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.
14 * Yuji SEKIYA @USAGI: Support default route on router node;
15 * remove ip6_null_entry from the top of
17 * Ville Nuorvala: Fixed routing subtrees.
20 #define pr_fmt(fmt) "IPv6: " fmt
22 #include <linux/errno.h>
23 #include <linux/types.h>
24 #include <linux/net.h>
25 #include <linux/route.h>
26 #include <linux/netdevice.h>
27 #include <linux/in6.h>
28 #include <linux/init.h>
29 #include <linux/list.h>
30 #include <linux/slab.h>
33 #include <net/ndisc.h>
34 #include <net/addrconf.h>
35 #include <net/lwtunnel.h>
37 #include <net/ip6_fib.h>
38 #include <net/ip6_route.h>
43 #define RT6_TRACE(x...) pr_debug(x)
45 #define RT6_TRACE(x...) do { ; } while (0)
48 static struct kmem_cache *fib6_node_kmem __read_mostly;
53 int (*func)(struct rt6_info *, void *arg);
58 static DEFINE_RWLOCK(fib6_walker_lock);
60 #ifdef CONFIG_IPV6_SUBTREES
61 #define FWS_INIT FWS_S
63 #define FWS_INIT FWS_L
66 static void fib6_prune_clones(struct net *net, struct fib6_node *fn);
67 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
68 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
69 static int fib6_walk(struct fib6_walker *w);
70 static int fib6_walk_continue(struct fib6_walker *w);
73 * A routing update causes an increase of the serial number on the
74 * affected subtree. This allows for cached routes to be asynchronously
75 * tested when modifications are made to the destination cache as a
76 * result of redirects, path MTU changes, etc.
79 static void fib6_gc_timer_cb(unsigned long arg);
81 static LIST_HEAD(fib6_walkers);
82 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
84 static void fib6_walker_link(struct fib6_walker *w)
86 write_lock_bh(&fib6_walker_lock);
87 list_add(&w->lh, &fib6_walkers);
88 write_unlock_bh(&fib6_walker_lock);
91 static void fib6_walker_unlink(struct fib6_walker *w)
93 write_lock_bh(&fib6_walker_lock);
95 write_unlock_bh(&fib6_walker_lock);
98 static int fib6_new_sernum(struct net *net)
103 old = atomic_read(&net->ipv6.fib6_sernum);
104 new = old < INT_MAX ? old + 1 : 1;
105 } while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
111 FIB6_NO_SERNUM_CHANGE = 0,
115 * Auxiliary address test functions for the radix tree.
117 * These assume a 32bit processor (although it will work on
124 #if defined(__LITTLE_ENDIAN)
125 # define BITOP_BE32_SWIZZLE (0x1F & ~7)
127 # define BITOP_BE32_SWIZZLE 0
130 static __be32 addr_bit_set(const void *token, int fn_bit)
132 const __be32 *addr = token;
135 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
136 * is optimized version of
137 * htonl(1 << ((~fn_bit)&0x1F))
138 * See include/asm-generic/bitops/le.h.
140 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
144 static struct fib6_node *node_alloc(void)
146 struct fib6_node *fn;
148 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
153 static void node_free(struct fib6_node *fn)
155 kmem_cache_free(fib6_node_kmem, fn);
158 static void rt6_free_pcpu(struct rt6_info *non_pcpu_rt)
162 if (!non_pcpu_rt->rt6i_pcpu)
165 for_each_possible_cpu(cpu) {
166 struct rt6_info **ppcpu_rt;
167 struct rt6_info *pcpu_rt;
169 ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
172 dst_free(&pcpu_rt->dst);
177 non_pcpu_rt->rt6i_pcpu = NULL;
180 static void rt6_release(struct rt6_info *rt)
182 if (atomic_dec_and_test(&rt->rt6i_ref)) {
188 static void fib6_link_table(struct net *net, struct fib6_table *tb)
193 * Initialize table lock at a single place to give lockdep a key,
194 * tables aren't visible prior to being linked to the list.
196 rwlock_init(&tb->tb6_lock);
198 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
201 * No protection necessary, this is the only list mutatation
202 * operation, tables never disappear once they exist.
