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 void node_free_immediate(struct fib6_node *fn)
158 kmem_cache_free(fib6_node_kmem, fn);
161 static void node_free_rcu(struct rcu_head *head)
163 struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
165 kmem_cache_free(fib6_node_kmem, fn);
168 static void node_free(struct fib6_node *fn)
170 call_rcu(&fn->rcu, node_free_rcu);
173 static __inline__ void rt6_release(struct rt6_info *rt)
175 if (atomic_dec_and_test(&rt->rt6i_ref))
179 static void fib6_link_table(struct net *net, struct fib6_table *tb)
184 * Initialize table lock at a single place to give lockdep a key,
185 * tables aren't visible prior to being linked to the list.
187 rwlock_init(&tb->tb6_lock);
189 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
192 * No protection necessary, this is the only list mutatation
193 * operation, tables never disappear once they exist.
195 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
198 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
200 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
202 struct fib6_table *table;
204 table = kzalloc(sizeof(*table), GFP_ATOMIC);
207 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
208 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
214 struct fib6_table *fib6_new_table(struct net *net, u32 id)
216 struct fib6_table *tb;
220 tb = fib6_get_table(net, id);
224 tb = fib6_alloc_table(net, id);
226 fib6_link_table(net, tb);
231 struct fib6_table *fib6_get_table(struct net *net, u32 id)
233 struct fib6_table *tb;
234 struct hlist_head *head;
235 struct hlist_node *node;
240 h = id & (FIB6_TABLE_HASHSZ - 1);
242 head = &net->ipv6.fib_table_hash[h];
243 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
244 if (tb->tb6_id == id) {
254 static void __net_init fib6_tables_init(struct net *net)
256 fib6_link_table(net, net->ipv6.fib6_main_tbl);
257 fib6_link_table(net, net->ipv6.fib6_local_tbl);
261 struct fib6_table *fib6_new_table(struct net *net, u32 id)
263 return fib6_get_table(net, id);
266 struct fib6_table *fib6_get_table(struct net *net, u32 id)
268 return net->ipv6.fib6_main_tbl;
271 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
272 int flags, pol_lookup_t lookup)
274 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
277 static void __net_init fib6_tables_init(struct net *net)
279 fib6_link_table(net, net->ipv6.fib6_main_tbl);
284 static int fib6_dump_node(struct fib6_walker_t *w)
289 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
290 res = rt6_dump_route(rt, w->args);
292 /* Frame is full, suspend walking */
302 static void fib6_dump_end(struct netlink_callback *cb)
304 struct fib6_walker_t *w = (void*)cb->args[2];
309 fib6_walker_unlink(w);
314 cb->done = (void*)cb->args[3];
318 static int fib6_dump_done(struct netlink_callback *cb)
321 return cb->done ? cb->done(cb) : 0;
324 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
325 struct netlink_callback *cb)
327 struct fib6_walker_t *w;
330 w = (void *)cb->args[2];
331 w->root = &table->tb6_root;
333 if (cb->args[4] == 0) {
337 read_lock_bh(&table->tb6_lock);
339 read_unlock_bh(&table->tb6_lock);
342 cb->args[5] = w->root->fn_sernum;
345 if (cb->args[5] != w->root->fn_sernum) {
346 /* Begin at the root if the tree changed */
347 cb->args[5] = w->root->fn_sernum;
354 read_lock_bh(&table->tb6_lock);
355 res = fib6_walk_continue(w);
356 read_unlock_bh(&table->tb6_lock);
358 fib6_walker_unlink(w);
366 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
368 struct net *net = sock_net(skb->sk);
370 unsigned int e = 0, s_e;
371 struct rt6_rtnl_dump_arg arg;
372 struct fib6_walker_t *w;
373 struct fib6_table *tb;
374 struct hlist_node *node;
375 struct hlist_head *head;
381 w = (void *)cb->args[2];
385 * 1. hook callback destructor.
387 cb->args[3] = (long)cb->done;
388 cb->done = fib6_dump_done;
391 * 2. allocate and initialize walker.
