close_port, kpacket_gen kmalloc oom, formard.c wake_sender/skb receive oom handling...
[cor_2_6_31.git] / net / ipv6 / ip6_fib.c
blob52ee1dced2ffe2ed28c86e6ef7dd355cfe18ee70
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 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
168 #define FIB_TABLE_HASHSZ 256
169 #else
170 #define FIB_TABLE_HASHSZ 1
171 #endif
173 static void fib6_link_table(struct net *net, struct fib6_table *tb)
175 unsigned int h;
178 * Initialize table lock at a single place to give lockdep a key,
179 * tables aren't visible prior to being linked to the list.
181 rwlock_init(&tb->tb6_lock);
183 h = tb->tb6_id & (FIB_TABLE_HASHSZ - 1);
186 * No protection necessary, this is the only list mutatation
187 * operation, tables never disappear once they exist.
189 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
192 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
194 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
196 struct fib6_table *table;
198 table = kzalloc(sizeof(*table), GFP_ATOMIC);
199 if (table != NULL) {
200 table->tb6_id = id;
201 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
202 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
205 return table;
208 struct fib6_table *fib6_new_table(struct net *net, u32 id)
210 struct fib6_table *tb;
212 if (id == 0)
213 id = RT6_TABLE_MAIN;
214 tb = fib6_get_table(net, id);
215 if (tb)
216 return tb;
218 tb = fib6_alloc_table(net, id);
219 if (tb != NULL)
220 fib6_link_table(net, tb);
222 return tb;
225 struct fib6_table *fib6_get_table(struct net *net, u32 id)
227 struct fib6_table *tb;
228 struct hlist_head *head;
229 struct hlist_node *node;
230 unsigned int h;
232 if (id == 0)
233 id = RT6_TABLE_MAIN;
234 h = id & (FIB_TABLE_HASHSZ - 1);
235 rcu_read_lock();
236 head = &net->ipv6.fib_table_hash[h];
237 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
238 if (tb->tb6_id == id) {
239 rcu_read_unlock();
240 return tb;
243 rcu_read_unlock();
245 return NULL;
248 static void fib6_tables_init(struct net *net)
250 fib6_link_table(net, net->ipv6.fib6_main_tbl);
251 fib6_link_table(net, net->ipv6.fib6_local_tbl);
253 #else
255 struct fib6_table *fib6_new_table(struct net *net, u32 id)
257 return fib6_get_table(net, id);
260 struct fib6_table *fib6_get_table(struct net *net, u32 id)
262 return net->ipv6.fib6_main_tbl;
265 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi *fl,
266 int flags, pol_lookup_t lookup)
268 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl, flags);
271 static void fib6_tables_init(struct net *net)
273 fib6_link_table(net, net->ipv6.fib6_main_tbl);
276 #endif
278 static int fib6_dump_node(struct fib6_walker_t *w)
280 int res;
281 struct rt6_info *rt;
283 for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) {
284 res = rt6_dump_route(rt, w->args);
285 if (res < 0) {
286 /* Frame is full, suspend walking */
287 w->leaf = rt;
288 return 1;
290 WARN_ON(res == 0);
292 w->leaf = NULL;
293 return 0;
296 static void fib6_dump_end(struct netlink_callback *cb)
298 struct fib6_walker_t *w = (void*)cb->args[2];
300 if (w) {
301 if (cb->args[4]) {
302 cb->args[4] = 0;
303 fib6_walker_unlink(w);
305 cb->args[2] = 0;
306 kfree(w);
308 cb->done = (void*)cb->args[3];
309 cb->args[1] = 3;
312 static int fib6_dump_done(struct netlink_callback *cb)
314 fib6_dump_end(cb);
315 return cb->done ? cb->done(cb) : 0;
318 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
319 struct netlink_callback *cb)
321 struct fib6_walker_t *w;
322 int res;
324 w = (void *)cb->args[2];
325 w->root = &table->tb6_root;
327 if (cb->args[4] == 0) {
328 read_lock_bh(&table->tb6_lock);
329 res = fib6_walk(w);
330 read_unlock_bh(&table->tb6_lock);
331 if (res > 0)
332 cb->args[4] = 1;
333 } else {
334 read_lock_bh(&table->tb6_lock);
335 res = fib6_walk_continue(w);
336 read_unlock_bh(&table->tb6_lock);
337 if (res <= 0) {
338 fib6_walker_unlink(w);
339 cb->args[4] = 0;
343 return res;
346 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
348 struct net *net = sock_net(skb->sk);
349 unsigned int h, s_h;
350 unsigned int e = 0, s_e;
351 struct rt6_rtnl_dump_arg arg;
352 struct fib6_walker_t *w;
353 struct fib6_table *tb;
354 struct hlist_node *node;
355 struct hlist_head *head;
356 int res = 0;
358 s_h = cb->args[0];
359 s_e = cb->args[1];
361 w = (void *)cb->args[2];
362 if (w == NULL) {
363 /* New dump:
365 * 1. hook callback destructor.