204 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
207 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
209 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
211 struct fib6_table *table;
213 table = kzalloc(sizeof(*table), GFP_ATOMIC);
216 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
217 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
218 inet_peer_base_init(&table->tb6_peers);
224 struct fib6_table *fib6_new_table(struct net *net, u32 id)
226 struct fib6_table *tb;
230 tb = fib6_get_table(net, id);
234 tb = fib6_alloc_table(net, id);
236 fib6_link_table(net, tb);
241 struct fib6_table *fib6_get_table(struct net *net, u32 id)
243 struct fib6_table *tb;
244 struct hlist_head *head;
249 h = id & (FIB6_TABLE_HASHSZ - 1);
251 head = &net->ipv6.fib_table_hash[h];
252 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
253 if (tb->tb6_id == id) {
263 static void __net_init fib6_tables_init(struct net *net)
265 fib6_link_table(net, net->ipv6.fib6_main_tbl);
266 fib6_link_table(net, net->ipv6.fib6_local_tbl);
270 struct fib6_table *fib6_new_table(struct net *net, u32 id)
272 return fib6_get_table(net, id);
275 struct fib6_table *fib6_get_table(struct net *net, u32 id)
277 return net->ipv6.fib6_main_tbl;
280 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
281 int flags, pol_lookup_t lookup)
283 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
286 static void __net_init fib6_tables_init(struct net *net)
288 fib6_link_table(net, net->ipv6.fib6_main_tbl);
293 static int fib6_dump_node(struct fib6_walker *w)
298 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
299 res = rt6_dump_route(rt, w->args);
301 /* Frame is full, suspend walking */
310 static void fib6_dump_end(struct netlink_callback *cb)
312 struct fib6_walker *w = (void *)cb->args[2];
317 fib6_walker_unlink(w);
322 cb->done = (void *)cb->args[3];
326 static int fib6_dump_done(struct netlink_callback *cb)
329 return cb->done ? cb->done(cb) : 0;
332 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
333 struct netlink_callback *cb)
335 struct fib6_walker *w;
338 w = (void *)cb->args[2];
339 w->root = &table->tb6_root;
341 if (cb->args[4] == 0) {
345 read_lock_bh(&table->tb6_lock);
347 read_unlock_bh(&table->tb6_lock);
350 cb->args[5] = w->root->fn_sernum;
353 if (cb->args[5] != w->root->fn_sernum) {
354 /* Begin at the root if the tree changed */
355 cb->args[5] = w->root->fn_sernum;
362 read_lock_bh(&table->tb6_lock);
363 res = fib6_walk_continue(w);
364 read_unlock_bh(&table->tb6_lock);
366 fib6_walker_unlink(w);
374 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
376 struct net *net = sock_net(skb->sk);
378 unsigned int e = 0, s_e;
379 struct rt6_rtnl_dump_arg arg;
380 struct fib6_walker *w;
381 struct fib6_table *tb;
382 struct hlist_head *head;
388 w = (void *)cb->args[2];
392 * 1. hook callback destructor.
394 cb->args[3] = (long)cb->done;
395 cb->done = fib6_dump_done;
398 * 2. allocate and initialize walker.
400 w = kzalloc(sizeof(*w), GFP_ATOMIC);
403 w->func = fib6_dump_node;
404 cb->args[2] = (long)w;
413 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
415 head = &net->ipv6.fib_table_hash[h];
416 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
419 res = fib6_dump_table(tb, skb, cb);
431 res = res < 0 ? res : skb->len;
440 * return the appropriate node for a routing tree "add" operation
441 * by either creating and inserting or by returning an existing
445 static struct fib6_node *fib6_add_1(struct fib6_node *root,
446 struct in6_addr *addr, int plen,
447 int offset, int allow_create,
448 int replace_required, int sernum)
450 struct fib6_node *fn, *in, *ln;
451 struct fib6_node *pn = NULL;
456 RT6_TRACE("fib6_add_1\n");
458 /* insert node in tree */
463 key = (struct rt6key *)((u8 *)fn->leaf + offset);
468 if (plen < fn->fn_bit ||
469 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
471 if (replace_required) {
472 pr_warn("Can't replace route, no match found\n");
473 return ERR_PTR(-ENOENT);
475 pr_warn("NLM_F_CREATE should be set when creating new route\n");
484 if (plen == fn->fn_bit) {
485 /* clean up an intermediate node */
486 if (!(fn->fn_flags & RTN_RTINFO)) {
487 rt6_release(fn->leaf);
491 fn->fn_sernum = sernum;
497 * We have more bits to go
500 /* Try to walk down on tree. */
501 fn->fn_sernum = sernum;
502 dir = addr_bit_set(addr, fn->fn_bit);
504 fn = dir ? fn->right : fn->left;
508 /* We should not create new node because
509 * NLM_F_REPLACE was specified without NLM_F_CREATE
510 * I assume it is safe to require NLM_F_CREATE when
511 * REPLACE flag is used! Later we may want to remove the
512 * check for replace_required, because according
513 * to netlink specification, NLM_F_CREATE
514 * MUST be specified if new route is created.
515 * That would keep IPv6 consistent with IPv4
517 if (replace_required) {
518 pr_warn("Can't replace route, no match found\n");
519 return ERR_PTR(-ENOENT);
521 pr_warn("NLM_F_CREATE should be set when creating new route\n");
524 * We walked to the bottom of tree.
