2 * Linux INET6 implementation
3 * Forwarding Information Database
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.
16 * Yuji SEKIYA @USAGI: Support default route on router node;
17 * remove ip6_null_entry from the top of
19 * Ville Nuorvala: Fixed routing subtrees.
21 #include <linux/errno.h>
22 #include <linux/types.h>
23 #include <linux/net.h>
24 #include <linux/route.h>
25 #include <linux/netdevice.h>
26 #include <linux/in6.h>
27 #include <linux/init.h>
28 #include <linux/list.h>
29 #include <linux/slab.h>
32 #include <net/ndisc.h>
33 #include <net/addrconf.h>
35 #include <net/ip6_fib.h>
36 #include <net/ip6_route.h>
41 #define RT6_TRACE(x...) printk(KERN_DEBUG x)
43 #define RT6_TRACE(x...) do { ; } while (0)
46 static struct kmem_cache * fib6_node_kmem __read_mostly;
50 #ifdef CONFIG_IPV6_SUBTREES
61 struct fib6_walker_t w;
63 int (*func)(struct rt6_info *, void *arg);
67 static DEFINE_RWLOCK(fib6_walker_lock);
69 #ifdef CONFIG_IPV6_SUBTREES
70 #define FWS_INIT FWS_S
72 #define FWS_INIT FWS_L
75 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
77 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
78 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
79 static int fib6_walk(struct fib6_walker_t *w);
80 static int fib6_walk_continue(struct fib6_walker_t *w);
83 * A routing update causes an increase of the serial number on the
84 * affected subtree. This allows for cached routes to be asynchronously
85 * tested when modifications are made to the destination cache as a
86 * result of redirects, path MTU changes, etc.
89 static __u32 rt_sernum;
91 static void fib6_gc_timer_cb(unsigned long arg);
93 static LIST_HEAD(fib6_walkers);
94 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
96 static inline void fib6_walker_link(struct fib6_walker_t *w)
98 write_lock_bh(&fib6_walker_lock);
99 list_add(&w->lh, &fib6_walkers);
100 write_unlock_bh(&fib6_walker_lock);
103 static inline void fib6_walker_unlink(struct fib6_walker_t *w)
105 write_lock_bh(&fib6_walker_lock);
107 write_unlock_bh(&fib6_walker_lock);
109 static __inline__ u32 fib6_new_sernum(void)
118 * Auxiliary address test functions for the radix tree.
120 * These assume a 32bit processor (although it will work on
127 #if defined(__LITTLE_ENDIAN)
128 # define BITOP_BE32_SWIZZLE (0x1F & ~7)
130 # define BITOP_BE32_SWIZZLE 0
133 static __inline__ __be32 addr_bit_set(const void *token, int fn_bit)
135 const __be32 *addr = token;
138 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
139 * is optimized version of
140 * htonl(1 << ((~fn_bit)&0x1F))
141 * See include/asm-generic/bitops/le.h.
143 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
147 static __inline__ struct fib6_node * node_alloc(void)
149 struct fib6_node *fn;
151 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
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))
167 static void fib6_link_table(struct net *net, struct fib6_table *tb)
172 * Initialize table lock at a single place to give lockdep a key,
173 * tables aren't visible prior to being linked to the list.
175 rwlock_init(&tb->tb6_lock);
177 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
180 * No protection necessary, this is the only list mutatation
181 * operation, tables never disappear once they exist.
