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/bpf.h>
19 #include <linux/errno.h>
20 #include <linux/types.h>
21 #include <linux/net.h>
22 #include <linux/route.h>
23 #include <linux/netdevice.h>
24 #include <linux/in6.h>
25 #include <linux/init.h>
26 #include <linux/list.h>
27 #include <linux/slab.h>
28
29 #include <net/ip.h>
30 #include <net/ipv6.h>
31 #include <net/ndisc.h>
32 #include <net/addrconf.h>
33 #include <net/lwtunnel.h>
34 #include <net/fib_notifier.h>
35
36 #include <net/ip_fib.h>
37 #include <net/ip6_fib.h>
38 #include <net/ip6_route.h>
39
40 static struct kmem_cache *fib6_node_kmem __read_mostly;
41
42 struct fib6_cleaner {
43 struct fib6_walker w;
44 struct net *net;
45 int (*func)(struct fib6_info *, void *arg);
46 int sernum;
47 void *arg;
48 bool skip_notify;
49 };
50
51 #ifdef CONFIG_IPV6_SUBTREES
52 #define FWS_INIT FWS_S
53 #else
54 #define FWS_INIT FWS_L
55 #endif
56
57 static struct fib6_info *fib6_find_prefix(struct net *net,
58 struct fib6_table *table,
59 struct fib6_node *fn);
60 static struct fib6_node *fib6_repair_tree(struct net *net,
61 struct fib6_table *table,
62 struct fib6_node *fn);
63 static int fib6_walk(struct net *net, struct fib6_walker *w);
64 static int fib6_walk_continue(struct fib6_walker *w);
65
66 /*
67 * A routing update causes an increase of the serial number on the
68 * affected subtree. This allows for cached routes to be asynchronously
69 * tested when modifications are made to the destination cache as a
70 * result of redirects, path MTU changes, etc.
71 */
72
73 static void fib6_gc_timer_cb(struct timer_list *t);
74
75 #define FOR_WALKERS(net, w) \
76 list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
77
78 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
79 {
80 write_lock_bh(&net->ipv6.fib6_walker_lock);
81 list_add(&w->lh, &net->ipv6.fib6_walkers);
82 write_unlock_bh(&net->ipv6.fib6_walker_lock);
83 }
84
85 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
86 {
87 write_lock_bh(&net->ipv6.fib6_walker_lock);
88 list_del(&w->lh);
89 write_unlock_bh(&net->ipv6.fib6_walker_lock);
90 }
91
92 static int fib6_new_sernum(struct net *net)
93 {
94 int new, old = atomic_read(&net->ipv6.fib6_sernum);
95
96 do {
97 new = old < INT_MAX ? old + 1 : 1;
98 } while (!atomic_try_cmpxchg(&net->ipv6.fib6_sernum, &old, new));
99
100 return new;
101 }
102
103 enum {
104 FIB6_NO_SERNUM_CHANGE = 0,
105 };
106
107 void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
108 {
109 struct fib6_node *fn;
110
111 fn = rcu_dereference_protected(f6i->fib6_node,
112 lockdep_is_held(&f6i->fib6_table->tb6_lock));
113 if (fn)
114 WRITE_ONCE(fn->fn_sernum, fib6_new_sernum(net));
115 }
116
117 /*
118 * Auxiliary address test functions for the radix tree.
119 *
120 * These assume a 32bit processor (although it will work on
121 * 64bit processors)
122 */
123
124 /*
125 * test bit
126 */
127 #if defined(__LITTLE_ENDIAN)
128 # define BITOP_BE32_SWIZZLE (0x1F & ~7)
129 #else
130 # define BITOP_BE32_SWIZZLE 0
131 #endif
132
133 static __be32 addr_bit_set(const void *token, int fn_bit)
134 {
135 const __be32 *addr = token;
136 /*
137 * Here,
138 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
139 * is optimized version of
140 * htonl(1 << ((~fn_bit)&0x1F))
141 * See include/asm-generic/bitops/le.h.
142 */
143 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
144 addr[fn_bit >> 5];
145 }
146
147 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags, bool with_fib6_nh)
148 {
149 struct fib6_info *f6i;
150 size_t sz = sizeof(*f6i);
151
152 if (with_fib6_nh)
153 sz += sizeof(struct fib6_nh);
154
155 f6i = kzalloc(sz, gfp_flags);
156 if (!f6i)
157 return NULL;
158
159 /* fib6_siblings is a union with nh_list, so this initializes both */
160 INIT_LIST_HEAD(&f6i->fib6_siblings);
161 refcount_set(&f6i->fib6_ref, 1);
162
163 INIT_HLIST_NODE(&f6i->gc_link);
164
165 return f6i;
166 }
167
168 void fib6_info_destroy_rcu(struct rcu_head *head)
169 {
170 struct fib6_info *f6i = container_of(head, struct fib6_info, rcu);
171
172 WARN_ON(f6i->fib6_node);
173
174 if (f6i->nh)
175 nexthop_put(f6i->nh);
176 else
177 fib6_nh_release(f6i->fib6_nh);
178
179 ip_fib_metrics_put(f6i->fib6_metrics);
180 kfree(f6i);
181 }
182 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
183
184 static struct fib6_node *node_alloc(struct net *net)
185 {
186 struct fib6_node *fn;
187
188 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
189 if (fn)
190 net->ipv6.rt6_stats->fib_nodes++;
191
192 return fn;
193 }
194
195 static void node_free_immediate(struct net *net, struct fib6_node *fn)
196 {
197 kmem_cache_free(fib6_node_kmem, fn);
198 net->ipv6.rt6_stats->fib_nodes--;
199 }
200
201 static void node_free_rcu(struct rcu_head *head)
202 {
203 struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
204
205 kmem_cache_free(fib6_node_kmem, fn);
206 }
207
208 static void node_free(struct net *net, struct fib6_node *fn)
209 {
210 call_rcu(&fn->rcu, node_free_rcu);
211 net->ipv6.rt6_stats->fib_nodes--;
212 }
213
214 static void fib6_free_table(struct fib6_table *table)
215 {
216 inetpeer_invalidate_tree(&table->tb6_peers);
217 kfree(table);
218 }
219
220 static void fib6_link_table(struct net *net, struct fib6_table *tb)
221 {
222 unsigned int h;
223
224 /*
225 * Initialize table lock at a single place to give lockdep a key,
226 * tables aren't visible prior to being linked to the list.
227 */
228 spin_lock_init(&tb->tb6_lock);
229 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
230
231 /*
232 * No protection necessary, this is the only list mutatation
233 * operation, tables never disappear once they exist.
234 */
235 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
236 }
237
238 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
239
240 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
241 {
242 struct fib6_table *table;
243
244 table = kzalloc(sizeof(*table), GFP_ATOMIC);
245 if (table) {
246 table->tb6_id = id;
247 rcu_assign_pointer(table->tb6_root.leaf,
248 net->ipv6.fib6_null_entry);
249 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
250 inet_peer_base_init(&table->tb6_peers);
251 INIT_HLIST_HEAD(&table->tb6_gc_hlist);
252 }
253
254 return table;
255 }
256
257 struct fib6_table *fib6_new_table(struct net *net, u32 id)
258 {
259 struct fib6_table *tb;
260
261 if (id == 0)
262 id = RT6_TABLE_MAIN;
263 tb = fib6_get_table(net, id);
264 if (tb)
265 return tb;
266
267 tb = fib6_alloc_table(net, id);
268 if (tb)
269 fib6_link_table(net, tb);
270
271 return tb;
272 }
273 EXPORT_SYMBOL_GPL(fib6_new_table);
274
275 struct fib6_table *fib6_get_table(struct net *net, u32 id)
276 {
277 struct fib6_table *tb;
278 struct hlist_head *head;
279 unsigned int h;
280
281 if (id == 0)
282 id = RT6_TABLE_MAIN;
283 h = id & (FIB6_TABLE_HASHSZ - 1);
284 rcu_read_lock();
285 head = &net->ipv6.fib_table_hash[h];
286 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
287 if (tb->tb6_id == id) {
288 rcu_read_unlock();
289 return tb;
290 }
291 }
292 rcu_read_unlock();
293
294 return NULL;
295 }
296 EXPORT_SYMBOL_GPL(fib6_get_table);
297
298 static void __net_init fib6_tables_init(struct net *net)
299 {
300 fib6_link_table(net, net->ipv6.fib6_main_tbl);
301 fib6_link_table(net, net->ipv6.fib6_local_tbl);
302 }
303 #else
304
305 struct fib6_table *fib6_new_table(struct net *net, u32 id)
306 {
307 return fib6_get_table(net, id);
308 }
309
310 struct fib6_table *fib6_get_table(struct net *net, u32 id)
311 {
312 return net->ipv6.fib6_main_tbl;
313 }
314
315 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
316 const struct sk_buff *skb,
317 int flags, pol_lookup_t lookup)
318 {
319 struct rt6_info *rt;
320
321 rt = pol_lookup_func(lookup,
322 net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
323 if (rt->dst.error == -EAGAIN) {
324 ip6_rt_put_flags(rt, flags);
325 rt = net->ipv6.ip6_null_entry;
326 if (!(flags & RT6_LOOKUP_F_DST_NOREF))
327 dst_hold(&rt->dst);
328 }
329
330 return &rt->dst;
331 }
332
333 /* called with rcu lock held; no reference taken on fib6_info */
334 int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
335 struct fib6_result *res, int flags)
336 {
337 return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6,
338 res, flags);
339 }
340
341 static void __net_init fib6_tables_init(struct net *net)
342 {
343 fib6_link_table(net, net->ipv6.fib6_main_tbl);
344 }
345
346 #endif
347
348 unsigned int fib6_tables_seq_read(struct net *net)
349 {
350 unsigned int h, fib_seq = 0;
351
352 rcu_read_lock();
353 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
354 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
355 struct fib6_table *tb;
356
357 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
358 fib_seq += tb->fib_seq;
359 }
360 rcu_read_unlock();
361
362 return fib_seq;
363 }
364
