Linux 2.6.33.13
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / ipv6 / ip6_fib.c
blob0e93ca56eb694c39914c2295b9394c1e83775001
1 /*
2 * Linux INET6 implementation
3 * Forwarding Information Database
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
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.
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.
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>
30 #ifdef CONFIG_PROC_FS
31 #include <linux/proc_fs.h>
32 #endif
34 #include <net/ipv6.h>
35 #include <net/ndisc.h>
36 #include <net/addrconf.h>
38 #include <net/ip6_fib.h>
39 #include <net/ip6_route.h>
41 #define RT6_DEBUG 2
43 #if RT6_DEBUG >= 3
44 #define RT6_TRACE(x...) printk(KERN_DEBUG x)
45 #else
46 #define RT6_TRACE(x...) do { ; } while (0)
47 #endif
49 static struct kmem_cache * fib6_node_kmem __read_mostly;
51 enum fib_walk_state_t
53 #ifdef CONFIG_IPV6_SUBTREES
54 FWS_S,
55 #endif
56 FWS_L,
57 FWS_R,
58 FWS_C,
59 FWS_U
62 struct fib6_cleaner_t
64 struct fib6_walker_t w;
65 struct net *net;
66 int (*func)(struct rt6_info *, void *arg);
67 void *arg;
70 static DEFINE_RWLOCK(fib6_walker_lock);
72 #ifdef CONFIG_IPV6_SUBTREES
73 #define FWS_INIT FWS_S
74 #else
75 #define FWS_INIT FWS_L
76 #endif
78 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
79 struct rt6_info *rt);
80 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
81 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
82 static int fib6_walk(struct fib6_walker_t *w);
83 static int fib6_walk_continue(struct fib6_walker_t *w);
86 * A routing update causes an increase of the serial number on the
87 * affected subtree. This allows for cached routes to be asynchronously
88 * tested when modifications are made to the destination cache as a
89 * result of redirects, path MTU changes, etc.
92 static __u32 rt_sernum;
94 static void fib6_gc_timer_cb(unsigned long arg);
96 static struct fib6_walker_t fib6_walker_list = {
97 .prev = &fib6_walker_list,
98 .next = &fib6_walker_list,
101 #define FOR_WALKERS(w) for ((w)=fib6_walker_list.next; (w) != &fib6_walker_list; (w)=(w)->next)
103 static inline void fib6_walker_link(struct fib6_walker_t *w)
105 write_lock_bh(&fib6_walker_lock);
106 w->next = fib6_walker_list.next;
107 w->prev = &fib6_walker_list;
108 w->next->prev = w;
109 w->prev->next = w;
110 write_unlock_bh(&fib6_walker_lock);
113 static inline void fib6_walker_unlink(struct fib6_walker_t *w)
115 write_lock_bh(&fib6_walker_lock);
116 w->next->prev = w->prev;
117 w->prev->next = w->next;
118 w->prev = w->next = w;
119 write_unlock_bh(&fib6_walker_lock);
121 static __inline__ u32 fib6_new_sernum(void)
123 u32 n = ++rt_sernum;
124 if ((__s32)n <= 0)
125 rt_sernum = n = 1;
126 return n;
130 * Auxiliary address test functions for the radix tree.
132 * These assume a 32bit processor (although it will work on
133 * 64bit processors)
137 * test bit
140 static __inline__ __be32 addr_bit_set(void *token, int fn_bit)
142 __be32 *addr = token;
144 return htonl(1 << ((~fn_bit)&0x1F)) & addr[fn_bit>>5];
147 static __inline__ struct fib6_node * node_alloc(void)
149 struct fib6_node *fn;
151 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
153 return fn;
156 static __inline__ void node_free(struct fib6_node * fn)
158 kmem_cache_free(fib6_node_kmem, fn);
161 static __inline__ void rt6_release(struct rt6_info *rt)
163 if (atomic_dec_and_test(&rt->rt6i_ref))
164 dst_free(&rt->u.dst);
167 static void fib6_link_table(struct net *net, struct fib6_table *tb)
169 unsigned int h;
172 * Initialize table lock at a single place to give lockdep a key,
173 * tables aren't visible prior to being linked to the list.
175 rwlock_init(&tb->tb6_lock);
177 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
180 * No protection necessary, this is the only list mutatation
181 * operation, tables never disappear once they exist.
