Loading...
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
14/*
15 * Changes:
16 * Yuji SEKIYA @USAGI: Support default route on router node;
17 * remove ip6_null_entry from the top of
18 * routing table.
19 * Ville Nuorvala: Fixed routing subtrees.
20 */
21#include <linux/errno.h>
22#include <linux/types.h>
23#include <linux/net.h>
24#include <linux/route.h>
25#include <linux/netdevice.h>
26#include <linux/in6.h>
27#include <linux/init.h>
28#include <linux/list.h>
29#include <linux/slab.h>
30
31#ifdef CONFIG_PROC_FS
32#include <linux/proc_fs.h>
33#endif
34
35#include <net/ipv6.h>
36#include <net/ndisc.h>
37#include <net/addrconf.h>
38
39#include <net/ip6_fib.h>
40#include <net/ip6_route.h>
41
42#define RT6_DEBUG 2
43
44#if RT6_DEBUG >= 3
45#define RT6_TRACE(x...) printk(KERN_DEBUG x)
46#else
47#define RT6_TRACE(x...) do { ; } while (0)
48#endif
49
50static struct kmem_cache * fib6_node_kmem __read_mostly;
51
52enum fib_walk_state_t
53{
54#ifdef CONFIG_IPV6_SUBTREES
55 FWS_S,
56#endif
57 FWS_L,
58 FWS_R,
59 FWS_C,
60 FWS_U
61};
62
63struct fib6_cleaner_t
64{
65 struct fib6_walker_t w;
66 struct net *net;
67 int (*func)(struct rt6_info *, void *arg);
68 void *arg;
69};
70
71static DEFINE_RWLOCK(fib6_walker_lock);
72
73#ifdef CONFIG_IPV6_SUBTREES
74#define FWS_INIT FWS_S
75#else
76#define FWS_INIT FWS_L
77#endif
78
79static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
80 struct rt6_info *rt);
81static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
82static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
83static int fib6_walk(struct fib6_walker_t *w);
84static int fib6_walk_continue(struct fib6_walker_t *w);
85
86/*
87 * A routing update causes an increase of the serial number on the
88 * affected subtree. This allows for cached routes to be asynchronously
89 * tested when modifications are made to the destination cache as a
90 * result of redirects, path MTU changes, etc.
91 */
92
93static __u32 rt_sernum;
94
95static void fib6_gc_timer_cb(unsigned long arg);
96
97static LIST_HEAD(fib6_walkers);
98#define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
99
100static inline void fib6_walker_link(struct fib6_walker_t *w)
101{
102 write_lock_bh(&fib6_walker_lock);
103 list_add(&w->lh, &fib6_walkers);
104 write_unlock_bh(&fib6_walker_lock);
105}
106
107static inline void fib6_walker_unlink(struct fib6_walker_t *w)
108{
109 write_lock_bh(&fib6_walker_lock);
110 list_del(&w->lh);
111 write_unlock_bh(&fib6_walker_lock);
112}
113static __inline__ u32 fib6_new_sernum(void)
114{
115 u32 n = ++rt_sernum;
116 if ((__s32)n <= 0)
117 rt_sernum = n = 1;
118 return n;
119}
120
121/*
122 * Auxiliary address test functions for the radix tree.
123 *
124 * These assume a 32bit processor (although it will work on
125 * 64bit processors)
126 */
127
128/*
129 * test bit
130 */
131#if defined(__LITTLE_ENDIAN)
132# define BITOP_BE32_SWIZZLE (0x1F & ~7)
133#else
134# define BITOP_BE32_SWIZZLE 0
135#endif
136
137static __inline__ __be32 addr_bit_set(const void *token, int fn_bit)
138{
139 const __be32 *addr = token;
140 /*
141 * Here,
142 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
143 * is optimized version of
144 * htonl(1 << ((~fn_bit)&0x1F))
145 * See include/asm-generic/bitops/le.h.
146 */
147 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
148 addr[fn_bit >> 5];
149}
150
151static __inline__ struct fib6_node * node_alloc(void)
152{
153 struct fib6_node *fn;
154
155 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
156
157 return fn;
158}
159
160static __inline__ void node_free(struct fib6_node * fn)
161{
162 kmem_cache_free(fib6_node_kmem, fn);
163}
164
165static __inline__ void rt6_release(struct rt6_info *rt)
166{
167 if (atomic_dec_and_test(&rt->rt6i_ref))
168 dst_free(&rt->dst);
169}
170
171static void fib6_link_table(struct net *net, struct fib6_table *tb)
172{
173 unsigned int h;
174
175 /*
176 * Initialize table lock at a single place to give lockdep a key,
177 * tables aren't visible prior to being linked to the list.
178 */
179 rwlock_init(&tb->tb6_lock);
180
181 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
182
183 /*
184 * No protection necessary, this is the only list mutatation
185 * operation, tables never disappear once they exist.
186 */
187 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
188}
189
190#ifdef CONFIG_IPV6_MULTIPLE_TABLES
191
192static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
193{
194 struct fib6_table *table;
195
196 table = kzalloc(sizeof(*table), GFP_ATOMIC);
197 if (table != NULL) {
198 table->tb6_id = id;
199 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
200 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
201 }
202
203 return table;
204}
205
206struct fib6_table *fib6_new_table(struct net *net, u32 id)
207{
208 struct fib6_table *tb;
209
210 if (id == 0)
211 id = RT6_TABLE_MAIN;
212 tb = fib6_get_table(net, id);
213 if (tb)
214 return tb;
215
216 tb = fib6_alloc_table(net, id);
217 if (tb != NULL)
218 fib6_link_table(net, tb);
219
220 return tb;
221}
222
223struct fib6_table *fib6_get_table(struct net *net, u32 id)
224{
225 struct fib6_table *tb;
226 struct hlist_head *head;
227 struct hlist_node *node;
228 unsigned int h;
229
230 if (id == 0)
231 id = RT6_TABLE_MAIN;
232 h = id & (FIB6_TABLE_HASHSZ - 1);
233 rcu_read_lock();
234 head = &net->ipv6.fib_table_hash[h];
235 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
236 if (tb->tb6_id == id) {
237 rcu_read_unlock();
238 return tb;
239 }
240 }
241 rcu_read_unlock();
242
243 return NULL;
244}
245
246static void __net_init fib6_tables_init(struct net *net)
247{
248 fib6_link_table(net, net->ipv6.fib6_main_tbl);
249 fib6_link_table(net, net->ipv6.fib6_local_tbl);
250}
251#else
252
253struct fib6_table *fib6_new_table(struct net *net, u32 id)
254{
255 return fib6_get_table(net, id);
256}
257
258struct fib6_table *fib6_get_table(struct net *net, u32 id)
259{
260 return net->ipv6.fib6_main_tbl;
261}
262
263struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
264 int flags, pol_lookup_t lookup)
265{
266 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
267}
268
269static void __net_init fib6_tables_init(struct net *net)
270{
271 fib6_link_table(net, net->ipv6.fib6_main_tbl);
272}
273
274#endif
275
276static int fib6_dump_node(struct fib6_walker_t *w)
277{
278 int res;
279 struct rt6_info *rt;
280
281 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
282 res = rt6_dump_route(rt, w->args);
283 if (res < 0) {
284 /* Frame is full, suspend walking */
285 w->leaf = rt;
286 return 1;
287 }
288 WARN_ON(res == 0);
289 }
290 w->leaf = NULL;
291 return 0;
292}
293
294static void fib6_dump_end(struct netlink_callback *cb)
295{
296 struct fib6_walker_t *w = (void*)cb->args[2];
297
298 if (w) {
299 if (cb->args[4]) {
300 cb->args[4] = 0;
301 fib6_walker_unlink(w);
302 }
303 cb->args[2] = 0;
304 kfree(w);
305 }
306 cb->done = (void*)cb->args[3];
307 cb->args[1] = 3;
308}
309
310static int fib6_dump_done(struct netlink_callback *cb)
311{
312 fib6_dump_end(cb);
313 return cb->done ? cb->done(cb) : 0;
314}
315
316static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
317 struct netlink_callback *cb)
318{
319 struct fib6_walker_t *w;
320 int res;
321
322 w = (void *)cb->args[2];
323 w->root = &table->tb6_root;
324
325 if (cb->args[4] == 0) {
326 w->count = 0;
327 w->skip = 0;
328
329 read_lock_bh(&table->tb6_lock);
330 res = fib6_walk(w);
331 read_unlock_bh(&table->tb6_lock);
332 if (res > 0) {
333 cb->args[4] = 1;
334 cb->args[5] = w->root->fn_sernum;
335 }
336 } else {
337 if (cb->args[5] != w->root->fn_sernum) {
338 /* Begin at the root if the tree changed */
339 cb->args[5] = w->root->fn_sernum;
340 w->state = FWS_INIT;
341 w->node = w->root;
342 w->skip = w->count;
343 } else
344 w->skip = 0;
345
346 read_lock_bh(&table->tb6_lock);
347 res = fib6_walk_continue(w);
348 read_unlock_bh(&table->tb6_lock);
349 if (res <= 0) {
350 fib6_walker_unlink(w);
351 cb->args[4] = 0;
352 }
353 }
354
355 return res;
356}
357
358static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
359{
360 struct net *net = sock_net(skb->sk);
361 unsigned int h, s_h;
362 unsigned int e = 0, s_e;
363 struct rt6_rtnl_dump_arg arg;
364 struct fib6_walker_t *w;
365 struct fib6_table *tb;
366 struct hlist_node *node;
367 struct hlist_head *head;
368 int res = 0;
369
370 s_h = cb->args[0];
371 s_e = cb->args[1];
372
373 w = (void *)cb->args[2];
374 if (w == NULL) {
375 /* New dump:
376 *
377 * 1. hook callback destructor.
378 */
379 cb->args[3] = (long)cb->done;
380 cb->done = fib6_dump_done;
381
382 /*
383 * 2. allocate and initialize walker.
384 */
385 w = kzalloc(sizeof(*w), GFP_ATOMIC);
386 if (w == NULL)
387 return -ENOMEM;
388 w->func = fib6_dump_node;
389 cb->args[2] = (long)w;
390 }
391
392 arg.skb = skb;
393 arg.cb = cb;
394 arg.net = net;
395 w->args = &arg;
396
397 rcu_read_lock();
398 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
399 e = 0;
400 head = &net->ipv6.fib_table_hash[h];
401 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
402 if (e < s_e)
403 goto next;
404 res = fib6_dump_table(tb, skb, cb);
405 if (res != 0)
406 goto out;
407next:
408 e++;
409 }
410 }
411out:
412 rcu_read_unlock();
413 cb->args[1] = e;
414 cb->args[0] = h;
415
416 res = res < 0 ? res : skb->len;
417 if (res <= 0)
418 fib6_dump_end(cb);
419 return res;
420}
421
422/*
423 * Routing Table
424 *
425 * return the appropriate node for a routing tree "add" operation
426 * by either creating and inserting or by returning an existing
427 * node.
428 */
429
430static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
431 int addrlen, int plen,
432 int offset)
433{
434 struct fib6_node *fn, *in, *ln;
435 struct fib6_node *pn = NULL;
436 struct rt6key *key;
437 int bit;
438 __be32 dir = 0;
439 __u32 sernum = fib6_new_sernum();
440
441 RT6_TRACE("fib6_add_1\n");
442
443 /* insert node in tree */
444
445 fn = root;
446
447 do {
448 key = (struct rt6key *)((u8 *)fn->leaf + offset);
449
450 /*
451 * Prefix match
452 */
453 if (plen < fn->fn_bit ||
454 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
455 goto insert_above;
456
457 /*
458 * Exact match ?
459 */
460
461 if (plen == fn->fn_bit) {
462 /* clean up an intermediate node */
463 if ((fn->fn_flags & RTN_RTINFO) == 0) {
464 rt6_release(fn->leaf);
465 fn->leaf = NULL;
466 }
467
468 fn->fn_sernum = sernum;
469
470 return fn;
471 }
472
473 /*
474 * We have more bits to go
475 */
476
477 /* Try to walk down on tree. */
478 fn->fn_sernum = sernum;
479 dir = addr_bit_set(addr, fn->fn_bit);
480 pn = fn;
481 fn = dir ? fn->right: fn->left;
482 } while (fn);
483
484 /*
485 * We walked to the bottom of tree.
486 * Create new leaf node without children.
487 */
488
489 ln = node_alloc();
490
491 if (ln == NULL)
492 return NULL;
493 ln->fn_bit = plen;
494
495 ln->parent = pn;
496 ln->fn_sernum = sernum;
497
498 if (dir)
499 pn->right = ln;
500 else
501 pn->left = ln;
502
503 return ln;
504
505
506insert_above:
507 /*
508 * split since we don't have a common prefix anymore or
509 * we have a less significant route.
510 * we've to insert an intermediate node on the list
511 * this new node will point to the one we need to create
512 * and the current
513 */
514
515 pn = fn->parent;
516
517 /* find 1st bit in difference between the 2 addrs.
518
519 See comment in __ipv6_addr_diff: bit may be an invalid value,
520 but if it is >= plen, the value is ignored in any case.
521 */
522
523 bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
524
525 /*
526 * (intermediate)[in]
527 * / \
528 * (new leaf node)[ln] (old node)[fn]
529 */
530 if (plen > bit) {
531 in = node_alloc();
532 ln = node_alloc();
533
534 if (in == NULL || ln == NULL) {
535 if (in)
536 node_free(in);
537 if (ln)
538 node_free(ln);
539 return NULL;
540 }
541
542 /*
543 * new intermediate node.
544 * RTN_RTINFO will
545 * be off since that an address that chooses one of
546 * the branches would not match less specific routes
547 * in the other branch
548 */
549
550 in->fn_bit = bit;
551
552 in->parent = pn;
553 in->leaf = fn->leaf;
554 atomic_inc(&in->leaf->rt6i_ref);
555
556 in->fn_sernum = sernum;
557
558 /* update parent pointer */
559 if (dir)
560 pn->right = in;
561 else
562 pn->left = in;
563
564 ln->fn_bit = plen;
565
566 ln->parent = in;
567 fn->parent = in;
568
569 ln->fn_sernum = sernum;
570
571 if (addr_bit_set(addr, bit)) {
572 in->right = ln;
573 in->left = fn;
574 } else {
575 in->left = ln;
576 in->right = fn;
577 }
578 } else { /* plen <= bit */
579
580 /*
581 * (new leaf node)[ln]
582 * / \
583 * (old node)[fn] NULL
584 */
585
586 ln = node_alloc();
587
588 if (ln == NULL)
589 return NULL;
590
591 ln->fn_bit = plen;
592
593 ln->parent = pn;
594
595 ln->fn_sernum = sernum;
596
597 if (dir)
598 pn->right = ln;
599 else
600 pn->left = ln;
601
602 if (addr_bit_set(&key->addr, plen))
603 ln->right = fn;
604 else
605 ln->left = fn;
606
607 fn->parent = ln;
608 }
609 return ln;
610}
611
612/*
613 * Insert routing information in a node.