525 * Create new leaf node without children.
531 return ERR_PTR(-ENOMEM);
535 ln->fn_sernum = sernum;
547 * split since we don't have a common prefix anymore or
548 * we have a less significant route.
549 * we've to insert an intermediate node on the list
550 * this new node will point to the one we need to create
556 /* find 1st bit in difference between the 2 addrs.
558 See comment in __ipv6_addr_diff: bit may be an invalid value,
559 but if it is >= plen, the value is ignored in any case.
562 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
567 * (new leaf node)[ln] (old node)[fn]
578 return ERR_PTR(-ENOMEM);
582 * new intermediate node.
584 * be off since that an address that chooses one of
585 * the branches would not match less specific routes
586 * in the other branch
593 atomic_inc(&in->leaf->rt6i_ref);
595 in->fn_sernum = sernum;
597 /* update parent pointer */
608 ln->fn_sernum = sernum;
610 if (addr_bit_set(addr, bit)) {
617 } else { /* plen <= bit */
620 * (new leaf node)[ln]
622 * (old node)[fn] NULL
628 return ERR_PTR(-ENOMEM);
634 ln->fn_sernum = sernum;
641 if (addr_bit_set(&key->addr, plen))
651 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
653 return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
657 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
661 for (i = 0; i < RTAX_MAX; i++) {
662 if (test_bit(i, mxc->mx_valid))
667 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
672 if (dst->flags & DST_HOST) {
673 u32 *mp = dst_metrics_write_ptr(dst);
678 fib6_copy_metrics(mp, mxc);
680 dst_init_metrics(dst, mxc->mx, false);
682 /* We've stolen mx now. */
689 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
692 if (atomic_read(&rt->rt6i_ref) != 1) {
693 /* This route is used as dummy address holder in some split
694 * nodes. It is not leaked, but it still holds other resources,
695 * which must be released in time. So, scan ascendant nodes
696 * and replace dummy references to this route with references
697 * to still alive ones.
700 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
701 fn->leaf = fib6_find_prefix(net, fn);
702 atomic_inc(&fn->leaf->rt6i_ref);
707 /* No more references are possible at this point. */
708 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
713 * Insert routing information in a node.
716 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
717 struct nl_info *info, struct mx6_config *mxc)
719 struct rt6_info *iter = NULL;
720 struct rt6_info **ins;
721 struct rt6_info **fallback_ins = NULL;
722 int replace = (info->nlh &&
723 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
724 int add = (!info->nlh ||
725 (info->nlh->nlmsg_flags & NLM_F_CREATE));
727 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
732 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
734 * Search for duplicates
737 if (iter->rt6i_metric == rt->rt6i_metric) {
739 * Same priority level
742 (info->nlh->nlmsg_flags & NLM_F_EXCL))
745 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
750 fallback_ins = fallback_ins ?: ins;
754 if (iter->dst.dev == rt->dst.dev &&
755 iter->rt6i_idev == rt->rt6i_idev &&
756 ipv6_addr_equal(&iter->rt6i_gateway,
757 &rt->rt6i_gateway)) {
758 if (rt->rt6i_nsiblings)
759 rt->rt6i_nsiblings = 0;
760 if (!(iter->rt6i_flags & RTF_EXPIRES))
762 if (!(rt->rt6i_flags & RTF_EXPIRES))
763 rt6_clean_expires(iter);
765 rt6_set_expires(iter, rt->dst.expires);
766 iter->rt6i_pmtu = rt->rt6i_pmtu;
769 /* If we have the same destination and the same metric,
770 * but not the same gateway, then the route we try to
771 * add is sibling to this route, increment our counter
772 * of siblings, and later we will add our route to the
774 * Only static routes (which don't have flag
775 * RTF_EXPIRES) are used for ECMPv6.
777 * To avoid long list, we only had siblings if the
778 * route have a gateway.
