2 * Linux INET6 implementation
3 * Forwarding Information Database
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.
22 #define pr_fmt(fmt) "IPv6: " fmt
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/net.h>
27 #include <linux/route.h>
28 #include <linux/netdevice.h>
29 #include <linux/in6.h>
30 #include <linux/init.h>
31 #include <linux/list.h>
32 #include <linux/slab.h>
35 #include <net/ndisc.h>
36 #include <net/addrconf.h>
38 #include <net/ip6_fib.h>
39 #include <net/ip6_route.h>
44 #define RT6_TRACE(x...) pr_debug(x)
46 #define RT6_TRACE(x...) do { ; } while (0)
49 static struct kmem_cache * fib6_node_kmem __read_mostly;
53 #ifdef CONFIG_IPV6_SUBTREES
64 struct fib6_walker_t w;
66 int (*func)(struct rt6_info *, void *arg);
70 static DEFINE_RWLOCK(fib6_walker_lock);
72 #ifdef CONFIG_IPV6_SUBTREES
73 #define FWS_INIT FWS_S
75 #define FWS_INIT FWS_L
78 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
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 LIST_HEAD(fib6_walkers);
97 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
99 static inline void fib6_walker_link(struct fib6_walker_t *w)
101 write_lock_bh(&fib6_walker_lock);
102 list_add(&w->lh, &fib6_walkers);
103 write_unlock_bh(&fib6_walker_lock);
106 static inline void fib6_walker_unlink(struct fib6_walker_t *w)
108 write_lock_bh(&fib6_walker_lock);
110 write_unlock_bh(&fib6_walker_lock);
112 static __inline__ u32 fib6_new_sernum(void)
121 * Auxiliary address test functions for the radix tree.
123 * These assume a 32bit processor (although it will work on
130 #if defined(__LITTLE_ENDIAN)
131 # define BITOP_BE32_SWIZZLE (0x1F & ~7)
133 # define BITOP_BE32_SWIZZLE 0
136 static __inline__ __be32 addr_bit_set(const void *token, int fn_bit)
138 const __be32 *addr = token;
141 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
142 * is optimized version of
143 * htonl(1 << ((~fn_bit)&0x1F))
144 * See include/asm-generic/bitops/le.h.
146 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
150 static __inline__ struct fib6_node * node_alloc(void)
152 struct fib6_node *fn;
154 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
159 static __inline__ void node_free(struct fib6_node * fn)
161 kmem_cache_free(fib6_node_kmem, fn);
164 static __inline__ void rt6_release(struct rt6_info *rt)
166 if (atomic_dec_and_test(&rt->rt6i_ref))
170 static void fib6_link_table(struct net *net, struct fib6_table *tb)
175 * Initialize table lock at a single place to give lockdep a key,
176 * tables aren't visible prior to being linked to the list.
178 rwlock_init(&tb->tb6_lock);
180 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
183 * No protection necessary, this is the only list mutatation
184 * operation, tables never disappear once they exist.
