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 non_pcpu_rt->rt6i_pcpu = NULL;
185 static void rt6_release(struct rt6_info *rt)
187 if (atomic_dec_and_test(&rt->rt6i_ref)) {
193 static void fib6_link_table(struct net *net, struct fib6_table *tb)
198 * Initialize table lock at a single place to give lockdep a key,
199 * tables aren't visible prior to being linked to the list.
201 rwlock_init(&tb->tb6_lock);
203 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
206 * No protection necessary, this is the only list mutatation
207 * operation, tables never disappear once they exist.
209 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
212 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
214 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
216 struct fib6_table *table;
218 table = kzalloc(sizeof(*table), GFP_ATOMIC);
221 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
222 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
223 inet_peer_base_init(&table->tb6_peers);
229 struct fib6_table *fib6_new_table(struct net *net, u32 id)
231 struct fib6_table *tb;
235 tb = fib6_get_table(net, id);
239 tb = fib6_alloc_table(net, id);
241 fib6_link_table(net, tb);
246 struct fib6_table *fib6_get_table(struct net *net, u32 id)
248 struct fib6_table *tb;
249 struct hlist_head *head;
254 h = id & (FIB6_TABLE_HASHSZ - 1);
256 head = &net->ipv6.fib_table_hash[h];
257 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
258 if (tb->tb6_id == id) {
267 EXPORT_SYMBOL_GPL(fib6_get_table);
269 static void __net_init fib6_tables_init(struct net *net)
271 fib6_link_table(net, net->ipv6.fib6_main_tbl);
272 fib6_link_table(net, net->ipv6.fib6_local_tbl);
276 struct fib6_table *fib6_new_table(struct net *net, u32 id)
278 return fib6_get_table(net, id);
281 struct fib6_table *fib6_get_table(struct net *net, u32 id)
283 return net->ipv6.fib6_main_tbl;
286 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
287 int flags, pol_lookup_t lookup)
289 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
292 static void __net_init fib6_tables_init(struct net *net)
294 fib6_link_table(net, net->ipv6.fib6_main_tbl);
299 static int fib6_dump_node(struct fib6_walker *w)
304 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
305 res = rt6_dump_route(rt, w->args);
307 /* Frame is full, suspend walking */
316 static void fib6_dump_end(struct netlink_callback *cb)
318 struct fib6_walker *w = (void *)cb->args[2];
323 fib6_walker_unlink(w);
328 cb->done = (void *)cb->args[3];
332 static int fib6_dump_done(struct netlink_callback *cb)
335 return cb->done ? cb->done(cb) : 0;
338 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
339 struct netlink_callback *cb)
341 struct fib6_walker *w;
344 w = (void *)cb->args[2];
345 w->root = &table->tb6_root;
347 if (cb->args[4] == 0) {
351 read_lock_bh(&table->tb6_lock);
353 read_unlock_bh(&table->tb6_lock);
356 cb->args[5] = w->root->fn_sernum;
359 if (cb->args[5] != w->root->fn_sernum) {
360 /* Begin at the root if the tree changed */
361 cb->args[5] = w->root->fn_sernum;
368 read_lock_bh(&table->tb6_lock);
369 res = fib6_walk_continue(w);
370 read_unlock_bh(&table->tb6_lock);
372 fib6_walker_unlink(w);
380 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
382 struct net *net = sock_net(skb->sk);
384 unsigned int e = 0, s_e;
385 struct rt6_rtnl_dump_arg arg;
386 struct fib6_walker *w;
387 struct fib6_table *tb;
388 struct hlist_head *head;
394 w = (void *)cb->args[2];
398 * 1. hook callback destructor.
400 cb->args[3] = (long)cb->done;
401 cb->done = fib6_dump_done;
404 * 2. allocate and initialize walker.