781 rt6_qualify_for_ecmp(iter))
782 rt->rt6i_nsiblings++;
785 if (iter->rt6i_metric > rt->rt6i_metric)
789 ins = &iter->dst.rt6_next;
792 if (fallback_ins && !found) {
793 /* No ECMP-able route found, replace first non-ECMP one */
799 /* Reset round-robin state, if necessary */
800 if (ins == &fn->leaf)
803 /* Link this route to others same route. */
804 if (rt->rt6i_nsiblings) {
805 unsigned int rt6i_nsiblings;
806 struct rt6_info *sibling, *temp_sibling;
808 /* Find the first route that have the same metric */
811 if (sibling->rt6i_metric == rt->rt6i_metric &&
812 rt6_qualify_for_ecmp(sibling)) {
813 list_add_tail(&rt->rt6i_siblings,
814 &sibling->rt6i_siblings);
817 sibling = sibling->dst.rt6_next;
819 /* For each sibling in the list, increment the counter of
820 * siblings. BUG() if counters does not match, list of siblings
824 list_for_each_entry_safe(sibling, temp_sibling,
825 &rt->rt6i_siblings, rt6i_siblings) {
826 sibling->rt6i_nsiblings++;
827 BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
830 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
838 pr_warn("NLM_F_CREATE should be set when creating new route\n");
841 err = fib6_commit_metrics(&rt->dst, mxc);
845 rt->dst.rt6_next = iter;
848 atomic_inc(&rt->rt6i_ref);
849 inet6_rt_notify(RTM_NEWROUTE, rt, info);
850 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
852 if (!(fn->fn_flags & RTN_RTINFO)) {
853 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
854 fn->fn_flags |= RTN_RTINFO;
863 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
867 err = fib6_commit_metrics(&rt->dst, mxc);
873 rt->dst.rt6_next = iter->dst.rt6_next;
874 atomic_inc(&rt->rt6i_ref);
875 inet6_rt_notify(RTM_NEWROUTE, rt, info);
876 if (!(fn->fn_flags & RTN_RTINFO)) {
877 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
878 fn->fn_flags |= RTN_RTINFO;
880 nsiblings = iter->rt6i_nsiblings;
881 fib6_purge_rt(iter, fn, info->nl_net);
885 /* Replacing an ECMP route, remove all siblings */
886 ins = &rt->dst.rt6_next;
889 if (rt6_qualify_for_ecmp(iter)) {
890 *ins = iter->dst.rt6_next;
891 fib6_purge_rt(iter, fn, info->nl_net);
895 ins = &iter->dst.rt6_next;
899 WARN_ON(nsiblings != 0);
906 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
908 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
909 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
910 mod_timer(&net->ipv6.ip6_fib_timer,
911 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
914 void fib6_force_start_gc(struct net *net)
916 if (!timer_pending(&net->ipv6.ip6_fib_timer))
917 mod_timer(&net->ipv6.ip6_fib_timer,
918 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
922 * Add routing information to the routing tree.
923 * <destination addr>/<source addr>
924 * with source addr info in sub-trees
927 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
928 struct nl_info *info, struct mx6_config *mxc)
930 struct fib6_node *fn, *pn = NULL;
932 int allow_create = 1;
933 int replace_required = 0;
934 int sernum = fib6_new_sernum(info->nl_net);
937 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
939 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
940 replace_required = 1;
942 if (!allow_create && !replace_required)
943 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
945 fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
946 offsetof(struct rt6_info, rt6i_dst), allow_create,
947 replace_required, sernum);
956 #ifdef CONFIG_IPV6_SUBTREES
957 if (rt->rt6i_src.plen) {
958 struct fib6_node *sn;
961 struct fib6_node *sfn;
973 /* Create subtree root node */
978 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
979 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
980 sfn->fn_flags = RTN_ROOT;
981 sfn->fn_sernum = sernum;
983 /* Now add the first leaf node to new subtree */
985 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
987 offsetof(struct rt6_info, rt6i_src),
988 allow_create, replace_required, sernum);
991 /* If it is failed, discard just allocated
992 root, and then (in st_failure) stale node
1000 /* Now link new subtree to main tree */
1004 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
1006 offsetof(struct rt6_info, rt6i_src),
1007 allow_create, replace_required, sernum);
1017 atomic_inc(&rt->rt6i_ref);
1023 err = fib6_add_rt2node(fn, rt, info, mxc);
1025 fib6_start_gc(info->nl_net, rt);
1026 if (!(rt->rt6i_flags & RTF_CACHE))
1027 fib6_prune_clones(info->nl_net, pn);
1032 #ifdef CONFIG_IPV6_SUBTREES
1034 * If fib6_add_1 has cleared the old leaf pointer in the
1035 * super-tree leaf node we have to find a new one for it.
1037 if (pn != fn && pn->leaf == rt) {
1039 atomic_dec(&rt->rt6i_ref);
1041 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
1042 pn->leaf = fib6_find_prefix(info->nl_net, pn);
1045 WARN_ON(pn->leaf == NULL);
1046 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
1049 atomic_inc(&pn->leaf->rt6i_ref);
1056 #ifdef CONFIG_IPV6_SUBTREES
1057 /* Subtree creation failed, probably main tree node
1058 is orphan. If it is, shoot it.