614 */
615
616static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
617 struct nl_info *info)
618{
619 struct rt6_info *iter = NULL;
620 struct rt6_info **ins;
621
622 ins = &fn->leaf;
623
624 for (iter = fn->leaf; iter; iter=iter->dst.rt6_next) {
625 /*
626 * Search for duplicates
627 */
628
629 if (iter->rt6i_metric == rt->rt6i_metric) {
630 /*
631 * Same priority level
632 */
633
634 if (iter->rt6i_dev == rt->rt6i_dev &&
635 iter->rt6i_idev == rt->rt6i_idev &&
636 ipv6_addr_equal(&iter->rt6i_gateway,
637 &rt->rt6i_gateway)) {
638 if (!(iter->rt6i_flags&RTF_EXPIRES))
639 return -EEXIST;
640 iter->rt6i_expires = rt->rt6i_expires;
641 if (!(rt->rt6i_flags&RTF_EXPIRES)) {
642 iter->rt6i_flags &= ~RTF_EXPIRES;
643 iter->rt6i_expires = 0;
644 }
645 return -EEXIST;
646 }
647 }
648
649 if (iter->rt6i_metric > rt->rt6i_metric)
650 break;
651
652 ins = &iter->dst.rt6_next;
653 }
654
655 /* Reset round-robin state, if necessary */
656 if (ins == &fn->leaf)
657 fn->rr_ptr = NULL;
658
659 /*
660 * insert node
661 */
662
663 rt->dst.rt6_next = iter;
664 *ins = rt;
665 rt->rt6i_node = fn;
666 atomic_inc(&rt->rt6i_ref);
667 inet6_rt_notify(RTM_NEWROUTE, rt, info);
668 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
669
670 if ((fn->fn_flags & RTN_RTINFO) == 0) {
671 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
672 fn->fn_flags |= RTN_RTINFO;
673 }
674
675 return 0;
676}
677
678static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
679{
680 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
681 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE)))
682 mod_timer(&net->ipv6.ip6_fib_timer,
683 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
684}
685
686void fib6_force_start_gc(struct net *net)
687{
688 if (!timer_pending(&net->ipv6.ip6_fib_timer))
689 mod_timer(&net->ipv6.ip6_fib_timer,
690 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
691}
692
693/*
694 * Add routing information to the routing tree.
695 * <destination addr>/<source addr>
696 * with source addr info in sub-trees
697 */
698
699int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
700{
701 struct fib6_node *fn, *pn = NULL;
702 int err = -ENOMEM;
703
704 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
705 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst));
706
707 if (fn == NULL)
708 goto out;
709
710 pn = fn;
711
712#ifdef CONFIG_IPV6_SUBTREES
713 if (rt->rt6i_src.plen) {
714 struct fib6_node *sn;
715
716 if (fn->subtree == NULL) {
717 struct fib6_node *sfn;
718
719 /*
720 * Create subtree.
721 *
722 * fn[main tree]
723 * |
724 * sfn[subtree root]
725 * \
726 * sn[new leaf node]
727 */
728
729 /* Create subtree root node */
730 sfn = node_alloc();
731 if (sfn == NULL)
732 goto st_failure;
733
734 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
735 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
736 sfn->fn_flags = RTN_ROOT;
737 sfn->fn_sernum = fib6_new_sernum();
738
739 /* Now add the first leaf node to new subtree */
740
741 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
742 sizeof(struct in6_addr), rt->rt6i_src.plen,
743 offsetof(struct rt6_info, rt6i_src));
744
745 if (sn == NULL) {
746 /* If it is failed, discard just allocated
747 root, and then (in st_failure) stale node
748 in main tree.
749 */
750 node_free(sfn);
751 goto st_failure;
752 }
753
754 /* Now link new subtree to main tree */
755 sfn->parent = fn;
756 fn->subtree = sfn;
757 } else {
758 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
759 sizeof(struct in6_addr), rt->rt6i_src.plen,
760 offsetof(struct rt6_info, rt6i_src));
761
762 if (sn == NULL)
763 goto st_failure;
764 }
765
766 if (fn->leaf == NULL) {
767 fn->leaf = rt;
768 atomic_inc(&rt->rt6i_ref);
769 }
770 fn = sn;
771 }
772#endif
773
774 err = fib6_add_rt2node(fn, rt, info);
775
776 if (err == 0) {
777 fib6_start_gc(info->nl_net, rt);
778 if (!(rt->rt6i_flags&RTF_CACHE))
779 fib6_prune_clones(info->nl_net, pn, rt);
780 }
781
782out:
783 if (err) {
784#ifdef CONFIG_IPV6_SUBTREES
785 /*
786 * If fib6_add_1 has cleared the old leaf pointer in the
787 * super-tree leaf node we have to find a new one for it.
788 */
789 if (pn != fn && pn->leaf == rt) {
790 pn->leaf = NULL;
791 atomic_dec(&rt->rt6i_ref);
792 }
793 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
794 pn->leaf = fib6_find_prefix(info->nl_net, pn);
795#if RT6_DEBUG >= 2
796 if (!pn->leaf) {
797 WARN_ON(pn->leaf == NULL);
798 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
799 }
800#endif
801 atomic_inc(&pn->leaf->rt6i_ref);
802 }
803#endif
804 dst_free(&rt->dst);
805 }
806 return err;
807
808#ifdef CONFIG_IPV6_SUBTREES
809 /* Subtree creation failed, probably main tree node
810 is orphan. If it is, shoot it.
811 */
812st_failure:
813 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
814 fib6_repair_tree(info->nl_net, fn);
815 dst_free(&rt->dst);
816 return err;
817#endif
818}
819
820/*
821 * Routing tree lookup
822 *
823 */
824
825struct lookup_args {
826 int offset; /* key offset on rt6_info */
827 const struct in6_addr *addr; /* search key */
828};
829
830static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
831 struct lookup_args *args)
832{
833 struct fib6_node *fn;
834 __be32 dir;
835
836 if (unlikely(args->offset == 0))
837 return NULL;
838
839 /*
840 * Descend on a tree
841 */
842
843 fn = root;
844
845 for (;;) {
846 struct fib6_node *next;
847
848 dir = addr_bit_set(args->addr, fn->fn_bit);
849
850 next = dir ? fn->right : fn->left;
851
852 if (next) {
853 fn = next;
854 continue;
855 }
856
857 break;
858 }
859
860 while(fn) {
861 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
862 struct rt6key *key;
863
864 key = (struct rt6key *) ((u8 *) fn->leaf +
865 args->offset);
866
867 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
868#ifdef CONFIG_IPV6_SUBTREES
869 if (fn->subtree)
870 fn = fib6_lookup_1(fn->subtree, args + 1);
871#endif
872 if (!fn || fn->fn_flags & RTN_RTINFO)
873 return fn;
874 }
875 }
876
877 if (fn->fn_flags & RTN_ROOT)
878 break;
879
880 fn = fn->parent;
881 }
882
883 return NULL;
884}
885
886struct fib6_node * fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
887 const struct in6_addr *saddr)
888{
889 struct fib6_node *fn;
890 struct lookup_args args[] = {
891 {
892 .offset = offsetof(struct rt6_info, rt6i_dst),
893 .addr = daddr,
894 },
895#ifdef CONFIG_IPV6_SUBTREES
896 {
897 .offset = offsetof(struct rt6_info, rt6i_src),
898 .addr = saddr,
899 },
900#endif
901 {
902 .offset = 0, /* sentinel */
903 }
904 };
905
906 fn = fib6_lookup_1(root, daddr ? args : args + 1);
907
908 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT)
909 fn = root;
910
911 return fn;
912}
913
914/*
915 * Get node with specified destination prefix (and source prefix,
916 * if subtrees are used)
917 */
918
919
920static struct fib6_node * fib6_locate_1(struct fib6_node *root,
921 const struct in6_addr *addr,
922 int plen, int offset)
923{
924 struct fib6_node *fn;
925
926 for (fn = root; fn ; ) {
927 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
928
929 /*
930 * Prefix match
931 */
932 if (plen < fn->fn_bit ||
933 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
934 return NULL;
935
936 if (plen == fn->fn_bit)
937 return fn;
938
939 /*
940 * We have more bits to go
941 */
942 if (addr_bit_set(addr, fn->fn_bit))
943 fn = fn->right;
944 else
945 fn = fn->left;
946 }
947 return NULL;
948}
949
950struct fib6_node * fib6_locate(struct fib6_node *root,
951 const struct in6_addr *daddr, int dst_len,
952 const struct in6_addr *saddr, int src_len)
953{
954 struct fib6_node *fn;
955
956 fn = fib6_locate_1(root, daddr, dst_len,
957 offsetof(struct rt6_info, rt6i_dst));
958
959#ifdef CONFIG_IPV6_SUBTREES
960 if (src_len) {
961 WARN_ON(saddr == NULL);
962 if (fn && fn->subtree)
963 fn = fib6_locate_1(fn->subtree, saddr, src_len,
964 offsetof(struct rt6_info, rt6i_src));
965 }
966#endif
967
968 if (fn && fn->fn_flags&RTN_RTINFO)
969 return fn;
970
971 return NULL;
972}
973
974
975/*
976 * Deletion
977 *
978 */
979
980static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
981{
982 if (fn->fn_flags&RTN_ROOT)
983 return net->ipv6.ip6_null_entry;
984
985 while(fn) {
986 if(fn->left)
987 return fn->left->leaf;
988
989 if(fn->right)
990 return fn->right->leaf;
991
992 fn = FIB6_SUBTREE(fn);
993 }
994 return NULL;
995}
996
997/*
998 * Called to trim the tree of intermediate nodes when possible. "fn"
999 * is the node we want to try and remove.
1000 */
1001
1002static struct fib6_node *fib6_repair_tree(struct net *net,
1003 struct fib6_node *fn)
1004{
1005 int children;
1006 int nstate;
1007 struct fib6_node *child, *pn;
1008 struct fib6_walker_t *w;
1009 int iter = 0;
1010
1011 for (;;) {
1012 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1013 iter++;
1014
1015 WARN_ON(fn->fn_flags & RTN_RTINFO);
1016 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1017 WARN_ON(fn->leaf != NULL);
1018
1019 children = 0;
1020 child = NULL;
1021 if (fn->right) child = fn->right, children |= 1;
1022 if (fn->left) child = fn->left, children |= 2;
1023
1024 if (children == 3 || FIB6_SUBTREE(fn)
1025#ifdef CONFIG_IPV6_SUBTREES
1026 /* Subtree root (i.e. fn) may have one child */
1027 || (children && fn->fn_flags&RTN_ROOT)
1028#endif
1029 ) {
1030 fn->leaf = fib6_find_prefix(net, fn);
1031#if RT6_DEBUG >= 2
1032 if (fn->leaf==NULL) {
1033 WARN_ON(!fn->leaf);
1034 fn->leaf = net->ipv6.ip6_null_entry;
1035 }
1036#endif
1037 atomic_inc(&fn->leaf->rt6i_ref);
1038 return fn->parent;
1039 }
1040
1041 pn = fn->parent;
1042#ifdef CONFIG_IPV6_SUBTREES
1043 if (FIB6_SUBTREE(pn) == fn) {
1044 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1045 FIB6_SUBTREE(pn) = NULL;
1046 nstate = FWS_L;
1047 } else {
1048 WARN_ON(fn->fn_flags & RTN_ROOT);
1049#endif
1050 if (pn->right == fn) pn->right = child;
1051 else if (pn->left == fn) pn->left = child;
1052#if RT6_DEBUG >= 2
1053 else
1054 WARN_ON(1);
1055#endif
1056 if (child)
1057 child->parent = pn;
1058 nstate = FWS_R;
1059#ifdef CONFIG_IPV6_SUBTREES
1060 }
1061#endif
1062
1063 read_lock(&fib6_walker_lock);
1064 FOR_WALKERS(w) {
1065 if (child == NULL) {
1066 if (w->root == fn) {
1067 w->root = w->node = NULL;
1068 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1069 } else if (w->node == fn) {
1070 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1071 w->node = pn;
1072 w->state = nstate;
1073 }
1074 } else {
1075 if (w->root == fn) {
1076 w->root = child;
1077 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1078 }
1079 if (w->node == fn) {
1080 w->node = child;
1081 if (children&2) {
1082 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1083 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1084 } else {
1085 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1086 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1087 }
1088 }
1089 }
1090 }
1091 read_unlock(&fib6_walker_lock);
1092
1093 node_free(fn);
1094 if (pn->fn_flags&RTN_RTINFO || FIB6_SUBTREE(pn))
1095 return pn;
1096
1097 rt6_release(pn->leaf);
1098 pn->leaf = NULL;
1099 fn = pn;
1100 }
1101}
1102
1103static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1104 struct nl_info *info)
1105{
1106 struct fib6_walker_t *w;
1107 struct rt6_info *rt = *rtp;
1108 struct net *net = info->nl_net;
1109
1110 RT6_TRACE("fib6_del_route\n");
1111
1112 /* Unlink it */
1113 *rtp = rt->dst.rt6_next;
1114 rt->rt6i_node = NULL;
1115 net->ipv6.rt6_stats->fib_rt_entries--;
1116 net->ipv6.rt6_stats->fib_discarded_routes++;
1117
1118 /* Reset round-robin state, if necessary */
1119 if (fn->rr_ptr == rt)
1120 fn->rr_ptr = NULL;
1121
1122 /* Adjust walkers */
1123 read_lock(&fib6_walker_lock);
1124 FOR_WALKERS(w) {
1125 if (w->state == FWS_C && w->leaf == rt) {
1126 RT6_TRACE("walker %p adjusted by delroute\n", w);
1127 w->leaf = rt->dst.rt6_next;
1128 if (w->leaf == NULL)
1129 w->state = FWS_U;
1130 }
1131 }
1132 read_unlock(&fib6_walker_lock);
1133
1134 rt->dst.rt6_next = NULL;
1135
1136 /* If it was last route, expunge its radix tree node */
1137 if (fn->leaf == NULL) {
1138 fn->fn_flags &= ~RTN_RTINFO;
1139 net->ipv6.rt6_stats->fib_route_nodes--;
1140 fn = fib6_repair_tree(net, fn);
1141 }
1142
1143 if (atomic_read(&rt->rt6i_ref) != 1) {
1144 /* This route is used as dummy address holder in some split
1145 * nodes. It is not leaked, but it still holds other resources,
1146 * which must be released in time. So, scan ascendant nodes
1147 * and replace dummy references to this route with references
1148 * to still alive ones.