406 w = kzalloc(sizeof(*w), GFP_ATOMIC);
409 w->func = fib6_dump_node;
410 cb->args[2] = (long)w;
419 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
421 head = &net->ipv6.fib_table_hash[h];
422 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
425 res = fib6_dump_table(tb, skb, cb);
437 res = res < 0 ? res : skb->len;
446 * return the appropriate node for a routing tree "add" operation
447 * by either creating and inserting or by returning an existing
451 static struct fib6_node *fib6_add_1(struct fib6_node *root,
452 struct in6_addr *addr, int plen,
453 int offset, int allow_create,
454 int replace_required, int sernum)
456 struct fib6_node *fn, *in, *ln;
457 struct fib6_node *pn = NULL;
462 RT6_TRACE("fib6_add_1\n");
464 /* insert node in tree */
469 key = (struct rt6key *)((u8 *)fn->leaf + offset);
474 if (plen < fn->fn_bit ||
475 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
477 if (replace_required) {
478 pr_warn("Can't replace route, no match found\n");
479 return ERR_PTR(-ENOENT);
481 pr_warn("NLM_F_CREATE should be set when creating new route\n");
490 if (plen == fn->fn_bit) {
491 /* clean up an intermediate node */
492 if (!(fn->fn_flags & RTN_RTINFO)) {
493 rt6_release(fn->leaf);
497 fn->fn_sernum = sernum;
503 * We have more bits to go
506 /* Try to walk down on tree. */
507 fn->fn_sernum = sernum;
508 dir = addr_bit_set(addr, fn->fn_bit);
510 fn = dir ? fn->right : fn->left;
514 /* We should not create new node because
515 * NLM_F_REPLACE was specified without NLM_F_CREATE
516 * I assume it is safe to require NLM_F_CREATE when
517 * REPLACE flag is used! Later we may want to remove the
518 * check for replace_required, because according
519 * to netlink specification, NLM_F_CREATE
520 * MUST be specified if new route is created.
521 * That would keep IPv6 consistent with IPv4
523 if (replace_required) {
524 pr_warn("Can't replace route, no match found\n");
525 return ERR_PTR(-ENOENT);
527 pr_warn("NLM_F_CREATE should be set when creating new route\n");
530 * We walked to the bottom of tree.
531 * Create new leaf node without children.
537 return ERR_PTR(-ENOMEM);
541 ln->fn_sernum = sernum;
553 * split since we don't have a common prefix anymore or
554 * we have a less significant route.
555 * we've to insert an intermediate node on the list
556 * this new node will point to the one we need to create
562 /* find 1st bit in difference between the 2 addrs.
564 See comment in __ipv6_addr_diff: bit may be an invalid value,
565 but if it is >= plen, the value is ignored in any case.
568 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
573 * (new leaf node)[ln] (old node)[fn]
584 return ERR_PTR(-ENOMEM);
588 * new intermediate node.
590 * be off since that an address that chooses one of
591 * the branches would not match less specific routes
592 * in the other branch
599 atomic_inc(&in->leaf->rt6i_ref);
601 in->fn_sernum = sernum;
603 /* update parent pointer */
614 ln->fn_sernum = sernum;
616 if (addr_bit_set(addr, bit)) {
623 } else { /* plen <= bit */
626 * (new leaf node)[ln]
628 * (old node)[fn] NULL
634 return ERR_PTR(-ENOMEM);
640 ln->fn_sernum = sernum;
647 if (addr_bit_set(&key->addr, plen))
657 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
659 return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
663 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
667 for (i = 0; i < RTAX_MAX; i++) {
668 if (test_bit(i, mxc->mx_valid))
673 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
678 if (dst->flags & DST_HOST) {
679 u32 *mp = dst_metrics_write_ptr(dst);
684 fib6_copy_metrics(mp, mxc);
686 dst_init_metrics(dst, mxc->mx, false);
688 /* We've stolen mx now. */
695 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
698 if (atomic_read(&rt->rt6i_ref) != 1) {
699 /* This route is used as dummy address holder in some split
700 * nodes. It is not leaked, but it still holds other resources,
701 * which must be released in time. So, scan ascendant nodes
702 * and replace dummy references to this route with references
703 * to still alive ones.
706 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
707 fn->leaf = fib6_find_prefix(net, fn);
708 atomic_inc(&fn->leaf->rt6i_ref);
713 /* No more references are possible at this point. */
714 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
719 * Insert routing information in a node.
