]> asedeno.scripts.mit.edu Git - linux.git/blob - net/ipv6/ip6_fib.c
net/smc: transfer fasync_list in case of fallback
[linux.git] / net / ipv6 / ip6_fib.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *      Linux INET6 implementation
4  *      Forwarding Information Database
5  *
6  *      Authors:
7  *      Pedro Roque             <roque@di.fc.ul.pt>
8  *
9  *      Changes:
10  *      Yuji SEKIYA @USAGI:     Support default route on router node;
11  *                              remove ip6_null_entry from the top of
12  *                              routing table.
13  *      Ville Nuorvala:         Fixed routing subtrees.
14  */
15
16 #define pr_fmt(fmt) "IPv6: " fmt
17
18 #include <linux/errno.h>
19 #include <linux/types.h>
20 #include <linux/net.h>
21 #include <linux/route.h>
22 #include <linux/netdevice.h>
23 #include <linux/in6.h>
24 #include <linux/init.h>
25 #include <linux/list.h>
26 #include <linux/slab.h>
27
28 #include <net/ip.h>
29 #include <net/ipv6.h>
30 #include <net/ndisc.h>
31 #include <net/addrconf.h>
32 #include <net/lwtunnel.h>
33 #include <net/fib_notifier.h>
34
35 #include <net/ip6_fib.h>
36 #include <net/ip6_route.h>
37
38 static struct kmem_cache *fib6_node_kmem __read_mostly;
39
40 struct fib6_cleaner {
41         struct fib6_walker w;
42         struct net *net;
43         int (*func)(struct fib6_info *, void *arg);
44         int sernum;
45         void *arg;
46         bool skip_notify;
47 };
48
49 #ifdef CONFIG_IPV6_SUBTREES
50 #define FWS_INIT FWS_S
51 #else
52 #define FWS_INIT FWS_L
53 #endif
54
55 static struct fib6_info *fib6_find_prefix(struct net *net,
56                                          struct fib6_table *table,
57                                          struct fib6_node *fn);
58 static struct fib6_node *fib6_repair_tree(struct net *net,
59                                           struct fib6_table *table,
60                                           struct fib6_node *fn);
61 static int fib6_walk(struct net *net, struct fib6_walker *w);
62 static int fib6_walk_continue(struct fib6_walker *w);
63
64 /*
65  *      A routing update causes an increase of the serial number on the
66  *      affected subtree. This allows for cached routes to be asynchronously
67  *      tested when modifications are made to the destination cache as a
68  *      result of redirects, path MTU changes, etc.
69  */
70
71 static void fib6_gc_timer_cb(struct timer_list *t);
72
73 #define FOR_WALKERS(net, w) \
74         list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
75
76 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
77 {
78         write_lock_bh(&net->ipv6.fib6_walker_lock);
79         list_add(&w->lh, &net->ipv6.fib6_walkers);
80         write_unlock_bh(&net->ipv6.fib6_walker_lock);
81 }
82
83 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
84 {
85         write_lock_bh(&net->ipv6.fib6_walker_lock);
86         list_del(&w->lh);
87         write_unlock_bh(&net->ipv6.fib6_walker_lock);
88 }
89
90 static int fib6_new_sernum(struct net *net)
91 {
92         int new, old;
93
94         do {
95                 old = atomic_read(&net->ipv6.fib6_sernum);
96                 new = old < INT_MAX ? old + 1 : 1;
97         } while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
98                                 old, new) != old);
99         return new;
100 }
101
102 enum {
103         FIB6_NO_SERNUM_CHANGE = 0,
104 };
105
106 void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
107 {
108         struct fib6_node *fn;
109
110         fn = rcu_dereference_protected(f6i->fib6_node,
111                         lockdep_is_held(&f6i->fib6_table->tb6_lock));
112         if (fn)
113                 fn->fn_sernum = fib6_new_sernum(net);
114 }
115
116 /*
117  *      Auxiliary address test functions for the radix tree.
118  *
119  *      These assume a 32bit processor (although it will work on
120  *      64bit processors)
121  */
122
123 /*
124  *      test bit
125  */
126 #if defined(__LITTLE_ENDIAN)
127 # define BITOP_BE32_SWIZZLE     (0x1F & ~7)
128 #else
129 # define BITOP_BE32_SWIZZLE     0
130 #endif
131
132 static __be32 addr_bit_set(const void *token, int fn_bit)
133 {
134         const __be32 *addr = token;
135         /*
136          * Here,
137          *      1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
138          * is optimized version of
139          *      htonl(1 << ((~fn_bit)&0x1F))
140          * See include/asm-generic/bitops/le.h.
141          */
142         return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
143                addr[fn_bit >> 5];
144 }
145
146 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags, bool with_fib6_nh)
147 {
148         struct fib6_info *f6i;
149         size_t sz = sizeof(*f6i);
150
151         if (with_fib6_nh)
152                 sz += sizeof(struct fib6_nh);
153
154         f6i = kzalloc(sz, gfp_flags);
155         if (!f6i)
156                 return NULL;
157
158         /* fib6_siblings is a union with nh_list, so this initializes both */
159         INIT_LIST_HEAD(&f6i->fib6_siblings);
160         refcount_set(&f6i->fib6_ref, 1);
161
162         return f6i;
163 }
164
165 void fib6_info_destroy_rcu(struct rcu_head *head)
166 {
167         struct fib6_info *f6i = container_of(head, struct fib6_info, rcu);
168
169         WARN_ON(f6i->fib6_node);
170
171         if (f6i->nh)
172                 nexthop_put(f6i->nh);
173         else
174                 fib6_nh_release(f6i->fib6_nh);
175
176         ip_fib_metrics_put(f6i->fib6_metrics);
177         kfree(f6i);
178 }
179 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
180
181 static struct fib6_node *node_alloc(struct net *net)
182 {
183         struct fib6_node *fn;
184
185         fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
186         if (fn)
187                 net->ipv6.rt6_stats->fib_nodes++;
188
189         return fn;
190 }
191
192 static void node_free_immediate(struct net *net, struct fib6_node *fn)
193 {
194         kmem_cache_free(fib6_node_kmem, fn);
195         net->ipv6.rt6_stats->fib_nodes--;
196 }
197
198 static void node_free_rcu(struct rcu_head *head)
199 {
200         struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
201
202         kmem_cache_free(fib6_node_kmem, fn);
203 }
204
205 static void node_free(struct net *net, struct fib6_node *fn)
206 {
207         call_rcu(&fn->rcu, node_free_rcu);
208         net->ipv6.rt6_stats->fib_nodes--;
209 }
210
211 static void fib6_free_table(struct fib6_table *table)
212 {
213         inetpeer_invalidate_tree(&table->tb6_peers);
214         kfree(table);
215 }
216
217 static void fib6_link_table(struct net *net, struct fib6_table *tb)
218 {
219         unsigned int h;
220
221         /*
222          * Initialize table lock at a single place to give lockdep a key,
223          * tables aren't visible prior to being linked to the list.
224          */
225         spin_lock_init(&tb->tb6_lock);
226         h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
227
228         /*
229          * No protection necessary, this is the only list mutatation
230          * operation, tables never disappear once they exist.
231          */
232         hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
233 }
234
235 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
236
237 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
238 {
239         struct fib6_table *table;
240
241         table = kzalloc(sizeof(*table), GFP_ATOMIC);
242         if (table) {
243                 table->tb6_id = id;
244                 rcu_assign_pointer(table->tb6_root.leaf,
245                                    net->ipv6.fib6_null_entry);
246                 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
247                 inet_peer_base_init(&table->tb6_peers);
248         }
249
250         return table;
251 }
252
253 struct fib6_table *fib6_new_table(struct net *net, u32 id)
254 {
255         struct fib6_table *tb;
256
257         if (id == 0)
258                 id = RT6_TABLE_MAIN;
259         tb = fib6_get_table(net, id);
260         if (tb)
261                 return tb;
262
263         tb = fib6_alloc_table(net, id);
264         if (tb)
265                 fib6_link_table(net, tb);
266
267         return tb;
268 }
269 EXPORT_SYMBOL_GPL(fib6_new_table);
270
271 struct fib6_table *fib6_get_table(struct net *net, u32 id)
272 {
273         struct fib6_table *tb;
274         struct hlist_head *head;
275         unsigned int h;
276
277         if (id == 0)
278                 id = RT6_TABLE_MAIN;
279         h = id & (FIB6_TABLE_HASHSZ - 1);
280         rcu_read_lock();
281         head = &net->ipv6.fib_table_hash[h];
282         hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
283                 if (tb->tb6_id == id) {
284                         rcu_read_unlock();
285                         return tb;
286                 }
287         }
288         rcu_read_unlock();
289
290         return NULL;
291 }
292 EXPORT_SYMBOL_GPL(fib6_get_table);
293
294 static void __net_init fib6_tables_init(struct net *net)
295 {
296         fib6_link_table(net, net->ipv6.fib6_main_tbl);
297         fib6_link_table(net, net->ipv6.fib6_local_tbl);
298 }
299 #else
300
301 struct fib6_table *fib6_new_table(struct net *net, u32 id)
302 {
303         return fib6_get_table(net, id);
304 }
305
306 struct fib6_table *fib6_get_table(struct net *net, u32 id)
307 {
308           return net->ipv6.fib6_main_tbl;
309 }
310
311 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
312                                    const struct sk_buff *skb,
313                                    int flags, pol_lookup_t lookup)
314 {
315         struct rt6_info *rt;
316
317         rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
318         if (rt->dst.error == -EAGAIN) {
319                 ip6_rt_put_flags(rt, flags);
320                 rt = net->ipv6.ip6_null_entry;
321                 if (!(flags & RT6_LOOKUP_F_DST_NOREF))
322                         dst_hold(&rt->dst);
323         }
324
325         return &rt->dst;
326 }
327
328 /* called with rcu lock held; no reference taken on fib6_info */
329 int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
330                 struct fib6_result *res, int flags)
331 {
332         return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6,
333                                  res, flags);
334 }
335
336 static void __net_init fib6_tables_init(struct net *net)
337 {
338         fib6_link_table(net, net->ipv6.fib6_main_tbl);
339 }
340
341 #endif
342
343 unsigned int fib6_tables_seq_read(struct net *net)
344 {
345         unsigned int h, fib_seq = 0;
346
347         rcu_read_lock();
348         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
349                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
350                 struct fib6_table *tb;
351
352                 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
353                         fib_seq += tb->fib_seq;
354         }
355         rcu_read_unlock();
356
357         return fib_seq;
358 }
359
360 static int call_fib6_entry_notifier(struct notifier_block *nb,
361                                     enum fib_event_type event_type,
362                                     struct fib6_info *rt,
363                                     struct netlink_ext_ack *extack)
364 {
365         struct fib6_entry_notifier_info info = {
366                 .info.extack = extack,
367                 .rt = rt,
368         };
369
370         return call_fib6_notifier(nb, event_type, &info.info);
371 }
372
373 static int call_fib6_multipath_entry_notifier(struct notifier_block *nb,
374                                               enum fib_event_type event_type,
375                                               struct fib6_info *rt,
376                                               unsigned int nsiblings,
377                                               struct netlink_ext_ack *extack)
378 {
