Linux 2.6.26-rc5
[linux-2.6/openmoko-kernel/knife-kernel.git] / net / ipv6 / ip6_fib.c
blob1ee4fa17c12959a14c835663abdb81ce349f8de4
1 /*
2 * Linux INET6 implementation
3 * Forwarding Information Database
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * $Id: ip6_fib.c,v 1.25 2001/10/31 21:55:55 davem Exp $
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
17 * Changes:
18 * Yuji SEKIYA @USAGI: Support default route on router node;
19 * remove ip6_null_entry from the top of
20 * routing table.
21 * Ville Nuorvala: Fixed routing subtrees.
23 #include <linux/errno.h>
24 #include <linux/types.h>
25 #include <linux/net.h>
26 #include <linux/route.h>
27 #include <linux/netdevice.h>
28 #include <linux/in6.h>
29 #include <linux/init.h>
30 #include <linux/list.h>
32 #ifdef CONFIG_PROC_FS
33 #include <linux/proc_fs.h>
34 #endif
36 #include <net/ipv6.h>
37 #include <net/ndisc.h>
38 #include <net/addrconf.h>
40 #include <net/ip6_fib.h>
41 #include <net/ip6_route.h>
43 #define RT6_DEBUG 2
45 #if RT6_DEBUG >= 3
46 #define RT6_TRACE(x...) printk(KERN_DEBUG x)
47 #else
48 #define RT6_TRACE(x...) do { ; } while (0)
49 #endif
51 static struct kmem_cache * fib6_node_kmem __read_mostly;
53 enum fib_walk_state_t
55 #ifdef CONFIG_IPV6_SUBTREES
56 FWS_S,
57 #endif
58 FWS_L,
59 FWS_R,
60 FWS_C,
61 FWS_U
64 struct fib6_cleaner_t
66 struct fib6_walker_t w;
67 struct net *net;
68 int (*func)(struct rt6_info *, void *arg);
69 void *arg;
72 static DEFINE_RWLOCK(fib6_walker_lock);
74 #ifdef CONFIG_IPV6_SUBTREES
75 #define FWS_INIT FWS_S
76 #else
77 #define FWS_INIT FWS_L
78 #endif
80 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
81 struct rt6_info *rt);
82 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
83 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
84 static int fib6_walk(struct fib6_walker_t *w);
85 static int fib6_walk_continue(struct fib6_walker_t *w);
88 * A routing update causes an increase of the serial number on the
89 * affected subtree. This allows for cached routes to be asynchronously
90 * tested when modifications are made to the destination cache as a
91 * result of redirects, path MTU changes, etc.
94 static __u32 rt_sernum;
96 static void fib6_gc_timer_cb(unsigned long arg);
98 static struct fib6_walker_t fib6_walker_list = {
99 .prev = &fib6_walker_list,
100 .next = &fib6_walker_list,
103 #define FOR_WALKERS(w) for ((w)=fib6_walker_list.next; (w) != &fib6_walker_list; (w)=(w)->next)
105 static inline void fib6_walker_link(struct fib6_walker_t *w)
107 write_lock_bh(&fib6_walker_lock);
108 w->next = fib6_walker_list.next;
109 w->prev = &fib6_walker_list;
110 w->next->prev = w;
111 w->prev->next = w;
112 write_unlock_bh(&fib6_walker_lock);
115 static inline void fib6_walker_unlink(struct fib6_walker_t *w)
117 write_lock_bh(&fib6_walker_lock);
118 w->next->prev = w->prev;
119 w->prev->next = w->next;
120 w->prev = w->next = w;
121 write_unlock_bh(&fib6_walker_lock);
123 static __inline__ u32 fib6_new_sernum(void)
125 u32 n = ++rt_sernum;
126 if ((__s32)n <= 0)
127 rt_sernum = n = 1;
128 return n;
132 * Auxiliary address test functions for the radix tree.
134 * These assume a 32bit processor (although it will work on
135 * 64bit processors)
139 * test bit
142 static __inline__ __be32 addr_bit_set(void *token, int fn_bit)
144 __be32 *addr = token;
146 return htonl(1 << ((~fn_bit)&0x1F)) & addr[fn_bit>>5];
149 static __inline__ struct fib6_node * node_alloc(void)
151 struct fib6_node *fn;
153 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
155 return fn;
158 static __inline__ void node_free(struct fib6_node * fn)
160 kmem_cache_free(fib6_node_kmem, fn);
163 static __inline__ void rt6_release(struct rt6_info *rt)
165 if (atomic_dec_and_test(&rt->rt6i_ref))
166 dst_free(&rt->u.dst);
169 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
170 #define FIB_TABLE_HASHSZ 256
171 #else
172 #define FIB_TABLE_HASHSZ 1
173 #endif
175 static void fib6_link_table(struct net *net, struct fib6_table *tb)
177 unsigned int h;
180 * Initialize table lock at a single place to give lockdep a key,
181 * tables aren't visible prior to being linked to the list.
183 rwlock_init(&tb->tb6_lock);
185 h = tb->tb6_id & (FIB_TABLE_HASHSZ - 1);
188 * No protection necessary, this is the only list mutatation
189 * operation, tables never disappear once they exist.
