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 = pol_lookup_func(lookup,
318 net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
319 if (rt->dst.error == -EAGAIN) {
320 ip6_rt_put_flags(rt, flags);
321 rt = net->ipv6.ip6_null_entry;
322 if (!(flags & RT6_LOOKUP_F_DST_NOREF))
323 dst_hold(&rt->dst);
324 }
325
326 return &rt->dst;
327}
328
329/* called with rcu lock held; no reference taken on fib6_info */
330int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
331 struct fib6_result *res, int flags)
332{
333 return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6,
334 res, flags);
335}
336
337static void __net_init fib6_tables_init(struct net *net)
338{
339 fib6_link_table(net, net->ipv6.fib6_main_tbl);
340}
341
342#endif
343
344unsigned int fib6_tables_seq_read(struct net *net)
345{
346 unsigned int h, fib_seq = 0;
347
348 rcu_read_lock();
349 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
350 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
351 struct fib6_table *tb;
352
353 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
354 fib_seq += tb->fib_seq;
355 }
356 rcu_read_unlock();
357
358 return fib_seq;
359}
360
361static int call_fib6_entry_notifier(struct notifier_block *nb,
362 enum fib_event_type event_type,
363 struct fib6_info *rt,
364 struct netlink_ext_ack *extack)
365{
366 struct fib6_entry_notifier_info info = {
367 .info.extack = extack,
368 .rt = rt,
369 };
370
371 return call_fib6_notifier(nb, event_type, &info.info);
372}
373
374static int call_fib6_multipath_entry_notifier(struct notifier_block *nb,
375 enum fib_event_type event_type,
376 struct fib6_info *rt,
377 unsigned int nsiblings,
378 struct netlink_ext_ack *extack)
379{
380 struct fib6_entry_notifier_info info = {
381 .info.extack = extack,
382 .rt = rt,
383 .nsiblings = nsiblings,
384 };
385
386 return call_fib6_notifier(nb, event_type, &info.info);
387}
388
389int call_fib6_entry_notifiers(struct net *net,
390 enum fib_event_type event_type,
391 struct fib6_info *rt,
392 struct netlink_ext_ack *extack)
393{
394 struct fib6_entry_notifier_info info = {
395 .info.extack = extack,
396 .rt = rt,
397 };
398
399 rt->fib6_table->fib_seq++;
400 return call_fib6_notifiers(net, event_type, &info.info);
401}
402
403int call_fib6_multipath_entry_notifiers(struct net *net,
404 enum fib_event_type event_type,
405 struct fib6_info *rt,
406 unsigned int nsiblings,
407 struct netlink_ext_ack *extack)
408{
409 struct fib6_entry_notifier_info info = {
410 .info.extack = extack,
411 .rt = rt,
412 .nsiblings = nsiblings,
413 };
414
415 rt->fib6_table->fib_seq++;
416 return call_fib6_notifiers(net, event_type, &info.info);
417}
418
419int call_fib6_entry_notifiers_replace(struct net *net, struct fib6_info *rt)
420{
421 struct fib6_entry_notifier_info info = {
422 .rt = rt,
423 .nsiblings = rt->fib6_nsiblings,
424 };
425
426 rt->fib6_table->fib_seq++;
427 return call_fib6_notifiers(net, FIB_EVENT_ENTRY_REPLACE, &info.info);
428}
429
430struct fib6_dump_arg {
431 struct net *net;
432 struct notifier_block *nb;
433 struct netlink_ext_ack *extack;
434};
435
436static int fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
437{
438 enum fib_event_type fib_event = FIB_EVENT_ENTRY_REPLACE;
439 int err;
440
441 if (!rt || rt == arg->net->ipv6.fib6_null_entry)
442 return 0;
443
444 if (rt->fib6_nsiblings)
445 err = call_fib6_multipath_entry_notifier(arg->nb, fib_event,
446 rt,
447 rt->fib6_nsiblings,
448 arg->extack);
449 else
450 err = call_fib6_entry_notifier(arg->nb, fib_event, rt,
451 arg->extack);
452
453 return err;
454}
455
456static int fib6_node_dump(struct fib6_walker *w)
457{
458 int err;
459
460 err = fib6_rt_dump(w->leaf, w->args);
461 w->leaf = NULL;
462 return err;
463}
464
465static int fib6_table_dump(struct net *net, struct fib6_table *tb,
466 struct fib6_walker *w)
467{
468 int err;
469
470 w->root = &tb->tb6_root;
471 spin_lock_bh(&tb->tb6_lock);
472 err = fib6_walk(net, w);
473 spin_unlock_bh(&tb->tb6_lock);
474 return err;
475}
476
477/* Called with rcu_read_lock() */
478int fib6_tables_dump(struct net *net, struct notifier_block *nb,
479 struct netlink_ext_ack *extack)
480{
481 struct fib6_dump_arg arg;
482 struct fib6_walker *w;
483 unsigned int h;
484 int err = 0;
485
486 w = kzalloc(sizeof(*w), GFP_ATOMIC);
487 if (!w)
488 return -ENOMEM;
489
490 w->func = fib6_node_dump;
491 arg.net = net;
492 arg.nb = nb;
493 arg.extack = extack;
494 w->args = &arg;
495
496 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
497 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
498 struct fib6_table *tb;
499
500 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
501 err = fib6_table_dump(net, tb, w);
502 if (err < 0)
503 goto out;
504 }
505 }
506
507out:
508 kfree(w);
509
510 return err;
511}
512
513static int fib6_dump_node(struct fib6_walker *w)
514{
515 int res;
516 struct fib6_info *rt;
517
518 for_each_fib6_walker_rt(w) {
519 res = rt6_dump_route(rt, w->args, w->skip_in_node);
520 if (res >= 0) {
521 /* Frame is full, suspend walking */
522 w->leaf = rt;
523
524 /* We'll restart from this node, so if some routes were
525 * already dumped, skip them next time.
526 */
527 w->skip_in_node += res;
528
529 return 1;
530 }
531 w->skip_in_node = 0;
532
533 /* Multipath routes are dumped in one route with the
534 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
535 * last sibling of this route (no need to dump the
536 * sibling routes again)
537 */
538 if (rt->fib6_nsiblings)
539 rt = list_last_entry(&rt->fib6_siblings,
540 struct fib6_info,
541 fib6_siblings);
542 }
543 w->leaf = NULL;
544 return 0;
545}
546
547static void fib6_dump_end(struct netlink_callback *cb)
548{
549 struct net *net = sock_net(cb->skb->sk);
550 struct fib6_walker *w = (void *)cb->args[2];
551
552 if (w) {
553 if (cb->args[4]) {
554 cb->args[4] = 0;
555 fib6_walker_unlink(net, w);
556 }
557 cb->args[2] = 0;
558 kfree(w);
559 }
560 cb->done = (void *)cb->args[3];
561 cb->args[1] = 3;
562}
563
564static int fib6_dump_done(struct netlink_callback *cb)
565{
566 fib6_dump_end(cb);
567 return cb->done ? cb->done(cb) : 0;
568}
569
570static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
571 struct netlink_callback *cb)
572{
573 struct net *net = sock_net(skb->sk);
574 struct fib6_walker *w;
575 int res;
576
577 w = (void *)cb->args[2];
578 w->root = &table->tb6_root;
579
580 if (cb->args[4] == 0) {
581 w->count = 0;
582 w->skip = 0;
583 w->skip_in_node = 0;
584
585 spin_lock_bh(&table->tb6_lock);
586 res = fib6_walk(net, w);
587 spin_unlock_bh(&table->tb6_lock);
588 if (res > 0) {
589 cb->args[4] = 1;
590 cb->args[5] = w->root->fn_sernum;
591 }
592 } else {
593 if (cb->args[5] != w->root->fn_sernum) {
594 /* Begin at the root if the tree changed */
595 cb->args[5] = w->root->fn_sernum;
596 w->state = FWS_INIT;
597 w->node = w->root;
598 w->skip = w->count;
599 w->skip_in_node = 0;
600 } else
601 w->skip = 0;
602
603 spin_lock_bh(&table->tb6_lock);
604 res = fib6_walk_continue(w);
605 spin_unlock_bh(&table->tb6_lock);
606 if (res <= 0) {
607 fib6_walker_unlink(net, w);
608 cb->args[4] = 0;
609 }
610 }
611
612 return res;
613}
614
615static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
616{
617 struct rt6_rtnl_dump_arg arg = { .filter.dump_exceptions = true,
618 .filter.dump_routes = true };
619 const struct nlmsghdr *nlh = cb->nlh;
620 struct net *net = sock_net(skb->sk);
621 unsigned int h, s_h;
622 unsigned int e = 0, s_e;
623 struct fib6_walker *w;
624 struct fib6_table *tb;
625 struct hlist_head *head;
626 int res = 0;
627
628 if (cb->strict_check) {
629 int err;
630
631 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
632 if (err < 0)
633 return err;
634 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
635 struct rtmsg *rtm = nlmsg_data(nlh);
636
637 if (rtm->rtm_flags & RTM_F_PREFIX)
638 arg.filter.flags = RTM_F_PREFIX;
639 }
640
641 w = (void *)cb->args[2];
642 if (!w) {
643 /* New dump:
644 *
645 * 1. hook callback destructor.
