ixgbe: fix ethtool register dump
[linux-2.6/mini2440.git] / net / ipv6 / ip6_fib.c
blob29c7c99e69f7611034354319b9e6452f26972635
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 cb->args[2] = 0;
302 kfree(w);
304 cb->done = (void*)cb->args[3];
305 cb->args[1] = 3;
308 static int fib6_dump_done(struct netlink_callback *cb)
310 fib6_dump_end(cb);
311 return cb->done ? cb->done(cb) : 0;
314 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
315 struct netlink_callback *cb)
317 struct fib6_walker_t *w;
318 int res;
320 w = (void *)cb->args[2];
321 w->root = &table->tb6_root;
323 if (cb->args[4] == 0) {
324 read_lock_bh(&table->tb6_lock);
325 res = fib6_walk(w);
326 read_unlock_bh(&table->tb6_lock);
327 if (res > 0)
328 cb->args[4] = 1;
329 } else {
330 read_lock_bh(&table->tb6_lock);
331 res = fib6_walk_continue(w);
332 read_unlock_bh(&table->tb6_lock);
333 if (res != 0) {
334 if (res < 0)
335 fib6_walker_unlink(w);
336 goto end;
338 fib6_walker_unlink(w);
339 cb->args[4] = 0;
341 end:
342 return res;
345 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
347 struct net *net = sock_net(skb->sk);
348 unsigned int h, s_h;
349 unsigned int e = 0, s_e;
350 struct rt6_rtnl_dump_arg arg;
351 struct fib6_walker_t *w;
352 struct fib6_table *tb;
353 struct hlist_node *node;
354 struct hlist_head *head;
355 int res = 0;
357 s_h = cb->args[0];
358 s_e = cb->args[1];
360 w = (void *)cb->args[2];
361 if (w == NULL) {
362 /* New dump:
364 * 1. hook callback destructor.
366 cb->args[3] = (long)cb->done;
367 cb->done = fib6_dump_done;
370 * 2. allocate and initialize walker.
372 w = kzalloc(sizeof(*w), GFP_ATOMIC);
373 if (w == NULL)
374 return -ENOMEM;
375 w->func = fib6_dump_node;
376 cb->args[2] = (long)w;
379 arg.skb = skb;
380 arg.cb = cb;
381 arg.net = net;
382 w->args = &arg;
384 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
385 e = 0;
386 head = &net->ipv6.fib_table_hash[h];
387 hlist_for_each_entry(tb, node, head, tb6_hlist) {
388 if (e < s_e)
389 goto next;
390 res = fib6_dump_table(tb, skb, cb);
391 if (res != 0)
392 goto out;
393 next:
394 e++;
397 out:
398 cb->args[1] = e;
399 cb->args[0] = h;
401 res = res < 0 ? res : skb->len;
402 if (res <= 0)
403 fib6_dump_end(cb);
404 return res;
408 * Routing Table
410 * return the appropriate node for a routing tree "add" operation
411 * by either creating and inserting or by returning an existing
412 * node.
415 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
416 int addrlen, int plen,
417 int offset)
419 struct fib6_node *fn, *in, *ln;
420 struct fib6_node *pn = NULL;
421 struct rt6key *key;
422 int bit;
423 __be32 dir = 0;
424 __u32 sernum = fib6_new_sernum();
426 RT6_TRACE("fib6_add_1\n");
428 /* insert node in tree */
430 fn = root;
432 do {
433 key = (struct rt6key *)((u8 *)fn->leaf + offset);
436 * Prefix match
438 if (plen < fn->fn_bit ||
439 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
440 goto insert_above;
443 * Exact match ?
446 if (plen == fn->fn_bit) {
447 /* clean up an intermediate node */
448 if ((fn->fn_flags & RTN_RTINFO) == 0) {
449 rt6_release(fn->leaf);
450 fn->leaf = NULL;
453 fn->fn_sernum = sernum;
455 return fn;
459 * We have more bits to go
462 /* Try to walk down on tree. */
463 fn->fn_sernum = sernum;
464 dir = addr_bit_set(addr, fn->fn_bit);
465 pn = fn;
466 fn = dir ? fn->right: fn->left;
467 } while (fn);
470 * We walked to the bottom of tree.
