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_rcu_free(struct rt6_info *rt)
160 call_rcu(&rt->dst.rcu_head, dst_rcu_free);
163 static void rt6_free_pcpu(struct rt6_info *non_pcpu_rt)
167 if (!non_pcpu_rt->rt6i_pcpu)
170 for_each_possible_cpu(cpu) {
171 struct rt6_info **ppcpu_rt;
172 struct rt6_info *pcpu_rt;
174 ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
177 rt6_rcu_free(pcpu_rt);
182 free_percpu(non_pcpu_rt->rt6i_pcpu);
183 non_pcpu_rt->rt6i_pcpu = NULL;
186 static void rt6_release(struct rt6_info *rt)
188 if (atomic_dec_and_test(&rt->rt6i_ref)) {
194 static void fib6_link_table(struct net *net, struct fib6_table *tb)
199 * Initialize table lock at a single place to give lockdep a key,
200 * tables aren't visible prior to being linked to the list.
202 rwlock_init(&tb->tb6_lock);
204 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
207 * No protection necessary, this is the only list mutatation
208 * operation, tables never disappear once they exist.
210 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
213 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
215 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
217 struct fib6_table *table;
219 table = kzalloc(sizeof(*table), GFP_ATOMIC);
222 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
223 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
224 inet_peer_base_init(&table->tb6_peers);
230 struct fib6_table *fib6_new_table(struct net *net, u32 id)
232 struct fib6_table *tb;
236 tb = fib6_get_table(net, id);
240 tb = fib6_alloc_table(net, id);
242 fib6_link_table(net, tb);
247 struct fib6_table *fib6_get_table(struct net *net, u32 id)
249 struct fib6_table *tb;
250 struct hlist_head *head;
255 h = id & (FIB6_TABLE_HASHSZ - 1);
257 head = &net->ipv6.fib_table_hash[h];
258 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
259 if (tb->tb6_id == id) {
268 EXPORT_SYMBOL_GPL(fib6_get_table);
270 static void __net_init fib6_tables_init(struct net *net)
272 fib6_link_table(net, net->ipv6.fib6_main_tbl);
273 fib6_link_table(net, net->ipv6.fib6_local_tbl);
277 struct fib6_table *fib6_new_table(struct net *net, u32 id)
279 return fib6_get_table(net, id);
282 struct fib6_table *fib6_get_table(struct net *net, u32 id)
284 return net->ipv6.fib6_main_tbl;
287 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
288 int flags, pol_lookup_t lookup)
292 rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
293 if (rt->dst.error == -EAGAIN) {
295 rt = net->ipv6.ip6_null_entry;
302 static void __net_init fib6_tables_init(struct net *net)
304 fib6_link_table(net, net->ipv6.fib6_main_tbl);
309 static int fib6_dump_node(struct fib6_walker *w)
314 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
315 res = rt6_dump_route(rt, w->args);
317 /* Frame is full, suspend walking */
326 static void fib6_dump_end(struct netlink_callback *cb)
328 struct fib6_walker *w = (void *)cb->args[2];
333 fib6_walker_unlink(w);
338 cb->done = (void *)cb->args[3];
342 static int fib6_dump_done(struct netlink_callback *cb)
345 return cb->done ? cb->done(cb) : 0;
348 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
349 struct netlink_callback *cb)
351 struct fib6_walker *w;
354 w = (void *)cb->args[2];
355 w->root = &table->tb6_root;
357 if (cb->args[4] == 0) {
361 read_lock_bh(&table->tb6_lock);
363 read_unlock_bh(&table->tb6_lock);
366 cb->args[5] = w->root->fn_sernum;
369 if (cb->args[5] != w->root->fn_sernum) {
370 /* Begin at the root if the tree changed */
371 cb->args[5] = w->root->fn_sernum;
378 read_lock_bh(&table->tb6_lock);
379 res = fib6_walk_continue(w);
380 read_unlock_bh(&table->tb6_lock);
382 fib6_walker_unlink(w);
390 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
392 struct net *net = sock_net(skb->sk);
394 unsigned int e = 0, s_e;
395 struct rt6_rtnl_dump_arg arg;
396 struct fib6_walker *w;
397 struct fib6_table *tb;
398 struct hlist_head *head;
404 w = (void *)cb->args[2];
408 * 1. hook callback destructor.
410 cb->args[3] = (long)cb->done;
411 cb->done = fib6_dump_done;
414 * 2. allocate and initialize walker.