379         struct fib6_entry_notifier_info info = {
380                 .info.extack = extack,
381                 .rt = rt,
382                 .nsiblings = nsiblings,
383         };
384
385         return call_fib6_notifier(nb, event_type, &info.info);
386 }
387
388 int call_fib6_entry_notifiers(struct net *net,
389                               enum fib_event_type event_type,
390                               struct fib6_info *rt,
391                               struct netlink_ext_ack *extack)
392 {
393         struct fib6_entry_notifier_info info = {
394                 .info.extack = extack,
395                 .rt = rt,
396         };
397
398         rt->fib6_table->fib_seq++;
399         return call_fib6_notifiers(net, event_type, &info.info);
400 }
401
402 int call_fib6_multipath_entry_notifiers(struct net *net,
403                                         enum fib_event_type event_type,
404                                         struct fib6_info *rt,
405                                         unsigned int nsiblings,
406                                         struct netlink_ext_ack *extack)
407 {
408         struct fib6_entry_notifier_info info = {
409                 .info.extack = extack,
410                 .rt = rt,
411                 .nsiblings = nsiblings,
412         };
413
414         rt->fib6_table->fib_seq++;
415         return call_fib6_notifiers(net, event_type, &info.info);
416 }
417
418 int call_fib6_entry_notifiers_replace(struct net *net, struct fib6_info *rt)
419 {
420         struct fib6_entry_notifier_info info = {
421                 .rt = rt,
422                 .nsiblings = rt->fib6_nsiblings,
423         };
424
425         rt->fib6_table->fib_seq++;
426         return call_fib6_notifiers(net, FIB_EVENT_ENTRY_REPLACE, &info.info);
427 }
428
429 struct fib6_dump_arg {
430         struct net *net;
431         struct notifier_block *nb;
432         struct netlink_ext_ack *extack;
433 };
434
435 static int fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
436 {
437         enum fib_event_type fib_event = FIB_EVENT_ENTRY_REPLACE;
438         int err;
439
440         if (!rt || rt == arg->net->ipv6.fib6_null_entry)
441                 return 0;
442
443         if (rt->fib6_nsiblings)
444                 err = call_fib6_multipath_entry_notifier(arg->nb, fib_event,
445                                                          rt,
446                                                          rt->fib6_nsiblings,
447                                                          arg->extack);
448         else
449                 err = call_fib6_entry_notifier(arg->nb, fib_event, rt,
450                                                arg->extack);
451
452         return err;
453 }
454
455 static int fib6_node_dump(struct fib6_walker *w)
456 {
457         int err;
458
459         err = fib6_rt_dump(w->leaf, w->args);
460         w->leaf = NULL;
461         return err;
462 }
463
464 static int fib6_table_dump(struct net *net, struct fib6_table *tb,
465                            struct fib6_walker *w)
466 {
467         int err;
468
469         w->root = &tb->tb6_root;
470         spin_lock_bh(&tb->tb6_lock);
471         err = fib6_walk(net, w);
472         spin_unlock_bh(&tb->tb6_lock);
473         return err;
474 }
475
476 /* Called with rcu_read_lock() */
477 int fib6_tables_dump(struct net *net, struct notifier_block *nb,
478                      struct netlink_ext_ack *extack)
479 {
480         struct fib6_dump_arg arg;
481         struct fib6_walker *w;
482         unsigned int h;
483         int err = 0;
484
485         w = kzalloc(sizeof(*w), GFP_ATOMIC);
486         if (!w)
487                 return -ENOMEM;
488
489         w->func = fib6_node_dump;
490         arg.net = net;
491         arg.nb = nb;
492         arg.extack = extack;
493         w->args = &arg;
494
495         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
496                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
497                 struct fib6_table *tb;
498
499                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
500                         err = fib6_table_dump(net, tb, w);
501                         if (err < 0)
502                                 goto out;
503                 }
504         }
505
506 out:
507         kfree(w);
508
509         return err;
510 }
511
512 static int fib6_dump_node(struct fib6_walker *w)
513 {
514         int res;
515         struct fib6_info *rt;
516
517         for_each_fib6_walker_rt(w) {
518                 res = rt6_dump_route(rt, w->args, w->skip_in_node);
519                 if (res >= 0) {
520                         /* Frame is full, suspend walking */
521                         w->leaf = rt;
522
523                         /* We'll restart from this node, so if some routes were
524                          * already dumped, skip them next time.
525                          */
526                         w->skip_in_node += res;
527
528                         return 1;
529                 }
530                 w->skip_in_node = 0;
531
532                 /* Multipath routes are dumped in one route with the
533                  * RTA_MULTIPATH attribute. Jump 'rt' to point to the
534                  * last sibling of this route (no need to dump the
535                  * sibling routes again)
536                  */
537                 if (rt->fib6_nsiblings)
538                         rt = list_last_entry(&rt->fib6_siblings,
539                                              struct fib6_info,
540                                              fib6_siblings);
541         }
542         w->leaf = NULL;
543         return 0;
544 }
545
546 static void fib6_dump_end(struct netlink_callback *cb)
547 {
548         struct net *net = sock_net(cb->skb->sk);
549         struct fib6_walker *w = (void *)cb->args[2];
550
551         if (w) {
552                 if (cb->args[4]) {
553                         cb->args[4] = 0;
554                         fib6_walker_unlink(net, w);
555                 }
556                 cb->args[2] = 0;
557                 kfree(w);
558         }
559         cb->done = (void *)cb->args[3];
560         cb->args[1] = 3;
561 }
562
563 static int fib6_dump_done(struct netlink_callback *cb)
564 {
565         fib6_dump_end(cb);
566         return cb->done ? cb->done(cb) : 0;
567 }
568
569 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
570                            struct netlink_callback *cb)
571 {
572         struct net *net = sock_net(skb->sk);
573         struct fib6_walker *w;
574         int res;
575
576         w = (void *)cb->args[2];
577         w->root = &table->tb6_root;
578
579         if (cb->args[4] == 0) {
580                 w->count = 0;
581                 w->skip = 0;
582                 w->skip_in_node = 0;
583
584                 spin_lock_bh(&table->tb6_lock);
585                 res = fib6_walk(net, w);
586                 spin_unlock_bh(&table->tb6_lock);
587                 if (res > 0) {
588                         cb->args[4] = 1;
589                         cb->args[5] = w->root->fn_sernum;
590                 }
591         } else {
592                 if (cb->args[5] != w->root->fn_sernum) {
593                         /* Begin at the root if the tree changed */
594                         cb->args[5] = w->root->fn_sernum;
595                         w->state = FWS_INIT;
596                         w->node = w->root;
597                         w->skip = w->count;
598                         w->skip_in_node = 0;
599                 } else
600                         w->skip = 0;
601
602                 spin_lock_bh(&table->tb6_lock);
603                 res = fib6_walk_continue(w);
604                 spin_unlock_bh(&table->tb6_lock);
605                 if (res <= 0) {
606                         fib6_walker_unlink(net, w);
607                         cb->args[4] = 0;
608                 }
609         }
610
611         return res;
612 }
613
614 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
615 {
616         struct rt6_rtnl_dump_arg arg = { .filter.dump_exceptions = true,
617                                          .filter.dump_routes = true };
618         const struct nlmsghdr *nlh = cb->nlh;
619         struct net *net = sock_net(skb->sk);
620         unsigned int h, s_h;
621         unsigned int e = 0, s_e;
622         struct fib6_walker *w;
623         struct fib6_table *tb;
624         struct hlist_head *head;
625         int res = 0;
626
627         if (cb->strict_check) {
628                 int err;
629
630                 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
631                 if (err < 0)
632                         return err;
633         } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
634                 struct rtmsg *rtm = nlmsg_data(nlh);
635
636                 if (rtm->rtm_flags & RTM_F_PREFIX)
637                         arg.filter.flags = RTM_F_PREFIX;
638         }
639
640         w = (void *)cb->args[2];
641         if (!w) {
642                 /* New dump:
643                  *
644                  * 1. hook callback destructor.
645                  */
646                 cb->args[3] = (long)cb->done;
647                 cb->done = fib6_dump_done;
648
649                 /*
650                  * 2. allocate and initialize walker.
651                  */
652                 w = kzalloc(sizeof(*w), GFP_ATOMIC);
653                 if (!w)
654                         return -ENOMEM;
655                 w->func = fib6_dump_node;
656                 cb->args[2] = (long)w;
657         }
658
659         arg.skb = skb;
660         arg.cb = cb;
661         arg.net = net;
662         w->args = &arg;
663
664         if (arg.filter.table_id) {
665                 tb = fib6_get_table(net, arg.filter.table_id);
666                 if (!tb) {
667                         if (arg.filter.dump_all_families)
668                                 goto out;
669
670                         NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
671                         return -ENOENT;
672                 }
673
674                 if (!cb->args[0]) {
675                         res = fib6_dump_table(tb, skb, cb);
676                         if (!res)
677                                 cb->args[0] = 1;
678                 }
679                 goto out;
680         }
681
682         s_h = cb->args[0];
683         s_e = cb->args[1];
684
685         rcu_read_lock();
686         for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
687                 e = 0;
688                 head = &net->ipv6.fib_table_hash[h];
689                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
690                         if (e < s_e)
691                                 goto next;
692                         res = fib6_dump_table(tb, skb, cb);
693                         if (res != 0)
694                                 goto out_unlock;
695 next:
696                         e++;
697                 }
698         }
699 out_unlock:
700         rcu_read_unlock();
701         cb->args[1] = e;
702         cb->args[0] = h;
703 out:
704         res = res < 0 ? res : skb->len;
705         if (res <= 0)
706                 fib6_dump_end(cb);
707         return res;
708 }
709
710 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
711 {
712         if (!f6i)
713                 return;
714
715         if (f6i->fib6_metrics == &dst_default_metrics) {
716                 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
717
718                 if (!p)
719                         return;
720
721                 refcount_set(&p->refcnt, 1);
722                 f6i->fib6_metrics = p;
723         }
724
725         f6i->fib6_metrics->metrics[metric - 1] = val;
726 }
727
728 /*
729  *      Routing Table
730  *
731  *      return the appropriate node for a routing tree "add" operation
732  *      by either creating and inserting or by returning an existing
733  *      node.