1149 */
1150 while (fn) {
1151 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) {
1152 fn->leaf = fib6_find_prefix(net, fn);
1153 atomic_inc(&fn->leaf->rt6i_ref);
1154 rt6_release(rt);
1155 }
1156 fn = fn->parent;
1157 }
1158 /* No more references are possible at this point. */
1159 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1160 }
1161
1162 inet6_rt_notify(RTM_DELROUTE, rt, info);
1163 rt6_release(rt);
1164}
1165
1166int fib6_del(struct rt6_info *rt, struct nl_info *info)
1167{
1168 struct net *net = info->nl_net;
1169 struct fib6_node *fn = rt->rt6i_node;
1170 struct rt6_info **rtp;
1171
1172#if RT6_DEBUG >= 2
1173 if (rt->dst.obsolete>0) {
1174 WARN_ON(fn != NULL);
1175 return -ENOENT;
1176 }
1177#endif
1178 if (fn == NULL || rt == net->ipv6.ip6_null_entry)
1179 return -ENOENT;
1180
1181 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1182
1183 if (!(rt->rt6i_flags&RTF_CACHE)) {
1184 struct fib6_node *pn = fn;
1185#ifdef CONFIG_IPV6_SUBTREES
1186 /* clones of this route might be in another subtree */
1187 if (rt->rt6i_src.plen) {
1188 while (!(pn->fn_flags&RTN_ROOT))
1189 pn = pn->parent;
1190 pn = pn->parent;
1191 }
1192#endif
1193 fib6_prune_clones(info->nl_net, pn, rt);
1194 }
1195
1196 /*
1197 * Walk the leaf entries looking for ourself
1198 */
1199
1200 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1201 if (*rtp == rt) {
1202 fib6_del_route(fn, rtp, info);
1203 return 0;
1204 }
1205 }
1206 return -ENOENT;
1207}
1208
1209/*
1210 * Tree traversal function.
1211 *
1212 * Certainly, it is not interrupt safe.
1213 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1214 * It means, that we can modify tree during walking
1215 * and use this function for garbage collection, clone pruning,
1216 * cleaning tree when a device goes down etc. etc.
1217 *
1218 * It guarantees that every node will be traversed,
1219 * and that it will be traversed only once.
1220 *
1221 * Callback function w->func may return:
1222 * 0 -> continue walking.
1223 * positive value -> walking is suspended (used by tree dumps,
1224 * and probably by gc, if it will be split to several slices)
1225 * negative value -> terminate walking.
1226 *
1227 * The function itself returns:
1228 * 0 -> walk is complete.
1229 * >0 -> walk is incomplete (i.e. suspended)
1230 * <0 -> walk is terminated by an error.
1231 */
1232
1233static int fib6_walk_continue(struct fib6_walker_t *w)
1234{
1235 struct fib6_node *fn, *pn;
1236
1237 for (;;) {
1238 fn = w->node;
1239 if (fn == NULL)
1240 return 0;
1241
1242 if (w->prune && fn != w->root &&
1243 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) {
1244 w->state = FWS_C;
1245 w->leaf = fn->leaf;
1246 }
1247 switch (w->state) {
1248#ifdef CONFIG_IPV6_SUBTREES
1249 case FWS_S:
1250 if (FIB6_SUBTREE(fn)) {
1251 w->node = FIB6_SUBTREE(fn);
1252 continue;
1253 }
1254 w->state = FWS_L;
1255#endif
1256 case FWS_L:
1257 if (fn->left) {
1258 w->node = fn->left;
1259 w->state = FWS_INIT;
1260 continue;
1261 }
1262 w->state = FWS_R;
1263 case FWS_R:
1264 if (fn->right) {
1265 w->node = fn->right;
1266 w->state = FWS_INIT;
1267 continue;
1268 }
1269 w->state = FWS_C;
1270 w->leaf = fn->leaf;
1271 case FWS_C:
1272 if (w->leaf && fn->fn_flags&RTN_RTINFO) {
1273 int err;
1274
1275 if (w->count < w->skip) {
1276 w->count++;
1277 continue;
1278 }
1279
1280 err = w->func(w);
1281 if (err)
1282 return err;
1283
1284 w->count++;
1285 continue;
1286 }
1287 w->state = FWS_U;
1288 case FWS_U:
1289 if (fn == w->root)
1290 return 0;
1291 pn = fn->parent;
1292 w->node = pn;
1293#ifdef CONFIG_IPV6_SUBTREES
1294 if (FIB6_SUBTREE(pn) == fn) {
1295 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1296 w->state = FWS_L;
1297 continue;
1298 }
1299#endif
1300 if (pn->left == fn) {
1301 w->state = FWS_R;
1302 continue;
1303 }
1304 if (pn->right == fn) {
1305 w->state = FWS_C;
1306 w->leaf = w->node->leaf;
1307 continue;
1308 }
1309#if RT6_DEBUG >= 2
1310 WARN_ON(1);
1311#endif
1312 }
1313 }
1314}
1315
1316static int fib6_walk(struct fib6_walker_t *w)
1317{
1318 int res;
1319
1320 w->state = FWS_INIT;
1321 w->node = w->root;
1322
1323 fib6_walker_link(w);
1324 res = fib6_walk_continue(w);
1325 if (res <= 0)
1326 fib6_walker_unlink(w);
1327 return res;
1328}
1329
1330static int fib6_clean_node(struct fib6_walker_t *w)
1331{
1332 int res;
1333 struct rt6_info *rt;
1334 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1335 struct nl_info info = {
1336 .nl_net = c->net,
1337 };
1338
1339 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1340 res = c->func(rt, c->arg);
1341 if (res < 0) {
1342 w->leaf = rt;
1343 res = fib6_del(rt, &info);
1344 if (res) {
1345#if RT6_DEBUG >= 2
1346 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res);
1347#endif
1348 continue;
1349 }
1350 return 0;
1351 }
1352 WARN_ON(res != 0);
1353 }
1354 w->leaf = rt;
1355 return 0;
1356}
1357
1358/*
1359 * Convenient frontend to tree walker.
1360 *
1361 * func is called on each route.
1362 * It may return -1 -> delete this route.
1363 * 0 -> continue walking
1364 *
1365 * prune==1 -> only immediate children of node (certainly,
1366 * ignoring pure split nodes) will be scanned.
1367 */
1368
1369static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1370 int (*func)(struct rt6_info *, void *arg),
1371 int prune, void *arg)
1372{
1373 struct fib6_cleaner_t c;
1374
1375 c.w.root = root;
1376 c.w.func = fib6_clean_node;
1377 c.w.prune = prune;
1378 c.w.count = 0;
1379 c.w.skip = 0;
1380 c.func = func;
1381 c.arg = arg;
1382 c.net = net;
1383
1384 fib6_walk(&c.w);
1385}
1386
1387void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1388 int prune, void *arg)
1389{
1390 struct fib6_table *table;
1391 struct hlist_node *node;
1392 struct hlist_head *head;
1393 unsigned int h;
1394
1395 rcu_read_lock();
1396 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1397 head = &net->ipv6.fib_table_hash[h];
1398 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1399 write_lock_bh(&table->tb6_lock);
1400 fib6_clean_tree(net, &table->tb6_root,
1401 func, prune, arg);
1402 write_unlock_bh(&table->tb6_lock);
1403 }
1404 }
1405 rcu_read_unlock();
1406}
1407
1408static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1409{
1410 if (rt->rt6i_flags & RTF_CACHE) {
1411 RT6_TRACE("pruning clone %p\n", rt);
1412 return -1;
1413 }
1414
1415 return 0;
1416}
1417
1418static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1419 struct rt6_info *rt)
1420{
1421 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1422}
1423
1424/*
1425 * Garbage collection
1426 */
1427
1428static struct fib6_gc_args
1429{
1430 int timeout;
1431 int more;
1432} gc_args;
1433
1434static int fib6_age(struct rt6_info *rt, void *arg)
1435{
1436 unsigned long now = jiffies;
1437
1438 /*
1439 * check addrconf expiration here.
1440 * Routes are expired even if they are in use.
1441 *
1442 * Also age clones. Note, that clones are aged out
1443 * only if they are not in use now.
1444 */
1445
1446 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) {
1447 if (time_after(now, rt->rt6i_expires)) {
1448 RT6_TRACE("expiring %p\n", rt);
1449 return -1;
1450 }
1451 gc_args.more++;
1452 } else if (rt->rt6i_flags & RTF_CACHE) {
1453 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1454 time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1455 RT6_TRACE("aging clone %p\n", rt);
1456 return -1;
1457 } else if ((rt->rt6i_flags & RTF_GATEWAY) &&
1458 (!(dst_get_neighbour_raw(&rt->dst)->flags & NTF_ROUTER))) {
1459 RT6_TRACE("purging route %p via non-router but gateway\n",
1460 rt);
1461 return -1;
1462 }
1463 gc_args.more++;
1464 }
1465
1466 return 0;
1467}
1468
1469static DEFINE_SPINLOCK(fib6_gc_lock);
1470
1471void fib6_run_gc(unsigned long expires, struct net *net)
1472{
1473 if (expires != ~0UL) {
1474 spin_lock_bh(&fib6_gc_lock);
1475 gc_args.timeout = expires ? (int)expires :
1476 net->ipv6.sysctl.ip6_rt_gc_interval;
1477 } else {
1478 if (!spin_trylock_bh(&fib6_gc_lock)) {
1479 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1480 return;
1481 }
1482 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1483 }
1484
1485 gc_args.more = icmp6_dst_gc();
1486
1487 fib6_clean_all(net, fib6_age, 0, NULL);
1488
1489 if (gc_args.more)
1490 mod_timer(&net->ipv6.ip6_fib_timer,
1491 round_jiffies(jiffies
1492 + net->ipv6.sysctl.ip6_rt_gc_interval));
1493 else
1494 del_timer(&net->ipv6.ip6_fib_timer);
1495 spin_unlock_bh(&fib6_gc_lock);
1496}
1497
1498static void fib6_gc_timer_cb(unsigned long arg)
1499{
1500 fib6_run_gc(0, (struct net *)arg);
1501}
1502
1503static int __net_init fib6_net_init(struct net *net)
1504{
1505 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1506
1507 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1508
1509 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1510 if (!net->ipv6.rt6_stats)
1511 goto out_timer;
1512
1513 /* Avoid false sharing : Use at least a full cache line */
1514 size = max_t(size_t, size, L1_CACHE_BYTES);
1515
1516 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1517 if (!net->ipv6.fib_table_hash)
1518 goto out_rt6_stats;
1519
1520 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1521 GFP_KERNEL);
1522 if (!net->ipv6.fib6_main_tbl)
1523 goto out_fib_table_hash;
1524
1525 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1526 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1527 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1528 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1529
1530#ifdef CONFIG_IPV6_MULTIPLE_TABLES
1531 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1532 GFP_KERNEL);
1533 if (!net->ipv6.fib6_local_tbl)
1534 goto out_fib6_main_tbl;
1535 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1536 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1537 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1538 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1539#endif
1540 fib6_tables_init(net);
1541
1542 return 0;
1543
1544#ifdef CONFIG_IPV6_MULTIPLE_TABLES
1545out_fib6_main_tbl:
1546 kfree(net->ipv6.fib6_main_tbl);
1547#endif
1548out_fib_table_hash:
1549 kfree(net->ipv6.fib_table_hash);
1550out_rt6_stats:
1551 kfree(net->ipv6.rt6_stats);
1552out_timer:
1553 return -ENOMEM;
1554 }
1555
1556static void fib6_net_exit(struct net *net)
1557{
1558 rt6_ifdown(net, NULL);
1559 del_timer_sync(&net->ipv6.ip6_fib_timer);
1560
1561#ifdef CONFIG_IPV6_MULTIPLE_TABLES
1562 kfree(net->ipv6.fib6_local_tbl);
1563#endif
1564 kfree(net->ipv6.fib6_main_tbl);
1565 kfree(net->ipv6.fib_table_hash);
1566 kfree(net->ipv6.rt6_stats);
1567}
1568
1569static struct pernet_operations fib6_net_ops = {
1570 .init = fib6_net_init,
1571 .exit = fib6_net_exit,
1572};
1573
1574int __init fib6_init(void)
1575{
1576 int ret = -ENOMEM;
1577
1578 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1579 sizeof(struct fib6_node),
1580 0, SLAB_HWCACHE_ALIGN,
1581 NULL);
1582 if (!fib6_node_kmem)
1583 goto out;
1584
1585 ret = register_pernet_subsys(&fib6_net_ops);
1586 if (ret)
1587 goto out_kmem_cache_create;
1588
1589 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1590 NULL);
1591 if (ret)
1592 goto out_unregister_subsys;
1593out:
1594 return ret;
1595
1596out_unregister_subsys:
1597 unregister_pernet_subsys(&fib6_net_ops);
1598out_kmem_cache_create:
1599 kmem_cache_destroy(fib6_node_kmem);
1600 goto out;
1601}
1602
1603void fib6_gc_cleanup(void)
1604{
1605 unregister_pernet_subsys(&fib6_net_ops);
1606 kmem_cache_destroy(fib6_node_kmem);
1607}
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(struct net *net, struct fib6_node *fn)
202{
203 kfree_rcu(fn, rcu);
204 net->ipv6.rt6_stats->fib_nodes--;
205}
206
207static void fib6_free_table(struct fib6_table *table)
208{
209 inetpeer_invalidate_tree(&table->tb6_peers);
210 kfree(table);
211}
212
213static void fib6_link_table(struct net *net, struct fib6_table *tb)
214{
215 unsigned int h;
216
217 /*
218 * Initialize table lock at a single place to give lockdep a key,
219 * tables aren't visible prior to being linked to the list.
220 */
221 spin_lock_init(&tb->tb6_lock);
222 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
223
224 /*
225 * No protection necessary, this is the only list mutatation
226 * operation, tables never disappear once they exist.