183 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
186 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
188 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
190 struct fib6_table *table;
192 table = kzalloc(sizeof(*table), GFP_ATOMIC);
195 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
196 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
202 struct fib6_table *fib6_new_table(struct net *net, u32 id)
204 struct fib6_table *tb;
208 tb = fib6_get_table(net, id);
212 tb = fib6_alloc_table(net, id);
214 fib6_link_table(net, tb);
219 struct fib6_table *fib6_get_table(struct net *net, u32 id)
221 struct fib6_table *tb;
222 struct hlist_head *head;
223 struct hlist_node *node;
228 h = id & (FIB6_TABLE_HASHSZ - 1);
230 head = &net->ipv6.fib_table_hash[h];
231 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
232 if (tb->tb6_id == id) {
242 static void __net_init fib6_tables_init(struct net *net)
244 fib6_link_table(net, net->ipv6.fib6_main_tbl);
245 fib6_link_table(net, net->ipv6.fib6_local_tbl);
249 struct fib6_table *fib6_new_table(struct net *net, u32 id)
251 return fib6_get_table(net, id);
254 struct fib6_table *fib6_get_table(struct net *net, u32 id)
256 return net->ipv6.fib6_main_tbl;
259 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
260 int flags, pol_lookup_t lookup)
262 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
265 static void __net_init fib6_tables_init(struct net *net)
267 fib6_link_table(net, net->ipv6.fib6_main_tbl);
272 static int fib6_dump_node(struct fib6_walker_t *w)
277 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
278 res = rt6_dump_route(rt, w->args);
280 /* Frame is full, suspend walking */
290 static void fib6_dump_end(struct netlink_callback *cb)
292 struct fib6_walker_t *w = (void*)cb->args[2];
297 fib6_walker_unlink(w);
302 cb->done = (void*)cb->args[3];
306 static int fib6_dump_done(struct netlink_callback *cb)
309 return cb->done ? cb->done(cb) : 0;
312 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
313 struct netlink_callback *cb)
315 struct fib6_walker_t *w;
318 w = (void *)cb->args[2];
319 w->root = &table->tb6_root;
321 if (cb->args[4] == 0) {
325 read_lock_bh(&table->tb6_lock);
327 read_unlock_bh(&table->tb6_lock);
330 cb->args[5] = w->root->fn_sernum;
333 if (cb->args[5] != w->root->fn_sernum) {
334 /* Begin at the root if the tree changed */
335 cb->args[5] = w->root->fn_sernum;
342 read_lock_bh(&table->tb6_lock);
343 res = fib6_walk_continue(w);
344 read_unlock_bh(&table->tb6_lock);
346 fib6_walker_unlink(w);
354 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
356 struct net *net = sock_net(skb->sk);
358 unsigned int e = 0, s_e;
359 struct rt6_rtnl_dump_arg arg;
360 struct fib6_walker_t *w;
361 struct fib6_table *tb;
362 struct hlist_node *node;
363 struct hlist_head *head;
369 w = (void *)cb->args[2];
373 * 1. hook callback destructor.
375 cb->args[3] = (long)cb->done;
376 cb->done = fib6_dump_done;
379 * 2. allocate and initialize walker.
381 w = kzalloc(sizeof(*w), GFP_ATOMIC);
384 w->func = fib6_dump_node;
385 cb->args[2] = (long)w;
394 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
396 head = &net->ipv6.fib_table_hash[h];
397 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
400 res = fib6_dump_table(tb, skb, cb);
412 res = res < 0 ? res : skb->len;
421 * return the appropriate node for a routing tree "add" operation
422 * by either creating and inserting or by returning an existing
426 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
427 int addrlen, int plen,
430 struct fib6_node *fn, *in, *ln;
431 struct fib6_node *pn = NULL;
435 __u32 sernum = fib6_new_sernum();
437 RT6_TRACE("fib6_add_1\n");
439 /* insert node in tree */
444 key = (struct rt6key *)((u8 *)fn->leaf + offset);
449 if (plen < fn->fn_bit ||
450 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
457 if (plen == fn->fn_bit) {
458 /* clean up an intermediate node */
459 if ((fn->fn_flags & RTN_RTINFO) == 0) {
460 rt6_release(fn->leaf);
464 fn->fn_sernum = sernum;
470 * We have more bits to go
473 /* Try to walk down on tree. */
474 fn->fn_sernum = sernum;
475 dir = addr_bit_set(addr, fn->fn_bit);
477 fn = dir ? fn->right: fn->left;
481 * We walked to the bottom of tree.
482 * Create new leaf node without children.