1061 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1062 fib6_repair_tree(info->nl_net, fn);
1069 * Routing tree lookup
1073 struct lookup_args {
1074 int offset; /* key offset on rt6_info */
1075 const struct in6_addr *addr; /* search key */
1078 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1079 struct lookup_args *args)
1081 struct fib6_node *fn;
1084 if (unlikely(args->offset == 0))
1094 struct fib6_node *next;
1096 dir = addr_bit_set(args->addr, fn->fn_bit);
1098 next = dir ? fn->right : fn->left;
1108 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1111 key = (struct rt6key *) ((u8 *) fn->leaf +
1114 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1115 #ifdef CONFIG_IPV6_SUBTREES
1117 struct fib6_node *sfn;
1118 sfn = fib6_lookup_1(fn->subtree,
1125 if (fn->fn_flags & RTN_RTINFO)
1129 #ifdef CONFIG_IPV6_SUBTREES
1132 if (fn->fn_flags & RTN_ROOT)
1141 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1142 const struct in6_addr *saddr)
1144 struct fib6_node *fn;
1145 struct lookup_args args[] = {
1147 .offset = offsetof(struct rt6_info, rt6i_dst),
1150 #ifdef CONFIG_IPV6_SUBTREES
1152 .offset = offsetof(struct rt6_info, rt6i_src),
1157 .offset = 0, /* sentinel */
1161 fn = fib6_lookup_1(root, daddr ? args : args + 1);
1162 if (!fn || fn->fn_flags & RTN_TL_ROOT)
1169 * Get node with specified destination prefix (and source prefix,
1170 * if subtrees are used)
1174 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1175 const struct in6_addr *addr,
1176 int plen, int offset)
1178 struct fib6_node *fn;
1180 for (fn = root; fn ; ) {
1181 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1186 if (plen < fn->fn_bit ||
1187 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1190 if (plen == fn->fn_bit)
1194 * We have more bits to go
1196 if (addr_bit_set(addr, fn->fn_bit))
1204 struct fib6_node *fib6_locate(struct fib6_node *root,
1205 const struct in6_addr *daddr, int dst_len,
1206 const struct in6_addr *saddr, int src_len)
1208 struct fib6_node *fn;
1210 fn = fib6_locate_1(root, daddr, dst_len,
1211 offsetof(struct rt6_info, rt6i_dst));
1213 #ifdef CONFIG_IPV6_SUBTREES
1215 WARN_ON(saddr == NULL);
1216 if (fn && fn->subtree)
1217 fn = fib6_locate_1(fn->subtree, saddr, src_len,
1218 offsetof(struct rt6_info, rt6i_src));
1222 if (fn && fn->fn_flags & RTN_RTINFO)
1234 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1236 if (fn->fn_flags & RTN_ROOT)
1237 return net->ipv6.ip6_null_entry;
1241 return fn->left->leaf;
1243 return fn->right->leaf;
1245 fn = FIB6_SUBTREE(fn);
1251 * Called to trim the tree of intermediate nodes when possible. "fn"
1252 * is the node we want to try and remove.
1255 static struct fib6_node *fib6_repair_tree(struct net *net,
1256 struct fib6_node *fn)
1260 struct fib6_node *child, *pn;
1261 struct fib6_walker *w;
1265 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1268 WARN_ON(fn->fn_flags & RTN_RTINFO);
1269 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1275 child = fn->right, children |= 1;
1277 child = fn->left, children |= 2;
1279 if (children == 3 || FIB6_SUBTREE(fn)
1280 #ifdef CONFIG_IPV6_SUBTREES
1281 /* Subtree root (i.e. fn) may have one child */
1282 || (children && fn->fn_flags & RTN_ROOT)
1285 fn->leaf = fib6_find_prefix(net, fn);
1289 fn->leaf = net->ipv6.ip6_null_entry;
1292 atomic_inc(&fn->leaf->rt6i_ref);
1297 #ifdef CONFIG_IPV6_SUBTREES
1298 if (FIB6_SUBTREE(pn) == fn) {
1299 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1300 FIB6_SUBTREE(pn) = NULL;
1303 WARN_ON(fn->fn_flags & RTN_ROOT);
1305 if (pn->right == fn)
1307 else if (pn->left == fn)
1316 #ifdef CONFIG_IPV6_SUBTREES
1320 read_lock(&fib6_walker_lock);
1323 if (w->root == fn) {
1324 w->root = w->node = NULL;
1325 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1326 } else if (w->node == fn) {
1327 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1332 if (w->root == fn) {
1334 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1336 if (w->node == fn) {
1339 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1340 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1342 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1343 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1348 read_unlock(&fib6_walker_lock);
1351 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1354 rt6_release(pn->leaf);
1360 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1361 struct nl_info *info)
