      Walk the leaf entries looking for ourself
1495          */
1496
1497         for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1498                 if (*rtp == rt) {
1499                         fib6_del_route(fn, rtp, info);
1500                         return 0;
1501                 }
1502         }
1503         return -ENOENT;
1504 }
1505
1506 /*
1507  *      Tree traversal function.
1508  *
1509  *      Certainly, it is not interrupt safe.
1510  *      However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1511  *      It means, that we can modify tree during walking
1512  *      and use this function for garbage collection, clone pruning,
1513  *      cleaning tree when a device goes down etc. etc.
1514  *
1515  *      It guarantees that every node will be traversed,
1516  *      and that it will be traversed only once.
1517  *
1518  *      Callback function w->func may return:
1519  *      0 -> continue walking.
1520  *      positive value -> walking is suspended (used by tree dumps,
1521  *      and probably by gc, if it will be split to several slices)
1522  *      negative value -> terminate walking.
1523  *
1524  *      The function itself returns:
1525  *      0   -> walk is complete.
1526  *      >0  -> walk is incomplete (i.e. suspended)
1527  *      <0  -> walk is terminated by an error.
1528  */
1529
1530 static int fib6_walk_continue(struct fib6_walker *w)
1531 {
1532         struct fib6_node *fn, *pn;
1533
1534         for (;;) {
1535                 fn = w->node;
1536                 if (!fn)
1537                         return 0;
1538
1539                 if (w->prune && fn != w->root &&
1540                     fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1541                         w->state = FWS_C;
1542                         w->leaf = fn->leaf;
1543                 }
1544                 switch (w->state) {
1545 #ifdef CONFIG_IPV6_SUBTREES
1546                 case FWS_S:
1547                         if (FIB6_SUBTREE(fn)) {
1548                                 w->node = FIB6_SUBTREE(fn);
1549                                 continue;
1550                         }
1551                         w->state = FWS_L;
1552 #endif
1553                 case FWS_L:
1554                         if (fn->left) {
1555                                 w->node = fn->left;
1556                                 w->state = FWS_INIT;
1557                                 continue;
1558                         }
1559                         w->state = FWS_R;
1560                 case FWS_R:
1561                         if (fn->right) {
1562                                 w->node = fn->right;
1563                                 w->state = FWS_INIT;
1564                                 continue;
1565                         }
1566                         w->state = FWS_C;
1567                         w->leaf = fn->leaf;
1568                 case FWS_C:
1569                         if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1570                                 int err;
1571
1572                                 if (w->skip) {
1573                                         w->skip--;
1574                                         goto skip;
1575                                 }
1576
1577                                 err = w->func(w);
1578                                 if (err)
1579                                         return err;
1580
1581                                 w->count++;
1582                                 continue;
1583                         }
1584 skip:
1585                         w->state = FWS_U;
1586                 case FWS_U:
1587                         if (fn == w->root)
1588                                 return 0;
1589                         pn = fn->parent;
1590                         w->node = pn;
1591 #ifdef CONFIG_IPV6_SUBTREES
1592                         if (FIB6_SUBTREE(pn) == fn) {
1593                                 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1594                                 w->state = FWS_L;
1595                                 continue;
1596                         }
1597 #endif
1598                         if (pn->left == fn) {
1599                                 w->state = FWS_R;
1600                                 continue;
1601                         }
1602                         if (pn->right == fn) {
1603                                 w->state = FWS_C;
1604                                 w->leaf = w->node->leaf;
1605                                 continue;
1606                         }
1607 #if RT6_DEBUG >= 2
1608                         WARN_ON(1);
1609 #endif
1610                 }
1611         }
1612 }
1613
1614 static int fib6_walk(struct fib6_walker *w)
1615 {
1616         int res;