646 */
647 cb->args[3] = (long)cb->done;
648 cb->done = fib6_dump_done;
649
650 /*
651 * 2. allocate and initialize walker.
652 */
653 w = kzalloc(sizeof(*w), GFP_ATOMIC);
654 if (!w)
655 return -ENOMEM;
656 w->func = fib6_dump_node;
657 cb->args[2] = (long)w;
658 }
659
660 arg.skb = skb;
661 arg.cb = cb;
662 arg.net = net;
663 w->args = &arg;
664
665 if (arg.filter.table_id) {
666 tb = fib6_get_table(net, arg.filter.table_id);
667 if (!tb) {
668 if (rtnl_msg_family(cb->nlh) != PF_INET6)
669 goto out;
670
671 NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
672 return -ENOENT;
673 }
674
675 if (!cb->args[0]) {
676 res = fib6_dump_table(tb, skb, cb);
677 if (!res)
678 cb->args[0] = 1;
679 }
680 goto out;
681 }
682
683 s_h = cb->args[0];
684 s_e = cb->args[1];
685
686 rcu_read_lock();
687 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
688 e = 0;
689 head = &net->ipv6.fib_table_hash[h];
690 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
691 if (e < s_e)
692 goto next;
693 res = fib6_dump_table(tb, skb, cb);
694 if (res != 0)
695 goto out_unlock;
696next:
697 e++;
698 }
699 }
700out_unlock:
701 rcu_read_unlock();
702 cb->args[1] = e;
703 cb->args[0] = h;
704out:
705 res = res < 0 ? res : skb->len;
706 if (res <= 0)
707 fib6_dump_end(cb);
708 return res;
709}
710
711void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
712{
713 if (!f6i)
714 return;
715
716 if (f6i->fib6_metrics == &dst_default_metrics) {
717 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
718
719 if (!p)
720 return;
721
722 refcount_set(&p->refcnt, 1);
723 f6i->fib6_metrics = p;
724 }
725
726 f6i->fib6_metrics->metrics[metric - 1] = val;
727}
728
729/*
730 * Routing Table
731 *
732 * return the appropriate node for a routing tree "add" operation
733 * by either creating and inserting or by returning an existing
734 * node.
735 */
736
737static struct fib6_node *fib6_add_1(struct net *net,
738 struct fib6_table *table,
739 struct fib6_node *root,
740 struct in6_addr *addr, int plen,
741 int offset, int allow_create,
742 int replace_required,
743 struct netlink_ext_ack *extack)
744{
745 struct fib6_node *fn, *in, *ln;
746 struct fib6_node *pn = NULL;
747 struct rt6key *key;
748 int bit;
749 __be32 dir = 0;
750
751 RT6_TRACE("fib6_add_1\n");
752
753 /* insert node in tree */
754
755 fn = root;
756
757 do {
758 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
759 lockdep_is_held(&table->tb6_lock));
760 key = (struct rt6key *)((u8 *)leaf + offset);
761
762 /*
763 * Prefix match
764 */
765 if (plen < fn->fn_bit ||
766 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
767 if (!allow_create) {
768 if (replace_required) {
769 NL_SET_ERR_MSG(extack,
770 "Can not replace route - no match found");
771 pr_warn("Can't replace route, no match found\n");
772 return ERR_PTR(-ENOENT);
773 }
774 pr_warn("NLM_F_CREATE should be set when creating new route\n");
775 }
776 goto insert_above;
777 }
778
779 /*
780 * Exact match ?
781 */
782
783 if (plen == fn->fn_bit) {
784 /* clean up an intermediate node */
785 if (!(fn->fn_flags & RTN_RTINFO)) {
786 RCU_INIT_POINTER(fn->leaf, NULL);
787 fib6_info_release(leaf);
788 /* remove null_entry in the root node */
789 } else if (fn->fn_flags & RTN_TL_ROOT &&
790 rcu_access_pointer(fn->leaf) ==
791 net->ipv6.fib6_null_entry) {
792 RCU_INIT_POINTER(fn->leaf, NULL);
793 }
794
795 return fn;
796 }
797
798 /*
799 * We have more bits to go
800 */
801
802 /* Try to walk down on tree. */
803 dir = addr_bit_set(addr, fn->fn_bit);
804 pn = fn;
805 fn = dir ?
806 rcu_dereference_protected(fn->right,
807 lockdep_is_held(&table->tb6_lock)) :
808 rcu_dereference_protected(fn->left,
809 lockdep_is_held(&table->tb6_lock));
810 } while (fn);
811
812 if (!allow_create) {
813 /* We should not create new node because
814 * NLM_F_REPLACE was specified without NLM_F_CREATE
815 * I assume it is safe to require NLM_F_CREATE when
816 * REPLACE flag is used! Later we may want to remove the
817 * check for replace_required, because according
818 * to netlink specification, NLM_F_CREATE
819 * MUST be specified if new route is created.
820 * That would keep IPv6 consistent with IPv4
821 */
822 if (replace_required) {
823 NL_SET_ERR_MSG(extack,
824 "Can not replace route - no match found");
825 pr_warn("Can't replace route, no match found\n");
826 return ERR_PTR(-ENOENT);
827 }
828 pr_warn("NLM_F_CREATE should be set when creating new route\n");
829 }
830 /*
831 * We walked to the bottom of tree.
832 * Create new leaf node without children.
833 */
834
835 ln = node_alloc(net);
836
837 if (!ln)
838 return ERR_PTR(-ENOMEM);
839 ln->fn_bit = plen;
840 RCU_INIT_POINTER(ln->parent, pn);
841
842 if (dir)
843 rcu_assign_pointer(pn->right, ln);
844 else
845 rcu_assign_pointer(pn->left, ln);
846
847 return ln;
848
849
850insert_above:
851 /*
852 * split since we don't have a common prefix anymore or
853 * we have a less significant route.
854 * we've to insert an intermediate node on the list
855 * this new node will point to the one we need to create
856 * and the current
857 */
858
859 pn = rcu_dereference_protected(fn->parent,
860 lockdep_is_held(&table->tb6_lock));
861
862 /* find 1st bit in difference between the 2 addrs.
863
864 See comment in __ipv6_addr_diff: bit may be an invalid value,
865 but if it is >= plen, the value is ignored in any case.
866 */
867
868 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
869
870 /*
871 * (intermediate)[in]
872 * / \
873 * (new leaf node)[ln] (old node)[fn]
874 */
875 if (plen > bit) {
876 in = node_alloc(net);
877 ln = node_alloc(net);
878
879 if (!in || !ln) {
880 if (in)
881 node_free_immediate(net, in);
882 if (ln)
883 node_free_immediate(net, ln);
884 return ERR_PTR(-ENOMEM);
885 }
886
887 /*
888 * new intermediate node.
889 * RTN_RTINFO will
890 * be off since that an address that chooses one of
891 * the branches would not match less specific routes
892 * in the other branch
893 */
894
895 in->fn_bit = bit;
896
897 RCU_INIT_POINTER(in->parent, pn);
898 in->leaf = fn->leaf;
899 fib6_info_hold(rcu_dereference_protected(in->leaf,
900 lockdep_is_held(&table->tb6_lock)));
901
902 /* update parent pointer */
903 if (dir)
904 rcu_assign_pointer(pn->right, in);
905 else
906 rcu_assign_pointer(pn->left, in);
907
908 ln->fn_bit = plen;
909
910 RCU_INIT_POINTER(ln->parent, in);
911 rcu_assign_pointer(fn->parent, in);
912
913 if (addr_bit_set(addr, bit)) {
914 rcu_assign_pointer(in->right, ln);
915 rcu_assign_pointer(in->left, fn);
916 } else {
917 rcu_assign_pointer(in->left, ln);
918 rcu_assign_pointer(in->right, fn);
919 }
920 } else { /* plen <= bit */
921
922 /*
923 * (new leaf node)[ln]
924 * / \
925 * (old node)[fn] NULL
926 */
927
928 ln = node_alloc(net);
929
930 if (!ln)
931 return ERR_PTR(-ENOMEM);
932
933 ln->fn_bit = plen;
934
935 RCU_INIT_POINTER(ln->parent, pn);
936
937 if (addr_bit_set(&key->addr, plen))
938 RCU_INIT_POINTER(ln->right, fn);
939 else
940 RCU_INIT_POINTER(ln->left, fn);
941
942 rcu_assign_pointer(fn->parent, ln);
943
944 if (dir)
945 rcu_assign_pointer(pn->right, ln);
946 else
947 rcu_assign_pointer(pn->left, ln);
948 }
949 return ln;
950}
951
952static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
953 const struct fib6_info *match,
954 const struct fib6_table *table)
955{
956 int cpu;
957
958 if (!fib6_nh->rt6i_pcpu)
959 return;
960
961 /* release the reference to this fib entry from
962 * all of its cached pcpu routes
963 */
964 for_each_possible_cpu(cpu) {
965 struct rt6_info **ppcpu_rt;
966 struct rt6_info *pcpu_rt;
967
968 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
969 pcpu_rt = *ppcpu_rt;
970
971 /* only dropping the 'from' reference if the cached route
972 * is using 'match'. The cached pcpu_rt->from only changes
973 * from a fib6_info to NULL (ip6_dst_destroy); it can never
974 * change from one fib6_info reference to another
975 */
976 if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
977 struct fib6_info *from;
978
979 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
980 fib6_info_release(from);
981 }
982 }
983}
984
985struct fib6_nh_pcpu_arg {
986 struct fib6_info *from;
987 const struct fib6_table *table;
988};
989
990static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg)
991{
992 struct fib6_nh_pcpu_arg *arg = _arg;
993
994 __fib6_drop_pcpu_from(nh, arg->from, arg->table);
995 return 0;
996}
997
998static void fib6_drop_pcpu_from(struct fib6_info *f6i,
999 const struct fib6_table *table)
1000{
1001 /* Make sure rt6_make_pcpu_route() wont add other percpu routes
1002 * while we are cleaning them here.