183 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
186 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
188 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
190 struct fib6_table *table;
192 table = kzalloc(sizeof(*table), GFP_ATOMIC);
193 if (table != NULL) {
194 table->tb6_id = id;
195 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
196 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
199 return table;
202 struct fib6_table *fib6_new_table(struct net *net, u32 id)
204 struct fib6_table *tb;
206 if (id == 0)
207 id = RT6_TABLE_MAIN;
208 tb = fib6_get_table(net, id);
209 if (tb)
210 return tb;
212 tb = fib6_alloc_table(net, id);
213 if (tb != NULL)
214 fib6_link_table(net, tb);
216 return tb;
219 struct fib6_table *fib6_get_table(struct net *net, u32 id)
221 struct fib6_table *tb;
222 struct hlist_head *head;
223 struct hlist_node *node;
224 unsigned int h;
226 if (id == 0)
227 id = RT6_TABLE_MAIN;
228 h = id & (FIB6_TABLE_HASHSZ - 1);
229 rcu_read_lock();
230 head = &net->ipv6.fib_table_hash[h];
231 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
232 if (tb->tb6_id == id) {
233 rcu_read_unlock();
234 return tb;
237 rcu_read_unlock();
239 return NULL;
242 static void fib6_tables_init(struct net *net)
244 fib6_link_table(net, net->ipv6.fib6_main_tbl);
245 fib6_link_table(net, net->ipv6.fib6_local_tbl);
247 #else
249 struct fib6_table *fib6_new_table(struct net *net, u32 id)
251 return fib6_get_table(net, id);
254 struct fib6_table *fib6_get_table(struct net *net, u32 id)
256 return net->ipv6.fib6_main_tbl;
259 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi *fl,
260 int flags, pol_lookup_t lookup)
262 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl, flags);
265 static void fib6_tables_init(struct net *net)
267 fib6_link_table(net, net->ipv6.fib6_main_tbl);
270 #endif
272 static int fib6_dump_node(struct fib6_walker_t *w)
274 int res;
275 struct rt6_info *rt;
277 for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) {
278 res = rt6_dump_route(rt, w->args);
279 if (res < 0) {
280 /* Frame is full, suspend walking */
281 w->leaf = rt;
282 return 1;
284 WARN_ON(res == 0);
286 w->leaf = NULL;
287 return 0;
290 static void fib6_dump_end(struct netlink_callback *cb)
292 struct fib6_walker_t *w = (void*)cb->args[2];
294 if (w) {
295 if (cb->args[4]) {
296 cb->args[4] = 0;
297 fib6_walker_unlink(w);
299 cb->args[2] = 0;
300 kfree(w);
302 cb->done = (void*)cb->args[3];
303 cb->args[1] = 3;
306 static int fib6_dump_done(struct netlink_callback *cb)
308 fib6_dump_end(cb);
309 return cb->done ? cb->done(cb) : 0;
312 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
313 struct netlink_callback *cb)
315 struct fib6_walker_t *w;
316 int res;
318 w = (void *)cb->args[2];
319 w->root = &table->tb6_root;
321 if (cb->args[4] == 0) {
322 read_lock_bh(&table->tb6_lock);
323 res = fib6_walk(w);
324 read_unlock_bh(&table->tb6_lock);
325 if (res > 0)
326 cb->args[4] = 1;
327 } else {
328 read_lock_bh(&table->tb6_lock);
329 res = fib6_walk_continue(w);
330 read_unlock_bh(&table->tb6_lock);
331 if (res <= 0) {
332 fib6_walker_unlink(w);
333 cb->args[4] = 0;
337 return res;
340 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
342 struct net *net = sock_net(skb->sk);
343 unsigned int h, s_h;
344 unsigned int e = 0, s_e;
345 struct rt6_rtnl_dump_arg arg;
346 struct fib6_walker_t *w;
347 struct fib6_table *tb;
348 struct hlist_node *node;
349 struct hlist_head *head;
350 int res = 0;
352 s_h = cb->args[0];
353 s_e = cb->args[1];
355 w = (void *)cb->args[2];
356 if (w == NULL) {
357 /* New dump:
359 * 1. hook callback destructor.
361 cb->args[3] = (long)cb->done;
362 cb->done = fib6_dump_done;
365 * 2. allocate and initialize walker.
367 w = kzalloc(sizeof(*w), GFP_ATOMIC);
368 if (w == NULL)
369 return -ENOMEM;
370 w->func = fib6_dump_node;
371 cb->args[2] = (long)w;
374 arg.skb = skb;
375 arg.cb = cb;
376 arg.net = net;
377 w->args = &arg;
379 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
380 e = 0;
381 head = &net->ipv6.fib_table_hash[h];
382 hlist_for_each_entry(tb, node, head, tb6_hlist) {
383 if (e < s_e)
384 goto next;
385 res = fib6_dump_table(tb, skb, cb);
386 if (res != 0)
387 goto out;
388 next:
389 e++;
392 out:
393 cb->args[1] = e;
394 cb->args[0] = h;
396 res = res < 0 ? res : skb->len;
397 if (res <= 0)
398 fib6_dump_end(cb);
399 return res;
403 * Routing Table
405 * return the appropriate node for a routing tree "add" operation
406 * by either creating and inserting or by returning an existing
407 * node.