365 static int call_fib6_entry_notifier(struct notifier_block *nb,
366 enum fib_event_type event_type,
367 struct fib6_info *rt,
368 struct netlink_ext_ack *extack)
369 {
370 struct fib6_entry_notifier_info info = {
371 .info.extack = extack,
372 .rt = rt,
373 };
374
375 return call_fib6_notifier(nb, event_type, &info.info);
376 }
377
378 static int call_fib6_multipath_entry_notifier(struct notifier_block *nb,
379 enum fib_event_type event_type,
380 struct fib6_info *rt,
381 unsigned int nsiblings,
382 struct netlink_ext_ack *extack)
383 {
384 struct fib6_entry_notifier_info info = {
385 .info.extack = extack,
386 .rt = rt,
387 .nsiblings = nsiblings,
388 };
389
390 return call_fib6_notifier(nb, event_type, &info.info);
391 }
392
393 int call_fib6_entry_notifiers(struct net *net,
394 enum fib_event_type event_type,
395 struct fib6_info *rt,
396 struct netlink_ext_ack *extack)
397 {
398 struct fib6_entry_notifier_info info = {
399 .info.extack = extack,
400 .rt = rt,
401 };
402
403 rt->fib6_table->fib_seq++;
404 return call_fib6_notifiers(net, event_type, &info.info);
405 }
406
407 int call_fib6_multipath_entry_notifiers(struct net *net,
408 enum fib_event_type event_type,
409 struct fib6_info *rt,
410 unsigned int nsiblings,
411 struct netlink_ext_ack *extack)
412 {
413 struct fib6_entry_notifier_info info = {
414 .info.extack = extack,
415 .rt = rt,
416 .nsiblings = nsiblings,
417 };
418
419 rt->fib6_table->fib_seq++;
420 return call_fib6_notifiers(net, event_type, &info.info);
421 }
422
423 int call_fib6_entry_notifiers_replace(struct net *net, struct fib6_info *rt)
424 {
425 struct fib6_entry_notifier_info info = {
426 .rt = rt,
427 .nsiblings = rt->fib6_nsiblings,
428 };
429
430 rt->fib6_table->fib_seq++;
431 return call_fib6_notifiers(net, FIB_EVENT_ENTRY_REPLACE, &info.info);
432 }
433
434 struct fib6_dump_arg {
435 struct net *net;
436 struct notifier_block *nb;
437 struct netlink_ext_ack *extack;
438 };
439
440 static int fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
441 {
442 enum fib_event_type fib_event = FIB_EVENT_ENTRY_REPLACE;
443 int err;
444
445 if (!rt || rt == arg->net->ipv6.fib6_null_entry)
446 return 0;
447
448 if (rt->fib6_nsiblings)
449 err = call_fib6_multipath_entry_notifier(arg->nb, fib_event,
450 rt,
451 rt->fib6_nsiblings,
452 arg->extack);
453 else
454 err = call_fib6_entry_notifier(arg->nb, fib_event, rt,
455 arg->extack);
456
457 return err;
458 }
459
460 static int fib6_node_dump(struct fib6_walker *w)
461 {
462 int err;
463
464 err = fib6_rt_dump(w->leaf, w->args);
465 w->leaf = NULL;
466 return err;
467 }
468
469 static int fib6_table_dump(struct net *net, struct fib6_table *tb,
470 struct fib6_walker *w)
471 {
472 int err;
473
474 w->root = &tb->tb6_root;
475 spin_lock_bh(&tb->tb6_lock);
476 err = fib6_walk(net, w);
477 spin_unlock_bh(&tb->tb6_lock);
478 return err;
479 }
480
481 /* Called with rcu_read_lock() */
482 int fib6_tables_dump(struct net *net, struct notifier_block *nb,
483 struct netlink_ext_ack *extack)
484 {
485 struct fib6_dump_arg arg;
486 struct fib6_walker *w;
487 unsigned int h;
488 int err = 0;
489
490 w = kzalloc(sizeof(*w), GFP_ATOMIC);
491 if (!w)
492 return -ENOMEM;
493
494 w->func = fib6_node_dump;
495 arg.net = net;
496 arg.nb = nb;
497 arg.extack = extack;
498 w->args = &arg;
499
500 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
501 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
502 struct fib6_table *tb;
503
504 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
505 err = fib6_table_dump(net, tb, w);
506 if (err)
507 goto out;
508 }
509 }
510
511 out:
512 kfree(w);
513
514 /* The tree traversal function should never return a positive value. */
515 return err > 0 ? -EINVAL : err;
516 }
517
518 static int fib6_dump_node(struct fib6_walker *w)
519 {
520 int res;
521 struct fib6_info *rt;
522
523 for_each_fib6_walker_rt(w) {
524 res = rt6_dump_route(rt, w->args, w->skip_in_node);
525 if (res >= 0) {
526 /* Frame is full, suspend walking */
527 w->leaf = rt;
528
529 /* We'll restart from this node, so if some routes were
530 * already dumped, skip them next time.
531 */
532 w->skip_in_node += res;
533
534 return 1;
535 }
536 w->skip_in_node = 0;
537
538 /* Multipath routes are dumped in one route with the
539 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
540 * last sibling of this route (no need to dump the
541 * sibling routes again)
542 */
543 if (rt->fib6_nsiblings)
544 rt = list_last_entry(&rt->fib6_siblings,
545 struct fib6_info,
546 fib6_siblings);
547 }
548 w->leaf = NULL;
549 return 0;
550 }
551
552 static void fib6_dump_end(struct netlink_callback *cb)
553 {
554 struct net *net = sock_net(cb->skb->sk);
555 struct fib6_walker *w = (void *)cb->args[2];
556
557 if (w) {
558 if (cb->args[4]) {
559 cb->args[4] = 0;
560 fib6_walker_unlink(net, w);
561 }
562 cb->args[2] = 0;
563 kfree(w);
564 }
565 cb->done = (void *)cb->args[3];
566 cb->args[1] = 3;
567 }
568
569 static int fib6_dump_done(struct netlink_callback *cb)
570 {
571 fib6_dump_end(cb);
572 return cb->done ? cb->done(cb) : 0;
573 }
574
575 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
576 struct netlink_callback *cb)
577 {
578 struct net *net = sock_net(skb->sk);
579 struct fib6_walker *w;
580 int res;
581
582 w = (void *)cb->args[2];
583 w->root = &table->tb6_root;
584
585 if (cb->args[4] == 0) {
586 w->count = 0;
587 w->skip = 0;
588 w->skip_in_node = 0;
589
590 spin_lock_bh(&table->tb6_lock);
591 res = fib6_walk(net, w);
592 spin_unlock_bh(&table->tb6_lock);
593 if (res > 0) {
594 cb->args[4] = 1;
595 cb->args[5] = READ_ONCE(w->root->fn_sernum);
596 }
597 } else {
598 int sernum = READ_ONCE(w->root->fn_sernum);
599 if (cb->args[5] != sernum) {
600 /* Begin at the root if the tree changed */
601 cb->args[5] = sernum;
602 w->state = FWS_INIT;
603 w->node = w->root;
604 w->skip = w->count;
605 w->skip_in_node = 0;
606 } else
607 w->skip = 0;
608
609 spin_lock_bh(&table->tb6_lock);
610 res = fib6_walk_continue(w);
611 spin_unlock_bh(&table->tb6_lock);
612 if (res <= 0) {
613 fib6_walker_unlink(net, w);
614 cb->args[4] = 0;
615 }
616 }
617
618 return res;
619 }
620
621 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
622 {
623 struct rt6_rtnl_dump_arg arg = {
624 .filter.dump_exceptions = true,
625 .filter.dump_routes = true,
626 .filter.rtnl_held = false,
627 };
628 const struct nlmsghdr *nlh = cb->nlh;
629 struct net *net = sock_net(skb->sk);
630 unsigned int e = 0, s_e;
631 struct hlist_head *head;
632 struct fib6_walker *w;
633 struct fib6_table *tb;
634 unsigned int h, s_h;
635 int err = 0;
636
637 rcu_read_lock();
638 if (cb->strict_check) {
639 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
640 if (err < 0)
641 goto unlock;
642 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
643 struct rtmsg *rtm = nlmsg_data(nlh);
644
645 if (rtm->rtm_flags & RTM_F_PREFIX)
646 arg.filter.flags = RTM_F_PREFIX;
647 }
648
649 w = (void *)cb->args[2];
650 if (!w) {
651 /* New dump:
652 *
653 * 1. allocate and initialize walker.
654 */
655 w = kzalloc(sizeof(*w), GFP_ATOMIC);
656 if (!w) {
657 err = -ENOMEM;
658 goto unlock;
659 }
660 w->func = fib6_dump_node;
661 cb->args[2] = (long)w;
662
663 /* 2. hook callback destructor.
664 */
665 cb->args[3] = (long)cb->done;
666 cb->done = fib6_dump_done;
667
668 }
669
670 arg.skb = skb;
671 arg.cb = cb;
672 arg.net = net;
673 w->args = &arg;
674
675 if (arg.filter.table_id) {
676 tb = fib6_get_table(net, arg.filter.table_id);
677 if (!tb) {
678 if (rtnl_msg_family(cb->nlh) != PF_INET6)
679 goto unlock;
680
681 NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
682 err = -ENOENT;
683 goto unlock;
684 }
685
686 if (!cb->args[0]) {
687 err = fib6_dump_table(tb, skb, cb);
688 if (!err)
689 cb->args[0] = 1;
690 }
691 goto unlock;
692 }
693
694 s_h = cb->args[0];
695 s_e = cb->args[1];
696
697 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
698 e = 0;
699 head = &net->ipv6.fib_table_hash[h];
700 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
701 if (e < s_e)
702 goto next;
703 err = fib6_dump_table(tb, skb, cb);
704 if (err != 0)
705 goto out;
706 next:
707 e++;
708 }
709 }
710 out:
711 cb->args[1] = e;
712 cb->args[0] = h;
713
714 unlock:
715 rcu_read_unlock();
716 if (err <= 0)
717 fib6_dump_end(cb);
718 return err;
719 }
720
721 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
722 {
723 if (!f6i)
724 return;
725
726 if (f6i->fib6_metrics == &dst_default_metrics) {
727 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
728
729 if (!p)
730 return;
731
732 refcount_set(&p->refcnt, 1);
733 f6i->fib6_metrics = p;
734 }
735
736 f6i->fib6_metrics->metrics[metric - 1] = val;
737 }
738
739 /*
740 * Routing Table
741 *
742 * return the appropriate node for a routing tree "add" operation
743 * by either creating and inserting or by returning an existing
744 * node.
