2 * Linux INET6 implementation
3 * Forwarding Information Database
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * $Id: ip6_fib.c,v 1.25 2001/10/31 21:55:55 davem Exp $
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
18 * Yuji SEKIYA @USAGI: Support default route on router node;
19 * remove ip6_null_entry from the top of
21 * Ville Nuorvala: Fixed routing subtrees.
23 #include <linux/errno.h>
24 #include <linux/types.h>
25 #include <linux/net.h>
26 #include <linux/route.h>
27 #include <linux/netdevice.h>
28 #include <linux/in6.h>
29 #include <linux/init.h>
30 #include <linux/list.h>
33 #include <linux/proc_fs.h>
37 #include <net/ndisc.h>
38 #include <net/addrconf.h>
40 #include <net/ip6_fib.h>
41 #include <net/ip6_route.h>
46 #define RT6_TRACE(x...) printk(KERN_DEBUG x)
48 #define RT6_TRACE(x...) do { ; } while (0)
51 struct rt6_statistics rt6_stats;
53 static struct kmem_cache * fib6_node_kmem __read_mostly;
57 #ifdef CONFIG_IPV6_SUBTREES
68 struct fib6_walker_t w;
69 int (*func)(struct rt6_info *, void *arg);
73 static DEFINE_RWLOCK(fib6_walker_lock);
75 #ifdef CONFIG_IPV6_SUBTREES
76 #define FWS_INIT FWS_S
78 #define FWS_INIT FWS_L
81 static void fib6_prune_clones(struct fib6_node *fn, struct rt6_info *rt);
82 static struct rt6_info * fib6_find_prefix(struct fib6_node *fn);
83 static struct fib6_node * fib6_repair_tree(struct fib6_node *fn);
84 static int fib6_walk(struct fib6_walker_t *w);
85 static int fib6_walk_continue(struct fib6_walker_t *w);
88 * A routing update causes an increase of the serial number on the
89 * affected subtree. This allows for cached routes to be asynchronously
90 * tested when modifications are made to the destination cache as a
91 * result of redirects, path MTU changes, etc.
94 static __u32 rt_sernum;
96 static DEFINE_TIMER(ip6_fib_timer, fib6_run_gc, 0, 0);
98 static struct fib6_walker_t fib6_walker_list = {
99 .prev = &fib6_walker_list,
100 .next = &fib6_walker_list,
103 #define FOR_WALKERS(w) for ((w)=fib6_walker_list.next; (w) != &fib6_walker_list; (w)=(w)->next)
105 static inline void fib6_walker_link(struct fib6_walker_t *w)
107 write_lock_bh(&fib6_walker_lock);
108 w->next = fib6_walker_list.next;
109 w->prev = &fib6_walker_list;
112 write_unlock_bh(&fib6_walker_lock);
115 static inline void fib6_walker_unlink(struct fib6_walker_t *w)
117 write_lock_bh(&fib6_walker_lock);
118 w->next->prev = w->prev;
119 w->prev->next = w->next;
120 w->prev = w->next = w;
121 write_unlock_bh(&fib6_walker_lock);
123 static __inline__ u32 fib6_new_sernum(void)
132 * Auxiliary address test functions for the radix tree.
134 * These assume a 32bit processor (although it will work on
142 static __inline__ __be32 addr_bit_set(void *token, int fn_bit)
144 __be32 *addr = token;
146 return htonl(1 << ((~fn_bit)&0x1F)) & addr[fn_bit>>5];
149 static __inline__ struct fib6_node * node_alloc(void)
151 struct fib6_node *fn;
153 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
158 static __inline__ void node_free(struct fib6_node * fn)
160 kmem_cache_free(fib6_node_kmem, fn);
163 static __inline__ void rt6_release(struct rt6_info *rt)
165 if (atomic_dec_and_test(&rt->rt6i_ref))
166 dst_free(&rt->u.dst);
169 static struct fib6_table fib6_main_tbl = {
170 .tb6_id = RT6_TABLE_MAIN,
172 .leaf = &ip6_null_entry,
173 .fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO,
177 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
178 #define FIB_TABLE_HASHSZ 256
180 #define FIB_TABLE_HASHSZ 1
182 static struct hlist_head fib_table_hash[FIB_TABLE_HASHSZ];
184 static void fib6_link_table(struct fib6_table *tb)
189 * Initialize table lock at a single place to give lockdep a key,
190 * tables aren't visible prior to being linked to the list.
