Merge remote-tracking branch 'sysctl/master'
[linux-2.6/next.git] / net / ipv6 / ip6_fib.c
blob320d91d20ad7ddfd981cee92161482f218c225b4
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>
29 #include <linux/slab.h>
31 #ifdef CONFIG_PROC_FS
32 #include <linux/proc_fs.h>
33 #endif
35 #include <net/ipv6.h>
36 #include <net/ndisc.h>
37 #include <net/addrconf.h>
39 #include <net/ip6_fib.h>
40 #include <net/ip6_route.h>
42 #define RT6_DEBUG 2
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
50 static struct kmem_cache * fib6_node_kmem __read_mostly;
52 enum fib_walk_state_t
54 #ifdef CONFIG_IPV6_SUBTREES
55 FWS_S,
56 #endif
57 FWS_L,
58 FWS_R,
59 FWS_C,
60 FWS_U
63 struct fib6_cleaner_t
65 struct fib6_walker_t w;
66 struct net *net;
67 int (*func)(struct rt6_info *, void *arg);
68 void *arg;
71 static DEFINE_RWLOCK(fib6_walker_lock);
73 #ifdef CONFIG_IPV6_SUBTREES
74 #define FWS_INIT FWS_S
75 #else
76 #define FWS_INIT FWS_L
77 #endif
79 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
80 struct rt6_info *rt);
81 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
82 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
83 static int fib6_walk(struct fib6_walker_t *w);
84 static int fib6_walk_continue(struct fib6_walker_t *w);
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.
93 static __u32 rt_sernum;
95 static void fib6_gc_timer_cb(unsigned long arg);
97 static LIST_HEAD(fib6_walkers);
98 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
100 static inline void fib6_walker_link(struct fib6_walker_t *w)
102 write_lock_bh(&fib6_walker_lock);
103 list_add(&w->lh, &fib6_walkers);
104 write_unlock_bh(&fib6_walker_lock);
107 static inline void fib6_walker_unlink(struct fib6_walker_t *w)
109 write_lock_bh(&fib6_walker_lock);
110 list_del(&w->lh);
111 write_unlock_bh(&fib6_walker_lock);
113 static __inline__ u32 fib6_new_sernum(void)
115 u32 n = ++rt_sernum;
116 if ((__s32)n <= 0)
117 rt_sernum = n = 1;
118 return n;
122 * Auxiliary address test functions for the radix tree.
124 * These assume a 32bit processor (although it will work on
125 * 64bit processors)
129 * test bit
131 #if defined(__LITTLE_ENDIAN)
132 # define BITOP_BE32_SWIZZLE (0x1F & ~7)
133 #else
134 # define BITOP_BE32_SWIZZLE 0
135 #endif
137 static __inline__ __be32 addr_bit_set(const void *token, int fn_bit)
139 const __be32 *addr = token;
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.
147 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
148 addr[fn_bit >> 5];
151 static __inline__ struct fib6_node * node_alloc(void)
153 struct fib6_node *fn;
155 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
157 return fn;
160 static __inline__ void node_free(struct fib6_node * fn)
162 kmem_cache_free(fib6_node_kmem, fn);
165 static __inline__ void rt6_release(struct rt6_info *rt)
167 if (atomic_dec_and_test(&rt->rt6i_ref))
168 dst_free(&rt->dst);
171 static void fib6_link_table(struct net *net, struct fib6_table *tb)
173 unsigned int h;
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.
179 rwlock_init(&tb->tb6_lock);
181 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
184 * No protection necessary, this is the only list mutatation
185 * operation, tables never disappear once they exist.