492 ln->fn_sernum = sernum;
504 * split since we don't have a common prefix anymore or
505 * we have a less significant route.
506 * we've to insert an intermediate node on the list
507 * this new node will point to the one we need to create
513 /* find 1st bit in difference between the 2 addrs.
515 See comment in __ipv6_addr_diff: bit may be an invalid value,
516 but if it is >= plen, the value is ignored in any case.
519 bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
524 * (new leaf node)[ln] (old node)[fn]
530 if (in == NULL || ln == NULL) {
539 * new intermediate node.
541 * be off since that an address that chooses one of
542 * the branches would not match less specific routes
543 * in the other branch
550 atomic_inc(&in->leaf->rt6i_ref);
552 in->fn_sernum = sernum;
554 /* update parent pointer */
565 ln->fn_sernum = sernum;
567 if (addr_bit_set(addr, bit)) {
574 } else { /* plen <= bit */
577 * (new leaf node)[ln]
579 * (old node)[fn] NULL
591 ln->fn_sernum = sernum;
598 if (addr_bit_set(&key->addr, plen))
609 * Insert routing information in a node.
612 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
613 struct nl_info *info)
615 struct rt6_info *iter = NULL;
616 struct rt6_info **ins;
620 for (iter = fn->leaf; iter; iter=iter->dst.rt6_next) {
622 * Search for duplicates
625 if (iter->rt6i_metric == rt->rt6i_metric) {
627 * Same priority level
630 if (iter->rt6i_dev == rt->rt6i_dev &&
631 iter->rt6i_idev == rt->rt6i_idev &&
632 ipv6_addr_equal(&iter->rt6i_gateway,
633 &rt->rt6i_gateway)) {
634 if (!(iter->rt6i_flags&RTF_EXPIRES))
636 iter->rt6i_expires = rt->rt6i_expires;
637 if (!(rt->rt6i_flags&RTF_EXPIRES)) {
638 iter->rt6i_flags &= ~RTF_EXPIRES;
639 iter->rt6i_expires = 0;
645 if (iter->rt6i_metric > rt->rt6i_metric)
648 ins = &iter->dst.rt6_next;
651 /* Reset round-robin state, if necessary */
652 if (ins == &fn->leaf)
659 rt->dst.rt6_next = iter;
662 atomic_inc(&rt->rt6i_ref);
663 inet6_rt_notify(RTM_NEWROUTE, rt, info);
664 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
666 if ((fn->fn_flags & RTN_RTINFO) == 0) {
667 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
668 fn->fn_flags |= RTN_RTINFO;
674 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
676 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
677 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE)))
678 mod_timer(&net->ipv6.ip6_fib_timer,
679 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
682 void fib6_force_start_gc(struct net *net)
684 if (!timer_pending(&net->ipv6.ip6_fib_timer))
685 mod_timer(&net->ipv6.ip6_fib_timer,
686 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
690 * Add routing information to the routing tree.
691 * <destination addr>/<source addr>
692 * with source addr info in sub-trees
695 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
697 struct fib6_node *fn, *pn = NULL;
700 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
701 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst));
708 #ifdef CONFIG_IPV6_SUBTREES
709 if (rt->rt6i_src.plen) {
710 struct fib6_node *sn;
712 if (fn->subtree == NULL) {
713 struct fib6_node *sfn;
725 /* Create subtree root node */
730 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
731 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
732 sfn->fn_flags = RTN_ROOT;
733 sfn->fn_sernum = fib6_new_sernum();
735 /* Now add the first leaf node to new subtree */
737 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
738 sizeof(struct in6_addr), rt->rt6i_src.plen,
739 offsetof(struct rt6_info, rt6i_src));
742 /* If it is failed, discard just allocated
743 root, and then (in st_failure) stale node
750 /* Now link new subtree to main tree */
754 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
755 sizeof(struct in6_addr), rt->rt6i_src.plen,
756 offsetof(struct rt6_info, rt6i_src));
762 if (fn->leaf == NULL) {
764 atomic_inc(&rt->rt6i_ref);
770 err = fib6_add_rt2node(fn, rt, info);
773 fib6_start_gc(info->nl_net, rt);
774 if (!(rt->rt6i_flags&RTF_CACHE))
775 fib6_prune_clones(info->nl_net, pn, rt);
780 #ifdef CONFIG_IPV6_SUBTREES
782 * If fib6_add_1 has cleared the old leaf pointer in the
783 * super-tree leaf node we have to find a new one for it.