410 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
411 int addrlen, int plen,
412 int offset)
414 struct fib6_node *fn, *in, *ln;
415 struct fib6_node *pn = NULL;
416 struct rt6key *key;
417 int bit;
418 __be32 dir = 0;
419 __u32 sernum = fib6_new_sernum();
421 RT6_TRACE("fib6_add_1\n");
423 /* insert node in tree */
425 fn = root;
427 do {
428 key = (struct rt6key *)((u8 *)fn->leaf + offset);
431 * Prefix match
433 if (plen < fn->fn_bit ||
434 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
435 goto insert_above;
438 * Exact match ?
441 if (plen == fn->fn_bit) {
442 /* clean up an intermediate node */
443 if ((fn->fn_flags & RTN_RTINFO) == 0) {
444 rt6_release(fn->leaf);
445 fn->leaf = NULL;
448 fn->fn_sernum = sernum;
450 return fn;
454 * We have more bits to go
457 /* Try to walk down on tree. */
458 fn->fn_sernum = sernum;
459 dir = addr_bit_set(addr, fn->fn_bit);
460 pn = fn;
461 fn = dir ? fn->right: fn->left;
462 } while (fn);
465 * We walked to the bottom of tree.
466 * Create new leaf node without children.
469 ln = node_alloc();
471 if (ln == NULL)
472 return NULL;
473 ln->fn_bit = plen;
475 ln->parent = pn;
476 ln->fn_sernum = sernum;
478 if (dir)
479 pn->right = ln;
480 else
481 pn->left = ln;
483 return ln;
486 insert_above:
488 * split since we don't have a common prefix anymore or
489 * we have a less significant route.
490 * we've to insert an intermediate node on the list
491 * this new node will point to the one we need to create
492 * and the current
495 pn = fn->parent;
497 /* find 1st bit in difference between the 2 addrs.
499 See comment in __ipv6_addr_diff: bit may be an invalid value,
500 but if it is >= plen, the value is ignored in any case.
503 bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
506 * (intermediate)[in]
507 * / \
508 * (new leaf node)[ln] (old node)[fn]
510 if (plen > bit) {
511 in = node_alloc();
512 ln = node_alloc();
514 if (in == NULL || ln == NULL) {
515 if (in)
516 node_free(in);
517 if (ln)
518 node_free(ln);
519 return NULL;
523 * new intermediate node.
524 * RTN_RTINFO will
525 * be off since that an address that chooses one of
526 * the branches would not match less specific routes
527 * in the other branch
530 in->fn_bit = bit;
532 in->parent = pn;
533 in->leaf = fn->leaf;
534 atomic_inc(&in->leaf->rt6i_ref);
536 in->fn_sernum = sernum;
538 /* update parent pointer */
539 if (dir)
540 pn->right = in;
541 else
542 pn->left = in;
544 ln->fn_bit = plen;
546 ln->parent = in;
547 fn->parent = in;
549 ln->fn_sernum = sernum;
551 if (addr_bit_set(addr, bit)) {
552 in->right = ln;
553 in->left = fn;
554 } else {
555 in->left = ln;
556 in->right = fn;
558 } else { /* plen <= bit */
561 * (new leaf node)[ln]
562 * / \
563 * (old node)[fn] NULL
566 ln = node_alloc();
568 if (ln == NULL)
569 return NULL;
571 ln->fn_bit = plen;
573 ln->parent = pn;
575 ln->fn_sernum = sernum;
577 if (dir)
578 pn->right = ln;
579 else
580 pn->left = ln;
582 if (addr_bit_set(&key->addr, plen))
583 ln->right = fn;
584 else
585 ln->left = fn;
587 fn->parent = ln;
589 return ln;
593 * Insert routing information in a node.
596 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
597 struct nl_info *info)
599 struct rt6_info *iter = NULL;
600 struct rt6_info **ins;
602 ins = &fn->leaf;
604 for (iter = fn->leaf; iter; iter=iter->u.dst.rt6_next) {
606 * Search for duplicates
609 if (iter->rt6i_metric == rt->rt6i_metric) {
611 * Same priority level
614 if (iter->rt6i_dev == rt->rt6i_dev &&
615 iter->rt6i_idev == rt->rt6i_idev &&
616 ipv6_addr_equal(&iter->rt6i_gateway,
617 &rt->rt6i_gateway)) {
618 if (!(iter->rt6i_flags&RTF_EXPIRES))
619 return -EEXIST;
620 iter->rt6i_expires = rt->rt6i_expires;
621 if (!(rt->rt6i_flags&RTF_EXPIRES)) {
622 iter->rt6i_flags &= ~RTF_EXPIRES;
623 iter->rt6i_expires = 0;
625 return -EEXIST;
629 if (iter->rt6i_metric > rt->rt6i_metric)
630 break;
632 ins = &iter->u.dst.rt6_next;
635 /* Reset round-robin state, if necessary */
636 if (ins == &fn->leaf)
637 fn->rr_ptr = NULL;
640 * insert node
643 rt->u.dst.rt6_next = iter;
644 *ins = rt;
645 rt->rt6i_node = fn;
646 atomic_inc(&rt->rt6i_ref);
647 inet6_rt_notify(RTM_NEWROUTE, rt, info);
648 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
650 if ((fn->fn_flags & RTN_RTINFO) == 0) {
651 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
652 fn->fn_flags |= RTN_RTINFO;
655 return 0;
658 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
660 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
661 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE)))
662 mod_timer(&net->ipv6.ip6_fib_timer,
663 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
666 void fib6_force_start_gc(struct net *net)
668 if (!timer_pending(&net->ipv6.ip6_fib_timer))
669 mod_timer(&net->ipv6.ip6_fib_timer,
670 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
674 * Add routing information to the routing tree.