187 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
190 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
192 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
194 struct fib6_table *table;
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;
203 return table;
206 struct fib6_table *fib6_new_table(struct net *net, u32 id)
208 struct fib6_table *tb;
210 if (id == 0)
211 id = RT6_TABLE_MAIN;
212 tb = fib6_get_table(net, id);
213 if (tb)
214 return tb;
216 tb = fib6_alloc_table(net, id);
217 if (tb != NULL)
218 fib6_link_table(net, tb);
220 return tb;
223 struct fib6_table *fib6_get_table(struct net *net, u32 id)
225 struct fib6_table *tb;
226 struct hlist_head *head;
227 struct hlist_node *node;
228 unsigned int h;
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;
241 rcu_read_unlock();
243 return NULL;
246 static void __net_init fib6_tables_init(struct net *net)
248 fib6_link_table(net, net->ipv6.fib6_main_tbl);
249 fib6_link_table(net, net->ipv6.fib6_local_tbl);
251 #else
253 struct fib6_table *fib6_new_table(struct net *net, u32 id)
255 return fib6_get_table(net, id);
258 struct fib6_table *fib6_get_table(struct net *net, u32 id)
260 return net->ipv6.fib6_main_tbl;
263 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
264 int flags, pol_lookup_t lookup)
266 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
269 static void __net_init fib6_tables_init(struct net *net)
271 fib6_link_table(net, net->ipv6.fib6_main_tbl);
274 #endif
276 static int fib6_dump_node(struct fib6_walker_t *w)
278 int res;
279 struct rt6_info *rt;
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;
288 WARN_ON(res == 0);
290 w->leaf = NULL;
291 return 0;
294 static void fib6_dump_end(struct netlink_callback *cb)
296 struct fib6_walker_t *w = (void*)cb->args[2];
298 if (w) {
299 if (cb->args[4]) {
300 cb->args[4] = 0;
301 fib6_walker_unlink(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 w->count = 0;
327 w->skip = 0;
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;
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;
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;
355 return res;
358 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
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;
370 s_h = cb->args[0];
371 s_e = cb->args[1];
373 w = (void *)cb->args[2];
374 if (w == NULL) {
375 /* New dump:
377 * 1. hook callback destructor.
379 cb->args[3] = (long)cb->done;
380 cb->done = fib6_dump_done;
383 * 2. allocate and initialize walker.
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;
392 arg.skb = skb;
393 arg.cb = cb;
394 arg.net = net;
395 w->args = &arg;
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;
407 next:
408 e++;
411 out:
412 rcu_read_unlock();
413 cb->args[1] = e;
414 cb->args[0] = h;
416 res = res < 0 ? res : skb->len;
417 if (res <= 0)
418 fib6_dump_end(cb);
419 return res;
423 * Routing Table
425 * return the appropriate node for a routing tree "add" operation
426 * by either creating and inserting or by returning an existing
427 * node.
430 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
431 int addrlen, int plen,
432 int offset)
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();
441 RT6_TRACE("fib6_add_1\n");
443 /* insert node in tree */
445 fn = root;
447 do {
448 key = (struct rt6key *)((u8 *)fn->leaf + offset);
451 * Prefix match
453 if (plen < fn->fn_bit ||
454 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
455 goto insert_above;
458 * Exact match ?
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;
468 fn->fn_sernum = sernum;
470 return fn;
474 * We have more bits to go
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);
485 * We walked to the bottom of tree.
486 * Create new leaf node without children.
489 ln = node_alloc();
491 if (ln == NULL)
492 return NULL;
493 ln->fn_bit = plen;
495 ln->parent = pn;
496 ln->fn_sernum = sernum;
498 if (dir)
499 pn->right = ln;
500 else
501 pn->left = ln;
503 return ln;
506 insert_above:
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
515 pn = fn->parent;
517 /* find 1st bit in difference between the 2 addrs.
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.
523 bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
526 * (intermediate)[in]
527 * / \
528 * (new leaf node)[ln] (old node)[fn]
530 if (plen > bit) {
531 in = node_alloc();
532 ln = node_alloc();
534 if (in == NULL || ln == NULL) {
535 if (in)
536 node_free(in);
537 if (ln)
538 node_free(ln);
539 return NULL;
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
550 in->fn_bit = bit;
552 in->parent = pn;
553 in->leaf = fn->leaf;
554 atomic_inc(&in->leaf->rt6i_ref);
556 in->fn_sernum = sernum;
558 /* update parent pointer */
559 if (dir)
560 pn->right = in;
561 else
562 pn->left = in;
564 ln->fn_bit = plen;
566 ln->parent = in;
567 fn->parent = in;
569 ln->fn_sernum = sernum;
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;
578 } else { /* plen <= bit */
581 * (new leaf node)[ln]
582 * / \
583 * (old node)[fn] NULL
586 ln = node_alloc();
588 if (ln == NULL)
589 return NULL;
591 ln->fn_bit = plen;
593 ln->parent = pn;
595 ln->fn_sernum = sernum;
597 if (dir)
598 pn->right = ln;
599 else
600 pn->left = ln;
602 if (addr_bit_set(&key->addr, plen))
603 ln->right = fn;
604 else
605 ln->left = fn;
607 fn->parent = ln;
609 return ln;
613 * Insert routing information in a node.