785 if (pn != fn && pn->leaf == rt) {
787 atomic_dec(&rt->rt6i_ref);
789 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
790 pn->leaf = fib6_find_prefix(info->nl_net, pn);
793 WARN_ON(pn->leaf == NULL);
794 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
797 atomic_inc(&pn->leaf->rt6i_ref);
804 #ifdef CONFIG_IPV6_SUBTREES
805 /* Subtree creation failed, probably main tree node
806 is orphan. If it is, shoot it.
809 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
810 fib6_repair_tree(info->nl_net, fn);
817 * Routing tree lookup
822 int offset; /* key offset on rt6_info */
823 const struct in6_addr *addr; /* search key */
826 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
827 struct lookup_args *args)
829 struct fib6_node *fn;
832 if (unlikely(args->offset == 0))
842 struct fib6_node *next;
844 dir = addr_bit_set(args->addr, fn->fn_bit);
846 next = dir ? fn->right : fn->left;
857 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
860 key = (struct rt6key *) ((u8 *) fn->leaf +
863 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
864 #ifdef CONFIG_IPV6_SUBTREES
866 struct fib6_node *sfn;
867 sfn = fib6_lookup_1(fn->subtree,
874 if (fn->fn_flags & RTN_RTINFO)
878 #ifdef CONFIG_IPV6_SUBTREES
881 if (fn->fn_flags & RTN_ROOT)
890 struct fib6_node * fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
891 const struct in6_addr *saddr)
893 struct fib6_node *fn;
894 struct lookup_args args[] = {
896 .offset = offsetof(struct rt6_info, rt6i_dst),
899 #ifdef CONFIG_IPV6_SUBTREES
901 .offset = offsetof(struct rt6_info, rt6i_src),
906 .offset = 0, /* sentinel */
910 fn = fib6_lookup_1(root, daddr ? args : args + 1);
912 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT)
919 * Get node with specified destination prefix (and source prefix,
920 * if subtrees are used)
924 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
925 const struct in6_addr *addr,
926 int plen, int offset)
928 struct fib6_node *fn;
930 for (fn = root; fn ; ) {
931 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
936 if (plen < fn->fn_bit ||
937 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
940 if (plen == fn->fn_bit)
944 * We have more bits to go
946 if (addr_bit_set(addr, fn->fn_bit))
954 struct fib6_node * fib6_locate(struct fib6_node *root,
955 const struct in6_addr *daddr, int dst_len,
956 const struct in6_addr *saddr, int src_len)
958 struct fib6_node *fn;
960 fn = fib6_locate_1(root, daddr, dst_len,
961 offsetof(struct rt6_info, rt6i_dst));
963 #ifdef CONFIG_IPV6_SUBTREES
965 WARN_ON(saddr == NULL);
966 if (fn && fn->subtree)
967 fn = fib6_locate_1(fn->subtree, saddr, src_len,
968 offsetof(struct rt6_info, rt6i_src));
972 if (fn && fn->fn_flags&RTN_RTINFO)
984 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
986 if (fn->fn_flags&RTN_ROOT)
987 return net->ipv6.ip6_null_entry;
991 return fn->left->leaf;
994 return fn->right->leaf;
996 fn = FIB6_SUBTREE(fn);
1002 * Called to trim the tree of intermediate nodes when possible. "fn"
1003 * is the node we want to try and remove.