1363 struct fib6_walker *w;
1364 struct rt6_info *rt = *rtp;
1365 struct net *net = info->nl_net;
1367 RT6_TRACE("fib6_del_route\n");
1370 *rtp = rt->dst.rt6_next;
1371 rt->rt6i_node = NULL;
1372 net->ipv6.rt6_stats->fib_rt_entries--;
1373 net->ipv6.rt6_stats->fib_discarded_routes++;
1375 /* Reset round-robin state, if necessary */
1376 if (fn->rr_ptr == rt)
1379 /* Remove this entry from other siblings */
1380 if (rt->rt6i_nsiblings) {
1381 struct rt6_info *sibling, *next_sibling;
1383 list_for_each_entry_safe(sibling, next_sibling,
1384 &rt->rt6i_siblings, rt6i_siblings)
1385 sibling->rt6i_nsiblings--;
1386 rt->rt6i_nsiblings = 0;
1387 list_del_init(&rt->rt6i_siblings);
1390 /* Adjust walkers */
1391 read_lock(&fib6_walker_lock);
1393 if (w->state == FWS_C && w->leaf == rt) {
1394 RT6_TRACE("walker %p adjusted by delroute\n", w);
1395 w->leaf = rt->dst.rt6_next;
1400 read_unlock(&fib6_walker_lock);
1402 rt->dst.rt6_next = NULL;
1404 /* If it was last route, expunge its radix tree node */
1406 fn->fn_flags &= ~RTN_RTINFO;
1407 net->ipv6.rt6_stats->fib_route_nodes--;
1408 fn = fib6_repair_tree(net, fn);
1411 fib6_purge_rt(rt, fn, net);
1413 inet6_rt_notify(RTM_DELROUTE, rt, info);
1417 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1419 struct net *net = info->nl_net;
1420 struct fib6_node *fn = rt->rt6i_node;
1421 struct rt6_info **rtp;
1424 if (rt->dst.obsolete > 0) {
1429 if (!fn || rt == net->ipv6.ip6_null_entry)
1432 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1434 if (!(rt->rt6i_flags & RTF_CACHE)) {
1435 struct fib6_node *pn = fn;
1436 #ifdef CONFIG_IPV6_SUBTREES
1437 /* clones of this route might be in another subtree */
1438 if (rt->rt6i_src.plen) {
1439 while (!(pn->fn_flags & RTN_ROOT))
1444 fib6_prune_clones(info->nl_net, pn);
1448 * Walk the leaf entries looking for ourself
1451 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1453 fib6_del_route(fn, rtp, info);
1461 * Tree traversal function.
1463 * Certainly, it is not interrupt safe.
1464 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1465 * It means, that we can modify tree during walking
1466 * and use this function for garbage collection, clone pruning,
1467 * cleaning tree when a device goes down etc. etc.
1469 * It guarantees that every node will be traversed,
1470 * and that it will be traversed only once.
1472 * Callback function w->func may return:
1473 * 0 -> continue walking.
1474 * positive value -> walking is suspended (used by tree dumps,
1475 * and probably by gc, if it will be split to several slices)
1476 * negative value -> terminate walking.
1478 * The function itself returns:
1479 * 0 -> walk is complete.
1480 * >0 -> walk is incomplete (i.e. suspended)
1481 * <0 -> walk is terminated by an error.
1484 static int fib6_walk_continue(struct fib6_walker *w)
1486 struct fib6_node *fn, *pn;
1493 if (w->prune && fn != w->root &&
1494 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1499 #ifdef CONFIG_IPV6_SUBTREES
1501 if (FIB6_SUBTREE(fn)) {
1502 w->node = FIB6_SUBTREE(fn);
1510 w->state = FWS_INIT;
1516 w->node = fn->right;
1517 w->state = FWS_INIT;
1523 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1545 #ifdef CONFIG_IPV6_SUBTREES
1546 if (FIB6_SUBTREE(pn) == fn) {
1547 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1552 if (pn->left == fn) {
1556 if (pn->right == fn) {
1558 w->leaf = w->node->leaf;
1568 static int fib6_walk(struct fib6_walker *w)
1572 w->state = FWS_INIT;
1575 fib6_walker_link(w);
1576 res = fib6_walk_continue(w);
1578 fib6_walker_unlink(w);
1582 static int fib6_clean_node(struct fib6_walker *w)
1585 struct rt6_info *rt;
1586 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1587 struct nl_info info = {
1591 if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1592 w->node->fn_sernum != c->sernum)
1593 w->node->fn_sernum = c->sernum;
1596 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1601 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1602 res = c->func(rt, c->arg);
1605 res = fib6_del(rt, &info);
1608 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1609 __func__, rt, rt->rt6i_node, res);
1622 * Convenient frontend to tree walker.
1624 * func is called on each route.
1625 * It may return -1 -> delete this route.
1626 * 0 -> continue walking
1628 * prune==1 -> only immediate children of node (certainly,
1629 * ignoring pure split nodes) will be scanned.