1003 */
1004 f6i->fib6_destroying = 1;
1005 mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
1006
1007 if (f6i->nh) {
1008 struct fib6_nh_pcpu_arg arg = {
1009 .from = f6i,
1010 .table = table
1011 };
1012
1013 nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from,
1014 &arg);
1015 } else {
1016 struct fib6_nh *fib6_nh;
1017
1018 fib6_nh = f6i->fib6_nh;
1019 __fib6_drop_pcpu_from(fib6_nh, f6i, table);
1020 }
1021}
1022
1023static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
1024 struct net *net)
1025{
1026 struct fib6_table *table = rt->fib6_table;
1027
1028 fib6_drop_pcpu_from(rt, table);
1029
1030 if (rt->nh && !list_empty(&rt->nh_list))
1031 list_del_init(&rt->nh_list);
1032
1033 if (refcount_read(&rt->fib6_ref) != 1) {
1034 /* This route is used as dummy address holder in some split
1035 * nodes. It is not leaked, but it still holds other resources,
1036 * which must be released in time. So, scan ascendant nodes
1037 * and replace dummy references to this route with references
1038 * to still alive ones.
1039 */
1040 while (fn) {
1041 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1042 lockdep_is_held(&table->tb6_lock));
1043 struct fib6_info *new_leaf;
1044 if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
1045 new_leaf = fib6_find_prefix(net, table, fn);
1046 fib6_info_hold(new_leaf);
1047
1048 rcu_assign_pointer(fn->leaf, new_leaf);
1049 fib6_info_release(rt);
1050 }
1051 fn = rcu_dereference_protected(fn->parent,
1052 lockdep_is_held(&table->tb6_lock));
1053 }
1054 }
1055}
1056
1057/*
1058 * Insert routing information in a node.
1059 */
1060
1061static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
1062 struct nl_info *info,
1063 struct netlink_ext_ack *extack)
1064{
1065 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1066 lockdep_is_held(&rt->fib6_table->tb6_lock));
1067 struct fib6_info *iter = NULL;
1068 struct fib6_info __rcu **ins;
1069 struct fib6_info __rcu **fallback_ins = NULL;
1070 int replace = (info->nlh &&
1071 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
1072 int add = (!info->nlh ||
1073 (info->nlh->nlmsg_flags & NLM_F_CREATE));
1074 int found = 0;
1075 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
1076 bool notify_sibling_rt = false;
1077 u16 nlflags = NLM_F_EXCL;
1078 int err;
1079
1080 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
1081 nlflags |= NLM_F_APPEND;
1082
1083 ins = &fn->leaf;
1084
1085 for (iter = leaf; iter;
1086 iter = rcu_dereference_protected(iter->fib6_next,
1087 lockdep_is_held(&rt->fib6_table->tb6_lock))) {
1088 /*
1089 * Search for duplicates
1090 */
1091
1092 if (iter->fib6_metric == rt->fib6_metric) {
1093 /*
1094 * Same priority level
1095 */
1096 if (info->nlh &&
1097 (info->nlh->nlmsg_flags & NLM_F_EXCL))
1098 return -EEXIST;
1099
1100 nlflags &= ~NLM_F_EXCL;
1101 if (replace) {
1102 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1103 found++;
1104 break;
1105 }
1106 fallback_ins = fallback_ins ?: ins;
1107 goto next_iter;
1108 }
1109
1110 if (rt6_duplicate_nexthop(iter, rt)) {
1111 if (rt->fib6_nsiblings)
1112 rt->fib6_nsiblings = 0;
1113 if (!(iter->fib6_flags & RTF_EXPIRES))
1114 return -EEXIST;
1115 if (!(rt->fib6_flags & RTF_EXPIRES))
1116 fib6_clean_expires(iter);
1117 else
1118 fib6_set_expires(iter, rt->expires);
1119
1120 if (rt->fib6_pmtu)
1121 fib6_metric_set(iter, RTAX_MTU,
1122 rt->fib6_pmtu);
1123 return -EEXIST;
1124 }
1125 /* If we have the same destination and the same metric,
1126 * but not the same gateway, then the route we try to
1127 * add is sibling to this route, increment our counter
1128 * of siblings, and later we will add our route to the
1129 * list.
1130 * Only static routes (which don't have flag
1131 * RTF_EXPIRES) are used for ECMPv6.
1132 *
1133 * To avoid long list, we only had siblings if the
1134 * route have a gateway.
1135 */
1136 if (rt_can_ecmp &&
1137 rt6_qualify_for_ecmp(iter))
1138 rt->fib6_nsiblings++;
1139 }
1140
1141 if (iter->fib6_metric > rt->fib6_metric)
1142 break;
1143
1144next_iter:
1145 ins = &iter->fib6_next;
1146 }
1147
1148 if (fallback_ins && !found) {
1149 /* No matching route with same ecmp-able-ness found, replace
1150 * first matching route
1151 */
1152 ins = fallback_ins;
1153 iter = rcu_dereference_protected(*ins,
1154 lockdep_is_held(&rt->fib6_table->tb6_lock));
1155 found++;
1156 }
1157
1158 /* Reset round-robin state, if necessary */
1159 if (ins == &fn->leaf)
1160 fn->rr_ptr = NULL;
1161
1162 /* Link this route to others same route. */
1163 if (rt->fib6_nsiblings) {
1164 unsigned int fib6_nsiblings;
1165 struct fib6_info *sibling, *temp_sibling;
1166
1167 /* Find the first route that have the same metric */
1168 sibling = leaf;
1169 notify_sibling_rt = true;
1170 while (sibling) {
1171 if (sibling->fib6_metric == rt->fib6_metric &&
1172 rt6_qualify_for_ecmp(sibling)) {
1173 list_add_tail(&rt->fib6_siblings,
1174 &sibling->fib6_siblings);
1175 break;
1176 }
1177 sibling = rcu_dereference_protected(sibling->fib6_next,
1178 lockdep_is_held(&rt->fib6_table->tb6_lock));
1179 notify_sibling_rt = false;
1180 }
1181 /* For each sibling in the list, increment the counter of
1182 * siblings. BUG() if counters does not match, list of siblings
1183 * is broken!
1184 */
1185 fib6_nsiblings = 0;
1186 list_for_each_entry_safe(sibling, temp_sibling,
1187 &rt->fib6_siblings, fib6_siblings) {
1188 sibling->fib6_nsiblings++;
1189 BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1190 fib6_nsiblings++;
1191 }
1192 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1193 rt6_multipath_rebalance(temp_sibling);
1194 }
1195
1196 /*
1197 * insert node
1198 */
1199 if (!replace) {
1200 if (!add)
1201 pr_warn("NLM_F_CREATE should be set when creating new route\n");
1202
1203add:
1204 nlflags |= NLM_F_CREATE;
1205
1206 /* The route should only be notified if it is the first
1207 * route in the node or if it is added as a sibling
1208 * route to the first route in the node.
1209 */
1210 if (!info->skip_notify_kernel &&
1211 (notify_sibling_rt || ins == &fn->leaf)) {
1212 enum fib_event_type fib_event;
1213
1214 if (notify_sibling_rt)
1215 fib_event = FIB_EVENT_ENTRY_APPEND;
1216 else
1217 fib_event = FIB_EVENT_ENTRY_REPLACE;
1218 err = call_fib6_entry_notifiers(info->nl_net,
1219 fib_event, rt,
1220 extack);
1221 if (err) {
1222 struct fib6_info *sibling, *next_sibling;
1223
1224 /* If the route has siblings, then it first
1225 * needs to be unlinked from them.