675 * <destination addr>/<source addr>
676 * with source addr info in sub-trees
679 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
681 struct fib6_node *fn, *pn = NULL;
682 int err = -ENOMEM;
684 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
685 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst));
687 if (fn == NULL)
688 goto out;
690 pn = fn;
692 #ifdef CONFIG_IPV6_SUBTREES
693 if (rt->rt6i_src.plen) {
694 struct fib6_node *sn;
696 if (fn->subtree == NULL) {
697 struct fib6_node *sfn;
700 * Create subtree.
702 * fn[main tree]
704 * sfn[subtree root]
706 * sn[new leaf node]
709 /* Create subtree root node */
710 sfn = node_alloc();
711 if (sfn == NULL)
712 goto st_failure;
714 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
715 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
716 sfn->fn_flags = RTN_ROOT;
717 sfn->fn_sernum = fib6_new_sernum();
719 /* Now add the first leaf node to new subtree */
721 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
722 sizeof(struct in6_addr), rt->rt6i_src.plen,
723 offsetof(struct rt6_info, rt6i_src));
725 if (sn == NULL) {
726 /* If it is failed, discard just allocated
727 root, and then (in st_failure) stale node
728 in main tree.
730 node_free(sfn);
731 goto st_failure;
734 /* Now link new subtree to main tree */
735 sfn->parent = fn;
736 fn->subtree = sfn;
737 } else {
738 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
739 sizeof(struct in6_addr), rt->rt6i_src.plen,
740 offsetof(struct rt6_info, rt6i_src));
742 if (sn == NULL)
743 goto st_failure;
746 if (fn->leaf == NULL) {
747 fn->leaf = rt;
748 atomic_inc(&rt->rt6i_ref);
750 fn = sn;
752 #endif
754 err = fib6_add_rt2node(fn, rt, info);
756 if (err == 0) {
757 fib6_start_gc(info->nl_net, rt);
758 if (!(rt->rt6i_flags&RTF_CACHE))
759 fib6_prune_clones(info->nl_net, pn, rt);
762 out:
763 if (err) {
764 #ifdef CONFIG_IPV6_SUBTREES
766 * If fib6_add_1 has cleared the old leaf pointer in the
767 * super-tree leaf node we have to find a new one for it.
769 if (pn != fn && pn->leaf == rt) {
770 pn->leaf = NULL;
771 atomic_dec(&rt->rt6i_ref);
773 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
774 pn->leaf = fib6_find_prefix(info->nl_net, pn);
775 #if RT6_DEBUG >= 2
776 if (!pn->leaf) {
777 WARN_ON(pn->leaf == NULL);
778 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
780 #endif
781 atomic_inc(&pn->leaf->rt6i_ref);
783 #endif
784 dst_free(&rt->u.dst);
786 return err;
788 #ifdef CONFIG_IPV6_SUBTREES
789 /* Subtree creation failed, probably main tree node
790 is orphan. If it is, shoot it.