722 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
723 struct nl_info *info, struct mx6_config *mxc)
725 struct rt6_info *iter = NULL;
726 struct rt6_info **ins;
727 struct rt6_info **fallback_ins = NULL;
728 int replace = (info->nlh &&
729 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
730 int add = (!info->nlh ||
731 (info->nlh->nlmsg_flags & NLM_F_CREATE));
733 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
738 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
740 * Search for duplicates
743 if (iter->rt6i_metric == rt->rt6i_metric) {
745 * Same priority level
748 (info->nlh->nlmsg_flags & NLM_F_EXCL))
751 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
756 fallback_ins = fallback_ins ?: ins;
760 if (iter->dst.dev == rt->dst.dev &&
761 iter->rt6i_idev == rt->rt6i_idev &&
762 ipv6_addr_equal(&iter->rt6i_gateway,
763 &rt->rt6i_gateway)) {
764 if (rt->rt6i_nsiblings)
765 rt->rt6i_nsiblings = 0;
766 if (!(iter->rt6i_flags & RTF_EXPIRES))
768 if (!(rt->rt6i_flags & RTF_EXPIRES))
769 rt6_clean_expires(iter);
771 rt6_set_expires(iter, rt->dst.expires);
772 iter->rt6i_pmtu = rt->rt6i_pmtu;
775 /* If we have the same destination and the same metric,
776 * but not the same gateway, then the route we try to
777 * add is sibling to this route, increment our counter
778 * of siblings, and later we will add our route to the
780 * Only static routes (which don't have flag
781 * RTF_EXPIRES) are used for ECMPv6.
783 * To avoid long list, we only had siblings if the
784 * route have a gateway.
787 rt6_qualify_for_ecmp(iter))
788 rt->rt6i_nsiblings++;
791 if (iter->rt6i_metric > rt->rt6i_metric)
795 ins = &iter->dst.rt6_next;
798 if (fallback_ins && !found) {
799 /* No ECMP-able route found, replace first non-ECMP one */
805 /* Reset round-robin state, if necessary */
806 if (ins == &fn->leaf)
809 /* Link this route to others same route. */
810 if (rt->rt6i_nsiblings) {
811 unsigned int rt6i_nsiblings;
812 struct rt6_info *sibling, *temp_sibling;
814 /* Find the first route that have the same metric */
817 if (sibling->rt6i_metric == rt->rt6i_metric &&
818 rt6_qualify_for_ecmp(sibling)) {
819 list_add_tail(&rt->rt6i_siblings,
820 &sibling->rt6i_siblings);
823 sibling = sibling->dst.rt6_next;
825 /* For each sibling in the list, increment the counter of
826 * siblings. BUG() if counters does not match, list of siblings
830 list_for_each_entry_safe(sibling, temp_sibling,
831 &rt->rt6i_siblings, rt6i_siblings) {
832 sibling->rt6i_nsiblings++;
833 BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
836 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
844 pr_warn("NLM_F_CREATE should be set when creating new route\n");
847 err = fib6_commit_metrics(&rt->dst, mxc);
851 rt->dst.rt6_next = iter;
854 atomic_inc(&rt->rt6i_ref);
855 inet6_rt_notify(RTM_NEWROUTE, rt, info, 0);
856 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
858 if (!(fn->fn_flags & RTN_RTINFO)) {
859 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
860 fn->fn_flags |= RTN_RTINFO;
869 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
873 err = fib6_commit_metrics(&rt->dst, mxc);
879 rt->dst.rt6_next = iter->dst.rt6_next;
880 atomic_inc(&rt->rt6i_ref);
881 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
882 if (!(fn->fn_flags & RTN_RTINFO)) {
883 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
884 fn->fn_flags |= RTN_RTINFO;
886 nsiblings = iter->rt6i_nsiblings;
887 fib6_purge_rt(iter, fn, info->nl_net);
891 /* Replacing an ECMP route, remove all siblings */
892 ins = &rt->dst.rt6_next;
895 if (rt6_qualify_for_ecmp(iter)) {
896 *ins = iter->dst.rt6_next;
897 fib6_purge_rt(iter, fn, info->nl_net);
901 ins = &iter->dst.rt6_next;
905 WARN_ON(nsiblings != 0);
912 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
914 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
915 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
916 mod_timer(&net->ipv6.ip6_fib_timer,
917 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
920 void fib6_force_start_gc(struct net *net)
922 if (!timer_pending(&net->ipv6.ip6_fib_timer))
923 mod_timer(&net->ipv6.ip6_fib_timer,
924 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
928 * Add routing information to the routing tree.