1617
1618         w->state = FWS_INIT;
1619         w->node = w->root;
1620
1621         fib6_walker_link(w);
1622         res = fib6_walk_continue(w);
1623         if (res <= 0)
1624                 fib6_walker_unlink(w);
1625         return res;
1626 }
1627
1628 static int fib6_clean_node(struct fib6_walker *w)
1629 {
1630         int res;
1631         struct rt6_info *rt;
1632         struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1633         struct nl_info info = {
1634                 .nl_net = c->net,
1635         };
1636
1637         if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1638             w->node->fn_sernum != c->sernum)
1639                 w->node->fn_sernum = c->sernum;
1640
1641         if (!c->func) {
1642                 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1643                 w->leaf = NULL;
1644                 return 0;
1645         }
1646
1647         for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1648                 res = c->func(rt, c->arg);
1649                 if (res < 0) {
1650                         w->leaf = rt;
1651                         res = fib6_del(rt, &info);
1652                         if (res) {
1653 #if RT6_DEBUG >= 2
1654                                 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1655                                          __func__, rt,
1656                                          rcu_access_pointer(rt->rt6i_node),
1657                                          res);
1658 #endif
1659                                 continue;
1660                         }
1661                         return 0;
1662                 }
1663                 WARN_ON(res != 0);
1664         }
1665         w->leaf = rt;
1666         return 0;
1667 }
1668
1669 /*
1670  *      Convenient frontend to tree walker.
1671  *
1672  *      func is called on each route.
1673  *              It may return -1 -> delete this route.
1674  *                            0  -> continue walking
1675  *
1676  *      prune==1 -> only immediate children of node (certainly,
1677  *      ignoring pure split nodes) will be scanned.
1678  */
1679
1680 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1681                             int (*func)(struct rt6_info *, void *arg),
1682                             bool prune, int sernum, void *arg)
1683 {
1684         struct fib6_cleaner c;
1685
1686         c.w.root = root;
1687         c.w.func = fib6_clean_node;
1688         c.w.prune = prune;
1689         c.w.count = 0;
1690         c.w.skip = 0;
1691         c.func = func;
1692         c.sernum = sernum;
1693         c.arg = arg;
1694         c.net = net;
1695
1696         fib6_walk(&c.w);
1697 }
1698
1699 static void __fib6_clean_all(struct net *net,
1700                              int (*func)(struct rt6_info *, void *),
1701                              int sernum, void *arg)
1702 {
1703         struct fib6_table *table;
1704         struct hlist_head *head;
1705         unsigned int h;
1706
1707         rcu_read_lock();
1708         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1709                 head = &net->ipv6.fib_table_hash[h];
1710                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1711                         write_lock_bh(&table->tb6_lock);
1712                         fib6_clean_tree(net, &table->tb6_root,
1713                                         func, false, sernum, arg);
1714                         write_unlock_bh(&table->tb6_lock);
1715                 }
1716         }
1717         rcu_read_unlock();
1718 }
1719
1720 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1721                     void *arg)
1722 {
1723         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1724 }
1725
1726 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1727 {
1728         if (rt->rt6i_flags & RTF_CACHE) {
1729                 RT6_TRACE("pruning clone %p\n", rt);
1730                 return -1;
1731         }
1732
1733         return 0;
1734 }
1735
1736 static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1737 {
1738         fib6_clean_tree(net, fn, fib6_prune_clone, true,
1739                         FIB6_NO_SERNUM_CHANGE, NULL);
1740 }
1741
1742 static void fib6_flush_trees(struct net *net)
1743 {
1744         int new_sernum = fib6_new_sernum(net);
1745
1746         __fib6_clean_all(net, NULL, new_sernum, NULL);
1747 }
1748
1749 /*
1750  *      Garbage collection
1751  */
1752
1753 static struct fib6_gc_args
1754 {
1755         int                     timeout;
1756         int                     more;
1757 } gc_args;
1758
1759 static int fib6_age(struct rt6_info *rt, void *arg)
1760 {
1761         unsigned long now = jiffies;
1762
1763         /*
1764          *      check addrconf expiration here.
