ipv6: Convert to use flowi6 where applicable.
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / ipv6 / ip6_fib.c
blob7548905e79e119fb2b6fe6df234a9c63ad6b7c0e
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(void *token, int fn_bit)
139 __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 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
398 e = 0;
399 head = &net->ipv6.fib_table_hash[h];
400 hlist_for_each_entry(tb, node, head, tb6_hlist) {
401 if (e < s_e)
402 goto next;
403 res = fib6_dump_table(tb, skb, cb);
404 if (res != 0)
405 goto out;
406 next:
407 e++;
410 out:
411 cb->args[1] = e;
412 cb->args[0] = h;
414 res = res < 0 ? res : skb->len;
415 if (res <= 0)
416 fib6_dump_end(cb);
417 return res;
421 * Routing Table
423 * return the appropriate node for a routing tree "add" operation
424 * by either creating and inserting or by returning an existing
425 * node.
428 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
429 int addrlen, int plen,
430 int offset)
432 struct fib6_node *fn, *in, *ln;
433 struct fib6_node *pn = NULL;
434 struct rt6key *key;
435 int bit;
436 __be32 dir = 0;
437 __u32 sernum = fib6_new_sernum();
439 RT6_TRACE("fib6_add_1\n");
441 /* insert node in tree */
443 fn = root;
445 do {
446 key = (struct rt6key *)((u8 *)fn->leaf + offset);
449 * Prefix match
451 if (plen < fn->fn_bit ||
452 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
453 goto insert_above;
456 * Exact match ?
459 if (plen == fn->fn_bit) {
460 /* clean up an intermediate node */
461 if ((fn->fn_flags & RTN_RTINFO) == 0) {
462 rt6_release(fn->leaf);
463 fn->leaf = NULL;
466 fn->fn_sernum = sernum;
468 return fn;
472 * We have more bits to go
475 /* Try to walk down on tree. */
476 fn->fn_sernum = sernum;
477 dir = addr_bit_set(addr, fn->fn_bit);
478 pn = fn;
479 fn = dir ? fn->right: fn->left;
480 } while (fn);
483 * We walked to the bottom of tree.
484 * Create new leaf node without children.
487 ln = node_alloc();
489 if (ln == NULL)
490 return NULL;
491 ln->fn_bit = plen;
493 ln->parent = pn;
494 ln->fn_sernum = sernum;
496 if (dir)
497 pn->right = ln;
498 else
499 pn->left = ln;
501 return ln;
504 insert_above:
506 * split since we don't have a common prefix anymore or
507 * we have a less significant route.
508 * we've to insert an intermediate node on the list
509 * this new node will point to the one we need to create
510 * and the current
513 pn = fn->parent;
515 /* find 1st bit in difference between the 2 addrs.
517 See comment in __ipv6_addr_diff: bit may be an invalid value,
518 but if it is >= plen, the value is ignored in any case.
521 bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
524 * (intermediate)[in]
525 * / \
526 * (new leaf node)[ln] (old node)[fn]
528 if (plen > bit) {
529 in = node_alloc();
530 ln = node_alloc();
532 if (in == NULL || ln == NULL) {
533 if (in)
534 node_free(in);
535 if (ln)
536 node_free(ln);
537 return NULL;
541 * new intermediate node.
542 * RTN_RTINFO will
543 * be off since that an address that chooses one of
544 * the branches would not match less specific routes
545 * in the other branch
548 in->fn_bit = bit;
550 in->parent = pn;
551 in->leaf = fn->leaf;
552 atomic_inc(&in->leaf->rt6i_ref);
554 in->fn_sernum = sernum;
556 /* update parent pointer */
557 if (dir)
558 pn->right = in;
559 else
560 pn->left = in;
562 ln->fn_bit = plen;
564 ln->parent = in;
565 fn->parent = in;
567 ln->fn_sernum = sernum;
569 if (addr_bit_set(addr, bit)) {
570 in->right = ln;
571 in->left = fn;
572 } else {
573 in->left = ln;
574 in->right = fn;
576 } else { /* plen <= bit */
579 * (new leaf node)[ln]
580 * / \
581 * (old node)[fn] NULL
584 ln = node_alloc();
586 if (ln == NULL)
587 return NULL;
589 ln->fn_bit = plen;
591 ln->parent = pn;
593 ln->fn_sernum = sernum;
595 if (dir)
596 pn->right = ln;
597 else
598 pn->left = ln;
600 if (addr_bit_set(&key->addr, plen))
601 ln->right = fn;
602 else
603 ln->left = fn;
605 fn->parent = ln;
607 return ln;
611 * Insert routing information in a node.