227 */
228 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
229}
230
231#ifdef CONFIG_IPV6_MULTIPLE_TABLES
232
233static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
234{
235 struct fib6_table *table;
236
237 table = kzalloc(sizeof(*table), GFP_ATOMIC);
238 if (table) {
239 table->tb6_id = id;
240 rcu_assign_pointer(table->tb6_root.leaf,
241 net->ipv6.fib6_null_entry);
242 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
243 inet_peer_base_init(&table->tb6_peers);
244 INIT_HLIST_HEAD(&table->tb6_gc_hlist);
245 }
246
247 return table;
248}
249
250struct fib6_table *fib6_new_table(struct net *net, u32 id)
251{
252 struct fib6_table *tb;
253
254 if (id == 0)
255 id = RT6_TABLE_MAIN;
256 tb = fib6_get_table(net, id);
257 if (tb)
258 return tb;
259
260 tb = fib6_alloc_table(net, id);
261 if (tb)
262 fib6_link_table(net, tb);
263
264 return tb;
265}
266EXPORT_SYMBOL_GPL(fib6_new_table);
267
268struct fib6_table *fib6_get_table(struct net *net, u32 id)
269{
270 struct fib6_table *tb;
271 struct hlist_head *head;
272 unsigned int h;
273
274 if (id == 0)
275 id = RT6_TABLE_MAIN;
276 h = id & (FIB6_TABLE_HASHSZ - 1);
277 rcu_read_lock();
278 head = &net->ipv6.fib_table_hash[h];
279 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
280 if (tb->tb6_id == id) {
281 rcu_read_unlock();
282 return tb;
283 }
284 }
285 rcu_read_unlock();
286
287 return NULL;
288}
289EXPORT_SYMBOL_GPL(fib6_get_table);
290
291static void __net_init fib6_tables_init(struct net *net)
292{
293 fib6_link_table(net, net->ipv6.fib6_main_tbl);
294 fib6_link_table(net, net->ipv6.fib6_local_tbl);
295}
296#else
297
298struct fib6_table *fib6_new_table(struct net *net, u32 id)
299{
300 return fib6_get_table(net, id);
301}
302
303struct fib6_table *fib6_get_table(struct net *net, u32 id)
304{
305 return net->ipv6.fib6_main_tbl;
306}
307
308struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
309 const struct sk_buff *skb,
310 int flags, pol_lookup_t lookup)
311{
312 struct rt6_info *rt;
313
314 rt = pol_lookup_func(lookup,
315 net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
316 if (rt->dst.error == -EAGAIN) {
317 ip6_rt_put_flags(rt, flags);
318 rt = net->ipv6.ip6_null_entry;
319 if (!(flags & RT6_LOOKUP_F_DST_NOREF))
320 dst_hold(&rt->dst);
321 }
322
323 return &rt->dst;
324}
325
326/* called with rcu lock held; no reference taken on fib6_info */
327int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
328 struct fib6_result *res, int flags)
329{
330 return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6,
331 res, flags);
332}
333
334static void __net_init fib6_tables_init(struct net *net)
335{
336 fib6_link_table(net, net->ipv6.fib6_main_tbl);
337}
338
339#endif
340
341unsigned int fib6_tables_seq_read(const struct net *net)
342{
343 unsigned int h, fib_seq = 0;
344
345 rcu_read_lock();
346 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
347 const struct hlist_head *head = &net->ipv6.fib_table_hash[h];
348 const struct fib6_table *tb;
349
350 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
351 fib_seq += READ_ONCE(tb->fib_seq);
352 }
353 rcu_read_unlock();
354
355 return fib_seq;
356}
357
358static int call_fib6_entry_notifier(struct notifier_block *nb,
359 enum fib_event_type event_type,
360 struct fib6_info *rt,
361 struct netlink_ext_ack *extack)
362{
363 struct fib6_entry_notifier_info info = {
364 .info.extack = extack,
365 .rt = rt,
366 };
367
368 return call_fib6_notifier(nb, event_type, &info.info);
369}
370
371static int call_fib6_multipath_entry_notifier(struct notifier_block *nb,
372 enum fib_event_type event_type,
373 struct fib6_info *rt,
374 unsigned int nsiblings,
375 struct netlink_ext_ack *extack)
376{
377 struct fib6_entry_notifier_info info = {
378 .info.extack = extack,
379 .rt = rt,
380 .nsiblings = nsiblings,
381 };
382
383 return call_fib6_notifier(nb, event_type, &info.info);
384}
385
386int call_fib6_entry_notifiers(struct net *net,
387 enum fib_event_type event_type,
388 struct fib6_info *rt,
389 struct netlink_ext_ack *extack)
390{
391 struct fib6_entry_notifier_info info = {
392 .info.extack = extack,
393 .rt = rt,
394 };
395
396 WRITE_ONCE(rt->fib6_table->fib_seq, rt->fib6_table->fib_seq + 1);
397 return call_fib6_notifiers(net, event_type, &info.info);
398}
399
400int call_fib6_multipath_entry_notifiers(struct net *net,
401 enum fib_event_type event_type,
402 struct fib6_info *rt,
403 unsigned int nsiblings,
404 struct netlink_ext_ack *extack)
405{
406 struct fib6_entry_notifier_info info = {
407 .info.extack = extack,
408 .rt = rt,
409 .nsiblings = nsiblings,
410 };
411
412 WRITE_ONCE(rt->fib6_table->fib_seq, rt->fib6_table->fib_seq + 1);
413 return call_fib6_notifiers(net, event_type, &info.info);
414}
415
416int call_fib6_entry_notifiers_replace(struct net *net, struct fib6_info *rt)
417{
418 struct fib6_entry_notifier_info info = {
419 .rt = rt,
420 .nsiblings = rt->fib6_nsiblings,
421 };
422
423 WRITE_ONCE(rt->fib6_table->fib_seq, rt->fib6_table->fib_seq + 1);
424 return call_fib6_notifiers(net, FIB_EVENT_ENTRY_REPLACE, &info.info);
425}
426
427struct fib6_dump_arg {
428 struct net *net;
429 struct notifier_block *nb;
430 struct netlink_ext_ack *extack;
431};
432
433static int fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
434{
435 enum fib_event_type fib_event = FIB_EVENT_ENTRY_REPLACE;
436 int err;
437
438 if (!rt || rt == arg->net->ipv6.fib6_null_entry)
439 return 0;
440
441 if (rt->fib6_nsiblings)
442 err = call_fib6_multipath_entry_notifier(arg->nb, fib_event,
443 rt,
444 rt->fib6_nsiblings,
445 arg->extack);
446 else
447 err = call_fib6_entry_notifier(arg->nb, fib_event, rt,
448 arg->extack);
449
450 return err;
451}
452
453static int fib6_node_dump(struct fib6_walker *w)
454{
455 int err;
456
457 err = fib6_rt_dump(w->leaf, w->args);
458 w->leaf = NULL;
459 return err;
460}
461
462static int fib6_table_dump(struct net *net, struct fib6_table *tb,
463 struct fib6_walker *w)
464{
465 int err;
466
467 w->root = &tb->tb6_root;
468 spin_lock_bh(&tb->tb6_lock);
469 err = fib6_walk(net, w);
470 spin_unlock_bh(&tb->tb6_lock);
471 return err;
472}
473
474/* Called with rcu_read_lock() */
475int fib6_tables_dump(struct net *net, struct notifier_block *nb,
476 struct netlink_ext_ack *extack)
477{
478 struct fib6_dump_arg arg;
479 struct fib6_walker *w;
480 unsigned int h;
481 int err = 0;
482
483 w = kzalloc(sizeof(*w), GFP_ATOMIC);
484 if (!w)
485 return -ENOMEM;
486
487 w->func = fib6_node_dump;
488 arg.net = net;
489 arg.nb = nb;
490 arg.extack = extack;
491 w->args = &arg;
492
493 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
494 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
495 struct fib6_table *tb;
496
497 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
498 err = fib6_table_dump(net, tb, w);
499 if (err)
500 goto out;
501 }
502 }
503
504out:
505 kfree(w);
506
507 /* The tree traversal function should never return a positive value. */
508 return err > 0 ? -EINVAL : err;
509}
510
511static int fib6_dump_node(struct fib6_walker *w)
512{
513 int res;
514 struct fib6_info *rt;
515
516 for_each_fib6_walker_rt(w) {
517 res = rt6_dump_route(rt, w->args, w->skip_in_node);
518 if (res >= 0) {
519 /* Frame is full, suspend walking */
520 w->leaf = rt;
521
522 /* We'll restart from this node, so if some routes were
523 * already dumped, skip them next time.
524 */
525 w->skip_in_node += res;
526
527 return 1;
528 }
529 w->skip_in_node = 0;
530
531 /* Multipath routes are dumped in one route with the
532 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
533 * last sibling of this route (no need to dump the
534 * sibling routes again)
535 */
536 if (rt->fib6_nsiblings)
537 rt = list_last_entry(&rt->fib6_siblings,
538 struct fib6_info,
539 fib6_siblings);
540 }
541 w->leaf = NULL;
542 return 0;
543}
544
545static void fib6_dump_end(struct netlink_callback *cb)
546{
547 struct net *net = sock_net(cb->skb->sk);
548 struct fib6_walker *w = (void *)cb->args[2];
549
550 if (w) {
551 if (cb->args[4]) {
552 cb->args[4] = 0;
553 fib6_walker_unlink(net, w);
554 }
555 cb->args[2] = 0;
556 kfree(w);
557 }
558 cb->done = (void *)cb->args[3];
559 cb->args[1] = 3;
560}
561
562static int fib6_dump_done(struct netlink_callback *cb)
563{
564 fib6_dump_end(cb);
565 return cb->done ? cb->done(cb) : 0;
566}
567
568static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
569 struct netlink_callback *cb)
570{
571 struct net *net = sock_net(skb->sk);
572 struct fib6_walker *w;
573 int res;
574
575 w = (void *)cb->args[2];
576 w->root = &table->tb6_root;
577
578 if (cb->args[4] == 0) {
579 w->count = 0;
580 w->skip = 0;
581 w->skip_in_node = 0;
582
583 spin_lock_bh(&table->tb6_lock);
584 res = fib6_walk(net, w);
585 spin_unlock_bh(&table->tb6_lock);
586 if (res > 0) {
587 cb->args[4] = 1;
588 cb->args[5] = READ_ONCE(w->root->fn_sernum);
589 }
590 } else {
591 int sernum = READ_ONCE(w->root->fn_sernum);
592 if (cb->args[5] != sernum) {
593 /* Begin at the root if the tree changed */
594 cb->args[5] = sernum;
595 w->state = FWS_INIT;
596 w->node = w->root;
597 w->skip = w->count;
598 w->skip_in_node = 0;
599 } else
600 w->skip = 0;
601
602 spin_lock_bh(&table->tb6_lock);
603 res = fib6_walk_continue(w);
604 spin_unlock_bh(&table->tb6_lock);
605 if (res <= 0) {
606 fib6_walker_unlink(net, w);
607 cb->args[4] = 0;
608 }
609 }
610
611 return res;
612}
613
614static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
615{
616 struct rt6_rtnl_dump_arg arg = {
617 .filter.dump_exceptions = true,
618 .filter.dump_routes = true,
619 .filter.rtnl_held = false,
620 };
621 const struct nlmsghdr *nlh = cb->nlh;
622 struct net *net = sock_net(skb->sk);
623 unsigned int e = 0, s_e;
624 struct hlist_head *head;
625 struct fib6_walker *w;
626 struct fib6_table *tb;
627 unsigned int h, s_h;
628 int err = 0;
629
630 rcu_read_lock();
631 if (cb->strict_check) {
632 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
633 if (err < 0)
634 goto unlock;
635 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
636 struct rtmsg *rtm = nlmsg_data(nlh);
637
638 if (rtm->rtm_flags & RTM_F_PREFIX)
639 arg.filter.flags = RTM_F_PREFIX;
640 }
641
642 w = (void *)cb->args[2];
643 if (!w) {
644 /* New dump:
645 *
646 * 1. allocate and initialize walker.
647 */
648 w = kzalloc(sizeof(*w), GFP_ATOMIC);
649 if (!w) {
650 err = -ENOMEM;
651 goto unlock;
652 }
653 w->func = fib6_dump_node;
654 cb->args[2] = (long)w;
655
656 /* 2. hook callback destructor.
657 */
658 cb->args[3] = (long)cb->done;
659 cb->done = fib6_dump_done;
660
661 }
662
663 arg.skb = skb;
664 arg.cb = cb;
665 arg.net = net;
666 w->args = &arg;
667
668 if (arg.filter.table_id) {
669 tb = fib6_get_table(net, arg.filter.table_id);
670 if (!tb) {
671 if (rtnl_msg_family(cb->nlh) != PF_INET6)
672 goto unlock;
673
674 NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
675 err = -ENOENT;
676 goto unlock;
677 }
678
679 if (!cb->args[0]) {
680 err = fib6_dump_table(tb, skb, cb);
681 if (!err)
682 cb->args[0] = 1;
683 }
684 goto unlock;
685 }
686
687 s_h = cb->args[0];
688 s_e = cb->args[1];
689
690 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
691 e = 0;
692 head = &net->ipv6.fib_table_hash[h];
693 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
694 if (e < s_e)
695 goto next;
696 err = fib6_dump_table(tb, skb, cb);
697 if (err != 0)
698 goto out;
699next:
700 e++;
701 }
702 }
703out:
704 cb->args[1] = e;
705 cb->args[0] = h;
706
707unlock:
708 rcu_read_unlock();
709 if (err <= 0)
710 fib6_dump_end(cb);
711 return err;
712}
713
714void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
715{
716 if (!f6i)
717 return;
718
719 if (f6i->fib6_metrics == &dst_default_metrics) {
720 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
721
722 if (!p)
723 return;
724
725 refcount_set(&p->refcnt, 1);
726 f6i->fib6_metrics = p;
727 }
728
729 f6i->fib6_metrics->metrics[metric - 1] = val;
730}
731
732/*
733 * Routing Table
734 *
735 * return the appropriate node for a routing tree "add" operation
736 * by either creating and inserting or by returning an existing
737 * node.
738 */
739
740static struct fib6_node *fib6_add_1(struct net *net,
741 struct fib6_table *table,
742 struct fib6_node *root,
743 struct in6_addr *addr, int plen,
744 int offset, int allow_create,
745 int replace_required,
746 struct netlink_ext_ack *extack)
747{
748 struct fib6_node *fn, *in, *ln;
749 struct fib6_node *pn = NULL;
750 struct rt6key *key;
751 int bit;
752 __be32 dir = 0;
753
754 /* insert node in tree */
755
756 fn = root;
757
758 do {
759 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
760 lockdep_is_held(&table->tb6_lock));
761 key = (struct rt6key *)((u8 *)leaf + offset);
762
763 /*
764 * Prefix match
765 */
766 if (plen < fn->fn_bit ||
767 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
768 if (!allow_create) {
769 if (replace_required) {
770 NL_SET_ERR_MSG(extack,
771 "Can not replace route - no match found");
772 pr_warn("Can't replace route, no match found\n");
773 return ERR_PTR(-ENOENT);
774 }
775 pr_warn("NLM_F_CREATE should be set when creating new route\n");
776 }
777 goto insert_above;
778 }
779
780 /*
781 * Exact match ?
782 */
783
784 if (plen == fn->fn_bit) {
785 /* clean up an intermediate node */
786 if (!(fn->fn_flags & RTN_RTINFO)) {
787 RCU_INIT_POINTER(fn->leaf, NULL);
788 fib6_info_release(leaf);
789 /* remove null_entry in the root node */
790 } else if (fn->fn_flags & RTN_TL_ROOT &&
791 rcu_access_pointer(fn->leaf) ==
792 net->ipv6.fib6_null_entry) {
793 RCU_INIT_POINTER(fn->leaf, NULL);
794 }
795
796 return fn;
797 }
798
799 /*
800 * We have more bits to go
801 */
802
803 /* Try to walk down on tree. */
804 dir = addr_bit_set(addr, fn->fn_bit);
805 pn = fn;
806 fn = dir ?
807 rcu_dereference_protected(fn->right,
808 lockdep_is_held(&table->tb6_lock)) :
809 rcu_dereference_protected(fn->left,
810 lockdep_is_held(&table->tb6_lock));
811 } while (fn);
812
813 if (!allow_create) {
814 /* We should not create new node because
815 * NLM_F_REPLACE was specified without NLM_F_CREATE
816 * I assume it is safe to require NLM_F_CREATE when
817 * REPLACE flag is used! Later we may want to remove the
818 * check for replace_required, because according
819 * to netlink specification, NLM_F_CREATE
820 * MUST be specified if new route is created.
821 * That would keep IPv6 consistent with IPv4
822 */
823 if (replace_required) {
824 NL_SET_ERR_MSG(extack,
825 "Can not replace route - no match found");
826 pr_warn("Can't replace route, no match found\n");
827 return ERR_PTR(-ENOENT);
828 }
829 pr_warn("NLM_F_CREATE should be set when creating new route\n");
830 }
831 /*
832 * We walked to the bottom of tree.