734  */
735
736 static struct fib6_node *fib6_add_1(struct net *net,
737                                     struct fib6_table *table,
738                                     struct fib6_node *root,
739                                     struct in6_addr *addr, int plen,
740                                     int offset, int allow_create,
741                                     int replace_required,
742                                     struct netlink_ext_ack *extack)
743 {
744         struct fib6_node *fn, *in, *ln;
745         struct fib6_node *pn = NULL;
746         struct rt6key *key;
747         int     bit;
748         __be32  dir = 0;
749
750         RT6_TRACE("fib6_add_1\n");
751
752         /* insert node in tree */
753
754         fn = root;
755
756         do {
757                 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
758                                             lockdep_is_held(&table->tb6_lock));
759                 key = (struct rt6key *)((u8 *)leaf + offset);
760
761                 /*
762                  *      Prefix match
763                  */
764                 if (plen < fn->fn_bit ||
765                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
766                         if (!allow_create) {
767                                 if (replace_required) {
768                                         NL_SET_ERR_MSG(extack,
769                                                        "Can not replace route - no match found");
770                                         pr_warn("Can't replace route, no match found\n");
771                                         return ERR_PTR(-ENOENT);
772                                 }
773                                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
774                         }
775                         goto insert_above;
776                 }
777
778                 /*
779                  *      Exact match ?
780                  */
781
782                 if (plen == fn->fn_bit) {
783                         /* clean up an intermediate node */
784                         if (!(fn->fn_flags & RTN_RTINFO)) {
785                                 RCU_INIT_POINTER(fn->leaf, NULL);
786                                 fib6_info_release(leaf);
787                         /* remove null_entry in the root node */
788                         } else if (fn->fn_flags & RTN_TL_ROOT &&
789                                    rcu_access_pointer(fn->leaf) ==
790                                    net->ipv6.fib6_null_entry) {
791                                 RCU_INIT_POINTER(fn->leaf, NULL);
792                         }
793
794                         return fn;
795                 }
796
797                 /*
798                  *      We have more bits to go
799                  */
800
801                 /* Try to walk down on tree. */
802                 dir = addr_bit_set(addr, fn->fn_bit);
803                 pn = fn;
804                 fn = dir ?
805                      rcu_dereference_protected(fn->right,
806                                         lockdep_is_held(&table->tb6_lock)) :
807                      rcu_dereference_protected(fn->left,
808                                         lockdep_is_held(&table->tb6_lock));
809         } while (fn);
810
811         if (!allow_create) {
812                 /* We should not create new node because
813                  * NLM_F_REPLACE was specified without NLM_F_CREATE
814                  * I assume it is safe to require NLM_F_CREATE when
815                  * REPLACE flag is used! Later we may want to remove the
816                  * check for replace_required, because according
817                  * to netlink specification, NLM_F_CREATE
818                  * MUST be specified if new route is created.
819                  * That would keep IPv6 consistent with IPv4
820                  */
821                 if (replace_required) {
822                         NL_SET_ERR_MSG(extack,
823                                        "Can not replace route - no match found");
824                         pr_warn("Can't replace route, no match found\n");
825                         return ERR_PTR(-ENOENT);
826                 }
827                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
828         }
829         /*
830          *      We walked to the bottom of tree.
831          *      Create new leaf node without children.
832          */
833
834         ln = node_alloc(net);
835
836         if (!ln)
837                 return ERR_PTR(-ENOMEM);
838         ln->fn_bit = plen;
839         RCU_INIT_POINTER(ln->parent, pn);
840
841         if (dir)
842                 rcu_assign_pointer(pn->right, ln);
843         else
844                 rcu_assign_pointer(pn->left, ln);
845
846         return ln;
847
848
849 insert_above:
850         /*
851          * split since we don't have a common prefix anymore or
852          * we have a less significant route.
853          * we've to insert an intermediate node on the list
854          * this new node will point to the one we need to create
855          * and the current
856          */
857
858         pn = rcu_dereference_protected(fn->parent,
859                                        lockdep_is_held(&table->tb6_lock));
860
861         /* find 1st bit in difference between the 2 addrs.
862
863            See comment in __ipv6_addr_diff: bit may be an invalid value,
864            but if it is >= plen, the value is ignored in any case.
865          */
866
867         bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
868
869         /*
870          *              (intermediate)[in]
871          *                /        \
872          *      (new leaf node)[ln] (old node)[fn]
873          */
874         if (plen > bit) {
875                 in = node_alloc(net);
876                 ln = node_alloc(net);
877
878                 if (!in || !ln) {
879                         if (in)
880                                 node_free_immediate(net, in);
881                         if (ln)
882                                 node_free_immediate(net, ln);
883                         return ERR_PTR(-ENOMEM);
884                 }
885
886                 /*
887                  * new intermediate node.
888                  * RTN_RTINFO will
889                  * be off since that an address that chooses one of
890                  * the branches would not match less specific routes
891                  * in the other branch
892                  */
893
894                 in->fn_bit = bit;
895
896                 RCU_INIT_POINTER(in->parent, pn);
897                 in->leaf = fn->leaf;
898                 fib6_info_hold(rcu_dereference_protected(in->leaf,
899                                 lockdep_is_held(&table->tb6_lock)));
900
901                 /* update parent pointer */
902                 if (dir)
903                         rcu_assign_pointer(pn->right, in);
904                 else
905                         rcu_assign_pointer(pn->left, in);
906
907                 ln->fn_bit = plen;
908
909                 RCU_INIT_POINTER(ln->parent, in);
910                 rcu_assign_pointer(fn->parent, in);
911
912                 if (addr_bit_set(addr, bit)) {
913                         rcu_assign_pointer(in->right, ln);
914                         rcu_assign_pointer(in->left, fn);
915                 } else {
916                         rcu_assign_pointer(in->left, ln);
917                         rcu_assign_pointer(in->right, fn);
918                 }
919         } else { /* plen <= bit */
920
921                 /*
922                  *              (new leaf node)[ln]
923                  *                /        \
924                  *           (old node)[fn] NULL
925                  */
926
927                 ln = node_alloc(net);
928
929                 if (!ln)
930                         return ERR_PTR(-ENOMEM);
931
932                 ln->fn_bit = plen;
933
934                 RCU_INIT_POINTER(ln->parent, pn);
935
936                 if (addr_bit_set(&key->addr, plen))
937                         RCU_INIT_POINTER(ln->right, fn);
938                 else
939                         RCU_INIT_POINTER(ln->left, fn);
940
941                 rcu_assign_pointer(fn->parent, ln);
942
943                 if (dir)
944                         rcu_assign_pointer(pn->right, ln);
945                 else
946                         rcu_assign_pointer(pn->left, ln);
947         }
948         return ln;
949 }
950
951 static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
952                                   const struct fib6_info *match,
953                                   const struct fib6_table *table)
954 {
955         int cpu;
956
957         if (!fib6_nh->rt6i_pcpu)
958                 return;
959
960         /* release the reference to this fib entry from
961          * all of its cached pcpu routes
962          */
963         for_each_possible_cpu(cpu) {
964                 struct rt6_info **ppcpu_rt;
965                 struct rt6_info *pcpu_rt;
966
967                 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
968                 pcpu_rt = *ppcpu_rt;
969
970                 /* only dropping the 'from' reference if the cached route
971                  * is using 'match'. The cached pcpu_rt->from only changes
972                  * from a fib6_info to NULL (ip6_dst_destroy); it can never
973                  * change from one fib6_info reference to another
974                  */
975                 if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
976                         struct fib6_info *from;
977
978                         from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
979                         fib6_info_release(from);
980                 }
981         }
982 }
983
984 struct fib6_nh_pcpu_arg {
985         struct fib6_info        *from;
986         const struct fib6_table *table;
987 };
988
989 static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg)
990 {
991         struct fib6_nh_pcpu_arg *arg = _arg;
992
993         __fib6_drop_pcpu_from(nh, arg->from, arg->table);
994         return 0;
995 }
996
997 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
998                                 const struct fib6_table *table)
999 {
1000         /* Make sure rt6_make_pcpu_route() wont add other percpu routes
1001          * while we are cleaning them here.
1002          */
1003         f6i->fib6_destroying = 1;
1004         mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
1005
1006         if (f6i->nh) {
1007                 struct fib6_nh_pcpu_arg arg = {
1008                         .from = f6i,
1009                         .table = table
1010                 };
1011
1012                 nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from,
1013                                          &arg);
1014         } else {
1015                 struct fib6_nh *fib6_nh;
1016
1017                 fib6_nh = f6i->fib6_nh;
1018                 __fib6_drop_pcpu_from(fib6_nh, f6i, table);
1019         }
1020 }
1021
1022 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
1023                           struct net *net)
1024 {
1025         struct fib6_table *table = rt->fib6_table;
1026
1027         fib6_drop_pcpu_from(rt, table);
1028
1029         if (rt->nh && !list_empty(&rt->nh_list))
1030                 list_del_init(&rt->nh_list);
1031
1032         if (refcount_read(&rt->fib6_ref) != 1) {
1033                 /* This route is used as dummy address holder in some split
1034                  * nodes. It is not leaked, but it still holds other resources,
1035                  * which must be released in time. So, scan ascendant nodes
1036                  * and replace dummy references to this route with references
1037                  * to still alive ones.
1038                  */
1039                 while (fn) {
1040                         struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1041                                             lockdep_is_held(&table->tb6_lock));
1042                         struct fib6_info *new_leaf;
1043                         if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
1044                                 new_leaf = fib6_find_prefix(net, table, fn);
1045                                 fib6_info_hold(new_leaf);
1046
1047                                 rcu_assign_pointer(fn->leaf, new_leaf);
1048                                 fib6_info_release(rt);
1049                         }
1050                         fn = rcu_dereference_protected(fn->parent,
1051                                     lockdep_is_held(&table->tb6_lock));
1052                 }
1053         }
1054 }
1055
1056 /*
1057  *      Insert routing information in a node.