929 * <destination addr>/<source addr>
930 * with source addr info in sub-trees
933 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
934 struct nl_info *info, struct mx6_config *mxc)
936 struct fib6_node *fn, *pn = NULL;
938 int allow_create = 1;
939 int replace_required = 0;
940 int sernum = fib6_new_sernum(info->nl_net);
942 if (WARN_ON_ONCE((rt->dst.flags & DST_NOCACHE) &&
943 !atomic_read(&rt->dst.__refcnt)))
947 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
949 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
950 replace_required = 1;
952 if (!allow_create && !replace_required)
953 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
955 fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
956 offsetof(struct rt6_info, rt6i_dst), allow_create,
957 replace_required, sernum);
966 #ifdef CONFIG_IPV6_SUBTREES
967 if (rt->rt6i_src.plen) {
968 struct fib6_node *sn;
971 struct fib6_node *sfn;
983 /* Create subtree root node */
988 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
989 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
990 sfn->fn_flags = RTN_ROOT;
991 sfn->fn_sernum = sernum;
993 /* Now add the first leaf node to new subtree */
995 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
997 offsetof(struct rt6_info, rt6i_src),
998 allow_create, replace_required, sernum);
1001 /* If it is failed, discard just allocated
1002 root, and then (in st_failure) stale node
1010 /* Now link new subtree to main tree */
1014 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
1016 offsetof(struct rt6_info, rt6i_src),
1017 allow_create, replace_required, sernum);
1027 atomic_inc(&rt->rt6i_ref);
1033 err = fib6_add_rt2node(fn, rt, info, mxc);
1035 fib6_start_gc(info->nl_net, rt);
1036 if (!(rt->rt6i_flags & RTF_CACHE))
1037 fib6_prune_clones(info->nl_net, pn);
1038 rt->dst.flags &= ~DST_NOCACHE;
1043 #ifdef CONFIG_IPV6_SUBTREES
1045 * If fib6_add_1 has cleared the old leaf pointer in the
1046 * super-tree leaf node we have to find a new one for it.
1048 if (pn != fn && pn->leaf == rt) {
1050 atomic_dec(&rt->rt6i_ref);
1052 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
1053 pn->leaf = fib6_find_prefix(info->nl_net, pn);
1056 WARN_ON(pn->leaf == NULL);
1057 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
1060 atomic_inc(&pn->leaf->rt6i_ref);
1063 if (!(rt->dst.flags & DST_NOCACHE))
1068 #ifdef CONFIG_IPV6_SUBTREES
1069 /* Subtree creation failed, probably main tree node
1070 is orphan. If it is, shoot it.
1073 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1074 fib6_repair_tree(info->nl_net, fn);
1075 if (!(rt->dst.flags & DST_NOCACHE))
1082 * Routing tree lookup
1086 struct lookup_args {
1087 int offset; /* key offset on rt6_info */
1088 const struct in6_addr *addr; /* search key */
1091 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1092 struct lookup_args *args)
1094 struct fib6_node *fn;
1097 if (unlikely(args->offset == 0))
1107 struct fib6_node *next;
1109 dir = addr_bit_set(args->addr, fn->fn_bit);
1111 next = dir ? fn->right : fn->left;
1121 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1124 key = (struct rt6key *) ((u8 *) fn->leaf +
1127 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1128 #ifdef CONFIG_IPV6_SUBTREES
1130 struct fib6_node *sfn;
1131 sfn = fib6_lookup_1(fn->subtree,
1138 if (fn->fn_flags & RTN_RTINFO)
1142 #ifdef CONFIG_IPV6_SUBTREES
1145 if (fn->fn_flags & RTN_ROOT)
1154 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1155 const struct in6_addr *saddr)
1157 struct fib6_node *fn;
1158 struct lookup_args args[] = {
1160 .offset = offsetof(struct rt6_info, rt6i_dst),
1163 #ifdef CONFIG_IPV6_SUBTREES
1165 .offset = offsetof(struct rt6_info, rt6i_src),
1170 .offset = 0, /* sentinel */
1174 fn = fib6_lookup_1(root, daddr ? args : args + 1);
1175 if (!fn || fn->fn_flags & RTN_TL_ROOT)
1182 * Get node with specified destination prefix (and source prefix,