191 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
194 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
196 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
198 struct fib6_table *table;
200 table = kzalloc(sizeof(*table), GFP_ATOMIC);
201 if (table != NULL) {
202 table->tb6_id = id;
203 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
204 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
207 return table;
210 struct fib6_table *fib6_new_table(struct net *net, u32 id)
212 struct fib6_table *tb;
214 if (id == 0)
215 id = RT6_TABLE_MAIN;
216 tb = fib6_get_table(net, id);
217 if (tb)
218 return tb;
220 tb = fib6_alloc_table(net, id);
221 if (tb != NULL)
222 fib6_link_table(net, tb);
224 return tb;
227 struct fib6_table *fib6_get_table(struct net *net, u32 id)
229 struct fib6_table *tb;
230 struct hlist_head *head;
231 struct hlist_node *node;
232 unsigned int h;
234 if (id == 0)
235 id = RT6_TABLE_MAIN;
236 h = id & (FIB_TABLE_HASHSZ - 1);
237 rcu_read_lock();
238 head = &net->ipv6.fib_table_hash[h];
239 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
240 if (tb->tb6_id == id) {
241 rcu_read_unlock();
242 return tb;
245 rcu_read_unlock();
247 return NULL;
250 static void fib6_tables_init(struct net *net)
252 fib6_link_table(net, net->ipv6.fib6_main_tbl);
253 fib6_link_table(net, net->ipv6.fib6_local_tbl);
255 #else
257 struct fib6_table *fib6_new_table(struct net *net, u32 id)
259 return fib6_get_table(net, id);
262 struct fib6_table *fib6_get_table(struct net *net, u32 id)
264 return net->ipv6.fib6_main_tbl;
267 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi *fl,
268 int flags, pol_lookup_t lookup)
270 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl, flags);
273 static void fib6_tables_init(struct net *net)
275 fib6_link_table(net, net->ipv6.fib6_main_tbl);
278 #endif
280 static int fib6_dump_node(struct fib6_walker_t *w)
282 int res;
283 struct rt6_info *rt;
285 for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) {
286 res = rt6_dump_route(rt, w->args);
287 if (res < 0) {
288 /* Frame is full, suspend walking */
289 w->leaf = rt;
290 return 1;
292 BUG_TRAP(res!=0);
294 w->leaf = NULL;
295 return 0;
298 static void fib6_dump_end(struct netlink_callback *cb)
300 struct fib6_walker_t *w = (void*)cb->args[2];
302 if (w) {
303 cb->args[2] = 0;
304 kfree(w);
306 cb->done = (void*)cb->args[3];
307 cb->args[1] = 3;
310 static int fib6_dump_done(struct netlink_callback *cb)
312 fib6_dump_end(cb);
313 return cb->done ? cb->done(cb) : 0;
316 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
317 struct netlink_callback *cb)
319 struct fib6_walker_t *w;
320 int res;
322 w = (void *)cb->args[2];
323 w->root = &table->tb6_root;
325 if (cb->args[4] == 0) {
326 read_lock_bh(&table->tb6_lock);
327 res = fib6_walk(w);
328 read_unlock_bh(&table->tb6_lock);
329 if (res > 0)
330 cb->args[4] = 1;
331 } else {
332 read_lock_bh(&table->tb6_lock);
333 res = fib6_walk_continue(w);
334 read_unlock_bh(&table->tb6_lock);
335 if (res != 0) {
336 if (res < 0)
337 fib6_walker_unlink(w);
338 goto end;
340 fib6_walker_unlink(w);
341 cb->args[4] = 0;
343 end:
344 return res;
347 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
349 struct net *net = sock_net(skb->sk);
350 unsigned int h, s_h;
351 unsigned int e = 0, s_e;
352 struct rt6_rtnl_dump_arg arg;
353 struct fib6_walker_t *w;
354 struct fib6_table *tb;
355 struct hlist_node *node;
356 struct hlist_head *head;
357 int res = 0;
359 s_h = cb->args[0];
360 s_e = cb->args[1];
362 w = (void *)cb->args[2];
363 if (w == NULL) {
364 /* New dump:
366 * 1. hook callback destructor.
368 cb->args[3] = (long)cb->done;
369 cb->done = fib6_dump_done;
372 * 2. allocate and initialize walker.
374 w = kzalloc(sizeof(*w), GFP_ATOMIC);
375 if (w == NULL)
376 return -ENOMEM;
377 w->func = fib6_dump_node;
378 cb->args[2] = (long)w;
381 arg.skb = skb;
382 arg.cb = cb;
383 w->args = &arg;
385 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
386 e = 0;
387 head = &net->ipv6.fib_table_hash[h];
388 hlist_for_each_entry(tb, node, head, tb6_hlist) {
389 if (e < s_e)
390 goto next;
391 res = fib6_dump_table(tb, skb, cb);
392 if (res != 0)
393 goto out;
394 next:
395 e++;
398 out:
399 cb->args[1] = e;
400 cb->args[0] = h;
402 res = res < 0 ? res : skb->len;
403 if (res <= 0)
404 fib6_dump_end(cb);
405 return res;
409 * Routing Table
411 * return the appropriate node for a routing tree "add" operation
412 * by either creating and inserting or by returning an existing
413 * node.