1765          *      Routes are expired even if they are in use.
1766          *
1767          *      Also age clones. Note, that clones are aged out
1768          *      only if they are not in use now.
1769          */
1770
1771         if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1772                 if (time_after(now, rt->dst.expires)) {
1773                         RT6_TRACE("expiring %p\n", rt);
1774                         return -1;
1775                 }
1776                 gc_args.more++;
1777         } else if (rt->rt6i_flags & RTF_CACHE) {
1778                 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1779                     time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1780                         RT6_TRACE("aging clone %p\n", rt);
1781                         return -1;
1782                 } else if (rt->rt6i_flags & RTF_GATEWAY) {
1783                         struct neighbour *neigh;
1784                         __u8 neigh_flags = 0;
1785
1786                         neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1787                         if (neigh) {
1788                                 neigh_flags = neigh->flags;
1789                                 neigh_release(neigh);
1790                         }
1791                         if (!(neigh_flags & NTF_ROUTER)) {
1792                                 RT6_TRACE("purging route %p via non-router but gateway\n",
1793                                           rt);
1794                                 return -1;
1795                         }
1796                 }
1797                 gc_args.more++;
1798         }
1799
1800         return 0;
1801 }
1802
1803 static DEFINE_SPINLOCK(fib6_gc_lock);
1804
1805 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1806 {
1807         unsigned long now;
1808
1809         if (force) {
1810                 spin_lock_bh(&fib6_gc_lock);
1811         } else if (!spin_trylock_bh(&fib6_gc_lock)) {
1812                 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1813                 return;
1814         }
1815         gc_args.timeout = expires ? (int)expires :
1816                           net->ipv6.sysctl.ip6_rt_gc_interval;
1817
1818         gc_args.more = icmp6_dst_gc();
1819
1820         fib6_clean_all(net, fib6_age, NULL);
1821         now = jiffies;
1822         net->ipv6.ip6_rt_last_gc = now;
1823
1824         if (gc_args.more)
1825                 mod_timer(&net->ipv6.ip6_fib_timer,
1826                           round_jiffies(now
1827                                         + net->ipv6.sysctl.ip6_rt_gc_interval));
1828         else
1829                 del_timer(&net->ipv6.ip6_fib_timer);
1830         spin_unlock_bh(&fib6_gc_lock);
1831 }
1832
1833 static void fib6_gc_timer_cb(unsigned long arg)
1834 {
1835         fib6_run_gc(0, (struct net *)arg, true);
1836 }
1837
1838 static int __net_init fib6_net_init(struct net *net)
1839 {
1840         size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1841
1842         setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1843
1844         net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1845         if (!net->ipv6.rt6_stats)
1846                 goto out_timer;
1847
1848         /* Avoid false sharing : Use at least a full cache line */
1849         size = max_t(size_t, size, L1_CACHE_BYTES);
1850
1851         net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1852         if (!net->ipv6.fib_table_hash)
1853                 goto out_rt6_stats;
1854
1855         net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1856                                           GFP_KERNEL);
1857         if (!net->ipv6.fib6_main_tbl)
1858                 goto out_fib_table_hash;
1859
1860         net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1861         net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1862         net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1863                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1864         inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1865
1866 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1867         net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1868                                            GFP_KERNEL);
1869         if (!net->ipv6.fib6_local_tbl)
1870                 goto out_fib6_main_tbl;