1006 static struct fib6_node *fib6_repair_tree(struct net *net,
1007 struct fib6_node *fn)
1011 struct fib6_node *child, *pn;
1012 struct fib6_walker_t *w;
1016 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1019 WARN_ON(fn->fn_flags & RTN_RTINFO);
1020 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1021 WARN_ON(fn->leaf != NULL);
1025 if (fn->right) child = fn->right, children |= 1;
1026 if (fn->left) child = fn->left, children |= 2;
1028 if (children == 3 || FIB6_SUBTREE(fn)
1029 #ifdef CONFIG_IPV6_SUBTREES
1030 /* Subtree root (i.e. fn) may have one child */
1031 || (children && fn->fn_flags&RTN_ROOT)
1034 fn->leaf = fib6_find_prefix(net, fn);
1036 if (fn->leaf==NULL) {
1038 fn->leaf = net->ipv6.ip6_null_entry;
1041 atomic_inc(&fn->leaf->rt6i_ref);
1046 #ifdef CONFIG_IPV6_SUBTREES
1047 if (FIB6_SUBTREE(pn) == fn) {
1048 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1049 FIB6_SUBTREE(pn) = NULL;
1052 WARN_ON(fn->fn_flags & RTN_ROOT);
1054 if (pn->right == fn) pn->right = child;
1055 else if (pn->left == fn) pn->left = child;
1063 #ifdef CONFIG_IPV6_SUBTREES
1067 read_lock(&fib6_walker_lock);
1069 if (child == NULL) {
1070 if (w->root == fn) {
1071 w->root = w->node = NULL;
1072 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1073 } else if (w->node == fn) {
1074 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1079 if (w->root == fn) {
1081 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1083 if (w->node == fn) {
1086 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1087 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1089 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1090 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1095 read_unlock(&fib6_walker_lock);
1098 if (pn->fn_flags&RTN_RTINFO || FIB6_SUBTREE(pn))
1101 rt6_release(pn->leaf);
1107 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1108 struct nl_info *info)
1110 struct fib6_walker_t *w;
1111 struct rt6_info *rt = *rtp;
1112 struct net *net = info->nl_net;
1114 RT6_TRACE("fib6_del_route\n");
1117 *rtp = rt->dst.rt6_next;
1118 rt->rt6i_node = NULL;
1119 net->ipv6.rt6_stats->fib_rt_entries--;
1120 net->ipv6.rt6_stats->fib_discarded_routes++;
1122 /* Reset round-robin state, if necessary */
1123 if (fn->rr_ptr == rt)
1126 /* Adjust walkers */
1127 read_lock(&fib6_walker_lock);
1129 if (w->state == FWS_C && w->leaf == rt) {
1130 RT6_TRACE("walker %p adjusted by delroute\n", w);
1131 w->leaf = rt->dst.rt6_next;
1132 if (w->leaf == NULL)
1136 read_unlock(&fib6_walker_lock);
1138 rt->dst.rt6_next = NULL;
1140 /* If it was last route, expunge its radix tree node */
1141 if (fn->leaf == NULL) {
1142 fn->fn_flags &= ~RTN_RTINFO;
1143 net->ipv6.rt6_stats->fib_route_nodes--;
1144 fn = fib6_repair_tree(net, fn);
1147 if (atomic_read(&rt->rt6i_ref) != 1) {
1148 /* This route is used as dummy address holder in some split
1149 * nodes. It is not leaked, but it still holds other resources,
1150 * which must be released in time. So, scan ascendant nodes
1151 * and replace dummy references to this route with references
1152 * to still alive ones.