833 * Create new leaf node without children.
834 */
835
836 ln = node_alloc(net);
837
838 if (!ln)
839 return ERR_PTR(-ENOMEM);
840 ln->fn_bit = plen;
841 RCU_INIT_POINTER(ln->parent, pn);
842
843 if (dir)
844 rcu_assign_pointer(pn->right, ln);
845 else
846 rcu_assign_pointer(pn->left, ln);
847
848 return ln;
849
850
851insert_above:
852 /*
853 * split since we don't have a common prefix anymore or
854 * we have a less significant route.
855 * we've to insert an intermediate node on the list
856 * this new node will point to the one we need to create
857 * and the current
858 */
859
860 pn = rcu_dereference_protected(fn->parent,
861 lockdep_is_held(&table->tb6_lock));
862
863 /* find 1st bit in difference between the 2 addrs.
864
865 See comment in __ipv6_addr_diff: bit may be an invalid value,
866 but if it is >= plen, the value is ignored in any case.
867 */
868
869 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
870
871 /*
872 * (intermediate)[in]
873 * / \
874 * (new leaf node)[ln] (old node)[fn]
875 */
876 if (plen > bit) {
877 in = node_alloc(net);
878 ln = node_alloc(net);
879
880 if (!in || !ln) {
881 if (in)
882 node_free_immediate(net, in);
883 if (ln)
884 node_free_immediate(net, ln);
885 return ERR_PTR(-ENOMEM);
886 }
887
888 /*
889 * new intermediate node.
890 * RTN_RTINFO will
891 * be off since that an address that chooses one of
892 * the branches would not match less specific routes
893 * in the other branch
894 */
895
896 in->fn_bit = bit;
897
898 RCU_INIT_POINTER(in->parent, pn);
899 in->leaf = fn->leaf;
900 fib6_info_hold(rcu_dereference_protected(in->leaf,
901 lockdep_is_held(&table->tb6_lock)));
902
903 /* update parent pointer */
904 if (dir)
905 rcu_assign_pointer(pn->right, in);
906 else
907 rcu_assign_pointer(pn->left, in);
908
909 ln->fn_bit = plen;
910
911 RCU_INIT_POINTER(ln->parent, in);
912 rcu_assign_pointer(fn->parent, in);
913
914 if (addr_bit_set(addr, bit)) {
915 rcu_assign_pointer(in->right, ln);
916 rcu_assign_pointer(in->left, fn);
917 } else {
918 rcu_assign_pointer(in->left, ln);
919 rcu_assign_pointer(in->right, fn);
920 }
921 } else { /* plen <= bit */
922
923 /*
924 * (new leaf node)[ln]
925 * / \
926 * (old node)[fn] NULL
927 */
928
929 ln = node_alloc(net);
930
931 if (!ln)
932 return ERR_PTR(-ENOMEM);
933
934 ln->fn_bit = plen;
935
936 RCU_INIT_POINTER(ln->parent, pn);
937
938 if (addr_bit_set(&key->addr, plen))
939 RCU_INIT_POINTER(ln->right, fn);
940 else
941 RCU_INIT_POINTER(ln->left, fn);
942
943 rcu_assign_pointer(fn->parent, ln);
944
945 if (dir)
946 rcu_assign_pointer(pn->right, ln);
947 else
948 rcu_assign_pointer(pn->left, ln);
949 }
950 return ln;
951}
952
953static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
954 const struct fib6_info *match,
955 const struct fib6_table *table)
956{
957 int cpu;
958
959 if (!fib6_nh->rt6i_pcpu)
960 return;
961
962 rcu_read_lock();
963 /* release the reference to this fib entry from
964 * all of its cached pcpu routes
965 */
966 for_each_possible_cpu(cpu) {
967 struct rt6_info **ppcpu_rt;
968 struct rt6_info *pcpu_rt;
969
970 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
971
972 /* Paired with xchg() in rt6_get_pcpu_route() */
973 pcpu_rt = READ_ONCE(*ppcpu_rt);
974
975 /* only dropping the 'from' reference if the cached route
976 * is using 'match'. The cached pcpu_rt->from only changes
977 * from a fib6_info to NULL (ip6_dst_destroy); it can never
978 * change from one fib6_info reference to another
979 */
980 if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
981 struct fib6_info *from;
982
983 from = unrcu_pointer(xchg(&pcpu_rt->from, NULL));
984 fib6_info_release(from);
985 }
986 }
987 rcu_read_unlock();
988}
989
990struct fib6_nh_pcpu_arg {
991 struct fib6_info *from;
992 const struct fib6_table *table;
993};
994
995static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg)
996{
997 struct fib6_nh_pcpu_arg *arg = _arg;
998
999 __fib6_drop_pcpu_from(nh, arg->from, arg->table);
1000 return 0;
1001}
1002
1003static void fib6_drop_pcpu_from(struct fib6_info *f6i,
1004 const struct fib6_table *table)
1005{
1006 /* Make sure rt6_make_pcpu_route() wont add other percpu routes
1007 * while we are cleaning them here.
1008 */
1009 f6i->fib6_destroying = 1;
1010 mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
1011
1012 if (f6i->nh) {
1013 struct fib6_nh_pcpu_arg arg = {
1014 .from = f6i,
1015 .table = table
1016 };
1017
1018 nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from,
1019 &arg);
1020 } else {
1021 struct fib6_nh *fib6_nh;
1022
1023 fib6_nh = f6i->fib6_nh;
1024 __fib6_drop_pcpu_from(fib6_nh, f6i, table);
1025 }
1026}
1027
1028static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
1029 struct net *net)
1030{
1031 struct fib6_table *table = rt->fib6_table;
1032
1033 /* Flush all cached dst in exception table */
1034 rt6_flush_exceptions(rt);
1035 fib6_drop_pcpu_from(rt, table);
1036
1037 if (rt->nh && !list_empty(&rt->nh_list))
1038 list_del_init(&rt->nh_list);
1039
1040 if (refcount_read(&rt->fib6_ref) != 1) {
1041 /* This route is used as dummy address holder in some split
1042 * nodes. It is not leaked, but it still holds other resources,
1043 * which must be released in time. So, scan ascendant nodes
1044 * and replace dummy references to this route with references
1045 * to still alive ones.
1046 */
1047 while (fn) {
1048 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1049 lockdep_is_held(&table->tb6_lock));
1050 struct fib6_info *new_leaf;
1051 if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
1052 new_leaf = fib6_find_prefix(net, table, fn);
1053 fib6_info_hold(new_leaf);
1054
1055 rcu_assign_pointer(fn->leaf, new_leaf);
1056 fib6_info_release(rt);
1057 }
1058 fn = rcu_dereference_protected(fn->parent,
1059 lockdep_is_held(&table->tb6_lock));
1060 }
1061 }
1062
1063 fib6_clean_expires(rt);
1064 fib6_remove_gc_list(rt);
1065}
1066
1067/*
1068 * Insert routing information in a node.
1069 */
1070
1071static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
1072 struct nl_info *info,
1073 struct netlink_ext_ack *extack)
1074{
1075 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1076 lockdep_is_held(&rt->fib6_table->tb6_lock));
1077 struct fib6_info *iter = NULL;
1078 struct fib6_info __rcu **ins;
1079 struct fib6_info __rcu **fallback_ins = NULL;
1080 int replace = (info->nlh &&
1081 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
1082 int add = (!info->nlh ||
1083 (info->nlh->nlmsg_flags & NLM_F_CREATE));
1084 int found = 0;
1085 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
1086 bool notify_sibling_rt = false;
1087 u16 nlflags = NLM_F_EXCL;
1088 int err;
1089
1090 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
1091 nlflags |= NLM_F_APPEND;
1092
1093 ins = &fn->leaf;
1094
1095 for (iter = leaf; iter;
1096 iter = rcu_dereference_protected(iter->fib6_next,
1097 lockdep_is_held(&rt->fib6_table->tb6_lock))) {
1098 /*
1099 * Search for duplicates
1100 */
1101
1102 if (iter->fib6_metric == rt->fib6_metric) {
1103 /*
1104 * Same priority level
1105 */
1106 if (info->nlh &&
1107 (info->nlh->nlmsg_flags & NLM_F_EXCL))
1108 return -EEXIST;
1109
1110 nlflags &= ~NLM_F_EXCL;
1111 if (replace) {
1112 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1113 found++;
1114 break;
1115 }
1116 fallback_ins = fallback_ins ?: ins;
1117 goto next_iter;
1118 }
1119
1120 if (rt6_duplicate_nexthop(iter, rt)) {
1121 if (rt->fib6_nsiblings)
1122 rt->fib6_nsiblings = 0;
1123 if (!(iter->fib6_flags & RTF_EXPIRES))
1124 return -EEXIST;
1125 if (!(rt->fib6_flags & RTF_EXPIRES)) {
1126 fib6_clean_expires(iter);
1127 fib6_remove_gc_list(iter);
1128 } else {
1129 fib6_set_expires(iter, rt->expires);
1130 fib6_add_gc_list(iter);
1131 }
1132
1133 if (rt->fib6_pmtu)
1134 fib6_metric_set(iter, RTAX_MTU,
1135 rt->fib6_pmtu);
1136 return -EEXIST;
1137 }
1138 /* If we have the same destination and the same metric,
1139 * but not the same gateway, then the route we try to
1140 * add is sibling to this route, increment our counter
1141 * of siblings, and later we will add our route to the
1142 * list.
1143 * Only static routes (which don't have flag
1144 * RTF_EXPIRES) are used for ECMPv6.
1145 *
1146 * To avoid long list, we only had siblings if the
1147 * route have a gateway.
1148 */
1149 if (rt_can_ecmp &&
1150 rt6_qualify_for_ecmp(iter))
1151 rt->fib6_nsiblings++;
1152 }
1153
1154 if (iter->fib6_metric > rt->fib6_metric)
1155 break;
1156
1157next_iter:
1158 ins = &iter->fib6_next;
1159 }
1160
1161 if (fallback_ins && !found) {
1162 /* No matching route with same ecmp-able-ness found, replace
1163 * first matching route
1164 */
1165 ins = fallback_ins;
1166 iter = rcu_dereference_protected(*ins,
1167 lockdep_is_held(&rt->fib6_table->tb6_lock));
1168 found++;
1169 }
1170
1171 /* Reset round-robin state, if necessary */
1172 if (ins == &fn->leaf)
1173 fn->rr_ptr = NULL;
1174
1175 /* Link this route to others same route. */
1176 if (rt->fib6_nsiblings) {
1177 unsigned int fib6_nsiblings;
1178 struct fib6_info *sibling, *temp_sibling;
1179
1180 /* Find the first route that have the same metric */
1181 sibling = leaf;
1182 notify_sibling_rt = true;
1183 while (sibling) {
1184 if (sibling->fib6_metric == rt->fib6_metric &&
1185 rt6_qualify_for_ecmp(sibling)) {
1186 list_add_tail_rcu(&rt->fib6_siblings,
1187 &sibling->fib6_siblings);
1188 break;
1189 }
1190 sibling = rcu_dereference_protected(sibling->fib6_next,
1191 lockdep_is_held(&rt->fib6_table->tb6_lock));
1192 notify_sibling_rt = false;
1193 }
1194 /* For each sibling in the list, increment the counter of
1195 * siblings. BUG() if counters does not match, list of siblings
1196 * is broken!
1197 */
1198 fib6_nsiblings = 0;
1199 list_for_each_entry_safe(sibling, temp_sibling,
1200 &rt->fib6_siblings, fib6_siblings) {
1201 sibling->fib6_nsiblings++;
1202 BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1203 fib6_nsiblings++;
1204 }
1205 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1206 rt6_multipath_rebalance(temp_sibling);
1207 }
1208
1209 /*
1210 * insert node
1211 */
1212 if (!replace) {
1213 if (!add)
1214 pr_warn("NLM_F_CREATE should be set when creating new route\n");
1215
1216add:
1217 nlflags |= NLM_F_CREATE;
1218
1219 /* The route should only be notified if it is the first
1220 * route in the node or if it is added as a sibling
1221 * route to the first route in the node.
1222 */
1223 if (!info->skip_notify_kernel &&
1224 (notify_sibling_rt || ins == &fn->leaf)) {
1225 enum fib_event_type fib_event;
1226
1227 if (notify_sibling_rt)
1228 fib_event = FIB_EVENT_ENTRY_APPEND;
1229 else
1230 fib_event = FIB_EVENT_ENTRY_REPLACE;
1231 err = call_fib6_entry_notifiers(info->nl_net,
1232 fib_event, rt,
1233 extack);
1234 if (err) {
1235 struct fib6_info *sibling, *next_sibling;
1236
1237 /* If the route has siblings, then it first
1238 * needs to be unlinked from them.