186 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
189 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
191 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
193 struct fib6_table *table;
195 table = kzalloc(sizeof(*table), GFP_ATOMIC);
198 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
199 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
200 inet_peer_base_init(&table->tb6_peers);
206 struct fib6_table *fib6_new_table(struct net *net, u32 id)
208 struct fib6_table *tb;
212 tb = fib6_get_table(net, id);
216 tb = fib6_alloc_table(net, id);
218 fib6_link_table(net, tb);
223 struct fib6_table *fib6_get_table(struct net *net, u32 id)
225 struct fib6_table *tb;
226 struct hlist_head *head;
227 struct hlist_node *node;
232 h = id & (FIB6_TABLE_HASHSZ - 1);
234 head = &net->ipv6.fib_table_hash[h];
235 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
236 if (tb->tb6_id == id) {
246 static void __net_init fib6_tables_init(struct net *net)
248 fib6_link_table(net, net->ipv6.fib6_main_tbl);
249 fib6_link_table(net, net->ipv6.fib6_local_tbl);
253 struct fib6_table *fib6_new_table(struct net *net, u32 id)
255 return fib6_get_table(net, id);
258 struct fib6_table *fib6_get_table(struct net *net, u32 id)
260 return net->ipv6.fib6_main_tbl;
263 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
264 int flags, pol_lookup_t lookup)
266 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
269 static void __net_init fib6_tables_init(struct net *net)
271 fib6_link_table(net, net->ipv6.fib6_main_tbl);
276 static int fib6_dump_node(struct fib6_walker_t *w)
281 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
282 res = rt6_dump_route(rt, w->args);
284 /* Frame is full, suspend walking */
294 static void fib6_dump_end(struct netlink_callback *cb)
296 struct fib6_walker_t *w = (void*)cb->args[2];
301 fib6_walker_unlink(w);
306 cb->done = (void*)cb->args[3];
310 static int fib6_dump_done(struct netlink_callback *cb)
313 return cb->done ? cb->done(cb) : 0;
316 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
317 struct netlink_callback *cb)
319 struct fib6_walker_t *w;
322 w = (void *)cb->args[2];
323 w->root = &table->tb6_root;
325 if (cb->args[4] == 0) {
329 read_lock_bh(&table->tb6_lock);
331 read_unlock_bh(&table->tb6_lock);
334 cb->args[5] = w->root->fn_sernum;
337 if (cb->args[5] != w->root->fn_sernum) {
338 /* Begin at the root if the tree changed */
339 cb->args[5] = w->root->fn_sernum;
346 read_lock_bh(&table->tb6_lock);
347 res = fib6_walk_continue(w);
348 read_unlock_bh(&table->tb6_lock);
350 fib6_walker_unlink(w);
358 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
360 struct net *net = sock_net(skb->sk);
362 unsigned int e = 0, s_e;
363 struct rt6_rtnl_dump_arg arg;
364 struct fib6_walker_t *w;
365 struct fib6_table *tb;
366 struct hlist_node *node;
367 struct hlist_head *head;
373 w = (void *)cb->args[2];
377 * 1. hook callback destructor.
379 cb->args[3] = (long)cb->done;
380 cb->done = fib6_dump_done;
383 * 2. allocate and initialize walker.
385 w = kzalloc(sizeof(*w), GFP_ATOMIC);
388 w->func = fib6_dump_node;
389 cb->args[2] = (long)w;
398 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
400 head = &net->ipv6.fib_table_hash[h];
401 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
404 res = fib6_dump_table(tb, skb, cb);
416 res = res < 0 ? res : skb->len;
425 * return the appropriate node for a routing tree "add" operation
426 * by either creating and inserting or by returning an existing
430 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
431 int addrlen, int plen,
432 int offset, int allow_create,
433 int replace_required)
435 struct fib6_node *fn, *in, *ln;
436 struct fib6_node *pn = NULL;
440 __u32 sernum = fib6_new_sernum();
442 RT6_TRACE("fib6_add_1\n");
444 /* insert node in tree */
449 key = (struct rt6key *)((u8 *)fn->leaf + offset);
454 if (plen < fn->fn_bit ||
455 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
457 if (replace_required) {
458 pr_warn("Can't replace route, no match found\n");
459 return ERR_PTR(-ENOENT);
461 pr_warn("NLM_F_CREATE should be set when creating new route\n");
470 if (plen == fn->fn_bit) {
471 /* clean up an intermediate node */
472 if (!(fn->fn_flags & RTN_RTINFO)) {
473 rt6_release(fn->leaf);
477 fn->fn_sernum = sernum;
483 * We have more bits to go
486 /* Try to walk down on tree. */
487 fn->fn_sernum = sernum;
488 dir = addr_bit_set(addr, fn->fn_bit);
490 fn = dir ? fn->right: fn->left;
494 /* We should not create new node because
495 * NLM_F_REPLACE was specified without NLM_F_CREATE
496 * I assume it is safe to require NLM_F_CREATE when
497 * REPLACE flag is used! Later we may want to remove the
498 * check for replace_required, because according
499 * to netlink specification, NLM_F_CREATE
500 * MUST be specified if new route is created.
501 * That would keep IPv6 consistent with IPv4
503 if (replace_required) {
504 pr_warn("Can't replace route, no match found\n");
505 return ERR_PTR(-ENOENT);
507 pr_warn("NLM_F_CREATE should be set when creating new route\n");
510 * We walked to the bottom of tree.