416 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
417 int addrlen, int plen,
418 int offset)
420 struct fib6_node *fn, *in, *ln;
421 struct fib6_node *pn = NULL;
422 struct rt6key *key;
423 int bit;
424 __be32 dir = 0;
425 __u32 sernum = fib6_new_sernum();
427 RT6_TRACE("fib6_add_1\n");
429 /* insert node in tree */
431 fn = root;
433 do {
434 key = (struct rt6key *)((u8 *)fn->leaf + offset);
437 * Prefix match
439 if (plen < fn->fn_bit ||
440 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
441 goto insert_above;
444 * Exact match ?
447 if (plen == fn->fn_bit) {
448 /* clean up an intermediate node */
449 if ((fn->fn_flags & RTN_RTINFO) == 0) {
450 rt6_release(fn->leaf);
451 fn->leaf = NULL;
454 fn->fn_sernum = sernum;
456 return fn;
460 * We have more bits to go
463 /* Try to walk down on tree. */
464 fn->fn_sernum = sernum;
465 dir = addr_bit_set(addr, fn->fn_bit);
466 pn = fn;
467 fn = dir ? fn->right: fn->left;
468 } while (fn);
471 * We walked to the bottom of tree.
472 * Create new leaf node without children.
475 ln = node_alloc();
477 if (ln == NULL)
478 return NULL;
479 ln->fn_bit = plen;
481 ln->parent = pn;
482 ln->fn_sernum = sernum;
484 if (dir)
485 pn->right = ln;
486 else
487 pn->left = ln;
489 return ln;
492 insert_above:
494 * split since we don't have a common prefix anymore or
495 * we have a less significant route.
496 * we've to insert an intermediate node on the list
497 * this new node will point to the one we need to create
498 * and the current
501 pn = fn->parent;
503 /* find 1st bit in difference between the 2 addrs.
505 See comment in __ipv6_addr_diff: bit may be an invalid value,
506 but if it is >= plen, the value is ignored in any case.
509 bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
512 * (intermediate)[in]
513 * / \
514 * (new leaf node)[ln] (old node)[fn]
516 if (plen > bit) {
517 in = node_alloc();
518 ln = node_alloc();
520 if (in == NULL || ln == NULL) {
521 if (in)
522 node_free(in);
523 if (ln)
524 node_free(ln);
525 return NULL;
529 * new intermediate node.
530 * RTN_RTINFO will
531 * be off since that an address that chooses one of
532 * the branches would not match less specific routes
533 * in the other branch
536 in->fn_bit = bit;
538 in->parent = pn;
539 in->leaf = fn->leaf;
540 atomic_inc(&in->leaf->rt6i_ref);
542 in->fn_sernum = sernum;
544 /* update parent pointer */
545 if (dir)
546 pn->right = in;
547 else
548 pn->left = in;
550 ln->fn_bit = plen;
552 ln->parent = in;
553 fn->parent = in;
555 ln->fn_sernum = sernum;
557 if (addr_bit_set(addr, bit)) {
558 in->right = ln;
559 in->left = fn;
560 } else {
561 in->left = ln;
562 in->right = fn;
564 } else { /* plen <= bit */
567 * (new leaf node)[ln]
568 * / \
569 * (old node)[fn] NULL
572 ln = node_alloc();
574 if (ln == NULL)
575 return NULL;
577 ln->fn_bit = plen;
579 ln->parent = pn;
581 ln->fn_sernum = sernum;
583 if (dir)
584 pn->right = ln;
585 else
586 pn->left = ln;
588 if (addr_bit_set(&key->addr, plen))
589 ln->right = fn;
590 else
591 ln->left = fn;
593 fn->parent = ln;
595 return ln;
599 * Insert routing information in a node.