192 rwlock_init(&tb->tb6_lock);
194 h = tb->tb6_id & (FIB_TABLE_HASHSZ - 1);
197 * No protection necessary, this is the only list mutatation
198 * operation, tables never disappear once they exist.
200 hlist_add_head_rcu(&tb->tb6_hlist, &fib_table_hash[h]);
203 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
204 static struct fib6_table fib6_local_tbl = {
205 .tb6_id = RT6_TABLE_LOCAL,
207 .leaf = &ip6_null_entry,
208 .fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO,
212 static struct fib6_table *fib6_alloc_table(u32 id)
214 struct fib6_table *table;
216 table = kzalloc(sizeof(*table), GFP_ATOMIC);
219 table->tb6_root.leaf = &ip6_null_entry;
220 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
226 struct fib6_table *fib6_new_table(u32 id)
228 struct fib6_table *tb;
232 tb = fib6_get_table(id);
236 tb = fib6_alloc_table(id);
243 struct fib6_table *fib6_get_table(u32 id)
245 struct fib6_table *tb;
246 struct hlist_node *node;
251 h = id & (FIB_TABLE_HASHSZ - 1);
253 hlist_for_each_entry_rcu(tb, node, &fib_table_hash[h], tb6_hlist) {
254 if (tb->tb6_id == id) {
264 static void __init fib6_tables_init(void)
266 fib6_link_table(&fib6_main_tbl);
267 fib6_link_table(&fib6_local_tbl);
272 struct fib6_table *fib6_new_table(u32 id)
274 return fib6_get_table(id);
277 struct fib6_table *fib6_get_table(u32 id)
279 return &fib6_main_tbl;
282 struct dst_entry *fib6_rule_lookup(struct flowi *fl, int flags,
285 return (struct dst_entry *) lookup(&fib6_main_tbl, fl, flags);
288 static void __init fib6_tables_init(void)
290 fib6_link_table(&fib6_main_tbl);
295 static int fib6_dump_node(struct fib6_walker_t *w)
300 for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) {
301 res = rt6_dump_route(rt, w->args);
303 /* Frame is full, suspend walking */
313 static void fib6_dump_end(struct netlink_callback *cb)
315 struct fib6_walker_t *w = (void*)cb->args[2];
321 cb->done = (void*)cb->args[3];
325 static int fib6_dump_done(struct netlink_callback *cb)
328 return cb->done ? cb->done(cb) : 0;
331 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
332 struct netlink_callback *cb)
334 struct fib6_walker_t *w;
337 w = (void *)cb->args[2];
338 w->root = &table->tb6_root;
340 if (cb->args[4] == 0) {
341 read_lock_bh(&table->tb6_lock);
343 read_unlock_bh(&table->tb6_lock);
347 read_lock_bh(&table->tb6_lock);
348 res = fib6_walk_continue(w);
349 read_unlock_bh(&table->tb6_lock);
352 fib6_walker_unlink(w);
355 fib6_walker_unlink(w);
362 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
365 unsigned int e = 0, s_e;
366 struct rt6_rtnl_dump_arg arg;
367 struct fib6_walker_t *w;
368 struct fib6_table *tb;
369 struct hlist_node *node;
375 w = (void *)cb->args[2];
379 * 1. hook callback destructor.
381 cb->args[3] = (long)cb->done;
382 cb->done = fib6_dump_done;
385 * 2. allocate and initialize walker.