792 st_failure:
793 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
794 fib6_repair_tree(info->nl_net, fn);
795 dst_free(&rt->u.dst);
796 return err;
797 #endif
801 * Routing tree lookup
805 struct lookup_args {
806 int offset; /* key offset on rt6_info */
807 struct in6_addr *addr; /* search key */
810 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
811 struct lookup_args *args)
813 struct fib6_node *fn;
814 __be32 dir;
816 if (unlikely(args->offset == 0))
817 return NULL;
820 * Descend on a tree
823 fn = root;
825 for (;;) {
826 struct fib6_node *next;
828 dir = addr_bit_set(args->addr, fn->fn_bit);
830 next = dir ? fn->right : fn->left;
832 if (next) {
833 fn = next;
834 continue;
837 break;
840 while(fn) {
841 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
842 struct rt6key *key;
844 key = (struct rt6key *) ((u8 *) fn->leaf +
845 args->offset);
847 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
848 #ifdef CONFIG_IPV6_SUBTREES
849 if (fn->subtree)
850 fn = fib6_lookup_1(fn->subtree, args + 1);
851 #endif
852 if (!fn || fn->fn_flags & RTN_RTINFO)
853 return fn;
857 if (fn->fn_flags & RTN_ROOT)
858 break;
860 fn = fn->parent;
863 return NULL;
866 struct fib6_node * fib6_lookup(struct fib6_node *root, struct in6_addr *daddr,
867 struct in6_addr *saddr)
869 struct fib6_node *fn;
870 struct lookup_args args[] = {
872 .offset = offsetof(struct rt6_info, rt6i_dst),
873 .addr = daddr,
875 #ifdef CONFIG_IPV6_SUBTREES
877 .offset = offsetof(struct rt6_info, rt6i_src),
878 .addr = saddr,
880 #endif
882 .offset = 0, /* sentinel */
886 fn = fib6_lookup_1(root, daddr ? args : args + 1);
888 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT)
889 fn = root;
891 return fn;
895 * Get node with specified destination prefix (and source prefix,
896 * if subtrees are used)
900 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
901 struct in6_addr *addr,
902 int plen, int offset)
904 struct fib6_node *fn;
906 for (fn = root; fn ; ) {
907 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
910 * Prefix match
912 if (plen < fn->fn_bit ||
913 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
914 return NULL;
916 if (plen == fn->fn_bit)
917 return fn;
920 * We have more bits to go
922 if (addr_bit_set(addr, fn->fn_bit))
923 fn = fn->right;
924 else
925 fn = fn->left;
927 return NULL;
930 struct fib6_node * fib6_locate(struct fib6_node *root,
931 struct in6_addr *daddr, int dst_len,
932 struct in6_addr *saddr, int src_len)
934 struct fib6_node *fn;
936 fn = fib6_locate_1(root, daddr, dst_len,
937 offsetof(struct rt6_info, rt6i_dst));
939 #ifdef CONFIG_IPV6_SUBTREES
940 if (src_len) {
941 WARN_ON(saddr == NULL);
942 if (fn && fn->subtree)
943 fn = fib6_locate_1(fn->subtree, saddr, src_len,
944 offsetof(struct rt6_info, rt6i_src));
946 #endif
948 if (fn && fn->fn_flags&RTN_RTINFO)
949 return fn;
951 return NULL;
956 * Deletion
960 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
962 if (fn->fn_flags&RTN_ROOT)
963 return net->ipv6.ip6_null_entry;
965 while(fn) {
966 if(fn->left)
967 return fn->left->leaf;
969 if(fn->right)
970 return fn->right->leaf;
972 fn = FIB6_SUBTREE(fn);
974 return NULL;
978 * Called to trim the tree of intermediate nodes when possible. "fn"
979 * is the node we want to try and remove.
982 static struct fib6_node *fib6_repair_tree(struct net *net,
983 struct fib6_node *fn)
985 int children;
986 int nstate;
987 struct fib6_node *child, *pn;
988 struct fib6_walker_t *w;
989 int iter = 0;
991 for (;;) {
992 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
993 iter++;
995 WARN_ON(fn->fn_flags & RTN_RTINFO);
996 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
997 WARN_ON(fn->leaf != NULL);
999 children = 0;
1000 child = NULL;
1001 if (fn->right) child = fn->right, children |= 1;
1002 if (fn->left) child = fn->left, children |= 2;
1004 if (children == 3 || FIB6_SUBTREE(fn)
1005 #ifdef CONFIG_IPV6_SUBTREES
1006 /* Subtree root (i.e. fn) may have one child */
1007 || (children && fn->fn_flags&RTN_ROOT)
1008 #endif
1010 fn->leaf = fib6_find_prefix(net, fn);
1011 #if RT6_DEBUG >= 2
1012 if (fn->leaf==NULL) {
1013 WARN_ON(!fn->leaf);
1014 fn->leaf = net->ipv6.ip6_null_entry;
1016 #endif
1017 atomic_inc(&fn->leaf->rt6i_ref);
1018 return fn->parent;
1021 pn = fn->parent;
1022 #ifdef CONFIG_IPV6_SUBTREES
1023 if (FIB6_SUBTREE(pn) == fn) {
1024 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1025 FIB6_SUBTREE(pn) = NULL;
1026 nstate = FWS_L;
1027 } else {
1028 WARN_ON(fn->fn_flags & RTN_ROOT);
1029 #endif
1030 if (pn->right == fn) pn->right = child;
1031 else if (pn->left == fn) pn->left = child;