602 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
603 struct nl_info *info)
605 struct rt6_info *iter = NULL;
606 struct rt6_info **ins;
608 ins = &fn->leaf;
610 for (iter = fn->leaf; iter; iter=iter->u.dst.rt6_next) {
612 * Search for duplicates
615 if (iter->rt6i_metric == rt->rt6i_metric) {
617 * Same priority level
620 if (iter->rt6i_dev == rt->rt6i_dev &&
621 iter->rt6i_idev == rt->rt6i_idev &&
622 ipv6_addr_equal(&iter->rt6i_gateway,
623 &rt->rt6i_gateway)) {
624 if (!(iter->rt6i_flags&RTF_EXPIRES))
625 return -EEXIST;
626 iter->rt6i_expires = rt->rt6i_expires;
627 if (!(rt->rt6i_flags&RTF_EXPIRES)) {
628 iter->rt6i_flags &= ~RTF_EXPIRES;
629 iter->rt6i_expires = 0;
631 return -EEXIST;
635 if (iter->rt6i_metric > rt->rt6i_metric)
636 break;
638 ins = &iter->u.dst.rt6_next;
641 /* Reset round-robin state, if necessary */
642 if (ins == &fn->leaf)
643 fn->rr_ptr = NULL;
646 * insert node
649 rt->u.dst.rt6_next = iter;
650 *ins = rt;
651 rt->rt6i_node = fn;
652 atomic_inc(&rt->rt6i_ref);
653 inet6_rt_notify(RTM_NEWROUTE, rt, info);
654 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
656 if ((fn->fn_flags & RTN_RTINFO) == 0) {
657 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
658 fn->fn_flags |= RTN_RTINFO;
661 return 0;
664 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
666 if (net->ipv6.ip6_fib_timer->expires == 0 &&
667 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE)))
668 mod_timer(net->ipv6.ip6_fib_timer, jiffies +
669 net->ipv6.sysctl.ip6_rt_gc_interval);
672 void fib6_force_start_gc(struct net *net)
674 if (net->ipv6.ip6_fib_timer->expires == 0)
675 mod_timer(net->ipv6.ip6_fib_timer, jiffies +
676 net->ipv6.sysctl.ip6_rt_gc_interval);
680 * Add routing information to the routing tree.
681 * <destination addr>/<source addr>
682 * with source addr info in sub-trees
685 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
687 struct fib6_node *fn, *pn = NULL;
688 int err = -ENOMEM;
690 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
691 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst));
693 if (fn == NULL)
694 goto out;
696 pn = fn;
698 #ifdef CONFIG_IPV6_SUBTREES
699 if (rt->rt6i_src.plen) {
700 struct fib6_node *sn;
702 if (fn->subtree == NULL) {
703 struct fib6_node *sfn;
706 * Create subtree.
708 * fn[main tree]
710 * sfn[subtree root]
712 * sn[new leaf node]
715 /* Create subtree root node */
716 sfn = node_alloc();
717 if (sfn == NULL)
718 goto st_failure;
720 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
721 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
722 sfn->fn_flags = RTN_ROOT;
723 sfn->fn_sernum = fib6_new_sernum();
725 /* Now add the first leaf node to new subtree */
727 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
728 sizeof(struct in6_addr), rt->rt6i_src.plen,
729 offsetof(struct rt6_info, rt6i_src));
731 if (sn == NULL) {
732 /* If it is failed, discard just allocated
733 root, and then (in st_failure) stale node
734 in main tree.
736 node_free(sfn);
737 goto st_failure;
740 /* Now link new subtree to main tree */
741 sfn->parent = fn;
742 fn->subtree = sfn;
743 } else {
744 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
745 sizeof(struct in6_addr), rt->rt6i_src.plen,
746 offsetof(struct rt6_info, rt6i_src));
748 if (sn == NULL)
749 goto st_failure;
752 if (fn->leaf == NULL) {
753 fn->leaf = rt;
754 atomic_inc(&rt->rt6i_ref);
756 fn = sn;
758 #endif
760 err = fib6_add_rt2node(fn, rt, info);
762 if (err == 0) {
763 fib6_start_gc(info->nl_net, rt);
764 if (!(rt->rt6i_flags&RTF_CACHE))
765 fib6_prune_clones(info->nl_net, pn, rt);
768 out:
769 if (err) {
770 #ifdef CONFIG_IPV6_SUBTREES
772 * If fib6_add_1 has cleared the old leaf pointer in the
773 * super-tree leaf node we have to find a new one for it.
775 if (pn != fn && pn->leaf == rt) {
776 pn->leaf = NULL;
777 atomic_dec(&rt->rt6i_ref);
779 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
780 pn->leaf = fib6_find_prefix(info->nl_net, pn);
781 #if RT6_DEBUG >= 2
782 if (!pn->leaf) {
783 BUG_TRAP(pn->leaf != NULL);
784 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
786 #endif
787 atomic_inc(&pn->leaf->rt6i_ref);
789 #endif
790 dst_free(&rt->u.dst);
792 return err;
794 #ifdef CONFIG_IPV6_SUBTREES
795 /* Subtree creation failed, probably main tree node
796 is orphan. If it is, shoot it.
