Loading...
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
40static struct kmem_cache *fib6_node_kmem __read_mostly;
41
42struct 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
57static struct fib6_info *fib6_find_prefix(struct net *net,
58 struct fib6_table *table,
59 struct fib6_node *fn);
60static struct fib6_node *fib6_repair_tree(struct net *net,
61 struct fib6_table *table,
62 struct fib6_node *fn);
63static int fib6_walk(struct net *net, struct fib6_walker *w);
64static 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
73static 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
78static 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
85static 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
92static 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
103enum {
104 FIB6_NO_SERNUM_CHANGE = 0,
105};
106
107void 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
133static __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
147struct 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
168void 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}
182EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
183
184static 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
195static 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
201static 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
208static 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
214static void fib6_free_table(struct fib6_table *table)
215{
216 inetpeer_invalidate_tree(&table->tb6_peers);
217 kfree(table);
218}
219
220static 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
240static 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
257struct 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}
273EXPORT_SYMBOL_GPL(fib6_new_table);
274
275struct 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}
296EXPORT_SYMBOL_GPL(fib6_get_table);
297
298static 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
305struct fib6_table *fib6_new_table(struct net *net, u32 id)
306{
307 return fib6_get_table(net, id);
308}
309
310struct fib6_table *fib6_get_table(struct net *net, u32 id)
311{
312 return net->ipv6.fib6_main_tbl;
313}
314
315struct 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 */
334int 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
341static 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
348unsigned 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
365static 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
378static 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
393int 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
407int 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
423int 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
434struct fib6_dump_arg {
435 struct net *net;
436 struct notifier_block *nb;
437 struct netlink_ext_ack *extack;
438};
439
440static 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
460static 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
469static 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() */
482int 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
511out:
512 kfree(w);
513
514 /* The tree traversal function should never return a positive value. */
515 return err > 0 ? -EINVAL : err;
516}
517
518static 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
552static 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
569static int fib6_dump_done(struct netlink_callback *cb)
570{
571 fib6_dump_end(cb);
572 return cb->done ? cb->done(cb) : 0;
573}
574
575static 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
621static 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 = true,
627 };
628 const struct nlmsghdr *nlh = cb->nlh;
629 struct net *net = sock_net(skb->sk);
630 unsigned int h, s_h;
631 unsigned int e = 0, s_e;
632 struct fib6_walker *w;
633 struct fib6_table *tb;
634 struct hlist_head *head;
635 int res = 0;
636
637 if (cb->strict_check) {
638 int err;
639
640 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
641 if (err < 0)
642 return err;
643 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
644 struct rtmsg *rtm = nlmsg_data(nlh);
645
646 if (rtm->rtm_flags & RTM_F_PREFIX)
647 arg.filter.flags = RTM_F_PREFIX;
648 }
649
650 w = (void *)cb->args[2];
651 if (!w) {
652 /* New dump:
653 *
654 * 1. allocate and initialize walker.
655 */
656 w = kzalloc(sizeof(*w), GFP_ATOMIC);
657 if (!w)
658 return -ENOMEM;
659 w->func = fib6_dump_node;
660 cb->args[2] = (long)w;
661
662 /* 2. hook callback destructor.
663 */
664 cb->args[3] = (long)cb->done;
665 cb->done = fib6_dump_done;
666
667 }
668
669 arg.skb = skb;
670 arg.cb = cb;
671 arg.net = net;
672 w->args = &arg;
673
674 if (arg.filter.table_id) {
675 tb = fib6_get_table(net, arg.filter.table_id);
676 if (!tb) {
677 if (rtnl_msg_family(cb->nlh) != PF_INET6)
678 goto out;
679
680 NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
681 return -ENOENT;
682 }
683
684 if (!cb->args[0]) {
685 res = fib6_dump_table(tb, skb, cb);
686 if (!res)
687 cb->args[0] = 1;
688 }
689 goto out;
690 }
691
692 s_h = cb->args[0];
693 s_e = cb->args[1];
694
695 rcu_read_lock();
696 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
697 e = 0;
698 head = &net->ipv6.fib_table_hash[h];
699 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
700 if (e < s_e)
701 goto next;
702 res = fib6_dump_table(tb, skb, cb);
703 if (res != 0)
704 goto out_unlock;
705next:
706 e++;
707 }
708 }
709out_unlock:
710 rcu_read_unlock();
711 cb->args[1] = e;
712 cb->args[0] = h;
713out:
714 res = res < 0 ? res : skb->len;
715 if (res <= 0)
716 fib6_dump_end(cb);
717 return res;
718}
719
720void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
721{
722 if (!f6i)
723 return;
724
725 if (f6i->fib6_metrics == &dst_default_metrics) {
726 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
727
728 if (!p)
729 return;
730
731 refcount_set(&p->refcnt, 1);
732 f6i->fib6_metrics = p;
733 }
734
735 f6i->fib6_metrics->metrics[metric - 1] = val;
736}
737
738/*
739 * Routing Table
740 *
741 * return the appropriate node for a routing tree "add" operation
742 * by either creating and inserting or by returning an existing
743 * node.
744 */
745
746static struct fib6_node *fib6_add_1(struct net *net,
747 struct fib6_table *table,
748 struct fib6_node *root,
749 struct in6_addr *addr, int plen,
750 int offset, int allow_create,
751 int replace_required,
752 struct netlink_ext_ack *extack)
753{
754 struct fib6_node *fn, *in, *ln;
755 struct fib6_node *pn = NULL;
756 struct rt6key *key;
757 int bit;
758 __be32 dir = 0;
759
760 /* insert node in tree */
761
762 fn = root;
763
764 do {
765 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
766 lockdep_is_held(&table->tb6_lock));
767 key = (struct rt6key *)((u8 *)leaf + offset);
768
769 /*
770 * Prefix match
771 */
772 if (plen < fn->fn_bit ||
773 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
774 if (!allow_create) {
775 if (replace_required) {
776 NL_SET_ERR_MSG(extack,
777 "Can not replace route - no match found");
778 pr_warn("Can't replace route, no match found\n");
779 return ERR_PTR(-ENOENT);
780 }
781 pr_warn("NLM_F_CREATE should be set when creating new route\n");
782 }
783 goto insert_above;
784 }
785
786 /*
787 * Exact match ?
788 */
789
790 if (plen == fn->fn_bit) {
791 /* clean up an intermediate node */
792 if (!(fn->fn_flags & RTN_RTINFO)) {
793 RCU_INIT_POINTER(fn->leaf, NULL);
794 fib6_info_release(leaf);
795 /* remove null_entry in the root node */
796 } else if (fn->fn_flags & RTN_TL_ROOT &&
797 rcu_access_pointer(fn->leaf) ==
798 net->ipv6.fib6_null_entry) {
799 RCU_INIT_POINTER(fn->leaf, NULL);
800 }
801
802 return fn;
803 }
804
805 /*
806 * We have more bits to go
807 */
808
809 /* Try to walk down on tree. */
810 dir = addr_bit_set(addr, fn->fn_bit);
811 pn = fn;
812 fn = dir ?
813 rcu_dereference_protected(fn->right,
814 lockdep_is_held(&table->tb6_lock)) :
815 rcu_dereference_protected(fn->left,
816 lockdep_is_held(&table->tb6_lock));
817 } while (fn);
818
819 if (!allow_create) {
820 /* We should not create new node because
821 * NLM_F_REPLACE was specified without NLM_F_CREATE
822 * I assume it is safe to require NLM_F_CREATE when
823 * REPLACE flag is used! Later we may want to remove the
824 * check for replace_required, because according
825 * to netlink specification, NLM_F_CREATE
826 * MUST be specified if new route is created.
827 * That would keep IPv6 consistent with IPv4
828 */
829 if (replace_required) {
830 NL_SET_ERR_MSG(extack,
831 "Can not replace route - no match found");
832 pr_warn("Can't replace route, no match found\n");
833 return ERR_PTR(-ENOENT);
834 }
835 pr_warn("NLM_F_CREATE should be set when creating new route\n");
836 }
837 /*
838 * We walked to the bottom of tree.
839 * Create new leaf node without children.
840 */
841
842 ln = node_alloc(net);
843
844 if (!ln)
845 return ERR_PTR(-ENOMEM);
846 ln->fn_bit = plen;
847 RCU_INIT_POINTER(ln->parent, pn);
848
849 if (dir)
850 rcu_assign_pointer(pn->right, ln);
851 else
852 rcu_assign_pointer(pn->left, ln);
853
854 return ln;
855
856
857insert_above:
858 /*
859 * split since we don't have a common prefix anymore or
860 * we have a less significant route.
861 * we've to insert an intermediate node on the list
862 * this new node will point to the one we need to create
863 * and the current
864 */
865
866 pn = rcu_dereference_protected(fn->parent,
867 lockdep_is_held(&table->tb6_lock));
868
869 /* find 1st bit in difference between the 2 addrs.
870
871 See comment in __ipv6_addr_diff: bit may be an invalid value,
872 but if it is >= plen, the value is ignored in any case.
873 */
874
875 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
876
877 /*
878 * (intermediate)[in]
879 * / \
880 * (new leaf node)[ln] (old node)[fn]
881 */
882 if (plen > bit) {
883 in = node_alloc(net);
884 ln = node_alloc(net);
885
886 if (!in || !ln) {
887 if (in)
888 node_free_immediate(net, in);
889 if (ln)
890 node_free_immediate(net, ln);
891 return ERR_PTR(-ENOMEM);
892 }
893
894 /*
895 * new intermediate node.
896 * RTN_RTINFO will
897 * be off since that an address that chooses one of
898 * the branches would not match less specific routes
899 * in the other branch
900 */
901
902 in->fn_bit = bit;
903
904 RCU_INIT_POINTER(in->parent, pn);
905 in->leaf = fn->leaf;
906 fib6_info_hold(rcu_dereference_protected(in->leaf,
907 lockdep_is_held(&table->tb6_lock)));
908
909 /* update parent pointer */
910 if (dir)
911 rcu_assign_pointer(pn->right, in);
912 else
913 rcu_assign_pointer(pn->left, in);
914
915 ln->fn_bit = plen;
916
917 RCU_INIT_POINTER(ln->parent, in);
918 rcu_assign_pointer(fn->parent, in);
919
920 if (addr_bit_set(addr, bit)) {
921 rcu_assign_pointer(in->right, ln);
922 rcu_assign_pointer(in->left, fn);
923 } else {
924 rcu_assign_pointer(in->left, ln);
925 rcu_assign_pointer(in->right, fn);
926 }
927 } else { /* plen <= bit */
928
929 /*
930 * (new leaf node)[ln]
931 * / \
932 * (old node)[fn] NULL
933 */
934
935 ln = node_alloc(net);
936
937 if (!ln)
938 return ERR_PTR(-ENOMEM);
939
940 ln->fn_bit = plen;
941
942 RCU_INIT_POINTER(ln->parent, pn);
943
944 if (addr_bit_set(&key->addr, plen))
945 RCU_INIT_POINTER(ln->right, fn);
946 else
947 RCU_INIT_POINTER(ln->left, fn);
948
949 rcu_assign_pointer(fn->parent, ln);
950
951 if (dir)
952 rcu_assign_pointer(pn->right, ln);
953 else
954 rcu_assign_pointer(pn->left, ln);
955 }
956 return ln;
957}
958
959static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
960 const struct fib6_info *match,
961 const struct fib6_table *table)
962{
963 int cpu;
964
965 if (!fib6_nh->rt6i_pcpu)
966 return;
967
968 /* release the reference to this fib entry from
969 * all of its cached pcpu routes
970 */
971 for_each_possible_cpu(cpu) {
972 struct rt6_info **ppcpu_rt;
973 struct rt6_info *pcpu_rt;
974
975 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
976 pcpu_rt = *ppcpu_rt;
977
978 /* only dropping the 'from' reference if the cached route
979 * is using 'match'. The cached pcpu_rt->from only changes
980 * from a fib6_info to NULL (ip6_dst_destroy); it can never
981 * change from one fib6_info reference to another
982 */
983 if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
984 struct fib6_info *from;
985
986 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
987 fib6_info_release(from);
988 }
989 }
990}
991
992struct fib6_nh_pcpu_arg {
993 struct fib6_info *from;
994 const struct fib6_table *table;
995};
996
997static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg)
998{
999 struct fib6_nh_pcpu_arg *arg = _arg;
1000
1001 __fib6_drop_pcpu_from(nh, arg->from, arg->table);
1002 return 0;
1003}
1004
1005static void fib6_drop_pcpu_from(struct fib6_info *f6i,
1006 const struct fib6_table *table)
1007{
1008 /* Make sure rt6_make_pcpu_route() wont add other percpu routes
1009 * while we are cleaning them here.
1010 */
1011 f6i->fib6_destroying = 1;
1012 mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
1013
1014 if (f6i->nh) {
1015 struct fib6_nh_pcpu_arg arg = {
1016 .from = f6i,
1017 .table = table
1018 };
1019
1020 nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from,
1021 &arg);
1022 } else {
1023 struct fib6_nh *fib6_nh;
1024
1025 fib6_nh = f6i->fib6_nh;
1026 __fib6_drop_pcpu_from(fib6_nh, f6i, table);
1027 }
1028}
1029
1030static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
1031 struct net *net)
1032{
1033 struct fib6_table *table = rt->fib6_table;
1034
1035 /* Flush all cached dst in exception table */
1036 rt6_flush_exceptions(rt);
1037 fib6_drop_pcpu_from(rt, table);
1038
1039 if (rt->nh && !list_empty(&rt->nh_list))
1040 list_del_init(&rt->nh_list);
1041
1042 if (refcount_read(&rt->fib6_ref) != 1) {
1043 /* This route is used as dummy address holder in some split
1044 * nodes. It is not leaked, but it still holds other resources,
1045 * which must be released in time. So, scan ascendant nodes
1046 * and replace dummy references to this route with references
1047 * to still alive ones.
1048 */
1049 while (fn) {
1050 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1051 lockdep_is_held(&table->tb6_lock));
1052 struct fib6_info *new_leaf;
1053 if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
1054 new_leaf = fib6_find_prefix(net, table, fn);
1055 fib6_info_hold(new_leaf);
1056
1057 rcu_assign_pointer(fn->leaf, new_leaf);
1058 fib6_info_release(rt);
1059 }
1060 fn = rcu_dereference_protected(fn->parent,
1061 lockdep_is_held(&table->tb6_lock));
1062 }
1063 }
1064
1065 fib6_clean_expires(rt);
1066 fib6_remove_gc_list(rt);
1067}
1068
1069/*
1070 * Insert routing information in a node.
1071 */
1072
1073static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
1074 struct nl_info *info,
1075 struct netlink_ext_ack *extack)
1076{
1077 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1078 lockdep_is_held(&rt->fib6_table->tb6_lock));
1079 struct fib6_info *iter = NULL;
1080 struct fib6_info __rcu **ins;
1081 struct fib6_info __rcu **fallback_ins = NULL;
1082 int replace = (info->nlh &&
1083 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
1084 int add = (!info->nlh ||
1085 (info->nlh->nlmsg_flags & NLM_F_CREATE));
1086 int found = 0;
1087 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
1088 bool notify_sibling_rt = false;
1089 u16 nlflags = NLM_F_EXCL;
1090 int err;
1091
1092 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
1093 nlflags |= NLM_F_APPEND;
1094
1095 ins = &fn->leaf;
1096
1097 for (iter = leaf; iter;
1098 iter = rcu_dereference_protected(iter->fib6_next,
1099 lockdep_is_held(&rt->fib6_table->tb6_lock))) {
1100 /*
1101 * Search for duplicates
1102 */
1103
1104 if (iter->fib6_metric == rt->fib6_metric) {
1105 /*
1106 * Same priority level
1107 */
1108 if (info->nlh &&
1109 (info->nlh->nlmsg_flags & NLM_F_EXCL))
1110 return -EEXIST;
1111
1112 nlflags &= ~NLM_F_EXCL;
1113 if (replace) {
1114 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1115 found++;
1116 break;
1117 }
1118 fallback_ins = fallback_ins ?: ins;
1119 goto next_iter;
1120 }
1121
1122 if (rt6_duplicate_nexthop(iter, rt)) {
1123 if (rt->fib6_nsiblings)
1124 rt->fib6_nsiblings = 0;
1125 if (!(iter->fib6_flags & RTF_EXPIRES))
1126 return -EEXIST;
1127 if (!(rt->fib6_flags & RTF_EXPIRES)) {
1128 fib6_clean_expires(iter);
1129 fib6_remove_gc_list(iter);
1130 } else {
1131 fib6_set_expires(iter, rt->expires);
1132 fib6_add_gc_list(iter);
1133 }
1134
1135 if (rt->fib6_pmtu)
1136 fib6_metric_set(iter, RTAX_MTU,
1137 rt->fib6_pmtu);
1138 return -EEXIST;
1139 }
1140 /* If we have the same destination and the same metric,
1141 * but not the same gateway, then the route we try to
1142 * add is sibling to this route, increment our counter
1143 * of siblings, and later we will add our route to the
1144 * list.
1145 * Only static routes (which don't have flag
1146 * RTF_EXPIRES) are used for ECMPv6.