387 w = kzalloc(sizeof(*w), GFP_ATOMIC);
390 w->func = fib6_dump_node;
391 cb->args[2] = (long)w;
398 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
400 hlist_for_each_entry(tb, node, &fib_table_hash[h], tb6_hlist) {
403 res = fib6_dump_table(tb, skb, cb);
414 res = res < 0 ? res : skb->len;
423 * return the appropriate node for a routing tree "add" operation
424 * by either creating and inserting or by returning an existing
428 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
429 int addrlen, int plen,
432 struct fib6_node *fn, *in, *ln;
433 struct fib6_node *pn = NULL;
437 __u32 sernum = fib6_new_sernum();
439 RT6_TRACE("fib6_add_1\n");
441 /* insert node in tree */
446 key = (struct rt6key *)((u8 *)fn->leaf + offset);
451 if (plen < fn->fn_bit ||
452 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
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);
466 fn->fn_sernum = sernum;
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);
479 fn = dir ? fn->right: fn->left;
483 * We walked to the bottom of tree.
484 * Create new leaf node without children.
494 ln->fn_sernum = sernum;
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
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);
526 * (new leaf node)[ln] (old node)[fn]
532 if (in == NULL || ln == NULL) {
541 * new intermediate node.
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
552 atomic_inc(&in->leaf->rt6i_ref);
554 in->fn_sernum = sernum;
556 /* update parent pointer */
567 ln->fn_sernum = sernum;
569 if (addr_bit_set(addr, bit)) {
576 } else { /* plen <= bit */
579 * (new leaf node)[ln]
581 * (old node)[fn] NULL
593 ln->fn_sernum = sernum;
600 if (addr_bit_set(&key->addr, plen))
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;
622 if (fn->fn_flags&RTN_TL_ROOT &&
623 fn->leaf == &ip6_null_entry &&
624 !(rt->rt6i_flags & (RTF_DEFAULT | RTF_ADDRCONF)) ){
626 rt->u.dst.rt6_next = NULL;
630 for (iter = fn->leaf; iter; iter=iter->u.dst.rt6_next) {
632 * Search for duplicates
635 if (iter->rt6i_metric == rt->rt6i_metric) {
637 * Same priority level
640 if (iter->rt6i_dev == rt->rt6i_dev &&
641 iter->rt6i_idev == rt->rt6i_idev &&
642 ipv6_addr_equal(&iter->rt6i_gateway,
643 &rt->rt6i_gateway)) {
644 if (!(iter->rt6i_flags&RTF_EXPIRES))
646 iter->rt6i_expires = rt->rt6i_expires;
647 if (!(rt->rt6i_flags&RTF_EXPIRES)) {
648 iter->rt6i_flags &= ~RTF_EXPIRES;
649 iter->rt6i_expires = 0;
655 if (iter->rt6i_metric > rt->rt6i_metric)
658 ins = &iter->u.dst.rt6_next;
661 /* Reset round-robin state, if necessary */
662 if (ins == &fn->leaf)
670 rt->u.dst.rt6_next = iter;
673 atomic_inc(&rt->rt6i_ref);
674 inet6_rt_notify(RTM_NEWROUTE, rt, info);
675 rt6_stats.fib_rt_entries++;
677 if ((fn->fn_flags & RTN_RTINFO) == 0) {
678 rt6_stats.fib_route_nodes++;
679 fn->fn_flags |= RTN_RTINFO;
685 static __inline__ void fib6_start_gc(struct rt6_info *rt)
687 if (ip6_fib_timer.expires == 0 &&
688 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE)))
689 mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval);
692 void fib6_force_start_gc(void)
694 if (ip6_fib_timer.expires == 0)
695 mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval);
699 * Add routing information to the routing tree.
