Loading...
1/*
2 * Linux INET6 implementation
3 * Forwarding Information Database
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 */
13
14/*
15 * Changes:
16 * Yuji SEKIYA @USAGI: Support default route on router node;
17 * remove ip6_null_entry from the top of
18 * routing table.
19 * Ville Nuorvala: Fixed routing subtrees.
20 */
21#include <linux/errno.h>
22#include <linux/types.h>
23#include <linux/net.h>
24#include <linux/route.h>
25#include <linux/netdevice.h>
26#include <linux/in6.h>
27#include <linux/init.h>
28#include <linux/list.h>
29#include <linux/slab.h>
30
31#ifdef CONFIG_PROC_FS
32#include <linux/proc_fs.h>
33#endif
34
35#include <net/ipv6.h>
36#include <net/ndisc.h>
37#include <net/addrconf.h>
38
39#include <net/ip6_fib.h>
40#include <net/ip6_route.h>
41
42#define RT6_DEBUG 2
43
44#if RT6_DEBUG >= 3
45#define RT6_TRACE(x...) printk(KERN_DEBUG x)
46#else
47#define RT6_TRACE(x...) do { ; } while (0)
48#endif
49
50static struct kmem_cache * fib6_node_kmem __read_mostly;
51
52enum fib_walk_state_t
53{
54#ifdef CONFIG_IPV6_SUBTREES
55 FWS_S,
56#endif
57 FWS_L,
58 FWS_R,
59 FWS_C,
60 FWS_U
61};
62
63struct fib6_cleaner_t
64{
65 struct fib6_walker_t w;
66 struct net *net;
67 int (*func)(struct rt6_info *, void *arg);
68 void *arg;
69};
70
71static DEFINE_RWLOCK(fib6_walker_lock);
72
73#ifdef CONFIG_IPV6_SUBTREES
74#define FWS_INIT FWS_S
75#else
76#define FWS_INIT FWS_L
77#endif
78
79static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
80 struct rt6_info *rt);
81static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
82static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
83static int fib6_walk(struct fib6_walker_t *w);
84static int fib6_walk_continue(struct fib6_walker_t *w);
85
86/*
87 * A routing update causes an increase of the serial number on the
88 * affected subtree. This allows for cached routes to be asynchronously
89 * tested when modifications are made to the destination cache as a
90 * result of redirects, path MTU changes, etc.
91 */
92
93static __u32 rt_sernum;
94
95static void fib6_gc_timer_cb(unsigned long arg);
96
97static LIST_HEAD(fib6_walkers);
98#define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
99
100static inline void fib6_walker_link(struct fib6_walker_t *w)
101{
102 write_lock_bh(&fib6_walker_lock);
103 list_add(&w->lh, &fib6_walkers);
104 write_unlock_bh(&fib6_walker_lock);
105}
106
107static inline void fib6_walker_unlink(struct fib6_walker_t *w)
108{
109 write_lock_bh(&fib6_walker_lock);
110 list_del(&w->lh);
111 write_unlock_bh(&fib6_walker_lock);
112}
113static __inline__ u32 fib6_new_sernum(void)
114{
115 u32 n = ++rt_sernum;
116 if ((__s32)n <= 0)
117 rt_sernum = n = 1;
118 return n;
119}
120
121/*
122 * Auxiliary address test functions for the radix tree.
123 *
124 * These assume a 32bit processor (although it will work on
125 * 64bit processors)
126 */
127
128/*
129 * test bit
130 */
131#if defined(__LITTLE_ENDIAN)
132# define BITOP_BE32_SWIZZLE (0x1F & ~7)
133#else
134# define BITOP_BE32_SWIZZLE 0
135#endif
136
137static __inline__ __be32 addr_bit_set(const void *token, int fn_bit)
138{
139 const __be32 *addr = token;
140 /*
141 * Here,
142 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
143 * is optimized version of
144 * htonl(1 << ((~fn_bit)&0x1F))
145 * See include/asm-generic/bitops/le.h.
146 */
147 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
148 addr[fn_bit >> 5];
149}
150
151static __inline__ struct fib6_node * node_alloc(void)
152{
153 struct fib6_node *fn;
154
155 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
156
157 return fn;
158}
159
160static __inline__ void node_free(struct fib6_node * fn)
161{
162 kmem_cache_free(fib6_node_kmem, fn);
163}
164
165static __inline__ void rt6_release(struct rt6_info *rt)
166{
167 if (atomic_dec_and_test(&rt->rt6i_ref))
168 dst_free(&rt->dst);
169}
170
171static void fib6_link_table(struct net *net, struct fib6_table *tb)
172{
173 unsigned int h;
174
175 /*
176 * Initialize table lock at a single place to give lockdep a key,
177 * tables aren't visible prior to being linked to the list.
178 */
179 rwlock_init(&tb->tb6_lock);
180
181 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
182
183 /*
184 * No protection necessary, this is the only list mutatation
185 * operation, tables never disappear once they exist.
186 */
187 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
188}
189
190#ifdef CONFIG_IPV6_MULTIPLE_TABLES
191
192static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
193{
194 struct fib6_table *table;
195
196 table = kzalloc(sizeof(*table), GFP_ATOMIC);
197 if (table != NULL) {
198 table->tb6_id = id;
199 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
200 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
201 }
202
203 return table;
204}
205
206struct fib6_table *fib6_new_table(struct net *net, u32 id)
207{
208 struct fib6_table *tb;
209
210 if (id == 0)
211 id = RT6_TABLE_MAIN;
212 tb = fib6_get_table(net, id);
213 if (tb)
214 return tb;
215
216 tb = fib6_alloc_table(net, id);
217 if (tb != NULL)
218 fib6_link_table(net, tb);
219
220 return tb;
221}
222
223struct fib6_table *fib6_get_table(struct net *net, u32 id)
224{
225 struct fib6_table *tb;
226 struct hlist_head *head;
227 struct hlist_node *node;
228 unsigned int h;
229
230 if (id == 0)
231 id = RT6_TABLE_MAIN;
232 h = id & (FIB6_TABLE_HASHSZ - 1);
233 rcu_read_lock();
234 head = &net->ipv6.fib_table_hash[h];
235 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
236 if (tb->tb6_id == id) {
237 rcu_read_unlock();
238 return tb;
239 }
240 }
241 rcu_read_unlock();
242
243 return NULL;
244}
245
246static void __net_init fib6_tables_init(struct net *net)
247{
248 fib6_link_table(net, net->ipv6.fib6_main_tbl);
249 fib6_link_table(net, net->ipv6.fib6_local_tbl);
250}
251#else
252
253struct fib6_table *fib6_new_table(struct net *net, u32 id)
254{
255 return fib6_get_table(net, id);
256}
257
258struct fib6_table *fib6_get_table(struct net *net, u32 id)
259{
260 return net->ipv6.fib6_main_tbl;
261}
262
263struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
264 int flags, pol_lookup_t lookup)
265{
266 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
267}
268
269static void __net_init fib6_tables_init(struct net *net)
270{
271 fib6_link_table(net, net->ipv6.fib6_main_tbl);
272}
273
274#endif
275
276static int fib6_dump_node(struct fib6_walker_t *w)
277{
278 int res;
279 struct rt6_info *rt;
280
281 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
282 res = rt6_dump_route(rt, w->args);
283 if (res < 0) {
284 /* Frame is full, suspend walking */
285 w->leaf = rt;
286 return 1;
287 }
288 WARN_ON(res == 0);
289 }
290 w->leaf = NULL;
291 return 0;
292}
293
294static void fib6_dump_end(struct netlink_callback *cb)
295{
296 struct fib6_walker_t *w = (void*)cb->args[2];
297
298 if (w) {
299 if (cb->args[4]) {
300 cb->args[4] = 0;
301 fib6_walker_unlink(w);
302 }
303 cb->args[2] = 0;
304 kfree(w);
305 }
306 cb->done = (void*)cb->args[3];
307 cb->args[1] = 3;
308}
309
310static int fib6_dump_done(struct netlink_callback *cb)
311{
312 fib6_dump_end(cb);
313 return cb->done ? cb->done(cb) : 0;
314}
315
316static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
317 struct netlink_callback *cb)
318{
319 struct fib6_walker_t *w;
320 int res;
321
322 w = (void *)cb->args[2];
323 w->root = &table->tb6_root;
324
325 if (cb->args[4] == 0) {
326 w->count = 0;
327 w->skip = 0;
328
329 read_lock_bh(&table->tb6_lock);
330 res = fib6_walk(w);
331 read_unlock_bh(&table->tb6_lock);
332 if (res > 0) {
333 cb->args[4] = 1;
334 cb->args[5] = w->root->fn_sernum;
335 }
336 } else {
337 if (cb->args[5] != w->root->fn_sernum) {
338 /* Begin at the root if the tree changed */
339 cb->args[5] = w->root->fn_sernum;
340 w->state = FWS_INIT;
341 w->node = w->root;
342 w->skip = w->count;
343 } else
344 w->skip = 0;
345
346 read_lock_bh(&table->tb6_lock);
347 res = fib6_walk_continue(w);
348 read_unlock_bh(&table->tb6_lock);
349 if (res <= 0) {
350 fib6_walker_unlink(w);
351 cb->args[4] = 0;
352 }
353 }
354
355 return res;
356}
357
358static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
359{
360 struct net *net = sock_net(skb->sk);
361 unsigned int h, s_h;
362 unsigned int e = 0, s_e;
363 struct rt6_rtnl_dump_arg arg;
364 struct fib6_walker_t *w;
365 struct fib6_table *tb;
366 struct hlist_node *node;
367 struct hlist_head *head;
368 int res = 0;
369
370 s_h = cb->args[0];
371 s_e = cb->args[1];
372
373 w = (void *)cb->args[2];
374 if (w == NULL) {
375 /* New dump:
376 *
377 * 1. hook callback destructor.
378 */
379 cb->args[3] = (long)cb->done;
380 cb->done = fib6_dump_done;
381
382 /*
383 * 2. allocate and initialize walker.
384 */
385 w = kzalloc(sizeof(*w), GFP_ATOMIC);
386 if (w == NULL)
387 return -ENOMEM;
388 w->func = fib6_dump_node;
389 cb->args[2] = (long)w;
390 }
391
392 arg.skb = skb;
393 arg.cb = cb;
394 arg.net = net;
395 w->args = &arg;
396
397 rcu_read_lock();
398 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
399 e = 0;
400 head = &net->ipv6.fib_table_hash[h];
401 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
402 if (e < s_e)
403 goto next;
404 res = fib6_dump_table(tb, skb, cb);
405 if (res != 0)
406 goto out;
407next:
408 e++;
409 }
410 }
411out:
412 rcu_read_unlock();
413 cb->args[1] = e;
414 cb->args[0] = h;
415
416 res = res < 0 ? res : skb->len;
417 if (res <= 0)
418 fib6_dump_end(cb);
419 return res;
420}
421
422/*
423 * Routing Table
424 *
425 * return the appropriate node for a routing tree "add" operation
426 * by either creating and inserting or by returning an existing
427 * node.
428 */
429
430static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
431 int addrlen, int plen,
432 int offset)
433{
434 struct fib6_node *fn, *in, *ln;
435 struct fib6_node *pn = NULL;
436 struct rt6key *key;
437 int bit;
438 __be32 dir = 0;
439 __u32 sernum = fib6_new_sernum();
440
441 RT6_TRACE("fib6_add_1\n");
442
443 /* insert node in tree */
444
445 fn = root;
446
447 do {
448 key = (struct rt6key *)((u8 *)fn->leaf + offset);
449
450 /*
451 * Prefix match
452 */
453 if (plen < fn->fn_bit ||
454 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
455 goto insert_above;
456
457 /*
458 * Exact match ?
459 */
460
461 if (plen == fn->fn_bit) {
462 /* clean up an intermediate node */
463 if ((fn->fn_flags & RTN_RTINFO) == 0) {
464 rt6_release(fn->leaf);
465 fn->leaf = NULL;
466 }
467
468 fn->fn_sernum = sernum;
469
470 return fn;
471 }
472
473 /*
474 * We have more bits to go
475 */
476
477 /* Try to walk down on tree. */
478 fn->fn_sernum = sernum;
479 dir = addr_bit_set(addr, fn->fn_bit);
480 pn = fn;
481 fn = dir ? fn->right: fn->left;
482 } while (fn);
483
484 /*
485 * We walked to the bottom of tree.
486 * Create new leaf node without children.
487 */
488
489 ln = node_alloc();
490
491 if (ln == NULL)
492 return NULL;
493 ln->fn_bit = plen;
494
495 ln->parent = pn;
496 ln->fn_sernum = sernum;
497
498 if (dir)
499 pn->right = ln;
500 else
501 pn->left = ln;
502
503 return ln;
504
505
506insert_above:
507 /*
508 * split since we don't have a common prefix anymore or
509 * we have a less significant route.
510 * we've to insert an intermediate node on the list
511 * this new node will point to the one we need to create
512 * and the current
513 */
514
515 pn = fn->parent;
516
517 /* find 1st bit in difference between the 2 addrs.
518
519 See comment in __ipv6_addr_diff: bit may be an invalid value,
520 but if it is >= plen, the value is ignored in any case.
521 */
522
523 bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
524
525 /*
526 * (intermediate)[in]
527 * / \
528 * (new leaf node)[ln] (old node)[fn]
529 */
530 if (plen > bit) {
531 in = node_alloc();
532 ln = node_alloc();
533
534 if (in == NULL || ln == NULL) {
535 if (in)
536 node_free(in);
537 if (ln)
538 node_free(ln);
539 return NULL;
540 }
541
542 /*
543 * new intermediate node.
544 * RTN_RTINFO will
545 * be off since that an address that chooses one of
546 * the branches would not match less specific routes
547 * in the other branch
548 */
549
550 in->fn_bit = bit;
551
552 in->parent = pn;
553 in->leaf = fn->leaf;
554 atomic_inc(&in->leaf->rt6i_ref);
555
556 in->fn_sernum = sernum;
557
558 /* update parent pointer */
559 if (dir)
560 pn->right = in;
561 else
562 pn->left = in;
563
564 ln->fn_bit = plen;
565
566 ln->parent = in;
567 fn->parent = in;
568
569 ln->fn_sernum = sernum;
570
571 if (addr_bit_set(addr, bit)) {
572 in->right = ln;
573 in->left = fn;
574 } else {
575 in->left = ln;
576 in->right = fn;
577 }
578 } else { /* plen <= bit */
579
580 /*
581 * (new leaf node)[ln]
582 * / \
583 * (old node)[fn] NULL
584 */
585
586 ln = node_alloc();
587
588 if (ln == NULL)
589 return NULL;
590
591 ln->fn_bit = plen;
592
593 ln->parent = pn;
594
595 ln->fn_sernum = sernum;
596
597 if (dir)
598 pn->right = ln;
599 else
600 pn->left = ln;
601
602 if (addr_bit_set(&key->addr, plen))
603 ln->right = fn;
604 else
605 ln->left = fn;
606
607 fn->parent = ln;
608 }
609 return ln;
610}
611
612/*
613 * Insert routing information in a node.