745 */
746
747 static struct fib6_node *fib6_add_1(struct net *net,
748 struct fib6_table *table,
749 struct fib6_node *root,
750 struct in6_addr *addr, int plen,
751 int offset, int allow_create,
752 int replace_required,
753 struct netlink_ext_ack *extack)
754 {
755 struct fib6_node *fn, *in, *ln;
756 struct fib6_node *pn = NULL;
757 struct rt6key *key;
758 int bit;
759 __be32 dir = 0;
760
761 /* insert node in tree */
762
763 fn = root;
764
765 do {
766 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
767 lockdep_is_held(&table->tb6_lock));
768 key = (struct rt6key *)((u8 *)leaf + offset);
769
770 /*
771 * Prefix match
772 */
773 if (plen < fn->fn_bit ||
774 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
775 if (!allow_create) {
776 if (replace_required) {
777 NL_SET_ERR_MSG(extack,
778 "Can not replace route - no match found");
779 pr_warn("Can't replace route, no match found\n");
780 return ERR_PTR(-ENOENT);
781 }
782 pr_warn("NLM_F_CREATE should be set when creating new route\n");
783 }
784 goto insert_above;
785 }
786
787 /*
788 * Exact match ?
789 */
790
791 if (plen == fn->fn_bit) {
792 /* clean up an intermediate node */
793 if (!(fn->fn_flags & RTN_RTINFO)) {
794 RCU_INIT_POINTER(fn->leaf, NULL);
795 fib6_info_release(leaf);
796 /* remove null_entry in the root node */
797 } else if (fn->fn_flags & RTN_TL_ROOT &&
798 rcu_access_pointer(fn->leaf) ==
799 net->ipv6.fib6_null_entry) {
800 RCU_INIT_POINTER(fn->leaf, NULL);
801 }
802
803 return fn;
804 }
805
806 /*
807 * We have more bits to go
808 */
809
810 /* Try to walk down on tree. */
811 dir = addr_bit_set(addr, fn->fn_bit);
812 pn = fn;
813 fn = dir ?
814 rcu_dereference_protected(fn->right,
815 lockdep_is_held(&table->tb6_lock)) :
816 rcu_dereference_protected(fn->left,
817 lockdep_is_held(&table->tb6_lock));
818 } while (fn);
819
820 if (!allow_create) {
821 /* We should not create new node because
822 * NLM_F_REPLACE was specified without NLM_F_CREATE
823 * I assume it is safe to require NLM_F_CREATE when
824 * REPLACE flag is used! Later we may want to remove the
825 * check for replace_required, because according
826 * to netlink specification, NLM_F_CREATE
827 * MUST be specified if new route is created.
828 * That would keep IPv6 consistent with IPv4
829 */
830 if (replace_required) {
831 NL_SET_ERR_MSG(extack,
832 "Can not replace route - no match found");
833 pr_warn("Can't replace route, no match found\n");
834 return ERR_PTR(-ENOENT);
835 }
836 pr_warn("NLM_F_CREATE should be set when creating new route\n");
837 }
838 /*
839 * We walked to the bottom of tree.
840 * Create new leaf node without children.
841 */
842
843 ln = node_alloc(net);
844
845 if (!ln)
846 return ERR_PTR(-ENOMEM);
847 ln->fn_bit = plen;
848 RCU_INIT_POINTER(ln->parent, pn);
849
850 if (dir)
851 rcu_assign_pointer(pn->right, ln);
852 else
853 rcu_assign_pointer(pn->left, ln);
854
855 return ln;
856
857
858 insert_above:
859 /*
860 * split since we don't have a common prefix anymore or
861 * we have a less significant route.
862 * we've to insert an intermediate node on the list
863 * this new node will point to the one we need to create
864 * and the current
865 */
866
867 pn = rcu_dereference_protected(fn->parent,
868 lockdep_is_held(&table->tb6_lock));
869
870 /* find 1st bit in difference between the 2 addrs.
871
872 See comment in __ipv6_addr_diff: bit may be an invalid value,
873 but if it is >= plen, the value is ignored in any case.
874 */
875
876 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
877
878 /*
879 * (intermediate)[in]
880 * / \
881 * (new leaf node)[ln] (old node)[fn]
882 */
883 if (plen > bit) {
884 in = node_alloc(net);
885 ln = node_alloc(net);
886
887 if (!in || !ln) {
888 if (in)
889 node_free_immediate(net, in);
890 if (ln)
891 node_free_immediate(net, ln);
892 return ERR_PTR(-ENOMEM);
893 }
894
895 /*
896 * new intermediate node.
897 * RTN_RTINFO will
898 * be off since that an address that chooses one of
899 * the branches would not match less specific routes
900 * in the other branch
901 */
902
903 in->fn_bit = bit;
904
905 RCU_INIT_POINTER(in->parent, pn);
906 in->leaf = fn->leaf;
907 fib6_info_hold(rcu_dereference_protected(in->leaf,
908 lockdep_is_held(&table->tb6_lock)));
909
910 /* update parent pointer */
911 if (dir)
912 rcu_assign_pointer(pn->right, in);
913 else
914 rcu_assign_pointer(pn->left, in);
915
916 ln->fn_bit = plen;
917
918 RCU_INIT_POINTER(ln->parent, in);
919 rcu_assign_pointer(fn->parent, in);
920
921 if (addr_bit_set(addr, bit)) {
922 rcu_assign_pointer(in->right, ln);
923 rcu_assign_pointer(in->left, fn);
924 } else {
925 rcu_assign_pointer(in->left, ln);
926 rcu_assign_pointer(in->right, fn);
927 }
928 } else { /* plen <= bit */
929
930 /*
931 * (new leaf node)[ln]
932 * / \
933 * (old node)[fn] NULL
934 */
935
936 ln = node_alloc(net);
937
938 if (!ln)
939 return ERR_PTR(-ENOMEM);
940
941 ln->fn_bit = plen;
942
943 RCU_INIT_POINTER(ln->parent, pn);
944
945 if (addr_bit_set(&key->addr, plen))
946 RCU_INIT_POINTER(ln->right, fn);
947 else
948 RCU_INIT_POINTER(ln->left, fn);
949
950 rcu_assign_pointer(fn->parent, ln);
951
952 if (dir)
953 rcu_assign_pointer(pn->right, ln);
954 else
955 rcu_assign_pointer(pn->left, ln);
956 }
957 return ln;
958 }
959
960 static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
961 const struct fib6_info *match,
962 const struct fib6_table *table)
963 {
964 int cpu;
965
966 if (!fib6_nh->rt6i_pcpu)
967 return;
968
969 rcu_read_lock();
970 /* release the reference to this fib entry from
971 * all of its cached pcpu routes
972 */
973 for_each_possible_cpu(cpu) {
974 struct rt6_info **ppcpu_rt;
975 struct rt6_info *pcpu_rt;
976
977 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
978
979 /* Paired with xchg() in rt6_get_pcpu_route() */
980 pcpu_rt = READ_ONCE(*ppcpu_rt);
981
982 /* only dropping the 'from' reference if the cached route
983 * is using 'match'. The cached pcpu_rt->from only changes
984 * from a fib6_info to NULL (ip6_dst_destroy); it can never
985 * change from one fib6_info reference to another
986 */
987 if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
988 struct fib6_info *from;
989
990 from = unrcu_pointer(xchg(&pcpu_rt->from, NULL));
991 fib6_info_release(from);
992 }
993 }
994 rcu_read_unlock();
995 }
996
997 struct fib6_nh_pcpu_arg {
998 struct fib6_info *from;
999 const struct fib6_table *table;
1000 };
1001
1002 static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg)
1003 {
1004 struct fib6_nh_pcpu_arg *arg = _arg;
1005
1006 __fib6_drop_pcpu_from(nh, arg->from, arg->table);
1007 return 0;
1008 }
1009
1010 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
1011 const struct fib6_table *table)
1012 {
1013 /* Make sure rt6_make_pcpu_route() wont add other percpu routes
1014 * while we are cleaning them here.
1015 */
1016 f6i->fib6_destroying = 1;
1017 mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
1018
1019 if (f6i->nh) {
1020 struct fib6_nh_pcpu_arg arg = {
1021 .from = f6i,
1022 .table = table
1023 };
1024
1025 nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from,
1026 &arg);
1027 } else {
1028 struct fib6_nh *fib6_nh;
1029
1030 fib6_nh = f6i->fib6_nh;
1031 __fib6_drop_pcpu_from(fib6_nh, f6i, table);
1032 }
1033 }
1034
1035 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
1036 struct net *net)
1037 {
1038 struct fib6_table *table = rt->fib6_table;
1039
1040 /* Flush all cached dst in exception table */
1041 rt6_flush_exceptions(rt);
1042 fib6_drop_pcpu_from(rt, table);
1043
1044 if (rt->nh && !list_empty(&rt->nh_list))
1045 list_del_init(&rt->nh_list);
1046
1047 if (refcount_read(&rt->fib6_ref) != 1) {
1048 /* This route is used as dummy address holder in some split
1049 * nodes. It is not leaked, but it still holds other resources,
1050 * which must be released in time. So, scan ascendant nodes
1051 * and replace dummy references to this route with references
1052 * to still alive ones.
1053 */
1054 while (fn) {
1055 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1056 lockdep_is_held(&table->tb6_lock));
1057 struct fib6_info *new_leaf;
1058 if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
1059 new_leaf = fib6_find_prefix(net, table, fn);
1060 fib6_info_hold(new_leaf);
1061
1062 rcu_assign_pointer(fn->leaf, new_leaf);
1063 fib6_info_release(rt);
1064 }
1065 fn = rcu_dereference_protected(fn->parent,
1066 lockdep_is_held(&table->tb6_lock));
1067 }
1068 }
1069
1070 fib6_clean_expires(rt);
1071 fib6_remove_gc_list(rt);
1072 }
1073
1074 /*
1075 * Insert routing information in a node.