367 cb->args[3] = (long)cb->done;
368 cb->done = fib6_dump_done;
371 * 2. allocate and initialize walker.
373 w = kzalloc(sizeof(*w), GFP_ATOMIC);
374 if (w == NULL)
375 return -ENOMEM;
376 w->func = fib6_dump_node;
377 cb->args[2] = (long)w;
380 arg.skb = skb;
381 arg.cb = cb;
382 arg.net = net;
383 w->args = &arg;
385 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
386 e = 0;
387 head = &net->ipv6.fib_table_hash[h];
388 hlist_for_each_entry(tb, node, head, tb6_hlist) {
389 if (e < s_e)
390 goto next;
391 res = fib6_dump_table(tb, skb, cb);
392 if (res != 0)
393 goto out;
394 next:
395 e++;
398 out:
399 cb->args[1] = e;
400 cb->args[0] = h;
402 res = res < 0 ? res : skb->len;
403 if (res <= 0)
404 fib6_dump_end(cb);
405 return res;
409 * Routing Table
411 * return the appropriate node for a routing tree "add" operation
412 * by either creating and inserting or by returning an existing
413 * node.
416 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
417 int addrlen, int plen,
418 int offset)
420 struct fib6_node *fn, *in, *ln;
421 struct fib6_node *pn = NULL;
422 struct rt6key *key;
423 int bit;
424 __be32 dir = 0;
425 __u32 sernum = fib6_new_sernum();
427 RT6_TRACE("fib6_add_1\n");
429 /* insert node in tree */
431 fn = root;
433 do {
434 key = (struct rt6key *)((u8 *)fn->leaf + offset);
437 * Prefix match
439 if (plen < fn->fn_bit ||
440 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
441 goto insert_above;
444 * Exact match ?
447 if (plen == fn->fn_bit) {
448 /* clean up an intermediate node */
449 if ((fn->fn_flags & RTN_RTINFO) == 0) {
450 rt6_release(fn->leaf);
451 fn->leaf = NULL;
454 fn->fn_sernum = sernum;
456 return fn;
460 * We have more bits to go
463 /* Try to walk down on tree. */
464 fn->fn_sernum = sernum;
465 dir = addr_bit_set(addr, fn->fn_bit);
466 pn = fn;
467 fn = dir ? fn->right: fn->left;
468 } while (fn);
471 * We walked to the bottom of tree.
472 * Create new leaf node without children.
475 ln = node_alloc();
477 if (ln == NULL)
478 return NULL;
479 ln->fn_bit = plen;
481 ln->parent = pn;
482 ln->fn_sernum = sernum;
484 if (dir)
485 pn->right = ln;
486 else
487 pn->left = ln;
489 return ln;
492 insert_above:
494 * split since we don't have a common prefix anymore or
495 * we have a less significant route.
496 * we've to insert an intermediate node on the list
497 * this new node will point to the one we need to create
498 * and the current
501 pn = fn->parent;
503 /* find 1st bit in difference between the 2 addrs.
505 See comment in __ipv6_addr_diff: bit may be an invalid value,
506 but if it is >= plen, the value is ignored in any case.
509 bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
512 * (intermediate)[in]
513 * / \
514 * (new leaf node)[ln] (old node)[fn]
516 if (plen > bit) {
517 in = node_alloc();
518 ln = node_alloc();
520 if (in == NULL || ln == NULL) {
521 if (in)
522 node_free(in);
523 if (ln)
524 node_free(ln);
525 return NULL;
529 * new intermediate node.