1226 */
1227 if (!rt->fib6_nsiblings)
1228 return err;
1229
1230 list_for_each_entry_safe(sibling, next_sibling,
1231 &rt->fib6_siblings,
1232 fib6_siblings)
1233 sibling->fib6_nsiblings--;
1234 rt->fib6_nsiblings = 0;
1235 list_del_init(&rt->fib6_siblings);
1236 rt6_multipath_rebalance(next_sibling);
1237 return err;
1238 }
1239 }
1240
1241 rcu_assign_pointer(rt->fib6_next, iter);
1242 fib6_info_hold(rt);
1243 rcu_assign_pointer(rt->fib6_node, fn);
1244 rcu_assign_pointer(*ins, rt);
1245 if (!info->skip_notify)
1246 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1247 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1248
1249 if (!(fn->fn_flags & RTN_RTINFO)) {
1250 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1251 fn->fn_flags |= RTN_RTINFO;
1252 }
1253
1254 } else {
1255 int nsiblings;
1256
1257 if (!found) {
1258 if (add)
1259 goto add;
1260 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1261 return -ENOENT;
1262 }
1263
1264 if (!info->skip_notify_kernel && ins == &fn->leaf) {
1265 err = call_fib6_entry_notifiers(info->nl_net,
1266 FIB_EVENT_ENTRY_REPLACE,
1267 rt, extack);
1268 if (err)
1269 return err;
1270 }
1271
1272 fib6_info_hold(rt);
1273 rcu_assign_pointer(rt->fib6_node, fn);
1274 rt->fib6_next = iter->fib6_next;
1275 rcu_assign_pointer(*ins, rt);
1276 if (!info->skip_notify)
1277 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1278 if (!(fn->fn_flags & RTN_RTINFO)) {
1279 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1280 fn->fn_flags |= RTN_RTINFO;
1281 }
1282 nsiblings = iter->fib6_nsiblings;
1283 iter->fib6_node = NULL;
1284 fib6_purge_rt(iter, fn, info->nl_net);
1285 if (rcu_access_pointer(fn->rr_ptr) == iter)
1286 fn->rr_ptr = NULL;
1287 fib6_info_release(iter);
1288
1289 if (nsiblings) {
1290 /* Replacing an ECMP route, remove all siblings */
1291 ins = &rt->fib6_next;
1292 iter = rcu_dereference_protected(*ins,
1293 lockdep_is_held(&rt->fib6_table->tb6_lock));
1294 while (iter) {
1295 if (iter->fib6_metric > rt->fib6_metric)
1296 break;
1297 if (rt6_qualify_for_ecmp(iter)) {
1298 *ins = iter->fib6_next;
1299 iter->fib6_node = NULL;
1300 fib6_purge_rt(iter, fn, info->nl_net);
1301 if (rcu_access_pointer(fn->rr_ptr) == iter)
1302 fn->rr_ptr = NULL;
1303 fib6_info_release(iter);
1304 nsiblings--;
1305 info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1306 } else {
1307 ins = &iter->fib6_next;
1308 }
1309 iter = rcu_dereference_protected(*ins,
1310 lockdep_is_held(&rt->fib6_table->tb6_lock));
1311 }
1312 WARN_ON(nsiblings != 0);
1313 }
1314 }
1315
1316 return 0;
1317}
1318
1319static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1320{
1321 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1322 (rt->fib6_flags & RTF_EXPIRES))
1323 mod_timer(&net->ipv6.ip6_fib_timer,
1324 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1325}
1326
1327void fib6_force_start_gc(struct net *net)
1328{
1329 if (!timer_pending(&net->ipv6.ip6_fib_timer))
1330 mod_timer(&net->ipv6.ip6_fib_timer,
1331 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1332}
1333
1334static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1335 int sernum)
1336{
1337 struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1338 lockdep_is_held(&rt->fib6_table->tb6_lock));
1339
1340 /* paired with smp_rmb() in rt6_get_cookie_safe() */
1341 smp_wmb();
1342 while (fn) {
1343 fn->fn_sernum = sernum;
1344 fn = rcu_dereference_protected(fn->parent,
1345 lockdep_is_held(&rt->fib6_table->tb6_lock));
1346 }
1347}
1348
1349void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1350{
1351 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1352}
1353
1354/* allow ipv4 to update sernum via ipv6_stub */
1355void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1356{
1357 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1358 fib6_update_sernum_upto_root(net, f6i);
1359 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1360}
1361
1362/*
1363 * Add routing information to the routing tree.
1364 * <destination addr>/<source addr>
1365 * with source addr info in sub-trees
1366 * Need to own table->tb6_lock
1367 */
1368
1369int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1370 struct nl_info *info, struct netlink_ext_ack *extack)
1371{
1372 struct fib6_table *table = rt->fib6_table;
1373 struct fib6_node *fn, *pn = NULL;
1374 int err = -ENOMEM;
1375 int allow_create = 1;
1376 int replace_required = 0;
1377 int sernum = fib6_new_sernum(info->nl_net);
1378
1379 if (info->nlh) {
1380 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1381 allow_create = 0;
1382 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1383 replace_required = 1;
1384 }
1385 if (!allow_create && !replace_required)
1386 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1387
1388 fn = fib6_add_1(info->nl_net, table, root,
1389 &rt->fib6_dst.addr, rt->fib6_dst.plen,
1390 offsetof(struct fib6_info, fib6_dst), allow_create,
1391 replace_required, extack);
1392 if (IS_ERR(fn)) {
1393 err = PTR_ERR(fn);
1394 fn = NULL;
1395 goto out;
1396 }
1397
1398 pn = fn;
1399
1400#ifdef CONFIG_IPV6_SUBTREES
1401 if (rt->fib6_src.plen) {
1402 struct fib6_node *sn;
1403
1404 if (!rcu_access_pointer(fn->subtree)) {
1405 struct fib6_node *sfn;
1406
1407 /*
1408 * Create subtree.
1409 *
1410 * fn[main tree]
1411 * |
1412 * sfn[subtree root]
1413 * \
1414 * sn[new leaf node]
1415 */
1416
1417 /* Create subtree root node */
1418 sfn = node_alloc(info->nl_net);
1419 if (!sfn)
1420 goto failure;
1421
1422 fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1423 rcu_assign_pointer(sfn->leaf,
1424 info->nl_net->ipv6.fib6_null_entry);
1425 sfn->fn_flags = RTN_ROOT;
1426
1427 /* Now add the first leaf node to new subtree */
1428
1429 sn = fib6_add_1(info->nl_net, table, sfn,
1430 &rt->fib6_src.addr, rt->fib6_src.plen,
1431 offsetof(struct fib6_info, fib6_src),
1432 allow_create, replace_required, extack);
1433
1434 if (IS_ERR(sn)) {
1435 /* If it is failed, discard just allocated
1436 root, and then (in failure) stale node
1437 in main tree.
1438 */
1439 node_free_immediate(info->nl_net, sfn);
1440 err = PTR_ERR(sn);
1441 goto failure;
1442 }
1443
1444 /* Now link new subtree to main tree */
1445 rcu_assign_pointer(sfn->parent, fn);
1446 rcu_assign_pointer(fn->subtree, sfn);
1447 } else {
1448 sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1449 &rt->fib6_src.addr, rt->fib6_src.plen,
1450 offsetof(struct fib6_info, fib6_src),
1451 allow_create, replace_required, extack);
1452
1453 if (IS_ERR(sn)) {
1454 err = PTR_ERR(sn);
1455 goto failure;
1456 }
1457 }
1458
1459 if (!rcu_access_pointer(fn->leaf)) {
1460 if (fn->fn_flags & RTN_TL_ROOT) {
1461 /* put back null_entry for root node */
1462 rcu_assign_pointer(fn->leaf,
1463 info->nl_net->ipv6.fib6_null_entry);
1464 } else {
1465 fib6_info_hold(rt);
1466 rcu_assign_pointer(fn->leaf, rt);
1467 }
1468 }
1469 fn = sn;
1470 }
1471#endif
1472
1473 err = fib6_add_rt2node(fn, rt, info, extack);
1474 if (!err) {
1475 if (rt->nh)
1476 list_add(&rt->nh_list, &rt->nh->f6i_list);
1477 __fib6_update_sernum_upto_root(rt, sernum);
1478 fib6_start_gc(info->nl_net, rt);
1479 }
1480
1481out:
1482 if (err) {
1483#ifdef CONFIG_IPV6_SUBTREES
1484 /*
1485 * If fib6_add_1 has cleared the old leaf pointer in the
1486 * super-tree leaf node we have to find a new one for it.
1487 */
1488 if (pn != fn) {
1489 struct fib6_info *pn_leaf =
1490 rcu_dereference_protected(pn->leaf,
1491 lockdep_is_held(&table->tb6_lock));
1492 if (pn_leaf == rt) {
1493 pn_leaf = NULL;
1494 RCU_INIT_POINTER(pn->leaf, NULL);
1495 fib6_info_release(rt);
1496 }
1497 if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1498 pn_leaf = fib6_find_prefix(info->nl_net, table,
1499 pn);
1500#if RT6_DEBUG >= 2
1501 if (!pn_leaf) {
1502 WARN_ON(!pn_leaf);
1503 pn_leaf =
1504 info->nl_net->ipv6.fib6_null_entry;
1505 }
1506#endif
1507 fib6_info_hold(pn_leaf);
1508 rcu_assign_pointer(pn->leaf, pn_leaf);
1509 }
1510 }
1511#endif
1512 goto failure;
1513 } else if (fib6_requires_src(rt)) {
1514 fib6_routes_require_src_inc(info->nl_net);
1515 }
1516 return err;
1517
1518failure:
1519 /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1520 * 1. fn is an intermediate node and we failed to add the new
1521 * route to it in both subtree creation failure and fib6_add_rt2node()
1522 * failure case.