471 * Create new leaf node without children.
474 ln = node_alloc();
476 if (ln == NULL)
477 return NULL;
478 ln->fn_bit = plen;
480 ln->parent = pn;
481 ln->fn_sernum = sernum;
483 if (dir)
484 pn->right = ln;
485 else
486 pn->left = ln;
488 return ln;
491 insert_above:
493 * split since we don't have a common prefix anymore or
494 * we have a less significant route.
495 * we've to insert an intermediate node on the list
496 * this new node will point to the one we need to create
497 * and the current
500 pn = fn->parent;
502 /* find 1st bit in difference between the 2 addrs.
504 See comment in __ipv6_addr_diff: bit may be an invalid value,
505 but if it is >= plen, the value is ignored in any case.
508 bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
511 * (intermediate)[in]
512 * / \
513 * (new leaf node)[ln] (old node)[fn]
515 if (plen > bit) {
516 in = node_alloc();
517 ln = node_alloc();
519 if (in == NULL || ln == NULL) {
520 if (in)
521 node_free(in);
522 if (ln)
523 node_free(ln);
524 return NULL;
528 * new intermediate node.
529 * RTN_RTINFO will
530 * be off since that an address that chooses one of
531 * the branches would not match less specific routes
532 * in the other branch
535 in->fn_bit = bit;
537 in->parent = pn;
538 in->leaf = fn->leaf;
539 atomic_inc(&in->leaf->rt6i_ref);
541 in->fn_sernum = sernum;
543 /* update parent pointer */
544 if (dir)
545 pn->right = in;
546 else
547 pn->left = in;
549 ln->fn_bit = plen;
551 ln->parent = in;
552 fn->parent = in;
554 ln->fn_sernum = sernum;
556 if (addr_bit_set(addr, bit)) {
557 in->right = ln;
558 in->left = fn;
559 } else {
560 in->left = ln;
561 in->right = fn;
563 } else { /* plen <= bit */
566 * (new leaf node)[ln]
567 * / \
568 * (old node)[fn] NULL
571 ln = node_alloc();
573 if (ln == NULL)
574 return NULL;
576 ln->fn_bit = plen;
578 ln->parent = pn;
580 ln->fn_sernum = sernum;
582 if (dir)
583 pn->right = ln;
584 else
585 pn->left = ln;
587 if (addr_bit_set(&key->addr, plen))
588 ln->right = fn;
589 else
590 ln->left = fn;
592 fn->parent = ln;
594 return ln;
598 * Insert routing information in a node.
601 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
602 struct nl_info *info)
604 struct rt6_info *iter = NULL;
605 struct rt6_info **ins;
607 ins = &fn->leaf;
609 for (iter = fn->leaf; iter; iter=iter->u.dst.rt6_next) {
611 * Search for duplicates
614 if (iter->rt6i_metric == rt->rt6i_metric) {
616 * Same priority level
619 if (iter->rt6i_dev == rt->rt6i_dev &&
620 iter->rt6i_idev == rt->rt6i_idev &&
621 ipv6_addr_equal(&iter->rt6i_gateway,
622 &rt->rt6i_gateway)) {
623 if (!(iter->rt6i_flags&RTF_EXPIRES))
624 return -EEXIST;
625 iter->rt6i_expires = rt->rt6i_expires;
626 if (!(rt->rt6i_flags&RTF_EXPIRES)) {
627 iter->rt6i_flags &= ~RTF_EXPIRES;
628 iter->rt6i_expires = 0;
630 return -EEXIST;
634 if (iter->rt6i_metric > rt->rt6i_metric)
635 break;
637 ins = &iter->u.dst.rt6_next;
640 /* Reset round-robin state, if necessary */
641 if (ins == &fn->leaf)
642 fn->rr_ptr = NULL;
645 * insert node
648 rt->u.dst.rt6_next = iter;
649 *ins = rt;
650 rt->rt6i_node = fn;
651 atomic_inc(&rt->rt6i_ref);
652 inet6_rt_notify(RTM_NEWROUTE, rt, info);
653 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
655 if ((fn->fn_flags & RTN_RTINFO) == 0) {
656 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
657 fn->fn_flags |= RTN_RTINFO;
660 return 0;
663 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
665 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
666 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE)))
667 mod_timer(&net->ipv6.ip6_fib_timer,
668 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
671 void fib6_force_start_gc(struct net *net)
673 if (!timer_pending(&net->ipv6.ip6_fib_timer))
674 mod_timer(&net->ipv6.ip6_fib_timer,
675 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
679 * Add routing information to the routing tree.