1076 */
1077
1078 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
1079 struct nl_info *info,
1080 struct netlink_ext_ack *extack)
1081 {
1082 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1083 lockdep_is_held(&rt->fib6_table->tb6_lock));
1084 struct fib6_info *iter = NULL;
1085 struct fib6_info __rcu **ins;
1086 struct fib6_info __rcu **fallback_ins = NULL;
1087 int replace = (info->nlh &&
1088 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
1089 int add = (!info->nlh ||
1090 (info->nlh->nlmsg_flags & NLM_F_CREATE));
1091 int found = 0;
1092 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
1093 bool notify_sibling_rt = false;
1094 u16 nlflags = NLM_F_EXCL;
1095 int err;
1096
1097 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
1098 nlflags |= NLM_F_APPEND;
1099
1100 ins = &fn->leaf;
1101
1102 for (iter = leaf; iter;
1103 iter = rcu_dereference_protected(iter->fib6_next,
1104 lockdep_is_held(&rt->fib6_table->tb6_lock))) {
1105 /*
1106 * Search for duplicates
1107 */
1108
1109 if (iter->fib6_metric == rt->fib6_metric) {
1110 /*
1111 * Same priority level
1112 */
1113 if (info->nlh &&
1114 (info->nlh->nlmsg_flags & NLM_F_EXCL))
1115 return -EEXIST;
1116
1117 nlflags &= ~NLM_F_EXCL;
1118 if (replace) {
1119 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1120 found++;
1121 break;
1122 }
1123 fallback_ins = fallback_ins ?: ins;
1124 goto next_iter;
1125 }
1126
1127 if (rt6_duplicate_nexthop(iter, rt)) {
1128 if (rt->fib6_nsiblings)
1129 rt->fib6_nsiblings = 0;
1130 if (!(iter->fib6_flags & RTF_EXPIRES))
1131 return -EEXIST;
1132 if (!(rt->fib6_flags & RTF_EXPIRES)) {
1133 fib6_clean_expires(iter);
1134 fib6_remove_gc_list(iter);
1135 } else {
1136 fib6_set_expires(iter, rt->expires);
1137 fib6_add_gc_list(iter);
1138 }
1139
1140 if (rt->fib6_pmtu)
1141 fib6_metric_set(iter, RTAX_MTU,
1142 rt->fib6_pmtu);
1143 return -EEXIST;
1144 }
1145 /* If we have the same destination and the same metric,
1146 * but not the same gateway, then the route we try to
1147 * add is sibling to this route, increment our counter
1148 * of siblings, and later we will add our route to the
1149 * list.
1150 * Only static routes (which don't have flag
1151 * RTF_EXPIRES) are used for ECMPv6.
1152 *
1153 * To avoid long list, we only had siblings if the
1154 * route have a gateway.
1155 */
1156 if (rt_can_ecmp &&
1157 rt6_qualify_for_ecmp(iter))
1158 rt->fib6_nsiblings++;
1159 }
1160
1161 if (iter->fib6_metric > rt->fib6_metric)
1162 break;
1163
1164 next_iter:
1165 ins = &iter->fib6_next;
1166 }
1167
1168 if (fallback_ins && !found) {
1169 /* No matching route with same ecmp-able-ness found, replace
1170 * first matching route
1171 */
1172 ins = fallback_ins;
1173 iter = rcu_dereference_protected(*ins,
1174 lockdep_is_held(&rt->fib6_table->tb6_lock));
1175 found++;
1176 }
1177
1178 /* Reset round-robin state, if necessary */
1179 if (ins == &fn->leaf)
1180 fn->rr_ptr = NULL;
1181
1182 /* Link this route to others same route. */
1183 if (rt->fib6_nsiblings) {
1184 unsigned int fib6_nsiblings;
1185 struct fib6_info *sibling, *temp_sibling;
1186
1187 /* Find the first route that have the same metric */
1188 sibling = leaf;
1189 notify_sibling_rt = true;
1190 while (sibling) {
1191 if (sibling->fib6_metric == rt->fib6_metric &&
1192 rt6_qualify_for_ecmp(sibling)) {
1193 list_add_tail(&rt->fib6_siblings,
1194 &sibling->fib6_siblings);
1195 break;
1196 }
1197 sibling = rcu_dereference_protected(sibling->fib6_next,
1198 lockdep_is_held(&rt->fib6_table->tb6_lock));
1199 notify_sibling_rt = false;
1200 }
1201 /* For each sibling in the list, increment the counter of
1202 * siblings. BUG() if counters does not match, list of siblings
1203 * is broken!
1204 */
1205 fib6_nsiblings = 0;
1206 list_for_each_entry_safe(sibling, temp_sibling,
1207 &rt->fib6_siblings, fib6_siblings) {
1208 sibling->fib6_nsiblings++;
1209 BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1210 fib6_nsiblings++;
1211 }
1212 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1213 rt6_multipath_rebalance(temp_sibling);
1214 }
1215
1216 /*
1217 * insert node
1218 */
1219 if (!replace) {
1220 if (!add)
1221 pr_warn("NLM_F_CREATE should be set when creating new route\n");
1222
1223 add:
1224 nlflags |= NLM_F_CREATE;
1225
1226 /* The route should only be notified if it is the first
1227 * route in the node or if it is added as a sibling
1228 * route to the first route in the node.
1229 */
1230 if (!info->skip_notify_kernel &&
1231 (notify_sibling_rt || ins == &fn->leaf)) {
1232 enum fib_event_type fib_event;
1233
1234 if (notify_sibling_rt)
1235 fib_event = FIB_EVENT_ENTRY_APPEND;
1236 else
1237 fib_event = FIB_EVENT_ENTRY_REPLACE;
1238 err = call_fib6_entry_notifiers(info->nl_net,
1239 fib_event, rt,
1240 extack);
1241 if (err) {
1242 struct fib6_info *sibling, *next_sibling;
1243
1244 /* If the route has siblings, then it first
1245 * needs to be unlinked from them.
1246 */
1247 if (!rt->fib6_nsiblings)
1248 return err;
1249
1250 list_for_each_entry_safe(sibling, next_sibling,
1251 &rt->fib6_siblings,
1252 fib6_siblings)
1253 sibling->fib6_nsiblings--;
1254 rt->fib6_nsiblings = 0;
1255 list_del_init(&rt->fib6_siblings);
1256 rt6_multipath_rebalance(next_sibling);
1257 return err;
1258 }
1259 }
1260
1261 rcu_assign_pointer(rt->fib6_next, iter);
1262 fib6_info_hold(rt);
1263 rcu_assign_pointer(rt->fib6_node, fn);
1264 rcu_assign_pointer(*ins, rt);
1265 if (!info->skip_notify)
1266 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1267 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1268
1269 if (!(fn->fn_flags & RTN_RTINFO)) {
1270 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1271 fn->fn_flags |= RTN_RTINFO;
1272 }
1273
1274 } else {
1275 int nsiblings;
1276
1277 if (!found) {
1278 if (add)
1279 goto add;
1280 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1281 return -ENOENT;
1282 }
1283
1284 if (!info->skip_notify_kernel && ins == &fn->leaf) {
1285 err = call_fib6_entry_notifiers(info->nl_net,
1286 FIB_EVENT_ENTRY_REPLACE,
1287 rt, extack);
1288 if (err)
1289 return err;
1290 }
1291
1292 fib6_info_hold(rt);
1293 rcu_assign_pointer(rt->fib6_node, fn);
1294 rt->fib6_next = iter->fib6_next;
1295 rcu_assign_pointer(*ins, rt);
1296 if (!info->skip_notify)
1297 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1298 if (!(fn->fn_flags & RTN_RTINFO)) {
1299 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1300 fn->fn_flags |= RTN_RTINFO;
1301 }
1302 nsiblings = iter->fib6_nsiblings;
1303 iter->fib6_node = NULL;
1304 fib6_purge_rt(iter, fn, info->nl_net);
1305 if (rcu_access_pointer(fn->rr_ptr) == iter)
1306 fn->rr_ptr = NULL;
1307 fib6_info_release(iter);
1308
1309 if (nsiblings) {
1310 /* Replacing an ECMP route, remove all siblings */
1311 ins = &rt->fib6_next;
1312 iter = rcu_dereference_protected(*ins,
1313 lockdep_is_held(&rt->fib6_table->tb6_lock));
1314 while (iter) {
1315 if (iter->fib6_metric > rt->fib6_metric)
1316 break;
1317 if (rt6_qualify_for_ecmp(iter)) {
1318 *ins = iter->fib6_next;
1319 iter->fib6_node = NULL;
1320 fib6_purge_rt(iter, fn, info->nl_net);
1321 if (rcu_access_pointer(fn->rr_ptr) == iter)
1322 fn->rr_ptr = NULL;
1323 fib6_info_release(iter);
1324 nsiblings--;
1325 info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1326 } else {
1327 ins = &iter->fib6_next;
1328 }
1329 iter = rcu_dereference_protected(*ins,
1330 lockdep_is_held(&rt->fib6_table->tb6_lock));
1331 }
1332 WARN_ON(nsiblings != 0);
1333 }
1334 }
1335
1336 return 0;
1337 }
1338
1339 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1340 {
1341 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1342 (rt->fib6_flags & RTF_EXPIRES))
1343 mod_timer(&net->ipv6.ip6_fib_timer,
1344 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1345 }
1346
1347 void fib6_force_start_gc(struct net *net)
1348 {
1349 if (!timer_pending(&net->ipv6.ip6_fib_timer))
1350 mod_timer(&net->ipv6.ip6_fib_timer,
1351 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1352 }
1353
1354 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1355 int sernum)
1356 {
1357 struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1358 lockdep_is_held(&rt->fib6_table->tb6_lock));
1359
1360 /* paired with smp_rmb() in fib6_get_cookie_safe() */
1361 smp_wmb();
1362 while (fn) {
1363 WRITE_ONCE(fn->fn_sernum, sernum);
1364 fn = rcu_dereference_protected(fn->parent,
1365 lockdep_is_held(&rt->fib6_table->tb6_lock));
1366 }
1367 }
1368
1369 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1370 {
1371 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1372 }
1373
1374 /* allow ipv4 to update sernum via ipv6_stub */
1375 void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1376 {
1377 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1378 fib6_update_sernum_upto_root(net, f6i);
1379 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1380 }
1381
1382 /*
1383 * Add routing information to the routing tree.
1384 * <destination addr>/<source addr>
1385 * with source addr info in sub-trees
1386 * Need to own table->tb6_lock
1387 */
1388
1389 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1390 struct nl_info *info, struct netlink_ext_ack *extack)
1391 {
1392 struct fib6_table *table = rt->fib6_table;
1393 struct fib6_node *fn;
1394 #ifdef CONFIG_IPV6_SUBTREES
1395 struct fib6_node *pn = NULL;
1396 #endif
1397 int err = -ENOMEM;
1398 int allow_create = 1;
1399 int replace_required = 0;
1400
1401 if (info->nlh) {
1402 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1403 allow_create = 0;
1404 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1405 replace_required = 1;
1406 }
1407 if (!allow_create && !replace_required)
1408 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1409
1410 fn = fib6_add_1(info->nl_net, table, root,
1411 &rt->fib6_dst.addr, rt->fib6_dst.plen,
1412 offsetof(struct fib6_info, fib6_dst), allow_create,
1413 replace_required, extack);
1414 if (IS_ERR(fn)) {
1415 err = PTR_ERR(fn);
1416 fn = NULL;
1417 goto out;
1418 }
1419
1420 #ifdef CONFIG_IPV6_SUBTREES
1421 pn = fn;
1422
1423 if (rt->fib6_src.plen) {
1424 struct fib6_node *sn;
1425
1426 if (!rcu_access_pointer(fn->subtree)) {
1427 struct fib6_node *sfn;
1428
1429 /*
1430 * Create subtree.