511 * Create new leaf node without children.
517 return ERR_PTR(-ENOMEM);
521 ln->fn_sernum = sernum;
533 * split since we don't have a common prefix anymore or
534 * we have a less significant route.
535 * we've to insert an intermediate node on the list
536 * this new node will point to the one we need to create
542 /* find 1st bit in difference between the 2 addrs.
544 See comment in __ipv6_addr_diff: bit may be an invalid value,
545 but if it is >= plen, the value is ignored in any case.
548 bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
553 * (new leaf node)[ln] (old node)[fn]
564 return ERR_PTR(-ENOMEM);
568 * new intermediate node.
570 * be off since that an address that chooses one of
571 * the branches would not match less specific routes
572 * in the other branch
579 atomic_inc(&in->leaf->rt6i_ref);
581 in->fn_sernum = sernum;
583 /* update parent pointer */
594 ln->fn_sernum = sernum;
596 if (addr_bit_set(addr, bit)) {
603 } else { /* plen <= bit */
606 * (new leaf node)[ln]
608 * (old node)[fn] NULL
614 return ERR_PTR(-ENOMEM);
620 ln->fn_sernum = sernum;
627 if (addr_bit_set(&key->addr, plen))
638 * Insert routing information in a node.
641 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
642 struct nl_info *info)
644 struct rt6_info *iter = NULL;
645 struct rt6_info **ins;
646 int replace = (info->nlh &&
647 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
648 int add = (!info->nlh ||
649 (info->nlh->nlmsg_flags & NLM_F_CREATE));
654 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
656 * Search for duplicates
659 if (iter->rt6i_metric == rt->rt6i_metric) {
661 * Same priority level
664 (info->nlh->nlmsg_flags & NLM_F_EXCL))
671 if (iter->dst.dev == rt->dst.dev &&
672 iter->rt6i_idev == rt->rt6i_idev &&
673 ipv6_addr_equal(&iter->rt6i_gateway,
674 &rt->rt6i_gateway)) {
675 if (!(iter->rt6i_flags & RTF_EXPIRES))
677 if (!(rt->rt6i_flags & RTF_EXPIRES))
678 rt6_clean_expires(iter);
680 rt6_set_expires(iter, rt->dst.expires);
685 if (iter->rt6i_metric > rt->rt6i_metric)
688 ins = &iter->dst.rt6_next;
691 /* Reset round-robin state, if necessary */
692 if (ins == &fn->leaf)
700 pr_warn("NLM_F_CREATE should be set when creating new route\n");
703 rt->dst.rt6_next = iter;
706 atomic_inc(&rt->rt6i_ref);
707 inet6_rt_notify(RTM_NEWROUTE, rt, info);
708 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
710 if (!(fn->fn_flags & RTN_RTINFO)) {
711 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
712 fn->fn_flags |= RTN_RTINFO;
719 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
724 rt->dst.rt6_next = iter->dst.rt6_next;
725 atomic_inc(&rt->rt6i_ref);
726 inet6_rt_notify(RTM_NEWROUTE, rt, info);
728 if (!(fn->fn_flags & RTN_RTINFO)) {
729 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
730 fn->fn_flags |= RTN_RTINFO;
737 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
739 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
740 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
741 mod_timer(&net->ipv6.ip6_fib_timer,
742 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
745 void fib6_force_start_gc(struct net *net)
747 if (!timer_pending(&net->ipv6.ip6_fib_timer))
748 mod_timer(&net->ipv6.ip6_fib_timer,
749 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
753 * Add routing information to the routing tree.