1523 * 2. fn is the root node in the table and we fail to add the first
1524 * default route to it.
1525 */
1526 if (fn &&
1527 (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1528 (fn->fn_flags & RTN_TL_ROOT &&
1529 !rcu_access_pointer(fn->leaf))))
1530 fib6_repair_tree(info->nl_net, table, fn);
1531 return err;
1532}
1533
1534/*
1535 * Routing tree lookup
1536 *
1537 */
1538
1539struct lookup_args {
1540 int offset; /* key offset on fib6_info */
1541 const struct in6_addr *addr; /* search key */
1542};
1543
1544static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1545 struct lookup_args *args)
1546{
1547 struct fib6_node *fn;
1548 __be32 dir;
1549
1550 if (unlikely(args->offset == 0))
1551 return NULL;
1552
1553 /*
1554 * Descend on a tree
1555 */
1556
1557 fn = root;
1558
1559 for (;;) {
1560 struct fib6_node *next;
1561
1562 dir = addr_bit_set(args->addr, fn->fn_bit);
1563
1564 next = dir ? rcu_dereference(fn->right) :
1565 rcu_dereference(fn->left);
1566
1567 if (next) {
1568 fn = next;
1569 continue;
1570 }
1571 break;
1572 }
1573
1574 while (fn) {
1575 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1576
1577 if (subtree || fn->fn_flags & RTN_RTINFO) {
1578 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1579 struct rt6key *key;
1580
1581 if (!leaf)
1582 goto backtrack;
1583
1584 key = (struct rt6key *) ((u8 *)leaf + args->offset);
1585
1586 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1587#ifdef CONFIG_IPV6_SUBTREES
1588 if (subtree) {
1589 struct fib6_node *sfn;
1590 sfn = fib6_node_lookup_1(subtree,
1591 args + 1);
1592 if (!sfn)
1593 goto backtrack;
1594 fn = sfn;
1595 }
1596#endif
1597 if (fn->fn_flags & RTN_RTINFO)
1598 return fn;
1599 }
1600 }
1601backtrack:
1602 if (fn->fn_flags & RTN_ROOT)
1603 break;
1604
1605 fn = rcu_dereference(fn->parent);
1606 }
1607
1608 return NULL;
1609}
1610
1611/* called with rcu_read_lock() held
1612 */
1613struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1614 const struct in6_addr *daddr,
1615 const struct in6_addr *saddr)
1616{
1617 struct fib6_node *fn;
1618 struct lookup_args args[] = {
1619 {
1620 .offset = offsetof(struct fib6_info, fib6_dst),
1621 .addr = daddr,
1622 },
1623#ifdef CONFIG_IPV6_SUBTREES
1624 {
1625 .offset = offsetof(struct fib6_info, fib6_src),
1626 .addr = saddr,
1627 },
1628#endif
1629 {
1630 .offset = 0, /* sentinel */
1631 }
1632 };
1633
1634 fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1635 if (!fn || fn->fn_flags & RTN_TL_ROOT)
1636 fn = root;
1637
1638 return fn;
1639}
1640
1641/*
1642 * Get node with specified destination prefix (and source prefix,
1643 * if subtrees are used)
1644 * exact_match == true means we try to find fn with exact match of
1645 * the passed in prefix addr
1646 * exact_match == false means we try to find fn with longest prefix
1647 * match of the passed in prefix addr. This is useful for finding fn
1648 * for cached route as it will be stored in the exception table under
1649 * the node with longest prefix length.
1650 */
1651
1652
1653static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1654 const struct in6_addr *addr,
1655 int plen, int offset,
1656 bool exact_match)
1657{
1658 struct fib6_node *fn, *prev = NULL;
1659
1660 for (fn = root; fn ; ) {
1661 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1662 struct rt6key *key;
1663
1664 /* This node is being deleted */
1665 if (!leaf) {
1666 if (plen <= fn->fn_bit)
1667 goto out;
1668 else
1669 goto next;
1670 }
1671
1672 key = (struct rt6key *)((u8 *)leaf + offset);
1673
1674 /*
1675 * Prefix match
1676 */
1677 if (plen < fn->fn_bit ||
1678 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1679 goto out;
1680
1681 if (plen == fn->fn_bit)
1682 return fn;
1683
1684 if (fn->fn_flags & RTN_RTINFO)
1685 prev = fn;
1686
1687next:
1688 /*
1689 * We have more bits to go
1690 */
1691 if (addr_bit_set(addr, fn->fn_bit))
1692 fn = rcu_dereference(fn->right);
1693 else
1694 fn = rcu_dereference(fn->left);
1695 }
1696out:
1697 if (exact_match)
1698 return NULL;
1699 else
1700 return prev;
1701}
1702
1703struct fib6_node *fib6_locate(struct fib6_node *root,
1704 const struct in6_addr *daddr, int dst_len,
1705 const struct in6_addr *saddr, int src_len,
1706 bool exact_match)
1707{
1708 struct fib6_node *fn;
1709
1710 fn = fib6_locate_1(root, daddr, dst_len,
1711 offsetof(struct fib6_info, fib6_dst),
1712 exact_match);
1713
1714#ifdef CONFIG_IPV6_SUBTREES
1715 if (src_len) {
1716 WARN_ON(saddr == NULL);
1717 if (fn) {
1718 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1719
1720 if (subtree) {
1721 fn = fib6_locate_1(subtree, saddr, src_len,
1722 offsetof(struct fib6_info, fib6_src),
1723 exact_match);
1724 }
1725 }
1726 }
1727#endif
1728
1729 if (fn && fn->fn_flags & RTN_RTINFO)
1730 return fn;
1731
1732 return NULL;
1733}
1734
1735
1736/*
1737 * Deletion
1738 *
1739 */
1740
1741static struct fib6_info *fib6_find_prefix(struct net *net,
1742 struct fib6_table *table,
1743 struct fib6_node *fn)
1744{
1745 struct fib6_node *child_left, *child_right;
1746
1747 if (fn->fn_flags & RTN_ROOT)
1748 return net->ipv6.fib6_null_entry;
1749
1750 while (fn) {
1751 child_left = rcu_dereference_protected(fn->left,
1752 lockdep_is_held(&table->tb6_lock));
1753 child_right = rcu_dereference_protected(fn->right,
1754 lockdep_is_held(&table->tb6_lock));
1755 if (child_left)
1756 return rcu_dereference_protected(child_left->leaf,
1757 lockdep_is_held(&table->tb6_lock));
1758 if (child_right)
1759 return rcu_dereference_protected(child_right->leaf,
1760 lockdep_is_held(&table->tb6_lock));
1761
1762 fn = FIB6_SUBTREE(fn);
1763 }
1764 return NULL;
1765}
1766
1767/*
1768 * Called to trim the tree of intermediate nodes when possible. "fn"
1769 * is the node we want to try and remove.