614 */
615
616static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
617 struct nl_info *info)
618{
619 struct rt6_info *iter = NULL;
620 struct rt6_info **ins;
621
622 ins = &fn->leaf;
623
624 for (iter = fn->leaf; iter; iter=iter->dst.rt6_next) {
625 /*
626 * Search for duplicates
627 */
628
629 if (iter->rt6i_metric == rt->rt6i_metric) {
630 /*
631 * Same priority level
632 */
633
634 if (iter->rt6i_dev == rt->rt6i_dev &&
635 iter->rt6i_idev == rt->rt6i_idev &&
636 ipv6_addr_equal(&iter->rt6i_gateway,
637 &rt->rt6i_gateway)) {
638 if (!(iter->rt6i_flags&RTF_EXPIRES))
639 return -EEXIST;
640 iter->rt6i_expires = rt->rt6i_expires;
641 if (!(rt->rt6i_flags&RTF_EXPIRES)) {
642 iter->rt6i_flags &= ~RTF_EXPIRES;
643 iter->rt6i_expires = 0;
644 }
645 return -EEXIST;
646 }
647 }
648
649 if (iter->rt6i_metric > rt->rt6i_metric)
650 break;
651
652 ins = &iter->dst.rt6_next;
653 }
654
655 /* Reset round-robin state, if necessary */
656 if (ins == &fn->leaf)
657 fn->rr_ptr = NULL;
658
659 /*
660 * insert node
661 */
662
663 rt->dst.rt6_next = iter;
664 *ins = rt;
665 rt->rt6i_node = fn;
666 atomic_inc(&rt->rt6i_ref);
667 inet6_rt_notify(RTM_NEWROUTE, rt, info);
668 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
669
670 if ((fn->fn_flags & RTN_RTINFO) == 0) {
671 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
672 fn->fn_flags |= RTN_RTINFO;
673 }
674
675 return 0;
676}
677
678static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
679{
680 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
681 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE)))
682 mod_timer(&net->ipv6.ip6_fib_timer,
683 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
684}
685
686void fib6_force_start_gc(struct net *net)
687{
688 if (!timer_pending(&net->ipv6.ip6_fib_timer))
689 mod_timer(&net->ipv6.ip6_fib_timer,
690 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
691}
692
693/*
694 * Add routing information to the routing tree.
695 * <destination addr>/<source addr>
696 * with source addr info in sub-trees
697 */
698
699int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
700{
701 struct fib6_node *fn, *pn = NULL;
702 int err = -ENOMEM;
703
704 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
705 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst));
706
707 if (fn == NULL)
708 goto out;
709
710 pn = fn;
711
712#ifdef CONFIG_IPV6_SUBTREES
713 if (rt->rt6i_src.plen) {
714 struct fib6_node *sn;
715
716 if (fn->subtree == NULL) {
717 struct fib6_node *sfn;
718
719 /*
720 * Create subtree.
721 *
722 * fn[main tree]
723 * |
724 * sfn[subtree root]
725 * \
726 * sn[new leaf node]
727 */
728
729 /* Create subtree root node */
730 sfn = node_alloc();
731 if (sfn == NULL)
732 goto st_failure;
733
734 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
735 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
736 sfn->fn_flags = RTN_ROOT;
737 sfn->fn_sernum = fib6_new_sernum();
738
739 /* Now add the first leaf node to new subtree */
740
741 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
742 sizeof(struct in6_addr), rt->rt6i_src.plen,
743 offsetof(struct rt6_info, rt6i_src));
744
745 if (sn == NULL) {
746 /* If it is failed, discard just allocated
747 root, and then (in st_failure) stale node
748 in main tree.
749 */
750 node_free(sfn);
751 goto st_failure;
752 }
753
754 /* Now link new subtree to main tree */
755 sfn->parent = fn;
756 fn->subtree = sfn;
757 } else {
758 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
759 sizeof(struct in6_addr), rt->rt6i_src.plen,
760 offsetof(struct rt6_info, rt6i_src));
761
762 if (sn == NULL)
763 goto st_failure;
764 }
765
766 if (fn->leaf == NULL) {
767 fn->leaf = rt;
768 atomic_inc(&rt->rt6i_ref);
769 }
770 fn = sn;
771 }
772#endif
773
774 err = fib6_add_rt2node(fn, rt, info);
775
776 if (err == 0) {
777 fib6_start_gc(info->nl_net, rt);
778 if (!(rt->rt6i_flags&RTF_CACHE))
779 fib6_prune_clones(info->nl_net, pn, rt);
780 }
781
782out:
783 if (err) {
784#ifdef CONFIG_IPV6_SUBTREES
785 /*
786 * If fib6_add_1 has cleared the old leaf pointer in the
787 * super-tree leaf node we have to find a new one for it.
788 */
789 if (pn != fn && pn->leaf == rt) {
790 pn->leaf = NULL;
791 atomic_dec(&rt->rt6i_ref);
792 }
793 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
794 pn->leaf = fib6_find_prefix(info->nl_net, pn);
795#if RT6_DEBUG >= 2
796 if (!pn->leaf) {
797 WARN_ON(pn->leaf == NULL);
798 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
799 }
800#endif
801 atomic_inc(&pn->leaf->rt6i_ref);
802 }
803#endif
804 dst_free(&rt->dst);
805 }
806 return err;
807
808#ifdef CONFIG_IPV6_SUBTREES
809 /* Subtree creation failed, probably main tree node
810 is orphan. If it is, shoot it.
811 */
812st_failure:
813 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
814 fib6_repair_tree(info->nl_net, fn);
815 dst_free(&rt->dst);
816 return err;
817#endif
818}
819
820/*
821 * Routing tree lookup
822 *
823 */
824
825struct lookup_args {
826 int offset; /* key offset on rt6_info */
827 const struct in6_addr *addr; /* search key */
828};
829
830static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
831 struct lookup_args *args)
832{
833 struct fib6_node *fn;
834 __be32 dir;
835
836 if (unlikely(args->offset == 0))
837 return NULL;
838
839 /*
840 * Descend on a tree
841 */
842
843 fn = root;
844
845 for (;;) {
846 struct fib6_node *next;
847
848 dir = addr_bit_set(args->addr, fn->fn_bit);
849
850 next = dir ? fn->right : fn->left;
851
852 if (next) {
853 fn = next;
854 continue;
855 }
856
857 break;
858 }
859
860 while(fn) {
861 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
862 struct rt6key *key;
863
864 key = (struct rt6key *) ((u8 *) fn->leaf +
865 args->offset);
866
867 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
868#ifdef CONFIG_IPV6_SUBTREES
869 if (fn->subtree)
870 fn = fib6_lookup_1(fn->subtree, args + 1);
871#endif
872 if (!fn || fn->fn_flags & RTN_RTINFO)
873 return fn;
874 }
875 }
876
877 if (fn->fn_flags & RTN_ROOT)
878 break;
879
880 fn = fn->parent;
881 }
882
883 return NULL;
884}
885
886struct fib6_node * fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
887 const struct in6_addr *saddr)
888{
889 struct fib6_node *fn;
890 struct lookup_args args[] = {
891 {
892 .offset = offsetof(struct rt6_info, rt6i_dst),
893 .addr = daddr,
894 },
895#ifdef CONFIG_IPV6_SUBTREES
896 {
897 .offset = offsetof(struct rt6_info, rt6i_src),
898 .addr = saddr,
899 },
900#endif
901 {
902 .offset = 0, /* sentinel */
903 }
904 };
905
906 fn = fib6_lookup_1(root, daddr ? args : args + 1);
907
908 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT)
909 fn = root;
910
911 return fn;
912}
913
914/*
915 * Get node with specified destination prefix (and source prefix,
916 * if subtrees are used)
917 */
918
919
920static struct fib6_node * fib6_locate_1(struct fib6_node *root,
921 const struct in6_addr *addr,
922 int plen, int offset)
923{
924 struct fib6_node *fn;
925
926 for (fn = root; fn ; ) {
927 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
928
929 /*
930 * Prefix match
931 */
932 if (plen < fn->fn_bit ||
933 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
934 return NULL;
935
936 if (plen == fn->fn_bit)
937 return fn;
938
939 /*
940 * We have more bits to go
941 */
942 if (addr_bit_set(addr, fn->fn_bit))
943 fn = fn->right;
944 else
945 fn = fn->left;
946 }
947 return NULL;
948}
949
950struct fib6_node * fib6_locate(struct fib6_node *root,
951 const struct in6_addr *daddr, int dst_len,
952 const struct in6_addr *saddr, int src_len)
953{
954 struct fib6_node *fn;
955
956 fn = fib6_locate_1(root, daddr, dst_len,
957 offsetof(struct rt6_info, rt6i_dst));
958
959#ifdef CONFIG_IPV6_SUBTREES
960 if (src_len) {
961 WARN_ON(saddr == NULL);
962 if (fn && fn->subtree)
963 fn = fib6_locate_1(fn->subtree, saddr, src_len,
964 offsetof(struct rt6_info, rt6i_src));
965 }
966#endif
967
968 if (fn && fn->fn_flags&RTN_RTINFO)
969 return fn;
970
971 return NULL;
972}
973
974
975/*
976 * Deletion
977 *
978 */
979
980static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
981{
982 if (fn->fn_flags&RTN_ROOT)
983 return net->ipv6.ip6_null_entry;
984
985 while(fn) {
986 if(fn->left)
987 return fn->left->leaf;
988
989 if(fn->right)
990 return fn->right->leaf;
991
992 fn = FIB6_SUBTREE(fn);
993 }
994 return NULL;
995}
996
997/*
998 * Called to trim the tree of intermediate nodes when possible. "fn"
999 * is the node we want to try and remove.
1000 */
1001
1002static struct fib6_node *fib6_repair_tree(struct net *net,
1003 struct fib6_node *fn)
1004{
1005 int children;
1006 int nstate;
1007 struct fib6_node *child, *pn;
1008 struct fib6_walker_t *w;
1009 int iter = 0;
1010
1011 for (;;) {
1012 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1013 iter++;
1014
1015 WARN_ON(fn->fn_flags & RTN_RTINFO);
1016 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1017 WARN_ON(fn->leaf != NULL);
1018
1019 children = 0;
1020 child = NULL;
1021 if (fn->right) child = fn->right, children |= 1;
1022 if (fn->left) child = fn->left, children |= 2;
1023
1024 if (children == 3 || FIB6_SUBTREE(fn)
1025#ifdef CONFIG_IPV6_SUBTREES
1026 /* Subtree root (i.e. fn) may have one child */
1027 || (children && fn->fn_flags&RTN_ROOT)
1028#endif
1029 ) {
1030 fn->leaf = fib6_find_prefix(net, fn);
1031#if RT6_DEBUG >= 2
1032 if (fn->leaf==NULL) {
1033 WARN_ON(!fn->leaf);
1034 fn->leaf = net->ipv6.ip6_null_entry;
1035 }
1036#endif
1037 atomic_inc(&fn->leaf->rt6i_ref);
1038 return fn->parent;
1039 }
1040
1041 pn = fn->parent;
1042#ifdef CONFIG_IPV6_SUBTREES
1043 if (FIB6_SUBTREE(pn) == fn) {
1044 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1045 FIB6_SUBTREE(pn) = NULL;
1046 nstate = FWS_L;
1047 } else {
1048 WARN_ON(fn->fn_flags & RTN_ROOT);
1049#endif
1050 if (pn->right == fn) pn->right = child;
1051 else if (pn->left == fn) pn->left = child;
1052#if RT6_DEBUG >= 2
1053 else
1054 WARN_ON(1);
1055#endif
1056 if (child)
1057 child->parent = pn;
1058 nstate = FWS_R;
1059#ifdef CONFIG_IPV6_SUBTREES
1060 }
1061#endif
1062
1063 read_lock(&fib6_walker_lock);
1064 FOR_WALKERS(w) {
1065 if (child == NULL) {
1066 if (w->root == fn) {
1067 w->root = w->node = NULL;
1068 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1069 } else if (w->node == fn) {
1070 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1071 w->node = pn;
1072 w->state = nstate;
1073 }
1074 } else {
1075 if (w->root == fn) {
1076 w->root = child;
1077 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1078 }
1079 if (w->node == fn) {
1080 w->node = child;
1081 if (children&2) {
1082 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1083 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1084 } else {
1085 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1086 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1087 }
1088 }
1089 }
1090 }
1091 read_unlock(&fib6_walker_lock);
1092
1093 node_free(fn);
1094 if (pn->fn_flags&RTN_RTINFO || FIB6_SUBTREE(pn))
1095 return pn;
1096
1097 rt6_release(pn->leaf);
1098 pn->leaf = NULL;
1099 fn = pn;
1100 }
1101}
1102
1103static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1104 struct nl_info *info)
1105{
1106 struct fib6_walker_t *w;
1107 struct rt6_info *rt = *rtp;
1108 struct net *net = info->nl_net;
1109
1110 RT6_TRACE("fib6_del_route\n");
1111
1112 /* Unlink it */
1113 *rtp = rt->dst.rt6_next;
1114 rt->rt6i_node = NULL;
1115 net->ipv6.rt6_stats->fib_rt_entries--;
1116 net->ipv6.rt6_stats->fib_discarded_routes++;
1117
1118 /* Reset round-robin state, if necessary */
1119 if (fn->rr_ptr == rt)
1120 fn->rr_ptr = NULL;
1121
1122 /* Adjust walkers */
1123 read_lock(&fib6_walker_lock);
1124 FOR_WALKERS(w) {
1125 if (w->state == FWS_C && w->leaf == rt) {
1126 RT6_TRACE("walker %p adjusted by delroute\n", w);
1127 w->leaf = rt->dst.rt6_next;
1128 if (w->leaf == NULL)
1129 w->state = FWS_U;
1130 }
1131 }
1132 read_unlock(&fib6_walker_lock);
1133
1134 rt->dst.rt6_next = NULL;
1135
1136 /* If it was last route, expunge its radix tree node */
1137 if (fn->leaf == NULL) {
1138 fn->fn_flags &= ~RTN_RTINFO;
1139 net->ipv6.rt6_stats->fib_route_nodes--;
1140 fn = fib6_repair_tree(net, fn);
1141 }
1142
1143 if (atomic_read(&rt->rt6i_ref) != 1) {
1144 /* This route is used as dummy address holder in some split
1145 * nodes. It is not leaked, but it still holds other resources,
1146 * which must be released in time. So, scan ascendant nodes
1147 * and replace dummy references to this route with references
1148 * to still alive ones.
1149 */
1150 while (fn) {
1151 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) {
1152 fn->leaf = fib6_find_prefix(net, fn);
1153 atomic_inc(&fn->leaf->rt6i_ref);
1154 rt6_release(rt);
1155 }
1156 fn = fn->parent;
1157 }
1158 /* No more references are possible at this point. */
1159 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1160 }
1161
1162 inet6_rt_notify(RTM_DELROUTE, rt, info);
1163 rt6_release(rt);
1164}
1165
1166int fib6_del(struct rt6_info *rt, struct nl_info *info)
1167{
1168 struct net *net = info->nl_net;
1169 struct fib6_node *fn = rt->rt6i_node;
1170 struct rt6_info **rtp;
1171
1172#if RT6_DEBUG >= 2
1173 if (rt->dst.obsolete>0) {
1174 WARN_ON(fn != NULL);
1175 return -ENOENT;
1176 }
1177#endif
1178 if (fn == NULL || rt == net->ipv6.ip6_null_entry)
1179 return -ENOENT;
1180
1181 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1182
1183 if (!(rt->rt6i_flags&RTF_CACHE)) {
1184 struct fib6_node *pn = fn;
1185#ifdef CONFIG_IPV6_SUBTREES
1186 /* clones of this route might be in another subtree */
1187 if (rt->rt6i_src.plen) {
1188 while (!(pn->fn_flags&RTN_ROOT))
1189 pn = pn->parent;
1190 pn = pn->parent;
1191 }
1192#endif
1193 fib6_prune_clones(info->nl_net, pn, rt);
1194 }
1195
1196 /*
1197 * Walk the leaf entries looking for ourself
1198 */
1199
1200 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1201 if (*rtp == rt) {
1202 fib6_del_route(fn, rtp, info);
1203 return 0;
1204 }
1205 }
1206 return -ENOENT;
1207}
1208
1209/*
1210 * Tree traversal function.