754 * <destination addr>/<source addr>
755 * with source addr info in sub-trees
758 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
760 struct fib6_node *fn, *pn = NULL;
762 int allow_create = 1;
763 int replace_required = 0;
766 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
768 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
769 replace_required = 1;
771 if (!allow_create && !replace_required)
772 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
774 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
775 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst),
776 allow_create, replace_required);
785 #ifdef CONFIG_IPV6_SUBTREES
786 if (rt->rt6i_src.plen) {
787 struct fib6_node *sn;
790 struct fib6_node *sfn;
802 /* Create subtree root node */
807 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
808 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
809 sfn->fn_flags = RTN_ROOT;
810 sfn->fn_sernum = fib6_new_sernum();
812 /* Now add the first leaf node to new subtree */
814 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
815 sizeof(struct in6_addr), rt->rt6i_src.plen,
816 offsetof(struct rt6_info, rt6i_src),
817 allow_create, replace_required);
820 /* If it is failed, discard just allocated
821 root, and then (in st_failure) stale node
829 /* Now link new subtree to main tree */
833 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
834 sizeof(struct in6_addr), rt->rt6i_src.plen,
835 offsetof(struct rt6_info, rt6i_src),
836 allow_create, replace_required);
846 atomic_inc(&rt->rt6i_ref);
852 err = fib6_add_rt2node(fn, rt, info);
854 fib6_start_gc(info->nl_net, rt);
855 if (!(rt->rt6i_flags & RTF_CACHE))
856 fib6_prune_clones(info->nl_net, pn, rt);
861 #ifdef CONFIG_IPV6_SUBTREES
863 * If fib6_add_1 has cleared the old leaf pointer in the
864 * super-tree leaf node we have to find a new one for it.
866 if (pn != fn && pn->leaf == rt) {
868 atomic_dec(&rt->rt6i_ref);
870 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
871 pn->leaf = fib6_find_prefix(info->nl_net, pn);
874 WARN_ON(pn->leaf == NULL);
875 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
878 atomic_inc(&pn->leaf->rt6i_ref);
885 #ifdef CONFIG_IPV6_SUBTREES
886 /* Subtree creation failed, probably main tree node
887 is orphan. If it is, shoot it.
890 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
891 fib6_repair_tree(info->nl_net, fn);
898 * Routing tree lookup
903 int offset; /* key offset on rt6_info */
904 const struct in6_addr *addr; /* search key */
907 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
908 struct lookup_args *args)
910 struct fib6_node *fn;
913 if (unlikely(args->offset == 0))
923 struct fib6_node *next;
925 dir = addr_bit_set(args->addr, fn->fn_bit);
927 next = dir ? fn->right : fn->left;
937 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
940 key = (struct rt6key *) ((u8 *) fn->leaf +
943 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
944 #ifdef CONFIG_IPV6_SUBTREES
946 fn = fib6_lookup_1(fn->subtree, args + 1);
948 if (!fn || fn->fn_flags & RTN_RTINFO)
953 if (fn->fn_flags & RTN_ROOT)
962 struct fib6_node * fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
963 const struct in6_addr *saddr)
965 struct fib6_node *fn;
966 struct lookup_args args[] = {
968 .offset = offsetof(struct rt6_info, rt6i_dst),
971 #ifdef CONFIG_IPV6_SUBTREES
973 .offset = offsetof(struct rt6_info, rt6i_src),
978 .offset = 0, /* sentinel */
982 fn = fib6_lookup_1(root, daddr ? args : args + 1);
983 if (!fn || fn->fn_flags & RTN_TL_ROOT)
990 * Get node with specified destination prefix (and source prefix,
991 * if subtrees are used)
995 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
996 const struct in6_addr *addr,
997 int plen, int offset)
999 struct fib6_node *fn;
1001 for (fn = root; fn ; ) {
1002 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1007 if (plen < fn->fn_bit ||
1008 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1011 if (plen == fn->fn_bit)
1015 * We have more bits to go
1017 if (addr_bit_set(addr, fn->fn_bit))
1025 struct fib6_node * fib6_locate(struct fib6_node *root,
1026 const struct in6_addr *daddr, int dst_len,
1027 const struct in6_addr *saddr, int src_len)
1029 struct fib6_node *fn;
1031 fn = fib6_locate_1(root, daddr, dst_len,
1032 offsetof(struct rt6_info, rt6i_dst));
1034 #ifdef CONFIG_IPV6_SUBTREES
1036 WARN_ON(saddr == NULL);
1037 if (fn && fn->subtree)
1038 fn = fib6_locate_1(fn->subtree, saddr, src_len,
1039 offsetof(struct rt6_info, rt6i_src));
1043 if (fn && fn->fn_flags & RTN_RTINFO)
1055 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1057 if (fn->fn_flags & RTN_ROOT)
1058 return net->ipv6.ip6_null_entry;
1062 return fn->left->leaf;
1064 return fn->right->leaf;
1066 fn = FIB6_SUBTREE(fn);
1072 * Called to trim the tree of intermediate nodes when possible. "fn"
1073 * is the node we want to try and remove.