1032 #if RT6_DEBUG >= 2
1033 else
1034 WARN_ON(1);
1035 #endif
1036 if (child)
1037 child->parent = pn;
1038 nstate = FWS_R;
1039 #ifdef CONFIG_IPV6_SUBTREES
1041 #endif
1043 read_lock(&fib6_walker_lock);
1044 FOR_WALKERS(w) {
1045 if (child == NULL) {
1046 if (w->root == fn) {
1047 w->root = w->node = NULL;
1048 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1049 } else if (w->node == fn) {
1050 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1051 w->node = pn;
1052 w->state = nstate;
1054 } else {
1055 if (w->root == fn) {
1056 w->root = child;
1057 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1059 if (w->node == fn) {
1060 w->node = child;
1061 if (children&2) {
1062 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1063 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1064 } else {
1065 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1066 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1071 read_unlock(&fib6_walker_lock);
1073 node_free(fn);
1074 if (pn->fn_flags&RTN_RTINFO || FIB6_SUBTREE(pn))
1075 return pn;
1077 rt6_release(pn->leaf);
1078 pn->leaf = NULL;
1079 fn = pn;
1083 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1084 struct nl_info *info)
1086 struct fib6_walker_t *w;
1087 struct rt6_info *rt = *rtp;
1088 struct net *net = info->nl_net;
1090 RT6_TRACE("fib6_del_route\n");
1092 /* Unlink it */
1093 *rtp = rt->u.dst.rt6_next;
1094 rt->rt6i_node = NULL;
1095 net->ipv6.rt6_stats->fib_rt_entries--;
1096 net->ipv6.rt6_stats->fib_discarded_routes++;
1098 /* Reset round-robin state, if necessary */
1099 if (fn->rr_ptr == rt)
1100 fn->rr_ptr = NULL;
1102 /* Adjust walkers */
1103 read_lock(&fib6_walker_lock);
1104 FOR_WALKERS(w) {
1105 if (w->state == FWS_C && w->leaf == rt) {
1106 RT6_TRACE("walker %p adjusted by delroute\n", w);
1107 w->leaf = rt->u.dst.rt6_next;
1108 if (w->leaf == NULL)
1109 w->state = FWS_U;
1112 read_unlock(&fib6_walker_lock);
1114 rt->u.dst.rt6_next = NULL;
1116 /* If it was last route, expunge its radix tree node */
1117 if (fn->leaf == NULL) {
1118 fn->fn_flags &= ~RTN_RTINFO;
1119 net->ipv6.rt6_stats->fib_route_nodes--;
1120 fn = fib6_repair_tree(net, fn);
1123 if (atomic_read(&rt->rt6i_ref) != 1) {
1124 /* This route is used as dummy address holder in some split
1125 * nodes. It is not leaked, but it still holds other resources,
1126 * which must be released in time. So, scan ascendant nodes
1127 * and replace dummy references to this route with references
1128 * to still alive ones.
1130 while (fn) {
1131 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) {
1132 fn->leaf = fib6_find_prefix(net, fn);
1133 atomic_inc(&fn->leaf->rt6i_ref);
1134 rt6_release(rt);
1136 fn = fn->parent;
1138 /* No more references are possible at this point. */
1139 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1142 inet6_rt_notify(RTM_DELROUTE, rt, info);
1143 rt6_release(rt);
1146 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1148 struct net *net = info->nl_net;
1149 struct fib6_node *fn = rt->rt6i_node;
1150 struct rt6_info **rtp;
1152 #if RT6_DEBUG >= 2
1153 if (rt->u.dst.obsolete>0) {
1154 WARN_ON(fn != NULL);
1155 return -ENOENT;
1157 #endif
1158 if (fn == NULL || rt == net->ipv6.ip6_null_entry)
1159 return -ENOENT;
1161 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1163 if (!(rt->rt6i_flags&RTF_CACHE)) {
1164 struct fib6_node *pn = fn;
1165 #ifdef CONFIG_IPV6_SUBTREES
1166 /* clones of this route might be in another subtree */
1167 if (rt->rt6i_src.plen) {
1168 while (!(pn->fn_flags&RTN_ROOT))
1169 pn = pn->parent;
1170 pn = pn->parent;
1172 #endif
1173 fib6_prune_clones(info->nl_net, pn, rt);
1177 * Walk the leaf entries looking for ourself
1180 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->u.dst.rt6_next) {
1181 if (*rtp == rt) {
1182 fib6_del_route(fn, rtp, info);
1183 return 0;
1186 return -ENOENT;
1190 * Tree traversal function.
1192 * Certainly, it is not interrupt safe.
1193 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1194 * It means, that we can modify tree during walking
1195 * and use this function for garbage collection, clone pruning,
1196 * cleaning tree when a device goes down etc. etc.
1198 * It guarantees that every node will be traversed,
1199 * and that it will be traversed only once.
1201 * Callback function w->func may return:
1202 * 0 -> continue walking.
1203 * positive value -> walking is suspended (used by tree dumps,
1204 * and probably by gc, if it will be split to several slices)
1205 * negative value -> terminate walking.
1207 * The function itself returns:
1208 * 0 -> walk is complete.