1239 */
1240 if (!rt->fib6_nsiblings)
1241 return err;
1242
1243 list_for_each_entry_safe(sibling, next_sibling,
1244 &rt->fib6_siblings,
1245 fib6_siblings)
1246 sibling->fib6_nsiblings--;
1247 rt->fib6_nsiblings = 0;
1248 list_del_rcu(&rt->fib6_siblings);
1249 rt6_multipath_rebalance(next_sibling);
1250 return err;
1251 }
1252 }
1253
1254 rcu_assign_pointer(rt->fib6_next, iter);
1255 fib6_info_hold(rt);
1256 rcu_assign_pointer(rt->fib6_node, fn);
1257 rcu_assign_pointer(*ins, rt);
1258 if (!info->skip_notify)
1259 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1260 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1261
1262 if (!(fn->fn_flags & RTN_RTINFO)) {
1263 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1264 fn->fn_flags |= RTN_RTINFO;
1265 }
1266
1267 } else {
1268 int nsiblings;
1269
1270 if (!found) {
1271 if (add)
1272 goto add;
1273 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1274 return -ENOENT;
1275 }
1276
1277 if (!info->skip_notify_kernel && ins == &fn->leaf) {
1278 err = call_fib6_entry_notifiers(info->nl_net,
1279 FIB_EVENT_ENTRY_REPLACE,
1280 rt, extack);
1281 if (err)
1282 return err;
1283 }
1284
1285 fib6_info_hold(rt);
1286 rcu_assign_pointer(rt->fib6_node, fn);
1287 rt->fib6_next = iter->fib6_next;
1288 rcu_assign_pointer(*ins, rt);
1289 if (!info->skip_notify)
1290 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1291 if (!(fn->fn_flags & RTN_RTINFO)) {
1292 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1293 fn->fn_flags |= RTN_RTINFO;
1294 }
1295 nsiblings = iter->fib6_nsiblings;
1296 iter->fib6_node = NULL;
1297 fib6_purge_rt(iter, fn, info->nl_net);
1298 if (rcu_access_pointer(fn->rr_ptr) == iter)
1299 fn->rr_ptr = NULL;
1300 fib6_info_release(iter);
1301
1302 if (nsiblings) {
1303 /* Replacing an ECMP route, remove all siblings */
1304 ins = &rt->fib6_next;
1305 iter = rcu_dereference_protected(*ins,
1306 lockdep_is_held(&rt->fib6_table->tb6_lock));
1307 while (iter) {
1308 if (iter->fib6_metric > rt->fib6_metric)
1309 break;
1310 if (rt6_qualify_for_ecmp(iter)) {
1311 *ins = iter->fib6_next;
1312 iter->fib6_node = NULL;
1313 fib6_purge_rt(iter, fn, info->nl_net);
1314 if (rcu_access_pointer(fn->rr_ptr) == iter)
1315 fn->rr_ptr = NULL;
1316 fib6_info_release(iter);
1317 nsiblings--;
1318 info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1319 } else {
1320 ins = &iter->fib6_next;
1321 }
1322 iter = rcu_dereference_protected(*ins,
1323 lockdep_is_held(&rt->fib6_table->tb6_lock));
1324 }
1325 WARN_ON(nsiblings != 0);
1326 }
1327 }
1328
1329 return 0;
1330}
1331
1332static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1333{
1334 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1335 (rt->fib6_flags & RTF_EXPIRES))
1336 mod_timer(&net->ipv6.ip6_fib_timer,
1337 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1338}
1339
1340void fib6_force_start_gc(struct net *net)
1341{
1342 if (!timer_pending(&net->ipv6.ip6_fib_timer))
1343 mod_timer(&net->ipv6.ip6_fib_timer,
1344 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1345}
1346
1347static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1348 int sernum)
1349{
1350 struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1351 lockdep_is_held(&rt->fib6_table->tb6_lock));
1352
1353 /* paired with smp_rmb() in fib6_get_cookie_safe() */
1354 smp_wmb();
1355 while (fn) {
1356 WRITE_ONCE(fn->fn_sernum, sernum);
1357 fn = rcu_dereference_protected(fn->parent,
1358 lockdep_is_held(&rt->fib6_table->tb6_lock));
1359 }
1360}
1361
1362void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1363{
1364 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1365}
1366
1367/* allow ipv4 to update sernum via ipv6_stub */
1368void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1369{
1370 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1371 fib6_update_sernum_upto_root(net, f6i);
1372 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1373}
1374
1375/*
1376 * Add routing information to the routing tree.
1377 * <destination addr>/<source addr>
1378 * with source addr info in sub-trees
1379 * Need to own table->tb6_lock
1380 */
1381
1382int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1383 struct nl_info *info, struct netlink_ext_ack *extack)
1384{
1385 struct fib6_table *table = rt->fib6_table;
1386 struct fib6_node *fn;
1387#ifdef CONFIG_IPV6_SUBTREES
1388 struct fib6_node *pn = NULL;
1389#endif
1390 int err = -ENOMEM;
1391 int allow_create = 1;
1392 int replace_required = 0;
1393
1394 if (info->nlh) {
1395 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1396 allow_create = 0;
1397 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1398 replace_required = 1;
1399 }
1400 if (!allow_create && !replace_required)
1401 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1402
1403 fn = fib6_add_1(info->nl_net, table, root,
1404 &rt->fib6_dst.addr, rt->fib6_dst.plen,
1405 offsetof(struct fib6_info, fib6_dst), allow_create,
1406 replace_required, extack);
1407 if (IS_ERR(fn)) {
1408 err = PTR_ERR(fn);
1409 fn = NULL;
1410 goto out;
1411 }
1412
1413#ifdef CONFIG_IPV6_SUBTREES
1414 pn = fn;
1415
1416 if (rt->fib6_src.plen) {
1417 struct fib6_node *sn;
1418
1419 if (!rcu_access_pointer(fn->subtree)) {
1420 struct fib6_node *sfn;
1421
1422 /*
1423 * Create subtree.
1424 *
1425 * fn[main tree]
1426 * |
1427 * sfn[subtree root]
1428 * \
1429 * sn[new leaf node]
1430 */
1431
1432 /* Create subtree root node */
1433 sfn = node_alloc(info->nl_net);
1434 if (!sfn)
1435 goto failure;
1436
1437 fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1438 rcu_assign_pointer(sfn->leaf,
1439 info->nl_net->ipv6.fib6_null_entry);
1440 sfn->fn_flags = RTN_ROOT;
1441
1442 /* Now add the first leaf node to new subtree */
1443
1444 sn = fib6_add_1(info->nl_net, table, sfn,
1445 &rt->fib6_src.addr, rt->fib6_src.plen,
1446 offsetof(struct fib6_info, fib6_src),
1447 allow_create, replace_required, extack);
1448
1449 if (IS_ERR(sn)) {
1450 /* If it is failed, discard just allocated
1451 root, and then (in failure) stale node
1452 in main tree.
1453 */
1454 node_free_immediate(info->nl_net, sfn);
1455 err = PTR_ERR(sn);
1456 goto failure;
1457 }
1458
1459 /* Now link new subtree to main tree */
1460 rcu_assign_pointer(sfn->parent, fn);
1461 rcu_assign_pointer(fn->subtree, sfn);
1462 } else {
1463 sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1464 &rt->fib6_src.addr, rt->fib6_src.plen,
1465 offsetof(struct fib6_info, fib6_src),
1466 allow_create, replace_required, extack);
1467
1468 if (IS_ERR(sn)) {
1469 err = PTR_ERR(sn);
1470 goto failure;
1471 }
1472 }
1473
1474 if (!rcu_access_pointer(fn->leaf)) {
1475 if (fn->fn_flags & RTN_TL_ROOT) {
1476 /* put back null_entry for root node */
1477 rcu_assign_pointer(fn->leaf,
1478 info->nl_net->ipv6.fib6_null_entry);
1479 } else {
1480 fib6_info_hold(rt);
1481 rcu_assign_pointer(fn->leaf, rt);
1482 }
1483 }
1484 fn = sn;
1485 }
1486#endif
1487
1488 err = fib6_add_rt2node(fn, rt, info, extack);
1489 if (!err) {
1490 if (rt->nh)
1491 list_add(&rt->nh_list, &rt->nh->f6i_list);
1492 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(info->nl_net));
1493
1494 if (rt->fib6_flags & RTF_EXPIRES)
1495 fib6_add_gc_list(rt);
1496
1497 fib6_start_gc(info->nl_net, rt);
1498 }
1499
1500out:
1501 if (err) {
1502#ifdef CONFIG_IPV6_SUBTREES
1503 /*
1504 * If fib6_add_1 has cleared the old leaf pointer in the
1505 * super-tree leaf node we have to find a new one for it.
1506 */
1507 if (pn != fn) {
1508 struct fib6_info *pn_leaf =
1509 rcu_dereference_protected(pn->leaf,
1510 lockdep_is_held(&table->tb6_lock));
1511 if (pn_leaf == rt) {
1512 pn_leaf = NULL;
1513 RCU_INIT_POINTER(pn->leaf, NULL);
1514 fib6_info_release(rt);
1515 }
1516 if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1517 pn_leaf = fib6_find_prefix(info->nl_net, table,
1518 pn);
1519 if (!pn_leaf)
1520 pn_leaf =
1521 info->nl_net->ipv6.fib6_null_entry;
1522 fib6_info_hold(pn_leaf);
1523 rcu_assign_pointer(pn->leaf, pn_leaf);
1524 }
1525 }
1526#endif
1527 goto failure;
1528 } else if (fib6_requires_src(rt)) {
1529 fib6_routes_require_src_inc(info->nl_net);
1530 }
1531 return err;
1532
1533failure:
1534 /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1535 * 1. fn is an intermediate node and we failed to add the new
1536 * route to it in both subtree creation failure and fib6_add_rt2node()
1537 * failure case.
1538 * 2. fn is the root node in the table and we fail to add the first
1539 * default route to it.
1540 */
1541 if (fn &&
1542 (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1543 (fn->fn_flags & RTN_TL_ROOT &&
1544 !rcu_access_pointer(fn->leaf))))
1545 fib6_repair_tree(info->nl_net, table, fn);
1546 return err;
1547}
1548
1549/*
1550 * Routing tree lookup
1551 *
1552 */
1553
1554struct lookup_args {
1555 int offset; /* key offset on fib6_info */
1556 const struct in6_addr *addr; /* search key */
1557};
1558
1559static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1560 struct lookup_args *args)
1561{
1562 struct fib6_node *fn;
1563 __be32 dir;
1564
1565 if (unlikely(args->offset == 0))
1566 return NULL;
1567
1568 /*
1569 * Descend on a tree
1570 */
1571
1572 fn = root;
1573
1574 for (;;) {
1575 struct fib6_node *next;
1576
1577 dir = addr_bit_set(args->addr, fn->fn_bit);
1578
1579 next = dir ? rcu_dereference(fn->right) :
1580 rcu_dereference(fn->left);
1581
1582 if (next) {
1583 fn = next;
1584 continue;
1585 }
1586 break;
1587 }
1588
1589 while (fn) {
1590 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1591
1592 if (subtree || fn->fn_flags & RTN_RTINFO) {
1593 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1594 struct rt6key *key;
1595
1596 if (!leaf)
1597 goto backtrack;
1598
1599 key = (struct rt6key *) ((u8 *)leaf + args->offset);
1600
1601 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1602#ifdef CONFIG_IPV6_SUBTREES
1603 if (subtree) {
1604 struct fib6_node *sfn;
1605 sfn = fib6_node_lookup_1(subtree,
1606 args + 1);
1607 if (!sfn)
1608 goto backtrack;
1609 fn = sfn;
1610 }
1611#endif
1612 if (fn->fn_flags & RTN_RTINFO)
1613 return fn;
1614 }
1615 }
1616backtrack:
1617 if (fn->fn_flags & RTN_ROOT)
1618 break;
1619
1620 fn = rcu_dereference(fn->parent);
1621 }
1622
1623 return NULL;
1624}
1625
1626/* called with rcu_read_lock() held
1627 */
1628struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1629 const struct in6_addr *daddr,
1630 const struct in6_addr *saddr)
1631{
1632 struct fib6_node *fn;
1633 struct lookup_args args[] = {
1634 {
1635 .offset = offsetof(struct fib6_info, fib6_dst),
1636 .addr = daddr,
1637 },
1638#ifdef CONFIG_IPV6_SUBTREES
1639 {
1640 .offset = offsetof(struct fib6_info, fib6_src),
1641 .addr = saddr,
1642 },
1643#endif
1644 {
1645 .offset = 0, /* sentinel */
1646 }
1647 };
1648
1649 fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1650 if (!fn || fn->fn_flags & RTN_TL_ROOT)
1651 fn = root;
1652
1653 return fn;
1654}
1655
1656/*
1657 * Get node with specified destination prefix (and source prefix,
1658 * if subtrees are used)
1659 * exact_match == true means we try to find fn with exact match of
1660 * the passed in prefix addr
1661 * exact_match == false means we try to find fn with longest prefix
1662 * match of the passed in prefix addr. This is useful for finding fn
1663 * for cached route as it will be stored in the exception table under
1664 * the node with longest prefix length.
1665 */
1666
1667
1668static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1669 const struct in6_addr *addr,
1670 int plen, int offset,
1671 bool exact_match)
1672{
1673 struct fib6_node *fn, *prev = NULL;
1674
1675 for (fn = root; fn ; ) {
1676 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1677 struct rt6key *key;
1678
1679 /* This node is being deleted */
1680 if (!leaf) {
1681 if (plen <= fn->fn_bit)
1682 goto out;
1683 else
1684 goto next;
1685 }
1686
1687 key = (struct rt6key *)((u8 *)leaf + offset);
1688
1689 /*
1690 * Prefix match
1691 */
1692 if (plen < fn->fn_bit ||
1693 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1694 goto out;
1695
1696 if (plen == fn->fn_bit)
1697 return fn;
1698
1699 if (fn->fn_flags & RTN_RTINFO)
1700 prev = fn;
1701
1702next:
1703 /*
1704 * We have more bits to go
1705 */
1706 if (addr_bit_set(addr, fn->fn_bit))
1707 fn = rcu_dereference(fn->right);
1708 else
1709 fn = rcu_dereference(fn->left);
1710 }
1711out:
1712 if (exact_match)
1713 return NULL;
1714 else
1715 return prev;
1716}
1717
1718struct fib6_node *fib6_locate(struct fib6_node *root,
1719 const struct in6_addr *daddr, int dst_len,
1720 const struct in6_addr *saddr, int src_len,
1721 bool exact_match)
1722{
1723 struct fib6_node *fn;
1724
1725 fn = fib6_locate_1(root, daddr, dst_len,
1726 offsetof(struct fib6_info, fib6_dst),
1727 exact_match);
1728
1729#ifdef CONFIG_IPV6_SUBTREES
1730 if (src_len) {
1731 WARN_ON(saddr == NULL);
1732 if (fn) {
1733 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1734
1735 if (subtree) {
1736 fn = fib6_locate_1(subtree, saddr, src_len,
1737 offsetof(struct fib6_info, fib6_src),
1738 exact_match);
1739 }
1740 }
1741 }
1742#endif
1743
1744 if (fn && fn->fn_flags & RTN_RTINFO)
1745 return fn;
1746
1747 return NULL;
1748}
1749
1750
1751/*
1752 * Deletion
1753 *
1754 */
1755
1756static struct fib6_info *fib6_find_prefix(struct net *net,
1757 struct fib6_table *table,
1758 struct fib6_node *fn)
1759{
1760 struct fib6_node *child_left, *child_right;
1761
1762 if (fn->fn_flags & RTN_ROOT)
1763 return net->ipv6.fib6_null_entry;
1764
1765 while (fn) {
1766 child_left = rcu_dereference_protected(fn->left,
1767 lockdep_is_held(&table->tb6_lock));
1768 child_right = rcu_dereference_protected(fn->right,
1769 lockdep_is_held(&table->tb6_lock));
1770 if (child_left)
1771 return rcu_dereference_protected(child_left->leaf,
1772 lockdep_is_held(&table->tb6_lock));
1773 if (child_right)
1774 return rcu_dereference_protected(child_right->leaf,
1775 lockdep_is_held(&table->tb6_lock));
1776
1777 fn = FIB6_SUBTREE(fn);
1778 }
1779 return NULL;
1780}
1781
1782/*
1783 * Called to trim the tree of intermediate nodes when possible. "fn"
1784 * is the node we want to try and remove.