700 * <destination addr>/<source addr>
701 * with source addr info in sub-trees
704 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
706 struct fib6_node *fn, *pn = NULL;
709 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
710 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst));
717 #ifdef CONFIG_IPV6_SUBTREES
718 if (rt->rt6i_src.plen) {
719 struct fib6_node *sn;
721 if (fn->subtree == NULL) {
722 struct fib6_node *sfn;
734 /* Create subtree root node */
739 sfn->leaf = &ip6_null_entry;
740 atomic_inc(&ip6_null_entry.rt6i_ref);
741 sfn->fn_flags = RTN_ROOT;
742 sfn->fn_sernum = fib6_new_sernum();
744 /* Now add the first leaf node to new subtree */
746 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
747 sizeof(struct in6_addr), rt->rt6i_src.plen,
748 offsetof(struct rt6_info, rt6i_src));
751 /* If it is failed, discard just allocated
752 root, and then (in st_failure) stale node
759 /* Now link new subtree to main tree */
763 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
764 sizeof(struct in6_addr), rt->rt6i_src.plen,
765 offsetof(struct rt6_info, rt6i_src));
771 if (fn->leaf == NULL) {
773 atomic_inc(&rt->rt6i_ref);
779 err = fib6_add_rt2node(fn, rt, info);
783 if (!(rt->rt6i_flags&RTF_CACHE))
784 fib6_prune_clones(pn, rt);
789 #ifdef CONFIG_IPV6_SUBTREES
791 * If fib6_add_1 has cleared the old leaf pointer in the
792 * super-tree leaf node we have to find a new one for it.
794 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
795 pn->leaf = fib6_find_prefix(pn);
798 BUG_TRAP(pn->leaf != NULL);
799 pn->leaf = &ip6_null_entry;
802 atomic_inc(&pn->leaf->rt6i_ref);
805 dst_free(&rt->u.dst);
809 #ifdef CONFIG_IPV6_SUBTREES
810 /* Subtree creation failed, probably main tree node
811 is orphan. If it is, shoot it.
814 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
815 fib6_repair_tree(fn);
816 dst_free(&rt->u.dst);
822 * Routing tree lookup
827 int offset; /* key offset on rt6_info */
828 struct in6_addr *addr; /* search key */
831 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
832 struct lookup_args *args)
834 struct fib6_node *fn;
837 if (unlikely(args->offset == 0))
847 struct fib6_node *next;
849 dir = addr_bit_set(args->addr, fn->fn_bit);
851 next = dir ? fn->right : fn->left;
862 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
865 key = (struct rt6key *) ((u8 *) fn->leaf +
868 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
869 #ifdef CONFIG_IPV6_SUBTREES
871 fn = fib6_lookup_1(fn->subtree, args + 1);
873 if (!fn || fn->fn_flags & RTN_RTINFO)
878 if (fn->fn_flags & RTN_ROOT)
887 struct fib6_node * fib6_lookup(struct fib6_node *root, struct in6_addr *daddr,
888 struct in6_addr *saddr)
890 struct fib6_node *fn;
891 struct lookup_args args[] = {
893 .offset = offsetof(struct rt6_info, rt6i_dst),
896 #ifdef CONFIG_IPV6_SUBTREES
898 .offset = offsetof(struct rt6_info, rt6i_src),
903 .offset = 0, /* sentinel */
907 fn = fib6_lookup_1(root, daddr ? args : args + 1);
909 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT)
916 * Get node with specified destination prefix (and source prefix,
917 * if subtrees are used)
921 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
922 struct in6_addr *addr,
923 int plen, int offset)
925 struct fib6_node *fn;
927 for (fn = root; fn ; ) {
928 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
933 if (plen < fn->fn_bit ||
934 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
937 if (plen == fn->fn_bit)
941 * We have more bits to go
943 if (addr_bit_set(addr, fn->fn_bit))
951 struct fib6_node * fib6_locate(struct fib6_node *root,
952 struct in6_addr *daddr, int dst_len,
953 struct in6_addr *saddr, int src_len)
955 struct fib6_node *fn;
957 fn = fib6_locate_1(root, daddr, dst_len,
958 offsetof(struct rt6_info, rt6i_dst));
960 #ifdef CONFIG_IPV6_SUBTREES
962 BUG_TRAP(saddr!=NULL);
963 if (fn && fn->subtree)
964 fn = fib6_locate_1(fn->subtree, saddr, src_len,
965 offsetof(struct rt6_info, rt6i_src));
969 if (fn && fn->fn_flags&RTN_RTINFO)
981 static struct rt6_info * fib6_find_prefix(struct fib6_node *fn)
983 if (fn->fn_flags&RTN_ROOT)
984 return &ip6_null_entry;
988 return fn->left->leaf;
991 return fn->right->leaf;
993 fn = FIB6_SUBTREE(fn);
999 * Called to trim the tree of intermediate nodes when possible. "fn"
1000 * is the node we want to try and remove.