614 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
615 struct nl_info *info)
617 struct rt6_info *iter = NULL;
618 struct rt6_info **ins;
620 ins = &fn->leaf;
622 for (iter = fn->leaf; iter; iter=iter->dst.rt6_next) {
624 * Search for duplicates
627 if (iter->rt6i_metric == rt->rt6i_metric) {
629 * Same priority level
632 if (iter->rt6i_dev == rt->rt6i_dev &&
633 iter->rt6i_idev == rt->rt6i_idev &&
634 ipv6_addr_equal(&iter->rt6i_gateway,
635 &rt->rt6i_gateway)) {
636 if (!(iter->rt6i_flags&RTF_EXPIRES))
637 return -EEXIST;
638 iter->rt6i_expires = rt->rt6i_expires;
639 if (!(rt->rt6i_flags&RTF_EXPIRES)) {
640 iter->rt6i_flags &= ~RTF_EXPIRES;
641 iter->rt6i_expires = 0;
643 return -EEXIST;
647 if (iter->rt6i_metric > rt->rt6i_metric)
648 break;
650 ins = &iter->dst.rt6_next;
653 /* Reset round-robin state, if necessary */
654 if (ins == &fn->leaf)
655 fn->rr_ptr = NULL;
658 * insert node
661 rt->dst.rt6_next = iter;
662 *ins = rt;
663 rt->rt6i_node = fn;
664 atomic_inc(&rt->rt6i_ref);
665 inet6_rt_notify(RTM_NEWROUTE, rt, info);
666 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
668 if ((fn->fn_flags & RTN_RTINFO) == 0) {
669 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
670 fn->fn_flags |= RTN_RTINFO;
673 return 0;
676 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
678 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
679 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE)))
680 mod_timer(&net->ipv6.ip6_fib_timer,
681 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
684 void fib6_force_start_gc(struct net *net)
686 if (!timer_pending(&net->ipv6.ip6_fib_timer))
687 mod_timer(&net->ipv6.ip6_fib_timer,
688 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
692 * Add routing information to the routing tree.
693 * <destination addr>/<source addr>
694 * with source addr info in sub-trees
697 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
699 struct fib6_node *fn, *pn = NULL;
700 int err = -ENOMEM;
702 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
703 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst));
705 if (fn == NULL)
706 goto out;
708 pn = fn;
710 #ifdef CONFIG_IPV6_SUBTREES
711 if (rt->rt6i_src.plen) {
712 struct fib6_node *sn;
714 if (fn->subtree == NULL) {
715 struct fib6_node *sfn;
718 * Create subtree.
720 * fn[main tree]
722 * sfn[subtree root]
724 * sn[new leaf node]
727 /* Create subtree root node */
728 sfn = node_alloc();
729 if (sfn == NULL)
730 goto st_failure;
732 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
733 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
734 sfn->fn_flags = RTN_ROOT;
735 sfn->fn_sernum = fib6_new_sernum();
737 /* Now add the first leaf node to new subtree */
739 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
740 sizeof(struct in6_addr), rt->rt6i_src.plen,
741 offsetof(struct rt6_info, rt6i_src));
743 if (sn == NULL) {
744 /* If it is failed, discard just allocated
745 root, and then (in st_failure) stale node
746 in main tree.
748 node_free(sfn);
749 goto st_failure;
752 /* Now link new subtree to main tree */
753 sfn->parent = fn;
754 fn->subtree = sfn;
755 } else {
756 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
757 sizeof(struct in6_addr), rt->rt6i_src.plen,
758 offsetof(struct rt6_info, rt6i_src));
760 if (sn == NULL)
761 goto st_failure;
764 if (fn->leaf == NULL) {
765 fn->leaf = rt;
766 atomic_inc(&rt->rt6i_ref);
768 fn = sn;
770 #endif
772 err = fib6_add_rt2node(fn, rt, info);
774 if (err == 0) {
775 fib6_start_gc(info->nl_net, rt);
776 if (!(rt->rt6i_flags&RTF_CACHE))
777 fib6_prune_clones(info->nl_net, pn, rt);
780 out:
781 if (err) {
782 #ifdef CONFIG_IPV6_SUBTREES
784 * If fib6_add_1 has cleared the old leaf pointer in the
785 * super-tree leaf node we have to find a new one for it.