1147 *
1148 * To avoid long list, we only had siblings if the
1149 * route have a gateway.
1150 */
1151 if (rt_can_ecmp &&
1152 rt6_qualify_for_ecmp(iter))
1153 rt->fib6_nsiblings++;
1154 }
1155
1156 if (iter->fib6_metric > rt->fib6_metric)
1157 break;
1158
1159next_iter:
1160 ins = &iter->fib6_next;
1161 }
1162
1163 if (fallback_ins && !found) {
1164 /* No matching route with same ecmp-able-ness found, replace
1165 * first matching route
1166 */
1167 ins = fallback_ins;
1168 iter = rcu_dereference_protected(*ins,
1169 lockdep_is_held(&rt->fib6_table->tb6_lock));
1170 found++;
1171 }
1172
1173 /* Reset round-robin state, if necessary */
1174 if (ins == &fn->leaf)
1175 fn->rr_ptr = NULL;
1176
1177 /* Link this route to others same route. */
1178 if (rt->fib6_nsiblings) {
1179 unsigned int fib6_nsiblings;
1180 struct fib6_info *sibling, *temp_sibling;
1181
1182 /* Find the first route that have the same metric */
1183 sibling = leaf;
1184 notify_sibling_rt = true;
1185 while (sibling) {
1186 if (sibling->fib6_metric == rt->fib6_metric &&
1187 rt6_qualify_for_ecmp(sibling)) {
1188 list_add_tail(&rt->fib6_siblings,
1189 &sibling->fib6_siblings);
1190 break;
1191 }
1192 sibling = rcu_dereference_protected(sibling->fib6_next,
1193 lockdep_is_held(&rt->fib6_table->tb6_lock));
1194 notify_sibling_rt = false;
1195 }
1196 /* For each sibling in the list, increment the counter of
1197 * siblings. BUG() if counters does not match, list of siblings
1198 * is broken!
1199 */
1200 fib6_nsiblings = 0;
1201 list_for_each_entry_safe(sibling, temp_sibling,
1202 &rt->fib6_siblings, fib6_siblings) {
1203 sibling->fib6_nsiblings++;
1204 BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1205 fib6_nsiblings++;
1206 }
1207 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1208 rt6_multipath_rebalance(temp_sibling);
1209 }
1210
1211 /*
1212 * insert node
1213 */
1214 if (!replace) {
1215 if (!add)
1216 pr_warn("NLM_F_CREATE should be set when creating new route\n");
1217
1218add:
1219 nlflags |= NLM_F_CREATE;
1220
1221 /* The route should only be notified if it is the first
1222 * route in the node or if it is added as a sibling
1223 * route to the first route in the node.
1224 */
1225 if (!info->skip_notify_kernel &&
1226 (notify_sibling_rt || ins == &fn->leaf)) {
1227 enum fib_event_type fib_event;
1228
1229 if (notify_sibling_rt)
1230 fib_event = FIB_EVENT_ENTRY_APPEND;
1231 else
1232 fib_event = FIB_EVENT_ENTRY_REPLACE;
1233 err = call_fib6_entry_notifiers(info->nl_net,
1234 fib_event, rt,
1235 extack);
1236 if (err) {
1237 struct fib6_info *sibling, *next_sibling;
1238
1239 /* If the route has siblings, then it first
1240 * needs to be unlinked from them.
1241 */
1242 if (!rt->fib6_nsiblings)
1243 return err;
1244
1245 list_for_each_entry_safe(sibling, next_sibling,
1246 &rt->fib6_siblings,
1247 fib6_siblings)
1248 sibling->fib6_nsiblings--;
1249 rt->fib6_nsiblings = 0;
1250 list_del_init(&rt->fib6_siblings);
1251 rt6_multipath_rebalance(next_sibling);
1252 return err;
1253 }
1254 }
1255
1256 rcu_assign_pointer(rt->fib6_next, iter);
1257 fib6_info_hold(rt);
1258 rcu_assign_pointer(rt->fib6_node, fn);
1259 rcu_assign_pointer(*ins, rt);
1260 if (!info->skip_notify)
1261 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1262 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1263
1264 if (!(fn->fn_flags & RTN_RTINFO)) {
1265 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1266 fn->fn_flags |= RTN_RTINFO;
1267 }
1268
1269 } else {
1270 int nsiblings;
1271
1272 if (!found) {
1273 if (add)
1274 goto add;
1275 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1276 return -ENOENT;
1277 }
1278
1279 if (!info->skip_notify_kernel && ins == &fn->leaf) {
1280 err = call_fib6_entry_notifiers(info->nl_net,
1281 FIB_EVENT_ENTRY_REPLACE,
1282 rt, extack);
1283 if (err)
1284 return err;
1285 }
1286
1287 fib6_info_hold(rt);
1288 rcu_assign_pointer(rt->fib6_node, fn);
1289 rt->fib6_next = iter->fib6_next;
1290 rcu_assign_pointer(*ins, rt);
1291 if (!info->skip_notify)
1292 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1293 if (!(fn->fn_flags & RTN_RTINFO)) {
1294 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1295 fn->fn_flags |= RTN_RTINFO;
1296 }
1297 nsiblings = iter->fib6_nsiblings;
1298 iter->fib6_node = NULL;
1299 fib6_purge_rt(iter, fn, info->nl_net);
1300 if (rcu_access_pointer(fn->rr_ptr) == iter)
1301 fn->rr_ptr = NULL;
1302 fib6_info_release(iter);
1303
1304 if (nsiblings) {
1305 /* Replacing an ECMP route, remove all siblings */
1306 ins = &rt->fib6_next;
1307 iter = rcu_dereference_protected(*ins,
1308 lockdep_is_held(&rt->fib6_table->tb6_lock));
1309 while (iter) {
1310 if (iter->fib6_metric > rt->fib6_metric)
1311 break;
1312 if (rt6_qualify_for_ecmp(iter)) {
1313 *ins = iter->fib6_next;
1314 iter->fib6_node = NULL;
1315 fib6_purge_rt(iter, fn, info->nl_net);
1316 if (rcu_access_pointer(fn->rr_ptr) == iter)
1317 fn->rr_ptr = NULL;
1318 fib6_info_release(iter);
1319 nsiblings--;
1320 info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1321 } else {
1322 ins = &iter->fib6_next;
1323 }
1324 iter = rcu_dereference_protected(*ins,
1325 lockdep_is_held(&rt->fib6_table->tb6_lock));
1326 }
1327 WARN_ON(nsiblings != 0);
1328 }
1329 }
1330
1331 return 0;
1332}
1333
1334static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1335{
1336 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1337 (rt->fib6_flags & RTF_EXPIRES))
1338 mod_timer(&net->ipv6.ip6_fib_timer,
1339 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1340}
1341
1342void fib6_force_start_gc(struct net *net)
1343{
1344 if (!timer_pending(&net->ipv6.ip6_fib_timer))
1345 mod_timer(&net->ipv6.ip6_fib_timer,
1346 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1347}
1348
1349static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1350 int sernum)
1351{
1352 struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1353 lockdep_is_held(&rt->fib6_table->tb6_lock));
1354
1355 /* paired with smp_rmb() in fib6_get_cookie_safe() */
1356 smp_wmb();
1357 while (fn) {
1358 WRITE_ONCE(fn->fn_sernum, sernum);
1359 fn = rcu_dereference_protected(fn->parent,
1360 lockdep_is_held(&rt->fib6_table->tb6_lock));
1361 }
1362}
1363
1364void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1365{
1366 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1367}
1368
1369/* allow ipv4 to update sernum via ipv6_stub */
1370void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1371{
1372 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1373 fib6_update_sernum_upto_root(net, f6i);
1374 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1375}
1376
1377/*
1378 * Add routing information to the routing tree.
1379 * <destination addr>/<source addr>
1380 * with source addr info in sub-trees
1381 * Need to own table->tb6_lock
1382 */
1383
1384int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1385 struct nl_info *info, struct netlink_ext_ack *extack)
1386{
1387 struct fib6_table *table = rt->fib6_table;
1388 struct fib6_node *fn;
1389#ifdef CONFIG_IPV6_SUBTREES
1390 struct fib6_node *pn = NULL;
1391#endif
1392 int err = -ENOMEM;
1393 int allow_create = 1;
1394 int replace_required = 0;
1395
1396 if (info->nlh) {
1397 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1398 allow_create = 0;
1399 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1400 replace_required = 1;
1401 }
1402 if (!allow_create && !replace_required)
1403 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1404
1405 fn = fib6_add_1(info->nl_net, table, root,
1406 &rt->fib6_dst.addr, rt->fib6_dst.plen,
1407 offsetof(struct fib6_info, fib6_dst), allow_create,
1408 replace_required, extack);
1409 if (IS_ERR(fn)) {
1410 err = PTR_ERR(fn);
1411 fn = NULL;
1412 goto out;
1413 }
1414
1415#ifdef CONFIG_IPV6_SUBTREES
1416 pn = fn;
1417
1418 if (rt->fib6_src.plen) {
1419 struct fib6_node *sn;
1420
1421 if (!rcu_access_pointer(fn->subtree)) {
1422 struct fib6_node *sfn;
1423
1424 /*
1425 * Create subtree.
1426 *
1427 * fn[main tree]
1428 * |
1429 * sfn[subtree root]
1430 * \
1431 * sn[new leaf node]
1432 */
1433
1434 /* Create subtree root node */
1435 sfn = node_alloc(info->nl_net);
1436 if (!sfn)
1437 goto failure;
1438
1439 fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1440 rcu_assign_pointer(sfn->leaf,
1441 info->nl_net->ipv6.fib6_null_entry);
1442 sfn->fn_flags = RTN_ROOT;
1443
1444 /* Now add the first leaf node to new subtree */
1445
1446 sn = fib6_add_1(info->nl_net, table, sfn,
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 /* If it is failed, discard just allocated
1453 root, and then (in failure) stale node
1454 in main tree.
1455 */
1456 node_free_immediate(info->nl_net, sfn);
1457 err = PTR_ERR(sn);
1458 goto failure;
1459 }
1460
1461 /* Now link new subtree to main tree */
1462 rcu_assign_pointer(sfn->parent, fn);
1463 rcu_assign_pointer(fn->subtree, sfn);
1464 } else {
1465 sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1466 &rt->fib6_src.addr, rt->fib6_src.plen,
1467 offsetof(struct fib6_info, fib6_src),
1468 allow_create, replace_required, extack);
1469
1470 if (IS_ERR(sn)) {
1471 err = PTR_ERR(sn);
1472 goto failure;
1473 }
1474 }
1475
1476 if (!rcu_access_pointer(fn->leaf)) {
1477 if (fn->fn_flags & RTN_TL_ROOT) {
1478 /* put back null_entry for root node */
1479 rcu_assign_pointer(fn->leaf,
1480 info->nl_net->ipv6.fib6_null_entry);
1481 } else {
1482 fib6_info_hold(rt);
1483 rcu_assign_pointer(fn->leaf, rt);
1484 }
1485 }
1486 fn = sn;
1487 }
1488#endif
1489
1490 err = fib6_add_rt2node(fn, rt, info, extack);
1491 if (!err) {
1492 if (rt->nh)
1493 list_add(&rt->nh_list, &rt->nh->f6i_list);
1494 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(info->nl_net));
1495
1496 if (rt->fib6_flags & RTF_EXPIRES)
1497 fib6_add_gc_list(rt);
1498
1499 fib6_start_gc(info->nl_net, rt);
1500 }
1501
1502out:
1503 if (err) {
1504#ifdef CONFIG_IPV6_SUBTREES
1505 /*
1506 * If fib6_add_1 has cleared the old leaf pointer in the
1507 * super-tree leaf node we have to find a new one for it.
1508 */
1509 if (pn != fn) {
1510 struct fib6_info *pn_leaf =
1511 rcu_dereference_protected(pn->leaf,
1512 lockdep_is_held(&table->tb6_lock));
1513 if (pn_leaf == rt) {
1514 pn_leaf = NULL;
1515 RCU_INIT_POINTER(pn->leaf, NULL);
1516 fib6_info_release(rt);
1517 }
1518 if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1519 pn_leaf = fib6_find_prefix(info->nl_net, table,
1520 pn);
1521 if (!pn_leaf)
1522 pn_leaf =
1523 info->nl_net->ipv6.fib6_null_entry;
1524 fib6_info_hold(pn_leaf);
1525 rcu_assign_pointer(pn->leaf, pn_leaf);
1526 }
1527 }
1528#endif
1529 goto failure;
1530 } else if (fib6_requires_src(rt)) {
1531 fib6_routes_require_src_inc(info->nl_net);
1532 }
1533 return err;
1534
1535failure:
1536 /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1537 * 1. fn is an intermediate node and we failed to add the new
1538 * route to it in both subtree creation failure and fib6_add_rt2node()
1539 * failure case.
1540 * 2. fn is the root node in the table and we fail to add the first
1541 * default route to it.
1542 */
1543 if (fn &&
1544 (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1545 (fn->fn_flags & RTN_TL_ROOT &&
1546 !rcu_access_pointer(fn->leaf))))
1547 fib6_repair_tree(info->nl_net, table, fn);
1548 return err;
1549}
1550
1551/*
1552 * Routing tree lookup
1553 *
1554 */
1555
1556struct lookup_args {
1557 int offset; /* key offset on fib6_info */
1558 const struct in6_addr *addr; /* search key */
1559};
1560
1561static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1562 struct lookup_args *args)
1563{
1564 struct fib6_node *fn;
1565 __be32 dir;
1566
1567 if (unlikely(args->offset == 0))
1568 return NULL;
1569
1570 /*
1571 * Descend on a tree
1572 */
1573
1574 fn = root;
1575
1576 for (;;) {
1577 struct fib6_node *next;
1578
1579 dir = addr_bit_set(args->addr, fn->fn_bit);
1580
1581 next = dir ? rcu_dereference(fn->right) :
1582 rcu_dereference(fn->left);
1583
1584 if (next) {
1585 fn = next;
1586 continue;
1587 }
1588 break;
1589 }
1590
1591 while (fn) {
1592 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1593
1594 if (subtree || fn->fn_flags & RTN_RTINFO) {
1595 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1596 struct rt6key *key;
1597
1598 if (!leaf)
1599 goto backtrack;
1600
1601 key = (struct rt6key *) ((u8 *)leaf + args->offset);
1602
1603 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1604#ifdef CONFIG_IPV6_SUBTREES
1605 if (subtree) {
1606 struct fib6_node *sfn;
1607 sfn = fib6_node_lookup_1(subtree,
1608 args + 1);
1609 if (!sfn)
1610 goto backtrack;
1611 fn = sfn;
1612 }
1613#endif
1614 if (fn->fn_flags & RTN_RTINFO)
1615 return fn;
1616 }
1617 }
1618backtrack:
1619 if (fn->fn_flags & RTN_ROOT)
1620 break;
1621
1622 fn = rcu_dereference(fn->parent);
1623 }
1624
1625 return NULL;
1626}
1627
1628/* called with rcu_read_lock() held
1629 */
1630struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1631 const struct in6_addr *daddr,
1632 const struct in6_addr *saddr)
1633{
1634 struct fib6_node *fn;
1635 struct lookup_args args[] = {
1636 {
1637 .offset = offsetof(struct fib6_info, fib6_dst),
1638 .addr = daddr,
1639 },
1640#ifdef CONFIG_IPV6_SUBTREES
1641 {
1642 .offset = offsetof(struct fib6_info, fib6_src),
1643 .addr = saddr,
1644 },
1645#endif
1646 {
1647 .offset = 0, /* sentinel */
1648 }
1649 };
1650
1651 fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1652 if (!fn || fn->fn_flags & RTN_TL_ROOT)
1653 fn = root;
1654
1655 return fn;
1656}
1657
1658/*
1659 * Get node with specified destination prefix (and source prefix,
1660 * if subtrees are used)
1661 * exact_match == true means we try to find fn with exact match of
1662 * the passed in prefix addr
1663 * exact_match == false means we try to find fn with longest prefix
1664 * match of the passed in prefix addr. This is useful for finding fn
1665 * for cached route as it will be stored in the exception table under
1666 * the node with longest prefix length.