680 * <destination addr>/<source addr>
681 * with source addr info in sub-trees
684 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
686 struct fib6_node *fn, *pn = NULL;
687 int err = -ENOMEM;
689 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
690 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst));
692 if (fn == NULL)
693 goto out;
695 pn = fn;
697 #ifdef CONFIG_IPV6_SUBTREES
698 if (rt->rt6i_src.plen) {
699 struct fib6_node *sn;
701 if (fn->subtree == NULL) {
702 struct fib6_node *sfn;
705 * Create subtree.
707 * fn[main tree]
709 * sfn[subtree root]
711 * sn[new leaf node]
714 /* Create subtree root node */
715 sfn = node_alloc();
716 if (sfn == NULL)
717 goto st_failure;
719 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
720 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
721 sfn->fn_flags = RTN_ROOT;
722 sfn->fn_sernum = fib6_new_sernum();
724 /* Now add the first leaf node to new subtree */
726 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
727 sizeof(struct in6_addr), rt->rt6i_src.plen,
728 offsetof(struct rt6_info, rt6i_src));
730 if (sn == NULL) {
731 /* If it is failed, discard just allocated
732 root, and then (in st_failure) stale node
733 in main tree.
735 node_free(sfn);
736 goto st_failure;
739 /* Now link new subtree to main tree */
740 sfn->parent = fn;
741 fn->subtree = sfn;
742 } else {
743 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
744 sizeof(struct in6_addr), rt->rt6i_src.plen,
745 offsetof(struct rt6_info, rt6i_src));
747 if (sn == NULL)
748 goto st_failure;
751 if (fn->leaf == NULL) {
752 fn->leaf = rt;
753 atomic_inc(&rt->rt6i_ref);
755 fn = sn;
757 #endif
759 err = fib6_add_rt2node(fn, rt, info);
761 if (err == 0) {
762 fib6_start_gc(info->nl_net, rt);
763 if (!(rt->rt6i_flags&RTF_CACHE))
764 fib6_prune_clones(info->nl_net, pn, rt);
767 out:
768 if (err) {
769 #ifdef CONFIG_IPV6_SUBTREES
771 * If fib6_add_1 has cleared the old leaf pointer in the
772 * super-tree leaf node we have to find a new one for it.
774 if (pn != fn && pn->leaf == rt) {
775 pn->leaf = NULL;
776 atomic_dec(&rt->rt6i_ref);
778 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
779 pn->leaf = fib6_find_prefix(info->nl_net, pn);
780 #if RT6_DEBUG >= 2
781 if (!pn->leaf) {
782 WARN_ON(pn->leaf == NULL);
783 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
785 #endif
786 atomic_inc(&pn->leaf->rt6i_ref);
788 #endif
789 dst_free(&rt->u.dst);
791 return err;
793 #ifdef CONFIG_IPV6_SUBTREES
794 /* Subtree creation failed, probably main tree node
795 is orphan. If it is, shoot it.