787 if (pn != fn && pn->leaf == rt) {
788 pn->leaf = NULL;
789 atomic_dec(&rt->rt6i_ref);
791 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
792 pn->leaf = fib6_find_prefix(info->nl_net, pn);
793 #if RT6_DEBUG >= 2
794 if (!pn->leaf) {
795 WARN_ON(pn->leaf == NULL);
796 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
798 #endif
799 atomic_inc(&pn->leaf->rt6i_ref);
801 #endif
802 dst_free(&rt->dst);
804 return err;
806 #ifdef CONFIG_IPV6_SUBTREES
807 /* Subtree creation failed, probably main tree node
808 is orphan. If it is, shoot it.
810 st_failure:
811 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
812 fib6_repair_tree(info->nl_net, fn);
813 dst_free(&rt->dst);
814 return err;
815 #endif
819 * Routing tree lookup
823 struct lookup_args {
824 int offset; /* key offset on rt6_info */
825 struct in6_addr *addr; /* search key */
828 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
829 struct lookup_args *args)
831 struct fib6_node *fn;
832 __be32 dir;
834 if (unlikely(args->offset == 0))
835 return NULL;
838 * Descend on a tree
841 fn = root;
843 for (;;) {
844 struct fib6_node *next;
846 dir = addr_bit_set(args->addr, fn->fn_bit);
848 next = dir ? fn->right : fn->left;
850 if (next) {
851 fn = next;
852 continue;
855 break;
858 while(fn) {
859 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
860 struct rt6key *key;
862 key = (struct rt6key *) ((u8 *) fn->leaf +
863 args->offset);
865 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
866 #ifdef CONFIG_IPV6_SUBTREES
867 if (fn->subtree)
868 fn = fib6_lookup_1(fn->subtree, args + 1);
869 #endif
870 if (!fn || fn->fn_flags & RTN_RTINFO)
871 return fn;
875 if (fn->fn_flags & RTN_ROOT)
876 break;
878 fn = fn->parent;
881 return NULL;
884 struct fib6_node * fib6_lookup(struct fib6_node *root, struct in6_addr *daddr,
885 struct in6_addr *saddr)
887 struct fib6_node *fn;
888 struct lookup_args args[] = {
890 .offset = offsetof(struct rt6_info, rt6i_dst),
891 .addr = daddr,
893 #ifdef CONFIG_IPV6_SUBTREES
895 .offset = offsetof(struct rt6_info, rt6i_src),
896 .addr = saddr,
898 #endif
900 .offset = 0, /* sentinel */
904 fn = fib6_lookup_1(root, daddr ? args : args + 1);
906 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT)
907 fn = root;
909 return fn;
913 * Get node with specified destination prefix (and source prefix,
914 * if subtrees are used)
918 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
919 struct in6_addr *addr,
920 int plen, int offset)
922 struct fib6_node *fn;
924 for (fn = root; fn ; ) {
925 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
928 * Prefix match
930 if (plen < fn->fn_bit ||
931 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
932 return NULL;
934 if (plen == fn->fn_bit)
935 return fn;
938 * We have more bits to go
940 if (addr_bit_set(addr, fn->fn_bit))
941 fn = fn->right;
942 else
943 fn = fn->left;
945 return NULL;
948 struct fib6_node * fib6_locate(struct fib6_node *root,
949 struct in6_addr *daddr, int dst_len,
950 struct in6_addr *saddr, int src_len)
952 struct fib6_node *fn;
954 fn = fib6_locate_1(root, daddr, dst_len,
955 offsetof(struct rt6_info, rt6i_dst));
957 #ifdef CONFIG_IPV6_SUBTREES
958 if (src_len) {
959 WARN_ON(saddr == NULL);
960 if (fn && fn->subtree)
961 fn = fib6_locate_1(fn->subtree, saddr, src_len,
962 offsetof(struct rt6_info, rt6i_src));
964 #endif
966 if (fn && fn->fn_flags&RTN_RTINFO)
967 return fn;
969 return NULL;
974 * Deletion
978 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
980 if (fn->fn_flags&RTN_ROOT)
981 return net->ipv6.ip6_null_entry;
983 while(fn) {
984 if(fn->left)
985 return fn->left->leaf;
987 if(fn->right)
988 return fn->right->leaf;
990 fn = FIB6_SUBTREE(fn);
992 return NULL;
996 * Called to trim the tree of intermediate nodes when possible. "fn"
997 * is the node we want to try and remove.