1667 */
1668
1669
1670static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1671 const struct in6_addr *addr,
1672 int plen, int offset,
1673 bool exact_match)
1674{
1675 struct fib6_node *fn, *prev = NULL;
1676
1677 for (fn = root; fn ; ) {
1678 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1679 struct rt6key *key;
1680
1681 /* This node is being deleted */
1682 if (!leaf) {
1683 if (plen <= fn->fn_bit)
1684 goto out;
1685 else
1686 goto next;
1687 }
1688
1689 key = (struct rt6key *)((u8 *)leaf + offset);
1690
1691 /*
1692 * Prefix match
1693 */
1694 if (plen < fn->fn_bit ||
1695 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1696 goto out;
1697
1698 if (plen == fn->fn_bit)
1699 return fn;
1700
1701 if (fn->fn_flags & RTN_RTINFO)
1702 prev = fn;
1703
1704next:
1705 /*
1706 * We have more bits to go
1707 */
1708 if (addr_bit_set(addr, fn->fn_bit))
1709 fn = rcu_dereference(fn->right);
1710 else
1711 fn = rcu_dereference(fn->left);
1712 }
1713out:
1714 if (exact_match)
1715 return NULL;
1716 else
1717 return prev;
1718}
1719
1720struct fib6_node *fib6_locate(struct fib6_node *root,
1721 const struct in6_addr *daddr, int dst_len,
1722 const struct in6_addr *saddr, int src_len,
1723 bool exact_match)
1724{
1725 struct fib6_node *fn;
1726
1727 fn = fib6_locate_1(root, daddr, dst_len,
1728 offsetof(struct fib6_info, fib6_dst),
1729 exact_match);
1730
1731#ifdef CONFIG_IPV6_SUBTREES
1732 if (src_len) {
1733 WARN_ON(saddr == NULL);
1734 if (fn) {
1735 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1736
1737 if (subtree) {
1738 fn = fib6_locate_1(subtree, saddr, src_len,
1739 offsetof(struct fib6_info, fib6_src),
1740 exact_match);
1741 }
1742 }
1743 }
1744#endif
1745
1746 if (fn && fn->fn_flags & RTN_RTINFO)
1747 return fn;
1748
1749 return NULL;
1750}
1751
1752
1753/*
1754 * Deletion
1755 *
1756 */
1757
1758static struct fib6_info *fib6_find_prefix(struct net *net,
1759 struct fib6_table *table,
1760 struct fib6_node *fn)
1761{
1762 struct fib6_node *child_left, *child_right;
1763
1764 if (fn->fn_flags & RTN_ROOT)
1765 return net->ipv6.fib6_null_entry;
1766
1767 while (fn) {
1768 child_left = rcu_dereference_protected(fn->left,
1769 lockdep_is_held(&table->tb6_lock));
1770 child_right = rcu_dereference_protected(fn->right,
1771 lockdep_is_held(&table->tb6_lock));
1772 if (child_left)
1773 return rcu_dereference_protected(child_left->leaf,
1774 lockdep_is_held(&table->tb6_lock));
1775 if (child_right)
1776 return rcu_dereference_protected(child_right->leaf,
1777 lockdep_is_held(&table->tb6_lock));
1778
1779 fn = FIB6_SUBTREE(fn);
1780 }
1781 return NULL;
1782}
1783
1784/*
1785 * Called to trim the tree of intermediate nodes when possible. "fn"
1786 * is the node we want to try and remove.
1787 * Need to own table->tb6_lock
1788 */
1789
1790static struct fib6_node *fib6_repair_tree(struct net *net,
1791 struct fib6_table *table,
1792 struct fib6_node *fn)
1793{
1794 int children;
1795 int nstate;
1796 struct fib6_node *child;
1797 struct fib6_walker *w;
1798 int iter = 0;
1799
1800 /* Set fn->leaf to null_entry for root node. */
1801 if (fn->fn_flags & RTN_TL_ROOT) {
1802 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1803 return fn;
1804 }
1805
1806 for (;;) {
1807 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1808 lockdep_is_held(&table->tb6_lock));
1809 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1810 lockdep_is_held(&table->tb6_lock));
1811 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1812 lockdep_is_held(&table->tb6_lock));
1813 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1814 lockdep_is_held(&table->tb6_lock));
1815 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1816 lockdep_is_held(&table->tb6_lock));
1817 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1818 lockdep_is_held(&table->tb6_lock));
1819 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1820 lockdep_is_held(&table->tb6_lock));
1821 struct fib6_info *new_fn_leaf;
1822
1823 pr_debug("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1824 iter++;
1825
1826 WARN_ON(fn->fn_flags & RTN_RTINFO);
1827 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1828 WARN_ON(fn_leaf);
1829
1830 children = 0;
1831 child = NULL;
1832 if (fn_r) {
1833 child = fn_r;
1834 children |= 1;
1835 }
1836 if (fn_l) {
1837 child = fn_l;
1838 children |= 2;
1839 }
1840
1841 if (children == 3 || FIB6_SUBTREE(fn)
1842#ifdef CONFIG_IPV6_SUBTREES
1843 /* Subtree root (i.e. fn) may have one child */
1844 || (children && fn->fn_flags & RTN_ROOT)
1845#endif
1846 ) {
1847 new_fn_leaf = fib6_find_prefix(net, table, fn);
1848#if RT6_DEBUG >= 2
1849 if (!new_fn_leaf) {
1850 WARN_ON(!new_fn_leaf);
1851 new_fn_leaf = net->ipv6.fib6_null_entry;
1852 }
1853#endif
1854 fib6_info_hold(new_fn_leaf);
1855 rcu_assign_pointer(fn->leaf, new_fn_leaf);
1856 return pn;
1857 }
1858
1859#ifdef CONFIG_IPV6_SUBTREES
1860 if (FIB6_SUBTREE(pn) == fn) {
1861 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1862 RCU_INIT_POINTER(pn->subtree, NULL);
1863 nstate = FWS_L;
1864 } else {
1865 WARN_ON(fn->fn_flags & RTN_ROOT);
1866#endif
1867 if (pn_r == fn)
1868 rcu_assign_pointer(pn->right, child);
1869 else if (pn_l == fn)
1870 rcu_assign_pointer(pn->left, child);
1871#if RT6_DEBUG >= 2
1872 else
1873 WARN_ON(1);
1874#endif
1875 if (child)
1876 rcu_assign_pointer(child->parent, pn);
1877 nstate = FWS_R;
1878#ifdef CONFIG_IPV6_SUBTREES
1879 }
1880#endif
1881
1882 read_lock(&net->ipv6.fib6_walker_lock);
1883 FOR_WALKERS(net, w) {
1884 if (!child) {
1885 if (w->node == fn) {
1886 pr_debug("W %p adjusted by delnode 1, s=%d/%d\n",
1887 w, w->state, nstate);
1888 w->node = pn;
1889 w->state = nstate;
1890 }
1891 } else {
1892 if (w->node == fn) {
1893 w->node = child;
1894 if (children&2) {
1895 pr_debug("W %p adjusted by delnode 2, s=%d\n",
1896 w, w->state);
1897 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1898 } else {
1899 pr_debug("W %p adjusted by delnode 2, s=%d\n",
1900 w, w->state);
1901 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1902 }
1903 }
1904 }
1905 }
1906 read_unlock(&net->ipv6.fib6_walker_lock);
1907
1908 node_free(net, fn);
1909 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1910 return pn;
1911
1912 RCU_INIT_POINTER(pn->leaf, NULL);
1913 fib6_info_release(pn_leaf);
1914 fn = pn;
1915 }
1916}
1917
1918static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1919 struct fib6_info __rcu **rtp, struct nl_info *info)
1920{
1921 struct fib6_info *leaf, *replace_rt = NULL;
1922 struct fib6_walker *w;
1923 struct fib6_info *rt = rcu_dereference_protected(*rtp,
1924 lockdep_is_held(&table->tb6_lock));
1925 struct net *net = info->nl_net;
1926 bool notify_del = false;
1927
1928 /* If the deleted route is the first in the node and it is not part of
1929 * a multipath route, then we need to replace it with the next route
1930 * in the node, if exists.
1931 */
1932 leaf = rcu_dereference_protected(fn->leaf,
1933 lockdep_is_held(&table->tb6_lock));
1934 if (leaf == rt && !rt->fib6_nsiblings) {
1935 if (rcu_access_pointer(rt->fib6_next))
1936 replace_rt = rcu_dereference_protected(rt->fib6_next,
1937 lockdep_is_held(&table->tb6_lock));
1938 else
1939 notify_del = true;
1940 }
1941
1942 /* Unlink it */
1943 *rtp = rt->fib6_next;
1944 rt->fib6_node = NULL;
1945 net->ipv6.rt6_stats->fib_rt_entries--;
1946 net->ipv6.rt6_stats->fib_discarded_routes++;
1947
1948 /* Reset round-robin state, if necessary */
1949 if (rcu_access_pointer(fn->rr_ptr) == rt)
1950 fn->rr_ptr = NULL;
1951
1952 /* Remove this entry from other siblings */
1953 if (rt->fib6_nsiblings) {
1954 struct fib6_info *sibling, *next_sibling;
1955
1956 /* The route is deleted from a multipath route. If this
1957 * multipath route is the first route in the node, then we need
1958 * to emit a delete notification. Otherwise, we need to skip
1959 * the notification.
1960 */
1961 if (rt->fib6_metric == leaf->fib6_metric &&
1962 rt6_qualify_for_ecmp(leaf))
1963 notify_del = true;
1964 list_for_each_entry_safe(sibling, next_sibling,
1965 &rt->fib6_siblings, fib6_siblings)
1966 sibling->fib6_nsiblings--;
1967 rt->fib6_nsiblings = 0;
1968 list_del_init(&rt->fib6_siblings);
1969 rt6_multipath_rebalance(next_sibling);
1970 }
1971
1972 /* Adjust walkers */
1973 read_lock(&net->ipv6.fib6_walker_lock);
1974 FOR_WALKERS(net, w) {
1975 if (w->state == FWS_C && w->leaf == rt) {
1976 pr_debug("walker %p adjusted by delroute\n", w);
1977 w->leaf = rcu_dereference_protected(rt->fib6_next,
1978 lockdep_is_held(&table->tb6_lock));
1979 if (!w->leaf)
1980 w->state = FWS_U;
1981 }
1982 }
1983 read_unlock(&net->ipv6.fib6_walker_lock);
1984
1985 /* If it was last route, call fib6_repair_tree() to:
1986 * 1. For root node, put back null_entry as how the table was created.
1987 * 2. For other nodes, expunge its radix tree node.
1988 */
1989 if (!rcu_access_pointer(fn->leaf)) {
1990 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1991 fn->fn_flags &= ~RTN_RTINFO;
1992 net->ipv6.rt6_stats->fib_route_nodes--;
1993 }
1994 fn = fib6_repair_tree(net, table, fn);
1995 }
1996
1997 fib6_purge_rt(rt, fn, net);
1998
1999 if (!info->skip_notify_kernel) {
2000 if (notify_del)
2001 call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
2002 rt, NULL);
2003 else if (replace_rt)
2004 call_fib6_entry_notifiers_replace(net, replace_rt);
2005 }
2006 if (!info->skip_notify)
2007 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
2008
2009 fib6_info_release(rt);
2010}
2011
2012/* Need to own table->tb6_lock */
2013int fib6_del(struct fib6_info *rt, struct nl_info *info)
2014{
2015 struct net *net = info->nl_net;
2016 struct fib6_info __rcu **rtp;
2017 struct fib6_info __rcu **rtp_next;
2018 struct fib6_table *table;
2019 struct fib6_node *fn;
2020
2021 if (rt == net->ipv6.fib6_null_entry)
2022 return -ENOENT;
2023
2024 table = rt->fib6_table;
2025 fn = rcu_dereference_protected(rt->fib6_node,
2026 lockdep_is_held(&table->tb6_lock));
2027 if (!fn)
2028 return -ENOENT;
2029
2030 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
2031
2032 /*
2033 * Walk the leaf entries looking for ourself
2034 */
2035
2036 for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
2037 struct fib6_info *cur = rcu_dereference_protected(*rtp,
2038 lockdep_is_held(&table->tb6_lock));
2039 if (rt == cur) {
2040 if (fib6_requires_src(cur))
2041 fib6_routes_require_src_dec(info->nl_net);
2042 fib6_del_route(table, fn, rtp, info);
2043 return 0;
2044 }
2045 rtp_next = &cur->fib6_next;
2046 }
2047 return -ENOENT;
2048}
2049
2050/*
2051 * Tree traversal function.
2052 *
2053 * Certainly, it is not interrupt safe.
2054 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
2055 * It means, that we can modify tree during walking
2056 * and use this function for garbage collection, clone pruning,
2057 * cleaning tree when a device goes down etc. etc.
2058 *
2059 * It guarantees that every node will be traversed,
2060 * and that it will be traversed only once.
2061 *
2062 * Callback function w->func may return:
2063 * 0 -> continue walking.
2064 * positive value -> walking is suspended (used by tree dumps,
2065 * and probably by gc, if it will be split to several slices)
2066 * negative value -> terminate walking.
2067 *
2068 * The function itself returns:
2069 * 0 -> walk is complete.
2070 * >0 -> walk is incomplete (i.e. suspended)
2071 * <0 -> walk is terminated by an error.
2072 *
2073 * This function is called with tb6_lock held.
2074 */
2075
2076static int fib6_walk_continue(struct fib6_walker *w)
2077{
2078 struct fib6_node *fn, *pn, *left, *right;
2079
2080 /* w->root should always be table->tb6_root */
2081 WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
2082
2083 for (;;) {
2084 fn = w->node;
2085 if (!fn)
2086 return 0;
2087
2088 switch (w->state) {
2089#ifdef CONFIG_IPV6_SUBTREES
2090 case FWS_S:
2091 if (FIB6_SUBTREE(fn)) {
2092 w->node = FIB6_SUBTREE(fn);
2093 continue;
2094 }
2095 w->state = FWS_L;
2096 fallthrough;
2097#endif
2098 case FWS_L:
2099 left = rcu_dereference_protected(fn->left, 1);
2100 if (left) {
2101 w->node = left;
2102 w->state = FWS_INIT;
2103 continue;
2104 }
2105 w->state = FWS_R;
2106 fallthrough;
2107 case FWS_R:
2108 right = rcu_dereference_protected(fn->right, 1);
2109 if (right) {
2110 w->node = right;
2111 w->state = FWS_INIT;
2112 continue;
2113 }
2114 w->state = FWS_C;
2115 w->leaf = rcu_dereference_protected(fn->leaf, 1);
2116 fallthrough;
2117 case FWS_C:
2118 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
2119 int err;
2120
2121 if (w->skip) {
2122 w->skip--;
2123 goto skip;
2124 }
2125
2126 err = w->func(w);
2127 if (err)
2128 return err;
2129
2130 w->count++;
2131 continue;
2132 }
2133skip:
2134 w->state = FWS_U;
2135 fallthrough;
2136 case FWS_U:
2137 if (fn == w->root)
2138 return 0;
2139 pn = rcu_dereference_protected(fn->parent, 1);
2140 left = rcu_dereference_protected(pn->left, 1);
2141 right = rcu_dereference_protected(pn->right, 1);
2142 w->node = pn;
2143#ifdef CONFIG_IPV6_SUBTREES
2144 if (FIB6_SUBTREE(pn) == fn) {
2145 WARN_ON(!(fn->fn_flags & RTN_ROOT));
2146 w->state = FWS_L;
2147 continue;
2148 }
2149#endif
2150 if (left == fn) {
2151 w->state = FWS_R;
2152 continue;
2153 }
2154 if (right == fn) {
2155 w->state = FWS_C;
2156 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
2157 continue;
2158 }
2159#if RT6_DEBUG >= 2
2160 WARN_ON(1);
2161#endif
2162 }
2163 }
2164}
2165
2166static int fib6_walk(struct net *net, struct fib6_walker *w)
2167{
2168 int res;
2169
2170 w->state = FWS_INIT;
2171 w->node = w->root;
2172
2173 fib6_walker_link(net, w);
2174 res = fib6_walk_continue(w);
2175 if (res <= 0)
2176 fib6_walker_unlink(net, w);
2177 return res;
2178}
2179
2180static int fib6_clean_node(struct fib6_walker *w)
2181{
2182 int res;
2183 struct fib6_info *rt;
2184 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
2185 struct nl_info info = {
2186 .nl_net = c->net,
2187 .skip_notify = c->skip_notify,
2188 };
2189
2190 if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
2191 READ_ONCE(w->node->fn_sernum) != c->sernum)
2192 WRITE_ONCE(w->node->fn_sernum, c->sernum);
2193
2194 if (!c->func) {
2195 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
2196 w->leaf = NULL;
2197 return 0;
2198 }
2199
2200 for_each_fib6_walker_rt(w) {
2201 res = c->func(rt, c->arg);
2202 if (res == -1) {
2203 w->leaf = rt;
2204 res = fib6_del(rt, &info);
2205 if (res) {
2206#if RT6_DEBUG >= 2
2207 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2208 __func__, rt,
2209 rcu_access_pointer(rt->fib6_node),
2210 res);
2211#endif
2212 continue;
2213 }
2214 return 0;
2215 } else if (res == -2) {
2216 if (WARN_ON(!rt->fib6_nsiblings))
2217 continue;
2218 rt = list_last_entry(&rt->fib6_siblings,
2219 struct fib6_info, fib6_siblings);
2220 continue;
2221 }
2222 WARN_ON(res != 0);
2223 }
2224 w->leaf = rt;
2225 return 0;
2226}
2227
2228/*
2229 * Convenient frontend to tree walker.
2230 *
2231 * func is called on each route.
2232 * It may return -2 -> skip multipath route.
2233 * -1 -> delete this route.