1211 *
1212 * Certainly, it is not interrupt safe.
1213 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1214 * It means, that we can modify tree during walking
1215 * and use this function for garbage collection, clone pruning,
1216 * cleaning tree when a device goes down etc. etc.
1217 *
1218 * It guarantees that every node will be traversed,
1219 * and that it will be traversed only once.
1220 *
1221 * Callback function w->func may return:
1222 * 0 -> continue walking.
1223 * positive value -> walking is suspended (used by tree dumps,
1224 * and probably by gc, if it will be split to several slices)
1225 * negative value -> terminate walking.
1226 *
1227 * The function itself returns:
1228 * 0 -> walk is complete.
1229 * >0 -> walk is incomplete (i.e. suspended)
1230 * <0 -> walk is terminated by an error.
1231 */
1232
1233static int fib6_walk_continue(struct fib6_walker_t *w)
1234{
1235 struct fib6_node *fn, *pn;
1236
1237 for (;;) {
1238 fn = w->node;
1239 if (fn == NULL)
1240 return 0;
1241
1242 if (w->prune && fn != w->root &&
1243 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) {
1244 w->state = FWS_C;
1245 w->leaf = fn->leaf;
1246 }
1247 switch (w->state) {
1248#ifdef CONFIG_IPV6_SUBTREES
1249 case FWS_S:
1250 if (FIB6_SUBTREE(fn)) {
1251 w->node = FIB6_SUBTREE(fn);
1252 continue;
1253 }
1254 w->state = FWS_L;
1255#endif
1256 case FWS_L:
1257 if (fn->left) {
1258 w->node = fn->left;
1259 w->state = FWS_INIT;
1260 continue;
1261 }
1262 w->state = FWS_R;
1263 case FWS_R:
1264 if (fn->right) {
1265 w->node = fn->right;
1266 w->state = FWS_INIT;
1267 continue;
1268 }
1269 w->state = FWS_C;
1270 w->leaf = fn->leaf;
1271 case FWS_C:
1272 if (w->leaf && fn->fn_flags&RTN_RTINFO) {
1273 int err;
1274
1275 if (w->count < w->skip) {
1276 w->count++;
1277 continue;
1278 }
1279
1280 err = w->func(w);
1281 if (err)
1282 return err;
1283
1284 w->count++;
1285 continue;
1286 }
1287 w->state = FWS_U;
1288 case FWS_U:
1289 if (fn == w->root)
1290 return 0;
1291 pn = fn->parent;
1292 w->node = pn;
1293#ifdef CONFIG_IPV6_SUBTREES
1294 if (FIB6_SUBTREE(pn) == fn) {
1295 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1296 w->state = FWS_L;
1297 continue;
1298 }
1299#endif
1300 if (pn->left == fn) {
1301 w->state = FWS_R;
1302 continue;
1303 }
1304 if (pn->right == fn) {
1305 w->state = FWS_C;
1306 w->leaf = w->node->leaf;
1307 continue;
1308 }
1309#if RT6_DEBUG >= 2
1310 WARN_ON(1);
1311#endif
1312 }
1313 }
1314}
1315
1316static int fib6_walk(struct fib6_walker_t *w)
1317{
1318 int res;
1319
1320 w->state = FWS_INIT;
1321 w->node = w->root;
1322
1323 fib6_walker_link(w);
1324 res = fib6_walk_continue(w);
1325 if (res <= 0)
1326 fib6_walker_unlink(w);
1327 return res;
1328}
1329
1330static int fib6_clean_node(struct fib6_walker_t *w)
1331{
1332 int res;
1333 struct rt6_info *rt;
1334 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1335 struct nl_info info = {
1336 .nl_net = c->net,
1337 };
1338
1339 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1340 res = c->func(rt, c->arg);
1341 if (res < 0) {
1342 w->leaf = rt;
1343 res = fib6_del(rt, &info);
1344 if (res) {
1345#if RT6_DEBUG >= 2
1346 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res);
1347#endif
1348 continue;
1349 }
1350 return 0;
1351 }
1352 WARN_ON(res != 0);
1353 }
1354 w->leaf = rt;
1355 return 0;
1356}
1357
1358/*
1359 * Convenient frontend to tree walker.
1360 *
1361 * func is called on each route.
1362 * It may return -1 -> delete this route.
1363 * 0 -> continue walking
1364 *
1365 * prune==1 -> only immediate children of node (certainly,
1366 * ignoring pure split nodes) will be scanned.
1367 */
1368
1369static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1370 int (*func)(struct rt6_info *, void *arg),
1371 int prune, void *arg)
1372{
1373 struct fib6_cleaner_t c;
1374
1375 c.w.root = root;
1376 c.w.func = fib6_clean_node;
1377 c.w.prune = prune;
1378 c.w.count = 0;
1379 c.w.skip = 0;
1380 c.func = func;
1381 c.arg = arg;
1382 c.net = net;
1383
1384 fib6_walk(&c.w);
1385}
1386
1387void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1388 int prune, void *arg)
1389{
1390 struct fib6_table *table;
1391 struct hlist_node *node;
1392 struct hlist_head *head;
1393 unsigned int h;
1394
1395 rcu_read_lock();
1396 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1397 head = &net->ipv6.fib_table_hash[h];
1398 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1399 write_lock_bh(&table->tb6_lock);
1400 fib6_clean_tree(net, &table->tb6_root,
1401 func, prune, arg);
1402 write_unlock_bh(&table->tb6_lock);
1403 }
1404 }
1405 rcu_read_unlock();
1406}
1407
1408static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1409{
1410 if (rt->rt6i_flags & RTF_CACHE) {
1411 RT6_TRACE("pruning clone %p\n", rt);
1412 return -1;
1413 }
1414
1415 return 0;
1416}
1417
1418static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1419 struct rt6_info *rt)
1420{
1421 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1422}
1423
1424/*
1425 * Garbage collection
1426 */
1427
1428static struct fib6_gc_args
1429{
1430 int timeout;
1431 int more;
1432} gc_args;
1433
1434static int fib6_age(struct rt6_info *rt, void *arg)
1435{
1436 unsigned long now = jiffies;
1437
1438 /*
1439 * check addrconf expiration here.
1440 * Routes are expired even if they are in use.
1441 *
1442 * Also age clones. Note, that clones are aged out
1443 * only if they are not in use now.
1444 */
1445
1446 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) {
1447 if (time_after(now, rt->rt6i_expires)) {
1448 RT6_TRACE("expiring %p\n", rt);
1449 return -1;
1450 }
1451 gc_args.more++;
1452 } else if (rt->rt6i_flags & RTF_CACHE) {
1453 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1454 time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1455 RT6_TRACE("aging clone %p\n", rt);
1456 return -1;
1457 } else if ((rt->rt6i_flags & RTF_GATEWAY) &&
1458 (!(dst_get_neighbour_raw(&rt->dst)->flags & NTF_ROUTER))) {
1459 RT6_TRACE("purging route %p via non-router but gateway\n",
1460 rt);
1461 return -1;
1462 }
1463 gc_args.more++;
1464 }
1465
1466 return 0;
1467}
1468
1469static DEFINE_SPINLOCK(fib6_gc_lock);
1470
1471void fib6_run_gc(unsigned long expires, struct net *net)
1472{
1473 if (expires != ~0UL) {
1474 spin_lock_bh(&fib6_gc_lock);
1475 gc_args.timeout = expires ? (int)expires :
1476 net->ipv6.sysctl.ip6_rt_gc_interval;
1477 } else {
1478 if (!spin_trylock_bh(&fib6_gc_lock)) {
1479 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1480 return;
1481 }
1482 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1483 }
1484
1485 gc_args.more = icmp6_dst_gc();
1486
1487 fib6_clean_all(net, fib6_age, 0, NULL);
1488
1489 if (gc_args.more)
1490 mod_timer(&net->ipv6.ip6_fib_timer,
1491 round_jiffies(jiffies
1492 + net->ipv6.sysctl.ip6_rt_gc_interval));
1493 else
1494 del_timer(&net->ipv6.ip6_fib_timer);
1495 spin_unlock_bh(&fib6_gc_lock);
1496}
1497
1498static void fib6_gc_timer_cb(unsigned long arg)
1499{
1500 fib6_run_gc(0, (struct net *)arg);
1501}
1502
1503static int __net_init fib6_net_init(struct net *net)
1504{
1505 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1506
1507 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1508
1509 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1510 if (!net->ipv6.rt6_stats)
1511 goto out_timer;
1512
1513 /* Avoid false sharing : Use at least a full cache line */
1514 size = max_t(size_t, size, L1_CACHE_BYTES);
1515
1516 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1517 if (!net->ipv6.fib_table_hash)
1518 goto out_rt6_stats;
1519
1520 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1521 GFP_KERNEL);
1522 if (!net->ipv6.fib6_main_tbl)
1523 goto out_fib_table_hash;
1524
1525 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1526 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1527 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1528 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1529
1530#ifdef CONFIG_IPV6_MULTIPLE_TABLES
1531 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1532 GFP_KERNEL);
1533 if (!net->ipv6.fib6_local_tbl)
1534 goto out_fib6_main_tbl;
1535 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1536 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1537 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1538 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1539#endif
1540 fib6_tables_init(net);
1541
1542 return 0;
1543
1544#ifdef CONFIG_IPV6_MULTIPLE_TABLES
1545out_fib6_main_tbl:
1546 kfree(net->ipv6.fib6_main_tbl);
1547#endif
1548out_fib_table_hash:
1549 kfree(net->ipv6.fib_table_hash);
1550out_rt6_stats:
1551 kfree(net->ipv6.rt6_stats);
1552out_timer:
1553 return -ENOMEM;
1554 }
1555
1556static void fib6_net_exit(struct net *net)
1557{
1558 rt6_ifdown(net, NULL);
1559 del_timer_sync(&net->ipv6.ip6_fib_timer);
1560
1561#ifdef CONFIG_IPV6_MULTIPLE_TABLES
1562 kfree(net->ipv6.fib6_local_tbl);
1563#endif
1564 kfree(net->ipv6.fib6_main_tbl);
1565 kfree(net->ipv6.fib_table_hash);
1566 kfree(net->ipv6.rt6_stats);
1567}
1568
1569static struct pernet_operations fib6_net_ops = {
1570 .init = fib6_net_init,
1571 .exit = fib6_net_exit,
1572};
1573
1574int __init fib6_init(void)
1575{
1576 int ret = -ENOMEM;
1577
1578 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1579 sizeof(struct fib6_node),
1580 0, SLAB_HWCACHE_ALIGN,
1581 NULL);
1582 if (!fib6_node_kmem)
1583 goto out;
1584
1585 ret = register_pernet_subsys(&fib6_net_ops);
1586 if (ret)
1587 goto out_kmem_cache_create;
1588
1589 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1590 NULL);
1591 if (ret)
1592 goto out_unregister_subsys;
1593out:
1594 return ret;
1595
1596out_unregister_subsys:
1597 unregister_pernet_subsys(&fib6_net_ops);
1598out_kmem_cache_create:
1599 kmem_cache_destroy(fib6_node_kmem);
1600 goto out;
1601}
1602
1603void fib6_gc_cleanup(void)
1604{
1605 unregister_pernet_subsys(&fib6_net_ops);
1606 kmem_cache_destroy(fib6_node_kmem);
1607}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Linux INET6 implementation
4 * Forwarding Information Database
5 *
6 * Authors:
7 * Pedro Roque <roque@di.fc.ul.pt>
8 *
9 * Changes:
10 * Yuji SEKIYA @USAGI: Support default route on router node;
11 * remove ip6_null_entry from the top of
12 * routing table.
13 * Ville Nuorvala: Fixed routing subtrees.
14 */
15
16#define pr_fmt(fmt) "IPv6: " fmt
17
18#include <linux/bpf.h>
19#include <linux/errno.h>
20#include <linux/types.h>
21#include <linux/net.h>
22#include <linux/route.h>
23#include <linux/netdevice.h>
24#include <linux/in6.h>
25#include <linux/init.h>
26#include <linux/list.h>
27#include <linux/slab.h>
28
29#include <net/ip.h>
30#include <net/ipv6.h>
31#include <net/ndisc.h>
32#include <net/addrconf.h>
33#include <net/lwtunnel.h>
34#include <net/fib_notifier.h>
35
36#include <net/ip_fib.h>
37#include <net/ip6_fib.h>
38#include <net/ip6_route.h>
39
40static struct kmem_cache *fib6_node_kmem __read_mostly;
41
42struct fib6_cleaner {
43 struct fib6_walker w;
44 struct net *net;
45 int (*func)(struct fib6_info *, void *arg);
46 int sernum;
47 void *arg;
48 bool skip_notify;
49};
50
51#ifdef CONFIG_IPV6_SUBTREES
52#define FWS_INIT FWS_S
53#else
54#define FWS_INIT FWS_L
55#endif
56
57static struct fib6_info *fib6_find_prefix(struct net *net,
58 struct fib6_table *table,
59 struct fib6_node *fn);
60static struct fib6_node *fib6_repair_tree(struct net *net,
61 struct fib6_table *table,
62 struct fib6_node *fn);
63static int fib6_walk(struct net *net, struct fib6_walker *w);
64static int fib6_walk_continue(struct fib6_walker *w);
65
66/*
67 * A routing update causes an increase of the serial number on the
68 * affected subtree. This allows for cached routes to be asynchronously
69 * tested when modifications are made to the destination cache as a
70 * result of redirects, path MTU changes, etc.
71 */
72
73static void fib6_gc_timer_cb(struct timer_list *t);
74
75#define FOR_WALKERS(net, w) \
76 list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
77
78static void fib6_walker_link(struct net *net, struct fib6_walker *w)
79{
80 write_lock_bh(&net->ipv6.fib6_walker_lock);
81 list_add(&w->lh, &net->ipv6.fib6_walkers);
82 write_unlock_bh(&net->ipv6.fib6_walker_lock);
83}
84
85static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
86{
87 write_lock_bh(&net->ipv6.fib6_walker_lock);
88 list_del(&w->lh);
89 write_unlock_bh(&net->ipv6.fib6_walker_lock);
90}
91
92static int fib6_new_sernum(struct net *net)
93{
94 int new, old = atomic_read(&net->ipv6.fib6_sernum);
95
96 do {
97 new = old < INT_MAX ? old + 1 : 1;
98 } while (!atomic_try_cmpxchg(&net->ipv6.fib6_sernum, &old, new));
99
100 return new;
101}
102
103enum {
104 FIB6_NO_SERNUM_CHANGE = 0,
105};
106
107void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
108{
109 struct fib6_node *fn;
110
111 fn = rcu_dereference_protected(f6i->fib6_node,
112 lockdep_is_held(&f6i->fib6_table->tb6_lock));
113 if (fn)
114 WRITE_ONCE(fn->fn_sernum, fib6_new_sernum(net));
115}
116
117/*
118 * Auxiliary address test functions for the radix tree.