1000 static struct fib6_node *fib6_repair_tree(struct net *net,
1001 struct fib6_node *fn)
1003 int children;
1004 int nstate;
1005 struct fib6_node *child, *pn;
1006 struct fib6_walker_t *w;
1007 int iter = 0;
1009 for (;;) {
1010 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1011 iter++;
1013 WARN_ON(fn->fn_flags & RTN_RTINFO);
1014 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1015 WARN_ON(fn->leaf != NULL);
1017 children = 0;
1018 child = NULL;
1019 if (fn->right) child = fn->right, children |= 1;
1020 if (fn->left) child = fn->left, children |= 2;
1022 if (children == 3 || FIB6_SUBTREE(fn)
1023 #ifdef CONFIG_IPV6_SUBTREES
1024 /* Subtree root (i.e. fn) may have one child */
1025 || (children && fn->fn_flags&RTN_ROOT)
1026 #endif
1028 fn->leaf = fib6_find_prefix(net, fn);
1029 #if RT6_DEBUG >= 2
1030 if (fn->leaf==NULL) {
1031 WARN_ON(!fn->leaf);
1032 fn->leaf = net->ipv6.ip6_null_entry;
1034 #endif
1035 atomic_inc(&fn->leaf->rt6i_ref);
1036 return fn->parent;
1039 pn = fn->parent;
1040 #ifdef CONFIG_IPV6_SUBTREES
1041 if (FIB6_SUBTREE(pn) == fn) {
1042 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1043 FIB6_SUBTREE(pn) = NULL;
1044 nstate = FWS_L;
1045 } else {
1046 WARN_ON(fn->fn_flags & RTN_ROOT);
1047 #endif
1048 if (pn->right == fn) pn->right = child;
1049 else if (pn->left == fn) pn->left = child;
1050 #if RT6_DEBUG >= 2
1051 else
1052 WARN_ON(1);
1053 #endif
1054 if (child)
1055 child->parent = pn;
1056 nstate = FWS_R;
1057 #ifdef CONFIG_IPV6_SUBTREES
1059 #endif
1061 read_lock(&fib6_walker_lock);
1062 FOR_WALKERS(w) {
1063 if (child == NULL) {
1064 if (w->root == fn) {
1065 w->root = w->node = NULL;
1066 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1067 } else if (w->node == fn) {
1068 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1069 w->node = pn;
1070 w->state = nstate;
1072 } else {
1073 if (w->root == fn) {
1074 w->root = child;
1075 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1077 if (w->node == fn) {
1078 w->node = child;
1079 if (children&2) {
1080 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1081 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1082 } else {
1083 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1084 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1089 read_unlock(&fib6_walker_lock);
1091 node_free(fn);
1092 if (pn->fn_flags&RTN_RTINFO || FIB6_SUBTREE(pn))
1093 return pn;
1095 rt6_release(pn->leaf);
1096 pn->leaf = NULL;
1097 fn = pn;
1101 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1102 struct nl_info *info)
1104 struct fib6_walker_t *w;
1105 struct rt6_info *rt = *rtp;
1106 struct net *net = info->nl_net;
1108 RT6_TRACE("fib6_del_route\n");
1110 /* Unlink it */
1111 *rtp = rt->dst.rt6_next;
1112 rt->rt6i_node = NULL;
1113 net->ipv6.rt6_stats->fib_rt_entries--;
1114 net->ipv6.rt6_stats->fib_discarded_routes++;
1116 /* Reset round-robin state, if necessary */
1117 if (fn->rr_ptr == rt)
1118 fn->rr_ptr = NULL;
1120 /* Adjust walkers */
1121 read_lock(&fib6_walker_lock);
1122 FOR_WALKERS(w) {
1123 if (w->state == FWS_C && w->leaf == rt) {
1124 RT6_TRACE("walker %p adjusted by delroute\n", w);
1125 w->leaf = rt->dst.rt6_next;
1126 if (w->leaf == NULL)
1127 w->state = FWS_U;
1130 read_unlock(&fib6_walker_lock);
1132 rt->dst.rt6_next = NULL;
1134 /* If it was last route, expunge its radix tree node */
1135 if (fn->leaf == NULL) {
1136 fn->fn_flags &= ~RTN_RTINFO;
1137 net->ipv6.rt6_stats->fib_route_nodes--;
1138 fn = fib6_repair_tree(net, fn);
1141 if (atomic_read(&rt->rt6i_ref) != 1) {
1142 /* This route is used as dummy address holder in some split
1143 * nodes. It is not leaked, but it still holds other resources,
1144 * which must be released in time. So, scan ascendant nodes
1145 * and replace dummy references to this route with references
1146 * to still alive ones.