1785 * Need to own table->tb6_lock
1786 */
1787
1788static struct fib6_node *fib6_repair_tree(struct net *net,
1789 struct fib6_table *table,
1790 struct fib6_node *fn)
1791{
1792 int children;
1793 int nstate;
1794 struct fib6_node *child;
1795 struct fib6_walker *w;
1796 int iter = 0;
1797
1798 /* Set fn->leaf to null_entry for root node. */
1799 if (fn->fn_flags & RTN_TL_ROOT) {
1800 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1801 return fn;
1802 }
1803
1804 for (;;) {
1805 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1806 lockdep_is_held(&table->tb6_lock));
1807 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1808 lockdep_is_held(&table->tb6_lock));
1809 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1810 lockdep_is_held(&table->tb6_lock));
1811 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1812 lockdep_is_held(&table->tb6_lock));
1813 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1814 lockdep_is_held(&table->tb6_lock));
1815 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1816 lockdep_is_held(&table->tb6_lock));
1817 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1818 lockdep_is_held(&table->tb6_lock));
1819 struct fib6_info *new_fn_leaf;
1820
1821 pr_debug("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1822 iter++;
1823
1824 WARN_ON(fn->fn_flags & RTN_RTINFO);
1825 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1826 WARN_ON(fn_leaf);
1827
1828 children = 0;
1829 child = NULL;
1830 if (fn_r) {
1831 child = fn_r;
1832 children |= 1;
1833 }
1834 if (fn_l) {
1835 child = fn_l;
1836 children |= 2;
1837 }
1838
1839 if (children == 3 || FIB6_SUBTREE(fn)
1840#ifdef CONFIG_IPV6_SUBTREES
1841 /* Subtree root (i.e. fn) may have one child */
1842 || (children && fn->fn_flags & RTN_ROOT)
1843#endif
1844 ) {
1845 new_fn_leaf = fib6_find_prefix(net, table, fn);
1846#if RT6_DEBUG >= 2
1847 if (!new_fn_leaf) {
1848 WARN_ON(!new_fn_leaf);
1849 new_fn_leaf = net->ipv6.fib6_null_entry;
1850 }
1851#endif
1852 fib6_info_hold(new_fn_leaf);
1853 rcu_assign_pointer(fn->leaf, new_fn_leaf);
1854 return pn;
1855 }
1856
1857#ifdef CONFIG_IPV6_SUBTREES
1858 if (FIB6_SUBTREE(pn) == fn) {
1859 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1860 RCU_INIT_POINTER(pn->subtree, NULL);
1861 nstate = FWS_L;
1862 } else {
1863 WARN_ON(fn->fn_flags & RTN_ROOT);
1864#endif
1865 if (pn_r == fn)
1866 rcu_assign_pointer(pn->right, child);
1867 else if (pn_l == fn)
1868 rcu_assign_pointer(pn->left, child);
1869#if RT6_DEBUG >= 2
1870 else
1871 WARN_ON(1);
1872#endif
1873 if (child)
1874 rcu_assign_pointer(child->parent, pn);
1875 nstate = FWS_R;
1876#ifdef CONFIG_IPV6_SUBTREES
1877 }
1878#endif
1879
1880 read_lock(&net->ipv6.fib6_walker_lock);
1881 FOR_WALKERS(net, w) {
1882 if (!child) {
1883 if (w->node == fn) {
1884 pr_debug("W %p adjusted by delnode 1, s=%d/%d\n",
1885 w, w->state, nstate);
1886 w->node = pn;
1887 w->state = nstate;
1888 }
1889 } else {
1890 if (w->node == fn) {
1891 w->node = child;
1892 if (children&2) {
1893 pr_debug("W %p adjusted by delnode 2, s=%d\n",
1894 w, w->state);
1895 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1896 } else {
1897 pr_debug("W %p adjusted by delnode 2, s=%d\n",
1898 w, w->state);
1899 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1900 }
1901 }
1902 }
1903 }
1904 read_unlock(&net->ipv6.fib6_walker_lock);
1905
1906 node_free(net, fn);
1907 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1908 return pn;
1909
1910 RCU_INIT_POINTER(pn->leaf, NULL);
1911 fib6_info_release(pn_leaf);
1912 fn = pn;
1913 }
1914}
1915
1916static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1917 struct fib6_info __rcu **rtp, struct nl_info *info)
1918{
1919 struct fib6_info *leaf, *replace_rt = NULL;
1920 struct fib6_walker *w;
1921 struct fib6_info *rt = rcu_dereference_protected(*rtp,
1922 lockdep_is_held(&table->tb6_lock));
1923 struct net *net = info->nl_net;
1924 bool notify_del = false;
1925
1926 /* If the deleted route is the first in the node and it is not part of
1927 * a multipath route, then we need to replace it with the next route
1928 * in the node, if exists.
1929 */
1930 leaf = rcu_dereference_protected(fn->leaf,
1931 lockdep_is_held(&table->tb6_lock));
1932 if (leaf == rt && !rt->fib6_nsiblings) {
1933 if (rcu_access_pointer(rt->fib6_next))
1934 replace_rt = rcu_dereference_protected(rt->fib6_next,
1935 lockdep_is_held(&table->tb6_lock));
1936 else
1937 notify_del = true;
1938 }
1939
1940 /* Unlink it */
1941 *rtp = rt->fib6_next;
1942 rt->fib6_node = NULL;
1943 net->ipv6.rt6_stats->fib_rt_entries--;
1944 net->ipv6.rt6_stats->fib_discarded_routes++;
1945
1946 /* Reset round-robin state, if necessary */
1947 if (rcu_access_pointer(fn->rr_ptr) == rt)
1948 fn->rr_ptr = NULL;
1949
1950 /* Remove this entry from other siblings */
1951 if (rt->fib6_nsiblings) {
1952 struct fib6_info *sibling, *next_sibling;
1953
1954 /* The route is deleted from a multipath route. If this
1955 * multipath route is the first route in the node, then we need
1956 * to emit a delete notification. Otherwise, we need to skip
1957 * the notification.
1958 */
1959 if (rt->fib6_metric == leaf->fib6_metric &&
1960 rt6_qualify_for_ecmp(leaf))
1961 notify_del = true;
1962 list_for_each_entry_safe(sibling, next_sibling,
1963 &rt->fib6_siblings, fib6_siblings)
1964 sibling->fib6_nsiblings--;
1965 rt->fib6_nsiblings = 0;
1966 list_del_rcu(&rt->fib6_siblings);
1967 rt6_multipath_rebalance(next_sibling);
1968 }
1969
1970 /* Adjust walkers */
1971 read_lock(&net->ipv6.fib6_walker_lock);
1972 FOR_WALKERS(net, w) {
1973 if (w->state == FWS_C && w->leaf == rt) {
1974 pr_debug("walker %p adjusted by delroute\n", w);
1975 w->leaf = rcu_dereference_protected(rt->fib6_next,
1976 lockdep_is_held(&table->tb6_lock));
1977 if (!w->leaf)
1978 w->state = FWS_U;
1979 }
1980 }
1981 read_unlock(&net->ipv6.fib6_walker_lock);
1982
1983 /* If it was last route, call fib6_repair_tree() to:
1984 * 1. For root node, put back null_entry as how the table was created.
1985 * 2. For other nodes, expunge its radix tree node.
1986 */
1987 if (!rcu_access_pointer(fn->leaf)) {
1988 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1989 fn->fn_flags &= ~RTN_RTINFO;
1990 net->ipv6.rt6_stats->fib_route_nodes--;
1991 }
1992 fn = fib6_repair_tree(net, table, fn);
1993 }
1994
1995 fib6_purge_rt(rt, fn, net);
1996
1997 if (!info->skip_notify_kernel) {
1998 if (notify_del)
1999 call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
2000 rt, NULL);
2001 else if (replace_rt)
2002 call_fib6_entry_notifiers_replace(net, replace_rt);
2003 }
2004 if (!info->skip_notify)
2005 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
2006
2007 fib6_info_release(rt);
2008}
2009
2010/* Need to own table->tb6_lock */
2011int fib6_del(struct fib6_info *rt, struct nl_info *info)
2012{
2013 struct net *net = info->nl_net;
2014 struct fib6_info __rcu **rtp;
2015 struct fib6_info __rcu **rtp_next;
2016 struct fib6_table *table;
2017 struct fib6_node *fn;
2018
2019 if (rt == net->ipv6.fib6_null_entry)
2020 return -ENOENT;
2021
2022 table = rt->fib6_table;
2023 fn = rcu_dereference_protected(rt->fib6_node,
2024 lockdep_is_held(&table->tb6_lock));
2025 if (!fn)
2026 return -ENOENT;
2027
2028 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
2029
2030 /*
2031 * Walk the leaf entries looking for ourself
2032 */
2033
2034 for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
2035 struct fib6_info *cur = rcu_dereference_protected(*rtp,
2036 lockdep_is_held(&table->tb6_lock));
2037 if (rt == cur) {
2038 if (fib6_requires_src(cur))
2039 fib6_routes_require_src_dec(info->nl_net);
2040 fib6_del_route(table, fn, rtp, info);
2041 return 0;
2042 }
2043 rtp_next = &cur->fib6_next;
2044 }
2045 return -ENOENT;
2046}
2047
2048/*
2049 * Tree traversal function.
2050 *
2051 * Certainly, it is not interrupt safe.
2052 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
2053 * It means, that we can modify tree during walking
2054 * and use this function for garbage collection, clone pruning,
2055 * cleaning tree when a device goes down etc. etc.
2056 *
2057 * It guarantees that every node will be traversed,
2058 * and that it will be traversed only once.
2059 *
2060 * Callback function w->func may return:
2061 * 0 -> continue walking.
2062 * positive value -> walking is suspended (used by tree dumps,
2063 * and probably by gc, if it will be split to several slices)
2064 * negative value -> terminate walking.
2065 *
2066 * The function itself returns:
2067 * 0 -> walk is complete.
2068 * >0 -> walk is incomplete (i.e. suspended)
2069 * <0 -> walk is terminated by an error.
2070 *
2071 * This function is called with tb6_lock held.
2072 */
2073
2074static int fib6_walk_continue(struct fib6_walker *w)
2075{
2076 struct fib6_node *fn, *pn, *left, *right;
2077
2078 /* w->root should always be table->tb6_root */
2079 WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
2080
2081 for (;;) {
2082 fn = w->node;
2083 if (!fn)
2084 return 0;
2085
2086 switch (w->state) {
2087#ifdef CONFIG_IPV6_SUBTREES
2088 case FWS_S:
2089 if (FIB6_SUBTREE(fn)) {
2090 w->node = FIB6_SUBTREE(fn);
2091 continue;
2092 }
2093 w->state = FWS_L;
2094 fallthrough;
2095#endif
2096 case FWS_L:
2097 left = rcu_dereference_protected(fn->left, 1);
2098 if (left) {
2099 w->node = left;
2100 w->state = FWS_INIT;
2101 continue;
2102 }
2103 w->state = FWS_R;
2104 fallthrough;
2105 case FWS_R:
2106 right = rcu_dereference_protected(fn->right, 1);
2107 if (right) {
2108 w->node = right;
2109 w->state = FWS_INIT;
2110 continue;
2111 }
2112 w->state = FWS_C;
2113 w->leaf = rcu_dereference_protected(fn->leaf, 1);
2114 fallthrough;
2115 case FWS_C:
2116 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
2117 int err;
2118
2119 if (w->skip) {
2120 w->skip--;
2121 goto skip;
2122 }
2123
2124 err = w->func(w);
2125 if (err)
2126 return err;
2127
2128 w->count++;
2129 continue;
2130 }
2131skip:
2132 w->state = FWS_U;
2133 fallthrough;
2134 case FWS_U:
2135 if (fn == w->root)
2136 return 0;
2137 pn = rcu_dereference_protected(fn->parent, 1);
2138 left = rcu_dereference_protected(pn->left, 1);
2139 right = rcu_dereference_protected(pn->right, 1);
2140 w->node = pn;
2141#ifdef CONFIG_IPV6_SUBTREES
2142 if (FIB6_SUBTREE(pn) == fn) {
2143 WARN_ON(!(fn->fn_flags & RTN_ROOT));
2144 w->state = FWS_L;
2145 continue;
2146 }
2147#endif
2148 if (left == fn) {
2149 w->state = FWS_R;
2150 continue;
2151 }
2152 if (right == fn) {
2153 w->state = FWS_C;
2154 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
2155 continue;
2156 }
2157#if RT6_DEBUG >= 2
2158 WARN_ON(1);
2159#endif
2160 }
2161 }
2162}
2163
2164static int fib6_walk(struct net *net, struct fib6_walker *w)
2165{
2166 int res;
2167
2168 w->state = FWS_INIT;
2169 w->node = w->root;
2170
2171 fib6_walker_link(net, w);
2172 res = fib6_walk_continue(w);
2173 if (res <= 0)
2174 fib6_walker_unlink(net, w);
2175 return res;
2176}
2177
2178static int fib6_clean_node(struct fib6_walker *w)
2179{
2180 int res;
2181 struct fib6_info *rt;
2182 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
2183 struct nl_info info = {
2184 .nl_net = c->net,
2185 .skip_notify = c->skip_notify,
2186 };
2187
2188 if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
2189 READ_ONCE(w->node->fn_sernum) != c->sernum)
2190 WRITE_ONCE(w->node->fn_sernum, c->sernum);
2191
2192 if (!c->func) {
2193 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
2194 w->leaf = NULL;
2195 return 0;
2196 }
2197
2198 for_each_fib6_walker_rt(w) {
2199 res = c->func(rt, c->arg);
2200 if (res == -1) {
2201 w->leaf = rt;
2202 res = fib6_del(rt, &info);
2203 if (res) {
2204#if RT6_DEBUG >= 2
2205 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2206 __func__, rt,
2207 rcu_access_pointer(rt->fib6_node),
2208 res);
2209#endif
2210 continue;
2211 }
2212 return 0;
2213 } else if (res == -2) {
2214 if (WARN_ON(!rt->fib6_nsiblings))
2215 continue;
2216 rt = list_last_entry(&rt->fib6_siblings,
2217 struct fib6_info, fib6_siblings);
2218 continue;
2219 }
2220 WARN_ON(res != 0);
2221 }
2222 w->leaf = rt;
2223 return 0;
2224}
2225
2226/*
2227 * Convenient frontend to tree walker.
2228 *
2229 * func is called on each route.
2230 * It may return -2 -> skip multipath route.
2231 * -1 -> delete this route.
2232 * 0 -> continue walking
2233 */
2234
2235static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2236 int (*func)(struct fib6_info *, void *arg),
2237 int sernum, void *arg, bool skip_notify)
2238{
2239 struct fib6_cleaner c;
2240
2241 c.w.root = root;
2242 c.w.func = fib6_clean_node;
2243 c.w.count = 0;
2244 c.w.skip = 0;
2245 c.w.skip_in_node = 0;
2246 c.func = func;
2247 c.sernum = sernum;
2248 c.arg = arg;
2249 c.net = net;
2250 c.skip_notify = skip_notify;
2251
2252 fib6_walk(net, &c.w);
2253}
2254
2255static void __fib6_clean_all(struct net *net,
2256 int (*func)(struct fib6_info *, void *),
2257 int sernum, void *arg, bool skip_notify)
2258{
2259 struct fib6_table *table;
2260 struct hlist_head *head;
2261 unsigned int h;
2262
2263 rcu_read_lock();
2264 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2265 head = &net->ipv6.fib_table_hash[h];
2266 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2267 spin_lock_bh(&table->tb6_lock);
2268 fib6_clean_tree(net, &table->tb6_root,
2269 func, sernum, arg, skip_notify);
2270 spin_unlock_bh(&table->tb6_lock);
2271 }
2272 }
2273 rcu_read_unlock();
2274}
2275
2276void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2277 void *arg)
2278{
2279 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2280}
2281
2282void fib6_clean_all_skip_notify(struct net *net,
2283 int (*func)(struct fib6_info *, void *),
2284 void *arg)
2285{
2286 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2287}
2288
2289static void fib6_flush_trees(struct net *net)
2290{
2291 int new_sernum = fib6_new_sernum(net);
2292
2293 __fib6_clean_all(net, NULL, new_sernum, NULL, false);
2294}
2295
2296/*
2297 * Garbage collection
2298 */
2299
2300static int fib6_age(struct fib6_info *rt, struct fib6_gc_args *gc_args)
2301{
2302 unsigned long now = jiffies;
2303
2304 /*
2305 * check addrconf expiration here.