616 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
617 struct nl_info *info)
619 struct rt6_info *iter = NULL;
620 struct rt6_info **ins;
622 ins = &fn->leaf;
624 for (iter = fn->leaf; iter; iter=iter->dst.rt6_next) {
626 * Search for duplicates
629 if (iter->rt6i_metric == rt->rt6i_metric) {
631 * Same priority level
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;
645 return -EEXIST;
649 if (iter->rt6i_metric > rt->rt6i_metric)
650 break;
652 ins = &iter->dst.rt6_next;
655 /* Reset round-robin state, if necessary */
656 if (ins == &fn->leaf)
657 fn->rr_ptr = NULL;
660 * insert node
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++;
670 if ((fn->fn_flags & RTN_RTINFO) == 0) {
671 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
672 fn->fn_flags |= RTN_RTINFO;
675 return 0;
678 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
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);
686 void fib6_force_start_gc(struct net *net)
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);
694 * Add routing information to the routing tree.
695 * <destination addr>/<source addr>
696 * with source addr info in sub-trees
699 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
701 struct fib6_node *fn, *pn = NULL;
702 int err = -ENOMEM;
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));
707 if (fn == NULL)
708 goto out;
710 pn = fn;
712 #ifdef CONFIG_IPV6_SUBTREES
713 if (rt->rt6i_src.plen) {
714 struct fib6_node *sn;
716 if (fn->subtree == NULL) {
717 struct fib6_node *sfn;
720 * Create subtree.
722 * fn[main tree]
724 * sfn[subtree root]
726 * sn[new leaf node]
729 /* Create subtree root node */
730 sfn = node_alloc();
731 if (sfn == NULL)
732 goto st_failure;
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();
739 /* Now add the first leaf node to new subtree */
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));
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.
750 node_free(sfn);
751 goto st_failure;
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));
762 if (sn == NULL)
763 goto st_failure;
766 if (fn->leaf == NULL) {
767 fn->leaf = rt;
768 atomic_inc(&rt->rt6i_ref);
770 fn = sn;
772 #endif
774 err = fib6_add_rt2node(fn, rt, info);
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);
782 out:
783 if (err) {
784 #ifdef CONFIG_IPV6_SUBTREES
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.
789 if (pn != fn && pn->leaf == rt) {
790 pn->leaf = NULL;
791 atomic_dec(&rt->rt6i_ref);
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;
800 #endif
801 atomic_inc(&pn->leaf->rt6i_ref);
803 #endif
804 dst_free(&rt->dst);
806 return err;
808 #ifdef CONFIG_IPV6_SUBTREES
809 /* Subtree creation failed, probably main tree node
810 is orphan. If it is, shoot it.
812 st_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
821 * Routing tree lookup
825 struct lookup_args {
826 int offset; /* key offset on rt6_info */
827 const struct in6_addr *addr; /* search key */
830 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
831 struct lookup_args *args)
833 struct fib6_node *fn;
834 __be32 dir;
836 if (unlikely(args->offset == 0))
837 return NULL;
840 * Descend on a tree
843 fn = root;
845 for (;;) {
846 struct fib6_node *next;
848 dir = addr_bit_set(args->addr, fn->fn_bit);
850 next = dir ? fn->right : fn->left;
852 if (next) {
853 fn = next;
854 continue;
857 break;
860 while(fn) {
861 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
862 struct rt6key *key;
864 key = (struct rt6key *) ((u8 *) fn->leaf +
865 args->offset);
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;
877 if (fn->fn_flags & RTN_ROOT)
878 break;
880 fn = fn->parent;
883 return NULL;
886 struct fib6_node * fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
887 const struct in6_addr *saddr)
889 struct fib6_node *fn;
890 struct lookup_args args[] = {
892 .offset = offsetof(struct rt6_info, rt6i_dst),
893 .addr = daddr,
895 #ifdef CONFIG_IPV6_SUBTREES
897 .offset = offsetof(struct rt6_info, rt6i_src),
898 .addr = saddr,
900 #endif
902 .offset = 0, /* sentinel */
906 fn = fib6_lookup_1(root, daddr ? args : args + 1);
908 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT)
909 fn = root;
911 return fn;
915 * Get node with specified destination prefix (and source prefix,
916 * if subtrees are used)
920 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
921 const struct in6_addr *addr,
922 int plen, int offset)
924 struct fib6_node *fn;
926 for (fn = root; fn ; ) {
927 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
930 * Prefix match
932 if (plen < fn->fn_bit ||
933 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
934 return NULL;
936 if (plen == fn->fn_bit)
937 return fn;
940 * We have more bits to go
942 if (addr_bit_set(addr, fn->fn_bit))
943 fn = fn->right;
944 else
945 fn = fn->left;
947 return NULL;
950 struct 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)
954 struct fib6_node *fn;
956 fn = fib6_locate_1(root, daddr, dst_len,
957 offsetof(struct rt6_info, rt6i_dst));
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));
966 #endif
968 if (fn && fn->fn_flags&RTN_RTINFO)
969 return fn;
971 return NULL;
976 * Deletion
980 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
982 if (fn->fn_flags&RTN_ROOT)
983 return net->ipv6.ip6_null_entry;
985 while(fn) {
986 if(fn->left)
987 return fn->left->leaf;
989 if(fn->right)
990 return fn->right->leaf;
992 fn = FIB6_SUBTREE(fn);
994 return NULL;
998 * Called to trim the tree of intermediate nodes when possible. "fn"
999 * is the node we want to try and remove.