119 *
120 * These assume a 32bit processor (although it will work on
121 * 64bit processors)
122 */
123
124/*
125 * test bit
126 */
127#if defined(__LITTLE_ENDIAN)
128# define BITOP_BE32_SWIZZLE (0x1F & ~7)
129#else
130# define BITOP_BE32_SWIZZLE 0
131#endif
132
133static __be32 addr_bit_set(const void *token, int fn_bit)
134{
135 const __be32 *addr = token;
136 /*
137 * Here,
138 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
139 * is optimized version of
140 * htonl(1 << ((~fn_bit)&0x1F))
141 * See include/asm-generic/bitops/le.h.
142 */
143 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
144 addr[fn_bit >> 5];
145}
146
147struct fib6_info *fib6_info_alloc(gfp_t gfp_flags, bool with_fib6_nh)
148{
149 struct fib6_info *f6i;
150 size_t sz = sizeof(*f6i);
151
152 if (with_fib6_nh)
153 sz += sizeof(struct fib6_nh);
154
155 f6i = kzalloc(sz, gfp_flags);
156 if (!f6i)
157 return NULL;
158
159 /* fib6_siblings is a union with nh_list, so this initializes both */
160 INIT_LIST_HEAD(&f6i->fib6_siblings);
161 refcount_set(&f6i->fib6_ref, 1);
162
163 return f6i;
164}
165
166void fib6_info_destroy_rcu(struct rcu_head *head)
167{
168 struct fib6_info *f6i = container_of(head, struct fib6_info, rcu);
169
170 WARN_ON(f6i->fib6_node);
171
172 if (f6i->nh)
173 nexthop_put(f6i->nh);
174 else
175 fib6_nh_release(f6i->fib6_nh);
176
177 ip_fib_metrics_put(f6i->fib6_metrics);
178 kfree(f6i);
179}
180EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
181
182static struct fib6_node *node_alloc(struct net *net)
183{
184 struct fib6_node *fn;
185
186 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
187 if (fn)
188 net->ipv6.rt6_stats->fib_nodes++;
189
190 return fn;
191}
192
193static void node_free_immediate(struct net *net, struct fib6_node *fn)
194{
195 kmem_cache_free(fib6_node_kmem, fn);
196 net->ipv6.rt6_stats->fib_nodes--;
197}
198
199static void node_free_rcu(struct rcu_head *head)
200{
201 struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
202
203 kmem_cache_free(fib6_node_kmem, fn);
204}
205
206static void node_free(struct net *net, struct fib6_node *fn)
207{
208 call_rcu(&fn->rcu, node_free_rcu);
209 net->ipv6.rt6_stats->fib_nodes--;
210}
211
212static void fib6_free_table(struct fib6_table *table)
213{
214 inetpeer_invalidate_tree(&table->tb6_peers);
215 kfree(table);
216}
217
218static void fib6_link_table(struct net *net, struct fib6_table *tb)
219{
220 unsigned int h;
221
222 /*
223 * Initialize table lock at a single place to give lockdep a key,
224 * tables aren't visible prior to being linked to the list.
225 */
226 spin_lock_init(&tb->tb6_lock);
227 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
228
229 /*
230 * No protection necessary, this is the only list mutatation
231 * operation, tables never disappear once they exist.
232 */
233 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
234}
235
236#ifdef CONFIG_IPV6_MULTIPLE_TABLES
237
238static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
239{
240 struct fib6_table *table;
241
242 table = kzalloc(sizeof(*table), GFP_ATOMIC);
243 if (table) {
244 table->tb6_id = id;
245 rcu_assign_pointer(table->tb6_root.leaf,
246 net->ipv6.fib6_null_entry);
247 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
248 inet_peer_base_init(&table->tb6_peers);
249 }
250
251 return table;
252}
253
254struct fib6_table *fib6_new_table(struct net *net, u32 id)
255{
256 struct fib6_table *tb;
257
258 if (id == 0)
259 id = RT6_TABLE_MAIN;
260 tb = fib6_get_table(net, id);
261 if (tb)
262 return tb;
263
264 tb = fib6_alloc_table(net, id);
265 if (tb)
266 fib6_link_table(net, tb);
267
268 return tb;
269}
270EXPORT_SYMBOL_GPL(fib6_new_table);
271
272struct fib6_table *fib6_get_table(struct net *net, u32 id)
273{
274 struct fib6_table *tb;
275 struct hlist_head *head;
276 unsigned int h;
277
278 if (id == 0)
279 id = RT6_TABLE_MAIN;
280 h = id & (FIB6_TABLE_HASHSZ - 1);
281 rcu_read_lock();
282 head = &net->ipv6.fib_table_hash[h];
283 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
284 if (tb->tb6_id == id) {
285 rcu_read_unlock();
286 return tb;
287 }
288 }
289 rcu_read_unlock();
290
291 return NULL;
292}
293EXPORT_SYMBOL_GPL(fib6_get_table);
294
295static void __net_init fib6_tables_init(struct net *net)
296{
297 fib6_link_table(net, net->ipv6.fib6_main_tbl);
298 fib6_link_table(net, net->ipv6.fib6_local_tbl);
299}
300#else
301
302struct fib6_table *fib6_new_table(struct net *net, u32 id)
303{
304 return fib6_get_table(net, id);
305}
306
307struct fib6_table *fib6_get_table(struct net *net, u32 id)
308{
309 return net->ipv6.fib6_main_tbl;
310}
311
312struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
313 const struct sk_buff *skb,
314 int flags, pol_lookup_t lookup)
315{
316 struct rt6_info *rt;
317
318 rt = pol_lookup_func(lookup,
319 net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
320 if (rt->dst.error == -EAGAIN) {
321 ip6_rt_put_flags(rt, flags);
322 rt = net->ipv6.ip6_null_entry;
323 if (!(flags & RT6_LOOKUP_F_DST_NOREF))
324 dst_hold(&rt->dst);
325 }
326
327 return &rt->dst;
328}
329
330/* called with rcu lock held; no reference taken on fib6_info */
331int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
332 struct fib6_result *res, int flags)
333{
334 return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6,
335 res, flags);
336}
337
338static void __net_init fib6_tables_init(struct net *net)
339{
340 fib6_link_table(net, net->ipv6.fib6_main_tbl);
341}
342
343#endif
344
345unsigned int fib6_tables_seq_read(struct net *net)
346{
347 unsigned int h, fib_seq = 0;
348
349 rcu_read_lock();
350 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
351 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
352 struct fib6_table *tb;
353
354 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
355 fib_seq += tb->fib_seq;
356 }
357 rcu_read_unlock();
358
359 return fib_seq;
360}
361
362static int call_fib6_entry_notifier(struct notifier_block *nb,
363 enum fib_event_type event_type,
364 struct fib6_info *rt,
365 struct netlink_ext_ack *extack)
366{
367 struct fib6_entry_notifier_info info = {
368 .info.extack = extack,
369 .rt = rt,
370 };
371
372 return call_fib6_notifier(nb, event_type, &info.info);
373}
374
375static int call_fib6_multipath_entry_notifier(struct notifier_block *nb,
376 enum fib_event_type event_type,
377 struct fib6_info *rt,
378 unsigned int nsiblings,
379 struct netlink_ext_ack *extack)
380{
381 struct fib6_entry_notifier_info info = {
382 .info.extack = extack,
383 .rt = rt,
384 .nsiblings = nsiblings,
385 };
386
387 return call_fib6_notifier(nb, event_type, &info.info);
388}
389
390int call_fib6_entry_notifiers(struct net *net,
391 enum fib_event_type event_type,
392 struct fib6_info *rt,
393 struct netlink_ext_ack *extack)
394{
395 struct fib6_entry_notifier_info info = {
396 .info.extack = extack,
397 .rt = rt,
398 };
399
400 rt->fib6_table->fib_seq++;
401 return call_fib6_notifiers(net, event_type, &info.info);
402}
403
404int call_fib6_multipath_entry_notifiers(struct net *net,
405 enum fib_event_type event_type,
406 struct fib6_info *rt,
407 unsigned int nsiblings,
408 struct netlink_ext_ack *extack)
409{
410 struct fib6_entry_notifier_info info = {
411 .info.extack = extack,
412 .rt = rt,
413 .nsiblings = nsiblings,
414 };
415
416 rt->fib6_table->fib_seq++;
417 return call_fib6_notifiers(net, event_type, &info.info);
418}
419
420int call_fib6_entry_notifiers_replace(struct net *net, struct fib6_info *rt)
421{
422 struct fib6_entry_notifier_info info = {
423 .rt = rt,
424 .nsiblings = rt->fib6_nsiblings,
425 };
426
427 rt->fib6_table->fib_seq++;
428 return call_fib6_notifiers(net, FIB_EVENT_ENTRY_REPLACE, &info.info);
429}
430
431struct fib6_dump_arg {
432 struct net *net;
433 struct notifier_block *nb;
434 struct netlink_ext_ack *extack;
435};
436
437static int fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
438{
439 enum fib_event_type fib_event = FIB_EVENT_ENTRY_REPLACE;
440 int err;
441
442 if (!rt || rt == arg->net->ipv6.fib6_null_entry)
443 return 0;
444
445 if (rt->fib6_nsiblings)
446 err = call_fib6_multipath_entry_notifier(arg->nb, fib_event,
447 rt,
448 rt->fib6_nsiblings,
449 arg->extack);
450 else
451 err = call_fib6_entry_notifier(arg->nb, fib_event, rt,
452 arg->extack);
453
454 return err;
455}
456
457static int fib6_node_dump(struct fib6_walker *w)
458{
459 int err;
460
461 err = fib6_rt_dump(w->leaf, w->args);
462 w->leaf = NULL;
463 return err;
464}
465
466static int fib6_table_dump(struct net *net, struct fib6_table *tb,
467 struct fib6_walker *w)
468{
469 int err;
470
471 w->root = &tb->tb6_root;
472 spin_lock_bh(&tb->tb6_lock);
473 err = fib6_walk(net, w);
474 spin_unlock_bh(&tb->tb6_lock);
475 return err;
476}
477
478/* Called with rcu_read_lock() */
479int fib6_tables_dump(struct net *net, struct notifier_block *nb,
480 struct netlink_ext_ack *extack)
481{
482 struct fib6_dump_arg arg;
483 struct fib6_walker *w;
484 unsigned int h;
485 int err = 0;
486
487 w = kzalloc(sizeof(*w), GFP_ATOMIC);
488 if (!w)
489 return -ENOMEM;
490
491 w->func = fib6_node_dump;
492 arg.net = net;
493 arg.nb = nb;
494 arg.extack = extack;
495 w->args = &arg;
496
497 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
498 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
499 struct fib6_table *tb;
500
501 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
502 err = fib6_table_dump(net, tb, w);
503 if (err)
504 goto out;
505 }
506 }
507
508out:
509 kfree(w);
510
511 /* The tree traversal function should never return a positive value. */
512 return err > 0 ? -EINVAL : err;
513}
514
515static int fib6_dump_node(struct fib6_walker *w)
516{
517 int res;
518 struct fib6_info *rt;
519
520 for_each_fib6_walker_rt(w) {
521 res = rt6_dump_route(rt, w->args, w->skip_in_node);
522 if (res >= 0) {
523 /* Frame is full, suspend walking */
524 w->leaf = rt;
525
526 /* We'll restart from this node, so if some routes were
527 * already dumped, skip them next time.
528 */
529 w->skip_in_node += res;
530
531 return 1;
532 }
533 w->skip_in_node = 0;
534
535 /* Multipath routes are dumped in one route with the
536 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
537 * last sibling of this route (no need to dump the
538 * sibling routes again)
539 */
540 if (rt->fib6_nsiblings)
541 rt = list_last_entry(&rt->fib6_siblings,
542 struct fib6_info,
543 fib6_siblings);
544 }
545 w->leaf = NULL;
546 return 0;
547}
548
549static void fib6_dump_end(struct netlink_callback *cb)
550{
551 struct net *net = sock_net(cb->skb->sk);
552 struct fib6_walker *w = (void *)cb->args[2];
553
554 if (w) {
555 if (cb->args[4]) {
556 cb->args[4] = 0;
557 fib6_walker_unlink(net, w);
558 }
559 cb->args[2] = 0;
560 kfree(w);
561 }
562 cb->done = (void *)cb->args[3];
563 cb->args[1] = 3;
564}
565
566static int fib6_dump_done(struct netlink_callback *cb)
567{
568 fib6_dump_end(cb);
569 return cb->done ? cb->done(cb) : 0;
570}
571
572static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
573 struct netlink_callback *cb)
574{
575 struct net *net = sock_net(skb->sk);
576 struct fib6_walker *w;
577 int res;
578
579 w = (void *)cb->args[2];
580 w->root = &table->tb6_root;
581
582 if (cb->args[4] == 0) {
583 w->count = 0;
584 w->skip = 0;
585 w->skip_in_node = 0;
586
587 spin_lock_bh(&table->tb6_lock);
588 res = fib6_walk(net, w);
589 spin_unlock_bh(&table->tb6_lock);
590 if (res > 0) {
591 cb->args[4] = 1;
592 cb->args[5] = READ_ONCE(w->root->fn_sernum);
593 }
594 } else {
595 int sernum = READ_ONCE(w->root->fn_sernum);
596 if (cb->args[5] != sernum) {
597 /* Begin at the root if the tree changed */
598 cb->args[5] = sernum;
599 w->state = FWS_INIT;
600 w->node = w->root;
601 w->skip = w->count;
602 w->skip_in_node = 0;
603 } else
604 w->skip = 0;
605
606 spin_lock_bh(&table->tb6_lock);
607 res = fib6_walk_continue(w);
608 spin_unlock_bh(&table->tb6_lock);
609 if (res <= 0) {
610 fib6_walker_unlink(net, w);
611 cb->args[4] = 0;
612 }
613 }
614
615 return res;
616}
617
618static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
619{
620 struct rt6_rtnl_dump_arg arg = { .filter.dump_exceptions = true,
621 .filter.dump_routes = true };
622 const struct nlmsghdr *nlh = cb->nlh;
623 struct net *net = sock_net(skb->sk);
624 unsigned int h, s_h;
625 unsigned int e = 0, s_e;
626 struct fib6_walker *w;
627 struct fib6_table *tb;
628 struct hlist_head *head;
629 int res = 0;
630
631 if (cb->strict_check) {
632 int err;
633
634 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
635 if (err < 0)
636 return err;
637 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
638 struct rtmsg *rtm = nlmsg_data(nlh);
639
640 if (rtm->rtm_flags & RTM_F_PREFIX)
641 arg.filter.flags = RTM_F_PREFIX;
642 }
643
644 w = (void *)cb->args[2];
645 if (!w) {
646 /* New dump:
647 *
648 * 1. hook callback destructor.
649 */
650 cb->args[3] = (long)cb->done;
651 cb->done = fib6_dump_done;
652
653 /*
654 * 2. allocate and initialize walker.