797 st_failure:
798 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
799 fib6_repair_tree(info->nl_net, fn);
800 dst_free(&rt->u.dst);
801 return err;
802 #endif
806 * Routing tree lookup
810 struct lookup_args {
811 int offset; /* key offset on rt6_info */
812 struct in6_addr *addr; /* search key */
815 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
816 struct lookup_args *args)
818 struct fib6_node *fn;
819 __be32 dir;
821 if (unlikely(args->offset == 0))
822 return NULL;
825 * Descend on a tree
828 fn = root;
830 for (;;) {
831 struct fib6_node *next;
833 dir = addr_bit_set(args->addr, fn->fn_bit);
835 next = dir ? fn->right : fn->left;
837 if (next) {
838 fn = next;
839 continue;
842 break;
845 while(fn) {
846 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
847 struct rt6key *key;
849 key = (struct rt6key *) ((u8 *) fn->leaf +
850 args->offset);
852 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
853 #ifdef CONFIG_IPV6_SUBTREES
854 if (fn->subtree)
855 fn = fib6_lookup_1(fn->subtree, args + 1);
856 #endif
857 if (!fn || fn->fn_flags & RTN_RTINFO)
858 return fn;
862 if (fn->fn_flags & RTN_ROOT)
863 break;
865 fn = fn->parent;
868 return NULL;
871 struct fib6_node * fib6_lookup(struct fib6_node *root, struct in6_addr *daddr,
872 struct in6_addr *saddr)
874 struct fib6_node *fn;
875 struct lookup_args args[] = {
877 .offset = offsetof(struct rt6_info, rt6i_dst),
878 .addr = daddr,
880 #ifdef CONFIG_IPV6_SUBTREES
882 .offset = offsetof(struct rt6_info, rt6i_src),
883 .addr = saddr,
885 #endif
887 .offset = 0, /* sentinel */
891 fn = fib6_lookup_1(root, daddr ? args : args + 1);
893 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT)
894 fn = root;
896 return fn;
900 * Get node with specified destination prefix (and source prefix,
901 * if subtrees are used)
905 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
906 struct in6_addr *addr,
907 int plen, int offset)
909 struct fib6_node *fn;
911 for (fn = root; fn ; ) {
912 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
915 * Prefix match
917 if (plen < fn->fn_bit ||
918 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
919 return NULL;
921 if (plen == fn->fn_bit)
922 return fn;
925 * We have more bits to go
927 if (addr_bit_set(addr, fn->fn_bit))
928 fn = fn->right;
929 else
930 fn = fn->left;
932 return NULL;
935 struct fib6_node * fib6_locate(struct fib6_node *root,
936 struct in6_addr *daddr, int dst_len,
937 struct in6_addr *saddr, int src_len)
939 struct fib6_node *fn;
941 fn = fib6_locate_1(root, daddr, dst_len,
942 offsetof(struct rt6_info, rt6i_dst));
944 #ifdef CONFIG_IPV6_SUBTREES
945 if (src_len) {
946 WARN_ON(saddr == NULL);
947 if (fn && fn->subtree)
948 fn = fib6_locate_1(fn->subtree, saddr, src_len,
949 offsetof(struct rt6_info, rt6i_src));
951 #endif
953 if (fn && fn->fn_flags&RTN_RTINFO)
954 return fn;
956 return NULL;
961 * Deletion
965 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
967 if (fn->fn_flags&RTN_ROOT)
968 return net->ipv6.ip6_null_entry;
970 while(fn) {
971 if(fn->left)
972 return fn->left->leaf;
974 if(fn->right)
975 return fn->right->leaf;
977 fn = FIB6_SUBTREE(fn);
979 return NULL;
983 * Called to trim the tree of intermediate nodes when possible. "fn"
984 * is the node we want to try and remove.