416 w = kzalloc(sizeof(*w), GFP_ATOMIC);
419 w->func = fib6_dump_node;
420 cb->args[2] = (long)w;
429 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
431 head = &net->ipv6.fib_table_hash[h];
432 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
435 res = fib6_dump_table(tb, skb, cb);
447 res = res < 0 ? res : skb->len;
456 * return the appropriate node for a routing tree "add" operation
457 * by either creating and inserting or by returning an existing
461 static struct fib6_node *fib6_add_1(struct fib6_node *root,
462 struct in6_addr *addr, int plen,
463 int offset, int allow_create,
464 int replace_required, int sernum)
466 struct fib6_node *fn, *in, *ln;
467 struct fib6_node *pn = NULL;
472 RT6_TRACE("fib6_add_1\n");
474 /* insert node in tree */
479 key = (struct rt6key *)((u8 *)fn->leaf + offset);
484 if (plen < fn->fn_bit ||
485 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
487 if (replace_required) {
488 pr_warn("Can't replace route, no match found\n");
489 return ERR_PTR(-ENOENT);
491 pr_warn("NLM_F_CREATE should be set when creating new route\n");
500 if (plen == fn->fn_bit) {
501 /* clean up an intermediate node */
502 if (!(fn->fn_flags & RTN_RTINFO)) {
503 rt6_release(fn->leaf);
507 fn->fn_sernum = sernum;
513 * We have more bits to go
516 /* Try to walk down on tree. */
517 fn->fn_sernum = sernum;
518 dir = addr_bit_set(addr, fn->fn_bit);
520 fn = dir ? fn->right : fn->left;
524 /* We should not create new node because
525 * NLM_F_REPLACE was specified without NLM_F_CREATE
526 * I assume it is safe to require NLM_F_CREATE when
527 * REPLACE flag is used! Later we may want to remove the
528 * check for replace_required, because according
529 * to netlink specification, NLM_F_CREATE
530 * MUST be specified if new route is created.
531 * That would keep IPv6 consistent with IPv4
533 if (replace_required) {
534 pr_warn("Can't replace route, no match found\n");
535 return ERR_PTR(-ENOENT);
537 pr_warn("NLM_F_CREATE should be set when creating new route\n");
540 * We walked to the bottom of tree.
541 * Create new leaf node without children.
547 return ERR_PTR(-ENOMEM);
551 ln->fn_sernum = sernum;
563 * split since we don't have a common prefix anymore or
564 * we have a less significant route.
565 * we've to insert an intermediate node on the list
566 * this new node will point to the one we need to create
572 /* find 1st bit in difference between the 2 addrs.
574 See comment in __ipv6_addr_diff: bit may be an invalid value,
575 but if it is >= plen, the value is ignored in any case.
578 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
583 * (new leaf node)[ln] (old node)[fn]
594 return ERR_PTR(-ENOMEM);
598 * new intermediate node.
600 * be off since that an address that chooses one of
601 * the branches would not match less specific routes
602 * in the other branch
609 atomic_inc(&in->leaf->rt6i_ref);
611 in->fn_sernum = sernum;
613 /* update parent pointer */
624 ln->fn_sernum = sernum;
626 if (addr_bit_set(addr, bit)) {
633 } else { /* plen <= bit */
636 * (new leaf node)[ln]
638 * (old node)[fn] NULL
644 return ERR_PTR(-ENOMEM);
650 ln->fn_sernum = sernum;
657 if (addr_bit_set(&key->addr, plen))
667 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
669 return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
673 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
677 for (i = 0; i < RTAX_MAX; i++) {
678 if (test_bit(i, mxc->mx_valid))
683 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
688 if (dst->flags & DST_HOST) {
689 u32 *mp = dst_metrics_write_ptr(dst);
694 fib6_copy_metrics(mp, mxc);
696 dst_init_metrics(dst, mxc->mx, false);
698 /* We've stolen mx now. */
705 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
708 if (atomic_read(&rt->rt6i_ref) != 1) {
709 /* This route is used as dummy address holder in some split
710 * nodes. It is not leaked, but it still holds other resources,
711 * which must be released in time. So, scan ascendant nodes
712 * and replace dummy references to this route with references
713 * to still alive ones.
716 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
717 fn->leaf = fib6_find_prefix(net, fn);
718 atomic_inc(&fn->leaf->rt6i_ref);
723 /* No more references are possible at this point. */
724 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
729 * Insert routing information in a node.