655 */
656 w = kzalloc(sizeof(*w), GFP_ATOMIC);
657 if (!w)
658 return -ENOMEM;
659 w->func = fib6_dump_node;
660 cb->args[2] = (long)w;
661 }
662
663 arg.skb = skb;
664 arg.cb = cb;
665 arg.net = net;
666 w->args = &arg;
667
668 if (arg.filter.table_id) {
669 tb = fib6_get_table(net, arg.filter.table_id);
670 if (!tb) {
671 if (rtnl_msg_family(cb->nlh) != PF_INET6)
672 goto out;
673
674 NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
675 return -ENOENT;
676 }
677
678 if (!cb->args[0]) {
679 res = fib6_dump_table(tb, skb, cb);
680 if (!res)
681 cb->args[0] = 1;
682 }
683 goto out;
684 }
685
686 s_h = cb->args[0];
687 s_e = cb->args[1];
688
689 rcu_read_lock();
690 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
691 e = 0;
692 head = &net->ipv6.fib_table_hash[h];
693 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
694 if (e < s_e)
695 goto next;
696 res = fib6_dump_table(tb, skb, cb);
697 if (res != 0)
698 goto out_unlock;
699next:
700 e++;
701 }
702 }
703out_unlock:
704 rcu_read_unlock();
705 cb->args[1] = e;
706 cb->args[0] = h;
707out:
708 res = res < 0 ? res : skb->len;
709 if (res <= 0)
710 fib6_dump_end(cb);
711 return res;
712}
713
714void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
715{
716 if (!f6i)
717 return;
718
719 if (f6i->fib6_metrics == &dst_default_metrics) {
720 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
721
722 if (!p)
723 return;
724
725 refcount_set(&p->refcnt, 1);
726 f6i->fib6_metrics = p;
727 }
728
729 f6i->fib6_metrics->metrics[metric - 1] = val;
730}
731
732/*
733 * Routing Table
734 *
735 * return the appropriate node for a routing tree "add" operation
736 * by either creating and inserting or by returning an existing
737 * node.
738 */
739
740static struct fib6_node *fib6_add_1(struct net *net,
741 struct fib6_table *table,
742 struct fib6_node *root,
743 struct in6_addr *addr, int plen,
744 int offset, int allow_create,
745 int replace_required,
746 struct netlink_ext_ack *extack)
747{
748 struct fib6_node *fn, *in, *ln;
749 struct fib6_node *pn = NULL;
750 struct rt6key *key;
751 int bit;
752 __be32 dir = 0;
753
754 RT6_TRACE("fib6_add_1\n");
755
756 /* insert node in tree */
757
758 fn = root;
759
760 do {
761 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
762 lockdep_is_held(&table->tb6_lock));
763 key = (struct rt6key *)((u8 *)leaf + offset);
764
765 /*
766 * Prefix match
767 */
768 if (plen < fn->fn_bit ||
769 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
770 if (!allow_create) {
771 if (replace_required) {
772 NL_SET_ERR_MSG(extack,
773 "Can not replace route - no match found");
774 pr_warn("Can't replace route, no match found\n");
775 return ERR_PTR(-ENOENT);
776 }
777 pr_warn("NLM_F_CREATE should be set when creating new route\n");
778 }
779 goto insert_above;
780 }
781
782 /*
783 * Exact match ?
784 */
785
786 if (plen == fn->fn_bit) {
787 /* clean up an intermediate node */
788 if (!(fn->fn_flags & RTN_RTINFO)) {
789 RCU_INIT_POINTER(fn->leaf, NULL);
790 fib6_info_release(leaf);
791 /* remove null_entry in the root node */
792 } else if (fn->fn_flags & RTN_TL_ROOT &&
793 rcu_access_pointer(fn->leaf) ==
794 net->ipv6.fib6_null_entry) {
795 RCU_INIT_POINTER(fn->leaf, NULL);
796 }
797
798 return fn;
799 }
800
801 /*
802 * We have more bits to go
803 */
804
805 /* Try to walk down on tree. */
806 dir = addr_bit_set(addr, fn->fn_bit);
807 pn = fn;
808 fn = dir ?
809 rcu_dereference_protected(fn->right,
810 lockdep_is_held(&table->tb6_lock)) :
811 rcu_dereference_protected(fn->left,
812 lockdep_is_held(&table->tb6_lock));
813 } while (fn);
814
815 if (!allow_create) {
816 /* We should not create new node because
817 * NLM_F_REPLACE was specified without NLM_F_CREATE
818 * I assume it is safe to require NLM_F_CREATE when
819 * REPLACE flag is used! Later we may want to remove the
820 * check for replace_required, because according
821 * to netlink specification, NLM_F_CREATE
822 * MUST be specified if new route is created.
823 * That would keep IPv6 consistent with IPv4
824 */
825 if (replace_required) {
826 NL_SET_ERR_MSG(extack,
827 "Can not replace route - no match found");
828 pr_warn("Can't replace route, no match found\n");
829 return ERR_PTR(-ENOENT);
830 }
831 pr_warn("NLM_F_CREATE should be set when creating new route\n");
832 }
833 /*
834 * We walked to the bottom of tree.
835 * Create new leaf node without children.
836 */
837
838 ln = node_alloc(net);
839
840 if (!ln)
841 return ERR_PTR(-ENOMEM);
842 ln->fn_bit = plen;
843 RCU_INIT_POINTER(ln->parent, pn);
844
845 if (dir)
846 rcu_assign_pointer(pn->right, ln);
847 else
848 rcu_assign_pointer(pn->left, ln);
849
850 return ln;
851
852
853insert_above:
854 /*
855 * split since we don't have a common prefix anymore or
856 * we have a less significant route.
857 * we've to insert an intermediate node on the list
858 * this new node will point to the one we need to create
859 * and the current
860 */
861
862 pn = rcu_dereference_protected(fn->parent,
863 lockdep_is_held(&table->tb6_lock));
864
865 /* find 1st bit in difference between the 2 addrs.
866
867 See comment in __ipv6_addr_diff: bit may be an invalid value,
868 but if it is >= plen, the value is ignored in any case.
869 */
870
871 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
872
873 /*
874 * (intermediate)[in]
875 * / \
876 * (new leaf node)[ln] (old node)[fn]
877 */
878 if (plen > bit) {
879 in = node_alloc(net);
880 ln = node_alloc(net);
881
882 if (!in || !ln) {
883 if (in)
884 node_free_immediate(net, in);
885 if (ln)
886 node_free_immediate(net, ln);
887 return ERR_PTR(-ENOMEM);
888 }
889
890 /*
891 * new intermediate node.
892 * RTN_RTINFO will
893 * be off since that an address that chooses one of
894 * the branches would not match less specific routes
895 * in the other branch
896 */
897
898 in->fn_bit = bit;
899
900 RCU_INIT_POINTER(in->parent, pn);
901 in->leaf = fn->leaf;
902 fib6_info_hold(rcu_dereference_protected(in->leaf,
903 lockdep_is_held(&table->tb6_lock)));
904
905 /* update parent pointer */
906 if (dir)
907 rcu_assign_pointer(pn->right, in);
908 else
909 rcu_assign_pointer(pn->left, in);
910
911 ln->fn_bit = plen;
912
913 RCU_INIT_POINTER(ln->parent, in);
914 rcu_assign_pointer(fn->parent, in);
915
916 if (addr_bit_set(addr, bit)) {
917 rcu_assign_pointer(in->right, ln);
918 rcu_assign_pointer(in->left, fn);
919 } else {
920 rcu_assign_pointer(in->left, ln);
921 rcu_assign_pointer(in->right, fn);
922 }
923 } else { /* plen <= bit */
924
925 /*
926 * (new leaf node)[ln]
927 * / \
928 * (old node)[fn] NULL
929 */
930
931 ln = node_alloc(net);
932
933 if (!ln)
934 return ERR_PTR(-ENOMEM);
935
936 ln->fn_bit = plen;
937
938 RCU_INIT_POINTER(ln->parent, pn);
939
940 if (addr_bit_set(&key->addr, plen))
941 RCU_INIT_POINTER(ln->right, fn);
942 else
943 RCU_INIT_POINTER(ln->left, fn);
944
945 rcu_assign_pointer(fn->parent, ln);
946
947 if (dir)
948 rcu_assign_pointer(pn->right, ln);
949 else
950 rcu_assign_pointer(pn->left, ln);
951 }
952 return ln;
953}
954
955static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
956 const struct fib6_info *match,
957 const struct fib6_table *table)
958{
959 int cpu;
960
961 if (!fib6_nh->rt6i_pcpu)
962 return;
963
964 /* release the reference to this fib entry from
965 * all of its cached pcpu routes
966 */
967 for_each_possible_cpu(cpu) {
968 struct rt6_info **ppcpu_rt;
969 struct rt6_info *pcpu_rt;
970
971 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
972 pcpu_rt = *ppcpu_rt;
973
974 /* only dropping the 'from' reference if the cached route
975 * is using 'match'. The cached pcpu_rt->from only changes
976 * from a fib6_info to NULL (ip6_dst_destroy); it can never
977 * change from one fib6_info reference to another
978 */
979 if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
980 struct fib6_info *from;
981
982 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
983 fib6_info_release(from);
984 }
985 }
986}
987
988struct fib6_nh_pcpu_arg {
989 struct fib6_info *from;
990 const struct fib6_table *table;
991};
992
993static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg)
994{
995 struct fib6_nh_pcpu_arg *arg = _arg;
996
997 __fib6_drop_pcpu_from(nh, arg->from, arg->table);
998 return 0;
999}
1000
1001static void fib6_drop_pcpu_from(struct fib6_info *f6i,
1002 const struct fib6_table *table)
1003{
1004 /* Make sure rt6_make_pcpu_route() wont add other percpu routes
1005 * while we are cleaning them here.
1006 */
1007 f6i->fib6_destroying = 1;
1008 mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
1009
1010 if (f6i->nh) {
1011 struct fib6_nh_pcpu_arg arg = {
1012 .from = f6i,
1013 .table = table
1014 };
1015
1016 nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from,
1017 &arg);
1018 } else {
1019 struct fib6_nh *fib6_nh;
1020
1021 fib6_nh = f6i->fib6_nh;
1022 __fib6_drop_pcpu_from(fib6_nh, f6i, table);
1023 }
1024}
1025
1026static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
1027 struct net *net)
1028{
1029 struct fib6_table *table = rt->fib6_table;
1030
1031 /* Flush all cached dst in exception table */
1032 rt6_flush_exceptions(rt);
1033 fib6_drop_pcpu_from(rt, table);
1034
1035 if (rt->nh && !list_empty(&rt->nh_list))
1036 list_del_init(&rt->nh_list);
1037
1038 if (refcount_read(&rt->fib6_ref) != 1) {
1039 /* This route is used as dummy address holder in some split
1040 * nodes. It is not leaked, but it still holds other resources,
1041 * which must be released in time. So, scan ascendant nodes
1042 * and replace dummy references to this route with references
1043 * to still alive ones.
1044 */
1045 while (fn) {
1046 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1047 lockdep_is_held(&table->tb6_lock));
1048 struct fib6_info *new_leaf;
1049 if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
1050 new_leaf = fib6_find_prefix(net, table, fn);
1051 fib6_info_hold(new_leaf);
1052
1053 rcu_assign_pointer(fn->leaf, new_leaf);
1054 fib6_info_release(rt);
1055 }
1056 fn = rcu_dereference_protected(fn->parent,
1057 lockdep_is_held(&table->tb6_lock));
1058 }
1059 }
1060}
1061
1062/*
1063 * Insert routing information in a node.
1064 */
1065
1066static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
1067 struct nl_info *info,
1068 struct netlink_ext_ack *extack)
1069{
1070 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1071 lockdep_is_held(&rt->fib6_table->tb6_lock));
1072 struct fib6_info *iter = NULL;
1073 struct fib6_info __rcu **ins;
1074 struct fib6_info __rcu **fallback_ins = NULL;
1075 int replace = (info->nlh &&
1076 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
1077 int add = (!info->nlh ||
1078 (info->nlh->nlmsg_flags & NLM_F_CREATE));
1079 int found = 0;
1080 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
1081 bool notify_sibling_rt = false;
1082 u16 nlflags = NLM_F_EXCL;
1083 int err;
1084
1085 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
1086 nlflags |= NLM_F_APPEND;
1087
1088 ins = &fn->leaf;
1089
1090 for (iter = leaf; iter;
1091 iter = rcu_dereference_protected(iter->fib6_next,
1092 lockdep_is_held(&rt->fib6_table->tb6_lock))) {
1093 /*
1094 * Search for duplicates
1095 */
1096
1097 if (iter->fib6_metric == rt->fib6_metric) {
1098 /*
1099 * Same priority level
1100 */
1101 if (info->nlh &&
1102 (info->nlh->nlmsg_flags & NLM_F_EXCL))
1103 return -EEXIST;
1104
1105 nlflags &= ~NLM_F_EXCL;
1106 if (replace) {
1107 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1108 found++;
1109 break;
1110 }
1111 fallback_ins = fallback_ins ?: ins;
1112 goto next_iter;
1113 }
1114
1115 if (rt6_duplicate_nexthop(iter, rt)) {
1116 if (rt->fib6_nsiblings)
1117 rt->fib6_nsiblings = 0;
1118 if (!(iter->fib6_flags & RTF_EXPIRES))
1119 return -EEXIST;
1120 if (!(rt->fib6_flags & RTF_EXPIRES))
1121 fib6_clean_expires(iter);
1122 else
1123 fib6_set_expires(iter, rt->expires);
1124
1125 if (rt->fib6_pmtu)
1126 fib6_metric_set(iter, RTAX_MTU,
1127 rt->fib6_pmtu);
1128 return -EEXIST;
1129 }
1130 /* If we have the same destination and the same metric,
1131 * but not the same gateway, then the route we try to
1132 * add is sibling to this route, increment our counter
1133 * of siblings, and later we will add our route to the
1134 * list.
1135 * Only static routes (which don't have flag
1136 * RTF_EXPIRES) are used for ECMPv6.
1137 *
1138 * To avoid long list, we only had siblings if the
1139 * route have a gateway.
1140 */
1141 if (rt_can_ecmp &&
1142 rt6_qualify_for_ecmp(iter))
1143 rt->fib6_nsiblings++;
1144 }
1145
1146 if (iter->fib6_metric > rt->fib6_metric)
1147 break;
1148
1149next_iter:
1150 ins = &iter->fib6_next;
1151 }
1152
1153 if (fallback_ins && !found) {
1154 /* No matching route with same ecmp-able-ness found, replace
1155 * first matching route
1156 */
1157 ins = fallback_ins;
1158 iter = rcu_dereference_protected(*ins,
1159 lockdep_is_held(&rt->fib6_table->tb6_lock));
1160 found++;
1161 }
1162
1163 /* Reset round-robin state, if necessary */
1164 if (ins == &fn->leaf)
1165 fn->rr_ptr = NULL;
1166
1167 /* Link this route to others same route. */
1168 if (rt->fib6_nsiblings) {
1169 unsigned int fib6_nsiblings;
1170 struct fib6_info *sibling, *temp_sibling;
1171
1172 /* Find the first route that have the same metric */
1173 sibling = leaf;
1174 notify_sibling_rt = true;
1175 while (sibling) {
1176 if (sibling->fib6_metric == rt->fib6_metric &&
1177 rt6_qualify_for_ecmp(sibling)) {
1178 list_add_tail(&rt->fib6_siblings,
1179 &sibling->fib6_siblings);
1180 break;
1181 }
1182 sibling = rcu_dereference_protected(sibling->fib6_next,
1183 lockdep_is_held(&rt->fib6_table->tb6_lock));
1184 notify_sibling_rt = false;
1185 }
1186 /* For each sibling in the list, increment the counter of
1187 * siblings. BUG() if counters does not match, list of siblings
1188 * is broken!