1058  */
1059
1060 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
1061                             struct nl_info *info,
1062                             struct netlink_ext_ack *extack)
1063 {
1064         struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1065                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1066         struct fib6_info *iter = NULL;
1067         struct fib6_info __rcu **ins;
1068         struct fib6_info __rcu **fallback_ins = NULL;
1069         int replace = (info->nlh &&
1070                        (info->nlh->nlmsg_flags & NLM_F_REPLACE));
1071         int add = (!info->nlh ||
1072                    (info->nlh->nlmsg_flags & NLM_F_CREATE));
1073         int found = 0;
1074         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
1075         bool notify_sibling_rt = false;
1076         u16 nlflags = NLM_F_EXCL;
1077         int err;
1078
1079         if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
1080                 nlflags |= NLM_F_APPEND;
1081
1082         ins = &fn->leaf;
1083
1084         for (iter = leaf; iter;
1085              iter = rcu_dereference_protected(iter->fib6_next,
1086                                 lockdep_is_held(&rt->fib6_table->tb6_lock))) {
1087                 /*
1088                  *      Search for duplicates
1089                  */
1090
1091                 if (iter->fib6_metric == rt->fib6_metric) {
1092                         /*
1093                          *      Same priority level
1094                          */
1095                         if (info->nlh &&
1096                             (info->nlh->nlmsg_flags & NLM_F_EXCL))
1097                                 return -EEXIST;
1098
1099                         nlflags &= ~NLM_F_EXCL;
1100                         if (replace) {
1101                                 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1102                                         found++;
1103                                         break;
1104                                 }
1105                                 if (rt_can_ecmp)
1106                                         fallback_ins = fallback_ins ?: ins;
1107                                 goto next_iter;
1108                         }
1109
1110                         if (rt6_duplicate_nexthop(iter, rt)) {
1111                                 if (rt->fib6_nsiblings)
1112                                         rt->fib6_nsiblings = 0;
1113                                 if (!(iter->fib6_flags & RTF_EXPIRES))
1114                                         return -EEXIST;
1115                                 if (!(rt->fib6_flags & RTF_EXPIRES))
1116                                         fib6_clean_expires(iter);
1117                                 else
1118                                         fib6_set_expires(iter, rt->expires);
1119
1120                                 if (rt->fib6_pmtu)
1121                                         fib6_metric_set(iter, RTAX_MTU,
1122                                                         rt->fib6_pmtu);
1123                                 return -EEXIST;
1124                         }
1125                         /* If we have the same destination and the same metric,
1126                          * but not the same gateway, then the route we try to
1127                          * add is sibling to this route, increment our counter
1128                          * of siblings, and later we will add our route to the
1129                          * list.
1130                          * Only static routes (which don't have flag
1131                          * RTF_EXPIRES) are used for ECMPv6.
1132                          *
1133                          * To avoid long list, we only had siblings if the
1134                          * route have a gateway.
1135                          */
1136                         if (rt_can_ecmp &&
1137                             rt6_qualify_for_ecmp(iter))
1138                                 rt->fib6_nsiblings++;
1139                 }
1140
1141                 if (iter->fib6_metric > rt->fib6_metric)
1142                         break;
1143
1144 next_iter:
1145                 ins = &iter->fib6_next;
1146         }
1147
1148         if (fallback_ins && !found) {
1149                 /* No ECMP-able route found, replace first non-ECMP one */
1150                 ins = fallback_ins;
1151                 iter = rcu_dereference_protected(*ins,
1152                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1153                 found++;
1154         }
1155
1156         /* Reset round-robin state, if necessary */
1157         if (ins == &fn->leaf)
1158                 fn->rr_ptr = NULL;
1159
1160         /* Link this route to others same route. */
1161         if (rt->fib6_nsiblings) {
1162                 unsigned int fib6_nsiblings;
1163                 struct fib6_info *sibling, *temp_sibling;
1164
1165                 /* Find the first route that have the same metric */
1166                 sibling = leaf;
1167                 notify_sibling_rt = true;
1168                 while (sibling) {
1169                         if (sibling->fib6_metric == rt->fib6_metric &&
1170                             rt6_qualify_for_ecmp(sibling)) {
1171                                 list_add_tail(&rt->fib6_siblings,
1172                                               &sibling->fib6_siblings);
1173                                 break;
1174                         }
1175                         sibling = rcu_dereference_protected(sibling->fib6_next,
1176                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1177                         notify_sibling_rt = false;
1178                 }
1179                 /* For each sibling in the list, increment the counter of
1180                  * siblings. BUG() if counters does not match, list of siblings
1181                  * is broken!
1182                  */
1183                 fib6_nsiblings = 0;
1184                 list_for_each_entry_safe(sibling, temp_sibling,
1185                                          &rt->fib6_siblings, fib6_siblings) {
1186                         sibling->fib6_nsiblings++;
1187                         BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1188                         fib6_nsiblings++;
1189                 }
1190                 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1191                 rt6_multipath_rebalance(temp_sibling);
1192         }
1193
1194         /*
1195          *      insert node
1196          */
1197         if (!replace) {
1198                 if (!add)
1199                         pr_warn("NLM_F_CREATE should be set when creating new route\n");
1200
1201 add:
1202                 nlflags |= NLM_F_CREATE;
1203
1204                 /* The route should only be notified if it is the first
1205                  * route in the node or if it is added as a sibling
1206                  * route to the first route in the node.
1207                  */
1208                 if (!info->skip_notify_kernel &&
1209                     (notify_sibling_rt || ins == &fn->leaf)) {
1210                         enum fib_event_type fib_event;
1211
1212                         if (notify_sibling_rt)
1213                                 fib_event = FIB_EVENT_ENTRY_APPEND;
1214                         else
1215                                 fib_event = FIB_EVENT_ENTRY_REPLACE;
1216                         err = call_fib6_entry_notifiers(info->nl_net,
1217                                                         fib_event, rt,
1218                                                         extack);
1219                         if (err) {
1220                                 struct fib6_info *sibling, *next_sibling;
1221
1222                                 /* If the route has siblings, then it first
1223                                  * needs to be unlinked from them.
1224                                  */
1225                                 if (!rt->fib6_nsiblings)
1226                                         return err;
1227
1228                                 list_for_each_entry_safe(sibling, next_sibling,
1229                                                          &rt->fib6_siblings,
1230                                                          fib6_siblings)
1231                                         sibling->fib6_nsiblings--;
1232                                 rt->fib6_nsiblings = 0;
1233                                 list_del_init(&rt->fib6_siblings);
1234                                 rt6_multipath_rebalance(next_sibling);
1235                                 return err;
1236                         }
1237                 }
1238
1239                 rcu_assign_pointer(rt->fib6_next, iter);
1240                 fib6_info_hold(rt);
1241                 rcu_assign_pointer(rt->fib6_node, fn);
1242                 rcu_assign_pointer(*ins, rt);
1243                 if (!info->skip_notify)
1244                         inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1245                 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1246
1247                 if (!(fn->fn_flags & RTN_RTINFO)) {
1248                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1249                         fn->fn_flags |= RTN_RTINFO;
1250                 }
1251
1252         } else {
1253                 int nsiblings;
1254
1255                 if (!found) {
1256                         if (add)
1257                                 goto add;
1258                         pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1259                         return -ENOENT;
1260                 }
1261
1262                 if (!info->skip_notify_kernel && ins == &fn->leaf) {
1263                         err = call_fib6_entry_notifiers(info->nl_net,
1264                                                         FIB_EVENT_ENTRY_REPLACE,
1265                                                         rt, extack);
1266                         if (err)
1267                                 return err;
1268                 }
1269
1270                 fib6_info_hold(rt);
1271                 rcu_assign_pointer(rt->fib6_node, fn);
1272                 rt->fib6_next = iter->fib6_next;
1273                 rcu_assign_pointer(*ins, rt);
1274                 if (!info->skip_notify)
1275                         inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1276                 if (!(fn->fn_flags & RTN_RTINFO)) {
1277                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1278                         fn->fn_flags |= RTN_RTINFO;
1279                 }
1280                 nsiblings = iter->fib6_nsiblings;
1281                 iter->fib6_node = NULL;
1282                 fib6_purge_rt(iter, fn, info->nl_net);
1283                 if (rcu_access_pointer(fn->rr_ptr) == iter)
1284                         fn->rr_ptr = NULL;
1285                 fib6_info_release(iter);
1286
1287                 if (nsiblings) {
1288                         /* Replacing an ECMP route, remove all siblings */
1289                         ins = &rt->fib6_next;
1290                         iter = rcu_dereference_protected(*ins,
1291                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1292                         while (iter) {
1293                                 if (iter->fib6_metric > rt->fib6_metric)
1294                                         break;
1295                                 if (rt6_qualify_for_ecmp(iter)) {
1296                                         *ins = iter->fib6_next;
1297                                         iter->fib6_node = NULL;
1298                                         fib6_purge_rt(iter, fn, info->nl_net);
1299                                         if (rcu_access_pointer(fn->rr_ptr) == iter)
1300                                                 fn->rr_ptr = NULL;
1301                                         fib6_info_release(iter);
1302                                         nsiblings--;
1303                                         info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1304                                 } else {
1305                                         ins = &iter->fib6_next;
1306                                 }
1307                                 iter = rcu_dereference_protected(*ins,
1308                                         lockdep_is_held(&rt->fib6_table->tb6_lock));
1309                         }
1310                         WARN_ON(nsiblings != 0);
1311                 }
1312         }
1313
1314         return 0;
1315 }
1316
1317 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1318 {
1319         if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1320             (rt->fib6_flags & RTF_EXPIRES))
1321                 mod_timer(&net->ipv6.ip6_fib_timer,
1322                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1323 }
1324
1325 void fib6_force_start_gc(struct net *net)
1326 {
1327         if (!timer_pending(&net->ipv6.ip6_fib_timer))
1328                 mod_timer(&net->ipv6.ip6_fib_timer,
1329                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1330 }
1331
1332 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1333                                            int sernum)
1334 {
1335         struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1336                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
1337
1338         /* paired with smp_rmb() in rt6_get_cookie_safe() */
1339         smp_wmb();
1340         while (fn) {
1341                 fn->fn_sernum = sernum;
1342                 fn = rcu_dereference_protected(fn->parent,
1343                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
1344         }
1345 }
1346
1347 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1348 {
1349         __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1350 }
1351
1352 /* allow ipv4 to update sernum via ipv6_stub */
1353 void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1354 {
1355         spin_lock_bh(&f6i->fib6_table->tb6_lock);
1356         fib6_update_sernum_upto_root(net, f6i);
1357         spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1358 }
1359
1360 /*
1361  *      Add routing information to the routing tree.