732 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
733 struct nl_info *info, struct mx6_config *mxc)
735 struct rt6_info *iter = NULL;
736 struct rt6_info **ins;
737 struct rt6_info **fallback_ins = NULL;
738 int replace = (info->nlh &&
739 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
740 int add = (!info->nlh ||
741 (info->nlh->nlmsg_flags & NLM_F_CREATE));
743 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
748 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
750 * Search for duplicates
753 if (iter->rt6i_metric == rt->rt6i_metric) {
755 * Same priority level
758 (info->nlh->nlmsg_flags & NLM_F_EXCL))
761 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
766 fallback_ins = fallback_ins ?: ins;
770 if (iter->dst.dev == rt->dst.dev &&
771 iter->rt6i_idev == rt->rt6i_idev &&
772 ipv6_addr_equal(&iter->rt6i_gateway,
773 &rt->rt6i_gateway)) {
774 if (rt->rt6i_nsiblings)
775 rt->rt6i_nsiblings = 0;
776 if (!(iter->rt6i_flags & RTF_EXPIRES))
778 if (!(rt->rt6i_flags & RTF_EXPIRES))
779 rt6_clean_expires(iter);
781 rt6_set_expires(iter, rt->dst.expires);
782 iter->rt6i_pmtu = rt->rt6i_pmtu;
785 /* If we have the same destination and the same metric,
786 * but not the same gateway, then the route we try to
787 * add is sibling to this route, increment our counter
788 * of siblings, and later we will add our route to the
790 * Only static routes (which don't have flag
791 * RTF_EXPIRES) are used for ECMPv6.
793 * To avoid long list, we only had siblings if the
794 * route have a gateway.
797 rt6_qualify_for_ecmp(iter))
798 rt->rt6i_nsiblings++;
801 if (iter->rt6i_metric > rt->rt6i_metric)
805 ins = &iter->dst.rt6_next;
808 if (fallback_ins && !found) {
809 /* No ECMP-able route found, replace first non-ECMP one */
815 /* Reset round-robin state, if necessary */
816 if (ins == &fn->leaf)
819 /* Link this route to others same route. */
820 if (rt->rt6i_nsiblings) {
821 unsigned int rt6i_nsiblings;
822 struct rt6_info *sibling, *temp_sibling;
824 /* Find the first route that have the same metric */
827 if (sibling->rt6i_metric == rt->rt6i_metric &&
828 rt6_qualify_for_ecmp(sibling)) {
829 list_add_tail(&rt->rt6i_siblings,
830 &sibling->rt6i_siblings);
833 sibling = sibling->dst.rt6_next;
835 /* For each sibling in the list, increment the counter of
836 * siblings. BUG() if counters does not match, list of siblings
840 list_for_each_entry_safe(sibling, temp_sibling,
841 &rt->rt6i_siblings, rt6i_siblings) {
842 sibling->rt6i_nsiblings++;
843 BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
846 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
854 pr_warn("NLM_F_CREATE should be set when creating new route\n");
857 err = fib6_commit_metrics(&rt->dst, mxc);
861 rt->dst.rt6_next = iter;
864 atomic_inc(&rt->rt6i_ref);
865 inet6_rt_notify(RTM_NEWROUTE, rt, info, 0);
866 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
868 if (!(fn->fn_flags & RTN_RTINFO)) {
869 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
870 fn->fn_flags |= RTN_RTINFO;
879 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
883 err = fib6_commit_metrics(&rt->dst, mxc);
889 rt->dst.rt6_next = iter->dst.rt6_next;
890 atomic_inc(&rt->rt6i_ref);
891 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
892 if (!(fn->fn_flags & RTN_RTINFO)) {
893 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
894 fn->fn_flags |= RTN_RTINFO;
896 nsiblings = iter->rt6i_nsiblings;
897 fib6_purge_rt(iter, fn, info->nl_net);
901 /* Replacing an ECMP route, remove all siblings */
902 ins = &rt->dst.rt6_next;
905 if (iter->rt6i_metric > rt->rt6i_metric)
907 if (rt6_qualify_for_ecmp(iter)) {
908 *ins = iter->dst.rt6_next;
909 fib6_purge_rt(iter, fn, info->nl_net);
913 ins = &iter->dst.rt6_next;
917 WARN_ON(nsiblings != 0);
924 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
926 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
927 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
928 mod_timer(&net->ipv6.ip6_fib_timer,
929 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
932 void fib6_force_start_gc(struct net *net)
934 if (!timer_pending(&net->ipv6.ip6_fib_timer))
935 mod_timer(&net->ipv6.ip6_fib_timer,
936 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
940 * Add routing information to the routing tree.