987 static struct fib6_node *fib6_repair_tree(struct net *net,
988 struct fib6_node *fn)
990 int children;
991 int nstate;
992 struct fib6_node *child, *pn;
993 struct fib6_walker_t *w;
994 int iter = 0;
996 for (;;) {
997 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
998 iter++;
1000 WARN_ON(fn->fn_flags & RTN_RTINFO);
1001 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1002 WARN_ON(fn->leaf != NULL);
1004 children = 0;
1005 child = NULL;
1006 if (fn->right) child = fn->right, children |= 1;
1007 if (fn->left) child = fn->left, children |= 2;
1009 if (children == 3 || FIB6_SUBTREE(fn)
1010 #ifdef CONFIG_IPV6_SUBTREES
1011 /* Subtree root (i.e. fn) may have one child */
1012 || (children && fn->fn_flags&RTN_ROOT)
1013 #endif
1015 fn->leaf = fib6_find_prefix(net, fn);
1016 #if RT6_DEBUG >= 2
1017 if (fn->leaf==NULL) {
1018 WARN_ON(!fn->leaf);
1019 fn->leaf = net->ipv6.ip6_null_entry;
1021 #endif
1022 atomic_inc(&fn->leaf->rt6i_ref);
1023 return fn->parent;
1026 pn = fn->parent;
1027 #ifdef CONFIG_IPV6_SUBTREES
1028 if (FIB6_SUBTREE(pn) == fn) {
1029 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1030 FIB6_SUBTREE(pn) = NULL;
1031 nstate = FWS_L;
1032 } else {
1033 WARN_ON(fn->fn_flags & RTN_ROOT);
1034 #endif
1035 if (pn->right == fn) pn->right = child;
1036 else if (pn->left == fn) pn->left = child;
1037 #if RT6_DEBUG >= 2
1038 else
1039 WARN_ON(1);
1040 #endif
1041 if (child)
1042 child->parent = pn;
1043 nstate = FWS_R;
1044 #ifdef CONFIG_IPV6_SUBTREES
1046 #endif
1048 read_lock(&fib6_walker_lock);
1049 FOR_WALKERS(w) {
1050 if (child == NULL) {
1051 if (w->root == fn) {
1052 w->root = w->node = NULL;
1053 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1054 } else if (w->node == fn) {
1055 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1056 w->node = pn;
1057 w->state = nstate;
1059 } else {
1060 if (w->root == fn) {
1061 w->root = child;
1062 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1064 if (w->node == fn) {
1065 w->node = child;
1066 if (children&2) {
1067 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1068 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1069 } else {
1070 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1071 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1076 read_unlock(&fib6_walker_lock);
1078 node_free(fn);
1079 if (pn->fn_flags&RTN_RTINFO || FIB6_SUBTREE(pn))
1080 return pn;
1082 rt6_release(pn->leaf);
1083 pn->leaf = NULL;
1084 fn = pn;
1088 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1089 struct nl_info *info)
1091 struct fib6_walker_t *w;
1092 struct rt6_info *rt = *rtp;
1093 struct net *net = info->nl_net;
1095 RT6_TRACE("fib6_del_route\n");
1097 /* Unlink it */
1098 *rtp = rt->u.dst.rt6_next;
1099 rt->rt6i_node = NULL;
1100 net->ipv6.rt6_stats->fib_rt_entries--;
1101 net->ipv6.rt6_stats->fib_discarded_routes++;
1103 /* Reset round-robin state, if necessary */
1104 if (fn->rr_ptr == rt)
1105 fn->rr_ptr = NULL;
1107 /* Adjust walkers */
1108 read_lock(&fib6_walker_lock);
1109 FOR_WALKERS(w) {
1110 if (w->state == FWS_C && w->leaf == rt) {
1111 RT6_TRACE("walker %p adjusted by delroute\n", w);
1112 w->leaf = rt->u.dst.rt6_next;
1113 if (w->leaf == NULL)
1114 w->state = FWS_U;
1117 read_unlock(&fib6_walker_lock);
1119 rt->u.dst.rt6_next = NULL;
1121 /* If it was last route, expunge its radix tree node */
1122 if (fn->leaf == NULL) {
1123 fn->fn_flags &= ~RTN_RTINFO;
1124 net->ipv6.rt6_stats->fib_route_nodes--;
1125 fn = fib6_repair_tree(net, fn);
1128 if (atomic_read(&rt->rt6i_ref) != 1) {
1129 /* This route is used as dummy address holder in some split
1130 * nodes. It is not leaked, but it still holds other resources,
1131 * which must be released in time. So, scan ascendant nodes
1132 * and replace dummy references to this route with references
1133 * to still alive ones.