1362  *      <destination addr>/<source addr>
1363  *      with source addr info in sub-trees
1364  *      Need to own table->tb6_lock
1365  */
1366
1367 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1368              struct nl_info *info, struct netlink_ext_ack *extack)
1369 {
1370         struct fib6_table *table = rt->fib6_table;
1371         struct fib6_node *fn, *pn = NULL;
1372         int err = -ENOMEM;
1373         int allow_create = 1;
1374         int replace_required = 0;
1375         int sernum = fib6_new_sernum(info->nl_net);
1376
1377         if (info->nlh) {
1378                 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1379                         allow_create = 0;
1380                 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1381                         replace_required = 1;
1382         }
1383         if (!allow_create && !replace_required)
1384                 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1385
1386         fn = fib6_add_1(info->nl_net, table, root,
1387                         &rt->fib6_dst.addr, rt->fib6_dst.plen,
1388                         offsetof(struct fib6_info, fib6_dst), allow_create,
1389                         replace_required, extack);
1390         if (IS_ERR(fn)) {
1391                 err = PTR_ERR(fn);
1392                 fn = NULL;
1393                 goto out;
1394         }
1395
1396         pn = fn;
1397
1398 #ifdef CONFIG_IPV6_SUBTREES
1399         if (rt->fib6_src.plen) {
1400                 struct fib6_node *sn;
1401
1402                 if (!rcu_access_pointer(fn->subtree)) {
1403                         struct fib6_node *sfn;
1404
1405                         /*
1406                          * Create subtree.
1407                          *
1408                          *              fn[main tree]
1409                          *              |
1410                          *              sfn[subtree root]
1411                          *                 \
1412                          *                  sn[new leaf node]
1413                          */
1414
1415                         /* Create subtree root node */
1416                         sfn = node_alloc(info->nl_net);
1417                         if (!sfn)
1418                                 goto failure;
1419
1420                         fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1421                         rcu_assign_pointer(sfn->leaf,
1422                                            info->nl_net->ipv6.fib6_null_entry);
1423                         sfn->fn_flags = RTN_ROOT;
1424
1425                         /* Now add the first leaf node to new subtree */
1426
1427                         sn = fib6_add_1(info->nl_net, table, sfn,
1428                                         &rt->fib6_src.addr, rt->fib6_src.plen,
1429                                         offsetof(struct fib6_info, fib6_src),
1430                                         allow_create, replace_required, extack);
1431
1432                         if (IS_ERR(sn)) {
1433                                 /* If it is failed, discard just allocated
1434                                    root, and then (in failure) stale node
1435                                    in main tree.
1436                                  */
1437                                 node_free_immediate(info->nl_net, sfn);
1438                                 err = PTR_ERR(sn);
1439                                 goto failure;
1440                         }
1441
1442                         /* Now link new subtree to main tree */
1443                         rcu_assign_pointer(sfn->parent, fn);
1444                         rcu_assign_pointer(fn->subtree, sfn);
1445                 } else {
1446                         sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1447                                         &rt->fib6_src.addr, rt->fib6_src.plen,
1448                                         offsetof(struct fib6_info, fib6_src),
1449                                         allow_create, replace_required, extack);
1450
1451                         if (IS_ERR(sn)) {
1452                                 err = PTR_ERR(sn);
1453                                 goto failure;
1454                         }
1455                 }
1456
1457                 if (!rcu_access_pointer(fn->leaf)) {
1458                         if (fn->fn_flags & RTN_TL_ROOT) {
1459                                 /* put back null_entry for root node */
1460                                 rcu_assign_pointer(fn->leaf,
1461                                             info->nl_net->ipv6.fib6_null_entry);
1462                         } else {
1463                                 fib6_info_hold(rt);
1464                                 rcu_assign_pointer(fn->leaf, rt);
1465                         }
1466                 }
1467                 fn = sn;
1468         }
1469 #endif
1470
1471         err = fib6_add_rt2node(fn, rt, info, extack);
1472         if (!err) {
1473                 if (rt->nh)
1474                         list_add(&rt->nh_list, &rt->nh->f6i_list);
1475                 __fib6_update_sernum_upto_root(rt, sernum);
1476                 fib6_start_gc(info->nl_net, rt);
1477         }
1478
1479 out:
1480         if (err) {
1481 #ifdef CONFIG_IPV6_SUBTREES
1482                 /*
1483                  * If fib6_add_1 has cleared the old leaf pointer in the
1484                  * super-tree leaf node we have to find a new one for it.
1485                  */
1486                 if (pn != fn) {
1487                         struct fib6_info *pn_leaf =
1488                                 rcu_dereference_protected(pn->leaf,
1489                                     lockdep_is_held(&table->tb6_lock));
1490                         if (pn_leaf == rt) {
1491                                 pn_leaf = NULL;
1492                                 RCU_INIT_POINTER(pn->leaf, NULL);
1493                                 fib6_info_release(rt);
1494                         }
1495                         if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1496                                 pn_leaf = fib6_find_prefix(info->nl_net, table,
1497                                                            pn);
1498 #if RT6_DEBUG >= 2
1499                                 if (!pn_leaf) {
1500                                         WARN_ON(!pn_leaf);
1501                                         pn_leaf =
1502                                             info->nl_net->ipv6.fib6_null_entry;
1503                                 }
1504 #endif
1505                                 fib6_info_hold(pn_leaf);
1506                                 rcu_assign_pointer(pn->leaf, pn_leaf);
1507                         }
1508                 }
1509 #endif
1510                 goto failure;
1511         } else if (fib6_requires_src(rt)) {
1512                 fib6_routes_require_src_inc(info->nl_net);
1513         }
1514         return err;
1515
1516 failure:
1517         /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1518          * 1. fn is an intermediate node and we failed to add the new
1519          * route to it in both subtree creation failure and fib6_add_rt2node()
1520          * failure case.
1521          * 2. fn is the root node in the table and we fail to add the first
1522          * default route to it.
1523          */
1524         if (fn &&
1525             (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1526              (fn->fn_flags & RTN_TL_ROOT &&
1527               !rcu_access_pointer(fn->leaf))))
1528                 fib6_repair_tree(info->nl_net, table, fn);
1529         return err;
1530 }
1531
1532 /*
1533  *      Routing tree lookup
1534  *
1535  */
1536
1537 struct lookup_args {
1538         int                     offset;         /* key offset on fib6_info */
1539         const struct in6_addr   *addr;          /* search key                   */
1540 };
1541
1542 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1543                                             struct lookup_args *args)
1544 {
1545         struct fib6_node *fn;
1546         __be32 dir;
1547
1548         if (unlikely(args->offset == 0))
1549                 return NULL;
1550
1551         /*
1552          *      Descend on a tree
1553          */
1554
1555         fn = root;
1556
1557         for (;;) {
1558                 struct fib6_node *next;
1559
1560                 dir = addr_bit_set(args->addr, fn->fn_bit);
1561
1562                 next = dir ? rcu_dereference(fn->right) :
1563                              rcu_dereference(fn->left);
1564
1565                 if (next) {
1566                         fn = next;
1567                         continue;
1568                 }
1569                 break;
1570         }
1571
1572         while (fn) {
1573                 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1574
1575                 if (subtree || fn->fn_flags & RTN_RTINFO) {
1576                         struct fib6_info *leaf = rcu_dereference(fn->leaf);
1577                         struct rt6key *key;
1578
1579                         if (!leaf)
1580                                 goto backtrack;
1581
1582                         key = (struct rt6key *) ((u8 *)leaf + args->offset);
1583
1584                         if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1585 #ifdef CONFIG_IPV6_SUBTREES
1586                                 if (subtree) {
1587                                         struct fib6_node *sfn;
1588                                         sfn = fib6_node_lookup_1(subtree,
1589                                                                  args + 1);
1590                                         if (!sfn)
1591                                                 goto backtrack;
1592                                         fn = sfn;
1593                                 }
1594 #endif
1595                                 if (fn->fn_flags & RTN_RTINFO)
1596                                         return fn;
1597                         }
1598                 }
1599 backtrack:
1600                 if (fn->fn_flags & RTN_ROOT)
1601                         break;
1602
1603                 fn = rcu_dereference(fn->parent);
1604         }
1605
1606         return NULL;
1607 }
1608
1609 /* called with rcu_read_lock() held
1610  */
1611 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1612                                    const struct in6_addr *daddr,
1613                                    const struct in6_addr *saddr)
1614 {
1615         struct fib6_node *fn;
1616         struct lookup_args args[] = {
1617                 {
1618                         .offset = offsetof(struct fib6_info, fib6_dst),
1619                         .addr = daddr,
1620                 },
1621 #ifdef CONFIG_IPV6_SUBTREES
1622                 {
1623                         .offset = offsetof(struct fib6_info, fib6_src),
1624                         .addr = saddr,
1625                 },
1626 #endif
1627                 {
1628                         .offset = 0,    /* sentinel */
1629                 }
1630         };
1631
1632         fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1633         if (!fn || fn->fn_flags & RTN_TL_ROOT)
1634                 fn = root;
1635
1636         return fn;
1637 }
1638
1639 /*
1640  *      Get node with specified destination prefix (and source prefix,
1641  *      if subtrees are used)
1642  *      exact_match == true means we try to find fn with exact match of
1643  *      the passed in prefix addr
1644  *      exact_match == false means we try to find fn with longest prefix
1645  *      match of the passed in prefix addr. This is useful for finding fn
1646  *      for cached route as it will be stored in the exception table under
1647  *      the node with longest prefix length.