1148 while (fn) {
1149 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) {
1150 fn->leaf = fib6_find_prefix(net, fn);
1151 atomic_inc(&fn->leaf->rt6i_ref);
1152 rt6_release(rt);
1154 fn = fn->parent;
1156 /* No more references are possible at this point. */
1157 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1160 inet6_rt_notify(RTM_DELROUTE, rt, info);
1161 rt6_release(rt);
1164 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1166 struct net *net = info->nl_net;
1167 struct fib6_node *fn = rt->rt6i_node;
1168 struct rt6_info **rtp;
1170 #if RT6_DEBUG >= 2
1171 if (rt->dst.obsolete>0) {
1172 WARN_ON(fn != NULL);
1173 return -ENOENT;
1175 #endif
1176 if (fn == NULL || rt == net->ipv6.ip6_null_entry)
1177 return -ENOENT;
1179 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1181 if (!(rt->rt6i_flags&RTF_CACHE)) {
1182 struct fib6_node *pn = fn;
1183 #ifdef CONFIG_IPV6_SUBTREES
1184 /* clones of this route might be in another subtree */
1185 if (rt->rt6i_src.plen) {
1186 while (!(pn->fn_flags&RTN_ROOT))
1187 pn = pn->parent;
1188 pn = pn->parent;
1190 #endif
1191 fib6_prune_clones(info->nl_net, pn, rt);
1195 * Walk the leaf entries looking for ourself
1198 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1199 if (*rtp == rt) {
1200 fib6_del_route(fn, rtp, info);
1201 return 0;
1204 return -ENOENT;
1208 * Tree traversal function.
1210 * Certainly, it is not interrupt safe.
1211 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1212 * It means, that we can modify tree during walking
1213 * and use this function for garbage collection, clone pruning,
1214 * cleaning tree when a device goes down etc. etc.
1216 * It guarantees that every node will be traversed,
1217 * and that it will be traversed only once.
1219 * Callback function w->func may return:
1220 * 0 -> continue walking.
1221 * positive value -> walking is suspended (used by tree dumps,
1222 * and probably by gc, if it will be split to several slices)
1223 * negative value -> terminate walking.
1225 * The function itself returns:
1226 * 0 -> walk is complete.
1227 * >0 -> walk is incomplete (i.e. suspended)
1228 * <0 -> walk is terminated by an error.