798 st_failure:
799 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
800 fib6_repair_tree(info->nl_net, fn);
801 dst_free(&rt->u.dst);
802 return err;
803 #endif
807 * Routing tree lookup
811 struct lookup_args {
812 int offset; /* key offset on rt6_info */
813 struct in6_addr *addr; /* search key */
816 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
817 struct lookup_args *args)
819 struct fib6_node *fn;
820 __be32 dir;
822 if (unlikely(args->offset == 0))
823 return NULL;
826 * Descend on a tree
829 fn = root;
831 for (;;) {
832 struct fib6_node *next;
834 dir = addr_bit_set(args->addr, fn->fn_bit);
836 next = dir ? fn->right : fn->left;
838 if (next) {
839 fn = next;
840 continue;
843 break;
846 while(fn) {
847 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
848 struct rt6key *key;
850 key = (struct rt6key *) ((u8 *) fn->leaf +
851 args->offset);
853 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
854 #ifdef CONFIG_IPV6_SUBTREES
855 if (fn->subtree)
856 fn = fib6_lookup_1(fn->subtree, args + 1);
857 #endif
858 if (!fn || fn->fn_flags & RTN_RTINFO)
859 return fn;
863 if (fn->fn_flags & RTN_ROOT)
864 break;
866 fn = fn->parent;
869 return NULL;
872 struct fib6_node * fib6_lookup(struct fib6_node *root, struct in6_addr *daddr,
873 struct in6_addr *saddr)
875 struct fib6_node *fn;
876 struct lookup_args args[] = {
878 .offset = offsetof(struct rt6_info, rt6i_dst),
879 .addr = daddr,
881 #ifdef CONFIG_IPV6_SUBTREES
883 .offset = offsetof(struct rt6_info, rt6i_src),
884 .addr = saddr,
886 #endif
888 .offset = 0, /* sentinel */
892 fn = fib6_lookup_1(root, daddr ? args : args + 1);
894 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT)
895 fn = root;
897 return fn;
901 * Get node with specified destination prefix (and source prefix,
902 * if subtrees are used)
906 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
907 struct in6_addr *addr,
908 int plen, int offset)
910 struct fib6_node *fn;
912 for (fn = root; fn ; ) {
913 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
916 * Prefix match
918 if (plen < fn->fn_bit ||
919 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
920 return NULL;
922 if (plen == fn->fn_bit)
923 return fn;
926 * We have more bits to go
928 if (addr_bit_set(addr, fn->fn_bit))
929 fn = fn->right;
930 else
931 fn = fn->left;
933 return NULL;
936 struct fib6_node * fib6_locate(struct fib6_node *root,
937 struct in6_addr *daddr, int dst_len,
938 struct in6_addr *saddr, int src_len)
940 struct fib6_node *fn;
942 fn = fib6_locate_1(root, daddr, dst_len,
943 offsetof(struct rt6_info, rt6i_dst));
945 #ifdef CONFIG_IPV6_SUBTREES
946 if (src_len) {
947 BUG_TRAP(saddr!=NULL);
948 if (fn && fn->subtree)
949 fn = fib6_locate_1(fn->subtree, saddr, src_len,
950 offsetof(struct rt6_info, rt6i_src));
952 #endif
954 if (fn && fn->fn_flags&RTN_RTINFO)
955 return fn;
957 return NULL;
962 * Deletion
966 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
968 if (fn->fn_flags&RTN_ROOT)
969 return net->ipv6.ip6_null_entry;
971 while(fn) {
972 if(fn->left)
973 return fn->left->leaf;
975 if(fn->right)
976 return fn->right->leaf;
978 fn = FIB6_SUBTREE(fn);
980 return NULL;
984 * Called to trim the tree of intermediate nodes when possible. "fn"
985 * is the node we want to try and remove.
988 static struct fib6_node *fib6_repair_tree(struct net *net,
989 struct fib6_node *fn)
991 int children;
992 int nstate;
993 struct fib6_node *child, *pn;
994 struct fib6_walker_t *w;
995 int iter = 0;
997 for (;;) {
998 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
999 iter++;
1001 BUG_TRAP(!(fn->fn_flags&RTN_RTINFO));
1002 BUG_TRAP(!(fn->fn_flags&RTN_TL_ROOT));
1003 BUG_TRAP(fn->leaf==NULL);
1005 children = 0;
1006 child = NULL;
1007 if (fn->right) child = fn->right, children |= 1;
1008 if (fn->left) child = fn->left, children |= 2;
1010 if (children == 3 || FIB6_SUBTREE(fn)
1011 #ifdef CONFIG_IPV6_SUBTREES
1012 /* Subtree root (i.e. fn) may have one child */
1013 || (children && fn->fn_flags&RTN_ROOT)
1014 #endif
1016 fn->leaf = fib6_find_prefix(net, fn);
1017 #if RT6_DEBUG >= 2
1018 if (fn->leaf==NULL) {
1019 BUG_TRAP(fn->leaf);
1020 fn->leaf = net->ipv6.ip6_null_entry;
1022 #endif
1023 atomic_inc(&fn->leaf->rt6i_ref);
1024 return fn->parent;
1027 pn = fn->parent;
1028 #ifdef CONFIG_IPV6_SUBTREES
1029 if (FIB6_SUBTREE(pn) == fn) {
1030 BUG_TRAP(fn->fn_flags&RTN_ROOT);
1031 FIB6_SUBTREE(pn) = NULL;
1032 nstate = FWS_L;
1033 } else {
1034 BUG_TRAP(!(fn->fn_flags&RTN_ROOT));
1035 #endif
1036 if (pn->right == fn) pn->right = child;
1037 else if (pn->left == fn) pn->left = child;
1038 #if RT6_DEBUG >= 2
1039 else BUG_TRAP(0);