1648  */
1649
1650
1651 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1652                                        const struct in6_addr *addr,
1653                                        int plen, int offset,
1654                                        bool exact_match)
1655 {
1656         struct fib6_node *fn, *prev = NULL;
1657
1658         for (fn = root; fn ; ) {
1659                 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1660                 struct rt6key *key;
1661
1662                 /* This node is being deleted */
1663                 if (!leaf) {
1664                         if (plen <= fn->fn_bit)
1665                                 goto out;
1666                         else
1667                                 goto next;
1668                 }
1669
1670                 key = (struct rt6key *)((u8 *)leaf + offset);
1671
1672                 /*
1673                  *      Prefix match
1674                  */
1675                 if (plen < fn->fn_bit ||
1676                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1677                         goto out;
1678
1679                 if (plen == fn->fn_bit)
1680                         return fn;
1681
1682                 if (fn->fn_flags & RTN_RTINFO)
1683                         prev = fn;
1684
1685 next:
1686                 /*
1687                  *      We have more bits to go
1688                  */
1689                 if (addr_bit_set(addr, fn->fn_bit))
1690                         fn = rcu_dereference(fn->right);
1691                 else
1692                         fn = rcu_dereference(fn->left);
1693         }
1694 out:
1695         if (exact_match)
1696                 return NULL;
1697         else
1698                 return prev;
1699 }
1700
1701 struct fib6_node *fib6_locate(struct fib6_node *root,
1702                               const struct in6_addr *daddr, int dst_len,
1703                               const struct in6_addr *saddr, int src_len,
1704                               bool exact_match)
1705 {
1706         struct fib6_node *fn;
1707
1708         fn = fib6_locate_1(root, daddr, dst_len,
1709                            offsetof(struct fib6_info, fib6_dst),
1710                            exact_match);
1711
1712 #ifdef CONFIG_IPV6_SUBTREES
1713         if (src_len) {
1714                 WARN_ON(saddr == NULL);
1715                 if (fn) {
1716                         struct fib6_node *subtree = FIB6_SUBTREE(fn);
1717
1718                         if (subtree) {
1719                                 fn = fib6_locate_1(subtree, saddr, src_len,
1720                                            offsetof(struct fib6_info, fib6_src),
1721                                            exact_match);
1722                         }
1723                 }
1724         }
1725 #endif
1726
1727         if (fn && fn->fn_flags & RTN_RTINFO)
1728                 return fn;
1729
1730         return NULL;
1731 }
1732
1733
1734 /*
1735  *      Deletion
1736  *
1737  */
1738
1739 static struct fib6_info *fib6_find_prefix(struct net *net,
1740                                          struct fib6_table *table,
1741                                          struct fib6_node *fn)
1742 {
1743         struct fib6_node *child_left, *child_right;
1744
1745         if (fn->fn_flags & RTN_ROOT)
1746                 return net->ipv6.fib6_null_entry;
1747
1748         while (fn) {
1749                 child_left = rcu_dereference_protected(fn->left,
1750                                     lockdep_is_held(&table->tb6_lock));
1751                 child_right = rcu_dereference_protected(fn->right,
1752                                     lockdep_is_held(&table->tb6_lock));
1753                 if (child_left)
1754                         return rcu_dereference_protected(child_left->leaf,
1755                                         lockdep_is_held(&table->tb6_lock));
1756                 if (child_right)
1757                         return rcu_dereference_protected(child_right->leaf,
1758                                         lockdep_is_held(&table->tb6_lock));
1759
1760                 fn = FIB6_SUBTREE(fn);
1761         }
1762         return NULL;
1763 }
1764
1765 /*
1766  *      Called to trim the tree of intermediate nodes when possible. "fn"
1767  *      is the node we want to try and remove.
1768  *      Need to own table->tb6_lock
1769  */
1770
1771 static struct fib6_node *fib6_repair_tree(struct net *net,
1772                                           struct fib6_table *table,
1773                                           struct fib6_node *fn)
1774 {
1775         int children;
1776         int nstate;
1777         struct fib6_node *child;
1778         struct fib6_walker *w;
1779         int iter = 0;
1780
1781         /* Set fn->leaf to null_entry for root node. */
1782         if (fn->fn_flags & RTN_TL_ROOT) {
1783                 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1784                 return fn;
1785         }
1786
1787         for (;;) {
1788                 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1789                                             lockdep_is_held(&table->tb6_lock));
1790                 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1791                                             lockdep_is_held(&table->tb6_lock));
1792                 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1793                                             lockdep_is_held(&table->tb6_lock));
1794                 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1795                                             lockdep_is_held(&table->tb6_lock));
1796                 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1797                                             lockdep_is_held(&table->tb6_lock));
1798                 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1799                                             lockdep_is_held(&table->tb6_lock));
1800                 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1801                                             lockdep_is_held(&table->tb6_lock));
1802                 struct fib6_info *new_fn_leaf;
1803
1804                 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1805                 iter++;
1806
1807                 WARN_ON(fn->fn_flags & RTN_RTINFO);
1808                 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1809                 WARN_ON(fn_leaf);
1810
1811                 children = 0;
1812                 child = NULL;
1813                 if (fn_r)
1814                         child = fn_r, children |= 1;
1815                 if (fn_l)
1816                         child = fn_l, children |= 2;
1817
1818                 if (children == 3 || FIB6_SUBTREE(fn)
1819 #ifdef CONFIG_IPV6_SUBTREES
1820                     /* Subtree root (i.e. fn) may have one child */
1821                     || (children && fn->fn_flags & RTN_ROOT)
1822 #endif
1823                     ) {
1824                         new_fn_leaf = fib6_find_prefix(net, table, fn);
1825 #if RT6_DEBUG >= 2
1826                         if (!new_fn_leaf) {
1827                                 WARN_ON(!new_fn_leaf);
1828                                 new_fn_leaf = net->ipv6.fib6_null_entry;
1829                         }
1830 #endif
1831                         fib6_info_hold(new_fn_leaf);
1832                         rcu_assign_pointer(fn->leaf, new_fn_leaf);
1833                         return pn;
1834                 }
1835
1836 #ifdef CONFIG_IPV6_SUBTREES
1837                 if (FIB6_SUBTREE(pn) == fn) {
1838                         WARN_ON(!(fn->fn_flags & RTN_ROOT));
1839                         RCU_INIT_POINTER(pn->subtree, NULL);
1840                         nstate = FWS_L;
1841                 } else {
1842                         WARN_ON(fn->fn_flags & RTN_ROOT);
1843 #endif
1844                         if (pn_r == fn)
1845                                 rcu_assign_pointer(pn->right, child);
1846                         else if (pn_l == fn)
1847                                 rcu_assign_pointer(pn->left, child);
1848 #if RT6_DEBUG >= 2
1849                         else
1850                                 WARN_ON(1);
1851 #endif
1852                         if (child)
1853                                 rcu_assign_pointer(child->parent, pn);
1854                         nstate = FWS_R;
1855 #ifdef CONFIG_IPV6_SUBTREES
1856                 }
1857 #endif
1858
1859                 read_lock(&net->ipv6.fib6_walker_lock);
1860                 FOR_WALKERS(net, w) {
1861                         if (!child) {
1862                                 if (w->node == fn) {
1863                                         RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1864                                         w->node = pn;
1865                                         w->state = nstate;
1866                                 }
1867                         } else {
1868                                 if (w->node == fn) {
1869                                         w->node = child;
1870                                         if (children&2) {
1871                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1872                                                 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1873                                         } else {
1874                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1875                                                 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1876                                         }
1877                                 }
1878                         }
1879                 }
1880                 read_unlock(&net->ipv6.fib6_walker_lock);
1881
1882                 node_free(net, fn);
1883                 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1884                         return pn;
1885
1886                 RCU_INIT_POINTER(pn->leaf, NULL);
1887                 fib6_info_release(pn_leaf);
1888                 fn = pn;
1889         }
1890 }
1891
1892 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1893                            struct fib6_info __rcu **rtp, struct nl_info *info)
1894 {
1895         struct fib6_info *leaf, *replace_rt = NULL;
1896         struct fib6_walker *w;
1897         struct fib6_info *rt = rcu_dereference_protected(*rtp,
1898                                     lockdep_is_held(&table->tb6_lock));
1899         struct net *net = info->nl_net;
1900         bool notify_del = false;
1901
1902         RT6_TRACE("fib6_del_route\n");
1903
1904         /* If the deleted route is the first in the node and it is not part of
1905          * a multipath route, then we need to replace it with the next route
1906          * in the node, if exists.
1907          */
1908         leaf = rcu_dereference_protected(fn->leaf,
1909                                          lockdep_is_held(&table->tb6_lock));
1910         if (leaf == rt && !rt->fib6_nsiblings) {
1911                 if (rcu_access_pointer(rt->fib6_next))
1912                         replace_rt = rcu_dereference_protected(rt->fib6_next,
1913                                             lockdep_is_held(&table->tb6_lock));
1914                 else
1915                         notify_del = true;
1916         }
1917
1918         /* Unlink it */
1919         *rtp = rt->fib6_next;
1920         rt->fib6_node = NULL;
1921         net->ipv6.rt6_stats->fib_rt_entries--;
1922         net->ipv6.rt6_stats->fib_discarded_routes++;
1923
1924         /* Flush all cached dst in exception table */
1925         rt6_flush_exceptions(rt);
1926
1927         /* Reset round-robin state, if necessary */
1928         if (rcu_access_pointer(fn->rr_ptr) == rt)
1929                 fn->rr_ptr = NULL;
1930
1931         /* Remove this entry from other siblings */
1932         if (rt->fib6_nsiblings) {
1933                 struct fib6_info *sibling, *next_sibling;
1934
1935                 /* The route is deleted from a multipath route. If this
1936                  * multipath route is the first route in the node, then we need
1937                  * to emit a delete notification. Otherwise, we need to skip
1938                  * the notification.
1939                  */
1940                 if (rt->fib6_metric == leaf->fib6_metric &&
1941                     rt6_qualify_for_ecmp(leaf))
1942                         notify_del = true;
1943                 list_for_each_entry_safe(sibling, next_sibling,
1944                                          &rt->fib6_siblings, fib6_siblings)
1945                         sibling->fib6_nsiblings--;
1946                 rt->fib6_nsiblings = 0;
1947                 list_del_init(&rt->fib6_siblings);
1948                 rt6_multipath_rebalance(next_sibling);
1949         }
1950
1951         /* Adjust walkers */
1952         read_lock(&net->ipv6.fib6_walker_lock);
1953         FOR_WALKERS(net, w) {
1954                 if (w->state == FWS_C && w->leaf == rt) {
1955                         RT6_TRACE("walker %p adjusted by delroute\n", w);
1956                         w->leaf = rcu_dereference_protected(rt->fib6_next,
1957                                             lockdep_is_held(&table->tb6_lock));
1958                         if (!w->leaf)
1959                                 w->state = FWS_U;
1960                 }
1961         }
1962         read_unlock(&net->ipv6.fib6_walker_lock);
1963
1964         /* If it was last route, call fib6_repair_tree() to:
1965          * 1. For root node, put back null_entry as how the table was created.