1189 */
1190 fib6_nsiblings = 0;
1191 list_for_each_entry_safe(sibling, temp_sibling,
1192 &rt->fib6_siblings, fib6_siblings) {
1193 sibling->fib6_nsiblings++;
1194 BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1195 fib6_nsiblings++;
1196 }
1197 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1198 rt6_multipath_rebalance(temp_sibling);
1199 }
1200
1201 /*
1202 * insert node
1203 */
1204 if (!replace) {
1205 if (!add)
1206 pr_warn("NLM_F_CREATE should be set when creating new route\n");
1207
1208add:
1209 nlflags |= NLM_F_CREATE;
1210
1211 /* The route should only be notified if it is the first
1212 * route in the node or if it is added as a sibling
1213 * route to the first route in the node.
1214 */
1215 if (!info->skip_notify_kernel &&
1216 (notify_sibling_rt || ins == &fn->leaf)) {
1217 enum fib_event_type fib_event;
1218
1219 if (notify_sibling_rt)
1220 fib_event = FIB_EVENT_ENTRY_APPEND;
1221 else
1222 fib_event = FIB_EVENT_ENTRY_REPLACE;
1223 err = call_fib6_entry_notifiers(info->nl_net,
1224 fib_event, rt,
1225 extack);
1226 if (err) {
1227 struct fib6_info *sibling, *next_sibling;
1228
1229 /* If the route has siblings, then it first
1230 * needs to be unlinked from them.
1231 */
1232 if (!rt->fib6_nsiblings)
1233 return err;
1234
1235 list_for_each_entry_safe(sibling, next_sibling,
1236 &rt->fib6_siblings,
1237 fib6_siblings)
1238 sibling->fib6_nsiblings--;
1239 rt->fib6_nsiblings = 0;
1240 list_del_init(&rt->fib6_siblings);
1241 rt6_multipath_rebalance(next_sibling);
1242 return err;
1243 }
1244 }
1245
1246 rcu_assign_pointer(rt->fib6_next, iter);
1247 fib6_info_hold(rt);
1248 rcu_assign_pointer(rt->fib6_node, fn);
1249 rcu_assign_pointer(*ins, rt);
1250 if (!info->skip_notify)
1251 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1252 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1253
1254 if (!(fn->fn_flags & RTN_RTINFO)) {
1255 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1256 fn->fn_flags |= RTN_RTINFO;
1257 }
1258
1259 } else {
1260 int nsiblings;
1261
1262 if (!found) {
1263 if (add)
1264 goto add;
1265 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1266 return -ENOENT;
1267 }
1268
1269 if (!info->skip_notify_kernel && ins == &fn->leaf) {
1270 err = call_fib6_entry_notifiers(info->nl_net,
1271 FIB_EVENT_ENTRY_REPLACE,
1272 rt, extack);
1273 if (err)
1274 return err;
1275 }
1276
1277 fib6_info_hold(rt);
1278 rcu_assign_pointer(rt->fib6_node, fn);
1279 rt->fib6_next = iter->fib6_next;
1280 rcu_assign_pointer(*ins, rt);
1281 if (!info->skip_notify)
1282 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1283 if (!(fn->fn_flags & RTN_RTINFO)) {
1284 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1285 fn->fn_flags |= RTN_RTINFO;
1286 }
1287 nsiblings = iter->fib6_nsiblings;
1288 iter->fib6_node = NULL;
1289 fib6_purge_rt(iter, fn, info->nl_net);
1290 if (rcu_access_pointer(fn->rr_ptr) == iter)
1291 fn->rr_ptr = NULL;
1292 fib6_info_release(iter);
1293
1294 if (nsiblings) {
1295 /* Replacing an ECMP route, remove all siblings */
1296 ins = &rt->fib6_next;
1297 iter = rcu_dereference_protected(*ins,
1298 lockdep_is_held(&rt->fib6_table->tb6_lock));
1299 while (iter) {
1300 if (iter->fib6_metric > rt->fib6_metric)
1301 break;
1302 if (rt6_qualify_for_ecmp(iter)) {
1303 *ins = iter->fib6_next;
1304 iter->fib6_node = NULL;
1305 fib6_purge_rt(iter, fn, info->nl_net);
1306 if (rcu_access_pointer(fn->rr_ptr) == iter)
1307 fn->rr_ptr = NULL;
1308 fib6_info_release(iter);
1309 nsiblings--;
1310 info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1311 } else {
1312 ins = &iter->fib6_next;
1313 }
1314 iter = rcu_dereference_protected(*ins,
1315 lockdep_is_held(&rt->fib6_table->tb6_lock));
1316 }
1317 WARN_ON(nsiblings != 0);
1318 }
1319 }
1320
1321 return 0;
1322}
1323
1324static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1325{
1326 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1327 (rt->fib6_flags & RTF_EXPIRES))
1328 mod_timer(&net->ipv6.ip6_fib_timer,
1329 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1330}
1331
1332void fib6_force_start_gc(struct net *net)
1333{
1334 if (!timer_pending(&net->ipv6.ip6_fib_timer))
1335 mod_timer(&net->ipv6.ip6_fib_timer,
1336 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1337}
1338
1339static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1340 int sernum)
1341{
1342 struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1343 lockdep_is_held(&rt->fib6_table->tb6_lock));
1344
1345 /* paired with smp_rmb() in fib6_get_cookie_safe() */
1346 smp_wmb();
1347 while (fn) {
1348 WRITE_ONCE(fn->fn_sernum, sernum);
1349 fn = rcu_dereference_protected(fn->parent,
1350 lockdep_is_held(&rt->fib6_table->tb6_lock));
1351 }
1352}
1353
1354void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1355{
1356 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1357}
1358
1359/* allow ipv4 to update sernum via ipv6_stub */
1360void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1361{
1362 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1363 fib6_update_sernum_upto_root(net, f6i);
1364 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1365}
1366
1367/*
1368 * Add routing information to the routing tree.
1369 * <destination addr>/<source addr>
1370 * with source addr info in sub-trees
1371 * Need to own table->tb6_lock
1372 */
1373
1374int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1375 struct nl_info *info, struct netlink_ext_ack *extack)
1376{
1377 struct fib6_table *table = rt->fib6_table;
1378 struct fib6_node *fn, *pn = NULL;
1379 int err = -ENOMEM;
1380 int allow_create = 1;
1381 int replace_required = 0;
1382
1383 if (info->nlh) {
1384 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1385 allow_create = 0;
1386 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1387 replace_required = 1;
1388 }
1389 if (!allow_create && !replace_required)
1390 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1391
1392 fn = fib6_add_1(info->nl_net, table, root,
1393 &rt->fib6_dst.addr, rt->fib6_dst.plen,
1394 offsetof(struct fib6_info, fib6_dst), allow_create,
1395 replace_required, extack);
1396 if (IS_ERR(fn)) {
1397 err = PTR_ERR(fn);
1398 fn = NULL;
1399 goto out;
1400 }
1401
1402 pn = fn;
1403
1404#ifdef CONFIG_IPV6_SUBTREES
1405 if (rt->fib6_src.plen) {
1406 struct fib6_node *sn;
1407
1408 if (!rcu_access_pointer(fn->subtree)) {
1409 struct fib6_node *sfn;
1410
1411 /*
1412 * Create subtree.
1413 *
1414 * fn[main tree]
1415 * |
1416 * sfn[subtree root]
1417 * \
1418 * sn[new leaf node]
1419 */
1420
1421 /* Create subtree root node */
1422 sfn = node_alloc(info->nl_net);
1423 if (!sfn)
1424 goto failure;
1425
1426 fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1427 rcu_assign_pointer(sfn->leaf,
1428 info->nl_net->ipv6.fib6_null_entry);
1429 sfn->fn_flags = RTN_ROOT;
1430
1431 /* Now add the first leaf node to new subtree */
1432
1433 sn = fib6_add_1(info->nl_net, table, sfn,
1434 &rt->fib6_src.addr, rt->fib6_src.plen,
1435 offsetof(struct fib6_info, fib6_src),
1436 allow_create, replace_required, extack);
1437
1438 if (IS_ERR(sn)) {
1439 /* If it is failed, discard just allocated
1440 root, and then (in failure) stale node
1441 in main tree.
1442 */
1443 node_free_immediate(info->nl_net, sfn);
1444 err = PTR_ERR(sn);
1445 goto failure;
1446 }
1447
1448 /* Now link new subtree to main tree */
1449 rcu_assign_pointer(sfn->parent, fn);
1450 rcu_assign_pointer(fn->subtree, sfn);
1451 } else {
1452 sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1453 &rt->fib6_src.addr, rt->fib6_src.plen,
1454 offsetof(struct fib6_info, fib6_src),
1455 allow_create, replace_required, extack);
1456
1457 if (IS_ERR(sn)) {
1458 err = PTR_ERR(sn);
1459 goto failure;
1460 }
1461 }
1462
1463 if (!rcu_access_pointer(fn->leaf)) {
1464 if (fn->fn_flags & RTN_TL_ROOT) {
1465 /* put back null_entry for root node */
1466 rcu_assign_pointer(fn->leaf,
1467 info->nl_net->ipv6.fib6_null_entry);
1468 } else {
1469 fib6_info_hold(rt);
1470 rcu_assign_pointer(fn->leaf, rt);
1471 }
1472 }
1473 fn = sn;
1474 }
1475#endif
1476
1477 err = fib6_add_rt2node(fn, rt, info, extack);
1478 if (!err) {
1479 if (rt->nh)
1480 list_add(&rt->nh_list, &rt->nh->f6i_list);
1481 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(info->nl_net));
1482 fib6_start_gc(info->nl_net, rt);
1483 }
1484
1485out:
1486 if (err) {
1487#ifdef CONFIG_IPV6_SUBTREES
1488 /*
1489 * If fib6_add_1 has cleared the old leaf pointer in the
1490 * super-tree leaf node we have to find a new one for it.
1491 */
1492 if (pn != fn) {
1493 struct fib6_info *pn_leaf =
1494 rcu_dereference_protected(pn->leaf,
1495 lockdep_is_held(&table->tb6_lock));
1496 if (pn_leaf == rt) {
1497 pn_leaf = NULL;
1498 RCU_INIT_POINTER(pn->leaf, NULL);
1499 fib6_info_release(rt);
1500 }
1501 if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1502 pn_leaf = fib6_find_prefix(info->nl_net, table,
1503 pn);
1504#if RT6_DEBUG >= 2
1505 if (!pn_leaf) {
1506 WARN_ON(!pn_leaf);
1507 pn_leaf =
1508 info->nl_net->ipv6.fib6_null_entry;
1509 }
1510#endif
1511 fib6_info_hold(pn_leaf);
1512 rcu_assign_pointer(pn->leaf, pn_leaf);
1513 }
1514 }
1515#endif
1516 goto failure;
1517 } else if (fib6_requires_src(rt)) {
1518 fib6_routes_require_src_inc(info->nl_net);
1519 }
1520 return err;
1521
1522failure:
1523 /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1524 * 1. fn is an intermediate node and we failed to add the new
1525 * route to it in both subtree creation failure and fib6_add_rt2node()
1526 * failure case.
1527 * 2. fn is the root node in the table and we fail to add the first
1528 * default route to it.
1529 */
1530 if (fn &&
1531 (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1532 (fn->fn_flags & RTN_TL_ROOT &&
1533 !rcu_access_pointer(fn->leaf))))
1534 fib6_repair_tree(info->nl_net, table, fn);
1535 return err;
1536}
1537
1538/*
1539 * Routing tree lookup
1540 *
1541 */
1542
1543struct lookup_args {
1544 int offset; /* key offset on fib6_info */
1545 const struct in6_addr *addr; /* search key */
1546};
1547
1548static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1549 struct lookup_args *args)
1550{
1551 struct fib6_node *fn;
1552 __be32 dir;
1553
1554 if (unlikely(args->offset == 0))
1555 return NULL;
1556
1557 /*
1558 * Descend on a tree
1559 */
1560
1561 fn = root;
1562
1563 for (;;) {
1564 struct fib6_node *next;
1565
1566 dir = addr_bit_set(args->addr, fn->fn_bit);
1567
1568 next = dir ? rcu_dereference(fn->right) :
1569 rcu_dereference(fn->left);
1570
1571 if (next) {
1572 fn = next;
1573 continue;
1574 }
1575 break;
1576 }
1577
1578 while (fn) {
1579 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1580
1581 if (subtree || fn->fn_flags & RTN_RTINFO) {
1582 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1583 struct rt6key *key;
1584
1585 if (!leaf)
1586 goto backtrack;
1587
1588 key = (struct rt6key *) ((u8 *)leaf + args->offset);
1589
1590 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1591#ifdef CONFIG_IPV6_SUBTREES
1592 if (subtree) {
1593 struct fib6_node *sfn;
1594 sfn = fib6_node_lookup_1(subtree,
1595 args + 1);
1596 if (!sfn)
1597 goto backtrack;
1598 fn = sfn;
1599 }
1600#endif
1601 if (fn->fn_flags & RTN_RTINFO)
1602 return fn;
1603 }
1604 }
1605backtrack:
1606 if (fn->fn_flags & RTN_ROOT)
1607 break;
1608
1609 fn = rcu_dereference(fn->parent);
1610 }
1611
1612 return NULL;
1613}
1614
1615/* called with rcu_read_lock() held
1616 */
1617struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1618 const struct in6_addr *daddr,
1619 const struct in6_addr *saddr)
1620{
1621 struct fib6_node *fn;
1622 struct lookup_args args[] = {
1623 {
1624 .offset = offsetof(struct fib6_info, fib6_dst),
1625 .addr = daddr,
1626 },
1627#ifdef CONFIG_IPV6_SUBTREES
1628 {
1629 .offset = offsetof(struct fib6_info, fib6_src),
1630 .addr = saddr,
1631 },
1632#endif
1633 {
1634 .offset = 0, /* sentinel */
1635 }
1636 };
1637
1638 fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1639 if (!fn || fn->fn_flags & RTN_TL_ROOT)
1640 fn = root;
1641
1642 return fn;
1643}
1644
1645/*
1646 * Get node with specified destination prefix (and source prefix,
1647 * if subtrees are used)
1648 * exact_match == true means we try to find fn with exact match of
1649 * the passed in prefix addr
1650 * exact_match == false means we try to find fn with longest prefix
1651 * match of the passed in prefix addr. This is useful for finding fn
1652 * for cached route as it will be stored in the exception table under
1653 * the node with longest prefix length.