941 * <destination addr>/<source addr>
942 * with source addr info in sub-trees
945 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
946 struct nl_info *info, struct mx6_config *mxc)
948 struct fib6_node *fn, *pn = NULL;
950 int allow_create = 1;
951 int replace_required = 0;
952 int sernum = fib6_new_sernum(info->nl_net);
954 if (WARN_ON_ONCE((rt->dst.flags & DST_NOCACHE) &&
955 !atomic_read(&rt->dst.__refcnt)))
959 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
961 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
962 replace_required = 1;
964 if (!allow_create && !replace_required)
965 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
967 fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
968 offsetof(struct rt6_info, rt6i_dst), allow_create,
969 replace_required, sernum);
978 #ifdef CONFIG_IPV6_SUBTREES
979 if (rt->rt6i_src.plen) {
980 struct fib6_node *sn;
983 struct fib6_node *sfn;
995 /* Create subtree root node */
1000 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
1001 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
1002 sfn->fn_flags = RTN_ROOT;
1003 sfn->fn_sernum = sernum;
1005 /* Now add the first leaf node to new subtree */
1007 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
1009 offsetof(struct rt6_info, rt6i_src),
1010 allow_create, replace_required, sernum);
1013 /* If it is failed, discard just allocated
1014 root, and then (in st_failure) stale node
1022 /* Now link new subtree to main tree */
1026 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
1028 offsetof(struct rt6_info, rt6i_src),
1029 allow_create, replace_required, sernum);
1039 atomic_inc(&rt->rt6i_ref);
1045 err = fib6_add_rt2node(fn, rt, info, mxc);
1047 fib6_start_gc(info->nl_net, rt);
1048 if (!(rt->rt6i_flags & RTF_CACHE))
1049 fib6_prune_clones(info->nl_net, pn);
1050 rt->dst.flags &= ~DST_NOCACHE;
1055 #ifdef CONFIG_IPV6_SUBTREES
1057 * If fib6_add_1 has cleared the old leaf pointer in the
1058 * super-tree leaf node we have to find a new one for it.
1060 if (pn != fn && pn->leaf == rt) {
1062 atomic_dec(&rt->rt6i_ref);
1064 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
1065 pn->leaf = fib6_find_prefix(info->nl_net, pn);
1068 WARN_ON(pn->leaf == NULL);
1069 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
1072 atomic_inc(&pn->leaf->rt6i_ref);
1075 if (!(rt->dst.flags & DST_NOCACHE))
1080 #ifdef CONFIG_IPV6_SUBTREES
1081 /* Subtree creation failed, probably main tree node
1082 is orphan. If it is, shoot it.
1085 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1086 fib6_repair_tree(info->nl_net, fn);
1087 if (!(rt->dst.flags & DST_NOCACHE))
1094 * Routing tree lookup
1098 struct lookup_args {
1099 int offset; /* key offset on rt6_info */
1100 const struct in6_addr *addr; /* search key */
1103 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1104 struct lookup_args *args)
1106 struct fib6_node *fn;
1109 if (unlikely(args->offset == 0))
1119 struct fib6_node *next;
1121 dir = addr_bit_set(args->addr, fn->fn_bit);
1123 next = dir ? fn->right : fn->left;
1133 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1136 key = (struct rt6key *) ((u8 *) fn->leaf +
1139 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1140 #ifdef CONFIG_IPV6_SUBTREES
1142 struct fib6_node *sfn;
1143 sfn = fib6_lookup_1(fn->subtree,
1150 if (fn->fn_flags & RTN_RTINFO)
1154 #ifdef CONFIG_IPV6_SUBTREES
1157 if (fn->fn_flags & RTN_ROOT)
1166 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1167 const struct in6_addr *saddr)
1169 struct fib6_node *fn;
1170 struct lookup_args args[] = {
1172 .offset = offsetof(struct rt6_info, rt6i_dst),
1175 #ifdef CONFIG_IPV6_SUBTREES
1177 .offset = offsetof(struct rt6_info, rt6i_src),
1182 .offset = 0, /* sentinel */
1186 fn = fib6_lookup_1(root, daddr ? args : args + 1);
1187 if (!fn || fn->fn_flags & RTN_TL_ROOT)
1194 * Get node with specified destination prefix (and source prefix,
1195 * if subtrees are used)
1199 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1200 const struct in6_addr *addr,
1201 int plen, int offset)
1203 struct fib6_node *fn;
1205 for (fn = root; fn ; ) {
1206 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1211 if (plen < fn->fn_bit ||
1212 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1215 if (plen == fn->fn_bit)
1219 * We have more bits to go
1221 if (addr_bit_set(addr, fn->fn_bit))
1229 struct fib6_node *fib6_locate(struct fib6_node *root,
1230 const struct in6_addr *daddr, int dst_len,