1966          * 2. For other nodes, expunge its radix tree node.
1967          */
1968         if (!rcu_access_pointer(fn->leaf)) {
1969                 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1970                         fn->fn_flags &= ~RTN_RTINFO;
1971                         net->ipv6.rt6_stats->fib_route_nodes--;
1972                 }
1973                 fn = fib6_repair_tree(net, table, fn);
1974         }
1975
1976         fib6_purge_rt(rt, fn, net);
1977
1978         if (!info->skip_notify_kernel) {
1979                 if (notify_del)
1980                         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1981                                                   rt, NULL);
1982                 else if (replace_rt)
1983                         call_fib6_entry_notifiers_replace(net, replace_rt);
1984         }
1985         if (!info->skip_notify)
1986                 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1987
1988         fib6_info_release(rt);
1989 }
1990
1991 /* Need to own table->tb6_lock */
1992 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1993 {
1994         struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1995                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1996         struct fib6_table *table = rt->fib6_table;
1997         struct net *net = info->nl_net;
1998         struct fib6_info __rcu **rtp;
1999         struct fib6_info __rcu **rtp_next;
2000
2001         if (!fn || rt == net->ipv6.fib6_null_entry)
2002                 return -ENOENT;
2003
2004         WARN_ON(!(fn->fn_flags & RTN_RTINFO));
2005
2006         /*
2007          *      Walk the leaf entries looking for ourself
2008          */
2009
2010         for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
2011                 struct fib6_info *cur = rcu_dereference_protected(*rtp,
2012                                         lockdep_is_held(&table->tb6_lock));
2013                 if (rt == cur) {
2014                         if (fib6_requires_src(cur))
2015                                 fib6_routes_require_src_dec(info->nl_net);
2016                         fib6_del_route(table, fn, rtp, info);
2017                         return 0;
2018                 }
2019                 rtp_next = &cur->fib6_next;
2020         }
2021         return -ENOENT;
2022 }
2023
2024 /*
2025  *      Tree traversal function.
2026  *
2027  *      Certainly, it is not interrupt safe.
2028  *      However, it is internally reenterable wrt itself and fib6_add/fib6_del.
2029  *      It means, that we can modify tree during walking
2030  *      and use this function for garbage collection, clone pruning,
2031  *      cleaning tree when a device goes down etc. etc.
2032  *
2033  *      It guarantees that every node will be traversed,
2034  *      and that it will be traversed only once.
2035  *
2036  *      Callback function w->func may return:
2037  *      0 -> continue walking.
2038  *      positive value -> walking is suspended (used by tree dumps,
2039  *      and probably by gc, if it will be split to several slices)
2040  *      negative value -> terminate walking.
2041  *
2042  *      The function itself returns:
2043  *      0   -> walk is complete.
2044  *      >0  -> walk is incomplete (i.e. suspended)
2045  *      <0  -> walk is terminated by an error.
2046  *
2047  *      This function is called with tb6_lock held.
2048  */
2049
2050 static int fib6_walk_continue(struct fib6_walker *w)
2051 {
2052         struct fib6_node *fn, *pn, *left, *right;
2053
2054         /* w->root should always be table->tb6_root */
2055         WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
2056
2057         for (;;) {
2058                 fn = w->node;
2059                 if (!fn)
2060                         return 0;
2061
2062                 switch (w->state) {
2063 #ifdef CONFIG_IPV6_SUBTREES
2064                 case FWS_S:
2065                         if (FIB6_SUBTREE(fn)) {
2066                                 w->node = FIB6_SUBTREE(fn);
2067                                 continue;
2068                         }
2069                         w->state = FWS_L;
2070 #endif
2071                         /* fall through */
2072                 case FWS_L:
2073                         left = rcu_dereference_protected(fn->left, 1);
2074                         if (left) {
2075                                 w->node = left;
2076                                 w->state = FWS_INIT;
2077                                 continue;
2078                         }
2079                         w->state = FWS_R;
2080                         /* fall through */
2081                 case FWS_R:
2082                         right = rcu_dereference_protected(fn->right, 1);
2083                         if (right) {
2084                                 w->node = right;
2085                                 w->state = FWS_INIT;
2086                                 continue;
2087                         }
2088                         w->state = FWS_C;
2089                         w->leaf = rcu_dereference_protected(fn->leaf, 1);
2090                         /* fall through */
2091                 case FWS_C:
2092                         if (w->leaf && fn->fn_flags & RTN_RTINFO) {
2093                                 int err;
2094
2095                                 if (w->skip) {
2096                                         w->skip--;
2097                                         goto skip;
2098                                 }
2099
2100                                 err = w->func(w);
2101                                 if (err)
2102                                         return err;
2103
2104                                 w->count++;
2105                                 continue;
2106                         }
2107 skip:
2108                         w->state = FWS_U;
2109                         /* fall through */
2110                 case FWS_U:
2111                         if (fn == w->root)
2112                                 return 0;
2113                         pn = rcu_dereference_protected(fn->parent, 1);
2114                         left = rcu_dereference_protected(pn->left, 1);
2115                         right = rcu_dereference_protected(pn->right, 1);
2116                         w->node = pn;
2117 #ifdef CONFIG_IPV6_SUBTREES
2118                         if (FIB6_SUBTREE(pn) == fn) {
2119                                 WARN_ON(!(fn->fn_flags & RTN_ROOT));
2120                                 w->state = FWS_L;
2121                                 continue;
2122                         }
2123 #endif
2124                         if (left == fn) {
2125                                 w->state = FWS_R;
2126                                 continue;
2127                         }
2128                         if (right == fn) {
2129                                 w->state = FWS_C;
2130                                 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
2131                                 continue;
2132                         }
2133 #if RT6_DEBUG >= 2
2134                         WARN_ON(1);
2135 #endif
2136                 }
2137         }
2138 }
2139
2140 static int fib6_walk(struct net *net, struct fib6_walker *w)
2141 {
2142         int res;
2143
2144         w->state = FWS_INIT;
2145         w->node = w->root;
2146
2147         fib6_walker_link(net, w);
2148         res = fib6_walk_continue(w);
2149         if (res <= 0)
2150                 fib6_walker_unlink(net, w);
2151         return res;
2152 }
2153
2154 static int fib6_clean_node(struct fib6_walker *w)
2155 {
2156         int res;
2157         struct fib6_info *rt;
2158         struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
2159         struct nl_info info = {
2160                 .nl_net = c->net,
2161                 .skip_notify = c->skip_notify,
2162         };
2163
2164         if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
2165             w->node->fn_sernum != c->sernum)
2166                 w->node->fn_sernum = c->sernum;
2167
2168         if (!c->func) {
2169                 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
2170                 w->leaf = NULL;
2171                 return 0;
2172         }
2173
2174         for_each_fib6_walker_rt(w) {
2175                 res = c->func(rt, c->arg);
2176                 if (res == -1) {
2177                         w->leaf = rt;
2178                         res = fib6_del(rt, &info);
2179                         if (res) {
2180 #if RT6_DEBUG >= 2
2181                                 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2182                                          __func__, rt,
2183                                          rcu_access_pointer(rt->fib6_node),
2184                                          res);
2185 #endif
2186                                 continue;
2187                         }
2188                         return 0;
2189                 } else if (res == -2) {
2190                         if (WARN_ON(!rt->fib6_nsiblings))
2191                                 continue;
2192                         rt = list_last_entry(&rt->fib6_siblings,
2193                                              struct fib6_info, fib6_siblings);
2194                         continue;
2195                 }
2196                 WARN_ON(res != 0);
2197         }
2198         w->leaf = rt;
2199         return 0;
2200 }
2201
2202 /*
2203  *      Convenient frontend to tree walker.
2204  *
2205  *      func is called on each route.
2206  *              It may return -2 -> skip multipath route.
2207  *                            -1 -> delete this route.
2208  *                            0  -> continue walking
2209  */
2210
2211 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2212                             int (*func)(struct fib6_info *, void *arg),
2213                             int sernum, void *arg, bool skip_notify)
2214 {
2215         struct fib6_cleaner c;
2216
2217         c.w.root = root;
2218         c.w.func = fib6_clean_node;
2219         c.w.count = 0;
2220         c.w.skip = 0;
2221         c.w.skip_in_node = 0;
2222         c.func = func;
2223         c.sernum = sernum;
2224         c.arg = arg;
2225         c.net = net;
2226         c.skip_notify = skip_notify;
2227
2228         fib6_walk(net, &c.w);
2229 }
2230
2231 static void __fib6_clean_all(struct net *net,
2232                              int (*func)(struct fib6_info *, void *),
2233                              int sernum, void *arg, bool skip_notify)
2234 {
2235         struct fib6_table *table;
2236         struct hlist_head *head;
2237         unsigned int h;
2238
2239         rcu_read_lock();
2240         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2241                 head = &net->ipv6.fib_table_hash[h];
2242                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2243                         spin_lock_bh(&table->tb6_lock);
2244                         fib6_clean_tree(net, &table->tb6_root,
2245                                         func, sernum, arg, skip_notify);
2246                         spin_unlock_bh(&table->tb6_lock);
2247                 }
2248         }
2249         rcu_read_unlock();
2250 }
2251
2252 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2253                     void *arg)
2254 {
2255         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2256 }
2257
2258 void fib6_clean_all_skip_notify(struct net *net,
2259                                 int (*func)(struct fib6_info *, void *),
2260                                 void *arg)
2261 {
2262         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2263 }
2264
2265 static void fib6_flush_trees(struct net *net)
2266 {
2267         int new_sernum = fib6_new_sernum(net);
2268
2269         __fib6_clean_all(net, NULL, new_sernum, NULL, false);
2270 }
2271
2272 /*
2273  *      Garbage collection
2274  */
2275
2276 static int fib6_age(struct fib6_info *rt, void *arg)
2277 {
2278         struct fib6_gc_args *gc_args = arg;
2279         unsigned long now = jiffies;
2280
2281         /*
2282          *      check addrconf expiration here.
2283          *      Routes are expired even if they are in use.
2284          */
2285
2286         if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2287                 if (time_after(now, rt->expires)) {
2288                         RT6_TRACE("expiring %p\n", rt);
2289                         return -1;
2290                 }
2291                 gc_args->more++;
2292         }
2293
2294         /*      Also age clones in the exception table.
2295          *      Note, that clones are aged out
2296          *      only if they are not in use now.