1871         net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1872         net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1873         net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1874                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1875         inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1876 #endif
1877         fib6_tables_init(net);
1878
1879         return 0;
1880
1881 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1882 out_fib6_main_tbl:
1883         kfree(net->ipv6.fib6_main_tbl);
1884 #endif
1885 out_fib_table_hash:
1886         kfree(net->ipv6.fib_table_hash);
1887 out_rt6_stats:
1888         kfree(net->ipv6.rt6_stats);
1889 out_timer:
1890         return -ENOMEM;
1891 }
1892
1893 static void fib6_net_exit(struct net *net)
1894 {
1895         unsigned int i;
1896
1897         rt6_ifdown(net, NULL);
1898         del_timer_sync(&net->ipv6.ip6_fib_timer);
1899
1900         for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
1901                 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
1902                 struct hlist_node *tmp;
1903                 struct fib6_table *tb;
1904
1905                 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
1906                         hlist_del(&tb->tb6_hlist);
1907                         fib6_free_table(tb);
1908                 }
1909         }
1910
1911         kfree(net->ipv6.fib_table_hash);
1912         kfree(net->ipv6.rt6_stats);
1913 }
1914
1915 static struct pernet_operations fib6_net_ops = {
1916         .init = fib6_net_init,
1917         .exit = fib6_net_exit,
1918 };
1919
1920 int __init fib6_init(void)
1921 {
1922         int ret = -ENOMEM;
1923
1924         fib6_node_kmem = kmem_cache_create("fib6_nodes",
1925                                            sizeof(struct fib6_node),
1926                                            0, SLAB_HWCACHE_ALIGN,
1927                                            NULL);
1928         if (!fib6_node_kmem)
1929                 goto out;
1930
1931         ret = register_pernet_subsys(&fib6_net_ops);
1932         if (ret)
1933                 goto out_kmem_cache_create;
1934
1935         ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1936                               NULL);
1937         if (ret)
1938                 goto out_unregister_subsys;
1939
1940         __fib6_flush_trees = fib6_flush_trees;
1941 out:
1942         return ret;
1943
1944 out_unregister_subsys:
1945         unregister_pernet_subsys(&fib6_net_ops);
1946 out_kmem_cache_create:
1947         kmem_cache_destroy(fib6_node_kmem);
1948         goto out;
1949 }
1950
1951 void fib6_gc_cleanup(void)
1952 {
1953         unregister_pernet_subsys(&fib6_net_ops);
1954         kmem_cache_destroy(fib6_node_kmem);
1955 }
1956
1957 #ifdef CONFIG_PROC_FS
1958
1959 struct ipv6_route_iter {
1960         struct seq_net_private p;
1961         struct fib6_walker w;
1962         loff_t skip;
1963         struct fib6_table *tbl;
1964         int sernum;
1965 };
1966
1967 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
1968 {
1969         struct rt6_info *rt = v;
1970         struct ipv6_route_iter *iter = seq->private;
1971
1972         seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
1973
1974 #ifdef CONFIG_IPV6_SUBTREES
1975         seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
1976 #else
1977         seq_puts(seq, "00000000000000000000000000000000 00 ");
1978 #endif
1979         if (rt->rt6i_flags & RTF_GATEWAY)
1980                 seq_printf(seq, "%pi6", &rt->rt6i_gateway);
1981         else
1982                 seq_puts(seq, "00000000000000000000000000000000");
1983
1984         seq_printf(seq, " %08x %08x %08x %08x %8s\n",
1985                    rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
1986                    rt->dst.__use, rt->rt6i_flags,
1987                    rt->dst.dev ? rt->dst.dev->name : "");
1988         iter->w.leaf = NULL;
1989         return 0;
1990 }
1991
1992 static int ipv6_route_yield(struct fib6_walker *w)
1993 {
1994         struct ipv6_route_iter *iter = w->args;
1995
1996         if (!iter->skip)
1997                 return 1;
1998
1999         do {
2000                 iter->w.leaf = iter->w.leaf->dst.rt6_next;
2001                 iter->skip--;
2002                 if (!iter->skip && iter->w.leaf)
2003                         return 1;
2004         } while (iter->w.leaf);
2005
2006         return 0;
2007 }