1231 const struct in6_addr *saddr, int src_len)
1233 struct fib6_node *fn;
1235 fn = fib6_locate_1(root, daddr, dst_len,
1236 offsetof(struct rt6_info, rt6i_dst));
1238 #ifdef CONFIG_IPV6_SUBTREES
1240 WARN_ON(saddr == NULL);
1241 if (fn && fn->subtree)
1242 fn = fib6_locate_1(fn->subtree, saddr, src_len,
1243 offsetof(struct rt6_info, rt6i_src));
1247 if (fn && fn->fn_flags & RTN_RTINFO)
1259 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1261 if (fn->fn_flags & RTN_ROOT)
1262 return net->ipv6.ip6_null_entry;
1266 return fn->left->leaf;
1268 return fn->right->leaf;
1270 fn = FIB6_SUBTREE(fn);
1276 * Called to trim the tree of intermediate nodes when possible. "fn"
1277 * is the node we want to try and remove.
1280 static struct fib6_node *fib6_repair_tree(struct net *net,
1281 struct fib6_node *fn)
1285 struct fib6_node *child, *pn;
1286 struct fib6_walker *w;
1290 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1293 WARN_ON(fn->fn_flags & RTN_RTINFO);
1294 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1300 child = fn->right, children |= 1;
1302 child = fn->left, children |= 2;
1304 if (children == 3 || FIB6_SUBTREE(fn)
1305 #ifdef CONFIG_IPV6_SUBTREES
1306 /* Subtree root (i.e. fn) may have one child */
1307 || (children && fn->fn_flags & RTN_ROOT)
1310 fn->leaf = fib6_find_prefix(net, fn);
1314 fn->leaf = net->ipv6.ip6_null_entry;
1317 atomic_inc(&fn->leaf->rt6i_ref);
1322 #ifdef CONFIG_IPV6_SUBTREES
1323 if (FIB6_SUBTREE(pn) == fn) {
1324 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1325 FIB6_SUBTREE(pn) = NULL;
1328 WARN_ON(fn->fn_flags & RTN_ROOT);
1330 if (pn->right == fn)
1332 else if (pn->left == fn)
1341 #ifdef CONFIG_IPV6_SUBTREES
1345 read_lock(&fib6_walker_lock);
1348 if (w->root == fn) {
1349 w->root = w->node = NULL;
1350 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1351 } else if (w->node == fn) {
1352 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1357 if (w->root == fn) {
1359 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1361 if (w->node == fn) {
1364 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1365 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1367 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1368 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1373 read_unlock(&fib6_walker_lock);
1376 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1379 rt6_release(pn->leaf);
1385 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1386 struct nl_info *info)
1388 struct fib6_walker *w;
1389 struct rt6_info *rt = *rtp;
1390 struct net *net = info->nl_net;
1392 RT6_TRACE("fib6_del_route\n");
1395 *rtp = rt->dst.rt6_next;
1396 rt->rt6i_node = NULL;
1397 net->ipv6.rt6_stats->fib_rt_entries--;
1398 net->ipv6.rt6_stats->fib_discarded_routes++;
1400 /* Reset round-robin state, if necessary */
1401 if (fn->rr_ptr == rt)
1404 /* Remove this entry from other siblings */
1405 if (rt->rt6i_nsiblings) {
1406 struct rt6_info *sibling, *next_sibling;
1408 list_for_each_entry_safe(sibling, next_sibling,
1409 &rt->rt6i_siblings, rt6i_siblings)
1410 sibling->rt6i_nsiblings--;
1411 rt->rt6i_nsiblings = 0;
1412 list_del_init(&rt->rt6i_siblings);
1415 /* Adjust walkers */
1416 read_lock(&fib6_walker_lock);
1418 if (w->state == FWS_C && w->leaf == rt) {
1419 RT6_TRACE("walker %p adjusted by delroute\n", w);
1420 w->leaf = rt->dst.rt6_next;
1425 read_unlock(&fib6_walker_lock);
1427 rt->dst.rt6_next = NULL;
1429 /* If it was last route, expunge its radix tree node */
1431 fn->fn_flags &= ~RTN_RTINFO;
1432 net->ipv6.rt6_stats->fib_route_nodes--;
1433 fn = fib6_repair_tree(net, fn);
1436 fib6_purge_rt(rt, fn, net);
1438 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1442 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1444 struct net *net = info->nl_net;
1445 struct fib6_node *fn = rt->rt6i_node;
1446 struct rt6_info **rtp;
1449 if (rt->dst.obsolete > 0) {
1454 if (!fn || rt == net->ipv6.ip6_null_entry)
1457 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1459 if (!(rt->rt6i_flags & RTF_CACHE)) {
1460 struct fib6_node *pn = fn;
1461 #ifdef CONFIG_IPV6_SUBTREES
1462 /* clones of this route might be in another subtree */
1463 if (rt->rt6i_src.plen) {
1464 while (!(pn->fn_flags & RTN_ROOT))
1469 fib6_prune_clones(info->nl_net, pn);
1473 * Walk the leaf entries looking for ourself
1476 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1478 fib6_del_route(fn, rtp, info);
1486 * Tree traversal function.