530 * RTN_RTINFO will
531 * be off since that an address that chooses one of
532 * the branches would not match less specific routes
533 * in the other branch
536 in->fn_bit = bit;
538 in->parent = pn;
539 in->leaf = fn->leaf;
540 atomic_inc(&in->leaf->rt6i_ref);
542 in->fn_sernum = sernum;
544 /* update parent pointer */
545 if (dir)
546 pn->right = in;
547 else
548 pn->left = in;
550 ln->fn_bit = plen;
552 ln->parent = in;
553 fn->parent = in;
555 ln->fn_sernum = sernum;
557 if (addr_bit_set(addr, bit)) {
558 in->right = ln;
559 in->left = fn;
560 } else {
561 in->left = ln;
562 in->right = fn;
564 } else { /* plen <= bit */
567 * (new leaf node)[ln]
568 * / \
569 * (old node)[fn] NULL
572 ln = node_alloc();
574 if (ln == NULL)
575 return NULL;
577 ln->fn_bit = plen;
579 ln->parent = pn;
581 ln->fn_sernum = sernum;
583 if (dir)
584 pn->right = ln;
585 else
586 pn->left = ln;
588 if (addr_bit_set(&key->addr, plen))
589 ln->right = fn;
590 else
591 ln->left = fn;
593 fn->parent = ln;
595 return ln;
599 * Insert routing information in a node.
602 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
603 struct nl_info *info)
605 struct rt6_info *iter = NULL;
606 struct rt6_info **ins;
608 ins = &fn->leaf;
610 for (iter = fn->leaf; iter; iter=iter->u.dst.rt6_next) {
612 * Search for duplicates
615 if (iter->rt6i_metric == rt->rt6i_metric) {
617 * Same priority level
620 if (iter->rt6i_dev == rt->rt6i_dev &&
621 iter->rt6i_idev == rt->rt6i_idev &&
622 ipv6_addr_equal(&iter->rt6i_gateway,
623 &rt->rt6i_gateway)) {
624 if (!(iter->rt6i_flags&RTF_EXPIRES))
625 return -EEXIST;
626 iter->rt6i_expires = rt->rt6i_expires;
627 if (!(rt->rt6i_flags&RTF_EXPIRES)) {
628 iter->rt6i_flags &= ~RTF_EXPIRES;
629 iter->rt6i_expires = 0;
631 return -EEXIST;
635 if (iter->rt6i_metric > rt->rt6i_metric)
636 break;
638 ins = &iter->u.dst.rt6_next;
641 /* Reset round-robin state, if necessary */
642 if (ins == &fn->leaf)
643 fn->rr_ptr = NULL;
646 * insert node
649 rt->u.dst.rt6_next = iter;
650 *ins = rt;
651 rt->rt6i_node = fn;
652 atomic_inc(&rt->rt6i_ref);
653 inet6_rt_notify(RTM_NEWROUTE, rt, info);
654 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
656 if ((fn->fn_flags & RTN_RTINFO) == 0) {
657 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
658 fn->fn_flags |= RTN_RTINFO;
661 return 0;
664 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
666 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
667 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE)))
668 mod_timer(&net->ipv6.ip6_fib_timer,
669 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
672 void fib6_force_start_gc(struct net *net)
674 if (!timer_pending(&net->ipv6.ip6_fib_timer))
675 mod_timer(&net->ipv6.ip6_fib_timer,
676 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
680 * Add routing information to the routing tree.
681 * <destination addr>/<source addr>
682 * with source addr info in sub-trees
685 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
687 struct fib6_node *fn, *pn = NULL;
688 int err = -ENOMEM;
690 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
691 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst));
693 if (fn == NULL)
694 goto out;
696 pn = fn;
698 #ifdef CONFIG_IPV6_SUBTREES
699 if (rt->rt6i_src.plen) {
700 struct fib6_node *sn;
702 if (fn->subtree == NULL) {
703 struct fib6_node *sfn;
706 * Create subtree.
708 * fn[main tree]
710 * sfn[subtree root]
712 * sn[new leaf node]
715 /* Create subtree root node */
716 sfn = node_alloc();
717 if (sfn == NULL)
718 goto st_failure;
720 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
721 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
722 sfn->fn_flags = RTN_ROOT;
723 sfn->fn_sernum = fib6_new_sernum();
725 /* Now add the first leaf node to new subtree */
727 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
728 sizeof(struct in6_addr), rt->rt6i_src.plen,
729 offsetof(struct rt6_info, rt6i_src));
731 if (sn == NULL) {
732 /* If it is failed, discard just allocated
733 root, and then (in st_failure) stale node
734 in main tree.