1632 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1633 int (*func)(struct rt6_info *, void *arg),
1634 bool prune, int sernum, void *arg)
1636 struct fib6_cleaner c;
1639 c.w.func = fib6_clean_node;
1651 static void __fib6_clean_all(struct net *net,
1652 int (*func)(struct rt6_info *, void *),
1653 int sernum, void *arg)
1655 struct fib6_table *table;
1656 struct hlist_head *head;
1660 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1661 head = &net->ipv6.fib_table_hash[h];
1662 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1663 write_lock_bh(&table->tb6_lock);
1664 fib6_clean_tree(net, &table->tb6_root,
1665 func, false, sernum, arg);
1666 write_unlock_bh(&table->tb6_lock);
1672 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1675 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1678 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1680 if (rt->rt6i_flags & RTF_CACHE) {
1681 RT6_TRACE("pruning clone %p\n", rt);
1688 static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1690 fib6_clean_tree(net, fn, fib6_prune_clone, true,
1691 FIB6_NO_SERNUM_CHANGE, NULL);
1694 static void fib6_flush_trees(struct net *net)
1696 int new_sernum = fib6_new_sernum(net);
1698 __fib6_clean_all(net, NULL, new_sernum, NULL);
1702 * Garbage collection
1705 static struct fib6_gc_args
1711 static int fib6_age(struct rt6_info *rt, void *arg)
1713 unsigned long now = jiffies;
1716 * check addrconf expiration here.
1717 * Routes are expired even if they are in use.
1719 * Also age clones. Note, that clones are aged out
1720 * only if they are not in use now.
1723 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1724 if (time_after(now, rt->dst.expires)) {
1725 RT6_TRACE("expiring %p\n", rt);
1729 } else if (rt->rt6i_flags & RTF_CACHE) {
1730 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1731 time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1732 RT6_TRACE("aging clone %p\n", rt);
1734 } else if (rt->rt6i_flags & RTF_GATEWAY) {
1735 struct neighbour *neigh;
1736 __u8 neigh_flags = 0;
1738 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1740 neigh_flags = neigh->flags;
1741 neigh_release(neigh);
1743 if (!(neigh_flags & NTF_ROUTER)) {
1744 RT6_TRACE("purging route %p via non-router but gateway\n",
1755 static DEFINE_SPINLOCK(fib6_gc_lock);
1757 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1762 spin_lock_bh(&fib6_gc_lock);
1763 } else if (!spin_trylock_bh(&fib6_gc_lock)) {
1764 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1767 gc_args.timeout = expires ? (int)expires :
1768 net->ipv6.sysctl.ip6_rt_gc_interval;
1770 gc_args.more = icmp6_dst_gc();
1772 fib6_clean_all(net, fib6_age, NULL);
1774 net->ipv6.ip6_rt_last_gc = now;
1777 mod_timer(&net->ipv6.ip6_fib_timer,
1779 + net->ipv6.sysctl.ip6_rt_gc_interval));
1781 del_timer(&net->ipv6.ip6_fib_timer);
1782 spin_unlock_bh(&fib6_gc_lock);
1785 static void fib6_gc_timer_cb(unsigned long arg)
1787 fib6_run_gc(0, (struct net *)arg, true);
1790 static int __net_init fib6_net_init(struct net *net)
1792 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1794 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1796 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1797 if (!net->ipv6.rt6_stats)
1800 /* Avoid false sharing : Use at least a full cache line */
1801 size = max_t(size_t, size, L1_CACHE_BYTES);
1803 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1804 if (!net->ipv6.fib_table_hash)
1807 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1809 if (!net->ipv6.fib6_main_tbl)
1810 goto out_fib_table_hash;
1812 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1813 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1814 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1815 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1816 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1818 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1819 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1821 if (!net->ipv6.fib6_local_tbl)
1822 goto out_fib6_main_tbl;
1823 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1824 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1825 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1826 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1827 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1829 fib6_tables_init(net);
1833 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1835 kfree(net->ipv6.fib6_main_tbl);
1838 kfree(net->ipv6.fib_table_hash);
1840 kfree(net->ipv6.rt6_stats);
1845 static void fib6_net_exit(struct net *net)
1847 rt6_ifdown(net, NULL);
1848 del_timer_sync(&net->ipv6.ip6_fib_timer);
1850 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1851 inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1852 kfree(net->ipv6.fib6_local_tbl);