1231 static int fib6_walk_continue(struct fib6_walker_t *w)
1233 struct fib6_node *fn, *pn;
1235 for (;;) {
1236 fn = w->node;
1237 if (fn == NULL)
1238 return 0;
1240 if (w->prune && fn != w->root &&
1241 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) {
1242 w->state = FWS_C;
1243 w->leaf = fn->leaf;
1245 switch (w->state) {
1246 #ifdef CONFIG_IPV6_SUBTREES
1247 case FWS_S:
1248 if (FIB6_SUBTREE(fn)) {
1249 w->node = FIB6_SUBTREE(fn);
1250 continue;
1252 w->state = FWS_L;
1253 #endif
1254 case FWS_L:
1255 if (fn->left) {
1256 w->node = fn->left;
1257 w->state = FWS_INIT;
1258 continue;
1260 w->state = FWS_R;
1261 case FWS_R:
1262 if (fn->right) {
1263 w->node = fn->right;
1264 w->state = FWS_INIT;
1265 continue;
1267 w->state = FWS_C;
1268 w->leaf = fn->leaf;
1269 case FWS_C:
1270 if (w->leaf && fn->fn_flags&RTN_RTINFO) {
1271 int err;
1273 if (w->count < w->skip) {
1274 w->count++;
1275 continue;
1278 err = w->func(w);
1279 if (err)
1280 return err;
1282 w->count++;
1283 continue;
1285 w->state = FWS_U;
1286 case FWS_U:
1287 if (fn == w->root)
1288 return 0;
1289 pn = fn->parent;
1290 w->node = pn;
1291 #ifdef CONFIG_IPV6_SUBTREES
1292 if (FIB6_SUBTREE(pn) == fn) {
1293 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1294 w->state = FWS_L;
1295 continue;
1297 #endif
1298 if (pn->left == fn) {
1299 w->state = FWS_R;
1300 continue;
1302 if (pn->right == fn) {
1303 w->state = FWS_C;
1304 w->leaf = w->node->leaf;
1305 continue;
1307 #if RT6_DEBUG >= 2
1308 WARN_ON(1);
1309 #endif
1314 static int fib6_walk(struct fib6_walker_t *w)
1316 int res;
1318 w->state = FWS_INIT;
1319 w->node = w->root;
1321 fib6_walker_link(w);
1322 res = fib6_walk_continue(w);
1323 if (res <= 0)
1324 fib6_walker_unlink(w);
1325 return res;
1328 static int fib6_clean_node(struct fib6_walker_t *w)
1330 int res;
1331 struct rt6_info *rt;
1332 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1333 struct nl_info info = {
1334 .nl_net = c->net,
1337 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1338 res = c->func(rt, c->arg);
1339 if (res < 0) {
1340 w->leaf = rt;
1341 res = fib6_del(rt, &info);
1342 if (res) {
1343 #if RT6_DEBUG >= 2
1344 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res);
1345 #endif
1346 continue;
1348 return 0;
1350 WARN_ON(res != 0);
1352 w->leaf = rt;
1353 return 0;
1357 * Convenient frontend to tree walker.
1359 * func is called on each route.
1360 * It may return -1 -> delete this route.
1361 * 0 -> continue walking
1363 * prune==1 -> only immediate children of node (certainly,
1364 * ignoring pure split nodes) will be scanned.
1367 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1368 int (*func)(struct rt6_info *, void *arg),
1369 int prune, void *arg)
1371 struct fib6_cleaner_t c;
1373 c.w.root = root;
1374 c.w.func = fib6_clean_node;
1375 c.w.prune = prune;
1376 c.w.count = 0;
1377 c.w.skip = 0;
1378 c.func = func;
1379 c.arg = arg;
1380 c.net = net;
1382 fib6_walk(&c.w);
1385 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1386 int prune, void *arg)
1388 struct fib6_table *table;
1389 struct hlist_node *node;
1390 struct hlist_head *head;
1391 unsigned int h;
1393 rcu_read_lock();
1394 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1395 head = &net->ipv6.fib_table_hash[h];
1396 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1397 write_lock_bh(&table->tb6_lock);
1398 fib6_clean_tree(net, &table->tb6_root,
1399 func, prune, arg);
1400 write_unlock_bh(&table->tb6_lock);
1403 rcu_read_unlock();
1406 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1408 if (rt->rt6i_flags & RTF_CACHE) {
1409 RT6_TRACE("pruning clone %p\n", rt);
1410 return -1;
1413 return 0;
1416 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1417 struct rt6_info *rt)
1419 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1423 * Garbage collection
1426 static struct fib6_gc_args
1428 int timeout;
1429 int more;
1430 } gc_args;
1432 static int fib6_age(struct rt6_info *rt, void *arg)
1434 unsigned long now = jiffies;
1437 * check addrconf expiration here.
1438 * Routes are expired even if they are in use.
1440 * Also age clones. Note, that clones are aged out
1441 * only if they are not in use now.