2234 * 0 -> continue walking
2235 */
2236
2237static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2238 int (*func)(struct fib6_info *, void *arg),
2239 int sernum, void *arg, bool skip_notify)
2240{
2241 struct fib6_cleaner c;
2242
2243 c.w.root = root;
2244 c.w.func = fib6_clean_node;
2245 c.w.count = 0;
2246 c.w.skip = 0;
2247 c.w.skip_in_node = 0;
2248 c.func = func;
2249 c.sernum = sernum;
2250 c.arg = arg;
2251 c.net = net;
2252 c.skip_notify = skip_notify;
2253
2254 fib6_walk(net, &c.w);
2255}
2256
2257static void __fib6_clean_all(struct net *net,
2258 int (*func)(struct fib6_info *, void *),
2259 int sernum, void *arg, bool skip_notify)
2260{
2261 struct fib6_table *table;
2262 struct hlist_head *head;
2263 unsigned int h;
2264
2265 rcu_read_lock();
2266 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2267 head = &net->ipv6.fib_table_hash[h];
2268 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2269 spin_lock_bh(&table->tb6_lock);
2270 fib6_clean_tree(net, &table->tb6_root,
2271 func, sernum, arg, skip_notify);
2272 spin_unlock_bh(&table->tb6_lock);
2273 }
2274 }
2275 rcu_read_unlock();
2276}
2277
2278void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2279 void *arg)
2280{
2281 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2282}
2283
2284void fib6_clean_all_skip_notify(struct net *net,
2285 int (*func)(struct fib6_info *, void *),
2286 void *arg)
2287{
2288 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2289}
2290
2291static void fib6_flush_trees(struct net *net)
2292{
2293 int new_sernum = fib6_new_sernum(net);
2294
2295 __fib6_clean_all(net, NULL, new_sernum, NULL, false);
2296}
2297
2298/*
2299 * Garbage collection
2300 */
2301
2302static int fib6_age(struct fib6_info *rt, struct fib6_gc_args *gc_args)
2303{
2304 unsigned long now = jiffies;
2305
2306 /*
2307 * check addrconf expiration here.
2308 * Routes are expired even if they are in use.
2309 */
2310
2311 if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2312 if (time_after(now, rt->expires)) {
2313 pr_debug("expiring %p\n", rt);
2314 return -1;
2315 }
2316 gc_args->more++;
2317 }
2318
2319 /* Also age clones in the exception table.
2320 * Note, that clones are aged out
2321 * only if they are not in use now.
2322 */
2323 rt6_age_exceptions(rt, gc_args, now);
2324
2325 return 0;
2326}
2327
2328static void fib6_gc_table(struct net *net,
2329 struct fib6_table *tb6,
2330 struct fib6_gc_args *gc_args)
2331{
2332 struct fib6_info *rt;
2333 struct hlist_node *n;
2334 struct nl_info info = {
2335 .nl_net = net,
2336 .skip_notify = false,
2337 };
2338
2339 hlist_for_each_entry_safe(rt, n, &tb6->tb6_gc_hlist, gc_link)
2340 if (fib6_age(rt, gc_args) == -1)
2341 fib6_del(rt, &info);
2342}
2343
2344static void fib6_gc_all(struct net *net, struct fib6_gc_args *gc_args)
2345{
2346 struct fib6_table *table;
2347 struct hlist_head *head;
2348 unsigned int h;
2349
2350 rcu_read_lock();
2351 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2352 head = &net->ipv6.fib_table_hash[h];
2353 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2354 spin_lock_bh(&table->tb6_lock);
2355
2356 fib6_gc_table(net, table, gc_args);
2357
2358 spin_unlock_bh(&table->tb6_lock);
2359 }
2360 }
2361 rcu_read_unlock();
2362}
2363
2364void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2365{
2366 struct fib6_gc_args gc_args;
2367 unsigned long now;
2368
2369 if (force) {
2370 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2371 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2372 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2373 return;
2374 }
2375 gc_args.timeout = expires ? (int)expires :
2376 net->ipv6.sysctl.ip6_rt_gc_interval;
2377 gc_args.more = 0;
2378
2379 fib6_gc_all(net, &gc_args);
2380 now = jiffies;
2381 net->ipv6.ip6_rt_last_gc = now;
2382
2383 if (gc_args.more)
2384 mod_timer(&net->ipv6.ip6_fib_timer,
2385 round_jiffies(now
2386 + net->ipv6.sysctl.ip6_rt_gc_interval));
2387 else
2388 del_timer(&net->ipv6.ip6_fib_timer);
2389 spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2390}
2391
2392static void fib6_gc_timer_cb(struct timer_list *t)
2393{
2394 struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2395
2396 fib6_run_gc(0, arg, true);
2397}
2398
2399static int __net_init fib6_net_init(struct net *net)
2400{
2401 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2402 int err;
2403
2404 err = fib6_notifier_init(net);
2405 if (err)
2406 return err;
2407
2408 /* Default to 3-tuple */
2409 net->ipv6.sysctl.multipath_hash_fields =
2410 FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK;
2411
2412 spin_lock_init(&net->ipv6.fib6_gc_lock);
2413 rwlock_init(&net->ipv6.fib6_walker_lock);
2414 INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2415 timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2416
2417 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2418 if (!net->ipv6.rt6_stats)
2419 goto out_notifier;
2420
2421 /* Avoid false sharing : Use at least a full cache line */
2422 size = max_t(size_t, size, L1_CACHE_BYTES);
2423
2424 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2425 if (!net->ipv6.fib_table_hash)
2426 goto out_rt6_stats;
2427
2428 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2429 GFP_KERNEL);
2430 if (!net->ipv6.fib6_main_tbl)
2431 goto out_fib_table_hash;
2432
2433 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2434 rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2435 net->ipv6.fib6_null_entry);
2436 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2437 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2438 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2439 INIT_HLIST_HEAD(&net->ipv6.fib6_main_tbl->tb6_gc_hlist);
2440
2441#ifdef CONFIG_IPV6_MULTIPLE_TABLES
2442 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2443 GFP_KERNEL);
2444 if (!net->ipv6.fib6_local_tbl)
2445 goto out_fib6_main_tbl;
2446 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2447 rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2448 net->ipv6.fib6_null_entry);
2449 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2450 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2451 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2452 INIT_HLIST_HEAD(&net->ipv6.fib6_local_tbl->tb6_gc_hlist);
2453#endif
2454 fib6_tables_init(net);
2455
2456 return 0;
2457
2458#ifdef CONFIG_IPV6_MULTIPLE_TABLES
2459out_fib6_main_tbl:
2460 kfree(net->ipv6.fib6_main_tbl);
2461#endif
2462out_fib_table_hash:
2463 kfree(net->ipv6.fib_table_hash);
2464out_rt6_stats:
2465 kfree(net->ipv6.rt6_stats);
2466out_notifier:
2467 fib6_notifier_exit(net);
2468 return -ENOMEM;
2469}
2470
2471static void fib6_net_exit(struct net *net)
2472{
2473 unsigned int i;
2474
2475 del_timer_sync(&net->ipv6.ip6_fib_timer);
2476
2477 for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2478 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2479 struct hlist_node *tmp;
2480 struct fib6_table *tb;
2481
2482 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2483 hlist_del(&tb->tb6_hlist);
2484 fib6_free_table(tb);
2485 }
2486 }
2487
2488 kfree(net->ipv6.fib_table_hash);
2489 kfree(net->ipv6.rt6_stats);
2490 fib6_notifier_exit(net);
2491}
2492
2493static struct pernet_operations fib6_net_ops = {
2494 .init = fib6_net_init,
2495 .exit = fib6_net_exit,
2496};
2497
2498int __init fib6_init(void)
2499{
2500 int ret = -ENOMEM;
2501
2502 fib6_node_kmem = KMEM_CACHE(fib6_node,
2503 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT);
2504 if (!fib6_node_kmem)
2505 goto out;
2506
2507 ret = register_pernet_subsys(&fib6_net_ops);
2508 if (ret)
2509 goto out_kmem_cache_create;
2510
2511 ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2512 inet6_dump_fib, 0);
2513 if (ret)
2514 goto out_unregister_subsys;
2515
2516 __fib6_flush_trees = fib6_flush_trees;
2517out:
2518 return ret;
2519
2520out_unregister_subsys:
2521 unregister_pernet_subsys(&fib6_net_ops);
2522out_kmem_cache_create:
2523 kmem_cache_destroy(fib6_node_kmem);
2524 goto out;
2525}
2526
2527void fib6_gc_cleanup(void)
2528{
2529 unregister_pernet_subsys(&fib6_net_ops);
2530 kmem_cache_destroy(fib6_node_kmem);
2531}
2532
2533#ifdef CONFIG_PROC_FS
2534static int ipv6_route_native_seq_show(struct seq_file *seq, void *v)
2535{
2536 struct fib6_info *rt = v;
2537 struct ipv6_route_iter *iter = seq->private;
2538 struct fib6_nh *fib6_nh = rt->fib6_nh;
2539 unsigned int flags = rt->fib6_flags;
2540 const struct net_device *dev;
2541
2542 if (rt->nh)
2543 fib6_nh = nexthop_fib6_nh(rt->nh);
2544
2545 seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2546
2547#ifdef CONFIG_IPV6_SUBTREES
2548 seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2549#else
2550 seq_puts(seq, "00000000000000000000000000000000 00 ");
2551#endif
2552 if (fib6_nh->fib_nh_gw_family) {
2553 flags |= RTF_GATEWAY;
2554 seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6);
2555 } else {
2556 seq_puts(seq, "00000000000000000000000000000000");
2557 }
2558
2559 dev = fib6_nh->fib_nh_dev;
2560 seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2561 rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2562 flags, dev ? dev->name : "");
2563 iter->w.leaf = NULL;
2564 return 0;
2565}
2566
2567static int ipv6_route_yield(struct fib6_walker *w)
2568{
2569 struct ipv6_route_iter *iter = w->args;
2570
2571 if (!iter->skip)
2572 return 1;
2573
2574 do {
2575 iter->w.leaf = rcu_dereference_protected(
2576 iter->w.leaf->fib6_next,
2577 lockdep_is_held(&iter->tbl->tb6_lock));
2578 iter->skip--;
2579 if (!iter->skip && iter->w.leaf)
2580 return 1;
2581 } while (iter->w.leaf);
2582
2583 return 0;
2584}
2585
2586static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2587 struct net *net)
2588{
2589 memset(&iter->w, 0, sizeof(iter->w));
2590 iter->w.func = ipv6_route_yield;
2591 iter->w.root = &iter->tbl->tb6_root;
2592 iter->w.state = FWS_INIT;
2593 iter->w.node = iter->w.root;
2594 iter->w.args = iter;
2595 iter->sernum = READ_ONCE(iter->w.root->fn_sernum);
2596 INIT_LIST_HEAD(&iter->w.lh);
2597 fib6_walker_link(net, &iter->w);
2598}
2599
2600static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2601 struct net *net)
2602{
2603 unsigned int h;
2604 struct hlist_node *node;
2605
2606 if (tbl) {
2607 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2608 node = rcu_dereference(hlist_next_rcu(&tbl->tb6_hlist));
2609 } else {
2610 h = 0;
2611 node = NULL;
2612 }
2613
2614 while (!node && h < FIB6_TABLE_HASHSZ) {
2615 node = rcu_dereference(
2616 hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2617 }
2618 return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2619}
2620
2621static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2622{
2623 int sernum = READ_ONCE(iter->w.root->fn_sernum);
2624
2625 if (iter->sernum != sernum) {
2626 iter->sernum = sernum;
2627 iter->w.state = FWS_INIT;
2628 iter->w.node = iter->w.root;
2629 WARN_ON(iter->w.skip);
2630 iter->w.skip = iter->w.count;
2631 }
2632}
2633
2634static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2635{
2636 int r;
2637 struct fib6_info *n;
2638 struct net *net = seq_file_net(seq);
2639 struct ipv6_route_iter *iter = seq->private;
2640
2641 ++(*pos);
2642 if (!v)
2643 goto iter_table;
2644
2645 n = rcu_dereference(((struct fib6_info *)v)->fib6_next);
2646 if (n)
2647 return n;
2648
2649iter_table:
2650 ipv6_route_check_sernum(iter);
2651 spin_lock_bh(&iter->tbl->tb6_lock);
2652 r = fib6_walk_continue(&iter->w);
2653 spin_unlock_bh(&iter->tbl->tb6_lock);
2654 if (r > 0) {
2655 return iter->w.leaf;
2656 } else if (r < 0) {
2657 fib6_walker_unlink(net, &iter->w);
2658 return NULL;
2659 }
2660 fib6_walker_unlink(net, &iter->w);
2661
2662 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2663 if (!iter->tbl)
2664 return NULL;
2665
2666 ipv6_route_seq_setup_walk(iter, net);
2667 goto iter_table;
2668}
2669
2670static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2671 __acquires(RCU)
2672{
2673 struct net *net = seq_file_net(seq);
2674 struct ipv6_route_iter *iter = seq->private;
2675
2676 rcu_read_lock();
2677 iter->tbl = ipv6_route_seq_next_table(NULL, net);
2678 iter->skip = *pos;
2679
2680 if (iter->tbl) {
2681 loff_t p = 0;
2682
2683 ipv6_route_seq_setup_walk(iter, net);
2684 return ipv6_route_seq_next(seq, NULL, &p);
2685 } else {
2686 return NULL;
2687 }
2688}
2689
2690static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2691{
2692 struct fib6_walker *w = &iter->w;
2693 return w->node && !(w->state == FWS_U && w->node == w->root);
2694}
2695
2696static void ipv6_route_native_seq_stop(struct seq_file *seq, void *v)
2697 __releases(RCU)
2698{
2699 struct net *net = seq_file_net(seq);
2700 struct ipv6_route_iter *iter = seq->private;
2701
2702 if (ipv6_route_iter_active(iter))
2703 fib6_walker_unlink(net, &iter->w);
2704
2705 rcu_read_unlock();
2706}
2707
2708#if IS_BUILTIN(CONFIG_IPV6) && defined(CONFIG_BPF_SYSCALL)
2709static int ipv6_route_prog_seq_show(struct bpf_prog *prog,
2710 struct bpf_iter_meta *meta,
2711 void *v)
2712{
2713 struct bpf_iter__ipv6_route ctx;
2714
2715 ctx.meta = meta;
2716 ctx.rt = v;
2717 return bpf_iter_run_prog(prog, &ctx);
2718}
2719
2720static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2721{
2722 struct ipv6_route_iter *iter = seq->private;
2723 struct bpf_iter_meta meta;
2724 struct bpf_prog *prog;
2725 int ret;
2726
2727 meta.seq = seq;
2728 prog = bpf_iter_get_info(&meta, false);
2729 if (!prog)
2730 return ipv6_route_native_seq_show(seq, v);
2731
2732 ret = ipv6_route_prog_seq_show(prog, &meta, v);
2733 iter->w.leaf = NULL;
2734
2735 return ret;
2736}
2737
2738static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2739{
2740 struct bpf_iter_meta meta;
2741 struct bpf_prog *prog;
2742
2743 if (!v) {
2744 meta.seq = seq;
2745 prog = bpf_iter_get_info(&meta, true);
2746 if (prog)
2747 (void)ipv6_route_prog_seq_show(prog, &meta, v);
2748 }
2749
2750 ipv6_route_native_seq_stop(seq, v);
2751}
2752#else
2753static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2754{
2755 return ipv6_route_native_seq_show(seq, v);
2756}
2757
2758static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2759{
2760 ipv6_route_native_seq_stop(seq, v);
2761}
2762#endif
2763
2764const struct seq_operations ipv6_route_seq_ops = {
2765 .start = ipv6_route_seq_start,
2766 .next = ipv6_route_seq_next,
2767 .stop = ipv6_route_seq_stop,
2768 .show = ipv6_route_seq_show
2769};
2770#endif /* CONFIG_PROC_FS */
1/*
2 * Linux INET6 implementation
3 * Forwarding Information Database
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 *
13 * Changes:
14 * Yuji SEKIYA @USAGI: Support default route on router node;
15 * remove ip6_null_entry from the top of
16 * routing table.
17 * Ville Nuorvala: Fixed routing subtrees.
18 */
19
20#define pr_fmt(fmt) "IPv6: " fmt
21
22#include <linux/errno.h>
23#include <linux/types.h>
24#include <linux/net.h>
25#include <linux/route.h>
26#include <linux/netdevice.h>
27#include <linux/in6.h>
28#include <linux/init.h>
29#include <linux/list.h>
30#include <linux/slab.h>
31
32#include <net/ipv6.h>
33#include <net/ndisc.h>
34#include <net/addrconf.h>
35#include <net/lwtunnel.h>
36#include <net/fib_notifier.h>
37
38#include <net/ip6_fib.h>
39#include <net/ip6_route.h>
40
41static struct kmem_cache *fib6_node_kmem __read_mostly;
42
43struct fib6_cleaner {
44 struct fib6_walker w;
45 struct net *net;
46 int (*func)(struct rt6_info *, void *arg);
47 int sernum;
48 void *arg;
49};
50
51#ifdef CONFIG_IPV6_SUBTREES
52#define FWS_INIT FWS_S
53#else
54#define FWS_INIT FWS_L
55#endif
56
57static struct rt6_info *fib6_find_prefix(struct net *net,
58 struct fib6_table *table,
59 struct fib6_node *fn);
60static struct fib6_node *fib6_repair_tree(struct net *net,
61 struct fib6_table *table,
62 struct fib6_node *fn);
63static int fib6_walk(struct net *net, struct fib6_walker *w);
64static 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
73static 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
78static 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
85static 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
92static int fib6_new_sernum(struct net *net)
93{
94 int new, old;
95
96 do {
97 old = atomic_read(&net->ipv6.fib6_sernum);
98 new = old < INT_MAX ? old + 1 : 1;
99 } while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
100 old, new) != old);
101 return new;
102}
103
104enum {
105 FIB6_NO_SERNUM_CHANGE = 0,
106};
107
108void fib6_update_sernum(struct rt6_info *rt)
109{
110 struct net *net = dev_net(rt->dst.dev);
111 struct fib6_node *fn;
112
113 fn = rcu_dereference_protected(rt->rt6i_node,
114 lockdep_is_held(&rt->rt6i_table->tb6_lock));
115 if (fn)
116 fn->fn_sernum = fib6_new_sernum(net);
117}
118
119/*
120 * Auxiliary address test functions for the radix tree.
