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