1488 * Certainly, it is not interrupt safe.
1489 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1490 * It means, that we can modify tree during walking
1491 * and use this function for garbage collection, clone pruning,
1492 * cleaning tree when a device goes down etc. etc.
1494 * It guarantees that every node will be traversed,
1495 * and that it will be traversed only once.
1497 * Callback function w->func may return:
1498 * 0 -> continue walking.
1499 * positive value -> walking is suspended (used by tree dumps,
1500 * and probably by gc, if it will be split to several slices)
1501 * negative value -> terminate walking.
1503 * The function itself returns:
1504 * 0 -> walk is complete.
1505 * >0 -> walk is incomplete (i.e. suspended)
1506 * <0 -> walk is terminated by an error.
1509 static int fib6_walk_continue(struct fib6_walker *w)
1511 struct fib6_node *fn, *pn;
1518 if (w->prune && fn != w->root &&
1519 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1524 #ifdef CONFIG_IPV6_SUBTREES
1526 if (FIB6_SUBTREE(fn)) {
1527 w->node = FIB6_SUBTREE(fn);
1535 w->state = FWS_INIT;
1541 w->node = fn->right;
1542 w->state = FWS_INIT;
1548 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1570 #ifdef CONFIG_IPV6_SUBTREES
1571 if (FIB6_SUBTREE(pn) == fn) {
1572 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1577 if (pn->left == fn) {
1581 if (pn->right == fn) {
1583 w->leaf = w->node->leaf;
1593 static int fib6_walk(struct fib6_walker *w)
1597 w->state = FWS_INIT;
1600 fib6_walker_link(w);
1601 res = fib6_walk_continue(w);
1603 fib6_walker_unlink(w);
1607 static int fib6_clean_node(struct fib6_walker *w)
1610 struct rt6_info *rt;
1611 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1612 struct nl_info info = {
1616 if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1617 w->node->fn_sernum != c->sernum)
1618 w->node->fn_sernum = c->sernum;
1621 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1626 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1627 res = c->func(rt, c->arg);
1630 res = fib6_del(rt, &info);
1633 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1634 __func__, rt, rt->rt6i_node, res);
1647 * Convenient frontend to tree walker.
1649 * func is called on each route.
1650 * It may return -1 -> delete this route.
1651 * 0 -> continue walking
1653 * prune==1 -> only immediate children of node (certainly,
1654 * ignoring pure split nodes) will be scanned.
1657 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1658 int (*func)(struct rt6_info *, void *arg),
1659 bool prune, int sernum, void *arg)
1661 struct fib6_cleaner c;
1664 c.w.func = fib6_clean_node;
1676 static void __fib6_clean_all(struct net *net,
1677 int (*func)(struct rt6_info *, void *),
1678 int sernum, void *arg)
1680 struct fib6_table *table;
1681 struct hlist_head *head;
1685 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1686 head = &net->ipv6.fib_table_hash[h];
1687 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1688 write_lock_bh(&table->tb6_lock);
1689 fib6_clean_tree(net, &table->tb6_root,
1690 func, false, sernum, arg);
1691 write_unlock_bh(&table->tb6_lock);
1697 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1700 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1703 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1705 if (rt->rt6i_flags & RTF_CACHE) {
1706 RT6_TRACE("pruning clone %p\n", rt);
1713 static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1715 fib6_clean_tree(net, fn, fib6_prune_clone, true,
1716 FIB6_NO_SERNUM_CHANGE, NULL);
1719 static void fib6_flush_trees(struct net *net)
1721 int new_sernum = fib6_new_sernum(net);
1723 __fib6_clean_all(net, NULL, new_sernum, NULL);
1727 * Garbage collection
1730 static struct fib6_gc_args
1736 static int fib6_age(struct rt6_info *rt, void *arg)
1738 unsigned long now = jiffies;
1741 * check addrconf expiration here.