1770 * Need to own table->tb6_lock
1771 */
1772
1773static struct fib6_node *fib6_repair_tree(struct net *net,
1774 struct fib6_table *table,
1775 struct fib6_node *fn)
1776{
1777 int children;
1778 int nstate;
1779 struct fib6_node *child;
1780 struct fib6_walker *w;
1781 int iter = 0;
1782
1783 /* Set fn->leaf to null_entry for root node. */
1784 if (fn->fn_flags & RTN_TL_ROOT) {
1785 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1786 return fn;
1787 }
1788
1789 for (;;) {
1790 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1791 lockdep_is_held(&table->tb6_lock));
1792 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1793 lockdep_is_held(&table->tb6_lock));
1794 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1795 lockdep_is_held(&table->tb6_lock));
1796 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1797 lockdep_is_held(&table->tb6_lock));
1798 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1799 lockdep_is_held(&table->tb6_lock));
1800 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1801 lockdep_is_held(&table->tb6_lock));
1802 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1803 lockdep_is_held(&table->tb6_lock));
1804 struct fib6_info *new_fn_leaf;
1805
1806 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1807 iter++;
1808
1809 WARN_ON(fn->fn_flags & RTN_RTINFO);
1810 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1811 WARN_ON(fn_leaf);
1812
1813 children = 0;
1814 child = NULL;
1815 if (fn_r)
1816 child = fn_r, children |= 1;
1817 if (fn_l)
1818 child = fn_l, children |= 2;
1819
1820 if (children == 3 || FIB6_SUBTREE(fn)
1821#ifdef CONFIG_IPV6_SUBTREES
1822 /* Subtree root (i.e. fn) may have one child */
1823 || (children && fn->fn_flags & RTN_ROOT)
1824#endif
1825 ) {
1826 new_fn_leaf = fib6_find_prefix(net, table, fn);
1827#if RT6_DEBUG >= 2
1828 if (!new_fn_leaf) {
1829 WARN_ON(!new_fn_leaf);
1830 new_fn_leaf = net->ipv6.fib6_null_entry;
1831 }
1832#endif
1833 fib6_info_hold(new_fn_leaf);
1834 rcu_assign_pointer(fn->leaf, new_fn_leaf);
1835 return pn;
1836 }
1837
1838#ifdef CONFIG_IPV6_SUBTREES
1839 if (FIB6_SUBTREE(pn) == fn) {
1840 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1841 RCU_INIT_POINTER(pn->subtree, NULL);
1842 nstate = FWS_L;
1843 } else {
1844 WARN_ON(fn->fn_flags & RTN_ROOT);
1845#endif
1846 if (pn_r == fn)
1847 rcu_assign_pointer(pn->right, child);
1848 else if (pn_l == fn)
1849 rcu_assign_pointer(pn->left, child);
1850#if RT6_DEBUG >= 2
1851 else
1852 WARN_ON(1);
1853#endif
1854 if (child)
1855 rcu_assign_pointer(child->parent, pn);
1856 nstate = FWS_R;
1857#ifdef CONFIG_IPV6_SUBTREES
1858 }
1859#endif
1860
1861 read_lock(&net->ipv6.fib6_walker_lock);
1862 FOR_WALKERS(net, w) {
1863 if (!child) {
1864 if (w->node == fn) {
1865 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1866 w->node = pn;
1867 w->state = nstate;
1868 }
1869 } else {
1870 if (w->node == fn) {
1871 w->node = child;
1872 if (children&2) {
1873 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1874 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1875 } else {
1876 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1877 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1878 }
1879 }
1880 }
1881 }
1882 read_unlock(&net->ipv6.fib6_walker_lock);
1883
1884 node_free(net, fn);
1885 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1886 return pn;
1887
1888 RCU_INIT_POINTER(pn->leaf, NULL);
1889 fib6_info_release(pn_leaf);
1890 fn = pn;
1891 }
1892}
1893
1894static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1895 struct fib6_info __rcu **rtp, struct nl_info *info)
1896{
1897 struct fib6_info *leaf, *replace_rt = NULL;
1898 struct fib6_walker *w;
1899 struct fib6_info *rt = rcu_dereference_protected(*rtp,
1900 lockdep_is_held(&table->tb6_lock));
1901 struct net *net = info->nl_net;
1902 bool notify_del = false;
1903
1904 RT6_TRACE("fib6_del_route\n");
1905
1906 /* If the deleted route is the first in the node and it is not part of
1907 * a multipath route, then we need to replace it with the next route
1908 * in the node, if exists.
1909 */
1910 leaf = rcu_dereference_protected(fn->leaf,
1911 lockdep_is_held(&table->tb6_lock));
1912 if (leaf == rt && !rt->fib6_nsiblings) {
1913 if (rcu_access_pointer(rt->fib6_next))
1914 replace_rt = rcu_dereference_protected(rt->fib6_next,
1915 lockdep_is_held(&table->tb6_lock));
1916 else
1917 notify_del = true;
1918 }
1919
1920 /* Unlink it */
1921 *rtp = rt->fib6_next;
1922 rt->fib6_node = NULL;
1923 net->ipv6.rt6_stats->fib_rt_entries--;
1924 net->ipv6.rt6_stats->fib_discarded_routes++;
1925
1926 /* Flush all cached dst in exception table */
1927 rt6_flush_exceptions(rt);
1928
1929 /* Reset round-robin state, if necessary */
1930 if (rcu_access_pointer(fn->rr_ptr) == rt)
1931 fn->rr_ptr = NULL;
1932
1933 /* Remove this entry from other siblings */
1934 if (rt->fib6_nsiblings) {
1935 struct fib6_info *sibling, *next_sibling;
1936
1937 /* The route is deleted from a multipath route. If this
1938 * multipath route is the first route in the node, then we need
1939 * to emit a delete notification. Otherwise, we need to skip
1940 * the notification.
1941 */
1942 if (rt->fib6_metric == leaf->fib6_metric &&
1943 rt6_qualify_for_ecmp(leaf))
1944 notify_del = true;
1945 list_for_each_entry_safe(sibling, next_sibling,
1946 &rt->fib6_siblings, fib6_siblings)
1947 sibling->fib6_nsiblings--;
1948 rt->fib6_nsiblings = 0;
1949 list_del_init(&rt->fib6_siblings);
1950 rt6_multipath_rebalance(next_sibling);
1951 }
1952
1953 /* Adjust walkers */
1954 read_lock(&net->ipv6.fib6_walker_lock);
1955 FOR_WALKERS(net, w) {
1956 if (w->state == FWS_C && w->leaf == rt) {
1957 RT6_TRACE("walker %p adjusted by delroute\n", w);
1958 w->leaf = rcu_dereference_protected(rt->fib6_next,
1959 lockdep_is_held(&table->tb6_lock));
1960 if (!w->leaf)
1961 w->state = FWS_U;
1962 }
1963 }
1964 read_unlock(&net->ipv6.fib6_walker_lock);
1965
1966 /* If it was last route, call fib6_repair_tree() to:
1967 * 1. For root node, put back null_entry as how the table was created.
1968 * 2. For other nodes, expunge its radix tree node.
1969 */
1970 if (!rcu_access_pointer(fn->leaf)) {
1971 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1972 fn->fn_flags &= ~RTN_RTINFO;
1973 net->ipv6.rt6_stats->fib_route_nodes--;
1974 }
1975 fn = fib6_repair_tree(net, table, fn);
1976 }
1977
1978 fib6_purge_rt(rt, fn, net);
1979
1980 if (!info->skip_notify_kernel) {
1981 if (notify_del)
1982 call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1983 rt, NULL);
1984 else if (replace_rt)
1985 call_fib6_entry_notifiers_replace(net, replace_rt);
1986 }
1987 if (!info->skip_notify)
1988 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1989
1990 fib6_info_release(rt);
1991}
1992
1993/* Need to own table->tb6_lock */
1994int fib6_del(struct fib6_info *rt, struct nl_info *info)
1995{
1996 struct net *net = info->nl_net;
1997 struct fib6_info __rcu **rtp;
1998 struct fib6_info __rcu **rtp_next;
1999 struct fib6_table *table;
2000 struct fib6_node *fn;
2001
2002 if (rt == net->ipv6.fib6_null_entry)
2003 return -ENOENT;
2004
2005 table = rt->fib6_table;
2006 fn = rcu_dereference_protected(rt->fib6_node,
2007 lockdep_is_held(&table->tb6_lock));
2008 if (!fn)
2009 return -ENOENT;
2010
2011 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
2012
2013 /*
2014 * Walk the leaf entries looking for ourself
2015 */
2016
2017 for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
2018 struct fib6_info *cur = rcu_dereference_protected(*rtp,
2019 lockdep_is_held(&table->tb6_lock));
2020 if (rt == cur) {
2021 if (fib6_requires_src(cur))
2022 fib6_routes_require_src_dec(info->nl_net);
2023 fib6_del_route(table, fn, rtp, info);
2024 return 0;
2025 }
2026 rtp_next = &cur->fib6_next;
2027 }
2028 return -ENOENT;
2029}
2030
2031/*
2032 * Tree traversal function.
2033 *
2034 * Certainly, it is not interrupt safe.
2035 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
2036 * It means, that we can modify tree during walking
2037 * and use this function for garbage collection, clone pruning,
2038 * cleaning tree when a device goes down etc. etc.
2039 *
2040 * It guarantees that every node will be traversed,
2041 * and that it will be traversed only once.
2042 *
2043 * Callback function w->func may return:
2044 * 0 -> continue walking.
2045 * positive value -> walking is suspended (used by tree dumps,
2046 * and probably by gc, if it will be split to several slices)
2047 * negative value -> terminate walking.
2048 *
2049 * The function itself returns:
2050 * 0 -> walk is complete.
2051 * >0 -> walk is incomplete (i.e. suspended)
2052 * <0 -> walk is terminated by an error.
2053 *
2054 * This function is called with tb6_lock held.