2297          */
2298         rt6_age_exceptions(rt, gc_args, now);
2299
2300         return 0;
2301 }
2302
2303 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2304 {
2305         struct fib6_gc_args gc_args;
2306         unsigned long now;
2307
2308         if (force) {
2309                 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2310         } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2311                 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2312                 return;
2313         }
2314         gc_args.timeout = expires ? (int)expires :
2315                           net->ipv6.sysctl.ip6_rt_gc_interval;
2316         gc_args.more = 0;
2317
2318         fib6_clean_all(net, fib6_age, &gc_args);
2319         now = jiffies;
2320         net->ipv6.ip6_rt_last_gc = now;
2321
2322         if (gc_args.more)
2323                 mod_timer(&net->ipv6.ip6_fib_timer,
2324                           round_jiffies(now
2325                                         + net->ipv6.sysctl.ip6_rt_gc_interval));
2326         else
2327                 del_timer(&net->ipv6.ip6_fib_timer);
2328         spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2329 }
2330
2331 static void fib6_gc_timer_cb(struct timer_list *t)
2332 {
2333         struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2334
2335         fib6_run_gc(0, arg, true);
2336 }
2337
2338 static int __net_init fib6_net_init(struct net *net)
2339 {
2340         size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2341         int err;
2342
2343         err = fib6_notifier_init(net);
2344         if (err)
2345                 return err;
2346
2347         spin_lock_init(&net->ipv6.fib6_gc_lock);
2348         rwlock_init(&net->ipv6.fib6_walker_lock);
2349         INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2350         timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2351
2352         net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2353         if (!net->ipv6.rt6_stats)
2354                 goto out_timer;
2355
2356         /* Avoid false sharing : Use at least a full cache line */
2357         size = max_t(size_t, size, L1_CACHE_BYTES);
2358
2359         net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2360         if (!net->ipv6.fib_table_hash)
2361                 goto out_rt6_stats;
2362
2363         net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2364                                           GFP_KERNEL);
2365         if (!net->ipv6.fib6_main_tbl)
2366                 goto out_fib_table_hash;
2367
2368         net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2369         rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2370                            net->ipv6.fib6_null_entry);
2371         net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2372                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2373         inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2374
2375 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2376         net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2377                                            GFP_KERNEL);
2378         if (!net->ipv6.fib6_local_tbl)
2379                 goto out_fib6_main_tbl;
2380         net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2381         rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2382                            net->ipv6.fib6_null_entry);
2383         net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2384                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2385         inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2386 #endif
2387         fib6_tables_init(net);
2388
2389         return 0;
2390
2391 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2392 out_fib6_main_tbl:
2393         kfree(net->ipv6.fib6_main_tbl);
2394 #endif
2395 out_fib_table_hash:
2396         kfree(net->ipv6.fib_table_hash);
2397 out_rt6_stats:
2398         kfree(net->ipv6.rt6_stats);
2399 out_timer:
2400         fib6_notifier_exit(net);
2401         return -ENOMEM;
2402 }
2403
2404 static void fib6_net_exit(struct net *net)
2405 {
2406         unsigned int i;
2407
2408         del_timer_sync(&net->ipv6.ip6_fib_timer);
2409
2410         for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2411                 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2412                 struct hlist_node *tmp;
2413                 struct fib6_table *tb;
2414
2415                 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2416                         hlist_del(&tb->tb6_hlist);
2417                         fib6_free_table(tb);
2418                 }
2419         }
2420
2421         kfree(net->ipv6.fib_table_hash);
2422         kfree(net->ipv6.rt6_stats);
2423         fib6_notifier_exit(net);
2424 }
2425
2426 static struct pernet_operations fib6_net_ops = {
2427         .init = fib6_net_init,
2428         .exit = fib6_net_exit,
2429 };
2430
2431 int __init fib6_init(void)
2432 {
2433         int ret = -ENOMEM;
2434
2435         fib6_node_kmem = kmem_cache_create("fib6_nodes",
2436                                            sizeof(struct fib6_node),
2437                                            0, SLAB_HWCACHE_ALIGN,
2438                                            NULL);
2439         if (!fib6_node_kmem)
2440                 goto out;
2441
2442         ret = register_pernet_subsys(&fib6_net_ops);
2443         if (ret)
2444                 goto out_kmem_cache_create;
2445
2446         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2447                                    inet6_dump_fib, 0);
2448         if (ret)
2449                 goto out_unregister_subsys;
2450
2451         __fib6_flush_trees = fib6_flush_trees;
2452 out:
2453         return ret;
2454
2455 out_unregister_subsys:
2456         unregister_pernet_subsys(&fib6_net_ops);
2457 out_kmem_cache_create:
2458         kmem_cache_destroy(fib6_node_kmem);
2459         goto out;
2460 }
2461
2462 void fib6_gc_cleanup(void)
2463 {
2464         unregister_pernet_subsys(&fib6_net_ops);
2465         kmem_cache_destroy(fib6_node_kmem);
2466 }
2467
2468 #ifdef CONFIG_PROC_FS
2469 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2470 {
2471         struct fib6_info *rt = v;
2472         struct ipv6_route_iter *iter = seq->private;
2473         struct fib6_nh *fib6_nh = rt->fib6_nh;
2474         unsigned int flags = rt->fib6_flags;
2475         const struct net_device *dev;
2476
2477         if (rt->nh)
2478                 fib6_nh = nexthop_fib6_nh(rt->nh);
2479
2480         seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2481
2482 #ifdef CONFIG_IPV6_SUBTREES
2483         seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2484 #else
2485         seq_puts(seq, "00000000000000000000000000000000 00 ");
2486 #endif
2487         if (fib6_nh->fib_nh_gw_family) {
2488                 flags |= RTF_GATEWAY;
2489                 seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6);
2490         } else {
2491                 seq_puts(seq, "00000000000000000000000000000000");
2492         }
2493
2494         dev = fib6_nh->fib_nh_dev;
2495         seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2496                    rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2497                    flags, dev ? dev->name : "");
2498         iter->w.leaf = NULL;
2499         return 0;
2500 }
2501
2502 static int ipv6_route_yield(struct fib6_walker *w)
2503 {
2504         struct ipv6_route_iter *iter = w->args;
2505
2506         if (!iter->skip)
2507                 return 1;
2508
2509         do {
2510                 iter->w.leaf = rcu_dereference_protected(
2511                                 iter->w.leaf->fib6_next,
2512                                 lockdep_is_held(&iter->tbl->tb6_lock));
2513                 iter->skip--;
2514                 if (!iter->skip && iter->w.leaf)
2515                         return 1;
2516         } while (iter->w.leaf);
2517
2518         return 0;
2519 }
2520
2521 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2522                                       struct net *net)
2523 {
2524         memset(&iter->w, 0, sizeof(iter->w));
2525         iter->w.func = ipv6_route_yield;
2526         iter->w.root = &iter->tbl->tb6_root;
2527         iter->w.state = FWS_INIT;
2528         iter->w.node = iter->w.root;
2529         iter->w.args = iter;
2530         iter->sernum = iter->w.root->fn_sernum;
2531         INIT_LIST_HEAD(&iter->w.lh);
2532         fib6_walker_link(net, &iter->w);
2533 }
2534
2535 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2536                                                     struct net *net)
2537 {
2538         unsigned int h;
2539         struct hlist_node *node;
2540
2541         if (tbl) {
2542                 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2543                 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2544         } else {
2545                 h = 0;
2546                 node = NULL;
2547         }
2548
2549         while (!node && h < FIB6_TABLE_HASHSZ) {
2550                 node = rcu_dereference_bh(
2551                         hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2552         }
2553         return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2554 }
2555
2556 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2557 {
2558         if (iter->sernum != iter->w.root->fn_sernum) {
2559                 iter->sernum = iter->w.root->fn_sernum;
2560                 iter->w.state = FWS_INIT;
2561                 iter->w.node = iter->w.root;
2562                 WARN_ON(iter->w.skip);
2563                 iter->w.skip = iter->w.count;
2564         }
2565 }
2566
2567 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2568 {
2569         int r;
2570         struct fib6_info *n;
2571         struct net *net = seq_file_net(seq);
2572         struct ipv6_route_iter *iter = seq->private;
2573
2574         ++(*pos);
2575         if (!v)
2576                 goto iter_table;
2577
2578         n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2579         if (n)
2580                 return n;
2581
2582 iter_table:
2583         ipv6_route_check_sernum(iter);
2584         spin_lock_bh(&iter->tbl->tb6_lock);
2585         r = fib6_walk_continue(&iter->w);
2586         spin_unlock_bh(&iter->tbl->tb6_lock);
2587         if (r > 0) {
2588                 return iter->w.leaf;
2589         } else if (r < 0) {
2590                 fib6_walker_unlink(net, &iter->w);
2591                 return NULL;
2592         }
2593         fib6_walker_unlink(net, &iter->w);
2594
2595         iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2596         if (!iter->tbl)
2597                 return NULL;
2598
2599         ipv6_route_seq_setup_walk(iter, net);
2600         goto iter_table;
2601 }
2602
2603 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2604         __acquires(RCU_BH)
2605 {
2606         struct net *net = seq_file_net(seq);
2607         struct ipv6_route_iter *iter = seq->private;
2608
2609         rcu_read_lock_bh();
2610         iter->tbl = ipv6_route_seq_next_table(NULL, net);
2611         iter->skip = *pos;
2612
2613         if (iter->tbl) {
2614                 ipv6_route_seq_setup_walk(iter, net);
2615                 return ipv6_route_seq_next(seq, NULL, pos);
2616         } else {
2617                 return NULL;
2618         }
2619 }
2620
2621 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2622 {
2623         struct fib6_walker *w = &iter->w;
2624         return w->node && !(w->state == FWS_U && w->node == w->root);
2625 }
2626
2627 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2628         __releases(RCU_BH)
2629 {
2630         struct net *net = seq_file_net(seq);
2631         struct ipv6_route_iter *iter = seq->private;
2632
2633         if (ipv6_route_iter_active(iter))
2634                 fib6_walker_unlink(net, &iter->w);
2635
2636         rcu_read_unlock_bh();
2637 }
2638
2639 const struct seq_operations ipv6_route_seq_ops = {
2640         .start  = ipv6_route_seq_start,
2641         .next   = ipv6_route_seq_next,
2642         .stop   = ipv6_route_seq_stop,
2643         .show   = ipv6_route_seq_show
2644 };
2645 #endif /* CONFIG_PROC_FS */