1135 while (fn) {
1136 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) {
1137 fn->leaf = fib6_find_prefix(net, fn);
1138 atomic_inc(&fn->leaf->rt6i_ref);
1139 rt6_release(rt);
1141 fn = fn->parent;
1143 /* No more references are possible at this point. */
1144 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1147 inet6_rt_notify(RTM_DELROUTE, rt, info);
1148 rt6_release(rt);
1151 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1153 struct net *net = info->nl_net;
1154 struct fib6_node *fn = rt->rt6i_node;
1155 struct rt6_info **rtp;
1157 #if RT6_DEBUG >= 2
1158 if (rt->u.dst.obsolete>0) {
1159 WARN_ON(fn != NULL);
1160 return -ENOENT;
1162 #endif
1163 if (fn == NULL || rt == net->ipv6.ip6_null_entry)
1164 return -ENOENT;
1166 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1168 if (!(rt->rt6i_flags&RTF_CACHE)) {
1169 struct fib6_node *pn = fn;
1170 #ifdef CONFIG_IPV6_SUBTREES
1171 /* clones of this route might be in another subtree */
1172 if (rt->rt6i_src.plen) {
1173 while (!(pn->fn_flags&RTN_ROOT))
1174 pn = pn->parent;
1175 pn = pn->parent;
1177 #endif
1178 fib6_prune_clones(info->nl_net, pn, rt);
1182 * Walk the leaf entries looking for ourself
1185 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->u.dst.rt6_next) {
1186 if (*rtp == rt) {
1187 fib6_del_route(fn, rtp, info);
1188 return 0;
1191 return -ENOENT;
1195 * Tree traversal function.
1197 * Certainly, it is not interrupt safe.
1198 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1199 * It means, that we can modify tree during walking
1200 * and use this function for garbage collection, clone pruning,
1201 * cleaning tree when a device goes down etc. etc.
1203 * It guarantees that every node will be traversed,
1204 * and that it will be traversed only once.
1206 * Callback function w->func may return:
1207 * 0 -> continue walking.
1208 * positive value -> walking is suspended (used by tree dumps,
1209 * and probably by gc, if it will be split to several slices)
1210 * negative value -> terminate walking.
1212 * The function itself returns:
1213 * 0 -> walk is complete.
1214 * >0 -> walk is incomplete (i.e. suspended)
1215 * <0 -> walk is terminated by an error.
1218 static int fib6_walk_continue(struct fib6_walker_t *w)
1220 struct fib6_node *fn, *pn;
1222 for (;;) {
1223 fn = w->node;
1224 if (fn == NULL)
1225 return 0;
1227 if (w->prune && fn != w->root &&
1228 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) {
1229 w->state = FWS_C;
1230 w->leaf = fn->leaf;
1232 switch (w->state) {
1233 #ifdef CONFIG_IPV6_SUBTREES
1234 case FWS_S:
1235 if (FIB6_SUBTREE(fn)) {
1236 w->node = FIB6_SUBTREE(fn);
1237 continue;
1239 w->state = FWS_L;
1240 #endif
1241 case FWS_L:
1242 if (fn->left) {
1243 w->node = fn->left;
1244 w->state = FWS_INIT;
1245 continue;
1247 w->state = FWS_R;
1248 case FWS_R:
1249 if (fn->right) {
1250 w->node = fn->right;
1251 w->state = FWS_INIT;
1252 continue;
1254 w->state = FWS_C;
1255 w->leaf = fn->leaf;
1256 case FWS_C:
1257 if (w->leaf && fn->fn_flags&RTN_RTINFO) {
1258 int err = w->func(w);
1259 if (err)
1260 return err;
1261 continue;
1263 w->state = FWS_U;