1854 inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1855 kfree(net->ipv6.fib6_main_tbl);
1856 kfree(net->ipv6.fib_table_hash);
1857 kfree(net->ipv6.rt6_stats);
1860 static struct pernet_operations fib6_net_ops = {
1861 .init = fib6_net_init,
1862 .exit = fib6_net_exit,
1865 int __init fib6_init(void)
1869 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1870 sizeof(struct fib6_node),
1871 0, SLAB_HWCACHE_ALIGN,
1873 if (!fib6_node_kmem)
1876 ret = register_pernet_subsys(&fib6_net_ops);
1878 goto out_kmem_cache_create;
1880 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1883 goto out_unregister_subsys;
1885 __fib6_flush_trees = fib6_flush_trees;
1889 out_unregister_subsys:
1890 unregister_pernet_subsys(&fib6_net_ops);
1891 out_kmem_cache_create:
1892 kmem_cache_destroy(fib6_node_kmem);
1896 void fib6_gc_cleanup(void)
1898 unregister_pernet_subsys(&fib6_net_ops);
1899 kmem_cache_destroy(fib6_node_kmem);
1902 #ifdef CONFIG_PROC_FS
1904 struct ipv6_route_iter {
1905 struct seq_net_private p;
1906 struct fib6_walker w;
1908 struct fib6_table *tbl;
1912 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
1914 struct rt6_info *rt = v;
1915 struct ipv6_route_iter *iter = seq->private;
1917 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
1919 #ifdef CONFIG_IPV6_SUBTREES
1920 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
1922 seq_puts(seq, "00000000000000000000000000000000 00 ");
1924 if (rt->rt6i_flags & RTF_GATEWAY)
1925 seq_printf(seq, "%pi6", &rt->rt6i_gateway);
1927 seq_puts(seq, "00000000000000000000000000000000");
1929 seq_printf(seq, " %08x %08x %08x %08x %8s\n",
1930 rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
1931 rt->dst.__use, rt->rt6i_flags,
1932 rt->dst.dev ? rt->dst.dev->name : "");
1933 iter->w.leaf = NULL;
1937 static int ipv6_route_yield(struct fib6_walker *w)
1939 struct ipv6_route_iter *iter = w->args;
1945 iter->w.leaf = iter->w.leaf->dst.rt6_next;
1947 if (!iter->skip && iter->w.leaf)
1949 } while (iter->w.leaf);
1954 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter)
1956 memset(&iter->w, 0, sizeof(iter->w));
1957 iter->w.func = ipv6_route_yield;
1958 iter->w.root = &iter->tbl->tb6_root;
1959 iter->w.state = FWS_INIT;
1960 iter->w.node = iter->w.root;
1961 iter->w.args = iter;
1962 iter->sernum = iter->w.root->fn_sernum;
1963 INIT_LIST_HEAD(&iter->w.lh);
1964 fib6_walker_link(&iter->w);
1967 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
1971 struct hlist_node *node;
1974 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
1975 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
1981 while (!node && h < FIB6_TABLE_HASHSZ) {
1982 node = rcu_dereference_bh(
1983 hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
1985 return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
1988 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
1990 if (iter->sernum != iter->w.root->fn_sernum) {
1991 iter->sernum = iter->w.root->fn_sernum;
1992 iter->w.state = FWS_INIT;
1993 iter->w.node = iter->w.root;
1994 WARN_ON(iter->w.skip);
1995 iter->w.skip = iter->w.count;
1999 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2003 struct net *net = seq_file_net(seq);
2004 struct ipv6_route_iter *iter = seq->private;
2009 n = ((struct rt6_info *)v)->dst.rt6_next;
2016 ipv6_route_check_sernum(iter);
2017 read_lock(&iter->tbl->tb6_lock);
2018 r = fib6_walk_continue(&iter->w);
2019 read_unlock(&iter->tbl->tb6_lock);
2023 return iter->w.leaf;
2025 fib6_walker_unlink(&iter->w);
2028 fib6_walker_unlink(&iter->w);
2030 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2034 ipv6_route_seq_setup_walk(iter);
2038 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2041 struct net *net = seq_file_net(seq);
2042 struct ipv6_route_iter *iter = seq->private;
2045 iter->tbl = ipv6_route_seq_next_table(NULL, net);
2049 ipv6_route_seq_setup_walk(iter);
2050 return ipv6_route_seq_next(seq, NULL, pos);
2056 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2058 struct fib6_walker *w = &iter->w;
2059 return w->node && !(w->state == FWS_U && w->node == w->root);
2062 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2065 struct ipv6_route_iter *iter = seq->private;
2067 if (ipv6_route_iter_active(iter))
2068 fib6_walker_unlink(&iter->w);
2070 rcu_read_unlock_bh();
2073 static const struct seq_operations ipv6_route_seq_ops = {
2074 .start = ipv6_route_seq_start,
2075 .next = ipv6_route_seq_next,
2076 .stop = ipv6_route_seq_stop,
2077 .show = ipv6_route_seq_show
2080 int ipv6_route_open(struct inode *inode, struct file *file)
2082 return seq_open_net(inode, file, &ipv6_route_seq_ops,
2083 sizeof(struct ipv6_route_iter));
2086 #endif /* CONFIG_PROC_FS */