1654 */
1655
1656
1657static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1658 const struct in6_addr *addr,
1659 int plen, int offset,
1660 bool exact_match)
1661{
1662 struct fib6_node *fn, *prev = NULL;
1663
1664 for (fn = root; fn ; ) {
1665 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1666 struct rt6key *key;
1667
1668 /* This node is being deleted */
1669 if (!leaf) {
1670 if (plen <= fn->fn_bit)
1671 goto out;
1672 else
1673 goto next;
1674 }
1675
1676 key = (struct rt6key *)((u8 *)leaf + offset);
1677
1678 /*
1679 * Prefix match
1680 */
1681 if (plen < fn->fn_bit ||
1682 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1683 goto out;
1684
1685 if (plen == fn->fn_bit)
1686 return fn;
1687
1688 if (fn->fn_flags & RTN_RTINFO)
1689 prev = fn;
1690
1691next:
1692 /*
1693 * We have more bits to go
1694 */
1695 if (addr_bit_set(addr, fn->fn_bit))
1696 fn = rcu_dereference(fn->right);
1697 else
1698 fn = rcu_dereference(fn->left);
1699 }
1700out:
1701 if (exact_match)
1702 return NULL;
1703 else
1704 return prev;
1705}
1706
1707struct fib6_node *fib6_locate(struct fib6_node *root,
1708 const struct in6_addr *daddr, int dst_len,
1709 const struct in6_addr *saddr, int src_len,
1710 bool exact_match)
1711{
1712 struct fib6_node *fn;
1713
1714 fn = fib6_locate_1(root, daddr, dst_len,
1715 offsetof(struct fib6_info, fib6_dst),
1716 exact_match);
1717
1718#ifdef CONFIG_IPV6_SUBTREES
1719 if (src_len) {
1720 WARN_ON(saddr == NULL);
1721 if (fn) {
1722 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1723
1724 if (subtree) {
1725 fn = fib6_locate_1(subtree, saddr, src_len,
1726 offsetof(struct fib6_info, fib6_src),
1727 exact_match);
1728 }
1729 }
1730 }
1731#endif
1732
1733 if (fn && fn->fn_flags & RTN_RTINFO)
1734 return fn;
1735
1736 return NULL;
1737}
1738
1739
1740/*
1741 * Deletion
1742 *
1743 */
1744
1745static struct fib6_info *fib6_find_prefix(struct net *net,
1746 struct fib6_table *table,
1747 struct fib6_node *fn)
1748{
1749 struct fib6_node *child_left, *child_right;
1750
1751 if (fn->fn_flags & RTN_ROOT)
1752 return net->ipv6.fib6_null_entry;
1753
1754 while (fn) {
1755 child_left = rcu_dereference_protected(fn->left,
1756 lockdep_is_held(&table->tb6_lock));
1757 child_right = rcu_dereference_protected(fn->right,
1758 lockdep_is_held(&table->tb6_lock));
1759 if (child_left)
1760 return rcu_dereference_protected(child_left->leaf,
1761 lockdep_is_held(&table->tb6_lock));
1762 if (child_right)
1763 return rcu_dereference_protected(child_right->leaf,
1764 lockdep_is_held(&table->tb6_lock));
1765
1766 fn = FIB6_SUBTREE(fn);
1767 }
1768 return NULL;
1769}
1770
1771/*
1772 * Called to trim the tree of intermediate nodes when possible. "fn"
1773 * is the node we want to try and remove.
1774 * Need to own table->tb6_lock
1775 */
1776
1777static struct fib6_node *fib6_repair_tree(struct net *net,
1778 struct fib6_table *table,
1779 struct fib6_node *fn)
1780{
1781 int children;
1782 int nstate;
1783 struct fib6_node *child;
1784 struct fib6_walker *w;
1785 int iter = 0;
1786
1787 /* Set fn->leaf to null_entry for root node. */
1788 if (fn->fn_flags & RTN_TL_ROOT) {
1789 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1790 return fn;
1791 }
1792
1793 for (;;) {
1794 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1795 lockdep_is_held(&table->tb6_lock));
1796 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1797 lockdep_is_held(&table->tb6_lock));
1798 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1799 lockdep_is_held(&table->tb6_lock));
1800 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1801 lockdep_is_held(&table->tb6_lock));
1802 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1803 lockdep_is_held(&table->tb6_lock));
1804 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1805 lockdep_is_held(&table->tb6_lock));
1806 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1807 lockdep_is_held(&table->tb6_lock));
1808 struct fib6_info *new_fn_leaf;
1809
1810 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1811 iter++;
1812
1813 WARN_ON(fn->fn_flags & RTN_RTINFO);
1814 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1815 WARN_ON(fn_leaf);
1816
1817 children = 0;
1818 child = NULL;
1819 if (fn_r) {
1820 child = fn_r;
1821 children |= 1;
1822 }
1823 if (fn_l) {
1824 child = fn_l;
1825 children |= 2;
1826 }
1827
1828 if (children == 3 || FIB6_SUBTREE(fn)
1829#ifdef CONFIG_IPV6_SUBTREES
1830 /* Subtree root (i.e. fn) may have one child */
1831 || (children && fn->fn_flags & RTN_ROOT)
1832#endif
1833 ) {
1834 new_fn_leaf = fib6_find_prefix(net, table, fn);
1835#if RT6_DEBUG >= 2
1836 if (!new_fn_leaf) {
1837 WARN_ON(!new_fn_leaf);
1838 new_fn_leaf = net->ipv6.fib6_null_entry;
1839 }
1840#endif
1841 fib6_info_hold(new_fn_leaf);
1842 rcu_assign_pointer(fn->leaf, new_fn_leaf);
1843 return pn;
1844 }
1845
1846#ifdef CONFIG_IPV6_SUBTREES
1847 if (FIB6_SUBTREE(pn) == fn) {
1848 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1849 RCU_INIT_POINTER(pn->subtree, NULL);
1850 nstate = FWS_L;
1851 } else {
1852 WARN_ON(fn->fn_flags & RTN_ROOT);
1853#endif
1854 if (pn_r == fn)
1855 rcu_assign_pointer(pn->right, child);
1856 else if (pn_l == fn)
1857 rcu_assign_pointer(pn->left, child);
1858#if RT6_DEBUG >= 2
1859 else
1860 WARN_ON(1);
1861#endif
1862 if (child)
1863 rcu_assign_pointer(child->parent, pn);
1864 nstate = FWS_R;
1865#ifdef CONFIG_IPV6_SUBTREES
1866 }
1867#endif
1868
1869 read_lock(&net->ipv6.fib6_walker_lock);
1870 FOR_WALKERS(net, w) {
1871 if (!child) {
1872 if (w->node == fn) {
1873 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1874 w->node = pn;
1875 w->state = nstate;
1876 }
1877 } else {
1878 if (w->node == fn) {
1879 w->node = child;
1880 if (children&2) {
1881 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1882 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1883 } else {
1884 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1885 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1886 }
1887 }
1888 }
1889 }
1890 read_unlock(&net->ipv6.fib6_walker_lock);
1891
1892 node_free(net, fn);
1893 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1894 return pn;
1895
1896 RCU_INIT_POINTER(pn->leaf, NULL);
1897 fib6_info_release(pn_leaf);
1898 fn = pn;
1899 }
1900}
1901
1902static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1903 struct fib6_info __rcu **rtp, struct nl_info *info)
1904{
1905 struct fib6_info *leaf, *replace_rt = NULL;
1906 struct fib6_walker *w;
1907 struct fib6_info *rt = rcu_dereference_protected(*rtp,
1908 lockdep_is_held(&table->tb6_lock));
1909 struct net *net = info->nl_net;
1910 bool notify_del = false;
1911
1912 RT6_TRACE("fib6_del_route\n");
1913
1914 /* If the deleted route is the first in the node and it is not part of
1915 * a multipath route, then we need to replace it with the next route
1916 * in the node, if exists.
1917 */
1918 leaf = rcu_dereference_protected(fn->leaf,
1919 lockdep_is_held(&table->tb6_lock));
1920 if (leaf == rt && !rt->fib6_nsiblings) {
1921 if (rcu_access_pointer(rt->fib6_next))
1922 replace_rt = rcu_dereference_protected(rt->fib6_next,
1923 lockdep_is_held(&table->tb6_lock));
1924 else
1925 notify_del = true;
1926 }
1927
1928 /* Unlink it */
1929 *rtp = rt->fib6_next;
1930 rt->fib6_node = NULL;
1931 net->ipv6.rt6_stats->fib_rt_entries--;
1932 net->ipv6.rt6_stats->fib_discarded_routes++;
1933
1934 /* Reset round-robin state, if necessary */
1935 if (rcu_access_pointer(fn->rr_ptr) == rt)
1936 fn->rr_ptr = NULL;
1937
1938 /* Remove this entry from other siblings */
1939 if (rt->fib6_nsiblings) {
1940 struct fib6_info *sibling, *next_sibling;
1941
1942 /* The route is deleted from a multipath route. If this
1943 * multipath route is the first route in the node, then we need
1944 * to emit a delete notification. Otherwise, we need to skip
1945 * the notification.
1946 */
1947 if (rt->fib6_metric == leaf->fib6_metric &&
1948 rt6_qualify_for_ecmp(leaf))
1949 notify_del = true;
1950 list_for_each_entry_safe(sibling, next_sibling,
1951 &rt->fib6_siblings, fib6_siblings)
1952 sibling->fib6_nsiblings--;
1953 rt->fib6_nsiblings = 0;
1954 list_del_init(&rt->fib6_siblings);
1955 rt6_multipath_rebalance(next_sibling);
1956 }
1957
1958 /* Adjust walkers */
1959 read_lock(&net->ipv6.fib6_walker_lock);
1960 FOR_WALKERS(net, w) {
1961 if (w->state == FWS_C && w->leaf == rt) {
1962 RT6_TRACE("walker %p adjusted by delroute\n", w);
1963 w->leaf = rcu_dereference_protected(rt->fib6_next,
1964 lockdep_is_held(&table->tb6_lock));
1965 if (!w->leaf)
1966 w->state = FWS_U;
1967 }
1968 }
1969 read_unlock(&net->ipv6.fib6_walker_lock);
1970
1971 /* If it was last route, call fib6_repair_tree() to:
1972 * 1. For root node, put back null_entry as how the table was created.
1973 * 2. For other nodes, expunge its radix tree node.
1974 */
1975 if (!rcu_access_pointer(fn->leaf)) {
1976 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1977 fn->fn_flags &= ~RTN_RTINFO;
1978 net->ipv6.rt6_stats->fib_route_nodes--;
1979 }
1980 fn = fib6_repair_tree(net, table, fn);
1981 }
1982
1983 fib6_purge_rt(rt, fn, net);
1984
1985 if (!info->skip_notify_kernel) {
1986 if (notify_del)
1987 call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1988 rt, NULL);
1989 else if (replace_rt)
1990 call_fib6_entry_notifiers_replace(net, replace_rt);
1991 }
1992 if (!info->skip_notify)
1993 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1994
1995 fib6_info_release(rt);
1996}
1997
1998/* Need to own table->tb6_lock */
1999int fib6_del(struct fib6_info *rt, struct nl_info *info)
2000{
2001 struct net *net = info->nl_net;
2002 struct fib6_info __rcu **rtp;
2003 struct fib6_info __rcu **rtp_next;
2004 struct fib6_table *table;
2005 struct fib6_node *fn;
2006
2007 if (rt == net->ipv6.fib6_null_entry)
2008 return -ENOENT;
2009
2010 table = rt->fib6_table;
2011 fn = rcu_dereference_protected(rt->fib6_node,
2012 lockdep_is_held(&table->tb6_lock));
2013 if (!fn)
2014 return -ENOENT;
2015
2016 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
2017
2018 /*
2019 * Walk the leaf entries looking for ourself
2020 */
2021
2022 for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
2023 struct fib6_info *cur = rcu_dereference_protected(*rtp,
2024 lockdep_is_held(&table->tb6_lock));
2025 if (rt == cur) {
2026 if (fib6_requires_src(cur))
2027 fib6_routes_require_src_dec(info->nl_net);
2028 fib6_del_route(table, fn, rtp, info);
2029 return 0;
2030 }
2031 rtp_next = &cur->fib6_next;
2032 }
2033 return -ENOENT;
2034}
2035
2036/*
2037 * Tree traversal function.
2038 *
2039 * Certainly, it is not interrupt safe.
2040 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
2041 * It means, that we can modify tree during walking
2042 * and use this function for garbage collection, clone pruning,
2043 * cleaning tree when a device goes down etc. etc.
2044 *
2045 * It guarantees that every node will be traversed,
2046 * and that it will be traversed only once.
2047 *
2048 * Callback function w->func may return:
2049 * 0 -> continue walking.
2050 * positive value -> walking is suspended (used by tree dumps,
2051 * and probably by gc, if it will be split to several slices)
2052 * negative value -> terminate walking.
2053 *
2054 * The function itself returns:
2055 * 0 -> walk is complete.
2056 * >0 -> walk is incomplete (i.e. suspended)
2057 * <0 -> walk is terminated by an error.
2058 *
2059 * This function is called with tb6_lock held.
2060 */
2061
2062static int fib6_walk_continue(struct fib6_walker *w)
2063{
2064 struct fib6_node *fn, *pn, *left, *right;
2065
2066 /* w->root should always be table->tb6_root */
2067 WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
2068
2069 for (;;) {
2070 fn = w->node;
2071 if (!fn)
2072 return 0;
2073
2074 switch (w->state) {
2075#ifdef CONFIG_IPV6_SUBTREES
2076 case FWS_S:
2077 if (FIB6_SUBTREE(fn)) {
2078 w->node = FIB6_SUBTREE(fn);
2079 continue;
2080 }
2081 w->state = FWS_L;
2082 fallthrough;
2083#endif
2084 case FWS_L:
2085 left = rcu_dereference_protected(fn->left, 1);
2086 if (left) {
2087 w->node = left;
2088 w->state = FWS_INIT;
2089 continue;
2090 }
2091 w->state = FWS_R;
2092 fallthrough;
2093 case FWS_R:
2094 right = rcu_dereference_protected(fn->right, 1);
2095 if (right) {
2096 w->node = right;
2097 w->state = FWS_INIT;
2098 continue;
2099 }
2100 w->state = FWS_C;
2101 w->leaf = rcu_dereference_protected(fn->leaf, 1);
2102 fallthrough;
2103 case FWS_C:
2104 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
2105 int err;
2106
2107 if (w->skip) {
2108 w->skip--;
2109 goto skip;
2110 }
2111
2112 err = w->func(w);
2113 if (err)
2114 return err;
2115
2116 w->count++;
2117 continue;
2118 }
2119skip:
2120 w->state = FWS_U;
2121 fallthrough;
2122 case FWS_U:
2123 if (fn == w->root)
2124 return 0;
2125 pn = rcu_dereference_protected(fn->parent, 1);
2126 left = rcu_dereference_protected(pn->left, 1);
2127 right = rcu_dereference_protected(pn->right, 1);
2128 w->node = pn;
2129#ifdef CONFIG_IPV6_SUBTREES
2130 if (FIB6_SUBTREE(pn) == fn) {
2131 WARN_ON(!(fn->fn_flags & RTN_ROOT));
2132 w->state = FWS_L;
2133 continue;
2134 }
2135#endif
2136 if (left == fn) {
2137 w->state = FWS_R;
2138 continue;
2139 }
2140 if (right == fn) {
2141 w->state = FWS_C;
2142 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
2143 continue;
2144 }
2145#if RT6_DEBUG >= 2
2146 WARN_ON(1);
2147#endif
2148 }
2149 }
2150}
2151
2152static int fib6_walk(struct net *net, struct fib6_walker *w)
2153{
2154 int res;
2155
2156 w->state = FWS_INIT;
2157 w->node = w->root;
2158
2159 fib6_walker_link(net, w);
2160 res = fib6_walk_continue(w);
2161 if (res <= 0)
2162 fib6_walker_unlink(net, w);
2163 return res;
2164}
2165
2166static int fib6_clean_node(struct fib6_walker *w)
2167{
2168 int res;
2169 struct fib6_info *rt;
2170 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
2171 struct nl_info info = {
2172 .nl_net = c->net,
2173 .skip_notify = c->skip_notify,
2174 };
2175
2176 if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
2177 READ_ONCE(w->node->fn_sernum) != c->sernum)
2178 WRITE_ONCE(w->node->fn_sernum, c->sernum);
2179
2180 if (!c->func) {
2181 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
2182 w->leaf = NULL;
2183 return 0;
2184 }
2185
2186 for_each_fib6_walker_rt(w) {
2187 res = c->func(rt, c->arg);
2188 if (res == -1) {
2189 w->leaf = rt;
2190 res = fib6_del(rt, &info);
2191 if (res) {
2192#if RT6_DEBUG >= 2
2193 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2194 __func__, rt,
2195 rcu_access_pointer(rt->fib6_node),
2196 res);
2197#endif
2198 continue;
2199 }
2200 return 0;
2201 } else if (res == -2) {
2202 if (WARN_ON(!rt->fib6_nsiblings))
2203 continue;
2204 rt = list_last_entry(&rt->fib6_siblings,
2205 struct fib6_info, fib6_siblings);
2206 continue;
2207 }
2208 WARN_ON(res != 0);
2209 }
2210 w->leaf = rt;
2211 return 0;
2212}
2213
2214/*
2215 * Convenient frontend to tree walker.