1264 case FWS_U:
1265 if (fn == w->root)
1266 return 0;
1267 pn = fn->parent;
1268 w->node = pn;
1269 #ifdef CONFIG_IPV6_SUBTREES
1270 if (FIB6_SUBTREE(pn) == fn) {
1271 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1272 w->state = FWS_L;
1273 continue;
1275 #endif
1276 if (pn->left == fn) {
1277 w->state = FWS_R;
1278 continue;
1280 if (pn->right == fn) {
1281 w->state = FWS_C;
1282 w->leaf = w->node->leaf;
1283 continue;
1285 #if RT6_DEBUG >= 2
1286 WARN_ON(1);
1287 #endif
1292 static int fib6_walk(struct fib6_walker_t *w)
1294 int res;
1296 w->state = FWS_INIT;
1297 w->node = w->root;
1299 fib6_walker_link(w);
1300 res = fib6_walk_continue(w);
1301 if (res <= 0)
1302 fib6_walker_unlink(w);
1303 return res;
1306 static int fib6_clean_node(struct fib6_walker_t *w)
1308 int res;
1309 struct rt6_info *rt;
1310 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1311 struct nl_info info = {
1312 .nl_net = c->net,
1315 for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) {
1316 res = c->func(rt, c->arg);
1317 if (res < 0) {
1318 w->leaf = rt;
1319 res = fib6_del(rt, &info);
1320 if (res) {
1321 #if RT6_DEBUG >= 2
1322 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res);
1323 #endif
1324 continue;
1326 return 0;
1328 WARN_ON(res != 0);
1330 w->leaf = rt;
1331 return 0;
1335 * Convenient frontend to tree walker.
1337 * func is called on each route.
1338 * It may return -1 -> delete this route.
1339 * 0 -> continue walking
1341 * prune==1 -> only immediate children of node (certainly,
1342 * ignoring pure split nodes) will be scanned.
1345 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1346 int (*func)(struct rt6_info *, void *arg),
1347 int prune, void *arg)
1349 struct fib6_cleaner_t c;
1351 c.w.root = root;
1352 c.w.func = fib6_clean_node;
1353 c.w.prune = prune;
1354 c.func = func;
1355 c.arg = arg;
1356 c.net = net;
1358 fib6_walk(&c.w);
1361 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1362 int prune, void *arg)
1364 struct fib6_table *table;
1365 struct hlist_node *node;
1366 struct hlist_head *head;
1367 unsigned int h;
1369 rcu_read_lock();
1370 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
1371 head = &net->ipv6.fib_table_hash[h];
1372 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1373 write_lock_bh(&table->tb6_lock);
1374 fib6_clean_tree(net, &table->tb6_root,
1375 func, prune, arg);
1376 write_unlock_bh(&table->tb6_lock);
1379 rcu_read_unlock();
1382 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1384 if (rt->rt6i_flags & RTF_CACHE) {
1385 RT6_TRACE("pruning clone %p\n", rt);
1386 return -1;
1389 return 0;
1392 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1393 struct rt6_info *rt)
1395 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1399 * Garbage collection
1402 static struct fib6_gc_args
1404 int timeout;
1405 int more;
1406 } gc_args;
1408 static int fib6_age(struct rt6_info *rt, void *arg)
1410 unsigned long now = jiffies;
1413 * check addrconf expiration here.
1414 * Routes are expired even if they are in use.
1416 * Also age clones. Note, that clones are aged out
1417 * only if they are not in use now.