121 *
122 * These assume a 32bit processor (although it will work on
123 * 64bit processors)
124 */
125
126/*
127 * test bit
128 */
129#if defined(__LITTLE_ENDIAN)
130# define BITOP_BE32_SWIZZLE (0x1F & ~7)
131#else
132# define BITOP_BE32_SWIZZLE 0
133#endif
134
135static __be32 addr_bit_set(const void *token, int fn_bit)
136{
137 const __be32 *addr = token;
138 /*
139 * Here,
140 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
141 * is optimized version of
142 * htonl(1 << ((~fn_bit)&0x1F))
143 * See include/asm-generic/bitops/le.h.
144 */
145 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
146 addr[fn_bit >> 5];
147}
148
149static struct fib6_node *node_alloc(struct net *net)
150{
151 struct fib6_node *fn;
152
153 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
154 if (fn)
155 net->ipv6.rt6_stats->fib_nodes++;
156
157 return fn;
158}
159
160static void node_free_immediate(struct net *net, struct fib6_node *fn)
161{
162 kmem_cache_free(fib6_node_kmem, fn);
163 net->ipv6.rt6_stats->fib_nodes--;
164}
165
166static void node_free_rcu(struct rcu_head *head)
167{
168 struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
169
170 kmem_cache_free(fib6_node_kmem, fn);
171}
172
173static void node_free(struct net *net, struct fib6_node *fn)
174{
175 call_rcu(&fn->rcu, node_free_rcu);
176 net->ipv6.rt6_stats->fib_nodes--;
177}
178
179void rt6_free_pcpu(struct rt6_info *non_pcpu_rt)
180{
181 int cpu;
182
183 if (!non_pcpu_rt->rt6i_pcpu)
184 return;
185
186 for_each_possible_cpu(cpu) {
187 struct rt6_info **ppcpu_rt;
188 struct rt6_info *pcpu_rt;
189
190 ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
191 pcpu_rt = *ppcpu_rt;
192 if (pcpu_rt) {
193 dst_dev_put(&pcpu_rt->dst);
194 dst_release(&pcpu_rt->dst);
195 *ppcpu_rt = NULL;
196 }
197 }
198}
199EXPORT_SYMBOL_GPL(rt6_free_pcpu);
200
201static void fib6_free_table(struct fib6_table *table)
202{
203 inetpeer_invalidate_tree(&table->tb6_peers);
204 kfree(table);
205}
206
207static void fib6_link_table(struct net *net, struct fib6_table *tb)
208{
209 unsigned int h;
210
211 /*
212 * Initialize table lock at a single place to give lockdep a key,
213 * tables aren't visible prior to being linked to the list.
214 */
215 spin_lock_init(&tb->tb6_lock);
216 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
217
218 /*
219 * No protection necessary, this is the only list mutatation
220 * operation, tables never disappear once they exist.
221 */
222 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
223}
224
225#ifdef CONFIG_IPV6_MULTIPLE_TABLES
226
227static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
228{
229 struct fib6_table *table;
230
231 table = kzalloc(sizeof(*table), GFP_ATOMIC);
232 if (table) {
233 table->tb6_id = id;
234 rcu_assign_pointer(table->tb6_root.leaf,
235 net->ipv6.ip6_null_entry);
236 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
237 inet_peer_base_init(&table->tb6_peers);
238 }
239
240 return table;
241}
242
243struct fib6_table *fib6_new_table(struct net *net, u32 id)
244{
245 struct fib6_table *tb;
246
247 if (id == 0)
248 id = RT6_TABLE_MAIN;
249 tb = fib6_get_table(net, id);
250 if (tb)
251 return tb;
252
253 tb = fib6_alloc_table(net, id);
254 if (tb)
255 fib6_link_table(net, tb);
256
257 return tb;
258}
259EXPORT_SYMBOL_GPL(fib6_new_table);
260
261struct fib6_table *fib6_get_table(struct net *net, u32 id)
262{
263 struct fib6_table *tb;
264 struct hlist_head *head;
265 unsigned int h;
266
267 if (id == 0)
268 id = RT6_TABLE_MAIN;
269 h = id & (FIB6_TABLE_HASHSZ - 1);
270 rcu_read_lock();
271 head = &net->ipv6.fib_table_hash[h];
272 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
273 if (tb->tb6_id == id) {
274 rcu_read_unlock();
275 return tb;
276 }
277 }
278 rcu_read_unlock();
279
280 return NULL;
281}
282EXPORT_SYMBOL_GPL(fib6_get_table);
283
284static void __net_init fib6_tables_init(struct net *net)
285{
286 fib6_link_table(net, net->ipv6.fib6_main_tbl);
287 fib6_link_table(net, net->ipv6.fib6_local_tbl);
288}
289#else
290
291struct fib6_table *fib6_new_table(struct net *net, u32 id)
292{
293 return fib6_get_table(net, id);
294}
295
296struct fib6_table *fib6_get_table(struct net *net, u32 id)
297{
298 return net->ipv6.fib6_main_tbl;
299}
300
301struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
302 const struct sk_buff *skb,
303 int flags, pol_lookup_t lookup)
304{
305 struct rt6_info *rt;
306
307 rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
308 if (rt->dst.error == -EAGAIN) {
309 ip6_rt_put(rt);
310 rt = net->ipv6.ip6_null_entry;
311 dst_hold(&rt->dst);
312 }
313
314 return &rt->dst;
315}
316
317static void __net_init fib6_tables_init(struct net *net)
318{
319 fib6_link_table(net, net->ipv6.fib6_main_tbl);
320}
321
322#endif
323
324unsigned int fib6_tables_seq_read(struct net *net)
325{
326 unsigned int h, fib_seq = 0;
327
328 rcu_read_lock();
329 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
330 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
331 struct fib6_table *tb;
332
333 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
334 fib_seq += tb->fib_seq;
335 }
336 rcu_read_unlock();
337
338 return fib_seq;
339}
340
341static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
342 enum fib_event_type event_type,
343 struct rt6_info *rt)
344{
345 struct fib6_entry_notifier_info info = {
346 .rt = rt,
347 };
348
349 return call_fib6_notifier(nb, net, event_type, &info.info);
350}
351
352static int call_fib6_entry_notifiers(struct net *net,
353 enum fib_event_type event_type,
354 struct rt6_info *rt,
355 struct netlink_ext_ack *extack)
356{
357 struct fib6_entry_notifier_info info = {
358 .info.extack = extack,
359 .rt = rt,
360 };
361
362 rt->rt6i_table->fib_seq++;
363 return call_fib6_notifiers(net, event_type, &info.info);
364}
365
366struct fib6_dump_arg {
367 struct net *net;
368 struct notifier_block *nb;
369};
370
371static void fib6_rt_dump(struct rt6_info *rt, struct fib6_dump_arg *arg)
372{
373 if (rt == arg->net->ipv6.ip6_null_entry)
374 return;
375 call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
376}
377
378static int fib6_node_dump(struct fib6_walker *w)
379{
380 struct rt6_info *rt;
381
382 for_each_fib6_walker_rt(w)
383 fib6_rt_dump(rt, w->args);
384 w->leaf = NULL;
385 return 0;
386}
387
388static void fib6_table_dump(struct net *net, struct fib6_table *tb,
389 struct fib6_walker *w)
390{
391 w->root = &tb->tb6_root;
392 spin_lock_bh(&tb->tb6_lock);
393 fib6_walk(net, w);
394 spin_unlock_bh(&tb->tb6_lock);
395}
396
397/* Called with rcu_read_lock() */
398int fib6_tables_dump(struct net *net, struct notifier_block *nb)
399{
400 struct fib6_dump_arg arg;
401 struct fib6_walker *w;
402 unsigned int h;
403
404 w = kzalloc(sizeof(*w), GFP_ATOMIC);
405 if (!w)
406 return -ENOMEM;
407
408 w->func = fib6_node_dump;
409 arg.net = net;
410 arg.nb = nb;
411 w->args = &arg;
412
413 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
414 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
415 struct fib6_table *tb;
416
417 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
418 fib6_table_dump(net, tb, w);
419 }
420
421 kfree(w);
422
423 return 0;
424}
425
426static int fib6_dump_node(struct fib6_walker *w)
427{
428 int res;
429 struct rt6_info *rt;
430
431 for_each_fib6_walker_rt(w) {
432 res = rt6_dump_route(rt, w->args);
433 if (res < 0) {
434 /* Frame is full, suspend walking */
435 w->leaf = rt;
436 return 1;
437 }
438
439 /* Multipath routes are dumped in one route with the
440 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
441 * last sibling of this route (no need to dump the
442 * sibling routes again)
443 */
444 if (rt->rt6i_nsiblings)
445 rt = list_last_entry(&rt->rt6i_siblings,
446 struct rt6_info,
447 rt6i_siblings);
448 }
449 w->leaf = NULL;
450 return 0;
451}
452
453static void fib6_dump_end(struct netlink_callback *cb)
454{
455 struct net *net = sock_net(cb->skb->sk);
456 struct fib6_walker *w = (void *)cb->args[2];
457
458 if (w) {
459 if (cb->args[4]) {
460 cb->args[4] = 0;
461 fib6_walker_unlink(net, w);
462 }
463 cb->args[2] = 0;
464 kfree(w);
465 }
466 cb->done = (void *)cb->args[3];
467 cb->args[1] = 3;
468}
469
470static int fib6_dump_done(struct netlink_callback *cb)
471{
472 fib6_dump_end(cb);
473 return cb->done ? cb->done(cb) : 0;
474}
475
476static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
477 struct netlink_callback *cb)
478{
479 struct net *net = sock_net(skb->sk);
480 struct fib6_walker *w;
481 int res;
482
483 w = (void *)cb->args[2];
484 w->root = &table->tb6_root;
485
486 if (cb->args[4] == 0) {
487 w->count = 0;
488 w->skip = 0;
489
490 spin_lock_bh(&table->tb6_lock);
491 res = fib6_walk(net, w);
492 spin_unlock_bh(&table->tb6_lock);
493 if (res > 0) {
494 cb->args[4] = 1;
495 cb->args[5] = w->root->fn_sernum;
496 }
497 } else {
498 if (cb->args[5] != w->root->fn_sernum) {
499 /* Begin at the root if the tree changed */
500 cb->args[5] = w->root->fn_sernum;
501 w->state = FWS_INIT;
502 w->node = w->root;
503 w->skip = w->count;
504 } else
505 w->skip = 0;
506
507 spin_lock_bh(&table->tb6_lock);
508 res = fib6_walk_continue(w);
509 spin_unlock_bh(&table->tb6_lock);
510 if (res <= 0) {
511 fib6_walker_unlink(net, w);
512 cb->args[4] = 0;
513 }
514 }
515
516 return res;
517}
518
519static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
520{
521 struct net *net = sock_net(skb->sk);
522 unsigned int h, s_h;
523 unsigned int e = 0, s_e;
524 struct rt6_rtnl_dump_arg arg;
525 struct fib6_walker *w;
526 struct fib6_table *tb;
527 struct hlist_head *head;
528 int res = 0;
529
530 s_h = cb->args[0];
531 s_e = cb->args[1];
532
533 w = (void *)cb->args[2];
534 if (!w) {
535 /* New dump:
536 *
537 * 1. hook callback destructor.
538 */
539 cb->args[3] = (long)cb->done;
540 cb->done = fib6_dump_done;
541
542 /*
543 * 2. allocate and initialize walker.
544 */
545 w = kzalloc(sizeof(*w), GFP_ATOMIC);
546 if (!w)
547 return -ENOMEM;
548 w->func = fib6_dump_node;
549 cb->args[2] = (long)w;
550 }
551
552 arg.skb = skb;
553 arg.cb = cb;
554 arg.net = net;
555 w->args = &arg;
556
557 rcu_read_lock();
558 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
559 e = 0;
560 head = &net->ipv6.fib_table_hash[h];
561 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
562 if (e < s_e)
563 goto next;
564 res = fib6_dump_table(tb, skb, cb);
565 if (res != 0)
566 goto out;
567next:
568 e++;
569 }
570 }
571out:
572 rcu_read_unlock();
573 cb->args[1] = e;
574 cb->args[0] = h;
575
576 res = res < 0 ? res : skb->len;
577 if (res <= 0)
578 fib6_dump_end(cb);
579 return res;
580}
581
582/*
583 * Routing Table
584 *
585 * return the appropriate node for a routing tree "add" operation
586 * by either creating and inserting or by returning an existing
587 * node.
588 */
589
590static struct fib6_node *fib6_add_1(struct net *net,
591 struct fib6_table *table,
592 struct fib6_node *root,
593 struct in6_addr *addr, int plen,
594 int offset, int allow_create,
595 int replace_required,
596 struct netlink_ext_ack *extack)
597{
598 struct fib6_node *fn, *in, *ln;
599 struct fib6_node *pn = NULL;
600 struct rt6key *key;
601 int bit;
602 __be32 dir = 0;
603
604 RT6_TRACE("fib6_add_1\n");
605
606 /* insert node in tree */
607
608 fn = root;
609
610 do {
611 struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
612 lockdep_is_held(&table->tb6_lock));
613 key = (struct rt6key *)((u8 *)leaf + offset);
614
615 /*
616 * Prefix match
617 */
618 if (plen < fn->fn_bit ||
619 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
620 if (!allow_create) {
621 if (replace_required) {
622 NL_SET_ERR_MSG(extack,
623 "Can not replace route - no match found");
624 pr_warn("Can't replace route, no match found\n");
625 return ERR_PTR(-ENOENT);
626 }
627 pr_warn("NLM_F_CREATE should be set when creating new route\n");
628 }
629 goto insert_above;
630 }
631
632 /*
633 * Exact match ?
634 */
635
636 if (plen == fn->fn_bit) {
637 /* clean up an intermediate node */
638 if (!(fn->fn_flags & RTN_RTINFO)) {
639 RCU_INIT_POINTER(fn->leaf, NULL);
640 rt6_release(leaf);
641 /* remove null_entry in the root node */
642 } else if (fn->fn_flags & RTN_TL_ROOT &&
643 rcu_access_pointer(fn->leaf) ==
644 net->ipv6.ip6_null_entry) {
645 RCU_INIT_POINTER(fn->leaf, NULL);
646 }
647
648 return fn;
649 }
650
651 /*
652 * We have more bits to go
653 */
654
655 /* Try to walk down on tree. */
656 dir = addr_bit_set(addr, fn->fn_bit);
657 pn = fn;
658 fn = dir ?
659 rcu_dereference_protected(fn->right,
660 lockdep_is_held(&table->tb6_lock)) :
661 rcu_dereference_protected(fn->left,
662 lockdep_is_held(&table->tb6_lock));
663 } while (fn);
664
665 if (!allow_create) {
666 /* We should not create new node because
667 * NLM_F_REPLACE was specified without NLM_F_CREATE
668 * I assume it is safe to require NLM_F_CREATE when
669 * REPLACE flag is used! Later we may want to remove the
670 * check for replace_required, because according
671 * to netlink specification, NLM_F_CREATE
672 * MUST be specified if new route is created.
673 * That would keep IPv6 consistent with IPv4
674 */
675 if (replace_required) {
676 NL_SET_ERR_MSG(extack,
677 "Can not replace route - no match found");
678 pr_warn("Can't replace route, no match found\n");
679 return ERR_PTR(-ENOENT);
680 }
681 pr_warn("NLM_F_CREATE should be set when creating new route\n");
682 }
683 /*
684 * We walked to the bottom of tree.
685 * Create new leaf node without children.
686 */
687
688 ln = node_alloc(net);
689
690 if (!ln)
691 return ERR_PTR(-ENOMEM);
692 ln->fn_bit = plen;
693 RCU_INIT_POINTER(ln->parent, pn);
694
695 if (dir)
696 rcu_assign_pointer(pn->right, ln);
697 else
698 rcu_assign_pointer(pn->left, ln);
699
700 return ln;
701
702
703insert_above:
704 /*
705 * split since we don't have a common prefix anymore or
706 * we have a less significant route.