1003 static struct fib6_node * fib6_repair_tree(struct fib6_node *fn)
1007 struct fib6_node *child, *pn;
1008 struct fib6_walker_t *w;
1012 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1015 BUG_TRAP(!(fn->fn_flags&RTN_RTINFO));
1016 BUG_TRAP(!(fn->fn_flags&RTN_TL_ROOT));
1017 BUG_TRAP(fn->leaf==NULL);
1021 if (fn->right) child = fn->right, children |= 1;
1022 if (fn->left) child = fn->left, children |= 2;
1024 if (children == 3 || FIB6_SUBTREE(fn)
1025 #ifdef CONFIG_IPV6_SUBTREES
1026 /* Subtree root (i.e. fn) may have one child */
1027 || (children && fn->fn_flags&RTN_ROOT)
1030 fn->leaf = fib6_find_prefix(fn);
1032 if (fn->leaf==NULL) {
1034 fn->leaf = &ip6_null_entry;
1037 atomic_inc(&fn->leaf->rt6i_ref);
1042 #ifdef CONFIG_IPV6_SUBTREES
1043 if (FIB6_SUBTREE(pn) == fn) {
1044 BUG_TRAP(fn->fn_flags&RTN_ROOT);
1045 FIB6_SUBTREE(pn) = NULL;
1048 BUG_TRAP(!(fn->fn_flags&RTN_ROOT));
1050 if (pn->right == fn) pn->right = child;
1051 else if (pn->left == fn) pn->left = child;
1058 #ifdef CONFIG_IPV6_SUBTREES
1062 read_lock(&fib6_walker_lock);
1064 if (child == NULL) {
1065 if (w->root == fn) {
1066 w->root = w->node = NULL;
1067 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1068 } else if (w->node == fn) {
1069 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1074 if (w->root == fn) {
1076 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1078 if (w->node == fn) {
1081 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1082 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1084 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1085 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1090 read_unlock(&fib6_walker_lock);
1093 if (pn->fn_flags&RTN_RTINFO || FIB6_SUBTREE(pn))
1096 rt6_release(pn->leaf);
1102 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1103 struct nl_info *info)
1105 struct fib6_walker_t *w;
1106 struct rt6_info *rt = *rtp;
1108 RT6_TRACE("fib6_del_route\n");
1111 *rtp = rt->u.dst.rt6_next;
1112 rt->rt6i_node = NULL;
1113 rt6_stats.fib_rt_entries--;
1114 rt6_stats.fib_discarded_routes++;
1116 /* Reset round-robin state, if necessary */
1117 if (fn->rr_ptr == rt)
1120 /* Adjust walkers */
1121 read_lock(&fib6_walker_lock);
1123 if (w->state == FWS_C && w->leaf == rt) {
1124 RT6_TRACE("walker %p adjusted by delroute\n", w);
1125 w->leaf = rt->u.dst.rt6_next;
1126 if (w->leaf == NULL)
1130 read_unlock(&fib6_walker_lock);
1132 rt->u.dst.rt6_next = NULL;
1134 if (fn->leaf == NULL && fn->fn_flags&RTN_TL_ROOT)
1135 fn->leaf = &ip6_null_entry;
1137 /* If it was last route, expunge its radix tree node */
1138 if (fn->leaf == NULL) {
1139 fn->fn_flags &= ~RTN_RTINFO;
1140 rt6_stats.fib_route_nodes--;
1141 fn = fib6_repair_tree(fn);
1144 if (atomic_read(&rt->rt6i_ref) != 1) {
1145 /* This route is used as dummy address holder in some split
1146 * nodes. It is not leaked, but it still holds other resources,
1147 * which must be released in time. So, scan ascendant nodes
1148 * and replace dummy references to this route with references
1149 * to still alive ones.