1742 * Routes are expired even if they are in use.
1744 * Also age clones. Note, that clones are aged out
1745 * only if they are not in use now.
1748 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1749 if (time_after(now, rt->dst.expires)) {
1750 RT6_TRACE("expiring %p\n", rt);
1754 } else if (rt->rt6i_flags & RTF_CACHE) {
1755 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1756 time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1757 RT6_TRACE("aging clone %p\n", rt);
1759 } else if (rt->rt6i_flags & RTF_GATEWAY) {
1760 struct neighbour *neigh;
1761 __u8 neigh_flags = 0;
1763 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1765 neigh_flags = neigh->flags;
1766 neigh_release(neigh);
1768 if (!(neigh_flags & NTF_ROUTER)) {
1769 RT6_TRACE("purging route %p via non-router but gateway\n",
1780 static DEFINE_SPINLOCK(fib6_gc_lock);
1782 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1787 spin_lock_bh(&fib6_gc_lock);
1788 } else if (!spin_trylock_bh(&fib6_gc_lock)) {
1789 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1792 gc_args.timeout = expires ? (int)expires :
1793 net->ipv6.sysctl.ip6_rt_gc_interval;
1795 gc_args.more = icmp6_dst_gc();
1797 fib6_clean_all(net, fib6_age, NULL);
1799 net->ipv6.ip6_rt_last_gc = now;
1802 mod_timer(&net->ipv6.ip6_fib_timer,
1804 + net->ipv6.sysctl.ip6_rt_gc_interval));
1806 del_timer(&net->ipv6.ip6_fib_timer);
1807 spin_unlock_bh(&fib6_gc_lock);
1810 static void fib6_gc_timer_cb(unsigned long arg)
1812 fib6_run_gc(0, (struct net *)arg, true);
1815 static int __net_init fib6_net_init(struct net *net)
1817 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1819 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1821 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1822 if (!net->ipv6.rt6_stats)
1825 /* Avoid false sharing : Use at least a full cache line */
1826 size = max_t(size_t, size, L1_CACHE_BYTES);
1828 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1829 if (!net->ipv6.fib_table_hash)
1832 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1834 if (!net->ipv6.fib6_main_tbl)
1835 goto out_fib_table_hash;
1837 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1838 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1839 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1840 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1841 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1843 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1844 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1846 if (!net->ipv6.fib6_local_tbl)
1847 goto out_fib6_main_tbl;
1848 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1849 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1850 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1851 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1852 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1854 fib6_tables_init(net);
1858 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1860 kfree(net->ipv6.fib6_main_tbl);
1863 kfree(net->ipv6.fib_table_hash);
1865 kfree(net->ipv6.rt6_stats);
1870 static void fib6_net_exit(struct net *net)
1872 rt6_ifdown(net, NULL);
1873 del_timer_sync(&net->ipv6.ip6_fib_timer);
1875 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1876 inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1877 kfree(net->ipv6.fib6_local_tbl);
1879 inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1880 kfree(net->ipv6.fib6_main_tbl);
1881 kfree(net->ipv6.fib_table_hash);
1882 kfree(net->ipv6.rt6_stats);
1885 static struct pernet_operations fib6_net_ops = {
1886 .init = fib6_net_init,
1887 .exit = fib6_net_exit,
1890 int __init fib6_init(void)
1894 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1895 sizeof(struct fib6_node),
1896 0, SLAB_HWCACHE_ALIGN,
1898 if (!fib6_node_kmem)
1901 ret = register_pernet_subsys(&fib6_net_ops);
1903 goto out_kmem_cache_create;
1905 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1908 goto out_unregister_subsys;
1910 __fib6_flush_trees = fib6_flush_trees;