1076 static struct fib6_node *fib6_repair_tree(struct net *net,
1077 struct fib6_node *fn)
1081 struct fib6_node *child, *pn;
1082 struct fib6_walker_t *w;
1086 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1089 WARN_ON(fn->fn_flags & RTN_RTINFO);
1090 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1091 WARN_ON(fn->leaf != NULL);
1095 if (fn->right) child = fn->right, children |= 1;
1096 if (fn->left) child = fn->left, children |= 2;
1098 if (children == 3 || FIB6_SUBTREE(fn)
1099 #ifdef CONFIG_IPV6_SUBTREES
1100 /* Subtree root (i.e. fn) may have one child */
1101 || (children && fn->fn_flags & RTN_ROOT)
1104 fn->leaf = fib6_find_prefix(net, fn);
1108 fn->leaf = net->ipv6.ip6_null_entry;
1111 atomic_inc(&fn->leaf->rt6i_ref);
1116 #ifdef CONFIG_IPV6_SUBTREES
1117 if (FIB6_SUBTREE(pn) == fn) {
1118 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1119 FIB6_SUBTREE(pn) = NULL;
1122 WARN_ON(fn->fn_flags & RTN_ROOT);
1124 if (pn->right == fn) pn->right = child;
1125 else if (pn->left == fn) pn->left = child;
1133 #ifdef CONFIG_IPV6_SUBTREES
1137 read_lock(&fib6_walker_lock);
1140 if (w->root == fn) {
1141 w->root = w->node = NULL;
1142 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1143 } else if (w->node == fn) {
1144 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1149 if (w->root == fn) {
1151 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1153 if (w->node == fn) {
1156 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1157 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1159 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1160 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1165 read_unlock(&fib6_walker_lock);
1168 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1171 rt6_release(pn->leaf);
1177 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1178 struct nl_info *info)
1180 struct fib6_walker_t *w;
1181 struct rt6_info *rt = *rtp;
1182 struct net *net = info->nl_net;
1184 RT6_TRACE("fib6_del_route\n");
1187 *rtp = rt->dst.rt6_next;
1188 rt->rt6i_node = NULL;
1189 net->ipv6.rt6_stats->fib_rt_entries--;
1190 net->ipv6.rt6_stats->fib_discarded_routes++;
1192 /* Reset round-robin state, if necessary */
1193 if (fn->rr_ptr == rt)
1196 /* Adjust walkers */
1197 read_lock(&fib6_walker_lock);
1199 if (w->state == FWS_C && w->leaf == rt) {
1200 RT6_TRACE("walker %p adjusted by delroute\n", w);
1201 w->leaf = rt->dst.rt6_next;
1206 read_unlock(&fib6_walker_lock);
1208 rt->dst.rt6_next = NULL;
1210 /* If it was last route, expunge its radix tree node */
1212 fn->fn_flags &= ~RTN_RTINFO;
1213 net->ipv6.rt6_stats->fib_route_nodes--;
1214 fn = fib6_repair_tree(net, fn);
1217 if (atomic_read(&rt->rt6i_ref) != 1) {
1218 /* This route is used as dummy address holder in some split
1219 * nodes. It is not leaked, but it still holds other resources,
1220 * which must be released in time. So, scan ascendant nodes
1221 * and replace dummy references to this route with references
1222 * to still alive ones.
1225 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
1226 fn->leaf = fib6_find_prefix(net, fn);
1227 atomic_inc(&fn->leaf->rt6i_ref);
1232 /* No more references are possible at this point. */
1233 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1236 inet6_rt_notify(RTM_DELROUTE, rt, info);
1240 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1242 struct net *net = info->nl_net;
1243 struct fib6_node *fn = rt->rt6i_node;
1244 struct rt6_info **rtp;
1247 if (rt->dst.obsolete>0) {
1248 WARN_ON(fn != NULL);
1252 if (!fn || rt == net->ipv6.ip6_null_entry)
1255 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1257 if (!(rt->rt6i_flags & RTF_CACHE)) {
1258 struct fib6_node *pn = fn;
1259 #ifdef CONFIG_IPV6_SUBTREES
1260 /* clones of this route might be in another subtree */
1261 if (rt->rt6i_src.plen) {
1262 while (!(pn->fn_flags & RTN_ROOT))
1267 fib6_prune_clones(info->nl_net, pn, rt);
1271 * Walk the leaf entries looking for ourself
1274 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1276 fib6_del_route(fn, rtp, info);
1284 * Tree traversal function.