736 node_free(sfn);
737 goto st_failure;
740 /* Now link new subtree to main tree */
741 sfn->parent = fn;
742 fn->subtree = sfn;
743 } else {
744 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
745 sizeof(struct in6_addr), rt->rt6i_src.plen,
746 offsetof(struct rt6_info, rt6i_src));
748 if (sn == NULL)
749 goto st_failure;
752 if (fn->leaf == NULL) {
753 fn->leaf = rt;
754 atomic_inc(&rt->rt6i_ref);
756 fn = sn;
758 #endif
760 err = fib6_add_rt2node(fn, rt, info);
762 if (err == 0) {
763 fib6_start_gc(info->nl_net, rt);
764 if (!(rt->rt6i_flags&RTF_CACHE))
765 fib6_prune_clones(info->nl_net, pn, rt);
768 out:
769 if (err) {
770 #ifdef CONFIG_IPV6_SUBTREES
772 * If fib6_add_1 has cleared the old leaf pointer in the
773 * super-tree leaf node we have to find a new one for it.
775 if (pn != fn && pn->leaf == rt) {
776 pn->leaf = NULL;
777 atomic_dec(&rt->rt6i_ref);
779 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
780 pn->leaf = fib6_find_prefix(info->nl_net, pn);
781 #if RT6_DEBUG >= 2
782 if (!pn->leaf) {
783 WARN_ON(pn->leaf == NULL);
784 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
786 #endif
787 atomic_inc(&pn->leaf->rt6i_ref);
789 #endif
790 dst_free(&rt->u.dst);
792 return err;
794 #ifdef CONFIG_IPV6_SUBTREES
795 /* Subtree creation failed, probably main tree node
796 is orphan. If it is, shoot it.
798 st_failure:
799 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
800 fib6_repair_tree(info->nl_net, fn);
801 dst_free(&rt->u.dst);
802 return err;
803 #endif
807 * Routing tree lookup
811 struct lookup_args {
812 int offset; /* key offset on rt6_info */
813 struct in6_addr *addr; /* search key */
816 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
817 struct lookup_args *args)
819 struct fib6_node *fn;
820 __be32 dir;
822 if (unlikely(args->offset == 0))
823 return NULL;
826 * Descend on a tree
829 fn = root;
831 for (;;) {
832 struct fib6_node *next;
834 dir = addr_bit_set(args->addr, fn->fn_bit);
836 next = dir ? fn->right : fn->left;
838 if (next) {
839 fn = next;
840 continue;
843 break;
846 while(fn) {
847 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
848 struct rt6key *key;
850 key = (struct rt6key *) ((u8 *) fn->leaf +
851 args->offset);
853 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
854 #ifdef CONFIG_IPV6_SUBTREES
855 if (fn->subtree)
856 fn = fib6_lookup_1(fn->subtree, args + 1);
857 #endif
858 if (!fn || fn->fn_flags & RTN_RTINFO)
859 return fn;
863 if (fn->fn_flags & RTN_ROOT)
864 break;
866 fn = fn->parent;
869 return NULL;
872 struct fib6_node * fib6_lookup(struct fib6_node *root, struct in6_addr *daddr,
873 struct in6_addr *saddr)
875 struct fib6_node *fn;
876 struct lookup_args args[] = {
878 .offset = offsetof(struct rt6_info, rt6i_dst),
879 .addr = daddr,
881 #ifdef CONFIG_IPV6_SUBTREES
883 .offset = offsetof(struct rt6_info, rt6i_src),
884 .addr = saddr,
886 #endif
888 .offset = 0, /* sentinel */
892 fn = fib6_lookup_1(root, daddr ? args : args + 1);
894 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT)
895 fn = root;
897 return fn;
901 * Get node with specified destination prefix (and source prefix,
902 * if subtrees are used)
906 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
907 struct in6_addr *addr,
908 int plen, int offset)
910 struct fib6_node *fn;
912 for (fn = root; fn ; ) {
913 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
916 * Prefix match
918 if (plen < fn->fn_bit ||
919 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
920 return NULL;
922 if (plen == fn->fn_bit)
923 return fn;
926 * We have more bits to go
928 if (addr_bit_set(addr, fn->fn_bit))
929 fn = fn->right;
930 else
931 fn = fn->left;
933 return NULL;
936 struct fib6_node * fib6_locate(struct fib6_node *root,
937 struct in6_addr *daddr, int dst_len,
938 struct in6_addr *saddr, int src_len)
940 struct fib6_node *fn;
942 fn = fib6_locate_1(root, daddr, dst_len,
943 offsetof(struct rt6_info, rt6i_dst));
945 #ifdef CONFIG_IPV6_SUBTREES
946 if (src_len) {
947 WARN_ON(saddr == NULL);
948 if (fn && fn->subtree)
949 fn = fib6_locate_1(fn->subtree, saddr, src_len,
950 offsetof(struct rt6_info, rt6i_src));
952 #endif
954 if (fn && fn->fn_flags&RTN_RTINFO)
955 return fn;
957 return NULL;
962 * Deletion
966 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
968 if (fn->fn_flags&RTN_ROOT)
969 return net->ipv6.ip6_null_entry;
971 while(fn) {
972 if(fn->left)
973 return fn->left->leaf;
975 if(fn->right)
976 return fn->right->leaf;
978 fn = FIB6_SUBTREE(fn);
980 return NULL;
984 * Called to trim the tree of intermediate nodes when possible. "fn"
985 * is the node we want to try and remove.