1209 * >0 -> walk is incomplete (i.e. suspended)
1210 * <0 -> walk is terminated by an error.
1213 static int fib6_walk_continue(struct fib6_walker_t *w)
1215 struct fib6_node *fn, *pn;
1217 for (;;) {
1218 fn = w->node;
1219 if (fn == NULL)
1220 return 0;
1222 if (w->prune && fn != w->root &&
1223 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) {
1224 w->state = FWS_C;
1225 w->leaf = fn->leaf;
1227 switch (w->state) {
1228 #ifdef CONFIG_IPV6_SUBTREES
1229 case FWS_S:
1230 if (FIB6_SUBTREE(fn)) {
1231 w->node = FIB6_SUBTREE(fn);
1232 continue;
1234 w->state = FWS_L;
1235 #endif
1236 case FWS_L:
1237 if (fn->left) {
1238 w->node = fn->left;
1239 w->state = FWS_INIT;
1240 continue;
1242 w->state = FWS_R;
1243 case FWS_R:
1244 if (fn->right) {
1245 w->node = fn->right;
1246 w->state = FWS_INIT;
1247 continue;
1249 w->state = FWS_C;
1250 w->leaf = fn->leaf;
1251 case FWS_C:
1252 if (w->leaf && fn->fn_flags&RTN_RTINFO) {
1253 int err = w->func(w);
1254 if (err)
1255 return err;
1256 continue;
1258 w->state = FWS_U;
1259 case FWS_U:
1260 if (fn == w->root)
1261 return 0;
1262 pn = fn->parent;
1263 w->node = pn;
1264 #ifdef CONFIG_IPV6_SUBTREES
1265 if (FIB6_SUBTREE(pn) == fn) {
1266 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1267 w->state = FWS_L;
1268 continue;
1270 #endif
1271 if (pn->left == fn) {
1272 w->state = FWS_R;
1273 continue;
1275 if (pn->right == fn) {
1276 w->state = FWS_C;
1277 w->leaf = w->node->leaf;
1278 continue;
1280 #if RT6_DEBUG >= 2
1281 WARN_ON(1);
1282 #endif
1287 static int fib6_walk(struct fib6_walker_t *w)
1289 int res;
1291 w->state = FWS_INIT;
1292 w->node = w->root;
1294 fib6_walker_link(w);
1295 res = fib6_walk_continue(w);
1296 if (res <= 0)
1297 fib6_walker_unlink(w);
1298 return res;
1301 static int fib6_clean_node(struct fib6_walker_t *w)
1303 int res;
1304 struct rt6_info *rt;
1305 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1306 struct nl_info info = {
1307 .nl_net = c->net,
1310 for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) {
1311 res = c->func(rt, c->arg);
1312 if (res < 0) {
1313 w->leaf = rt;
1314 res = fib6_del(rt, &info);
1315 if (res) {
1316 #if RT6_DEBUG >= 2
1317 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res);
1318 #endif
1319 continue;
1321 return 0;
1323 WARN_ON(res != 0);
1325 w->leaf = rt;
1326 return 0;
1330 * Convenient frontend to tree walker.
1332 * func is called on each route.
1333 * It may return -1 -> delete this route.
1334 * 0 -> continue walking
1336 * prune==1 -> only immediate children of node (certainly,
1337 * ignoring pure split nodes) will be scanned.
1340 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1341 int (*func)(struct rt6_info *, void *arg),
1342 int prune, void *arg)
1344 struct fib6_cleaner_t c;
1346 c.w.root = root;
1347 c.w.func = fib6_clean_node;
1348 c.w.prune = prune;
1349 c.func = func;
1350 c.arg = arg;
1351 c.net = net;
1353 fib6_walk(&c.w);
1356 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1357 int prune, void *arg)
1359 struct fib6_table *table;
1360 struct hlist_node *node;
1361 struct hlist_head *head;
1362 unsigned int h;
1364 rcu_read_lock();
1365 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1366 head = &net->ipv6.fib_table_hash[h];
1367 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1368 write_lock_bh(&table->tb6_lock);
1369 fib6_clean_tree(net, &table->tb6_root,
1370 func, prune, arg);
1371 write_unlock_bh(&table->tb6_lock);
1374 rcu_read_unlock();
1377 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1379 if (rt->rt6i_flags & RTF_CACHE) {
1380 RT6_TRACE("pruning clone %p\n", rt);
1381 return -1;
1384 return 0;
1387 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1388 struct rt6_info *rt)
1390 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1394 * Garbage collection
1397 static struct fib6_gc_args
1399 int timeout;
1400 int more;
1401 } gc_args;
1403 static int fib6_age(struct rt6_info *rt, void *arg)
1405 unsigned long now = jiffies;
1408 * check addrconf expiration here.
1409 * Routes are expired even if they are in use.