1152 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) {
1153 fn->leaf = fib6_find_prefix(fn);
1154 atomic_inc(&fn->leaf->rt6i_ref);
1159 /* No more references are possible at this point. */
1160 if (atomic_read(&rt->rt6i_ref) != 1) BUG();
1163 inet6_rt_notify(RTM_DELROUTE, rt, info);
1167 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1169 struct fib6_node *fn = rt->rt6i_node;
1170 struct rt6_info **rtp;
1173 if (rt->u.dst.obsolete>0) {
1178 if (fn == NULL || rt == &ip6_null_entry)
1181 BUG_TRAP(fn->fn_flags&RTN_RTINFO);
1183 if (!(rt->rt6i_flags&RTF_CACHE)) {
1184 struct fib6_node *pn = fn;
1185 #ifdef CONFIG_IPV6_SUBTREES
1186 /* clones of this route might be in another subtree */
1187 if (rt->rt6i_src.plen) {
1188 while (!(pn->fn_flags&RTN_ROOT))
1193 fib6_prune_clones(pn, rt);
1197 * Walk the leaf entries looking for ourself
1200 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->u.dst.rt6_next) {
1202 fib6_del_route(fn, rtp, info);
1210 * Tree traversal function.
1212 * Certainly, it is not interrupt safe.
1213 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1214 * It means, that we can modify tree during walking
1215 * and use this function for garbage collection, clone pruning,
1216 * cleaning tree when a device goes down etc. etc.
1218 * It guarantees that every node will be traversed,
1219 * and that it will be traversed only once.
1221 * Callback function w->func may return:
1222 * 0 -> continue walking.
1223 * positive value -> walking is suspended (used by tree dumps,
1224 * and probably by gc, if it will be split to several slices)
1225 * negative value -> terminate walking.
1227 * The function itself returns:
1228 * 0 -> walk is complete.
1229 * >0 -> walk is incomplete (i.e. suspended)
1230 * <0 -> walk is terminated by an error.
1233 static int fib6_walk_continue(struct fib6_walker_t *w)
1235 struct fib6_node *fn, *pn;
1242 if (w->prune && fn != w->root &&
1243 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) {
1248 #ifdef CONFIG_IPV6_SUBTREES
1250 if (FIB6_SUBTREE(fn)) {
1251 w->node = FIB6_SUBTREE(fn);
1259 w->state = FWS_INIT;
1265 w->node = fn->right;
1266 w->state = FWS_INIT;
1272 if (w->leaf && fn->fn_flags&RTN_RTINFO) {
1273 int err = w->func(w);
1284 #ifdef CONFIG_IPV6_SUBTREES
1285 if (FIB6_SUBTREE(pn) == fn) {
1286 BUG_TRAP(fn->fn_flags&RTN_ROOT);
1291 if (pn->left == fn) {
1295 if (pn->right == fn) {
1297 w->leaf = w->node->leaf;
1307 static int fib6_walk(struct fib6_walker_t *w)
1311 w->state = FWS_INIT;
1314 fib6_walker_link(w);
1315 res = fib6_walk_continue(w);
1317 fib6_walker_unlink(w);
1321 static int fib6_clean_node(struct fib6_walker_t *w)
1324 struct rt6_info *rt;
1325 struct fib6_cleaner_t *c = (struct fib6_cleaner_t*)w;
1327 for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) {
1328 res = c->func(rt, c->arg);
1331 res = fib6_del(rt, NULL);
1334 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res);
1347 * Convenient frontend to tree walker.