2216 *
2217 * func is called on each route.
2218 * It may return -2 -> skip multipath route.
2219 * -1 -> delete this route.
2220 * 0 -> continue walking
2221 */
2222
2223static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2224 int (*func)(struct fib6_info *, void *arg),
2225 int sernum, void *arg, bool skip_notify)
2226{
2227 struct fib6_cleaner c;
2228
2229 c.w.root = root;
2230 c.w.func = fib6_clean_node;
2231 c.w.count = 0;
2232 c.w.skip = 0;
2233 c.w.skip_in_node = 0;
2234 c.func = func;
2235 c.sernum = sernum;
2236 c.arg = arg;
2237 c.net = net;
2238 c.skip_notify = skip_notify;
2239
2240 fib6_walk(net, &c.w);
2241}
2242
2243static void __fib6_clean_all(struct net *net,
2244 int (*func)(struct fib6_info *, void *),
2245 int sernum, void *arg, bool skip_notify)
2246{
2247 struct fib6_table *table;
2248 struct hlist_head *head;
2249 unsigned int h;
2250
2251 rcu_read_lock();
2252 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2253 head = &net->ipv6.fib_table_hash[h];
2254 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2255 spin_lock_bh(&table->tb6_lock);
2256 fib6_clean_tree(net, &table->tb6_root,
2257 func, sernum, arg, skip_notify);
2258 spin_unlock_bh(&table->tb6_lock);
2259 }
2260 }
2261 rcu_read_unlock();
2262}
2263
2264void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2265 void *arg)
2266{
2267 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2268}
2269
2270void fib6_clean_all_skip_notify(struct net *net,
2271 int (*func)(struct fib6_info *, void *),
2272 void *arg)
2273{
2274 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2275}
2276
2277static void fib6_flush_trees(struct net *net)
2278{
2279 int new_sernum = fib6_new_sernum(net);
2280
2281 __fib6_clean_all(net, NULL, new_sernum, NULL, false);
2282}
2283
2284/*
2285 * Garbage collection
2286 */
2287
2288static int fib6_age(struct fib6_info *rt, void *arg)
2289{
2290 struct fib6_gc_args *gc_args = arg;
2291 unsigned long now = jiffies;
2292
2293 /*
2294 * check addrconf expiration here.
2295 * Routes are expired even if they are in use.
2296 */
2297
2298 if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2299 if (time_after(now, rt->expires)) {
2300 RT6_TRACE("expiring %p\n", rt);
2301 return -1;
2302 }
2303 gc_args->more++;
2304 }
2305
2306 /* Also age clones in the exception table.
2307 * Note, that clones are aged out
2308 * only if they are not in use now.
2309 */
2310 rt6_age_exceptions(rt, gc_args, now);
2311
2312 return 0;
2313}
2314
2315void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2316{
2317 struct fib6_gc_args gc_args;
2318 unsigned long now;
2319
2320 if (force) {
2321 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2322 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2323 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2324 return;
2325 }
2326 gc_args.timeout = expires ? (int)expires :
2327 net->ipv6.sysctl.ip6_rt_gc_interval;
2328 gc_args.more = 0;
2329
2330 fib6_clean_all(net, fib6_age, &gc_args);
2331 now = jiffies;
2332 net->ipv6.ip6_rt_last_gc = now;
2333
2334 if (gc_args.more)
2335 mod_timer(&net->ipv6.ip6_fib_timer,
2336 round_jiffies(now
2337 + net->ipv6.sysctl.ip6_rt_gc_interval));
2338 else
2339 del_timer(&net->ipv6.ip6_fib_timer);
2340 spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2341}
2342
2343static void fib6_gc_timer_cb(struct timer_list *t)
2344{
2345 struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2346
2347 fib6_run_gc(0, arg, true);
2348}
2349
2350static int __net_init fib6_net_init(struct net *net)
2351{
2352 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2353 int err;
2354
2355 err = fib6_notifier_init(net);
2356 if (err)
2357 return err;
2358
2359 /* Default to 3-tuple */
2360 net->ipv6.sysctl.multipath_hash_fields =
2361 FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK;
2362
2363 spin_lock_init(&net->ipv6.fib6_gc_lock);
2364 rwlock_init(&net->ipv6.fib6_walker_lock);
2365 INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2366 timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2367
2368 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2369 if (!net->ipv6.rt6_stats)
2370 goto out_notifier;
2371
2372 /* Avoid false sharing : Use at least a full cache line */
2373 size = max_t(size_t, size, L1_CACHE_BYTES);
2374
2375 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2376 if (!net->ipv6.fib_table_hash)
2377 goto out_rt6_stats;
2378
2379 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2380 GFP_KERNEL);
2381 if (!net->ipv6.fib6_main_tbl)
2382 goto out_fib_table_hash;
2383
2384 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2385 rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2386 net->ipv6.fib6_null_entry);
2387 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2388 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2389 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2390
2391#ifdef CONFIG_IPV6_MULTIPLE_TABLES
2392 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2393 GFP_KERNEL);
2394 if (!net->ipv6.fib6_local_tbl)
2395 goto out_fib6_main_tbl;
2396 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2397 rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2398 net->ipv6.fib6_null_entry);
2399 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2400 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2401 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2402#endif
2403 fib6_tables_init(net);
2404
2405 return 0;
2406
2407#ifdef CONFIG_IPV6_MULTIPLE_TABLES
2408out_fib6_main_tbl:
2409 kfree(net->ipv6.fib6_main_tbl);
2410#endif
2411out_fib_table_hash:
2412 kfree(net->ipv6.fib_table_hash);
2413out_rt6_stats:
2414 kfree(net->ipv6.rt6_stats);
2415out_notifier:
2416 fib6_notifier_exit(net);
2417 return -ENOMEM;
2418}
2419
2420static void fib6_net_exit(struct net *net)
2421{
2422 unsigned int i;
2423
2424 del_timer_sync(&net->ipv6.ip6_fib_timer);
2425
2426 for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2427 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2428 struct hlist_node *tmp;
2429 struct fib6_table *tb;
2430
2431 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2432 hlist_del(&tb->tb6_hlist);
2433 fib6_free_table(tb);
2434 }
2435 }
2436
2437 kfree(net->ipv6.fib_table_hash);
2438 kfree(net->ipv6.rt6_stats);
2439 fib6_notifier_exit(net);
2440}
2441
2442static struct pernet_operations fib6_net_ops = {
2443 .init = fib6_net_init,
2444 .exit = fib6_net_exit,
2445};
2446
2447int __init fib6_init(void)
2448{
2449 int ret = -ENOMEM;
2450
2451 fib6_node_kmem = kmem_cache_create("fib6_nodes",
2452 sizeof(struct fib6_node), 0,
2453 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT,
2454 NULL);
2455 if (!fib6_node_kmem)
2456 goto out;
2457
2458 ret = register_pernet_subsys(&fib6_net_ops);
2459 if (ret)
2460 goto out_kmem_cache_create;
2461
2462 ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2463 inet6_dump_fib, 0);
2464 if (ret)
2465 goto out_unregister_subsys;
2466
2467 __fib6_flush_trees = fib6_flush_trees;
2468out:
2469 return ret;
2470
2471out_unregister_subsys:
2472 unregister_pernet_subsys(&fib6_net_ops);
2473out_kmem_cache_create:
2474 kmem_cache_destroy(fib6_node_kmem);
2475 goto out;
2476}
2477
2478void fib6_gc_cleanup(void)
2479{
2480 unregister_pernet_subsys(&fib6_net_ops);
2481 kmem_cache_destroy(fib6_node_kmem);
2482}
2483
2484#ifdef CONFIG_PROC_FS
2485static int ipv6_route_native_seq_show(struct seq_file *seq, void *v)
2486{
2487 struct fib6_info *rt = v;
2488 struct ipv6_route_iter *iter = seq->private;
2489 struct fib6_nh *fib6_nh = rt->fib6_nh;
2490 unsigned int flags = rt->fib6_flags;
2491 const struct net_device *dev;
2492
2493 if (rt->nh)
2494 fib6_nh = nexthop_fib6_nh_bh(rt->nh);
2495
2496 seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2497
2498#ifdef CONFIG_IPV6_SUBTREES
2499 seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2500#else
2501 seq_puts(seq, "00000000000000000000000000000000 00 ");
2502#endif
2503 if (fib6_nh->fib_nh_gw_family) {
2504 flags |= RTF_GATEWAY;
2505 seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6);
2506 } else {
2507 seq_puts(seq, "00000000000000000000000000000000");
2508 }
2509
2510 dev = fib6_nh->fib_nh_dev;
2511 seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2512 rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2513 flags, dev ? dev->name : "");
2514 iter->w.leaf = NULL;
2515 return 0;
2516}
2517
2518static int ipv6_route_yield(struct fib6_walker *w)
2519{
2520 struct ipv6_route_iter *iter = w->args;
2521
2522 if (!iter->skip)
2523 return 1;
2524
2525 do {
2526 iter->w.leaf = rcu_dereference_protected(
2527 iter->w.leaf->fib6_next,
2528 lockdep_is_held(&iter->tbl->tb6_lock));
2529 iter->skip--;
2530 if (!iter->skip && iter->w.leaf)
2531 return 1;
2532 } while (iter->w.leaf);
2533
2534 return 0;
2535}
2536
2537static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2538 struct net *net)
2539{
2540 memset(&iter->w, 0, sizeof(iter->w));
2541 iter->w.func = ipv6_route_yield;
2542 iter->w.root = &iter->tbl->tb6_root;
2543 iter->w.state = FWS_INIT;
2544 iter->w.node = iter->w.root;
2545 iter->w.args = iter;
2546 iter->sernum = READ_ONCE(iter->w.root->fn_sernum);
2547 INIT_LIST_HEAD(&iter->w.lh);
2548 fib6_walker_link(net, &iter->w);
2549}
2550
2551static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2552 struct net *net)
2553{
2554 unsigned int h;
2555 struct hlist_node *node;
2556
2557 if (tbl) {
2558 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2559 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2560 } else {
2561 h = 0;
2562 node = NULL;
2563 }
2564
2565 while (!node && h < FIB6_TABLE_HASHSZ) {
2566 node = rcu_dereference_bh(
2567 hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2568 }
2569 return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2570}
2571
2572static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2573{
2574 int sernum = READ_ONCE(iter->w.root->fn_sernum);
2575
2576 if (iter->sernum != sernum) {
2577 iter->sernum = sernum;
2578 iter->w.state = FWS_INIT;
2579 iter->w.node = iter->w.root;
2580 WARN_ON(iter->w.skip);
2581 iter->w.skip = iter->w.count;
2582 }
2583}
2584
2585static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2586{
2587 int r;
2588 struct fib6_info *n;
2589 struct net *net = seq_file_net(seq);
2590 struct ipv6_route_iter *iter = seq->private;
2591
2592 ++(*pos);
2593 if (!v)
2594 goto iter_table;
2595
2596 n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2597 if (n)
2598 return n;
2599
2600iter_table:
2601 ipv6_route_check_sernum(iter);
2602 spin_lock_bh(&iter->tbl->tb6_lock);
2603 r = fib6_walk_continue(&iter->w);
2604 spin_unlock_bh(&iter->tbl->tb6_lock);
2605 if (r > 0) {
2606 return iter->w.leaf;
2607 } else if (r < 0) {
2608 fib6_walker_unlink(net, &iter->w);
2609 return NULL;
2610 }
2611 fib6_walker_unlink(net, &iter->w);
2612
2613 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2614 if (!iter->tbl)
2615 return NULL;
2616
2617 ipv6_route_seq_setup_walk(iter, net);
2618 goto iter_table;
2619}
2620
2621static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2622 __acquires(RCU_BH)
2623{
2624 struct net *net = seq_file_net(seq);
2625 struct ipv6_route_iter *iter = seq->private;
2626
2627 rcu_read_lock_bh();
2628 iter->tbl = ipv6_route_seq_next_table(NULL, net);
2629 iter->skip = *pos;
2630
2631 if (iter->tbl) {
2632 loff_t p = 0;
2633
2634 ipv6_route_seq_setup_walk(iter, net);
2635 return ipv6_route_seq_next(seq, NULL, &p);
2636 } else {
2637 return NULL;
2638 }
2639}
2640
2641static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2642{
2643 struct fib6_walker *w = &iter->w;
2644 return w->node && !(w->state == FWS_U && w->node == w->root);
2645}
2646
2647static void ipv6_route_native_seq_stop(struct seq_file *seq, void *v)
2648 __releases(RCU_BH)
2649{
2650 struct net *net = seq_file_net(seq);
2651 struct ipv6_route_iter *iter = seq->private;
2652
2653 if (ipv6_route_iter_active(iter))
2654 fib6_walker_unlink(net, &iter->w);
2655
2656 rcu_read_unlock_bh();
2657}
2658
2659#if IS_BUILTIN(CONFIG_IPV6) && defined(CONFIG_BPF_SYSCALL)
2660static int ipv6_route_prog_seq_show(struct bpf_prog *prog,
2661 struct bpf_iter_meta *meta,
2662 void *v)
2663{
2664 struct bpf_iter__ipv6_route ctx;
2665
2666 ctx.meta = meta;
2667 ctx.rt = v;
2668 return bpf_iter_run_prog(prog, &ctx);
2669}
2670
2671static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2672{
2673 struct ipv6_route_iter *iter = seq->private;
2674 struct bpf_iter_meta meta;
2675 struct bpf_prog *prog;
2676 int ret;
2677
2678 meta.seq = seq;
2679 prog = bpf_iter_get_info(&meta, false);
2680 if (!prog)
2681 return ipv6_route_native_seq_show(seq, v);
2682
2683 ret = ipv6_route_prog_seq_show(prog, &meta, v);
2684 iter->w.leaf = NULL;
2685
2686 return ret;
2687}
2688
2689static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2690{
2691 struct bpf_iter_meta meta;
2692 struct bpf_prog *prog;
2693
2694 if (!v) {
2695 meta.seq = seq;
2696 prog = bpf_iter_get_info(&meta, true);
2697 if (prog)
2698 (void)ipv6_route_prog_seq_show(prog, &meta, v);
2699 }
2700
2701 ipv6_route_native_seq_stop(seq, v);
2702}
2703#else
2704static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2705{
2706 return ipv6_route_native_seq_show(seq, v);
2707}
2708
2709static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2710{
2711 ipv6_route_native_seq_stop(seq, v);
2712}
2713#endif
2714
2715const struct seq_operations ipv6_route_seq_ops = {
2716 .start = ipv6_route_seq_start,
2717 .next = ipv6_route_seq_next,
2718 .stop = ipv6_route_seq_stop,
2719 .show = ipv6_route_seq_show
2720};
2721#endif /* CONFIG_PROC_FS */