1155 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) {
1156 fn->leaf = fib6_find_prefix(net, fn);
1157 atomic_inc(&fn->leaf->rt6i_ref);
1162 /* No more references are possible at this point. */
1163 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1166 inet6_rt_notify(RTM_DELROUTE, rt, info);
1170 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1172 struct net *net = info->nl_net;
1173 struct fib6_node *fn = rt->rt6i_node;
1174 struct rt6_info **rtp;
1177 if (rt->dst.obsolete>0) {
1178 WARN_ON(fn != NULL);
1182 if (fn == NULL || rt == net->ipv6.ip6_null_entry)
1185 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1187 if (!(rt->rt6i_flags&RTF_CACHE)) {
1188 struct fib6_node *pn = fn;
1189 #ifdef CONFIG_IPV6_SUBTREES
1190 /* clones of this route might be in another subtree */
1191 if (rt->rt6i_src.plen) {
1192 while (!(pn->fn_flags&RTN_ROOT))
1197 fib6_prune_clones(info->nl_net, pn, rt);
1201 * Walk the leaf entries looking for ourself
1204 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1206 fib6_del_route(fn, rtp, info);
1214 * Tree traversal function.
1216 * Certainly, it is not interrupt safe.
1217 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1218 * It means, that we can modify tree during walking
1219 * and use this function for garbage collection, clone pruning,
1220 * cleaning tree when a device goes down etc. etc.
1222 * It guarantees that every node will be traversed,
1223 * and that it will be traversed only once.
1225 * Callback function w->func may return:
1226 * 0 -> continue walking.
1227 * positive value -> walking is suspended (used by tree dumps,
1228 * and probably by gc, if it will be split to several slices)
1229 * negative value -> terminate walking.
1231 * The function itself returns:
1232 * 0 -> walk is complete.
1233 * >0 -> walk is incomplete (i.e. suspended)
1234 * <0 -> walk is terminated by an error.
1237 static int fib6_walk_continue(struct fib6_walker_t *w)
1239 struct fib6_node *fn, *pn;
1246 if (w->prune && fn != w->root &&
1247 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) {
1252 #ifdef CONFIG_IPV6_SUBTREES
1254 if (FIB6_SUBTREE(fn)) {
1255 w->node = FIB6_SUBTREE(fn);
1263 w->state = FWS_INIT;
1269 w->node = fn->right;
1270 w->state = FWS_INIT;
1276 if (w->leaf && fn->fn_flags&RTN_RTINFO) {
1279 if (w->count < w->skip) {
1297 #ifdef CONFIG_IPV6_SUBTREES
1298 if (FIB6_SUBTREE(pn) == fn) {
1299 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1304 if (pn->left == fn) {
1308 if (pn->right == fn) {
1310 w->leaf = w->node->leaf;
1320 static int fib6_walk(struct fib6_walker_t *w)
1324 w->state = FWS_INIT;
1327 fib6_walker_link(w);
1328 res = fib6_walk_continue(w);
1330 fib6_walker_unlink(w);
1334 static int fib6_clean_node(struct fib6_walker_t *w)
1337 struct rt6_info *rt;
1338 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1339 struct nl_info info = {
1343 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1344 res = c->func(rt, c->arg);
1347 res = fib6_del(rt, &info);
1350 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res);
1363 * Convenient frontend to tree walker.
1365 * func is called on each route.
1366 * It may return -1 -> delete this route.
1367 * 0 -> continue walking
1369 * prune==1 -> only immediate children of node (certainly,
1370 * ignoring pure split nodes) will be scanned.
1373 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1374 int (*func)(struct rt6_info *, void *arg),
1375 int prune, void *arg)
1377 struct fib6_cleaner_t c;
1380 c.w.func = fib6_clean_node;
1391 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1392 int prune, void *arg)
1394 struct fib6_table *table;
1395 struct hlist_node *node;
1396 struct hlist_head *head;
1400 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1401 head = &net->ipv6.fib_table_hash[h];
1402 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1403 write_lock_bh(&table->tb6_lock);
1404 fib6_clean_tree(net, &table->tb6_root,
1406 write_unlock_bh(&table->tb6_lock);
1412 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1414 if (rt->rt6i_flags & RTF_CACHE) {
1415 RT6_TRACE("pruning clone %p\n", rt);
1422 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1423 struct rt6_info *rt)
1425 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1429 * Garbage collection
1432 static struct fib6_gc_args
1438 static int fib6_age(struct rt6_info *rt, void *arg)
1440 unsigned long now = jiffies;
1443 * check addrconf expiration here.