988 static struct fib6_node *fib6_repair_tree(struct net *net,
989 struct fib6_node *fn)
991 int children;
992 int nstate;
993 struct fib6_node *child, *pn;
994 struct fib6_walker_t *w;
995 int iter = 0;
997 for (;;) {
998 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
999 iter++;
1001 WARN_ON(fn->fn_flags & RTN_RTINFO);
1002 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1003 WARN_ON(fn->leaf != NULL);
1005 children = 0;
1006 child = NULL;
1007 if (fn->right) child = fn->right, children |= 1;
1008 if (fn->left) child = fn->left, children |= 2;
1010 if (children == 3 || FIB6_SUBTREE(fn)
1011 #ifdef CONFIG_IPV6_SUBTREES
1012 /* Subtree root (i.e. fn) may have one child */
1013 || (children && fn->fn_flags&RTN_ROOT)
1014 #endif
1016 fn->leaf = fib6_find_prefix(net, fn);
1017 #if RT6_DEBUG >= 2
1018 if (fn->leaf==NULL) {
1019 WARN_ON(!fn->leaf);
1020 fn->leaf = net->ipv6.ip6_null_entry;
1022 #endif
1023 atomic_inc(&fn->leaf->rt6i_ref);
1024 return fn->parent;
1027 pn = fn->parent;
1028 #ifdef CONFIG_IPV6_SUBTREES
1029 if (FIB6_SUBTREE(pn) == fn) {
1030 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1031 FIB6_SUBTREE(pn) = NULL;
1032 nstate = FWS_L;
1033 } else {
1034 WARN_ON(fn->fn_flags & RTN_ROOT);
1035 #endif
1036 if (pn->right == fn) pn->right = child;
1037 else if (pn->left == fn) pn->left = child;
1038 #if RT6_DEBUG >= 2
1039 else
1040 WARN_ON(1);
1041 #endif
1042 if (child)
1043 child->parent = pn;
1044 nstate = FWS_R;
1045 #ifdef CONFIG_IPV6_SUBTREES
1047 #endif
1049 read_lock(&fib6_walker_lock);
1050 FOR_WALKERS(w) {
1051 if (child == NULL) {
1052 if (w->root == fn) {
1053 w->root = w->node = NULL;
1054 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1055 } else if (w->node == fn) {
1056 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1057 w->node = pn;
1058 w->state = nstate;
1060 } else {
1061 if (w->root == fn) {
1062 w->root = child;
1063 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1065 if (w->node == fn) {
1066 w->node = child;
1067 if (children&2) {
1068 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1069 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1070 } else {
1071 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1072 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1077 read_unlock(&fib6_walker_lock);
1079 node_free(fn);
1080 if (pn->fn_flags&RTN_RTINFO || FIB6_SUBTREE(pn))
1081 return pn;
1083 rt6_release(pn->leaf);
1084 pn->leaf = NULL;
1085 fn = pn;
1089 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1090 struct nl_info *info)
1092 struct fib6_walker_t *w;
1093 struct rt6_info *rt = *rtp;
1094 struct net *net = info->nl_net;
1096 RT6_TRACE("fib6_del_route\n");
1098 /* Unlink it */
1099 *rtp = rt->u.dst.rt6_next;
1100 rt->rt6i_node = NULL;
1101 net->ipv6.rt6_stats->fib_rt_entries--;
1102 net->ipv6.rt6_stats->fib_discarded_routes++;
1104 /* Reset round-robin state, if necessary */
1105 if (fn->rr_ptr == rt)
1106 fn->rr_ptr = NULL;
1108 /* Adjust walkers */
1109 read_lock(&fib6_walker_lock);
1110 FOR_WALKERS(w) {
1111 if (w->state == FWS_C && w->leaf == rt) {
1112 RT6_TRACE("walker %p adjusted by delroute\n", w);
1113 w->leaf = rt->u.dst.rt6_next;
1114 if (w->leaf == NULL)
1115 w->state = FWS_U;
1118 read_unlock(&fib6_walker_lock);
1120 rt->u.dst.rt6_next = NULL;
1122 /* If it was last route, expunge its radix tree node */
1123 if (fn->leaf == NULL) {
1124 fn->fn_flags &= ~RTN_RTINFO;
1125 net->ipv6.rt6_stats->fib_route_nodes--;
1126 fn = fib6_repair_tree(net, fn);
1129 if (atomic_read(&rt->rt6i_ref) != 1) {
1130 /* This route is used as dummy address holder in some split
1131 * nodes. It is not leaked, but it still holds other resources,
1132 * which must be released in time. So, scan ascendant nodes
1133 * and replace dummy references to this route with references
1134 * to still alive ones.