2008
2009 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter)
2010 {
2011         memset(&iter->w, 0, sizeof(iter->w));
2012         iter->w.func = ipv6_route_yield;
2013         iter->w.root = &iter->tbl->tb6_root;
2014         iter->w.state = FWS_INIT;
2015         iter->w.node = iter->w.root;
2016         iter->w.args = iter;
2017         iter->sernum = iter->w.root->fn_sernum;
2018         INIT_LIST_HEAD(&iter->w.lh);
2019         fib6_walker_link(&iter->w);
2020 }
2021
2022 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2023                                                     struct net *net)
2024 {
2025         unsigned int h;
2026         struct hlist_node *node;
2027
2028         if (tbl) {
2029                 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2030                 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2031         } else {
2032                 h = 0;
2033                 node = NULL;
2034         }
2035
2036         while (!node && h < FIB6_TABLE_HASHSZ) {
2037                 node = rcu_dereference_bh(
2038                         hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2039         }
2040         return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2041 }
2042
2043 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2044 {
2045         if (iter->sernum != iter->w.root->fn_sernum) {
2046                 iter->sernum = iter->w.root->fn_sernum;
2047                 iter->w.state = FWS_INIT;
2048                 iter->w.node = iter->w.root;
2049                 WARN_ON(iter->w.skip);
2050                 iter->w.skip = iter->w.count;
2051         }
2052 }
2053
2054 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2055 {
2056         int r;
2057         struct rt6_info *n;
2058         struct net *net = seq_file_net(seq);
2059         struct ipv6_route_iter *iter = seq->private;
2060
2061         if (!v)
2062                 goto iter_table;
2063
2064         n = ((struct rt6_info *)v)->dst.rt6_next;
2065         if (n) {
2066                 ++*pos;
2067                 return n;
2068         }
2069
2070 iter_table:
2071         ipv6_route_check_sernum(iter);
2072         read_lock(&iter->tbl->tb6_lock);
2073         r = fib6_walk_continue(&iter->w);
2074         read_unlock(&iter->tbl->tb6_lock);
2075         if (r > 0) {
2076                 if (v)
2077                         ++*pos;
2078                 return iter->w.leaf;
2079         } else if (r < 0) {
2080                 fib6_walker_unlink(&iter->w);
2081                 return NULL;
2082         }
2083         fib6_walker_unlink(&iter->w);
2084
2085         iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2086         if (!iter->tbl)
2087                 return NULL;
2088
2089         ipv6_route_seq_setup_walk(iter);
2090         goto iter_table;
2091 }
2092
2093 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2094         __acquires(RCU_BH)
2095 {
2096         struct net *net = seq_file_net(seq);
2097         struct ipv6_route_iter *iter = seq->private;
2098
2099         rcu_read_lock_bh();
2100         iter->tbl = ipv6_route_seq_next_table(NULL, net);
2101         iter->skip = *pos;
2102
2103         if (iter->tbl) {
2104                 ipv6_route_seq_setup_walk(iter);
2105                 return ipv6_route_seq_next(seq, NULL, pos);
2106         } else {
2107                 return NULL;
2108         }
2109 }
2110
2111 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2112 {
2113         struct fib6_walker *w = &iter->w;
2114         return w->node && !(w->state == FWS_U && w->node == w->root);
2115 }
2116
2117 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2118         __releases(RCU_BH)
2119 {
2120         struct ipv6_route_iter *iter = seq->private;
2121
2122         if (ipv6_route_iter_active(iter))
2123                 fib6_walker_unlink(&iter->w);
2124
2125         rcu_read_unlock_bh();
2126 }
2127
2128 static const struct seq_operations ipv6_route_seq_ops = {
2129         .start  = ipv6_route_seq_start,
2130         .next   = ipv6_route_seq_next,
2131         .stop   = ipv6_route_seq_stop,
2132         .show   = ipv6_route_seq_show
2133 };
2134
2135 int ipv6_route_open(struct inode *inode, struct file *file)
2136 {
2137         return seq_open_net(inode, file, &ipv6_route_seq_ops,
2138                             sizeof(struct ipv6_route_iter));
2139 }
2140
2141 #endif /* CONFIG_PROC_FS */