393 w = kzalloc(sizeof(*w), GFP_ATOMIC);
396 w->func = fib6_dump_node;
397 cb->args[2] = (long)w;
406 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
408 head = &net->ipv6.fib_table_hash[h];
409 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
412 res = fib6_dump_table(tb, skb, cb);
424 res = res < 0 ? res : skb->len;
433 * return the appropriate node for a routing tree "add" operation
434 * by either creating and inserting or by returning an existing
438 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
439 int addrlen, int plen,
442 struct fib6_node *fn, *in, *ln;
443 struct fib6_node *pn = NULL;
447 __u32 sernum = fib6_new_sernum();
449 RT6_TRACE("fib6_add_1\n");
451 /* insert node in tree */
456 key = (struct rt6key *)((u8 *)fn->leaf + offset);
461 if (plen < fn->fn_bit ||
462 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
469 if (plen == fn->fn_bit) {
470 /* clean up an intermediate node */
471 if ((fn->fn_flags & RTN_RTINFO) == 0) {
472 rt6_release(fn->leaf);
476 fn->fn_sernum = sernum;
482 * We have more bits to go
485 /* Try to walk down on tree. */
486 fn->fn_sernum = sernum;
487 dir = addr_bit_set(addr, fn->fn_bit);
489 fn = dir ? fn->right: fn->left;
493 * We walked to the bottom of tree.
494 * Create new leaf node without children.
504 ln->fn_sernum = sernum;
516 * split since we don't have a common prefix anymore or
517 * we have a less significant route.
518 * we've to insert an intermediate node on the list
519 * this new node will point to the one we need to create
525 /* find 1st bit in difference between the 2 addrs.
527 See comment in __ipv6_addr_diff: bit may be an invalid value,
528 but if it is >= plen, the value is ignored in any case.
531 bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
536 * (new leaf node)[ln] (old node)[fn]
542 if (in == NULL || ln == NULL) {
544 node_free_immediate(in);
546 node_free_immediate(ln);
551 * new intermediate node.
553 * be off since that an address that chooses one of
554 * the branches would not match less specific routes
555 * in the other branch
562 atomic_inc(&in->leaf->rt6i_ref);
564 in->fn_sernum = sernum;
566 /* update parent pointer */
577 ln->fn_sernum = sernum;
579 if (addr_bit_set(addr, bit)) {
586 } else { /* plen <= bit */
589 * (new leaf node)[ln]
591 * (old node)[fn] NULL
603 ln->fn_sernum = sernum;
610 if (addr_bit_set(&key->addr, plen))
621 * Insert routing information in a node.
624 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
625 struct nl_info *info)
627 struct rt6_info *iter = NULL;
628 struct rt6_info **ins;
632 for (iter = fn->leaf; iter; iter=iter->dst.rt6_next) {
634 * Search for duplicates
637 if (iter->rt6i_metric == rt->rt6i_metric) {
639 * Same priority level
642 if (iter->rt6i_dev == rt->rt6i_dev &&
643 iter->rt6i_idev == rt->rt6i_idev &&
644 ipv6_addr_equal(&iter->rt6i_gateway,
645 &rt->rt6i_gateway)) {
646 if (!(iter->rt6i_flags&RTF_EXPIRES))
648 iter->rt6i_expires = rt->rt6i_expires;
649 if (!(rt->rt6i_flags&RTF_EXPIRES)) {
650 iter->rt6i_flags &= ~RTF_EXPIRES;
651 iter->rt6i_expires = 0;
657 if (iter->rt6i_metric > rt->rt6i_metric)
660 ins = &iter->dst.rt6_next;
663 /* Reset round-robin state, if necessary */
664 if (ins == &fn->leaf)
671 rt->dst.rt6_next = iter;
673 rcu_assign_pointer(rt->rt6i_node, fn);
674 atomic_inc(&rt->rt6i_ref);
675 inet6_rt_notify(RTM_NEWROUTE, rt, info);
676 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
678 if ((fn->fn_flags & RTN_RTINFO) == 0) {
679 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
680 fn->fn_flags |= RTN_RTINFO;
686 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
688 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
689 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE)))
690 mod_timer(&net->ipv6.ip6_fib_timer,
691 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
694 void fib6_force_start_gc(struct net *net)
696 if (!timer_pending(&net->ipv6.ip6_fib_timer))
697 mod_timer(&net->ipv6.ip6_fib_timer,
698 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
702 * Add routing information to the routing tree.