1431 *
1432 * fn[main tree]
1433 * |
1434 * sfn[subtree root]
1435 * \
1436 * sn[new leaf node]
1437 */
1438
1439 /* Create subtree root node */
1440 sfn = node_alloc(info->nl_net);
1441 if (!sfn)
1442 goto failure;
1443
1444 fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1445 rcu_assign_pointer(sfn->leaf,
1446 info->nl_net->ipv6.fib6_null_entry);
1447 sfn->fn_flags = RTN_ROOT;
1448
1449 /* Now add the first leaf node to new subtree */
1450
1451 sn = fib6_add_1(info->nl_net, table, sfn,
1452 &rt->fib6_src.addr, rt->fib6_src.plen,
1453 offsetof(struct fib6_info, fib6_src),
1454 allow_create, replace_required, extack);
1455
1456 if (IS_ERR(sn)) {
1457 /* If it is failed, discard just allocated
1458 root, and then (in failure) stale node
1459 in main tree.
1460 */
1461 node_free_immediate(info->nl_net, sfn);
1462 err = PTR_ERR(sn);
1463 goto failure;
1464 }
1465
1466 /* Now link new subtree to main tree */
1467 rcu_assign_pointer(sfn->parent, fn);
1468 rcu_assign_pointer(fn->subtree, sfn);
1469 } else {
1470 sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1471 &rt->fib6_src.addr, rt->fib6_src.plen,
1472 offsetof(struct fib6_info, fib6_src),
1473 allow_create, replace_required, extack);
1474
1475 if (IS_ERR(sn)) {
1476 err = PTR_ERR(sn);
1477 goto failure;
1478 }
1479 }
1480
1481 if (!rcu_access_pointer(fn->leaf)) {
1482 if (fn->fn_flags & RTN_TL_ROOT) {
1483 /* put back null_entry for root node */
1484 rcu_assign_pointer(fn->leaf,
1485 info->nl_net->ipv6.fib6_null_entry);
1486 } else {
1487 fib6_info_hold(rt);
1488 rcu_assign_pointer(fn->leaf, rt);
1489 }
1490 }
1491 fn = sn;
1492 }
1493 #endif
1494
1495 err = fib6_add_rt2node(fn, rt, info, extack);
1496 if (!err) {
1497 if (rt->nh)
1498 list_add(&rt->nh_list, &rt->nh->f6i_list);
1499 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(info->nl_net));
1500
1501 if (rt->fib6_flags & RTF_EXPIRES)
1502 fib6_add_gc_list(rt);
1503
1504 fib6_start_gc(info->nl_net, rt);
1505 }
1506
1507 out:
1508 if (err) {
1509 #ifdef CONFIG_IPV6_SUBTREES
1510 /*
1511 * If fib6_add_1 has cleared the old leaf pointer in the
1512 * super-tree leaf node we have to find a new one for it.
1513 */
1514 if (pn != fn) {
1515 struct fib6_info *pn_leaf =
1516 rcu_dereference_protected(pn->leaf,
1517 lockdep_is_held(&table->tb6_lock));
1518 if (pn_leaf == rt) {
1519 pn_leaf = NULL;
1520 RCU_INIT_POINTER(pn->leaf, NULL);
1521 fib6_info_release(rt);
1522 }
1523 if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1524 pn_leaf = fib6_find_prefix(info->nl_net, table,
1525 pn);
1526 if (!pn_leaf)
1527 pn_leaf =
1528 info->nl_net->ipv6.fib6_null_entry;
1529 fib6_info_hold(pn_leaf);
1530 rcu_assign_pointer(pn->leaf, pn_leaf);
1531 }
1532 }
1533 #endif
1534 goto failure;
1535 } else if (fib6_requires_src(rt)) {
1536 fib6_routes_require_src_inc(info->nl_net);
1537 }
1538 return err;
1539
1540 failure:
1541 /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1542 * 1. fn is an intermediate node and we failed to add the new
1543 * route to it in both subtree creation failure and fib6_add_rt2node()
1544 * failure case.
1545 * 2. fn is the root node in the table and we fail to add the first
1546 * default route to it.
1547 */
1548 if (fn &&
1549 (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1550 (fn->fn_flags & RTN_TL_ROOT &&
1551 !rcu_access_pointer(fn->leaf))))
1552 fib6_repair_tree(info->nl_net, table, fn);
1553 return err;
1554 }
1555
1556 /*
1557 * Routing tree lookup
1558 *
1559 */
1560
1561 struct lookup_args {
1562 int offset; /* key offset on fib6_info */
1563 const struct in6_addr *addr; /* search key */
1564 };
1565
1566 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1567 struct lookup_args *args)
1568 {
1569 struct fib6_node *fn;
1570 __be32 dir;
1571
1572 if (unlikely(args->offset == 0))
1573 return NULL;
1574
1575 /*
1576 * Descend on a tree
1577 */
1578
1579 fn = root;
1580
1581 for (;;) {
1582 struct fib6_node *next;
1583
1584 dir = addr_bit_set(args->addr, fn->fn_bit);
1585
1586 next = dir ? rcu_dereference(fn->right) :
1587 rcu_dereference(fn->left);
1588
1589 if (next) {
1590 fn = next;
1591 continue;
1592 }
1593 break;
1594 }
1595
1596 while (fn) {
1597 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1598
1599 if (subtree || fn->fn_flags & RTN_RTINFO) {
1600 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1601 struct rt6key *key;
1602
1603 if (!leaf)
1604 goto backtrack;
1605
1606 key = (struct rt6key *) ((u8 *)leaf + args->offset);
1607
1608 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1609 #ifdef CONFIG_IPV6_SUBTREES
1610 if (subtree) {
1611 struct fib6_node *sfn;
1612 sfn = fib6_node_lookup_1(subtree,
1613 args + 1);
1614 if (!sfn)
1615 goto backtrack;
1616 fn = sfn;
1617 }
1618 #endif
1619 if (fn->fn_flags & RTN_RTINFO)
1620 return fn;
1621 }
1622 }
1623 backtrack:
1624 if (fn->fn_flags & RTN_ROOT)
1625 break;
1626
1627 fn = rcu_dereference(fn->parent);
1628 }
1629
1630 return NULL;
1631 }
1632
1633 /* called with rcu_read_lock() held
1634 */
1635 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1636 const struct in6_addr *daddr,
1637 const struct in6_addr *saddr)
1638 {
1639 struct fib6_node *fn;
1640 struct lookup_args args[] = {
1641 {
1642 .offset = offsetof(struct fib6_info, fib6_dst),
1643 .addr = daddr,
1644 },
1645 #ifdef CONFIG_IPV6_SUBTREES
1646 {
1647 .offset = offsetof(struct fib6_info, fib6_src),
1648 .addr = saddr,
1649 },
1650 #endif
1651 {
1652 .offset = 0, /* sentinel */
1653 }
1654 };
1655
1656 fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1657 if (!fn || fn->fn_flags & RTN_TL_ROOT)
1658 fn = root;
1659
1660 return fn;
1661 }
1662
1663 /*
1664 * Get node with specified destination prefix (and source prefix,
1665 * if subtrees are used)
1666 * exact_match == true means we try to find fn with exact match of
1667 * the passed in prefix addr
1668 * exact_match == false means we try to find fn with longest prefix
1669 * match of the passed in prefix addr. This is useful for finding fn
1670 * for cached route as it will be stored in the exception table under
1671 * the node with longest prefix length.
1672 */
1673
1674
1675 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1676 const struct in6_addr *addr,
1677 int plen, int offset,
1678 bool exact_match)
1679 {
1680 struct fib6_node *fn, *prev = NULL;
1681
1682 for (fn = root; fn ; ) {
1683 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1684 struct rt6key *key;
1685
1686 /* This node is being deleted */
1687 if (!leaf) {
1688 if (plen <= fn->fn_bit)
1689 goto out;
1690 else
1691 goto next;
1692 }
1693
1694 key = (struct rt6key *)((u8 *)leaf + offset);
1695
1696 /*
1697 * Prefix match
1698 */
1699 if (plen < fn->fn_bit ||
1700 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1701 goto out;
1702
1703 if (plen == fn->fn_bit)
1704 return fn;
1705
1706 if (fn->fn_flags & RTN_RTINFO)
1707 prev = fn;
1708
1709 next:
1710 /*
1711 * We have more bits to go
1712 */
1713 if (addr_bit_set(addr, fn->fn_bit))
1714 fn = rcu_dereference(fn->right);
1715 else
1716 fn = rcu_dereference(fn->left);
1717 }
1718 out:
1719 if (exact_match)
1720 return NULL;
1721 else
1722 return prev;
1723 }
1724
1725 struct fib6_node *fib6_locate(struct fib6_node *root,
1726 const struct in6_addr *daddr, int dst_len,
1727 const struct in6_addr *saddr, int src_len,
1728 bool exact_match)
1729 {
1730 struct fib6_node *fn;
1731
1732 fn = fib6_locate_1(root, daddr, dst_len,
1733 offsetof(struct fib6_info, fib6_dst),
1734 exact_match);
1735
1736 #ifdef CONFIG_IPV6_SUBTREES
1737 if (src_len) {
1738 WARN_ON(saddr == NULL);
1739 if (fn) {
1740 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1741
1742 if (subtree) {
1743 fn = fib6_locate_1(subtree, saddr, src_len,
1744 offsetof(struct fib6_info, fib6_src),
1745 exact_match);
1746 }
1747 }
1748 }
1749 #endif
1750
1751 if (fn && fn->fn_flags & RTN_RTINFO)
1752 return fn;
1753
1754 return NULL;
1755 }
1756
1757
1758 /*
1759 * Deletion
1760 *
1761 */
1762
1763 static struct fib6_info *fib6_find_prefix(struct net *net,
1764 struct fib6_table *table,
1765 struct fib6_node *fn)
1766 {
1767 struct fib6_node *child_left, *child_right;
1768
1769 if (fn->fn_flags & RTN_ROOT)
1770 return net->ipv6.fib6_null_entry;
1771
1772 while (fn) {
1773 child_left = rcu_dereference_protected(fn->left,
1774 lockdep_is_held(&table->tb6_lock));
1775 child_right = rcu_dereference_protected(fn->right,
1776 lockdep_is_held(&table->tb6_lock));
1777 if (child_left)
1778 return rcu_dereference_protected(child_left->leaf,
1779 lockdep_is_held(&table->tb6_lock));
1780 if (child_right)
1781 return rcu_dereference_protected(child_right->leaf,
1782 lockdep_is_held(&table->tb6_lock));
1783
1784 fn = FIB6_SUBTREE(fn);
1785 }
1786 return NULL;
1787 }
1788
1789 /*
1790 * Called to trim the tree of intermediate nodes when possible. "fn"
1791 * is the node we want to try and remove.