2306 * Routes are expired even if they are in use.
2307 */
2308
2309 if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2310 if (time_after(now, rt->expires)) {
2311 pr_debug("expiring %p\n", rt);
2312 return -1;
2313 }
2314 gc_args->more++;
2315 }
2316
2317 /* Also age clones in the exception table.
2318 * Note, that clones are aged out
2319 * only if they are not in use now.
2320 */
2321 rt6_age_exceptions(rt, gc_args, now);
2322
2323 return 0;
2324}
2325
2326static void fib6_gc_table(struct net *net,
2327 struct fib6_table *tb6,
2328 struct fib6_gc_args *gc_args)
2329{
2330 struct fib6_info *rt;
2331 struct hlist_node *n;
2332 struct nl_info info = {
2333 .nl_net = net,
2334 .skip_notify = false,
2335 };
2336
2337 hlist_for_each_entry_safe(rt, n, &tb6->tb6_gc_hlist, gc_link)
2338 if (fib6_age(rt, gc_args) == -1)
2339 fib6_del(rt, &info);
2340}
2341
2342static void fib6_gc_all(struct net *net, struct fib6_gc_args *gc_args)
2343{
2344 struct fib6_table *table;
2345 struct hlist_head *head;
2346 unsigned int h;
2347
2348 rcu_read_lock();
2349 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2350 head = &net->ipv6.fib_table_hash[h];
2351 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2352 spin_lock_bh(&table->tb6_lock);
2353
2354 fib6_gc_table(net, table, gc_args);
2355
2356 spin_unlock_bh(&table->tb6_lock);
2357 }
2358 }
2359 rcu_read_unlock();
2360}
2361
2362void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2363{
2364 struct fib6_gc_args gc_args;
2365 unsigned long now;
2366
2367 if (force) {
2368 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2369 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2370 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2371 return;
2372 }
2373 gc_args.timeout = expires ? (int)expires :
2374 net->ipv6.sysctl.ip6_rt_gc_interval;
2375 gc_args.more = 0;
2376
2377 fib6_gc_all(net, &gc_args);
2378 now = jiffies;
2379 net->ipv6.ip6_rt_last_gc = now;
2380
2381 if (gc_args.more)
2382 mod_timer(&net->ipv6.ip6_fib_timer,
2383 round_jiffies(now
2384 + net->ipv6.sysctl.ip6_rt_gc_interval));
2385 else
2386 del_timer(&net->ipv6.ip6_fib_timer);
2387 spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2388}
2389
2390static void fib6_gc_timer_cb(struct timer_list *t)
2391{
2392 struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2393
2394 fib6_run_gc(0, arg, true);
2395}
2396
2397static int __net_init fib6_net_init(struct net *net)
2398{
2399 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2400 int err;
2401
2402 err = fib6_notifier_init(net);
2403 if (err)
2404 return err;
2405
2406 /* Default to 3-tuple */
2407 net->ipv6.sysctl.multipath_hash_fields =
2408 FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK;
2409
2410 spin_lock_init(&net->ipv6.fib6_gc_lock);
2411 rwlock_init(&net->ipv6.fib6_walker_lock);
2412 INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2413 timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2414
2415 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2416 if (!net->ipv6.rt6_stats)
2417 goto out_notifier;
2418
2419 /* Avoid false sharing : Use at least a full cache line */
2420 size = max_t(size_t, size, L1_CACHE_BYTES);
2421
2422 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2423 if (!net->ipv6.fib_table_hash)
2424 goto out_rt6_stats;
2425
2426 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2427 GFP_KERNEL);
2428 if (!net->ipv6.fib6_main_tbl)
2429 goto out_fib_table_hash;
2430
2431 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2432 rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2433 net->ipv6.fib6_null_entry);
2434 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2435 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2436 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2437 INIT_HLIST_HEAD(&net->ipv6.fib6_main_tbl->tb6_gc_hlist);
2438
2439#ifdef CONFIG_IPV6_MULTIPLE_TABLES
2440 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2441 GFP_KERNEL);
2442 if (!net->ipv6.fib6_local_tbl)
2443 goto out_fib6_main_tbl;
2444 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2445 rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2446 net->ipv6.fib6_null_entry);
2447 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2448 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2449 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2450 INIT_HLIST_HEAD(&net->ipv6.fib6_local_tbl->tb6_gc_hlist);
2451#endif
2452 fib6_tables_init(net);
2453
2454 return 0;
2455
2456#ifdef CONFIG_IPV6_MULTIPLE_TABLES
2457out_fib6_main_tbl:
2458 kfree(net->ipv6.fib6_main_tbl);
2459#endif
2460out_fib_table_hash:
2461 kfree(net->ipv6.fib_table_hash);
2462out_rt6_stats:
2463 kfree(net->ipv6.rt6_stats);
2464out_notifier:
2465 fib6_notifier_exit(net);
2466 return -ENOMEM;
2467}
2468
2469static void fib6_net_exit(struct net *net)
2470{
2471 unsigned int i;
2472
2473 del_timer_sync(&net->ipv6.ip6_fib_timer);
2474
2475 for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2476 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2477 struct hlist_node *tmp;
2478 struct fib6_table *tb;
2479
2480 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2481 hlist_del(&tb->tb6_hlist);
2482 fib6_free_table(tb);
2483 }
2484 }
2485
2486 kfree(net->ipv6.fib_table_hash);
2487 kfree(net->ipv6.rt6_stats);
2488 fib6_notifier_exit(net);
2489}
2490
2491static struct pernet_operations fib6_net_ops = {
2492 .init = fib6_net_init,
2493 .exit = fib6_net_exit,
2494};
2495
2496static const struct rtnl_msg_handler fib6_rtnl_msg_handlers[] __initconst_or_module = {
2497 {.owner = THIS_MODULE, .protocol = PF_INET6, .msgtype = RTM_GETROUTE,
2498 .dumpit = inet6_dump_fib,
2499 .flags = RTNL_FLAG_DUMP_UNLOCKED | RTNL_FLAG_DUMP_SPLIT_NLM_DONE},
2500};
2501
2502int __init fib6_init(void)
2503{
2504 int ret = -ENOMEM;
2505
2506 fib6_node_kmem = KMEM_CACHE(fib6_node,
2507 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT);
2508 if (!fib6_node_kmem)
2509 goto out;
2510
2511 ret = register_pernet_subsys(&fib6_net_ops);
2512 if (ret)
2513 goto out_kmem_cache_create;
2514
2515 ret = rtnl_register_many(fib6_rtnl_msg_handlers);
2516 if (ret)
2517 goto out_unregister_subsys;
2518
2519 __fib6_flush_trees = fib6_flush_trees;
2520out:
2521 return ret;
2522
2523out_unregister_subsys:
2524 unregister_pernet_subsys(&fib6_net_ops);
2525out_kmem_cache_create:
2526 kmem_cache_destroy(fib6_node_kmem);
2527 goto out;
2528}
2529
2530void fib6_gc_cleanup(void)
2531{
2532 unregister_pernet_subsys(&fib6_net_ops);
2533 kmem_cache_destroy(fib6_node_kmem);
2534}
2535
2536#ifdef CONFIG_PROC_FS
2537static int ipv6_route_native_seq_show(struct seq_file *seq, void *v)
2538{
2539 struct fib6_info *rt = v;
2540 struct ipv6_route_iter *iter = seq->private;
2541 struct fib6_nh *fib6_nh = rt->fib6_nh;
2542 unsigned int flags = rt->fib6_flags;
2543 const struct net_device *dev;
2544
2545 if (rt->nh)
2546 fib6_nh = nexthop_fib6_nh(rt->nh);
2547
2548 seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2549
2550#ifdef CONFIG_IPV6_SUBTREES
2551 seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2552#else
2553 seq_puts(seq, "00000000000000000000000000000000 00 ");
2554#endif
2555 if (fib6_nh->fib_nh_gw_family) {
2556 flags |= RTF_GATEWAY;
2557 seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6);
2558 } else {
2559 seq_puts(seq, "00000000000000000000000000000000");
2560 }
2561
2562 dev = fib6_nh->fib_nh_dev;
2563 seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2564 rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2565 flags, dev ? dev->name : "");
2566 iter->w.leaf = NULL;
2567 return 0;
2568}
2569
2570static int ipv6_route_yield(struct fib6_walker *w)
2571{
2572 struct ipv6_route_iter *iter = w->args;
2573
2574 if (!iter->skip)
2575 return 1;
2576
2577 do {
2578 iter->w.leaf = rcu_dereference_protected(
2579 iter->w.leaf->fib6_next,
2580 lockdep_is_held(&iter->tbl->tb6_lock));
2581 iter->skip--;
2582 if (!iter->skip && iter->w.leaf)
2583 return 1;
2584 } while (iter->w.leaf);
2585
2586 return 0;
2587}
2588
2589static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2590 struct net *net)
2591{
2592 memset(&iter->w, 0, sizeof(iter->w));
2593 iter->w.func = ipv6_route_yield;
2594 iter->w.root = &iter->tbl->tb6_root;
2595 iter->w.state = FWS_INIT;
2596 iter->w.node = iter->w.root;
2597 iter->w.args = iter;
2598 iter->sernum = READ_ONCE(iter->w.root->fn_sernum);
2599 INIT_LIST_HEAD(&iter->w.lh);
2600 fib6_walker_link(net, &iter->w);
2601}
2602
2603static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2604 struct net *net)
2605{
2606 unsigned int h;
2607 struct hlist_node *node;
2608
2609 if (tbl) {
2610 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2611 node = rcu_dereference(hlist_next_rcu(&tbl->tb6_hlist));
2612 } else {
2613 h = 0;
2614 node = NULL;
2615 }
2616
2617 while (!node && h < FIB6_TABLE_HASHSZ) {
2618 node = rcu_dereference(
2619 hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2620 }
2621 return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2622}
2623
2624static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2625{
2626 int sernum = READ_ONCE(iter->w.root->fn_sernum);
2627
2628 if (iter->sernum != sernum) {
2629 iter->sernum = sernum;
2630 iter->w.state = FWS_INIT;
2631 iter->w.node = iter->w.root;
2632 WARN_ON(iter->w.skip);
2633 iter->w.skip = iter->w.count;
2634 }
2635}
2636
2637static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2638{
2639 int r;
2640 struct fib6_info *n;
2641 struct net *net = seq_file_net(seq);
2642 struct ipv6_route_iter *iter = seq->private;
2643
2644 ++(*pos);
2645 if (!v)
2646 goto iter_table;
2647
2648 n = rcu_dereference(((struct fib6_info *)v)->fib6_next);
2649 if (n)
2650 return n;
2651
2652iter_table:
2653 ipv6_route_check_sernum(iter);
2654 spin_lock_bh(&iter->tbl->tb6_lock);
2655 r = fib6_walk_continue(&iter->w);
2656 spin_unlock_bh(&iter->tbl->tb6_lock);
2657 if (r > 0) {
2658 return iter->w.leaf;
2659 } else if (r < 0) {
2660 fib6_walker_unlink(net, &iter->w);
2661 return NULL;
2662 }
2663 fib6_walker_unlink(net, &iter->w);
2664
2665 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2666 if (!iter->tbl)
2667 return NULL;
2668
2669 ipv6_route_seq_setup_walk(iter, net);
2670 goto iter_table;
2671}
2672
2673static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2674 __acquires(RCU)
2675{
2676 struct net *net = seq_file_net(seq);
2677 struct ipv6_route_iter *iter = seq->private;
2678
2679 rcu_read_lock();
2680 iter->tbl = ipv6_route_seq_next_table(NULL, net);
2681 iter->skip = *pos;
2682
2683 if (iter->tbl) {
2684 loff_t p = 0;
2685
2686 ipv6_route_seq_setup_walk(iter, net);
2687 return ipv6_route_seq_next(seq, NULL, &p);
2688 } else {
2689 return NULL;
2690 }
2691}
2692
2693static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2694{
2695 struct fib6_walker *w = &iter->w;
2696 return w->node && !(w->state == FWS_U && w->node == w->root);
2697}
2698
2699static void ipv6_route_native_seq_stop(struct seq_file *seq, void *v)
2700 __releases(RCU)
2701{
2702 struct net *net = seq_file_net(seq);
2703 struct ipv6_route_iter *iter = seq->private;
2704
2705 if (ipv6_route_iter_active(iter))
2706 fib6_walker_unlink(net, &iter->w);
2707
2708 rcu_read_unlock();
2709}
2710
2711#if IS_BUILTIN(CONFIG_IPV6) && defined(CONFIG_BPF_SYSCALL)
2712static int ipv6_route_prog_seq_show(struct bpf_prog *prog,
2713 struct bpf_iter_meta *meta,
2714 void *v)
2715{
2716 struct bpf_iter__ipv6_route ctx;
2717
2718 ctx.meta = meta;
2719 ctx.rt = v;
2720 return bpf_iter_run_prog(prog, &ctx);
2721}
2722
2723static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2724{
2725 struct ipv6_route_iter *iter = seq->private;
2726 struct bpf_iter_meta meta;
2727 struct bpf_prog *prog;
2728 int ret;
2729
2730 meta.seq = seq;
2731 prog = bpf_iter_get_info(&meta, false);
2732 if (!prog)
2733 return ipv6_route_native_seq_show(seq, v);
2734
2735 ret = ipv6_route_prog_seq_show(prog, &meta, v);
2736 iter->w.leaf = NULL;
2737
2738 return ret;
2739}
2740
2741static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2742{
2743 struct bpf_iter_meta meta;
2744 struct bpf_prog *prog;
2745
2746 if (!v) {
2747 meta.seq = seq;
2748 prog = bpf_iter_get_info(&meta, true);
2749 if (prog)
2750 (void)ipv6_route_prog_seq_show(prog, &meta, v);
2751 }
2752
2753 ipv6_route_native_seq_stop(seq, v);
2754}
2755#else
2756static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2757{
2758 return ipv6_route_native_seq_show(seq, v);
2759}
2760
2761static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2762{
2763 ipv6_route_native_seq_stop(seq, v);
2764}
2765#endif
2766
2767const struct seq_operations ipv6_route_seq_ops = {
2768 .start = ipv6_route_seq_start,
2769 .next = ipv6_route_seq_next,
2770 .stop = ipv6_route_seq_stop,
2771 .show = ipv6_route_seq_show
2772};
2773#endif /* CONFIG_PROC_FS */