1136 while (fn) {
1137 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) {
1138 fn->leaf = fib6_find_prefix(net, fn);
1139 atomic_inc(&fn->leaf->rt6i_ref);
1140 rt6_release(rt);
1142 fn = fn->parent;
1144 /* No more references are possible at this point. */
1145 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1148 inet6_rt_notify(RTM_DELROUTE, rt, info);
1149 rt6_release(rt);
1152 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1154 struct net *net = info->nl_net;
1155 struct fib6_node *fn = rt->rt6i_node;
1156 struct rt6_info **rtp;
1158 #if RT6_DEBUG >= 2
1159 if (rt->u.dst.obsolete>0) {
1160 WARN_ON(fn != NULL);
1161 return -ENOENT;
1163 #endif
1164 if (fn == NULL || rt == net->ipv6.ip6_null_entry)
1165 return -ENOENT;
1167 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1169 if (!(rt->rt6i_flags&RTF_CACHE)) {
1170 struct fib6_node *pn = fn;
1171 #ifdef CONFIG_IPV6_SUBTREES
1172 /* clones of this route might be in another subtree */
1173 if (rt->rt6i_src.plen) {
1174 while (!(pn->fn_flags&RTN_ROOT))
1175 pn = pn->parent;
1176 pn = pn->parent;
1178 #endif
1179 fib6_prune_clones(info->nl_net, pn, rt);
1183 * Walk the leaf entries looking for ourself
1186 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->u.dst.rt6_next) {
1187 if (*rtp == rt) {
1188 fib6_del_route(fn, rtp, info);
1189 return 0;
1192 return -ENOENT;
1196 * Tree traversal function.
1198 * Certainly, it is not interrupt safe.
1199 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1200 * It means, that we can modify tree during walking
1201 * and use this function for garbage collection, clone pruning,
1202 * cleaning tree when a device goes down etc. etc.
1204 * It guarantees that every node will be traversed,
1205 * and that it will be traversed only once.
1207 * Callback function w->func may return:
1208 * 0 -> continue walking.
1209 * positive value -> walking is suspended (used by tree dumps,
1210 * and probably by gc, if it will be split to several slices)
1211 * negative value -> terminate walking.
1213 * The function itself returns:
1214 * 0 -> walk is complete.
1215 * >0 -> walk is incomplete (i.e. suspended)
1216 * <0 -> walk is terminated by an error.