1792 * Need to own table->tb6_lock
1793 */
1794
1795 static struct fib6_node *fib6_repair_tree(struct net *net,
1796 struct fib6_table *table,
1797 struct fib6_node *fn)
1798 {
1799 int children;
1800 int nstate;
1801 struct fib6_node *child;
1802 struct fib6_walker *w;
1803 int iter = 0;
1804
1805 /* Set fn->leaf to null_entry for root node. */
1806 if (fn->fn_flags & RTN_TL_ROOT) {
1807 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1808 return fn;
1809 }
1810
1811 for (;;) {
1812 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1813 lockdep_is_held(&table->tb6_lock));
1814 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1815 lockdep_is_held(&table->tb6_lock));
1816 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1817 lockdep_is_held(&table->tb6_lock));
1818 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1819 lockdep_is_held(&table->tb6_lock));
1820 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1821 lockdep_is_held(&table->tb6_lock));
1822 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1823 lockdep_is_held(&table->tb6_lock));
1824 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1825 lockdep_is_held(&table->tb6_lock));
1826 struct fib6_info *new_fn_leaf;
1827
1828 pr_debug("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1829 iter++;
1830
1831 WARN_ON(fn->fn_flags & RTN_RTINFO);
1832 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1833 WARN_ON(fn_leaf);
1834
1835 children = 0;
1836 child = NULL;
1837 if (fn_r) {
1838 child = fn_r;
1839 children |= 1;
1840 }
1841 if (fn_l) {
1842 child = fn_l;
1843 children |= 2;
1844 }
1845
1846 if (children == 3 || FIB6_SUBTREE(fn)
1847 #ifdef CONFIG_IPV6_SUBTREES
1848 /* Subtree root (i.e. fn) may have one child */
1849 || (children && fn->fn_flags & RTN_ROOT)
1850 #endif
1851 ) {
1852 new_fn_leaf = fib6_find_prefix(net, table, fn);
1853 #if RT6_DEBUG >= 2
1854 if (!new_fn_leaf) {
1855 WARN_ON(!new_fn_leaf);
1856 new_fn_leaf = net->ipv6.fib6_null_entry;
1857 }
1858 #endif
1859 fib6_info_hold(new_fn_leaf);
1860 rcu_assign_pointer(fn->leaf, new_fn_leaf);
1861 return pn;
1862 }
1863
1864 #ifdef CONFIG_IPV6_SUBTREES
1865 if (FIB6_SUBTREE(pn) == fn) {
1866 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1867 RCU_INIT_POINTER(pn->subtree, NULL);
1868 nstate = FWS_L;
1869 } else {
1870 WARN_ON(fn->fn_flags & RTN_ROOT);
1871 #endif
1872 if (pn_r == fn)
1873 rcu_assign_pointer(pn->right, child);
1874 else if (pn_l == fn)
1875 rcu_assign_pointer(pn->left, child);
1876 #if RT6_DEBUG >= 2
1877 else
1878 WARN_ON(1);
1879 #endif
1880 if (child)
1881 rcu_assign_pointer(child->parent, pn);
1882 nstate = FWS_R;
1883 #ifdef CONFIG_IPV6_SUBTREES
1884 }
1885 #endif
1886
1887 read_lock(&net->ipv6.fib6_walker_lock);
1888 FOR_WALKERS(net, w) {
1889 if (!child) {
1890 if (w->node == fn) {
1891 pr_debug("W %p adjusted by delnode 1, s=%d/%d\n",
1892 w, w->state, nstate);
1893 w->node = pn;
1894 w->state = nstate;
1895 }
1896 } else {
1897 if (w->node == fn) {
1898 w->node = child;
1899 if (children&2) {
1900 pr_debug("W %p adjusted by delnode 2, s=%d\n",
1901 w, w->state);
1902 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1903 } else {
1904 pr_debug("W %p adjusted by delnode 2, s=%d\n",
1905 w, w->state);
1906 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1907 }
1908 }
1909 }
1910 }
1911 read_unlock(&net->ipv6.fib6_walker_lock);
1912
1913 node_free(net, fn);
1914 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1915 return pn;
1916
1917 RCU_INIT_POINTER(pn->leaf, NULL);
1918 fib6_info_release(pn_leaf);
1919 fn = pn;
1920 }
1921 }
1922
1923 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1924 struct fib6_info __rcu **rtp, struct nl_info *info)
1925 {
1926 struct fib6_info *leaf, *replace_rt = NULL;
1927 struct fib6_walker *w;
1928 struct fib6_info *rt = rcu_dereference_protected(*rtp,
1929 lockdep_is_held(&table->tb6_lock));
1930 struct net *net = info->nl_net;
1931 bool notify_del = false;
1932
1933 /* If the deleted route is the first in the node and it is not part of
1934 * a multipath route, then we need to replace it with the next route
1935 * in the node, if exists.
1936 */
1937 leaf = rcu_dereference_protected(fn->leaf,
1938 lockdep_is_held(&table->tb6_lock));
1939 if (leaf == rt && !rt->fib6_nsiblings) {
1940 if (rcu_access_pointer(rt->fib6_next))
1941 replace_rt = rcu_dereference_protected(rt->fib6_next,
1942 lockdep_is_held(&table->tb6_lock));
1943 else
1944 notify_del = true;
1945 }
1946
1947 /* Unlink it */
1948 *rtp = rt->fib6_next;
1949 rt->fib6_node = NULL;
1950 net->ipv6.rt6_stats->fib_rt_entries--;
1951 net->ipv6.rt6_stats->fib_discarded_routes++;
1952
1953 /* Reset round-robin state, if necessary */
1954 if (rcu_access_pointer(fn->rr_ptr) == rt)
1955 fn->rr_ptr = NULL;
1956
1957 /* Remove this entry from other siblings */
1958 if (rt->fib6_nsiblings) {
1959 struct fib6_info *sibling, *next_sibling;
1960
1961 /* The route is deleted from a multipath route. If this
1962 * multipath route is the first route in the node, then we need
1963 * to emit a delete notification. Otherwise, we need to skip
1964 * the notification.
1965 */
1966 if (rt->fib6_metric == leaf->fib6_metric &&
1967 rt6_qualify_for_ecmp(leaf))
1968 notify_del = true;
1969 list_for_each_entry_safe(sibling, next_sibling,
1970 &rt->fib6_siblings, fib6_siblings)
1971 sibling->fib6_nsiblings--;
1972 rt->fib6_nsiblings = 0;
1973 list_del_init(&rt->fib6_siblings);
1974 rt6_multipath_rebalance(next_sibling);
1975 }
1976
1977 /* Adjust walkers */
1978 read_lock(&net->ipv6.fib6_walker_lock);
1979 FOR_WALKERS(net, w) {
1980 if (w->state == FWS_C && w->leaf == rt) {
1981 pr_debug("walker %p adjusted by delroute\n", w);
1982 w->leaf = rcu_dereference_protected(rt->fib6_next,
1983 lockdep_is_held(&table->tb6_lock));
1984 if (!w->leaf)
1985 w->state = FWS_U;
1986 }
1987 }
1988 read_unlock(&net->ipv6.fib6_walker_lock);
1989
1990 /* If it was last route, call fib6_repair_tree() to:
1991 * 1. For root node, put back null_entry as how the table was created.
1992 * 2. For other nodes, expunge its radix tree node.
1993 */
1994 if (!rcu_access_pointer(fn->leaf)) {
1995 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1996 fn->fn_flags &= ~RTN_RTINFO;
1997 net->ipv6.rt6_stats->fib_route_nodes--;
1998 }
1999 fn = fib6_repair_tree(net, table, fn);
2000 }
2001
2002 fib6_purge_rt(rt, fn, net);
2003
2004 if (!info->skip_notify_kernel) {
2005 if (notify_del)
2006 call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
2007 rt, NULL);
2008 else if (replace_rt)
2009 call_fib6_entry_notifiers_replace(net, replace_rt);
2010 }
2011 if (!info->skip_notify)
2012 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
2013
2014 fib6_info_release(rt);
2015 }
2016
2017 /* Need to own table->tb6_lock */
2018 int fib6_del(struct fib6_info *rt, struct nl_info *info)
2019 {
2020 struct net *net = info->nl_net;
2021 struct fib6_info __rcu **rtp;
2022 struct fib6_info __rcu **rtp_next;
2023 struct fib6_table *table;
2024 struct fib6_node *fn;
2025
2026 if (rt == net->ipv6.fib6_null_entry)
2027 return -ENOENT;
2028
2029 table = rt->fib6_table;
2030 fn = rcu_dereference_protected(rt->fib6_node,
2031 lockdep_is_held(&table->tb6_lock));
2032 if (!fn)
2033 return -ENOENT;
2034
2035 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
2036
2037 /*
2038 * Walk the leaf entries looking for ourself
2039 */
2040
2041 for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
2042 struct fib6_info *cur = rcu_dereference_protected(*rtp,
2043 lockdep_is_held(&table->tb6_lock));
2044 if (rt == cur) {
2045 if (fib6_requires_src(cur))
2046 fib6_routes_require_src_dec(info->nl_net);
2047 fib6_del_route(table, fn, rtp, info);
2048 return 0;
2049 }
2050 rtp_next = &cur->fib6_next;
2051 }
2052 return -ENOENT;
2053 }
2054
2055 /*
2056 * Tree traversal function.
2057 *
2058 * Certainly, it is not interrupt safe.
2059 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
2060 * It means, that we can modify tree during walking
2061 * and use this function for garbage collection, clone pruning,
2062 * cleaning tree when a device goes down etc. etc.
2063 *
2064 * It guarantees that every node will be traversed,
2065 * and that it will be traversed only once.
2066 *
2067 * Callback function w->func may return:
2068 * 0 -> continue walking.
2069 * positive value -> walking is suspended (used by tree dumps,
2070 * and probably by gc, if it will be split to several slices)
2071 * negative value -> terminate walking.
2072 *
2073 * The function itself returns:
2074 * 0 -> walk is complete.