703 * <destination addr>/<source addr>
704 * with source addr info in sub-trees
707 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
709 struct fib6_node *fn, *pn = NULL;
712 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
713 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst));
720 #ifdef CONFIG_IPV6_SUBTREES
721 if (rt->rt6i_src.plen) {
722 struct fib6_node *sn;
724 if (fn->subtree == NULL) {
725 struct fib6_node *sfn;
737 /* Create subtree root node */
742 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
743 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
744 sfn->fn_flags = RTN_ROOT;
745 sfn->fn_sernum = fib6_new_sernum();
747 /* Now add the first leaf node to new subtree */
749 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
750 sizeof(struct in6_addr), rt->rt6i_src.plen,
751 offsetof(struct rt6_info, rt6i_src));
754 /* If it is failed, discard just allocated
755 root, and then (in st_failure) stale node
758 node_free_immediate(sfn);
762 /* Now link new subtree to main tree */
766 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
767 sizeof(struct in6_addr), rt->rt6i_src.plen,
768 offsetof(struct rt6_info, rt6i_src));
774 if (fn->leaf == NULL) {
776 atomic_inc(&rt->rt6i_ref);
782 err = fib6_add_rt2node(fn, rt, info);
785 fib6_start_gc(info->nl_net, rt);
786 if (!(rt->rt6i_flags&RTF_CACHE))
787 fib6_prune_clones(info->nl_net, pn, rt);
792 #ifdef CONFIG_IPV6_SUBTREES
794 * If fib6_add_1 has cleared the old leaf pointer in the
795 * super-tree leaf node we have to find a new one for it.
797 if (pn != fn && pn->leaf == rt) {
799 atomic_dec(&rt->rt6i_ref);
801 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
802 pn->leaf = fib6_find_prefix(info->nl_net, pn);
805 WARN_ON(pn->leaf == NULL);
806 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
809 atomic_inc(&pn->leaf->rt6i_ref);
816 #ifdef CONFIG_IPV6_SUBTREES
817 /* Subtree creation failed, probably main tree node
818 is orphan. If it is, shoot it.
821 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
822 fib6_repair_tree(info->nl_net, fn);
829 * Routing tree lookup
834 int offset; /* key offset on rt6_info */
835 const struct in6_addr *addr; /* search key */
838 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
839 struct lookup_args *args)
841 struct fib6_node *fn;
844 if (unlikely(args->offset == 0))
854 struct fib6_node *next;
856 dir = addr_bit_set(args->addr, fn->fn_bit);
858 next = dir ? fn->right : fn->left;
869 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
872 key = (struct rt6key *) ((u8 *) fn->leaf +
875 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
876 #ifdef CONFIG_IPV6_SUBTREES
878 struct fib6_node *sfn;
879 sfn = fib6_lookup_1(fn->subtree,
886 if (fn->fn_flags & RTN_RTINFO)
890 #ifdef CONFIG_IPV6_SUBTREES
893 if (fn->fn_flags & RTN_ROOT)
902 struct fib6_node * fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
903 const struct in6_addr *saddr)
905 struct fib6_node *fn;
906 struct lookup_args args[] = {
908 .offset = offsetof(struct rt6_info, rt6i_dst),
911 #ifdef CONFIG_IPV6_SUBTREES
913 .offset = offsetof(struct rt6_info, rt6i_src),
918 .offset = 0, /* sentinel */
922 fn = fib6_lookup_1(root, daddr ? args : args + 1);
924 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT)
931 * Get node with specified destination prefix (and source prefix,
932 * if subtrees are used)
936 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
937 const struct in6_addr *addr,
938 int plen, int offset)
940 struct fib6_node *fn;
942 for (fn = root; fn ; ) {
943 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
948 if (plen < fn->fn_bit ||
949 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
952 if (plen == fn->fn_bit)
956 * We have more bits to go
958 if (addr_bit_set(addr, fn->fn_bit))
966 struct fib6_node * fib6_locate(struct fib6_node *root,
967 const struct in6_addr *daddr, int dst_len,
968 const struct in6_addr *saddr, int src_len)
970 struct fib6_node *fn;
972 fn = fib6_locate_1(root, daddr, dst_len,
973 offsetof(struct rt6_info, rt6i_dst));
975 #ifdef CONFIG_IPV6_SUBTREES
977 WARN_ON(saddr == NULL);
978 if (fn && fn->subtree)
979 fn = fib6_locate_1(fn->subtree, saddr, src_len,
980 offsetof(struct rt6_info, rt6i_src));
984 if (fn && fn->fn_flags&RTN_RTINFO)
996 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
998 if (fn->fn_flags&RTN_ROOT)
999 return net->ipv6.ip6_null_entry;
1003 return fn->left->leaf;
1006 return fn->right->leaf;
1008 fn = FIB6_SUBTREE(fn);
1014 * Called to trim the tree of intermediate nodes when possible. "fn"
1015 * is the node we want to try and remove.