1286 * Certainly, it is not interrupt safe.
1287 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1288 * It means, that we can modify tree during walking
1289 * and use this function for garbage collection, clone pruning,
1290 * cleaning tree when a device goes down etc. etc.
1292 * It guarantees that every node will be traversed,
1293 * and that it will be traversed only once.
1295 * Callback function w->func may return:
1296 * 0 -> continue walking.
1297 * positive value -> walking is suspended (used by tree dumps,
1298 * and probably by gc, if it will be split to several slices)
1299 * negative value -> terminate walking.
1301 * The function itself returns:
1302 * 0 -> walk is complete.
1303 * >0 -> walk is incomplete (i.e. suspended)
1304 * <0 -> walk is terminated by an error.
1307 static int fib6_walk_continue(struct fib6_walker_t *w)
1309 struct fib6_node *fn, *pn;
1316 if (w->prune && fn != w->root &&
1317 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1322 #ifdef CONFIG_IPV6_SUBTREES
1324 if (FIB6_SUBTREE(fn)) {
1325 w->node = FIB6_SUBTREE(fn);
1333 w->state = FWS_INIT;
1339 w->node = fn->right;
1340 w->state = FWS_INIT;
1346 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1367 #ifdef CONFIG_IPV6_SUBTREES
1368 if (FIB6_SUBTREE(pn) == fn) {
1369 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1374 if (pn->left == fn) {
1378 if (pn->right == fn) {
1380 w->leaf = w->node->leaf;
1390 static int fib6_walk(struct fib6_walker_t *w)
1394 w->state = FWS_INIT;
1397 fib6_walker_link(w);
1398 res = fib6_walk_continue(w);
1400 fib6_walker_unlink(w);
1404 static int fib6_clean_node(struct fib6_walker_t *w)
1407 struct rt6_info *rt;
1408 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1409 struct nl_info info = {
1413 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1414 res = c->func(rt, c->arg);
1417 res = fib6_del(rt, &info);
1420 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1421 __func__, rt, rt->rt6i_node, res);
1434 * Convenient frontend to tree walker.
1436 * func is called on each route.
1437 * It may return -1 -> delete this route.
1438 * 0 -> continue walking
1440 * prune==1 -> only immediate children of node (certainly,
1441 * ignoring pure split nodes) will be scanned.
1444 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1445 int (*func)(struct rt6_info *, void *arg),
1446 int prune, void *arg)
1448 struct fib6_cleaner_t c;
1451 c.w.func = fib6_clean_node;
1462 void fib6_clean_all_ro(struct net *net, int (*func)(struct rt6_info *, void *arg),
1463 int prune, void *arg)
1465 struct fib6_table *table;
1466 struct hlist_node *node;
1467 struct hlist_head *head;
1471 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1472 head = &net->ipv6.fib_table_hash[h];
1473 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1474 read_lock_bh(&table->tb6_lock);
1475 fib6_clean_tree(net, &table->tb6_root,
1477 read_unlock_bh(&table->tb6_lock);
1482 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1483 int prune, void *arg)
1485 struct fib6_table *table;
1486 struct hlist_node *node;
1487 struct hlist_head *head;
1491 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1492 head = &net->ipv6.fib_table_hash[h];
1493 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1494 write_lock_bh(&table->tb6_lock);
1495 fib6_clean_tree(net, &table->tb6_root,
1497 write_unlock_bh(&table->tb6_lock);
1503 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1505 if (rt->rt6i_flags & RTF_CACHE) {
1506 RT6_TRACE("pruning clone %p\n", rt);
1513 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1514 struct rt6_info *rt)
1516 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1520 * Garbage collection
1523 static struct fib6_gc_args
1529 static int fib6_age(struct rt6_info *rt, void *arg)
1531 unsigned long now = jiffies;
1534 * check addrconf expiration here.