707 * we've to insert an intermediate node on the list
708 * this new node will point to the one we need to create
709 * and the current
710 */
711
712 pn = rcu_dereference_protected(fn->parent,
713 lockdep_is_held(&table->tb6_lock));
714
715 /* find 1st bit in difference between the 2 addrs.
716
717 See comment in __ipv6_addr_diff: bit may be an invalid value,
718 but if it is >= plen, the value is ignored in any case.
719 */
720
721 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
722
723 /*
724 * (intermediate)[in]
725 * / \
726 * (new leaf node)[ln] (old node)[fn]
727 */
728 if (plen > bit) {
729 in = node_alloc(net);
730 ln = node_alloc(net);
731
732 if (!in || !ln) {
733 if (in)
734 node_free_immediate(net, in);
735 if (ln)
736 node_free_immediate(net, ln);
737 return ERR_PTR(-ENOMEM);
738 }
739
740 /*
741 * new intermediate node.
742 * RTN_RTINFO will
743 * be off since that an address that chooses one of
744 * the branches would not match less specific routes
745 * in the other branch
746 */
747
748 in->fn_bit = bit;
749
750 RCU_INIT_POINTER(in->parent, pn);
751 in->leaf = fn->leaf;
752 atomic_inc(&rcu_dereference_protected(in->leaf,
753 lockdep_is_held(&table->tb6_lock))->rt6i_ref);
754
755 /* update parent pointer */
756 if (dir)
757 rcu_assign_pointer(pn->right, in);
758 else
759 rcu_assign_pointer(pn->left, in);
760
761 ln->fn_bit = plen;
762
763 RCU_INIT_POINTER(ln->parent, in);
764 rcu_assign_pointer(fn->parent, in);
765
766 if (addr_bit_set(addr, bit)) {
767 rcu_assign_pointer(in->right, ln);
768 rcu_assign_pointer(in->left, fn);
769 } else {
770 rcu_assign_pointer(in->left, ln);
771 rcu_assign_pointer(in->right, fn);
772 }
773 } else { /* plen <= bit */
774
775 /*
776 * (new leaf node)[ln]
777 * / \
778 * (old node)[fn] NULL
779 */
780
781 ln = node_alloc(net);
782
783 if (!ln)
784 return ERR_PTR(-ENOMEM);
785
786 ln->fn_bit = plen;
787
788 RCU_INIT_POINTER(ln->parent, pn);
789
790 if (addr_bit_set(&key->addr, plen))
791 RCU_INIT_POINTER(ln->right, fn);
792 else
793 RCU_INIT_POINTER(ln->left, fn);
794
795 rcu_assign_pointer(fn->parent, ln);
796
797 if (dir)
798 rcu_assign_pointer(pn->right, ln);
799 else
800 rcu_assign_pointer(pn->left, ln);
801 }
802 return ln;
803}
804
805static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
806{
807 int i;
808
809 for (i = 0; i < RTAX_MAX; i++) {
810 if (test_bit(i, mxc->mx_valid))
811 mp[i] = mxc->mx[i];
812 }
813}
814
815static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
816{
817 if (!mxc->mx)
818 return 0;
819
820 if (dst->flags & DST_HOST) {
821 u32 *mp = dst_metrics_write_ptr(dst);
822
823 if (unlikely(!mp))
824 return -ENOMEM;
825
826 fib6_copy_metrics(mp, mxc);
827 } else {
828 dst_init_metrics(dst, mxc->mx, false);
829
830 /* We've stolen mx now. */
831 mxc->mx = NULL;
832 }
833
834 return 0;
835}
836
837static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
838 struct net *net)
839{
840 struct fib6_table *table = rt->rt6i_table;
841
842 if (atomic_read(&rt->rt6i_ref) != 1) {
843 /* This route is used as dummy address holder in some split
844 * nodes. It is not leaked, but it still holds other resources,
845 * which must be released in time. So, scan ascendant nodes
846 * and replace dummy references to this route with references
847 * to still alive ones.
848 */
849 while (fn) {
850 struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
851 lockdep_is_held(&table->tb6_lock));
852 struct rt6_info *new_leaf;
853 if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
854 new_leaf = fib6_find_prefix(net, table, fn);
855 atomic_inc(&new_leaf->rt6i_ref);
856 rcu_assign_pointer(fn->leaf, new_leaf);
857 rt6_release(rt);
858 }
859 fn = rcu_dereference_protected(fn->parent,
860 lockdep_is_held(&table->tb6_lock));
861 }
862 }
863}
864
865/*
866 * Insert routing information in a node.
867 */
868
869static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
870 struct nl_info *info, struct mx6_config *mxc,
871 struct netlink_ext_ack *extack)
872{
873 struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
874 lockdep_is_held(&rt->rt6i_table->tb6_lock));
875 struct rt6_info *iter = NULL;
876 struct rt6_info __rcu **ins;
877 struct rt6_info __rcu **fallback_ins = NULL;
878 int replace = (info->nlh &&
879 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
880 int add = (!info->nlh ||
881 (info->nlh->nlmsg_flags & NLM_F_CREATE));
882 int found = 0;
883 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
884 u16 nlflags = NLM_F_EXCL;
885 int err;
886
887 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
888 nlflags |= NLM_F_APPEND;
889
890 ins = &fn->leaf;
891
892 for (iter = leaf; iter;
893 iter = rcu_dereference_protected(iter->rt6_next,
894 lockdep_is_held(&rt->rt6i_table->tb6_lock))) {
895 /*
896 * Search for duplicates
897 */
898
899 if (iter->rt6i_metric == rt->rt6i_metric) {
900 /*
901 * Same priority level
902 */
903 if (info->nlh &&
904 (info->nlh->nlmsg_flags & NLM_F_EXCL))
905 return -EEXIST;
906
907 nlflags &= ~NLM_F_EXCL;
908 if (replace) {
909 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
910 found++;
911 break;
912 }
913 if (rt_can_ecmp)
914 fallback_ins = fallback_ins ?: ins;
915 goto next_iter;
916 }
917
918 if (rt6_duplicate_nexthop(iter, rt)) {
919 if (rt->rt6i_nsiblings)
920 rt->rt6i_nsiblings = 0;
921 if (!(iter->rt6i_flags & RTF_EXPIRES))
922 return -EEXIST;
923 if (!(rt->rt6i_flags & RTF_EXPIRES))
924 rt6_clean_expires(iter);
925 else
926 rt6_set_expires(iter, rt->dst.expires);
927 iter->rt6i_pmtu = rt->rt6i_pmtu;
928 return -EEXIST;
929 }
930 /* If we have the same destination and the same metric,
931 * but not the same gateway, then the route we try to
932 * add is sibling to this route, increment our counter
933 * of siblings, and later we will add our route to the
934 * list.
935 * Only static routes (which don't have flag
936 * RTF_EXPIRES) are used for ECMPv6.
937 *
938 * To avoid long list, we only had siblings if the
939 * route have a gateway.
940 */
941 if (rt_can_ecmp &&
942 rt6_qualify_for_ecmp(iter))
943 rt->rt6i_nsiblings++;
944 }
945
946 if (iter->rt6i_metric > rt->rt6i_metric)
947 break;
948
949next_iter:
950 ins = &iter->rt6_next;
951 }
952
953 if (fallback_ins && !found) {
954 /* No ECMP-able route found, replace first non-ECMP one */
955 ins = fallback_ins;
956 iter = rcu_dereference_protected(*ins,
957 lockdep_is_held(&rt->rt6i_table->tb6_lock));
958 found++;
959 }
960
961 /* Reset round-robin state, if necessary */
962 if (ins == &fn->leaf)
963 fn->rr_ptr = NULL;
964
965 /* Link this route to others same route. */
966 if (rt->rt6i_nsiblings) {
967 unsigned int rt6i_nsiblings;
968 struct rt6_info *sibling, *temp_sibling;
969
970 /* Find the first route that have the same metric */
971 sibling = leaf;
972 while (sibling) {
973 if (sibling->rt6i_metric == rt->rt6i_metric &&
974 rt6_qualify_for_ecmp(sibling)) {
975 list_add_tail(&rt->rt6i_siblings,
976 &sibling->rt6i_siblings);
977 break;
978 }
979 sibling = rcu_dereference_protected(sibling->rt6_next,
980 lockdep_is_held(&rt->rt6i_table->tb6_lock));
981 }
982 /* For each sibling in the list, increment the counter of
983 * siblings. BUG() if counters does not match, list of siblings
984 * is broken!
985 */
986 rt6i_nsiblings = 0;
987 list_for_each_entry_safe(sibling, temp_sibling,
988 &rt->rt6i_siblings, rt6i_siblings) {
989 sibling->rt6i_nsiblings++;
990 BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
991 rt6i_nsiblings++;
992 }
993 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
994 rt6_multipath_rebalance(temp_sibling);
995 }
996
997 /*
998 * insert node
999 */
1000 if (!replace) {
1001 if (!add)
1002 pr_warn("NLM_F_CREATE should be set when creating new route\n");
1003
1004add:
1005 nlflags |= NLM_F_CREATE;
1006 err = fib6_commit_metrics(&rt->dst, mxc);
1007 if (err)
1008 return err;
1009
1010 err = call_fib6_entry_notifiers(info->nl_net,
1011 FIB_EVENT_ENTRY_ADD,
1012 rt, extack);
1013 if (err)
1014 return err;
1015
1016 rcu_assign_pointer(rt->rt6_next, iter);
1017 atomic_inc(&rt->rt6i_ref);
1018 rcu_assign_pointer(rt->rt6i_node, fn);
1019 rcu_assign_pointer(*ins, rt);
1020 if (!info->skip_notify)
1021 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1022 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1023
1024 if (!(fn->fn_flags & RTN_RTINFO)) {
1025 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1026 fn->fn_flags |= RTN_RTINFO;
1027 }
1028
1029 } else {
1030 int nsiblings;
1031
1032 if (!found) {
1033 if (add)
1034 goto add;
1035 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1036 return -ENOENT;
1037 }
1038
1039 err = fib6_commit_metrics(&rt->dst, mxc);
1040 if (err)
1041 return err;
1042
1043 err = call_fib6_entry_notifiers(info->nl_net,
1044 FIB_EVENT_ENTRY_REPLACE,
1045 rt, extack);
1046 if (err)
1047 return err;
1048
1049 atomic_inc(&rt->rt6i_ref);
1050 rcu_assign_pointer(rt->rt6i_node, fn);
1051 rt->rt6_next = iter->rt6_next;
1052 rcu_assign_pointer(*ins, rt);
1053 if (!info->skip_notify)
1054 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1055 if (!(fn->fn_flags & RTN_RTINFO)) {
1056 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1057 fn->fn_flags |= RTN_RTINFO;
1058 }
1059 nsiblings = iter->rt6i_nsiblings;
1060 iter->rt6i_node = NULL;
1061 fib6_purge_rt(iter, fn, info->nl_net);
1062 if (rcu_access_pointer(fn->rr_ptr) == iter)
1063 fn->rr_ptr = NULL;
1064 rt6_release(iter);
1065
1066 if (nsiblings) {
1067 /* Replacing an ECMP route, remove all siblings */
1068 ins = &rt->rt6_next;
1069 iter = rcu_dereference_protected(*ins,
1070 lockdep_is_held(&rt->rt6i_table->tb6_lock));
1071 while (iter) {
1072 if (iter->rt6i_metric > rt->rt6i_metric)
1073 break;
1074 if (rt6_qualify_for_ecmp(iter)) {
1075 *ins = iter->rt6_next;
1076 iter->rt6i_node = NULL;
1077 fib6_purge_rt(iter, fn, info->nl_net);
1078 if (rcu_access_pointer(fn->rr_ptr) == iter)
1079 fn->rr_ptr = NULL;
1080 rt6_release(iter);
1081 nsiblings--;
1082 info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1083 } else {
1084 ins = &iter->rt6_next;
1085 }
1086 iter = rcu_dereference_protected(*ins,
1087 lockdep_is_held(&rt->rt6i_table->tb6_lock));
1088 }
1089 WARN_ON(nsiblings != 0);
1090 }
1091 }
1092
1093 return 0;
1094}
1095
1096static void fib6_start_gc(struct net *net, struct rt6_info *rt)
1097{
1098 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1099 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
1100 mod_timer(&net->ipv6.ip6_fib_timer,
1101 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1102}
1103
1104void fib6_force_start_gc(struct net *net)
1105{
1106 if (!timer_pending(&net->ipv6.ip6_fib_timer))
1107 mod_timer(&net->ipv6.ip6_fib_timer,
1108 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1109}
1110
1111static void __fib6_update_sernum_upto_root(struct rt6_info *rt,
1112 int sernum)
1113{
1114 struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1115 lockdep_is_held(&rt->rt6i_table->tb6_lock));
1116
1117 /* paired with smp_rmb() in rt6_get_cookie_safe() */
1118 smp_wmb();
1119 while (fn) {
1120 fn->fn_sernum = sernum;
1121 fn = rcu_dereference_protected(fn->parent,
1122 lockdep_is_held(&rt->rt6i_table->tb6_lock));
1123 }
1124}
1125
1126void fib6_update_sernum_upto_root(struct net *net, struct rt6_info *rt)
1127{
1128 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1129}
1130
1131/*
1132 * Add routing information to the routing tree.
1133 * <destination addr>/<source addr>
1134 * with source addr info in sub-trees
1135 * Need to own table->tb6_lock
1136 */
1137
1138int fib6_add(struct fib6_node *root, struct rt6_info *rt,
1139 struct nl_info *info, struct mx6_config *mxc,
1140 struct netlink_ext_ack *extack)
1141{
1142 struct fib6_table *table = rt->rt6i_table;
1143 struct fib6_node *fn, *pn = NULL;
1144 int err = -ENOMEM;
1145 int allow_create = 1;
1146 int replace_required = 0;
1147 int sernum = fib6_new_sernum(info->nl_net);
1148
1149 if (WARN_ON_ONCE(!atomic_read(&rt->dst.__refcnt)))
1150 return -EINVAL;
1151 if (WARN_ON_ONCE(rt->rt6i_flags & RTF_CACHE))
1152 return -EINVAL;
1153
1154 if (info->nlh) {
1155 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1156 allow_create = 0;
1157 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1158 replace_required = 1;
1159 }
1160 if (!allow_create && !replace_required)
1161 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1162
1163 fn = fib6_add_1(info->nl_net, table, root,
1164 &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
1165 offsetof(struct rt6_info, rt6i_dst), allow_create,
1166 replace_required, extack);
1167 if (IS_ERR(fn)) {
1168 err = PTR_ERR(fn);
1169 fn = NULL;
1170 goto out;
1171 }
1172
1173 pn = fn;
1174
1175#ifdef CONFIG_IPV6_SUBTREES
1176 if (rt->rt6i_src.plen) {
1177 struct fib6_node *sn;
1178
1179 if (!rcu_access_pointer(fn->subtree)) {
1180 struct fib6_node *sfn;
1181
1182 /*
1183 * Create subtree.
1184 *
1185 * fn[main tree]
1186 * |
1187 * sfn[subtree root]
1188 * \
1189 * sn[new leaf node]
1190 */
1191
1192 /* Create subtree root node */
1193 sfn = node_alloc(info->nl_net);
1194 if (!sfn)
1195 goto failure;
1196
1197 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
1198 rcu_assign_pointer(sfn->leaf,
1199 info->nl_net->ipv6.ip6_null_entry);
1200 sfn->fn_flags = RTN_ROOT;
1201
1202 /* Now add the first leaf node to new subtree */
1203
1204 sn = fib6_add_1(info->nl_net, table, sfn,
1205 &rt->rt6i_src.addr, rt->rt6i_src.plen,
1206 offsetof(struct rt6_info, rt6i_src),
1207 allow_create, replace_required, extack);
1208
1209 if (IS_ERR(sn)) {
1210 /* If it is failed, discard just allocated
1211 root, and then (in failure) stale node
1212 in main tree.
1213 */
1214 node_free_immediate(info->nl_net, sfn);
1215 err = PTR_ERR(sn);
1216 goto failure;
1217 }
1218
1219 /* Now link new subtree to main tree */
1220 rcu_assign_pointer(sfn->parent, fn);
1221 rcu_assign_pointer(fn->subtree, sfn);
1222 } else {
1223 sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1224 &rt->rt6i_src.addr, rt->rt6i_src.plen,
1225 offsetof(struct rt6_info, rt6i_src),
1226 allow_create, replace_required, extack);
1227
1228 if (IS_ERR(sn)) {
1229 err = PTR_ERR(sn);
1230 goto failure;
1231 }
1232 }
1233
1234 if (!rcu_access_pointer(fn->leaf)) {
1235 if (fn->fn_flags & RTN_TL_ROOT) {
1236 /* put back null_entry for root node */
1237 rcu_assign_pointer(fn->leaf,
1238 info->nl_net->ipv6.ip6_null_entry);
1239 } else {
1240 atomic_inc(&rt->rt6i_ref);
1241 rcu_assign_pointer(fn->leaf, rt);
1242 }
1243 }
1244 fn = sn;
1245 }
1246#endif
1247
1248 err = fib6_add_rt2node(fn, rt, info, mxc, extack);
1249 if (!err) {
1250 __fib6_update_sernum_upto_root(rt, sernum);
1251 fib6_start_gc(info->nl_net, rt);
1252 }
1253
1254out:
1255 if (err) {
1256#ifdef CONFIG_IPV6_SUBTREES
1257 /*
1258 * If fib6_add_1 has cleared the old leaf pointer in the
1259 * super-tree leaf node we have to find a new one for it.