2075 * >0 -> walk is incomplete (i.e. suspended)
2076 * <0 -> walk is terminated by an error.
2077 *
2078 * This function is called with tb6_lock held.
2079 */
2080
2081 static int fib6_walk_continue(struct fib6_walker *w)
2082 {
2083 struct fib6_node *fn, *pn, *left, *right;
2084
2085 /* w->root should always be table->tb6_root */
2086 WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
2087
2088 for (;;) {
2089 fn = w->node;
2090 if (!fn)
2091 return 0;
2092
2093 switch (w->state) {
2094 #ifdef CONFIG_IPV6_SUBTREES
2095 case FWS_S:
2096 if (FIB6_SUBTREE(fn)) {
2097 w->node = FIB6_SUBTREE(fn);
2098 continue;
2099 }
2100 w->state = FWS_L;
2101 fallthrough;
2102 #endif
2103 case FWS_L:
2104 left = rcu_dereference_protected(fn->left, 1);
2105 if (left) {
2106 w->node = left;
2107 w->state = FWS_INIT;
2108 continue;
2109 }
2110 w->state = FWS_R;
2111 fallthrough;
2112 case FWS_R:
2113 right = rcu_dereference_protected(fn->right, 1);
2114 if (right) {
2115 w->node = right;
2116 w->state = FWS_INIT;
2117 continue;
2118 }
2119 w->state = FWS_C;
2120 w->leaf = rcu_dereference_protected(fn->leaf, 1);
2121 fallthrough;
2122 case FWS_C:
2123 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
2124 int err;
2125
2126 if (w->skip) {
2127 w->skip--;
2128 goto skip;
2129 }
2130
2131 err = w->func(w);
2132 if (err)
2133 return err;
2134
2135 w->count++;
2136 continue;
2137 }
2138 skip:
2139 w->state = FWS_U;
2140 fallthrough;
2141 case FWS_U:
2142 if (fn == w->root)
2143 return 0;
2144 pn = rcu_dereference_protected(fn->parent, 1);
2145 left = rcu_dereference_protected(pn->left, 1);
2146 right = rcu_dereference_protected(pn->right, 1);
2147 w->node = pn;
2148 #ifdef CONFIG_IPV6_SUBTREES
2149 if (FIB6_SUBTREE(pn) == fn) {
2150 WARN_ON(!(fn->fn_flags & RTN_ROOT));
2151 w->state = FWS_L;
2152 continue;
2153 }
2154 #endif
2155 if (left == fn) {
2156 w->state = FWS_R;
2157 continue;
2158 }
2159 if (right == fn) {
2160 w->state = FWS_C;
2161 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
2162 continue;
2163 }
2164 #if RT6_DEBUG >= 2
2165 WARN_ON(1);
2166 #endif
2167 }
2168 }
2169 }
2170
2171 static int fib6_walk(struct net *net, struct fib6_walker *w)
2172 {
2173 int res;
2174
2175 w->state = FWS_INIT;
2176 w->node = w->root;
2177
2178 fib6_walker_link(net, w);
2179 res = fib6_walk_continue(w);
2180 if (res <= 0)
2181 fib6_walker_unlink(net, w);
2182 return res;
2183 }
2184
2185 static int fib6_clean_node(struct fib6_walker *w)
2186 {
2187 int res;
2188 struct fib6_info *rt;
2189 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
2190 struct nl_info info = {
2191 .nl_net = c->net,
2192 .skip_notify = c->skip_notify,
2193 };
2194
2195 if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
2196 READ_ONCE(w->node->fn_sernum) != c->sernum)
2197 WRITE_ONCE(w->node->fn_sernum, c->sernum);
2198
2199 if (!c->func) {
2200 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
2201 w->leaf = NULL;
2202 return 0;
2203 }
2204
2205 for_each_fib6_walker_rt(w) {
2206 res = c->func(rt, c->arg);
2207 if (res == -1) {
2208 w->leaf = rt;
2209 res = fib6_del(rt, &info);
2210 if (res) {
2211 #if RT6_DEBUG >= 2
2212 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2213 __func__, rt,
2214 rcu_access_pointer(rt->fib6_node),
2215 res);
2216 #endif
2217 continue;
2218 }
2219 return 0;
2220 } else if (res == -2) {
2221 if (WARN_ON(!rt->fib6_nsiblings))
2222 continue;
2223 rt = list_last_entry(&rt->fib6_siblings,
2224 struct fib6_info, fib6_siblings);
2225 continue;
2226 }
2227 WARN_ON(res != 0);
2228 }
2229 w->leaf = rt;
2230 return 0;
2231 }
2232
2233 /*
2234 * Convenient frontend to tree walker.
2235 *
2236 * func is called on each route.
2237 * It may return -2 -> skip multipath route.
2238 * -1 -> delete this route.
2239 * 0 -> continue walking
2240 */
2241
2242 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2243 int (*func)(struct fib6_info *, void *arg),
2244 int sernum, void *arg, bool skip_notify)
2245 {
2246 struct fib6_cleaner c;
2247
2248 c.w.root = root;
2249 c.w.func = fib6_clean_node;
2250 c.w.count = 0;
2251 c.w.skip = 0;
2252 c.w.skip_in_node = 0;
2253 c.func = func;
2254 c.sernum = sernum;
2255 c.arg = arg;
2256 c.net = net;
2257 c.skip_notify = skip_notify;
2258
2259 fib6_walk(net, &c.w);
2260 }
2261
2262 static void __fib6_clean_all(struct net *net,
2263 int (*func)(struct fib6_info *, void *),
2264 int sernum, void *arg, bool skip_notify)
2265 {
2266 struct fib6_table *table;
2267 struct hlist_head *head;
2268 unsigned int h;
2269
2270 rcu_read_lock();
2271 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2272 head = &net->ipv6.fib_table_hash[h];
2273 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2274 spin_lock_bh(&table->tb6_lock);
2275 fib6_clean_tree(net, &table->tb6_root,
2276 func, sernum, arg, skip_notify);
2277 spin_unlock_bh(&table->tb6_lock);
2278 }
2279 }
2280 rcu_read_unlock();
2281 }
2282
2283 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2284 void *arg)
2285 {
2286 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2287 }
2288
2289 void fib6_clean_all_skip_notify(struct net *net,
2290 int (*func)(struct fib6_info *, void *),
2291 void *arg)
2292 {
2293 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2294 }
2295
2296 static void fib6_flush_trees(struct net *net)
2297 {
2298 int new_sernum = fib6_new_sernum(net);
2299
2300 __fib6_clean_all(net, NULL, new_sernum, NULL, false);
2301 }
2302
2303 /*
2304 * Garbage collection
2305 */
2306
2307 static int fib6_age(struct fib6_info *rt, struct fib6_gc_args *gc_args)
2308 {
2309 unsigned long now = jiffies;
2310
2311 /*
2312 * check addrconf expiration here.
2313 * Routes are expired even if they are in use.
2314 */
2315
2316 if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2317 if (time_after(now, rt->expires)) {
2318 pr_debug("expiring %p\n", rt);
2319 return -1;
2320 }
2321 gc_args->more++;
2322 }
2323
2324 /* Also age clones in the exception table.
2325 * Note, that clones are aged out
2326 * only if they are not in use now.
2327 */
2328 rt6_age_exceptions(rt, gc_args, now);
2329
2330 return 0;
2331 }
2332
2333 static void fib6_gc_table(struct net *net,
2334 struct fib6_table *tb6,
2335 struct fib6_gc_args *gc_args)
2336 {
2337 struct fib6_info *rt;
2338 struct hlist_node *n;
2339 struct nl_info info = {
2340 .nl_net = net,
2341 .skip_notify = false,
2342 };
2343
2344 hlist_for_each_entry_safe(rt, n, &tb6->tb6_gc_hlist, gc_link)
2345 if (fib6_age(rt, gc_args) == -1)
2346 fib6_del(rt, &info);
2347 }
2348
2349 static void fib6_gc_all(struct net *net, struct fib6_gc_args *gc_args)
2350 {
2351 struct fib6_table *table;
2352 struct hlist_head *head;
2353 unsigned int h;
2354
2355 rcu_read_lock();
2356 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2357 head = &net->ipv6.fib_table_hash[h];
2358 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2359 spin_lock_bh(&table->tb6_lock);
2360
2361 fib6_gc_table(net, table, gc_args);
2362
2363 spin_unlock_bh(&table->tb6_lock);
2364 }
2365 }
2366 rcu_read_unlock();
2367 }
2368
2369 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2370 {
2371 struct fib6_gc_args gc_args;
2372 unsigned long now;
2373
2374 if (force) {
2375 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2376 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2377 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2378 return;
2379 }
2380 gc_args.timeout = expires ? (int)expires :
2381 net->ipv6.sysctl.ip6_rt_gc_interval;
2382 gc_args.more = 0;
2383
2384 fib6_gc_all(net, &gc_args);
2385 now = jiffies;
2386 net->ipv6.ip6_rt_last_gc = now;
2387
2388 if (gc_args.more)
2389 mod_timer(&net->ipv6.ip6_fib_timer,
2390 round_jiffies(now
2391 + net->ipv6.sysctl.ip6_rt_gc_interval));
2392 else
2393 del_timer(&net->ipv6.ip6_fib_timer);
2394 spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2395 }
2396
2397 static void fib6_gc_timer_cb(struct timer_list *t)
2398 {
2399 struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2400
2401 fib6_run_gc(0, arg, true);
2402 }
2403
2404 static int __net_init fib6_net_init(struct net *net)
2405 {
2406 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2407 int err;
2408
2409 err = fib6_notifier_init(net);
2410 if (err)
2411 return err;
2412
2413 /* Default to 3-tuple */
2414 net->ipv6.sysctl.multipath_hash_fields =
2415 FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK;
2416
2417 spin_lock_init(&net->ipv6.fib6_gc_lock);
2418 rwlock_init(&net->ipv6.fib6_walker_lock);
2419 INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2420 timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2421
2422 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2423 if (!net->ipv6.rt6_stats)
2424 goto out_notifier;
2425
2426 /* Avoid false sharing : Use at least a full cache line */
2427 size = max_t(size_t, size, L1_CACHE_BYTES);
2428
2429 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2430 if (!net->ipv6.fib_table_hash)
2431 goto out_rt6_stats;
2432
2433 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2434 GFP_KERNEL);
2435 if (!net->ipv6.fib6_main_tbl)
2436 goto out_fib_table_hash;
2437
2438 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2439 rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2440 net->ipv6.fib6_null_entry);