1444 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) {
1445 if (time_after(now, rt->rt6i_expires)) {
1446 RT6_TRACE("expiring %p\n", rt);
1447 return -1;
1449 gc_args.more++;
1450 } else if (rt->rt6i_flags & RTF_CACHE) {
1451 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1452 time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1453 RT6_TRACE("aging clone %p\n", rt);
1454 return -1;
1455 } else if ((rt->rt6i_flags & RTF_GATEWAY) &&
1456 (!(rt->rt6i_nexthop->flags & NTF_ROUTER))) {
1457 RT6_TRACE("purging route %p via non-router but gateway\n",
1458 rt);
1459 return -1;
1461 gc_args.more++;
1464 return 0;
1467 static DEFINE_SPINLOCK(fib6_gc_lock);
1469 void fib6_run_gc(unsigned long expires, struct net *net)
1471 if (expires != ~0UL) {
1472 spin_lock_bh(&fib6_gc_lock);
1473 gc_args.timeout = expires ? (int)expires :
1474 net->ipv6.sysctl.ip6_rt_gc_interval;
1475 } else {
1476 if (!spin_trylock_bh(&fib6_gc_lock)) {
1477 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1478 return;
1480 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1483 gc_args.more = icmp6_dst_gc();
1485 fib6_clean_all(net, fib6_age, 0, NULL);
1487 if (gc_args.more)
1488 mod_timer(&net->ipv6.ip6_fib_timer,
1489 round_jiffies(jiffies
1490 + net->ipv6.sysctl.ip6_rt_gc_interval));
1491 else
1492 del_timer(&net->ipv6.ip6_fib_timer);
1493 spin_unlock_bh(&fib6_gc_lock);
1496 static void fib6_gc_timer_cb(unsigned long arg)
1498 fib6_run_gc(0, (struct net *)arg);
1501 static int __net_init fib6_net_init(struct net *net)
1503 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1505 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1507 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1508 if (!net->ipv6.rt6_stats)
1509 goto out_timer;
1511 /* Avoid false sharing : Use at least a full cache line */
1512 size = max_t(size_t, size, L1_CACHE_BYTES);
1514 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1515 if (!net->ipv6.fib_table_hash)
1516 goto out_rt6_stats;
1518 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1519 GFP_KERNEL);
1520 if (!net->ipv6.fib6_main_tbl)
1521 goto out_fib_table_hash;
1523 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1524 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1525 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1526 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1528 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1529 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1530 GFP_KERNEL);
1531 if (!net->ipv6.fib6_local_tbl)
1532 goto out_fib6_main_tbl;
1533 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1534 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1535 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1536 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1537 #endif
1538 fib6_tables_init(net);
1540 return 0;
1542 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1543 out_fib6_main_tbl:
1544 kfree(net->ipv6.fib6_main_tbl);
1545 #endif
1546 out_fib_table_hash:
1547 kfree(net->ipv6.fib_table_hash);
1548 out_rt6_stats:
1549 kfree(net->ipv6.rt6_stats);
1550 out_timer:
1551 return -ENOMEM;
1554 static void fib6_net_exit(struct net *net)
1556 rt6_ifdown(net, NULL);
1557 del_timer_sync(&net->ipv6.ip6_fib_timer);
1559 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1560 kfree(net->ipv6.fib6_local_tbl);
1561 #endif
1562 kfree(net->ipv6.fib6_main_tbl);
1563 kfree(net->ipv6.fib_table_hash);
1564 kfree(net->ipv6.rt6_stats);
1567 static struct pernet_operations fib6_net_ops = {
1568 .init = fib6_net_init,
1569 .exit = fib6_net_exit,
1572 int __init fib6_init(void)
1574 int ret = -ENOMEM;
1576 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1577 sizeof(struct fib6_node),
1578 0, SLAB_HWCACHE_ALIGN,
1579 NULL);
1580 if (!fib6_node_kmem)
1581 goto out;
1583 ret = register_pernet_subsys(&fib6_net_ops);
1584 if (ret)
1585 goto out_kmem_cache_create;
1587 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib);
1588 if (ret)
1589 goto out_unregister_subsys;
1590 out:
1591 return ret;
1593 out_unregister_subsys:
1594 unregister_pernet_subsys(&fib6_net_ops);
1595 out_kmem_cache_create:
1596 kmem_cache_destroy(fib6_node_kmem);
1597 goto out;
1600 void fib6_gc_cleanup(void)
1602 unregister_pernet_subsys(&fib6_net_ops);
1603 kmem_cache_destroy(fib6_node_kmem);