1018 static struct fib6_node *fib6_repair_tree(struct net *net,
1019 struct fib6_node *fn)
1023 struct fib6_node *child, *pn;
1024 struct fib6_walker_t *w;
1028 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1031 WARN_ON(fn->fn_flags & RTN_RTINFO);
1032 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1033 WARN_ON(fn->leaf != NULL);
1037 if (fn->right) child = fn->right, children |= 1;
1038 if (fn->left) child = fn->left, children |= 2;
1040 if (children == 3 || FIB6_SUBTREE(fn)
1041 #ifdef CONFIG_IPV6_SUBTREES
1042 /* Subtree root (i.e. fn) may have one child */
1043 || (children && fn->fn_flags&RTN_ROOT)
1046 fn->leaf = fib6_find_prefix(net, fn);
1048 if (fn->leaf==NULL) {
1050 fn->leaf = net->ipv6.ip6_null_entry;
1053 atomic_inc(&fn->leaf->rt6i_ref);
1058 #ifdef CONFIG_IPV6_SUBTREES
1059 if (FIB6_SUBTREE(pn) == fn) {
1060 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1061 FIB6_SUBTREE(pn) = NULL;
1064 WARN_ON(fn->fn_flags & RTN_ROOT);
1066 if (pn->right == fn) pn->right = child;
1067 else if (pn->left == fn) pn->left = child;
1075 #ifdef CONFIG_IPV6_SUBTREES
1079 read_lock(&fib6_walker_lock);
1081 if (child == NULL) {
1082 if (w->root == fn) {
1083 w->root = w->node = NULL;
1084 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1085 } else if (w->node == fn) {
1086 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1091 if (w->root == fn) {
1093 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1095 if (w->node == fn) {
1098 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1099 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1101 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1102 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1107 read_unlock(&fib6_walker_lock);
1110 if (pn->fn_flags&RTN_RTINFO || FIB6_SUBTREE(pn))
1113 rt6_release(pn->leaf);
1119 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1120 struct nl_info *info)
1122 struct fib6_walker_t *w;
1123 struct rt6_info *rt = *rtp;
1124 struct net *net = info->nl_net;
1126 RT6_TRACE("fib6_del_route\n");
1129 *rtp = rt->dst.rt6_next;
1130 rt->rt6i_node = NULL;
1131 net->ipv6.rt6_stats->fib_rt_entries--;
1132 net->ipv6.rt6_stats->fib_discarded_routes++;
1134 /* Reset round-robin state, if necessary */
1135 if (fn->rr_ptr == rt)
1138 /* Adjust walkers */
1139 read_lock(&fib6_walker_lock);
1141 if (w->state == FWS_C && w->leaf == rt) {
1142 RT6_TRACE("walker %p adjusted by delroute\n", w);
1143 w->leaf = rt->dst.rt6_next;
1144 if (w->leaf == NULL)
1148 read_unlock(&fib6_walker_lock);
1150 rt->dst.rt6_next = NULL;
1152 /* If it was last route, expunge its radix tree node */
1153 if (fn->leaf == NULL) {
1154 fn->fn_flags &= ~RTN_RTINFO;
1155 net->ipv6.rt6_stats->fib_route_nodes--;
1156 fn = fib6_repair_tree(net, fn);
1159 if (atomic_read(&rt->rt6i_ref) != 1) {
1160 /* This route is used as dummy address holder in some split
1161 * nodes. It is not leaked, but it still holds other resources,
1162 * which must be released in time. So, scan ascendant nodes
1163 * and replace dummy references to this route with references
1164 * to still alive ones.