2441 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2442 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2443 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2444 INIT_HLIST_HEAD(&net->ipv6.fib6_main_tbl->tb6_gc_hlist);
2445
2446 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2447 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2448 GFP_KERNEL);
2449 if (!net->ipv6.fib6_local_tbl)
2450 goto out_fib6_main_tbl;
2451 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2452 rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2453 net->ipv6.fib6_null_entry);
2454 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2455 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2456 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2457 INIT_HLIST_HEAD(&net->ipv6.fib6_local_tbl->tb6_gc_hlist);
2458 #endif
2459 fib6_tables_init(net);
2460
2461 return 0;
2462
2463 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2464 out_fib6_main_tbl:
2465 kfree(net->ipv6.fib6_main_tbl);
2466 #endif
2467 out_fib_table_hash:
2468 kfree(net->ipv6.fib_table_hash);
2469 out_rt6_stats:
2470 kfree(net->ipv6.rt6_stats);
2471 out_notifier:
2472 fib6_notifier_exit(net);
2473 return -ENOMEM;
2474 }
2475
2476 static void fib6_net_exit(struct net *net)
2477 {
2478 unsigned int i;
2479
2480 del_timer_sync(&net->ipv6.ip6_fib_timer);
2481
2482 for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2483 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2484 struct hlist_node *tmp;
2485 struct fib6_table *tb;
2486
2487 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2488 hlist_del(&tb->tb6_hlist);
2489 fib6_free_table(tb);
2490 }
2491 }
2492
2493 kfree(net->ipv6.fib_table_hash);
2494 kfree(net->ipv6.rt6_stats);
2495 fib6_notifier_exit(net);
2496 }
2497
2498 static struct pernet_operations fib6_net_ops = {
2499 .init = fib6_net_init,
2500 .exit = fib6_net_exit,
2501 };
2502
2503 int __init fib6_init(void)
2504 {
2505 int ret = -ENOMEM;
2506
2507 fib6_node_kmem = KMEM_CACHE(fib6_node,
2508 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT);
2509 if (!fib6_node_kmem)
2510 goto out;
2511
2512 ret = register_pernet_subsys(&fib6_net_ops);
2513 if (ret)
2514 goto out_kmem_cache_create;
2515
2516 ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2517 inet6_dump_fib, RTNL_FLAG_DUMP_UNLOCKED |
2518 RTNL_FLAG_DUMP_SPLIT_NLM_DONE);
2519 if (ret)
2520 goto out_unregister_subsys;
2521
2522 __fib6_flush_trees = fib6_flush_trees;
2523 out:
2524 return ret;
2525
2526 out_unregister_subsys:
2527 unregister_pernet_subsys(&fib6_net_ops);
2528 out_kmem_cache_create:
2529 kmem_cache_destroy(fib6_node_kmem);
2530 goto out;
2531 }
2532
2533 void fib6_gc_cleanup(void)
2534 {
2535 unregister_pernet_subsys(&fib6_net_ops);
2536 kmem_cache_destroy(fib6_node_kmem);
2537 }
2538
2539 #ifdef CONFIG_PROC_FS
2540 static int ipv6_route_native_seq_show(struct seq_file *seq, void *v)
2541 {
2542 struct fib6_info *rt = v;
2543 struct ipv6_route_iter *iter = seq->private;
2544 struct fib6_nh *fib6_nh = rt->fib6_nh;
2545 unsigned int flags = rt->fib6_flags;
2546 const struct net_device *dev;
2547
2548 if (rt->nh)
2549 fib6_nh = nexthop_fib6_nh(rt->nh);
2550
2551 seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2552
2553 #ifdef CONFIG_IPV6_SUBTREES
2554 seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2555 #else
2556 seq_puts(seq, "00000000000000000000000000000000 00 ");
2557 #endif
2558 if (fib6_nh->fib_nh_gw_family) {
2559 flags |= RTF_GATEWAY;
2560 seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6);
2561 } else {
2562 seq_puts(seq, "00000000000000000000000000000000");
2563 }
2564
2565 dev = fib6_nh->fib_nh_dev;
2566 seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2567 rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2568 flags, dev ? dev->name : "");
2569 iter->w.leaf = NULL;
2570 return 0;
2571 }
2572
2573 static int ipv6_route_yield(struct fib6_walker *w)
2574 {
2575 struct ipv6_route_iter *iter = w->args;
2576
2577 if (!iter->skip)
2578 return 1;
2579
2580 do {
2581 iter->w.leaf = rcu_dereference_protected(
2582 iter->w.leaf->fib6_next,
2583 lockdep_is_held(&iter->tbl->tb6_lock));
2584 iter->skip--;
2585 if (!iter->skip && iter->w.leaf)
2586 return 1;
2587 } while (iter->w.leaf);
2588
2589 return 0;
2590 }
2591
2592 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2593 struct net *net)
2594 {
2595 memset(&iter->w, 0, sizeof(iter->w));
2596 iter->w.func = ipv6_route_yield;
2597 iter->w.root = &iter->tbl->tb6_root;
2598 iter->w.state = FWS_INIT;
2599 iter->w.node = iter->w.root;
2600 iter->w.args = iter;
2601 iter->sernum = READ_ONCE(iter->w.root->fn_sernum);
2602 INIT_LIST_HEAD(&iter->w.lh);
2603 fib6_walker_link(net, &iter->w);
2604 }
2605
2606 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2607 struct net *net)
2608 {
2609 unsigned int h;
2610 struct hlist_node *node;
2611
2612 if (tbl) {
2613 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2614 node = rcu_dereference(hlist_next_rcu(&tbl->tb6_hlist));
2615 } else {
2616 h = 0;
2617 node = NULL;
2618 }
2619
2620 while (!node && h < FIB6_TABLE_HASHSZ) {
2621 node = rcu_dereference(
2622 hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2623 }
2624 return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2625 }
2626
2627 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2628 {
2629 int sernum = READ_ONCE(iter->w.root->fn_sernum);
2630
2631 if (iter->sernum != sernum) {
2632 iter->sernum = sernum;
2633 iter->w.state = FWS_INIT;
2634 iter->w.node = iter->w.root;
2635 WARN_ON(iter->w.skip);
2636 iter->w.skip = iter->w.count;
2637 }
2638 }
2639
2640 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2641 {
2642 int r;
2643 struct fib6_info *n;
2644 struct net *net = seq_file_net(seq);
2645 struct ipv6_route_iter *iter = seq->private;
2646
2647 ++(*pos);
2648 if (!v)
2649 goto iter_table;
2650
2651 n = rcu_dereference(((struct fib6_info *)v)->fib6_next);
2652 if (n)
2653 return n;
2654
2655 iter_table:
2656 ipv6_route_check_sernum(iter);
2657 spin_lock_bh(&iter->tbl->tb6_lock);
2658 r = fib6_walk_continue(&iter->w);
2659 spin_unlock_bh(&iter->tbl->tb6_lock);
2660 if (r > 0) {
2661 return iter->w.leaf;
2662 } else if (r < 0) {
2663 fib6_walker_unlink(net, &iter->w);
2664 return NULL;
2665 }
2666 fib6_walker_unlink(net, &iter->w);
2667
2668 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2669 if (!iter->tbl)
2670 return NULL;
2671
2672 ipv6_route_seq_setup_walk(iter, net);
2673 goto iter_table;
2674 }
2675
2676 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2677 __acquires(RCU)
2678 {
2679 struct net *net = seq_file_net(seq);
2680 struct ipv6_route_iter *iter = seq->private;
2681
2682 rcu_read_lock();
2683 iter->tbl = ipv6_route_seq_next_table(NULL, net);
2684 iter->skip = *pos;
2685
2686 if (iter->tbl) {
2687 loff_t p = 0;
2688
2689 ipv6_route_seq_setup_walk(iter, net);
2690 return ipv6_route_seq_next(seq, NULL, &p);
2691 } else {
2692 return NULL;
2693 }
2694 }
2695
2696 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2697 {
2698 struct fib6_walker *w = &iter->w;
2699 return w->node && !(w->state == FWS_U && w->node == w->root);
2700 }
2701
2702 static void ipv6_route_native_seq_stop(struct seq_file *seq, void *v)
2703 __releases(RCU)
2704 {
2705 struct net *net = seq_file_net(seq);
2706 struct ipv6_route_iter *iter = seq->private;
2707
2708 if (ipv6_route_iter_active(iter))
2709 fib6_walker_unlink(net, &iter->w);
2710
2711 rcu_read_unlock();
2712 }
2713
2714 #if IS_BUILTIN(CONFIG_IPV6) && defined(CONFIG_BPF_SYSCALL)
2715 static int ipv6_route_prog_seq_show(struct bpf_prog *prog,
2716 struct bpf_iter_meta *meta,
2717 void *v)
2718 {
2719 struct bpf_iter__ipv6_route ctx;
2720
2721 ctx.meta = meta;
2722 ctx.rt = v;
2723 return bpf_iter_run_prog(prog, &ctx);
2724 }
2725
2726 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2727 {
2728 struct ipv6_route_iter *iter = seq->private;
2729 struct bpf_iter_meta meta;
2730 struct bpf_prog *prog;
2731 int ret;
2732
2733 meta.seq = seq;
2734 prog = bpf_iter_get_info(&meta, false);
2735 if (!prog)
2736 return ipv6_route_native_seq_show(seq, v);
2737
2738 ret = ipv6_route_prog_seq_show(prog, &meta, v);
2739 iter->w.leaf = NULL;
2740
2741 return ret;
2742 }
2743
2744 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2745 {
2746 struct bpf_iter_meta meta;
2747 struct bpf_prog *prog;
2748
2749 if (!v) {
2750 meta.seq = seq;
2751 prog = bpf_iter_get_info(&meta, true);
2752 if (prog)
2753 (void)ipv6_route_prog_seq_show(prog, &meta, v);
2754 }
2755
2756 ipv6_route_native_seq_stop(seq, v);
2757 }
2758 #else
2759 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2760 {
2761 return ipv6_route_native_seq_show(seq, v);
2762 }
2763
2764 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2765 {
2766 ipv6_route_native_seq_stop(seq, v);
2767 }
2768 #endif
2769
2770 const struct seq_operations ipv6_route_seq_ops = {
2771 .start = ipv6_route_seq_start,
2772 .next = ipv6_route_seq_next,
2773 .stop = ipv6_route_seq_stop,
2774 .show = ipv6_route_seq_show
2775 };
2776 #endif /* CONFIG_PROC_FS */
2777
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.