1420 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) {
1421 if (time_after(now, rt->rt6i_expires)) {
1422 RT6_TRACE("expiring %p\n", rt);
1423 return -1;
1425 gc_args.more++;
1426 } else if (rt->rt6i_flags & RTF_CACHE) {
1427 if (atomic_read(&rt->u.dst.__refcnt) == 0 &&
1428 time_after_eq(now, rt->u.dst.lastuse + gc_args.timeout)) {
1429 RT6_TRACE("aging clone %p\n", rt);
1430 return -1;
1431 } else if ((rt->rt6i_flags & RTF_GATEWAY) &&
1432 (!(rt->rt6i_nexthop->flags & NTF_ROUTER))) {
1433 RT6_TRACE("purging route %p via non-router but gateway\n",
1434 rt);
1435 return -1;
1437 gc_args.more++;
1440 return 0;
1443 static DEFINE_SPINLOCK(fib6_gc_lock);
1445 void fib6_run_gc(unsigned long expires, struct net *net)
1447 if (expires != ~0UL) {
1448 spin_lock_bh(&fib6_gc_lock);
1449 gc_args.timeout = expires ? (int)expires :
1450 net->ipv6.sysctl.ip6_rt_gc_interval;
1451 } else {
1452 if (!spin_trylock_bh(&fib6_gc_lock)) {
1453 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1454 return;
1456 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1459 gc_args.more = icmp6_dst_gc();
1461 fib6_clean_all(net, fib6_age, 0, NULL);
1463 if (gc_args.more)
1464 mod_timer(&net->ipv6.ip6_fib_timer,
1465 round_jiffies(jiffies
1466 + net->ipv6.sysctl.ip6_rt_gc_interval));
1467 else
1468 del_timer(&net->ipv6.ip6_fib_timer);
1469 spin_unlock_bh(&fib6_gc_lock);
1472 static void fib6_gc_timer_cb(unsigned long arg)
1474 fib6_run_gc(0, (struct net *)arg);
1477 static int fib6_net_init(struct net *net)
1479 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1481 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1482 if (!net->ipv6.rt6_stats)
1483 goto out_timer;
1485 net->ipv6.fib_table_hash = kcalloc(FIB_TABLE_HASHSZ,
1486 sizeof(*net->ipv6.fib_table_hash),
1487 GFP_KERNEL);
1488 if (!net->ipv6.fib_table_hash)
1489 goto out_rt6_stats;
1491 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1492 GFP_KERNEL);
1493 if (!net->ipv6.fib6_main_tbl)
1494 goto out_fib_table_hash;
1496 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1497 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1498 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1499 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1501 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1502 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1503 GFP_KERNEL);
1504 if (!net->ipv6.fib6_local_tbl)
1505 goto out_fib6_main_tbl;
1506 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1507 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1508 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1509 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1510 #endif
1511 fib6_tables_init(net);
1513 return 0;
1515 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1516 out_fib6_main_tbl:
1517 kfree(net->ipv6.fib6_main_tbl);
1518 #endif
1519 out_fib_table_hash:
1520 kfree(net->ipv6.fib_table_hash);
1521 out_rt6_stats:
1522 kfree(net->ipv6.rt6_stats);
1523 out_timer:
1524 return -ENOMEM;
1527 static void fib6_net_exit(struct net *net)
1529 rt6_ifdown(net, NULL);
1530 del_timer_sync(&net->ipv6.ip6_fib_timer);
1532 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1533 kfree(net->ipv6.fib6_local_tbl);
1534 #endif
1535 kfree(net->ipv6.fib6_main_tbl);
1536 kfree(net->ipv6.fib_table_hash);
1537 kfree(net->ipv6.rt6_stats);
1540 static struct pernet_operations fib6_net_ops = {
1541 .init = fib6_net_init,
1542 .exit = fib6_net_exit,
1545 int __init fib6_init(void)
1547 int ret = -ENOMEM;
1549 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1550 sizeof(struct fib6_node),
1551 0, SLAB_HWCACHE_ALIGN,
1552 NULL);
1553 if (!fib6_node_kmem)
1554 goto out;
1556 ret = register_pernet_subsys(&fib6_net_ops);
1557 if (ret)
1558 goto out_kmem_cache_create;
1560 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib);
1561 if (ret)
1562 goto out_unregister_subsys;
1563 out:
1564 return ret;
1566 out_unregister_subsys:
1567 unregister_pernet_subsys(&fib6_net_ops);
1568 out_kmem_cache_create:
1569 kmem_cache_destroy(fib6_node_kmem);
1570 goto out;
1573 void fib6_gc_cleanup(void)
1575 unregister_pernet_subsys(&fib6_net_ops);
1576 kmem_cache_destroy(fib6_node_kmem);