2055 */
2056
2057static int fib6_walk_continue(struct fib6_walker *w)
2058{
2059 struct fib6_node *fn, *pn, *left, *right;
2060
2061 /* w->root should always be table->tb6_root */
2062 WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
2063
2064 for (;;) {
2065 fn = w->node;
2066 if (!fn)
2067 return 0;
2068
2069 switch (w->state) {
2070#ifdef CONFIG_IPV6_SUBTREES
2071 case FWS_S:
2072 if (FIB6_SUBTREE(fn)) {
2073 w->node = FIB6_SUBTREE(fn);
2074 continue;
2075 }
2076 w->state = FWS_L;
2077 fallthrough;
2078#endif
2079 case FWS_L:
2080 left = rcu_dereference_protected(fn->left, 1);
2081 if (left) {
2082 w->node = left;
2083 w->state = FWS_INIT;
2084 continue;
2085 }
2086 w->state = FWS_R;
2087 fallthrough;
2088 case FWS_R:
2089 right = rcu_dereference_protected(fn->right, 1);
2090 if (right) {
2091 w->node = right;
2092 w->state = FWS_INIT;
2093 continue;
2094 }
2095 w->state = FWS_C;
2096 w->leaf = rcu_dereference_protected(fn->leaf, 1);
2097 fallthrough;
2098 case FWS_C:
2099 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
2100 int err;
2101
2102 if (w->skip) {
2103 w->skip--;
2104 goto skip;
2105 }
2106
2107 err = w->func(w);
2108 if (err)
2109 return err;
2110
2111 w->count++;
2112 continue;
2113 }
2114skip:
2115 w->state = FWS_U;
2116 fallthrough;
2117 case FWS_U:
2118 if (fn == w->root)
2119 return 0;
2120 pn = rcu_dereference_protected(fn->parent, 1);
2121 left = rcu_dereference_protected(pn->left, 1);
2122 right = rcu_dereference_protected(pn->right, 1);
2123 w->node = pn;
2124#ifdef CONFIG_IPV6_SUBTREES
2125 if (FIB6_SUBTREE(pn) == fn) {
2126 WARN_ON(!(fn->fn_flags & RTN_ROOT));
2127 w->state = FWS_L;
2128 continue;
2129 }
2130#endif
2131 if (left == fn) {
2132 w->state = FWS_R;
2133 continue;
2134 }
2135 if (right == fn) {
2136 w->state = FWS_C;
2137 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
2138 continue;
2139 }
2140#if RT6_DEBUG >= 2
2141 WARN_ON(1);
2142#endif
2143 }
2144 }
2145}
2146
2147static int fib6_walk(struct net *net, struct fib6_walker *w)
2148{
2149 int res;
2150
2151 w->state = FWS_INIT;
2152 w->node = w->root;
2153
2154 fib6_walker_link(net, w);
2155 res = fib6_walk_continue(w);
2156 if (res <= 0)
2157 fib6_walker_unlink(net, w);
2158 return res;
2159}
2160
2161static int fib6_clean_node(struct fib6_walker *w)
2162{
2163 int res;
2164 struct fib6_info *rt;
2165 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
2166 struct nl_info info = {
2167 .nl_net = c->net,
2168 .skip_notify = c->skip_notify,
2169 };
2170
2171 if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
2172 w->node->fn_sernum != c->sernum)
2173 w->node->fn_sernum = c->sernum;
2174
2175 if (!c->func) {
2176 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
2177 w->leaf = NULL;
2178 return 0;
2179 }
2180
2181 for_each_fib6_walker_rt(w) {
2182 res = c->func(rt, c->arg);
2183 if (res == -1) {
2184 w->leaf = rt;
2185 res = fib6_del(rt, &info);
2186 if (res) {
2187#if RT6_DEBUG >= 2
2188 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2189 __func__, rt,
2190 rcu_access_pointer(rt->fib6_node),
2191 res);
2192#endif
2193 continue;
2194 }
2195 return 0;
2196 } else if (res == -2) {
2197 if (WARN_ON(!rt->fib6_nsiblings))
2198 continue;
2199 rt = list_last_entry(&rt->fib6_siblings,
2200 struct fib6_info, fib6_siblings);
2201 continue;
2202 }
2203 WARN_ON(res != 0);
2204 }
2205 w->leaf = rt;
2206 return 0;
2207}
2208
2209/*
2210 * Convenient frontend to tree walker.
2211 *
2212 * func is called on each route.
2213 * It may return -2 -> skip multipath route.
2214 * -1 -> delete this route.
2215 * 0 -> continue walking
2216 */
2217
2218static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2219 int (*func)(struct fib6_info *, void *arg),
2220 int sernum, void *arg, bool skip_notify)
2221{
2222 struct fib6_cleaner c;
2223
2224 c.w.root = root;
2225 c.w.func = fib6_clean_node;
2226 c.w.count = 0;
2227 c.w.skip = 0;
2228 c.w.skip_in_node = 0;
2229 c.func = func;
2230 c.sernum = sernum;
2231 c.arg = arg;
2232 c.net = net;
2233 c.skip_notify = skip_notify;
2234
2235 fib6_walk(net, &c.w);
2236}
2237
2238static void __fib6_clean_all(struct net *net,
2239 int (*func)(struct fib6_info *, void *),
2240 int sernum, void *arg, bool skip_notify)
2241{
2242 struct fib6_table *table;
2243 struct hlist_head *head;
2244 unsigned int h;
2245
2246 rcu_read_lock();
2247 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2248 head = &net->ipv6.fib_table_hash[h];
2249 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2250 spin_lock_bh(&table->tb6_lock);
2251 fib6_clean_tree(net, &table->tb6_root,
2252 func, sernum, arg, skip_notify);
2253 spin_unlock_bh(&table->tb6_lock);
2254 }
2255 }
2256 rcu_read_unlock();
2257}
2258
2259void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2260 void *arg)
2261{
2262 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2263}
2264
2265void fib6_clean_all_skip_notify(struct net *net,
2266 int (*func)(struct fib6_info *, void *),
2267 void *arg)
2268{
2269 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2270}
2271
2272static void fib6_flush_trees(struct net *net)
2273{
2274 int new_sernum = fib6_new_sernum(net);
2275
2276 __fib6_clean_all(net, NULL, new_sernum, NULL, false);
2277}
2278
2279/*
2280 * Garbage collection
2281 */
2282
2283static int fib6_age(struct fib6_info *rt, void *arg)
2284{
2285 struct fib6_gc_args *gc_args = arg;
2286 unsigned long now = jiffies;
2287
2288 /*
2289 * check addrconf expiration here.
2290 * Routes are expired even if they are in use.
2291 */
2292
2293 if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2294 if (time_after(now, rt->expires)) {
2295 RT6_TRACE("expiring %p\n", rt);
2296 return -1;
2297 }
2298 gc_args->more++;
2299 }
2300
2301 /* Also age clones in the exception table.
2302 * Note, that clones are aged out
2303 * only if they are not in use now.
2304 */
2305 rt6_age_exceptions(rt, gc_args, now);
2306
2307 return 0;
2308}
2309
2310void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2311{
2312 struct fib6_gc_args gc_args;
2313 unsigned long now;
2314
2315 if (force) {
2316 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2317 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2318 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2319 return;
2320 }
2321 gc_args.timeout = expires ? (int)expires :
2322 net->ipv6.sysctl.ip6_rt_gc_interval;
2323 gc_args.more = 0;
2324
2325 fib6_clean_all(net, fib6_age, &gc_args);
2326 now = jiffies;
2327 net->ipv6.ip6_rt_last_gc = now;
2328
2329 if (gc_args.more)
2330 mod_timer(&net->ipv6.ip6_fib_timer,
2331 round_jiffies(now
2332 + net->ipv6.sysctl.ip6_rt_gc_interval));
2333 else
2334 del_timer(&net->ipv6.ip6_fib_timer);
2335 spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2336}
2337
2338static void fib6_gc_timer_cb(struct timer_list *t)
2339{
2340 struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2341
2342 fib6_run_gc(0, arg, true);
2343}
2344
2345static int __net_init fib6_net_init(struct net *net)
2346{
2347 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2348 int err;
2349
2350 err = fib6_notifier_init(net);
2351 if (err)
2352 return err;
2353
2354 spin_lock_init(&net->ipv6.fib6_gc_lock);
2355 rwlock_init(&net->ipv6.fib6_walker_lock);
2356 INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2357 timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2358
2359 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2360 if (!net->ipv6.rt6_stats)
2361 goto out_timer;
2362
2363 /* Avoid false sharing : Use at least a full cache line */
2364 size = max_t(size_t, size, L1_CACHE_BYTES);
2365
2366 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2367 if (!net->ipv6.fib_table_hash)
2368 goto out_rt6_stats;
2369
2370 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2371 GFP_KERNEL);
2372 if (!net->ipv6.fib6_main_tbl)
2373 goto out_fib_table_hash;
2374
2375 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2376 rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2377 net->ipv6.fib6_null_entry);
2378 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2379 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2380 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2381
2382#ifdef CONFIG_IPV6_MULTIPLE_TABLES
2383 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2384 GFP_KERNEL);
2385 if (!net->ipv6.fib6_local_tbl)
2386 goto out_fib6_main_tbl;