1167 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) {
1168 fn->leaf = fib6_find_prefix(net, fn);
1169 atomic_inc(&fn->leaf->rt6i_ref);
1174 /* No more references are possible at this point. */
1175 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1178 inet6_rt_notify(RTM_DELROUTE, rt, info);
1182 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1184 struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1185 lockdep_is_held(&rt->rt6i_table->tb6_lock));
1186 struct net *net = info->nl_net;
1187 struct rt6_info **rtp;
1190 if (rt->dst.obsolete>0) {
1191 WARN_ON(fn != NULL);
1195 if (fn == NULL || rt == net->ipv6.ip6_null_entry)
1198 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1200 if (!(rt->rt6i_flags&RTF_CACHE)) {
1201 struct fib6_node *pn = fn;
1202 #ifdef CONFIG_IPV6_SUBTREES
1203 /* clones of this route might be in another subtree */
1204 if (rt->rt6i_src.plen) {
1205 while (!(pn->fn_flags&RTN_ROOT))
1210 fib6_prune_clones(info->nl_net, pn, rt);
1214 * Walk the leaf entries looking for ourself
1217 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1219 fib6_del_route(fn, rtp, info);
1227 * Tree traversal function.
1229 * Certainly, it is not interrupt safe.
1230 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1231 * It means, that we can modify tree during walking
1232 * and use this function for garbage collection, clone pruning,
1233 * cleaning tree when a device goes down etc. etc.
1235 * It guarantees that every node will be traversed,
1236 * and that it will be traversed only once.
1238 * Callback function w->func may return:
1239 * 0 -> continue walking.
1240 * positive value -> walking is suspended (used by tree dumps,
1241 * and probably by gc, if it will be split to several slices)
1242 * negative value -> terminate walking.
1244 * The function itself returns:
1245 * 0 -> walk is complete.
1246 * >0 -> walk is incomplete (i.e. suspended)
1247 * <0 -> walk is terminated by an error.
1250 static int fib6_walk_continue(struct fib6_walker_t *w)
1252 struct fib6_node *fn, *pn;
1259 if (w->prune && fn != w->root &&
1260 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) {
1265 #ifdef CONFIG_IPV6_SUBTREES
1267 if (FIB6_SUBTREE(fn)) {
1268 w->node = FIB6_SUBTREE(fn);
1276 w->state = FWS_INIT;
1282 w->node = fn->right;
1283 w->state = FWS_INIT;
1289 if (w->leaf && fn->fn_flags&RTN_RTINFO) {
1311 #ifdef CONFIG_IPV6_SUBTREES
1312 if (FIB6_SUBTREE(pn) == fn) {
1313 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1318 if (pn->left == fn) {
1322 if (pn->right == fn) {
1324 w->leaf = w->node->leaf;
1334 static int fib6_walk(struct fib6_walker_t *w)
1338 w->state = FWS_INIT;
1341 fib6_walker_link(w);
1342 res = fib6_walk_continue(w);
1344 fib6_walker_unlink(w);
1348 static int fib6_clean_node(struct fib6_walker_t *w)
1351 struct rt6_info *rt;
1352 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1353 struct nl_info info = {
1357 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1358 res = c->func(rt, c->arg);
1361 res = fib6_del(rt, &info);
1364 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1366 rcu_access_pointer(rt->rt6i_node),
1380 * Convenient frontend to tree walker.
1382 * func is called on each route.
1383 * It may return -1 -> delete this route.
1384 * 0 -> continue walking
1386 * prune==1 -> only immediate children of node (certainly,
1387 * ignoring pure split nodes) will be scanned.
1390 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1391 int (*func)(struct rt6_info *, void *arg),
1392 int prune, void *arg)
1394 struct fib6_cleaner_t c;
1397 c.w.func = fib6_clean_node;
1408 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1409 int prune, void *arg)
1411 struct fib6_table *table;
1412 struct hlist_node *node;
1413 struct hlist_head *head;
1417 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1418 head = &net->ipv6.fib_table_hash[h];
1419 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1420 write_lock_bh(&table->tb6_lock);
1421 fib6_clean_tree(net, &table->tb6_root,
1423 write_unlock_bh(&table->tb6_lock);
1429 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1431 if (rt->rt6i_flags & RTF_CACHE) {
1432 RT6_TRACE("pruning clone %p\n", rt);
1439 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1440 struct rt6_info *rt)
1442 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1446 * Garbage collection
1449 static struct fib6_gc_args
1455 static int fib6_age(struct rt6_info *rt, void *arg)
1457 unsigned long now = jiffies;
1460 * check addrconf expiration here.