1219 static int fib6_walk_continue(struct fib6_walker_t *w)
1221 struct fib6_node *fn, *pn;
1223 for (;;) {
1224 fn = w->node;
1225 if (fn == NULL)
1226 return 0;
1228 if (w->prune && fn != w->root &&
1229 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) {
1230 w->state = FWS_C;
1231 w->leaf = fn->leaf;
1233 switch (w->state) {
1234 #ifdef CONFIG_IPV6_SUBTREES
1235 case FWS_S:
1236 if (FIB6_SUBTREE(fn)) {
1237 w->node = FIB6_SUBTREE(fn);
1238 continue;
1240 w->state = FWS_L;
1241 #endif
1242 case FWS_L:
1243 if (fn->left) {
1244 w->node = fn->left;
1245 w->state = FWS_INIT;
1246 continue;
1248 w->state = FWS_R;
1249 case FWS_R:
1250 if (fn->right) {
1251 w->node = fn->right;
1252 w->state = FWS_INIT;
1253 continue;
1255 w->state = FWS_C;
1256 w->leaf = fn->leaf;
1257 case FWS_C:
1258 if (w->leaf && fn->fn_flags&RTN_RTINFO) {
1259 int err = w->func(w);
1260 if (err)
1261 return err;
1262 continue;
1264 w->state = FWS_U;
1265 case FWS_U:
1266 if (fn == w->root)
1267 return 0;
1268 pn = fn->parent;
1269 w->node = pn;
1270 #ifdef CONFIG_IPV6_SUBTREES
1271 if (FIB6_SUBTREE(pn) == fn) {
1272 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1273 w->state = FWS_L;
1274 continue;
1276 #endif
1277 if (pn->left == fn) {
1278 w->state = FWS_R;
1279 continue;
1281 if (pn->right == fn) {
1282 w->state = FWS_C;
1283 w->leaf = w->node->leaf;
1284 continue;
1286 #if RT6_DEBUG >= 2
1287 WARN_ON(1);
1288 #endif
1293 static int fib6_walk(struct fib6_walker_t *w)
1295 int res;
1297 w->state = FWS_INIT;
1298 w->node = w->root;
1300 fib6_walker_link(w);
1301 res = fib6_walk_continue(w);
1302 if (res <= 0)
1303 fib6_walker_unlink(w);
1304 return res;
1307 static int fib6_clean_node(struct fib6_walker_t *w)
1309 int res;
1310 struct rt6_info *rt;
1311 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1312 struct nl_info info = {
1313 .nl_net = c->net,
1316 for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) {
1317 res = c->func(rt, c->arg);
1318 if (res < 0) {
1319 w->leaf = rt;
1320 res = fib6_del(rt, &info);
1321 if (res) {
1322 #if RT6_DEBUG >= 2
1323 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res);
1324 #endif
1325 continue;
1327 return 0;
1329 WARN_ON(res != 0);
1331 w->leaf = rt;
1332 return 0;
1336 * Convenient frontend to tree walker.
1338 * func is called on each route.
1339 * It may return -1 -> delete this route.
1340 * 0 -> continue walking
1342 * prune==1 -> only immediate children of node (certainly,
1343 * ignoring pure split nodes) will be scanned.
1346 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1347 int (*func)(struct rt6_info *, void *arg),
1348 int prune, void *arg)
1350 struct fib6_cleaner_t c;
1352 c.w.root = root;
1353 c.w.func = fib6_clean_node;
1354 c.w.prune = prune;
1355 c.func = func;
1356 c.arg = arg;
1357 c.net = net;
1359 fib6_walk(&c.w);
1362 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1363 int prune, void *arg)
1365 struct fib6_table *table;
1366 struct hlist_node *node;
1367 struct hlist_head *head;
1368 unsigned int h;
1370 rcu_read_lock();
1371 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
1372 head = &net->ipv6.fib_table_hash[h];
1373 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1374 write_lock_bh(&table->tb6_lock);
1375 fib6_clean_tree(net, &table->tb6_root,
1376 func, prune, arg);
1377 write_unlock_bh(&table->tb6_lock);
1380 rcu_read_unlock();
1383 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1385 if (rt->rt6i_flags & RTF_CACHE) {
1386 RT6_TRACE("pruning clone %p\n", rt);
1387 return -1;
1390 return 0;
1393 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1394 struct rt6_info *rt)
1396 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1400 * Garbage collection
1403 static struct fib6_gc_args
1405 int timeout;
1406 int more;
1407 } gc_args;
1409 static int fib6_age(struct rt6_info *rt, void *arg)
1411 unsigned long now = jiffies;
1414 * check addrconf expiration here.
1415 * Routes are expired even if they are in use.
1417 * Also age clones. Note, that clones are aged out
1418 * only if they are not in use now.