1349 * func is called on each route.
1350 * It may return -1 -> delete this route.
1351 * 0 -> continue walking
1353 * prune==1 -> only immediate children of node (certainly,
1354 * ignoring pure split nodes) will be scanned.
1357 static void fib6_clean_tree(struct fib6_node *root,
1358 int (*func)(struct rt6_info *, void *arg),
1359 int prune, void *arg)
1361 struct fib6_cleaner_t c;
1364 c.w.func = fib6_clean_node;
1372 void fib6_clean_all(int (*func)(struct rt6_info *, void *arg),
1373 int prune, void *arg)
1375 struct fib6_table *table;
1376 struct hlist_node *node;
1380 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
1381 hlist_for_each_entry_rcu(table, node, &fib_table_hash[h],
1383 write_lock_bh(&table->tb6_lock);
1384 fib6_clean_tree(&table->tb6_root, func, prune, arg);
1385 write_unlock_bh(&table->tb6_lock);
1391 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1393 if (rt->rt6i_flags & RTF_CACHE) {
1394 RT6_TRACE("pruning clone %p\n", rt);
1401 static void fib6_prune_clones(struct fib6_node *fn, struct rt6_info *rt)
1403 fib6_clean_tree(fn, fib6_prune_clone, 1, rt);
1407 * Garbage collection
1410 static struct fib6_gc_args
1416 static int fib6_age(struct rt6_info *rt, void *arg)
1418 unsigned long now = jiffies;
1421 * check addrconf expiration here.
1422 * Routes are expired even if they are in use.
1424 * Also age clones. Note, that clones are aged out
1425 * only if they are not in use now.
1428 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) {
1429 if (time_after(now, rt->rt6i_expires)) {
1430 RT6_TRACE("expiring %p\n", rt);
1434 } else if (rt->rt6i_flags & RTF_CACHE) {
1435 if (atomic_read(&rt->u.dst.__refcnt) == 0 &&
1436 time_after_eq(now, rt->u.dst.lastuse + gc_args.timeout)) {
1437 RT6_TRACE("aging clone %p\n", rt);
1439 } else if ((rt->rt6i_flags & RTF_GATEWAY) &&
1440 (!(rt->rt6i_nexthop->flags & NTF_ROUTER))) {
1441 RT6_TRACE("purging route %p via non-router but gateway\n",
1451 static DEFINE_SPINLOCK(fib6_gc_lock);
1453 void fib6_run_gc(unsigned long dummy)
1455 if (dummy != ~0UL) {
1456 spin_lock_bh(&fib6_gc_lock);
1457 gc_args.timeout = dummy ? (int)dummy : ip6_rt_gc_interval;
1460 if (!spin_trylock(&fib6_gc_lock)) {
1461 mod_timer(&ip6_fib_timer, jiffies + HZ);
1465 gc_args.timeout = ip6_rt_gc_interval;
1469 ndisc_dst_gc(&gc_args.more);
1470 fib6_clean_all(fib6_age, 0, NULL);
1473 mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval);
1475 del_timer(&ip6_fib_timer);
1476 ip6_fib_timer.expires = 0;
1478 spin_unlock_bh(&fib6_gc_lock);
1481 void __init fib6_init(void)
1483 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1484 sizeof(struct fib6_node),
1485 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1490 __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib);
1493 void fib6_gc_cleanup(void)
1495 del_timer(&ip6_fib_timer);
1496 kmem_cache_destroy(fib6_node_kmem);