1040 #endif
1041 if (child)
1042 child->parent = pn;
1043 nstate = FWS_R;
1044 #ifdef CONFIG_IPV6_SUBTREES
1046 #endif
1048 read_lock(&fib6_walker_lock);
1049 FOR_WALKERS(w) {
1050 if (child == NULL) {
1051 if (w->root == fn) {
1052 w->root = w->node = NULL;
1053 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1054 } else if (w->node == fn) {
1055 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1056 w->node = pn;
1057 w->state = nstate;
1059 } else {
1060 if (w->root == fn) {
1061 w->root = child;
1062 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1064 if (w->node == fn) {
1065 w->node = child;
1066 if (children&2) {
1067 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1068 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1069 } else {
1070 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1071 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1076 read_unlock(&fib6_walker_lock);
1078 node_free(fn);
1079 if (pn->fn_flags&RTN_RTINFO || FIB6_SUBTREE(pn))
1080 return pn;
1082 rt6_release(pn->leaf);
1083 pn->leaf = NULL;
1084 fn = pn;
1088 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1089 struct nl_info *info)
1091 struct fib6_walker_t *w;
1092 struct rt6_info *rt = *rtp;
1093 struct net *net = info->nl_net;
1095 RT6_TRACE("fib6_del_route\n");
1097 /* Unlink it */
1098 *rtp = rt->u.dst.rt6_next;
1099 rt->rt6i_node = NULL;
1100 net->ipv6.rt6_stats->fib_rt_entries--;
1101 net->ipv6.rt6_stats->fib_discarded_routes++;
1103 /* Reset round-robin state, if necessary */
1104 if (fn->rr_ptr == rt)
1105 fn->rr_ptr = NULL;
1107 /* Adjust walkers */
1108 read_lock(&fib6_walker_lock);
1109 FOR_WALKERS(w) {
1110 if (w->state == FWS_C && w->leaf == rt) {
1111 RT6_TRACE("walker %p adjusted by delroute\n", w);
1112 w->leaf = rt->u.dst.rt6_next;
1113 if (w->leaf == NULL)
1114 w->state = FWS_U;
1117 read_unlock(&fib6_walker_lock);
1119 rt->u.dst.rt6_next = NULL;
1121 /* If it was last route, expunge its radix tree node */
1122 if (fn->leaf == NULL) {
1123 fn->fn_flags &= ~RTN_RTINFO;
1124 net->ipv6.rt6_stats->fib_route_nodes--;
1125 fn = fib6_repair_tree(net, fn);
1128 if (atomic_read(&rt->rt6i_ref) != 1) {
1129 /* This route is used as dummy address holder in some split
1130 * nodes. It is not leaked, but it still holds other resources,
1131 * which must be released in time. So, scan ascendant nodes
1132 * and replace dummy references to this route with references
1133 * to still alive ones.
1135 while (fn) {
1136 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) {
1137 fn->leaf = fib6_find_prefix(net, fn);
1138 atomic_inc(&fn->leaf->rt6i_ref);
1139 rt6_release(rt);
1141 fn = fn->parent;
1143 /* No more references are possible at this point. */
1144 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1147 inet6_rt_notify(RTM_DELROUTE, rt, info);
1148 rt6_release(rt);
1151 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1153 struct net *net = info->nl_net;
1154 struct fib6_node *fn = rt->rt6i_node;
1155 struct rt6_info **rtp;
1157 #if RT6_DEBUG >= 2
1158 if (rt->u.dst.obsolete>0) {
1159 BUG_TRAP(fn==NULL);
1160 return -ENOENT;
1162 #endif
1163 if (fn == NULL || rt == net->ipv6.ip6_null_entry)
1164 return -ENOENT;
1166 BUG_TRAP(fn->fn_flags&RTN_RTINFO);
1168 if (!(rt->rt6i_flags&RTF_CACHE)) {
1169 struct fib6_node *pn = fn;
1170 #ifdef CONFIG_IPV6_SUBTREES
1171 /* clones of this route might be in another subtree */
1172 if (rt->rt6i_src.plen) {
1173 while (!(pn->fn_flags&RTN_ROOT))
1174 pn = pn->parent;
1175 pn = pn->parent;
1177 #endif
1178 fib6_prune_clones(info->nl_net, pn, rt);
1182 * Walk the leaf entries looking for ourself
1185 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->u.dst.rt6_next) {
1186 if (*rtp == rt) {
1187 fib6_del_route(fn, rtp, info);
1188 return 0;
1191 return -ENOENT;
1195 * Tree traversal function.
1197 * Certainly, it is not interrupt safe.
1198 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1199 * It means, that we can modify tree during walking
1200 * and use this function for garbage collection, clone pruning,
1201 * cleaning tree when a device goes down etc. etc.
1203 * It guarantees that every node will be traversed,
1204 * and that it will be traversed only once.
1206 * Callback function w->func may return:
1207 * 0 -> continue walking.
1208 * positive value -> walking is suspended (used by tree dumps,
1209 * and probably by gc, if it will be split to several slices)
1210 * negative value -> terminate walking.
1212 * The function itself returns:
1213 * 0 -> walk is complete.
1214 * >0 -> walk is incomplete (i.e. suspended)
1215 * <0 -> walk is terminated by an error.