1411 * Also age clones. Note, that clones are aged out
1412 * only if they are not in use now.
1415 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) {
1416 if (time_after(now, rt->rt6i_expires)) {
1417 RT6_TRACE("expiring %p\n", rt);
1418 return -1;
1420 gc_args.more++;
1421 } else if (rt->rt6i_flags & RTF_CACHE) {
1422 if (atomic_read(&rt->u.dst.__refcnt) == 0 &&
1423 time_after_eq(now, rt->u.dst.lastuse + gc_args.timeout)) {
1424 RT6_TRACE("aging clone %p\n", rt);
1425 return -1;
1426 } else if ((rt->rt6i_flags & RTF_GATEWAY) &&
1427 (!(rt->rt6i_nexthop->flags & NTF_ROUTER))) {
1428 RT6_TRACE("purging route %p via non-router but gateway\n",
1429 rt);
1430 return -1;
1432 gc_args.more++;
1435 return 0;
1438 static DEFINE_SPINLOCK(fib6_gc_lock);
1440 void fib6_run_gc(unsigned long expires, struct net *net)
1442 if (expires != ~0UL) {
1443 spin_lock_bh(&fib6_gc_lock);
1444 gc_args.timeout = expires ? (int)expires :
1445 net->ipv6.sysctl.ip6_rt_gc_interval;
1446 } else {
1447 if (!spin_trylock_bh(&fib6_gc_lock)) {
1448 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1449 return;
1451 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1454 gc_args.more = icmp6_dst_gc();
1456 fib6_clean_all(net, fib6_age, 0, NULL);
1458 if (gc_args.more)
1459 mod_timer(&net->ipv6.ip6_fib_timer,
1460 round_jiffies(jiffies
1461 + net->ipv6.sysctl.ip6_rt_gc_interval));
1462 else
1463 del_timer(&net->ipv6.ip6_fib_timer);
1464 spin_unlock_bh(&fib6_gc_lock);
1467 static void fib6_gc_timer_cb(unsigned long arg)
1469 fib6_run_gc(0, (struct net *)arg);
1472 static int fib6_net_init(struct net *net)
1474 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1476 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1477 if (!net->ipv6.rt6_stats)
1478 goto out_timer;
1480 net->ipv6.fib_table_hash = kcalloc(FIB6_TABLE_HASHSZ,
1481 sizeof(*net->ipv6.fib_table_hash),
1482 GFP_KERNEL);
1483 if (!net->ipv6.fib_table_hash)
1484 goto out_rt6_stats;
1486 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1487 GFP_KERNEL);
1488 if (!net->ipv6.fib6_main_tbl)
1489 goto out_fib_table_hash;
1491 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1492 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1493 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1494 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1496 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1497 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1498 GFP_KERNEL);
1499 if (!net->ipv6.fib6_local_tbl)
1500 goto out_fib6_main_tbl;
1501 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1502 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1503 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1504 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1505 #endif
1506 fib6_tables_init(net);
1508 return 0;
1510 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1511 out_fib6_main_tbl:
1512 kfree(net->ipv6.fib6_main_tbl);
1513 #endif
1514 out_fib_table_hash:
1515 kfree(net->ipv6.fib_table_hash);
1516 out_rt6_stats:
1517 kfree(net->ipv6.rt6_stats);
1518 out_timer:
1519 return -ENOMEM;
1522 static void fib6_net_exit(struct net *net)
1524 rt6_ifdown(net, NULL);
1525 del_timer_sync(&net->ipv6.ip6_fib_timer);
1527 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1528 kfree(net->ipv6.fib6_local_tbl);
1529 #endif
1530 kfree(net->ipv6.fib6_main_tbl);
1531 kfree(net->ipv6.fib_table_hash);
1532 kfree(net->ipv6.rt6_stats);
1535 static struct pernet_operations fib6_net_ops = {
1536 .init = fib6_net_init,
1537 .exit = fib6_net_exit,
1540 int __init fib6_init(void)
1542 int ret = -ENOMEM;
1544 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1545 sizeof(struct fib6_node),
1546 0, SLAB_HWCACHE_ALIGN,
1547 NULL);
1548 if (!fib6_node_kmem)
1549 goto out;
1551 ret = register_pernet_subsys(&fib6_net_ops);
1552 if (ret)
1553 goto out_kmem_cache_create;
1555 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib);
1556 if (ret)
1557 goto out_unregister_subsys;
1558 out:
1559 return ret;
1561 out_unregister_subsys:
1562 unregister_pernet_subsys(&fib6_net_ops);
1563 out_kmem_cache_create:
1564 kmem_cache_destroy(fib6_node_kmem);
1565 goto out;
1568 void fib6_gc_cleanup(void)
1570 unregister_pernet_subsys(&fib6_net_ops);
1571 kmem_cache_destroy(fib6_node_kmem);