1421 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) {
1422 if (time_after(now, rt->rt6i_expires)) {
1423 RT6_TRACE("expiring %p\n", rt);
1424 return -1;
1426 gc_args.more++;
1427 } else if (rt->rt6i_flags & RTF_CACHE) {
1428 if (atomic_read(&rt->u.dst.__refcnt) == 0 &&
1429 time_after_eq(now, rt->u.dst.lastuse + gc_args.timeout)) {
1430 RT6_TRACE("aging clone %p\n", rt);
1431 return -1;
1432 } else if ((rt->rt6i_flags & RTF_GATEWAY) &&
1433 (!(rt->rt6i_nexthop->flags & NTF_ROUTER))) {
1434 RT6_TRACE("purging route %p via non-router but gateway\n",
1435 rt);
1436 return -1;
1438 gc_args.more++;
1441 return 0;
1444 static DEFINE_SPINLOCK(fib6_gc_lock);
1446 void fib6_run_gc(unsigned long expires, struct net *net)
1448 if (expires != ~0UL) {
1449 spin_lock_bh(&fib6_gc_lock);
1450 gc_args.timeout = expires ? (int)expires :
1451 net->ipv6.sysctl.ip6_rt_gc_interval;
1452 } else {
1453 if (!spin_trylock_bh(&fib6_gc_lock)) {
1454 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1455 return;
1457 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1460 gc_args.more = icmp6_dst_gc();
1462 fib6_clean_all(net, fib6_age, 0, NULL);
1464 if (gc_args.more)
1465 mod_timer(&net->ipv6.ip6_fib_timer,
1466 round_jiffies(jiffies
1467 + net->ipv6.sysctl.ip6_rt_gc_interval));
1468 else
1469 del_timer(&net->ipv6.ip6_fib_timer);
1470 spin_unlock_bh(&fib6_gc_lock);
1473 static void fib6_gc_timer_cb(unsigned long arg)
1475 fib6_run_gc(0, (struct net *)arg);
1478 static int fib6_net_init(struct net *net)
1480 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1482 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1483 if (!net->ipv6.rt6_stats)
1484 goto out_timer;
1486 net->ipv6.fib_table_hash = kcalloc(FIB_TABLE_HASHSZ,
1487 sizeof(*net->ipv6.fib_table_hash),
1488 GFP_KERNEL);
1489 if (!net->ipv6.fib_table_hash)
1490 goto out_rt6_stats;
1492 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1493 GFP_KERNEL);
1494 if (!net->ipv6.fib6_main_tbl)
1495 goto out_fib_table_hash;
1497 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1498 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1499 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1500 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1502 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1503 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1504 GFP_KERNEL);
1505 if (!net->ipv6.fib6_local_tbl)
1506 goto out_fib6_main_tbl;
1507 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1508 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1509 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1510 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1511 #endif
1512 fib6_tables_init(net);
1514 return 0;
1516 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1517 out_fib6_main_tbl:
1518 kfree(net->ipv6.fib6_main_tbl);
1519 #endif
1520 out_fib_table_hash:
1521 kfree(net->ipv6.fib_table_hash);
1522 out_rt6_stats:
1523 kfree(net->ipv6.rt6_stats);
1524 out_timer:
1525 return -ENOMEM;
1528 static void fib6_net_exit(struct net *net)
1530 rt6_ifdown(net, NULL);
1531 del_timer_sync(&net->ipv6.ip6_fib_timer);
1533 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1534 kfree(net->ipv6.fib6_local_tbl);
1535 #endif
1536 kfree(net->ipv6.fib6_main_tbl);
1537 kfree(net->ipv6.fib_table_hash);
1538 kfree(net->ipv6.rt6_stats);
1541 static struct pernet_operations fib6_net_ops = {
1542 .init = fib6_net_init,
1543 .exit = fib6_net_exit,
1546 int __init fib6_init(void)
1548 int ret = -ENOMEM;
1550 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1551 sizeof(struct fib6_node),
1552 0, SLAB_HWCACHE_ALIGN,
1553 NULL);
1554 if (!fib6_node_kmem)
1555 goto out;
1557 ret = register_pernet_subsys(&fib6_net_ops);
1558 if (ret)
1559 goto out_kmem_cache_create;
1561 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib);
1562 if (ret)
1563 goto out_unregister_subsys;
1564 out:
1565 return ret;
1567 out_unregister_subsys:
1568 unregister_pernet_subsys(&fib6_net_ops);
1569 out_kmem_cache_create:
1570 kmem_cache_destroy(fib6_node_kmem);
1571 goto out;
1574 void fib6_gc_cleanup(void)
1576 unregister_pernet_subsys(&fib6_net_ops);
1577 kmem_cache_destroy(fib6_node_kmem);