1461 * Routes are expired even if they are in use.
1463 * Also age clones. Note, that clones are aged out
1464 * only if they are not in use now.
1467 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) {
1468 if (time_after(now, rt->rt6i_expires)) {
1469 RT6_TRACE("expiring %p\n", rt);
1473 } else if (rt->rt6i_flags & RTF_CACHE) {
1474 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1475 time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1476 RT6_TRACE("aging clone %p\n", rt);
1478 } else if ((rt->rt6i_flags & RTF_GATEWAY) &&
1479 (!(dst_get_neighbour_raw(&rt->dst)->flags & NTF_ROUTER))) {
1480 RT6_TRACE("purging route %p via non-router but gateway\n",
1490 static DEFINE_SPINLOCK(fib6_gc_lock);
1492 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1497 spin_lock_bh(&fib6_gc_lock);
1498 } else if (!spin_trylock_bh(&fib6_gc_lock)) {
1499 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1502 gc_args.timeout = expires ? (int)expires :
1503 net->ipv6.sysctl.ip6_rt_gc_interval;
1505 gc_args.more = icmp6_dst_gc();
1507 fib6_clean_all(net, fib6_age, 0, NULL);
1509 net->ipv6.ip6_rt_last_gc = now;
1512 mod_timer(&net->ipv6.ip6_fib_timer,
1514 + net->ipv6.sysctl.ip6_rt_gc_interval));
1516 del_timer(&net->ipv6.ip6_fib_timer);
1517 spin_unlock_bh(&fib6_gc_lock);
1520 static void fib6_gc_timer_cb(unsigned long arg)
1522 fib6_run_gc(0, (struct net *)arg, true);
1525 static int __net_init fib6_net_init(struct net *net)
1527 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1529 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1531 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1532 if (!net->ipv6.rt6_stats)
1535 /* Avoid false sharing : Use at least a full cache line */
1536 size = max_t(size_t, size, L1_CACHE_BYTES);
1538 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1539 if (!net->ipv6.fib_table_hash)
1542 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1544 if (!net->ipv6.fib6_main_tbl)
1545 goto out_fib_table_hash;
1547 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1548 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1549 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1550 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1552 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1553 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1555 if (!net->ipv6.fib6_local_tbl)
1556 goto out_fib6_main_tbl;
1557 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1558 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1559 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1560 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1562 fib6_tables_init(net);
1566 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1568 kfree(net->ipv6.fib6_main_tbl);
1571 kfree(net->ipv6.fib_table_hash);
1573 kfree(net->ipv6.rt6_stats);
1578 static void fib6_net_exit(struct net *net)
1582 rt6_ifdown(net, NULL);
1583 del_timer_sync(&net->ipv6.ip6_fib_timer);
1585 for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
1586 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
1587 struct hlist_node *node, *tmp;
1588 struct fib6_table *tb;
1590 hlist_for_each_entry_safe(tb, node, tmp, head, tb6_hlist) {
1591 hlist_del(&tb->tb6_hlist);
1596 kfree(net->ipv6.fib_table_hash);
1597 kfree(net->ipv6.rt6_stats);
1600 static struct pernet_operations fib6_net_ops = {
1601 .init = fib6_net_init,
1602 .exit = fib6_net_exit,
1605 int __init fib6_init(void)
1609 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1610 sizeof(struct fib6_node),
1611 0, SLAB_HWCACHE_ALIGN,
1613 if (!fib6_node_kmem)
1616 ret = register_pernet_subsys(&fib6_net_ops);
1618 goto out_kmem_cache_create;
1620 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1623 goto out_unregister_subsys;
1627 out_unregister_subsys:
1628 unregister_pernet_subsys(&fib6_net_ops);
1629 out_kmem_cache_create:
1630 kmem_cache_destroy(fib6_node_kmem);
1634 void fib6_gc_cleanup(void)
1636 unregister_pernet_subsys(&fib6_net_ops);
1637 kmem_cache_destroy(fib6_node_kmem);