1218 static int fib6_walk_continue(struct fib6_walker_t *w)
1220 struct fib6_node *fn, *pn;
1222 for (;;) {
1223 fn = w->node;
1224 if (fn == NULL)
1225 return 0;
1227 if (w->prune && fn != w->root &&
1228 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) {
1229 w->state = FWS_C;
1230 w->leaf = fn->leaf;
1232 switch (w->state) {
1233 #ifdef CONFIG_IPV6_SUBTREES
1234 case FWS_S:
1235 if (FIB6_SUBTREE(fn)) {
1236 w->node = FIB6_SUBTREE(fn);
1237 continue;
1239 w->state = FWS_L;
1240 #endif
1241 case FWS_L:
1242 if (fn->left) {
1243 w->node = fn->left;
1244 w->state = FWS_INIT;
1245 continue;
1247 w->state = FWS_R;
1248 case FWS_R:
1249 if (fn->right) {
1250 w->node = fn->right;
1251 w->state = FWS_INIT;
1252 continue;
1254 w->state = FWS_C;
1255 w->leaf = fn->leaf;
1256 case FWS_C:
1257 if (w->leaf && fn->fn_flags&RTN_RTINFO) {
1258 int err = w->func(w);
1259 if (err)
1260 return err;
1261 continue;
1263 w->state = FWS_U;
1264 case FWS_U:
1265 if (fn == w->root)
1266 return 0;
1267 pn = fn->parent;
1268 w->node = pn;
1269 #ifdef CONFIG_IPV6_SUBTREES
1270 if (FIB6_SUBTREE(pn) == fn) {
1271 BUG_TRAP(fn->fn_flags&RTN_ROOT);
1272 w->state = FWS_L;
1273 continue;
1275 #endif
1276 if (pn->left == fn) {
1277 w->state = FWS_R;
1278 continue;
1280 if (pn->right == fn) {
1281 w->state = FWS_C;
1282 w->leaf = w->node->leaf;
1283 continue;
1285 #if RT6_DEBUG >= 2
1286 BUG_TRAP(0);
1287 #endif
1292 static int fib6_walk(struct fib6_walker_t *w)
1294 int res;
1296 w->state = FWS_INIT;
1297 w->node = w->root;
1299 fib6_walker_link(w);
1300 res = fib6_walk_continue(w);
1301 if (res <= 0)
1302 fib6_walker_unlink(w);
1303 return res;
1306 static int fib6_clean_node(struct fib6_walker_t *w)
1308 int res;
1309 struct rt6_info *rt;
1310 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1311 struct nl_info info = {
1312 .nl_net = c->net,
1315 for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) {
1316 res = c->func(rt, c->arg);
1317 if (res < 0) {
1318 w->leaf = rt;
1319 res = fib6_del(rt, &info);
1320 if (res) {
1321 #if RT6_DEBUG >= 2
1322 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res);
1323 #endif
1324 continue;
1326 return 0;
1328 BUG_TRAP(res==0);
1330 w->leaf = rt;
1331 return 0;
1335 * Convenient frontend to tree walker.
1337 * func is called on each route.
1338 * It may return -1 -> delete this route.
1339 * 0 -> continue walking
1341 * prune==1 -> only immediate children of node (certainly,
1342 * ignoring pure split nodes) will be scanned.
1345 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1346 int (*func)(struct rt6_info *, void *arg),
1347 int prune, void *arg)
1349 struct fib6_cleaner_t c;
1351 c.w.root = root;
1352 c.w.func = fib6_clean_node;
1353 c.w.prune = prune;
1354 c.func = func;
1355 c.arg = arg;
1356 c.net = net;
1358 fib6_walk(&c.w);
1361 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1362 int prune, void *arg)
1364 struct fib6_table *table;
1365 struct hlist_node *node;
1366 struct hlist_head *head;
1367 unsigned int h;
1369 rcu_read_lock();
1370 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
1371 head = &net->ipv6.fib_table_hash[h];
1372 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1373 write_lock_bh(&table->tb6_lock);
1374 fib6_clean_tree(net, &table->tb6_root,
1375 func, prune, arg);
1376 write_unlock_bh(&table->tb6_lock);
1379 rcu_read_unlock();
1382 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1384 if (rt->rt6i_flags & RTF_CACHE) {
1385 RT6_TRACE("pruning clone %p\n", rt);
1386 return -1;
1389 return 0;
1392 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1393 struct rt6_info *rt)
1395 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1399 * Garbage collection
1402 static struct fib6_gc_args
1404 int timeout;
1405 int more;
1406 } gc_args;
1408 static int fib6_age(struct rt6_info *rt, void *arg)
1410 unsigned long now = jiffies;
1413 * check addrconf expiration here.
1414 * Routes are expired even if they are in use.
1416 * Also age clones. Note, that clones are aged out
1417 * only if they are not in use now.