1002 static struct fib6_node *fib6_repair_tree(struct net *net,
1003 struct fib6_node *fn)
1005 int children;
1006 int nstate;
1007 struct fib6_node *child, *pn;
1008 struct fib6_walker_t *w;
1009 int iter = 0;
1011 for (;;) {
1012 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1013 iter++;
1015 WARN_ON(fn->fn_flags & RTN_RTINFO);
1016 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1017 WARN_ON(fn->leaf != NULL);
1019 children = 0;
1020 child = NULL;
1021 if (fn->right) child = fn->right, children |= 1;
1022 if (fn->left) child = fn->left, children |= 2;
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
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;
1036 #endif
1037 atomic_inc(&fn->leaf->rt6i_ref);
1038 return fn->parent;
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
1061 #endif
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;
1074 } else {
1075 if (w->root == fn) {
1076 w->root = child;
1077 RT6_TRACE("W %p adjusted by delroot 2\n", w);
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;
1091 read_unlock(&fib6_walker_lock);
1093 node_free(fn);
1094 if (pn->fn_flags&RTN_RTINFO || FIB6_SUBTREE(pn))
1095 return pn;
1097 rt6_release(pn->leaf);
1098 pn->leaf = NULL;
1099 fn = pn;
1103 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1104 struct nl_info *info)
1106 struct fib6_walker_t *w;
1107 struct rt6_info *rt = *rtp;
1108 struct net *net = info->nl_net;
1110 RT6_TRACE("fib6_del_route\n");
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++;
1118 /* Reset round-robin state, if necessary */
1119 if (fn->rr_ptr == rt)
1120 fn->rr_ptr = NULL;
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;
1132 read_unlock(&fib6_walker_lock);
1134 rt->dst.rt6_next = NULL;
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);
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.
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);
1156 fn = fn->parent;
1158 /* No more references are possible at this point. */
1159 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1162 inet6_rt_notify(RTM_DELROUTE, rt, info);
1163 rt6_release(rt);
1166 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1168 struct net *net = info->nl_net;
1169 struct fib6_node *fn = rt->rt6i_node;
1170 struct rt6_info **rtp;
1172 #if RT6_DEBUG >= 2
1173 if (rt->dst.obsolete>0) {
1174 WARN_ON(fn != NULL);
1175 return -ENOENT;
1177 #endif
1178 if (fn == NULL || rt == net->ipv6.ip6_null_entry)
1179 return -ENOENT;
1181 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
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;
1192 #endif
1193 fib6_prune_clones(info->nl_net, pn, rt);
1197 * Walk the leaf entries looking for ourself
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;
1206 return -ENOENT;
1210 * Tree traversal function.
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.
1218 * It guarantees that every node will be traversed,
1219 * and that it will be traversed only once.
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.
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.