2387 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2388 rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2389 net->ipv6.fib6_null_entry);
2390 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2391 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2392 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2393#endif
2394 fib6_tables_init(net);
2395
2396 return 0;
2397
2398#ifdef CONFIG_IPV6_MULTIPLE_TABLES
2399out_fib6_main_tbl:
2400 kfree(net->ipv6.fib6_main_tbl);
2401#endif
2402out_fib_table_hash:
2403 kfree(net->ipv6.fib_table_hash);
2404out_rt6_stats:
2405 kfree(net->ipv6.rt6_stats);
2406out_timer:
2407 fib6_notifier_exit(net);
2408 return -ENOMEM;
2409}
2410
2411static void fib6_net_exit(struct net *net)
2412{
2413 unsigned int i;
2414
2415 del_timer_sync(&net->ipv6.ip6_fib_timer);
2416
2417 for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2418 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2419 struct hlist_node *tmp;
2420 struct fib6_table *tb;
2421
2422 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2423 hlist_del(&tb->tb6_hlist);
2424 fib6_free_table(tb);
2425 }
2426 }
2427
2428 kfree(net->ipv6.fib_table_hash);
2429 kfree(net->ipv6.rt6_stats);
2430 fib6_notifier_exit(net);
2431}
2432
2433static struct pernet_operations fib6_net_ops = {
2434 .init = fib6_net_init,
2435 .exit = fib6_net_exit,
2436};
2437
2438int __init fib6_init(void)
2439{
2440 int ret = -ENOMEM;
2441
2442 fib6_node_kmem = kmem_cache_create("fib6_nodes",
2443 sizeof(struct fib6_node),
2444 0, SLAB_HWCACHE_ALIGN,
2445 NULL);
2446 if (!fib6_node_kmem)
2447 goto out;
2448
2449 ret = register_pernet_subsys(&fib6_net_ops);
2450 if (ret)
2451 goto out_kmem_cache_create;
2452
2453 ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2454 inet6_dump_fib, 0);
2455 if (ret)
2456 goto out_unregister_subsys;
2457
2458 __fib6_flush_trees = fib6_flush_trees;
2459out:
2460 return ret;
2461
2462out_unregister_subsys:
2463 unregister_pernet_subsys(&fib6_net_ops);
2464out_kmem_cache_create:
2465 kmem_cache_destroy(fib6_node_kmem);
2466 goto out;
2467}
2468
2469void fib6_gc_cleanup(void)
2470{
2471 unregister_pernet_subsys(&fib6_net_ops);
2472 kmem_cache_destroy(fib6_node_kmem);
2473}
2474
2475#ifdef CONFIG_PROC_FS
2476static int ipv6_route_native_seq_show(struct seq_file *seq, void *v)
2477{
2478 struct fib6_info *rt = v;
2479 struct ipv6_route_iter *iter = seq->private;
2480 struct fib6_nh *fib6_nh = rt->fib6_nh;
2481 unsigned int flags = rt->fib6_flags;
2482 const struct net_device *dev;
2483
2484 if (rt->nh)
2485 fib6_nh = nexthop_fib6_nh(rt->nh);
2486
2487 seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2488
2489#ifdef CONFIG_IPV6_SUBTREES
2490 seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2491#else
2492 seq_puts(seq, "00000000000000000000000000000000 00 ");
2493#endif
2494 if (fib6_nh->fib_nh_gw_family) {
2495 flags |= RTF_GATEWAY;
2496 seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6);
2497 } else {
2498 seq_puts(seq, "00000000000000000000000000000000");
2499 }
2500
2501 dev = fib6_nh->fib_nh_dev;
2502 seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2503 rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2504 flags, dev ? dev->name : "");
2505 iter->w.leaf = NULL;
2506 return 0;
2507}
2508
2509static int ipv6_route_yield(struct fib6_walker *w)
2510{
2511 struct ipv6_route_iter *iter = w->args;
2512
2513 if (!iter->skip)
2514 return 1;
2515
2516 do {
2517 iter->w.leaf = rcu_dereference_protected(
2518 iter->w.leaf->fib6_next,
2519 lockdep_is_held(&iter->tbl->tb6_lock));
2520 iter->skip--;
2521 if (!iter->skip && iter->w.leaf)
2522 return 1;
2523 } while (iter->w.leaf);
2524
2525 return 0;
2526}
2527
2528static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2529 struct net *net)
2530{
2531 memset(&iter->w, 0, sizeof(iter->w));
2532 iter->w.func = ipv6_route_yield;
2533 iter->w.root = &iter->tbl->tb6_root;
2534 iter->w.state = FWS_INIT;
2535 iter->w.node = iter->w.root;
2536 iter->w.args = iter;
2537 iter->sernum = iter->w.root->fn_sernum;
2538 INIT_LIST_HEAD(&iter->w.lh);
2539 fib6_walker_link(net, &iter->w);
2540}
2541
2542static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2543 struct net *net)
2544{
2545 unsigned int h;
2546 struct hlist_node *node;
2547
2548 if (tbl) {
2549 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2550 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2551 } else {
2552 h = 0;
2553 node = NULL;
2554 }
2555
2556 while (!node && h < FIB6_TABLE_HASHSZ) {
2557 node = rcu_dereference_bh(
2558 hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2559 }
2560 return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2561}
2562
2563static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2564{
2565 if (iter->sernum != iter->w.root->fn_sernum) {
2566 iter->sernum = iter->w.root->fn_sernum;
2567 iter->w.state = FWS_INIT;
2568 iter->w.node = iter->w.root;
2569 WARN_ON(iter->w.skip);
2570 iter->w.skip = iter->w.count;
2571 }
2572}
2573
2574static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2575{
2576 int r;
2577 struct fib6_info *n;
2578 struct net *net = seq_file_net(seq);
2579 struct ipv6_route_iter *iter = seq->private;
2580
2581 ++(*pos);
2582 if (!v)
2583 goto iter_table;
2584
2585 n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2586 if (n)
2587 return n;
2588
2589iter_table:
2590 ipv6_route_check_sernum(iter);
2591 spin_lock_bh(&iter->tbl->tb6_lock);
2592 r = fib6_walk_continue(&iter->w);
2593 spin_unlock_bh(&iter->tbl->tb6_lock);
2594 if (r > 0) {
2595 return iter->w.leaf;
2596 } else if (r < 0) {
2597 fib6_walker_unlink(net, &iter->w);
2598 return NULL;
2599 }
2600 fib6_walker_unlink(net, &iter->w);
2601
2602 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2603 if (!iter->tbl)
2604 return NULL;
2605
2606 ipv6_route_seq_setup_walk(iter, net);
2607 goto iter_table;
2608}
2609
2610static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2611 __acquires(RCU_BH)
2612{
2613 struct net *net = seq_file_net(seq);
2614 struct ipv6_route_iter *iter = seq->private;
2615
2616 rcu_read_lock_bh();
2617 iter->tbl = ipv6_route_seq_next_table(NULL, net);
2618 iter->skip = *pos;
2619
2620 if (iter->tbl) {
2621 ipv6_route_seq_setup_walk(iter, net);
2622 return ipv6_route_seq_next(seq, NULL, pos);
2623 } else {
2624 return NULL;
2625 }
2626}
2627
2628static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2629{
2630 struct fib6_walker *w = &iter->w;
2631 return w->node && !(w->state == FWS_U && w->node == w->root);
2632}
2633
2634static void ipv6_route_native_seq_stop(struct seq_file *seq, void *v)
2635 __releases(RCU_BH)
2636{
2637 struct net *net = seq_file_net(seq);
2638 struct ipv6_route_iter *iter = seq->private;
2639
2640 if (ipv6_route_iter_active(iter))
2641 fib6_walker_unlink(net, &iter->w);
2642
2643 rcu_read_unlock_bh();
2644}
2645
2646#if IS_BUILTIN(CONFIG_IPV6) && defined(CONFIG_BPF_SYSCALL)
2647static int ipv6_route_prog_seq_show(struct bpf_prog *prog,
2648 struct bpf_iter_meta *meta,
2649 void *v)
2650{
2651 struct bpf_iter__ipv6_route ctx;
2652
2653 ctx.meta = meta;
2654 ctx.rt = v;
2655 return bpf_iter_run_prog(prog, &ctx);
2656}
2657
2658static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2659{
2660 struct ipv6_route_iter *iter = seq->private;
2661 struct bpf_iter_meta meta;
2662 struct bpf_prog *prog;
2663 int ret;
2664
2665 meta.seq = seq;
2666 prog = bpf_iter_get_info(&meta, false);
2667 if (!prog)
2668 return ipv6_route_native_seq_show(seq, v);
2669
2670 ret = ipv6_route_prog_seq_show(prog, &meta, v);
2671 iter->w.leaf = NULL;
2672
2673 return ret;
2674}
2675
2676static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2677{
2678 struct bpf_iter_meta meta;
2679 struct bpf_prog *prog;
2680
2681 if (!v) {
2682 meta.seq = seq;
2683 prog = bpf_iter_get_info(&meta, true);
2684 if (prog)
2685 (void)ipv6_route_prog_seq_show(prog, &meta, v);
2686 }
2687
2688 ipv6_route_native_seq_stop(seq, v);
2689}
2690#else
2691static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2692{
2693 return ipv6_route_native_seq_show(seq, v);
2694}
2695
2696static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2697{
2698 ipv6_route_native_seq_stop(seq, v);
2699}
2700#endif
2701
2702const struct seq_operations ipv6_route_seq_ops = {
2703 .start = ipv6_route_seq_start,
2704 .next = ipv6_route_seq_next,
2705 .stop = ipv6_route_seq_stop,
2706 .show = ipv6_route_seq_show
2707};
2708#endif /* CONFIG_PROC_FS */