1420 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) {
1421 if (time_after(now, rt->rt6i_expires)) {
1422 RT6_TRACE("expiring %p\n", rt);
1423 return -1;
1425 gc_args.more++;
1426 } else if (rt->rt6i_flags & RTF_CACHE) {
1427 if (atomic_read(&rt->u.dst.__refcnt) == 0 &&
1428 time_after_eq(now, rt->u.dst.lastuse + gc_args.timeout)) {
1429 RT6_TRACE("aging clone %p\n", rt);
1430 return -1;
1431 } else if ((rt->rt6i_flags & RTF_GATEWAY) &&
1432 (!(rt->rt6i_nexthop->flags & NTF_ROUTER))) {
1433 RT6_TRACE("purging route %p via non-router but gateway\n",
1434 rt);
1435 return -1;
1437 gc_args.more++;
1440 return 0;
1443 static DEFINE_SPINLOCK(fib6_gc_lock);
1445 void fib6_run_gc(unsigned long expires, struct net *net)
1447 if (expires != ~0UL) {
1448 spin_lock_bh(&fib6_gc_lock);
1449 gc_args.timeout = expires ? (int)expires :
1450 net->ipv6.sysctl.ip6_rt_gc_interval;
1451 } else {
1452 local_bh_disable();
1453 if (!spin_trylock(&fib6_gc_lock)) {
1454 mod_timer(net->ipv6.ip6_fib_timer, jiffies + HZ);
1455 local_bh_enable();
1456 return;
1458 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1460 gc_args.more = 0;
1462 icmp6_dst_gc(&gc_args.more);
1464 fib6_clean_all(net, fib6_age, 0, NULL);
1466 if (gc_args.more)
1467 mod_timer(net->ipv6.ip6_fib_timer, jiffies +
1468 net->ipv6.sysctl.ip6_rt_gc_interval);
1469 else {
1470 del_timer(net->ipv6.ip6_fib_timer);
1471 net->ipv6.ip6_fib_timer->expires = 0;
1473 spin_unlock_bh(&fib6_gc_lock);
1476 static void fib6_gc_timer_cb(unsigned long arg)
1478 fib6_run_gc(0, (struct net *)arg);
1481 static int fib6_net_init(struct net *net)
1483 int ret;
1484 struct timer_list *timer;
1486 ret = -ENOMEM;
1487 timer = kzalloc(sizeof(*timer), GFP_KERNEL);
1488 if (!timer)
1489 goto out;
1491 setup_timer(timer, fib6_gc_timer_cb, (unsigned long)net);
1492 net->ipv6.ip6_fib_timer = timer;
1494 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1495 if (!net->ipv6.rt6_stats)
1496 goto out_timer;
1498 net->ipv6.fib_table_hash =
1499 kzalloc(sizeof(*net->ipv6.fib_table_hash)*FIB_TABLE_HASHSZ,
1500 GFP_KERNEL);
1501 if (!net->ipv6.fib_table_hash)
1502 goto out_rt6_stats;
1504 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1505 GFP_KERNEL);
1506 if (!net->ipv6.fib6_main_tbl)
1507 goto out_fib_table_hash;
1509 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1510 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1511 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1512 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1514 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1515 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1516 GFP_KERNEL);
1517 if (!net->ipv6.fib6_local_tbl)
1518 goto out_fib6_main_tbl;
1519 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1520 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1521 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1522 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1523 #endif
1524 fib6_tables_init(net);
1526 ret = 0;
1527 out:
1528 return ret;
1530 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1531 out_fib6_main_tbl:
1532 kfree(net->ipv6.fib6_main_tbl);
1533 #endif
1534 out_fib_table_hash:
1535 kfree(net->ipv6.fib_table_hash);
1536 out_rt6_stats:
1537 kfree(net->ipv6.rt6_stats);
1538 out_timer:
1539 kfree(timer);
1540 goto out;
1543 static void fib6_net_exit(struct net *net)
1545 rt6_ifdown(net, NULL);
1546 del_timer_sync(net->ipv6.ip6_fib_timer);
1547 kfree(net->ipv6.ip6_fib_timer);
1548 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1549 kfree(net->ipv6.fib6_local_tbl);
1550 #endif
1551 kfree(net->ipv6.fib6_main_tbl);
1552 kfree(net->ipv6.fib_table_hash);
1553 kfree(net->ipv6.rt6_stats);
1556 static struct pernet_operations fib6_net_ops = {
1557 .init = fib6_net_init,
1558 .exit = fib6_net_exit,
1561 int __init fib6_init(void)
1563 int ret = -ENOMEM;
1565 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1566 sizeof(struct fib6_node),
1567 0, SLAB_HWCACHE_ALIGN,
1568 NULL);
1569 if (!fib6_node_kmem)
1570 goto out;
1572 ret = register_pernet_subsys(&fib6_net_ops);
1573 if (ret)
1574 goto out_kmem_cache_create;
1576 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib);
1577 if (ret)
1578 goto out_unregister_subsys;
1579 out:
1580 return ret;
1582 out_unregister_subsys:
1583 unregister_pernet_subsys(&fib6_net_ops);
1584 out_kmem_cache_create:
1585 kmem_cache_destroy(fib6_node_kmem);
1586 goto out;
1589 void fib6_gc_cleanup(void)
1591 unregister_pernet_subsys(&fib6_net_ops);
1592 kmem_cache_destroy(fib6_node_kmem);