1233 static int fib6_walk_continue(struct fib6_walker_t *w)
1235 struct fib6_node *fn, *pn;
1237 for (;;) {
1238 fn = w->node;
1239 if (fn == NULL)
1240 return 0;
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;
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;
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;
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;
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;
1275 if (w->count < w->skip) {
1276 w->count++;
1277 continue;
1280 err = w->func(w);
1281 if (err)
1282 return err;
1284 w->count++;
1285 continue;
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;
1299 #endif
1300 if (pn->left == fn) {
1301 w->state = FWS_R;
1302 continue;
1304 if (pn->right == fn) {
1305 w->state = FWS_C;
1306 w->leaf = w->node->leaf;
1307 continue;
1309 #if RT6_DEBUG >= 2
1310 WARN_ON(1);
1311 #endif
1316 static int fib6_walk(struct fib6_walker_t *w)
1318 int res;
1320 w->state = FWS_INIT;
1321 w->node = w->root;
1323 fib6_walker_link(w);
1324 res = fib6_walk_continue(w);
1325 if (res <= 0)
1326 fib6_walker_unlink(w);
1327 return res;
1330 static int fib6_clean_node(struct fib6_walker_t *w)
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,
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;
1350 return 0;
1352 WARN_ON(res != 0);
1354 w->leaf = rt;
1355 return 0;
1359 * Convenient frontend to tree walker.
1361 * func is called on each route.
1362 * It may return -1 -> delete this route.
1363 * 0 -> continue walking
1365 * prune==1 -> only immediate children of node (certainly,
1366 * ignoring pure split nodes) will be scanned.
1369 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1370 int (*func)(struct rt6_info *, void *arg),
1371 int prune, void *arg)
1373 struct fib6_cleaner_t c;
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;
1384 fib6_walk(&c.w);
1387 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1388 int prune, void *arg)
1390 struct fib6_table *table;
1391 struct hlist_node *node;
1392 struct hlist_head *head;
1393 unsigned int h;
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);
1405 rcu_read_unlock();
1408 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1410 if (rt->rt6i_flags & RTF_CACHE) {
1411 RT6_TRACE("pruning clone %p\n", rt);
1412 return -1;
1415 return 0;
1418 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1419 struct rt6_info *rt)
1421 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1425 * Garbage collection
1428 static struct fib6_gc_args
1430 int timeout;
1431 int more;
1432 } gc_args;
1434 static int fib6_age(struct rt6_info *rt, void *arg)
1436 unsigned long now = jiffies;
1439 * check addrconf expiration here.
1440 * Routes are expired even if they are in use.
1442 * Also age clones. Note, that clones are aged out
1443 * only if they are not in use now.
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;
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;
1463 gc_args.more++;
1466 return 0;
1469 static DEFINE_SPINLOCK(fib6_gc_lock);
1471 void fib6_run_gc(unsigned long expires, struct net *net)
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;
1482 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1485 gc_args.more = icmp6_dst_gc();
1487 fib6_clean_all(net, fib6_age, 0, NULL);
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);
1498 static void fib6_gc_timer_cb(unsigned long arg)
1500 fib6_run_gc(0, (struct net *)arg);
1503 static int __net_init fib6_net_init(struct net *net)
1505 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1507 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1509 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1510 if (!net->ipv6.rt6_stats)
1511 goto out_timer;
1513 /* Avoid false sharing : Use at least a full cache line */
1514 size = max_t(size_t, size, L1_CACHE_BYTES);
1516 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1517 if (!net->ipv6.fib_table_hash)
1518 goto out_rt6_stats;
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;
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;
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);
1542 return 0;
1544 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1545 out_fib6_main_tbl:
1546 kfree(net->ipv6.fib6_main_tbl);
1547 #endif
1548 out_fib_table_hash:
1549 kfree(net->ipv6.fib_table_hash);
1550 out_rt6_stats:
1551 kfree(net->ipv6.rt6_stats);
1552 out_timer:
1553 return -ENOMEM;
1556 static void fib6_net_exit(struct net *net)
1558 rt6_ifdown(net, NULL);
1559 del_timer_sync(&net->ipv6.ip6_fib_timer);
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);
1569 static struct pernet_operations fib6_net_ops = {
1570 .init = fib6_net_init,
1571 .exit = fib6_net_exit,
1574 int __init fib6_init(void)
1576 int ret = -ENOMEM;
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;
1585 ret = register_pernet_subsys(&fib6_net_ops);
1586 if (ret)
1587 goto out_kmem_cache_create;
1589 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1590 NULL);
1591 if (ret)
1592 goto out_unregister_subsys;
1593 out:
1594 return ret;
1596 out_unregister_subsys:
1597 unregister_pernet_subsys(&fib6_net_ops);
1598 out_kmem_cache_create:
1599 kmem_cache_destroy(fib6_node_kmem);
1600 goto out;
1603 void fib6_gc_cleanup(void)
1605 unregister_pernet_subsys(&fib6_net_ops);
1606 kmem_cache_destroy(fib6_node_kmem);