1444 * Routes are expired even if they are in use.
1446 * Also age clones. Note, that clones are aged out
1447 * only if they are not in use now.
1450 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) {
1451 if (time_after(now, rt->rt6i_expires)) {
1452 RT6_TRACE("expiring %p\n", rt);
1456 } else if (rt->rt6i_flags & RTF_CACHE) {
1457 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1458 time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1459 RT6_TRACE("aging clone %p\n", rt);
1461 } else if ((rt->rt6i_flags & RTF_GATEWAY) &&
1462 (!(dst_get_neighbour_raw(&rt->dst)->flags & NTF_ROUTER))) {
1463 RT6_TRACE("purging route %p via non-router but gateway\n",
1473 static DEFINE_SPINLOCK(fib6_gc_lock);
1475 void fib6_run_gc(unsigned long expires, struct net *net)
1477 if (expires != ~0UL) {
1478 spin_lock_bh(&fib6_gc_lock);
1479 gc_args.timeout = expires ? (int)expires :
1480 net->ipv6.sysctl.ip6_rt_gc_interval;
1482 if (!spin_trylock_bh(&fib6_gc_lock)) {
1483 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1486 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1489 gc_args.more = icmp6_dst_gc();
1491 fib6_clean_all(net, fib6_age, 0, NULL);
1494 mod_timer(&net->ipv6.ip6_fib_timer,
1495 round_jiffies(jiffies
1496 + net->ipv6.sysctl.ip6_rt_gc_interval));
1498 del_timer(&net->ipv6.ip6_fib_timer);
1499 spin_unlock_bh(&fib6_gc_lock);
1502 static void fib6_gc_timer_cb(unsigned long arg)
1504 fib6_run_gc(0, (struct net *)arg);
1507 static int __net_init fib6_net_init(struct net *net)
1509 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1511 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1513 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1514 if (!net->ipv6.rt6_stats)
1517 /* Avoid false sharing : Use at least a full cache line */
1518 size = max_t(size_t, size, L1_CACHE_BYTES);
1520 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1521 if (!net->ipv6.fib_table_hash)
1524 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1526 if (!net->ipv6.fib6_main_tbl)
1527 goto out_fib_table_hash;
1529 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1530 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1531 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1532 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1534 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1535 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1537 if (!net->ipv6.fib6_local_tbl)
1538 goto out_fib6_main_tbl;
1539 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1540 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1541 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1542 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1544 fib6_tables_init(net);
1548 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1550 kfree(net->ipv6.fib6_main_tbl);
1553 kfree(net->ipv6.fib_table_hash);
1555 kfree(net->ipv6.rt6_stats);
1560 static void fib6_net_exit(struct net *net)
1562 rt6_ifdown(net, NULL);
1563 del_timer_sync(&net->ipv6.ip6_fib_timer);
1565 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1566 kfree(net->ipv6.fib6_local_tbl);
1568 kfree(net->ipv6.fib6_main_tbl);
1569 kfree(net->ipv6.fib_table_hash);
1570 kfree(net->ipv6.rt6_stats);
1573 static struct pernet_operations fib6_net_ops = {
1574 .init = fib6_net_init,
1575 .exit = fib6_net_exit,
1578 int __init fib6_init(void)
1582 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1583 sizeof(struct fib6_node),
1584 0, SLAB_HWCACHE_ALIGN,
1586 if (!fib6_node_kmem)
1589 ret = register_pernet_subsys(&fib6_net_ops);
1591 goto out_kmem_cache_create;
1593 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1596 goto out_unregister_subsys;
1600 out_unregister_subsys:
1601 unregister_pernet_subsys(&fib6_net_ops);
1602 out_kmem_cache_create:
1603 kmem_cache_destroy(fib6_node_kmem);
1607 void fib6_gc_cleanup(void)
1609 unregister_pernet_subsys(&fib6_net_ops);
1610 kmem_cache_destroy(fib6_node_kmem);