1260 */
1261 if (pn != fn) {
1262 struct rt6_info *pn_leaf =
1263 rcu_dereference_protected(pn->leaf,
1264 lockdep_is_held(&table->tb6_lock));
1265 if (pn_leaf == rt) {
1266 pn_leaf = NULL;
1267 RCU_INIT_POINTER(pn->leaf, NULL);
1268 atomic_dec(&rt->rt6i_ref);
1269 }
1270 if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1271 pn_leaf = fib6_find_prefix(info->nl_net, table,
1272 pn);
1273#if RT6_DEBUG >= 2
1274 if (!pn_leaf) {
1275 WARN_ON(!pn_leaf);
1276 pn_leaf =
1277 info->nl_net->ipv6.ip6_null_entry;
1278 }
1279#endif
1280 atomic_inc(&pn_leaf->rt6i_ref);
1281 rcu_assign_pointer(pn->leaf, pn_leaf);
1282 }
1283 }
1284#endif
1285 goto failure;
1286 }
1287 return err;
1288
1289failure:
1290 /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1291 * 1. fn is an intermediate node and we failed to add the new
1292 * route to it in both subtree creation failure and fib6_add_rt2node()
1293 * failure case.
1294 * 2. fn is the root node in the table and we fail to add the first
1295 * default route to it.
1296 */
1297 if (fn &&
1298 (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1299 (fn->fn_flags & RTN_TL_ROOT &&
1300 !rcu_access_pointer(fn->leaf))))
1301 fib6_repair_tree(info->nl_net, table, fn);
1302 /* Always release dst as dst->__refcnt is guaranteed
1303 * to be taken before entering this function
1304 */
1305 dst_release_immediate(&rt->dst);
1306 return err;
1307}
1308
1309/*
1310 * Routing tree lookup
1311 *
1312 */
1313
1314struct lookup_args {
1315 int offset; /* key offset on rt6_info */
1316 const struct in6_addr *addr; /* search key */
1317};
1318
1319static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1320 struct lookup_args *args)
1321{
1322 struct fib6_node *fn;
1323 __be32 dir;
1324
1325 if (unlikely(args->offset == 0))
1326 return NULL;
1327
1328 /*
1329 * Descend on a tree
1330 */
1331
1332 fn = root;
1333
1334 for (;;) {
1335 struct fib6_node *next;
1336
1337 dir = addr_bit_set(args->addr, fn->fn_bit);
1338
1339 next = dir ? rcu_dereference(fn->right) :
1340 rcu_dereference(fn->left);
1341
1342 if (next) {
1343 fn = next;
1344 continue;
1345 }
1346 break;
1347 }
1348
1349 while (fn) {
1350 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1351
1352 if (subtree || fn->fn_flags & RTN_RTINFO) {
1353 struct rt6_info *leaf = rcu_dereference(fn->leaf);
1354 struct rt6key *key;
1355
1356 if (!leaf)
1357 goto backtrack;
1358
1359 key = (struct rt6key *) ((u8 *)leaf + args->offset);
1360
1361 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1362#ifdef CONFIG_IPV6_SUBTREES
1363 if (subtree) {
1364 struct fib6_node *sfn;
1365 sfn = fib6_lookup_1(subtree, args + 1);
1366 if (!sfn)
1367 goto backtrack;
1368 fn = sfn;
1369 }
1370#endif
1371 if (fn->fn_flags & RTN_RTINFO)
1372 return fn;
1373 }
1374 }
1375backtrack:
1376 if (fn->fn_flags & RTN_ROOT)
1377 break;
1378
1379 fn = rcu_dereference(fn->parent);
1380 }
1381
1382 return NULL;
1383}
1384
1385/* called with rcu_read_lock() held
1386 */
1387struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1388 const struct in6_addr *saddr)
1389{
1390 struct fib6_node *fn;
1391 struct lookup_args args[] = {
1392 {
1393 .offset = offsetof(struct rt6_info, rt6i_dst),
1394 .addr = daddr,
1395 },
1396#ifdef CONFIG_IPV6_SUBTREES
1397 {
1398 .offset = offsetof(struct rt6_info, rt6i_src),
1399 .addr = saddr,
1400 },
1401#endif
1402 {
1403 .offset = 0, /* sentinel */
1404 }
1405 };
1406
1407 fn = fib6_lookup_1(root, daddr ? args : args + 1);
1408 if (!fn || fn->fn_flags & RTN_TL_ROOT)
1409 fn = root;
1410
1411 return fn;
1412}
1413
1414/*
1415 * Get node with specified destination prefix (and source prefix,
1416 * if subtrees are used)
1417 * exact_match == true means we try to find fn with exact match of
1418 * the passed in prefix addr
1419 * exact_match == false means we try to find fn with longest prefix
1420 * match of the passed in prefix addr. This is useful for finding fn
1421 * for cached route as it will be stored in the exception table under
1422 * the node with longest prefix length.
1423 */
1424
1425
1426static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1427 const struct in6_addr *addr,
1428 int plen, int offset,
1429 bool exact_match)
1430{
1431 struct fib6_node *fn, *prev = NULL;
1432
1433 for (fn = root; fn ; ) {
1434 struct rt6_info *leaf = rcu_dereference(fn->leaf);
1435 struct rt6key *key;
1436
1437 /* This node is being deleted */
1438 if (!leaf) {
1439 if (plen <= fn->fn_bit)
1440 goto out;
1441 else
1442 goto next;
1443 }
1444
1445 key = (struct rt6key *)((u8 *)leaf + offset);
1446
1447 /*
1448 * Prefix match
1449 */
1450 if (plen < fn->fn_bit ||
1451 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1452 goto out;
1453
1454 if (plen == fn->fn_bit)
1455 return fn;
1456
1457 prev = fn;
1458
1459next:
1460 /*
1461 * We have more bits to go
1462 */
1463 if (addr_bit_set(addr, fn->fn_bit))
1464 fn = rcu_dereference(fn->right);
1465 else
1466 fn = rcu_dereference(fn->left);
1467 }
1468out:
1469 if (exact_match)
1470 return NULL;
1471 else
1472 return prev;
1473}
1474
1475struct fib6_node *fib6_locate(struct fib6_node *root,
1476 const struct in6_addr *daddr, int dst_len,
1477 const struct in6_addr *saddr, int src_len,
1478 bool exact_match)
1479{
1480 struct fib6_node *fn;
1481
1482 fn = fib6_locate_1(root, daddr, dst_len,
1483 offsetof(struct rt6_info, rt6i_dst),
1484 exact_match);
1485
1486#ifdef CONFIG_IPV6_SUBTREES
1487 if (src_len) {
1488 WARN_ON(saddr == NULL);
1489 if (fn) {
1490 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1491
1492 if (subtree) {
1493 fn = fib6_locate_1(subtree, saddr, src_len,
1494 offsetof(struct rt6_info, rt6i_src),
1495 exact_match);
1496 }
1497 }
1498 }
1499#endif
1500
1501 if (fn && fn->fn_flags & RTN_RTINFO)
1502 return fn;
1503
1504 return NULL;
1505}
1506
1507
1508/*
1509 * Deletion
1510 *
1511 */
1512
1513static struct rt6_info *fib6_find_prefix(struct net *net,
1514 struct fib6_table *table,
1515 struct fib6_node *fn)
1516{
1517 struct fib6_node *child_left, *child_right;
1518
1519 if (fn->fn_flags & RTN_ROOT)
1520 return net->ipv6.ip6_null_entry;
1521
1522 while (fn) {
1523 child_left = rcu_dereference_protected(fn->left,
1524 lockdep_is_held(&table->tb6_lock));
1525 child_right = rcu_dereference_protected(fn->right,
1526 lockdep_is_held(&table->tb6_lock));
1527 if (child_left)
1528 return rcu_dereference_protected(child_left->leaf,
1529 lockdep_is_held(&table->tb6_lock));
1530 if (child_right)
1531 return rcu_dereference_protected(child_right->leaf,
1532 lockdep_is_held(&table->tb6_lock));
1533
1534 fn = FIB6_SUBTREE(fn);
1535 }
1536 return NULL;
1537}
1538
1539/*
1540 * Called to trim the tree of intermediate nodes when possible. "fn"
1541 * is the node we want to try and remove.
1542 * Need to own table->tb6_lock
1543 */
1544
1545static struct fib6_node *fib6_repair_tree(struct net *net,
1546 struct fib6_table *table,
1547 struct fib6_node *fn)
1548{
1549 int children;
1550 int nstate;
1551 struct fib6_node *child;
1552 struct fib6_walker *w;
1553 int iter = 0;
1554
1555 /* Set fn->leaf to null_entry for root node. */
1556 if (fn->fn_flags & RTN_TL_ROOT) {
1557 rcu_assign_pointer(fn->leaf, net->ipv6.ip6_null_entry);
1558 return fn;
1559 }
1560
1561 for (;;) {
1562 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1563 lockdep_is_held(&table->tb6_lock));
1564 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1565 lockdep_is_held(&table->tb6_lock));
1566 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1567 lockdep_is_held(&table->tb6_lock));
1568 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1569 lockdep_is_held(&table->tb6_lock));
1570 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1571 lockdep_is_held(&table->tb6_lock));
1572 struct rt6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1573 lockdep_is_held(&table->tb6_lock));
1574 struct rt6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1575 lockdep_is_held(&table->tb6_lock));
1576 struct rt6_info *new_fn_leaf;
1577
1578 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1579 iter++;
1580
1581 WARN_ON(fn->fn_flags & RTN_RTINFO);
1582 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1583 WARN_ON(fn_leaf);
1584
1585 children = 0;
1586 child = NULL;
1587 if (fn_r)
1588 child = fn_r, children |= 1;
1589 if (fn_l)
1590 child = fn_l, children |= 2;
1591
1592 if (children == 3 || FIB6_SUBTREE(fn)
1593#ifdef CONFIG_IPV6_SUBTREES
1594 /* Subtree root (i.e. fn) may have one child */
1595 || (children && fn->fn_flags & RTN_ROOT)
1596#endif
1597 ) {
1598 new_fn_leaf = fib6_find_prefix(net, table, fn);
1599#if RT6_DEBUG >= 2
1600 if (!new_fn_leaf) {
1601 WARN_ON(!new_fn_leaf);
1602 new_fn_leaf = net->ipv6.ip6_null_entry;
1603 }
1604#endif
1605 atomic_inc(&new_fn_leaf->rt6i_ref);
1606 rcu_assign_pointer(fn->leaf, new_fn_leaf);
1607 return pn;
1608 }
1609
1610#ifdef CONFIG_IPV6_SUBTREES
1611 if (FIB6_SUBTREE(pn) == fn) {
1612 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1613 RCU_INIT_POINTER(pn->subtree, NULL);
1614 nstate = FWS_L;
1615 } else {
1616 WARN_ON(fn->fn_flags & RTN_ROOT);
1617#endif
1618 if (pn_r == fn)
1619 rcu_assign_pointer(pn->right, child);
1620 else if (pn_l == fn)
1621 rcu_assign_pointer(pn->left, child);
1622#if RT6_DEBUG >= 2
1623 else
1624 WARN_ON(1);
1625#endif
1626 if (child)
1627 rcu_assign_pointer(child->parent, pn);
1628 nstate = FWS_R;
1629#ifdef CONFIG_IPV6_SUBTREES
1630 }
1631#endif
1632
1633 read_lock(&net->ipv6.fib6_walker_lock);
1634 FOR_WALKERS(net, w) {
1635 if (!child) {
1636 if (w->node == fn) {
1637 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1638 w->node = pn;
1639 w->state = nstate;
1640 }
1641 } else {
1642 if (w->node == fn) {
1643 w->node = child;
1644 if (children&2) {
1645 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1646 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1647 } else {
1648 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1649 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1650 }
1651 }
1652 }
1653 }
1654 read_unlock(&net->ipv6.fib6_walker_lock);
1655
1656 node_free(net, fn);
1657 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1658 return pn;
1659
1660 RCU_INIT_POINTER(pn->leaf, NULL);
1661 rt6_release(pn_leaf);
1662 fn = pn;
1663 }
1664}
1665
1666static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1667 struct rt6_info __rcu **rtp, struct nl_info *info)
1668{
1669 struct fib6_walker *w;
1670 struct rt6_info *rt = rcu_dereference_protected(*rtp,
1671 lockdep_is_held(&table->tb6_lock));
1672 struct net *net = info->nl_net;
1673
1674 RT6_TRACE("fib6_del_route\n");
1675
1676 WARN_ON_ONCE(rt->rt6i_flags & RTF_CACHE);
1677
1678 /* Unlink it */
1679 *rtp = rt->rt6_next;
1680 rt->rt6i_node = NULL;
1681 net->ipv6.rt6_stats->fib_rt_entries--;
1682 net->ipv6.rt6_stats->fib_discarded_routes++;
1683
1684 /* Flush all cached dst in exception table */
1685 rt6_flush_exceptions(rt);
1686
1687 /* Reset round-robin state, if necessary */
1688 if (rcu_access_pointer(fn->rr_ptr) == rt)
1689 fn->rr_ptr = NULL;
1690
1691 /* Remove this entry from other siblings */
1692 if (rt->rt6i_nsiblings) {
1693 struct rt6_info *sibling, *next_sibling;
1694
1695 list_for_each_entry_safe(sibling, next_sibling,
1696 &rt->rt6i_siblings, rt6i_siblings)
1697 sibling->rt6i_nsiblings--;
1698 rt->rt6i_nsiblings = 0;
1699 list_del_init(&rt->rt6i_siblings);
1700 rt6_multipath_rebalance(next_sibling);
1701 }
1702
1703 /* Adjust walkers */
1704 read_lock(&net->ipv6.fib6_walker_lock);
1705 FOR_WALKERS(net, w) {
1706 if (w->state == FWS_C && w->leaf == rt) {
1707 RT6_TRACE("walker %p adjusted by delroute\n", w);
1708 w->leaf = rcu_dereference_protected(rt->rt6_next,
1709 lockdep_is_held(&table->tb6_lock));
1710 if (!w->leaf)
1711 w->state = FWS_U;
1712 }
1713 }
1714 read_unlock(&net->ipv6.fib6_walker_lock);
1715
1716 /* If it was last route, call fib6_repair_tree() to:
1717 * 1. For root node, put back null_entry as how the table was created.
1718 * 2. For other nodes, expunge its radix tree node.
1719 */
1720 if (!rcu_access_pointer(fn->leaf)) {
1721 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1722 fn->fn_flags &= ~RTN_RTINFO;
1723 net->ipv6.rt6_stats->fib_route_nodes--;
1724 }
1725 fn = fib6_repair_tree(net, table, fn);
1726 }
1727
1728 fib6_purge_rt(rt, fn, net);
1729
1730 call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1731 if (!info->skip_notify)
1732 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1733 rt6_release(rt);
1734}
1735
1736/* Need to own table->tb6_lock */
1737int fib6_del(struct rt6_info *rt, struct nl_info *info)
1738{
1739 struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1740 lockdep_is_held(&rt->rt6i_table->tb6_lock));
1741 struct fib6_table *table = rt->rt6i_table;
1742 struct net *net = info->nl_net;
1743 struct rt6_info __rcu **rtp;
1744 struct rt6_info __rcu **rtp_next;
1745
1746#if RT6_DEBUG >= 2
1747 if (rt->dst.obsolete > 0) {
1748 WARN_ON(fn);
1749 return -ENOENT;
1750 }
1751#endif
1752 if (!fn || rt == net->ipv6.ip6_null_entry)
1753 return -ENOENT;
1754
1755 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1756
1757 /* remove cached dst from exception table */
1758 if (rt->rt6i_flags & RTF_CACHE)
1759 return rt6_remove_exception_rt(rt);
1760
1761 /*
1762 * Walk the leaf entries looking for ourself
1763 */
1764
1765 for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1766 struct rt6_info *cur = rcu_dereference_protected(*rtp,
1767 lockdep_is_held(&table->tb6_lock));
1768 if (rt == cur) {
1769 fib6_del_route(table, fn, rtp, info);
1770 return 0;
1771 }
1772 rtp_next = &cur->rt6_next;
1773 }
1774 return -ENOENT;
1775}
1776
1777/*
1778 * Tree traversal function.
1779 *
1780 * Certainly, it is not interrupt safe.
1781 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1782 * It means, that we can modify tree during walking
1783 * and use this function for garbage collection, clone pruning,
1784 * cleaning tree when a device goes down etc. etc.
1785 *
1786 * It guarantees that every node will be traversed,
1787 * and that it will be traversed only once.
1788 *
1789 * Callback function w->func may return:
1790 * 0 -> continue walking.
1791 * positive value -> walking is suspended (used by tree dumps,
1792 * and probably by gc, if it will be split to several slices)
1793 * negative value -> terminate walking.