1535 * Routes are expired even if they are in use.
1537 * Also age clones. Note, that clones are aged out
1538 * only if they are not in use now.
1541 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1542 if (time_after(now, rt->dst.expires)) {
1543 RT6_TRACE("expiring %p\n", rt);
1547 } else if (rt->rt6i_flags & RTF_CACHE) {
1548 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1549 time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1550 RT6_TRACE("aging clone %p\n", rt);
1552 } else if (rt->rt6i_flags & RTF_GATEWAY) {
1553 struct neighbour *neigh;
1554 __u8 neigh_flags = 0;
1556 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1558 neigh_flags = neigh->flags;
1559 neigh_release(neigh);
1561 if (!(neigh_flags & NTF_ROUTER)) {
1562 RT6_TRACE("purging route %p via non-router but gateway\n",
1573 static DEFINE_SPINLOCK(fib6_gc_lock);
1575 void fib6_run_gc(unsigned long expires, struct net *net)
1577 if (expires != ~0UL) {
1578 spin_lock_bh(&fib6_gc_lock);
1579 gc_args.timeout = expires ? (int)expires :
1580 net->ipv6.sysctl.ip6_rt_gc_interval;
1582 if (!spin_trylock_bh(&fib6_gc_lock)) {
1583 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1586 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1589 gc_args.more = icmp6_dst_gc();
1591 fib6_clean_all(net, fib6_age, 0, NULL);
1594 mod_timer(&net->ipv6.ip6_fib_timer,
1595 round_jiffies(jiffies
1596 + net->ipv6.sysctl.ip6_rt_gc_interval));
1598 del_timer(&net->ipv6.ip6_fib_timer);
1599 spin_unlock_bh(&fib6_gc_lock);
1602 static void fib6_gc_timer_cb(unsigned long arg)
1604 fib6_run_gc(0, (struct net *)arg);
1607 static int __net_init fib6_net_init(struct net *net)
1609 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1611 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1613 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1614 if (!net->ipv6.rt6_stats)
1617 /* Avoid false sharing : Use at least a full cache line */
1618 size = max_t(size_t, size, L1_CACHE_BYTES);
1620 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1621 if (!net->ipv6.fib_table_hash)
1624 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1626 if (!net->ipv6.fib6_main_tbl)
1627 goto out_fib_table_hash;
1629 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1630 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1631 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1632 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1633 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1635 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1636 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1638 if (!net->ipv6.fib6_local_tbl)
1639 goto out_fib6_main_tbl;
1640 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1641 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1642 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1643 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1644 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1646 fib6_tables_init(net);
1650 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1652 kfree(net->ipv6.fib6_main_tbl);
1655 kfree(net->ipv6.fib_table_hash);
1657 kfree(net->ipv6.rt6_stats);
1662 static void fib6_net_exit(struct net *net)
1664 rt6_ifdown(net, NULL);
1665 del_timer_sync(&net->ipv6.ip6_fib_timer);
1667 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1668 inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1669 kfree(net->ipv6.fib6_local_tbl);
1671 inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1672 kfree(net->ipv6.fib6_main_tbl);
1673 kfree(net->ipv6.fib_table_hash);
1674 kfree(net->ipv6.rt6_stats);
1677 static struct pernet_operations fib6_net_ops = {
1678 .init = fib6_net_init,
1679 .exit = fib6_net_exit,
1682 int __init fib6_init(void)
1686 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1687 sizeof(struct fib6_node),
1688 0, SLAB_HWCACHE_ALIGN,
1690 if (!fib6_node_kmem)
1693 ret = register_pernet_subsys(&fib6_net_ops);
1695 goto out_kmem_cache_create;
1697 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1700 goto out_unregister_subsys;
1704 out_unregister_subsys:
1705 unregister_pernet_subsys(&fib6_net_ops);
1706 out_kmem_cache_create:
1707 kmem_cache_destroy(fib6_node_kmem);
1711 void fib6_gc_cleanup(void)
1713 unregister_pernet_subsys(&fib6_net_ops);
1714 kmem_cache_destroy(fib6_node_kmem);