1914 out_unregister_subsys:
1915 unregister_pernet_subsys(&fib6_net_ops);
1916 out_kmem_cache_create:
1917 kmem_cache_destroy(fib6_node_kmem);
1921 void fib6_gc_cleanup(void)
1923 unregister_pernet_subsys(&fib6_net_ops);
1924 kmem_cache_destroy(fib6_node_kmem);
1927 #ifdef CONFIG_PROC_FS
1929 struct ipv6_route_iter {
1930 struct seq_net_private p;
1931 struct fib6_walker w;
1933 struct fib6_table *tbl;
1937 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
1939 struct rt6_info *rt = v;
1940 struct ipv6_route_iter *iter = seq->private;
1942 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
1944 #ifdef CONFIG_IPV6_SUBTREES
1945 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
1947 seq_puts(seq, "00000000000000000000000000000000 00 ");
1949 if (rt->rt6i_flags & RTF_GATEWAY)
1950 seq_printf(seq, "%pi6", &rt->rt6i_gateway);
1952 seq_puts(seq, "00000000000000000000000000000000");
1954 seq_printf(seq, " %08x %08x %08x %08x %8s\n",
1955 rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
1956 rt->dst.__use, rt->rt6i_flags,
1957 rt->dst.dev ? rt->dst.dev->name : "");
1958 iter->w.leaf = NULL;
1962 static int ipv6_route_yield(struct fib6_walker *w)
1964 struct ipv6_route_iter *iter = w->args;
1970 iter->w.leaf = iter->w.leaf->dst.rt6_next;
1972 if (!iter->skip && iter->w.leaf)
1974 } while (iter->w.leaf);
1979 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter)
1981 memset(&iter->w, 0, sizeof(iter->w));
1982 iter->w.func = ipv6_route_yield;
1983 iter->w.root = &iter->tbl->tb6_root;
1984 iter->w.state = FWS_INIT;
1985 iter->w.node = iter->w.root;
1986 iter->w.args = iter;
1987 iter->sernum = iter->w.root->fn_sernum;
1988 INIT_LIST_HEAD(&iter->w.lh);
1989 fib6_walker_link(&iter->w);
1992 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
1996 struct hlist_node *node;
1999 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2000 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2006 while (!node && h < FIB6_TABLE_HASHSZ) {
2007 node = rcu_dereference_bh(
2008 hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2010 return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2013 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2015 if (iter->sernum != iter->w.root->fn_sernum) {
2016 iter->sernum = iter->w.root->fn_sernum;
2017 iter->w.state = FWS_INIT;
2018 iter->w.node = iter->w.root;
2019 WARN_ON(iter->w.skip);
2020 iter->w.skip = iter->w.count;
2024 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2028 struct net *net = seq_file_net(seq);
2029 struct ipv6_route_iter *iter = seq->private;
2034 n = ((struct rt6_info *)v)->dst.rt6_next;
2041 ipv6_route_check_sernum(iter);
2042 read_lock(&iter->tbl->tb6_lock);
2043 r = fib6_walk_continue(&iter->w);
2044 read_unlock(&iter->tbl->tb6_lock);
2048 return iter->w.leaf;
2050 fib6_walker_unlink(&iter->w);
2053 fib6_walker_unlink(&iter->w);
2055 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2059 ipv6_route_seq_setup_walk(iter);
2063 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2066 struct net *net = seq_file_net(seq);
2067 struct ipv6_route_iter *iter = seq->private;
2070 iter->tbl = ipv6_route_seq_next_table(NULL, net);
2074 ipv6_route_seq_setup_walk(iter);
2075 return ipv6_route_seq_next(seq, NULL, pos);
2081 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2083 struct fib6_walker *w = &iter->w;
2084 return w->node && !(w->state == FWS_U && w->node == w->root);
2087 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2090 struct ipv6_route_iter *iter = seq->private;
2092 if (ipv6_route_iter_active(iter))
2093 fib6_walker_unlink(&iter->w);
2095 rcu_read_unlock_bh();
2098 static const struct seq_operations ipv6_route_seq_ops = {
2099 .start = ipv6_route_seq_start,
2100 .next = ipv6_route_seq_next,
2101 .stop = ipv6_route_seq_stop,
2102 .show = ipv6_route_seq_show
2105 int ipv6_route_open(struct inode *inode, struct file *file)
2107 return seq_open_net(inode, file, &ipv6_route_seq_ops,
2108 sizeof(struct ipv6_route_iter));
2111 #endif /* CONFIG_PROC_FS */