1794 *
1795 * The function itself returns:
1796 * 0 -> walk is complete.
1797 * >0 -> walk is incomplete (i.e. suspended)
1798 * <0 -> walk is terminated by an error.
1799 *
1800 * This function is called with tb6_lock held.
1801 */
1802
1803static int fib6_walk_continue(struct fib6_walker *w)
1804{
1805 struct fib6_node *fn, *pn, *left, *right;
1806
1807 /* w->root should always be table->tb6_root */
1808 WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1809
1810 for (;;) {
1811 fn = w->node;
1812 if (!fn)
1813 return 0;
1814
1815 switch (w->state) {
1816#ifdef CONFIG_IPV6_SUBTREES
1817 case FWS_S:
1818 if (FIB6_SUBTREE(fn)) {
1819 w->node = FIB6_SUBTREE(fn);
1820 continue;
1821 }
1822 w->state = FWS_L;
1823#endif
1824 /* fall through */
1825 case FWS_L:
1826 left = rcu_dereference_protected(fn->left, 1);
1827 if (left) {
1828 w->node = left;
1829 w->state = FWS_INIT;
1830 continue;
1831 }
1832 w->state = FWS_R;
1833 /* fall through */
1834 case FWS_R:
1835 right = rcu_dereference_protected(fn->right, 1);
1836 if (right) {
1837 w->node = right;
1838 w->state = FWS_INIT;
1839 continue;
1840 }
1841 w->state = FWS_C;
1842 w->leaf = rcu_dereference_protected(fn->leaf, 1);
1843 /* fall through */
1844 case FWS_C:
1845 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1846 int err;
1847
1848 if (w->skip) {
1849 w->skip--;
1850 goto skip;
1851 }
1852
1853 err = w->func(w);
1854 if (err)
1855 return err;
1856
1857 w->count++;
1858 continue;
1859 }
1860skip:
1861 w->state = FWS_U;
1862 /* fall through */
1863 case FWS_U:
1864 if (fn == w->root)
1865 return 0;
1866 pn = rcu_dereference_protected(fn->parent, 1);
1867 left = rcu_dereference_protected(pn->left, 1);
1868 right = rcu_dereference_protected(pn->right, 1);
1869 w->node = pn;
1870#ifdef CONFIG_IPV6_SUBTREES
1871 if (FIB6_SUBTREE(pn) == fn) {
1872 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1873 w->state = FWS_L;
1874 continue;
1875 }
1876#endif
1877 if (left == fn) {
1878 w->state = FWS_R;
1879 continue;
1880 }
1881 if (right == fn) {
1882 w->state = FWS_C;
1883 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1884 continue;
1885 }
1886#if RT6_DEBUG >= 2
1887 WARN_ON(1);
1888#endif
1889 }
1890 }
1891}
1892
1893static int fib6_walk(struct net *net, struct fib6_walker *w)
1894{
1895 int res;
1896
1897 w->state = FWS_INIT;
1898 w->node = w->root;
1899
1900 fib6_walker_link(net, w);
1901 res = fib6_walk_continue(w);
1902 if (res <= 0)
1903 fib6_walker_unlink(net, w);
1904 return res;
1905}
1906
1907static int fib6_clean_node(struct fib6_walker *w)
1908{
1909 int res;
1910 struct rt6_info *rt;
1911 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1912 struct nl_info info = {
1913 .nl_net = c->net,
1914 };
1915
1916 if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1917 w->node->fn_sernum != c->sernum)
1918 w->node->fn_sernum = c->sernum;
1919
1920 if (!c->func) {
1921 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1922 w->leaf = NULL;
1923 return 0;
1924 }
1925
1926 for_each_fib6_walker_rt(w) {
1927 res = c->func(rt, c->arg);
1928 if (res == -1) {
1929 w->leaf = rt;
1930 res = fib6_del(rt, &info);
1931 if (res) {
1932#if RT6_DEBUG >= 2
1933 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1934 __func__, rt,
1935 rcu_access_pointer(rt->rt6i_node),
1936 res);
1937#endif
1938 continue;
1939 }
1940 return 0;
1941 } else if (res == -2) {
1942 if (WARN_ON(!rt->rt6i_nsiblings))
1943 continue;
1944 rt = list_last_entry(&rt->rt6i_siblings,
1945 struct rt6_info, rt6i_siblings);
1946 continue;
1947 }
1948 WARN_ON(res != 0);
1949 }
1950 w->leaf = rt;
1951 return 0;
1952}
1953
1954/*
1955 * Convenient frontend to tree walker.
1956 *
1957 * func is called on each route.
1958 * It may return -2 -> skip multipath route.
1959 * -1 -> delete this route.
1960 * 0 -> continue walking
1961 */
1962
1963static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1964 int (*func)(struct rt6_info *, void *arg),
1965 int sernum, void *arg)
1966{
1967 struct fib6_cleaner c;
1968
1969 c.w.root = root;
1970 c.w.func = fib6_clean_node;
1971 c.w.count = 0;
1972 c.w.skip = 0;
1973 c.func = func;
1974 c.sernum = sernum;
1975 c.arg = arg;
1976 c.net = net;
1977
1978 fib6_walk(net, &c.w);
1979}
1980
1981static void __fib6_clean_all(struct net *net,
1982 int (*func)(struct rt6_info *, void *),
1983 int sernum, void *arg)
1984{
1985 struct fib6_table *table;
1986 struct hlist_head *head;
1987 unsigned int h;
1988
1989 rcu_read_lock();
1990 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1991 head = &net->ipv6.fib_table_hash[h];
1992 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1993 spin_lock_bh(&table->tb6_lock);
1994 fib6_clean_tree(net, &table->tb6_root,
1995 func, sernum, arg);
1996 spin_unlock_bh(&table->tb6_lock);
1997 }
1998 }
1999 rcu_read_unlock();
2000}
2001
2002void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
2003 void *arg)
2004{
2005 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
2006}
2007
2008static void fib6_flush_trees(struct net *net)
2009{
2010 int new_sernum = fib6_new_sernum(net);
2011
2012 __fib6_clean_all(net, NULL, new_sernum, NULL);
2013}
2014
2015/*
2016 * Garbage collection
2017 */
2018
2019static int fib6_age(struct rt6_info *rt, void *arg)
2020{
2021 struct fib6_gc_args *gc_args = arg;
2022 unsigned long now = jiffies;
2023
2024 /*
2025 * check addrconf expiration here.
2026 * Routes are expired even if they are in use.
2027 */
2028
2029 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
2030 if (time_after(now, rt->dst.expires)) {
2031 RT6_TRACE("expiring %p\n", rt);
2032 return -1;
2033 }
2034 gc_args->more++;
2035 }
2036
2037 /* Also age clones in the exception table.
2038 * Note, that clones are aged out
2039 * only if they are not in use now.
2040 */
2041 rt6_age_exceptions(rt, gc_args, now);
2042
2043 return 0;
2044}
2045
2046void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2047{
2048 struct fib6_gc_args gc_args;
2049 unsigned long now;
2050
2051 if (force) {
2052 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2053 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2054 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2055 return;
2056 }
2057 gc_args.timeout = expires ? (int)expires :
2058 net->ipv6.sysctl.ip6_rt_gc_interval;
2059 gc_args.more = 0;
2060
2061 fib6_clean_all(net, fib6_age, &gc_args);
2062 now = jiffies;
2063 net->ipv6.ip6_rt_last_gc = now;
2064
2065 if (gc_args.more)
2066 mod_timer(&net->ipv6.ip6_fib_timer,
2067 round_jiffies(now
2068 + net->ipv6.sysctl.ip6_rt_gc_interval));
2069 else
2070 del_timer(&net->ipv6.ip6_fib_timer);
2071 spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2072}
2073
2074static void fib6_gc_timer_cb(struct timer_list *t)
2075{
2076 struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2077
2078 fib6_run_gc(0, arg, true);
2079}
2080
2081static int __net_init fib6_net_init(struct net *net)
2082{
2083 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2084 int err;
2085
2086 err = fib6_notifier_init(net);
2087 if (err)
2088 return err;
2089
2090 spin_lock_init(&net->ipv6.fib6_gc_lock);
2091 rwlock_init(&net->ipv6.fib6_walker_lock);
2092 INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2093 timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2094
2095 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2096 if (!net->ipv6.rt6_stats)
2097 goto out_timer;
2098
2099 /* Avoid false sharing : Use at least a full cache line */
2100 size = max_t(size_t, size, L1_CACHE_BYTES);
2101
2102 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2103 if (!net->ipv6.fib_table_hash)
2104 goto out_rt6_stats;
2105
2106 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2107 GFP_KERNEL);
2108 if (!net->ipv6.fib6_main_tbl)
2109 goto out_fib_table_hash;
2110
2111 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2112 rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2113 net->ipv6.ip6_null_entry);
2114 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2115 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2116 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2117
2118#ifdef CONFIG_IPV6_MULTIPLE_TABLES
2119 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2120 GFP_KERNEL);
2121 if (!net->ipv6.fib6_local_tbl)
2122 goto out_fib6_main_tbl;
2123 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2124 rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2125 net->ipv6.ip6_null_entry);
2126 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2127 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2128 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2129#endif
2130 fib6_tables_init(net);
2131
2132 return 0;
2133
2134#ifdef CONFIG_IPV6_MULTIPLE_TABLES
2135out_fib6_main_tbl:
2136 kfree(net->ipv6.fib6_main_tbl);
2137#endif
2138out_fib_table_hash:
2139 kfree(net->ipv6.fib_table_hash);
2140out_rt6_stats:
2141 kfree(net->ipv6.rt6_stats);
2142out_timer:
2143 fib6_notifier_exit(net);
2144 return -ENOMEM;
2145}
2146
2147static void fib6_net_exit(struct net *net)
2148{
2149 unsigned int i;
2150
2151 del_timer_sync(&net->ipv6.ip6_fib_timer);
2152
2153 for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2154 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2155 struct hlist_node *tmp;
2156 struct fib6_table *tb;
2157
2158 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2159 hlist_del(&tb->tb6_hlist);
2160 fib6_free_table(tb);
2161 }
2162 }
2163
2164 kfree(net->ipv6.fib_table_hash);
2165 kfree(net->ipv6.rt6_stats);
2166 fib6_notifier_exit(net);
2167}
2168
2169static struct pernet_operations fib6_net_ops = {
2170 .init = fib6_net_init,
2171 .exit = fib6_net_exit,
2172};
2173
2174int __init fib6_init(void)
2175{
2176 int ret = -ENOMEM;
2177
2178 fib6_node_kmem = kmem_cache_create("fib6_nodes",
2179 sizeof(struct fib6_node),
2180 0, SLAB_HWCACHE_ALIGN,
2181 NULL);
2182 if (!fib6_node_kmem)
2183 goto out;
2184
2185 ret = register_pernet_subsys(&fib6_net_ops);
2186 if (ret)
2187 goto out_kmem_cache_create;
2188
2189 ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2190 inet6_dump_fib, 0);
2191 if (ret)
2192 goto out_unregister_subsys;
2193
2194 __fib6_flush_trees = fib6_flush_trees;
2195out:
2196 return ret;
2197
2198out_unregister_subsys:
2199 unregister_pernet_subsys(&fib6_net_ops);
2200out_kmem_cache_create:
2201 kmem_cache_destroy(fib6_node_kmem);
2202 goto out;
2203}
2204
2205void fib6_gc_cleanup(void)
2206{
2207 unregister_pernet_subsys(&fib6_net_ops);
2208 kmem_cache_destroy(fib6_node_kmem);
2209}
2210
2211#ifdef CONFIG_PROC_FS
2212
2213struct ipv6_route_iter {
2214 struct seq_net_private p;
2215 struct fib6_walker w;
2216 loff_t skip;
2217 struct fib6_table *tbl;
2218 int sernum;
2219};
2220
2221static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2222{
2223 struct rt6_info *rt = v;
2224 struct ipv6_route_iter *iter = seq->private;
2225
2226 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
2227
2228#ifdef CONFIG_IPV6_SUBTREES
2229 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
2230#else
2231 seq_puts(seq, "00000000000000000000000000000000 00 ");
2232#endif
2233 if (rt->rt6i_flags & RTF_GATEWAY)
2234 seq_printf(seq, "%pi6", &rt->rt6i_gateway);
2235 else
2236 seq_puts(seq, "00000000000000000000000000000000");
2237
2238 seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2239 rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
2240 rt->dst.__use, rt->rt6i_flags,
2241 rt->dst.dev ? rt->dst.dev->name : "");
2242 iter->w.leaf = NULL;
2243 return 0;
2244}
2245
2246static int ipv6_route_yield(struct fib6_walker *w)
2247{
2248 struct ipv6_route_iter *iter = w->args;
2249
2250 if (!iter->skip)
2251 return 1;
2252
2253 do {
2254 iter->w.leaf = rcu_dereference_protected(
2255 iter->w.leaf->rt6_next,
2256 lockdep_is_held(&iter->tbl->tb6_lock));
2257 iter->skip--;
2258 if (!iter->skip && iter->w.leaf)
2259 return 1;
2260 } while (iter->w.leaf);
2261
2262 return 0;
2263}
2264
2265static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2266 struct net *net)
2267{
2268 memset(&iter->w, 0, sizeof(iter->w));
2269 iter->w.func = ipv6_route_yield;
2270 iter->w.root = &iter->tbl->tb6_root;
2271 iter->w.state = FWS_INIT;
2272 iter->w.node = iter->w.root;
2273 iter->w.args = iter;
2274 iter->sernum = iter->w.root->fn_sernum;
2275 INIT_LIST_HEAD(&iter->w.lh);
2276 fib6_walker_link(net, &iter->w);
2277}
2278
2279static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2280 struct net *net)
2281{
2282 unsigned int h;
2283 struct hlist_node *node;
2284
2285 if (tbl) {
2286 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2287 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2288 } else {
2289 h = 0;
2290 node = NULL;
2291 }
2292
2293 while (!node && h < FIB6_TABLE_HASHSZ) {
2294 node = rcu_dereference_bh(
2295 hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2296 }
2297 return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2298}
2299
2300static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2301{
2302 if (iter->sernum != iter->w.root->fn_sernum) {
2303 iter->sernum = iter->w.root->fn_sernum;
2304 iter->w.state = FWS_INIT;
2305 iter->w.node = iter->w.root;
2306 WARN_ON(iter->w.skip);
2307 iter->w.skip = iter->w.count;
2308 }
2309}
2310
2311static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2312{
2313 int r;
2314 struct rt6_info *n;
2315 struct net *net = seq_file_net(seq);
2316 struct ipv6_route_iter *iter = seq->private;
2317
2318 if (!v)
2319 goto iter_table;
2320
2321 n = rcu_dereference_bh(((struct rt6_info *)v)->rt6_next);
2322 if (n) {
2323 ++*pos;
2324 return n;
2325 }
2326
2327iter_table:
2328 ipv6_route_check_sernum(iter);
2329 spin_lock_bh(&iter->tbl->tb6_lock);
2330 r = fib6_walk_continue(&iter->w);
2331 spin_unlock_bh(&iter->tbl->tb6_lock);
2332 if (r > 0) {
2333 if (v)
2334 ++*pos;
2335 return iter->w.leaf;
2336 } else if (r < 0) {
2337 fib6_walker_unlink(net, &iter->w);
2338 return NULL;
2339 }
2340 fib6_walker_unlink(net, &iter->w);
2341
2342 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2343 if (!iter->tbl)
2344 return NULL;
2345
2346 ipv6_route_seq_setup_walk(iter, net);
2347 goto iter_table;
2348}
2349
2350static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2351 __acquires(RCU_BH)
2352{
2353 struct net *net = seq_file_net(seq);
2354 struct ipv6_route_iter *iter = seq->private;
2355
2356 rcu_read_lock_bh();
2357 iter->tbl = ipv6_route_seq_next_table(NULL, net);
2358 iter->skip = *pos;
2359
2360 if (iter->tbl) {
2361 ipv6_route_seq_setup_walk(iter, net);
2362 return ipv6_route_seq_next(seq, NULL, pos);
2363 } else {
2364 return NULL;
2365 }
2366}
2367
2368static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2369{
2370 struct fib6_walker *w = &iter->w;
2371 return w->node && !(w->state == FWS_U && w->node == w->root);
2372}
2373
2374static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2375 __releases(RCU_BH)
2376{
2377 struct net *net = seq_file_net(seq);
2378 struct ipv6_route_iter *iter = seq->private;
2379
2380 if (ipv6_route_iter_active(iter))
2381 fib6_walker_unlink(net, &iter->w);
2382
2383 rcu_read_unlock_bh();
2384}
2385
2386static const struct seq_operations ipv6_route_seq_ops = {
2387 .start = ipv6_route_seq_start,
2388 .next = ipv6_route_seq_next,
2389 .stop = ipv6_route_seq_stop,
2390 .show = ipv6_route_seq_show
2391};
2392
2393int ipv6_route_open(struct inode *inode, struct file *file)
2394{
2395 return seq_open_net(inode, file, &ipv6_route_seq_ops,
2396 sizeof(struct ipv6_route_iter));
2397}
2398
2399#endif /* CONFIG_PROC_FS */