1183 * if subtrees are used)
1187 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1188 const struct in6_addr *addr,
1189 int plen, int offset)
1191 struct fib6_node *fn;
1193 for (fn = root; fn ; ) {
1194 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1199 if (plen < fn->fn_bit ||
1200 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1203 if (plen == fn->fn_bit)
1207 * We have more bits to go
1209 if (addr_bit_set(addr, fn->fn_bit))
1217 struct fib6_node *fib6_locate(struct fib6_node *root,
1218 const struct in6_addr *daddr, int dst_len,
1219 const struct in6_addr *saddr, int src_len)
1221 struct fib6_node *fn;
1223 fn = fib6_locate_1(root, daddr, dst_len,
1224 offsetof(struct rt6_info, rt6i_dst));
1226 #ifdef CONFIG_IPV6_SUBTREES
1228 WARN_ON(saddr == NULL);
1229 if (fn && fn->subtree)
1230 fn = fib6_locate_1(fn->subtree, saddr, src_len,
1231 offsetof(struct rt6_info, rt6i_src));
1235 if (fn && fn->fn_flags & RTN_RTINFO)
1247 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1249 if (fn->fn_flags & RTN_ROOT)
1250 return net->ipv6.ip6_null_entry;
1254 return fn->left->leaf;
1256 return fn->right->leaf;
1258 fn = FIB6_SUBTREE(fn);
1264 * Called to trim the tree of intermediate nodes when possible. "fn"
1265 * is the node we want to try and remove.
1268 static struct fib6_node *fib6_repair_tree(struct net *net,
1269 struct fib6_node *fn)
1273 struct fib6_node *child, *pn;
1274 struct fib6_walker *w;
1278 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1281 WARN_ON(fn->fn_flags & RTN_RTINFO);
1282 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1288 child = fn->right, children |= 1;
1290 child = fn->left, children |= 2;
1292 if (children == 3 || FIB6_SUBTREE(fn)
1293 #ifdef CONFIG_IPV6_SUBTREES
1294 /* Subtree root (i.e. fn) may have one child */
1295 || (children && fn->fn_flags & RTN_ROOT)
1298 fn->leaf = fib6_find_prefix(net, fn);
1302 fn->leaf = net->ipv6.ip6_null_entry;
1305 atomic_inc(&fn->leaf->rt6i_ref);
1310 #ifdef CONFIG_IPV6_SUBTREES
1311 if (FIB6_SUBTREE(pn) == fn) {
1312 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1313 FIB6_SUBTREE(pn) = NULL;
1316 WARN_ON(fn->fn_flags & RTN_ROOT);
1318 if (pn->right == fn)
1320 else if (pn->left == fn)
1329 #ifdef CONFIG_IPV6_SUBTREES
1333 read_lock(&fib6_walker_lock);
1336 if (w->root == fn) {
1337 w->root = w->node = NULL;
1338 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1339 } else if (w->node == fn) {
1340 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1345 if (w->root == fn) {
1347 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1349 if (w->node == fn) {
1352 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1353 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1355 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1356 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1361 read_unlock(&fib6_walker_lock);
1364 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1367 rt6_release(pn->leaf);
1373 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1374 struct nl_info *info)
1376 struct fib6_walker *w;
1377 struct rt6_info *rt = *rtp;
1378 struct net *net = info->nl_net;
1380 RT6_TRACE("fib6_del_route\n");
1383 *rtp = rt->dst.rt6_next;
1384 rt->rt6i_node = NULL;
1385 net->ipv6.rt6_stats->fib_rt_entries--;
1386 net->ipv6.rt6_stats->fib_discarded_routes++;
1388 /* Reset round-robin state, if necessary */
1389 if (fn->rr_ptr == rt)
1392 /* Remove this entry from other siblings */
1393 if (rt->rt6i_nsiblings) {
1394 struct rt6_info *sibling, *next_sibling;
1396 list_for_each_entry_safe(sibling, next_sibling,
1397 &rt->rt6i_siblings, rt6i_siblings)
1398 sibling->rt6i_nsiblings--;
1399 rt->rt6i_nsiblings = 0;
1400 list_del_init(&rt->rt6i_siblings);
1403 /* Adjust walkers */
1404 read_lock(&fib6_walker_lock);
1406 if (w->state == FWS_C && w->leaf == rt) {
1407 RT6_TRACE("walker %p adjusted by delroute\n", w);
1408 w->leaf = rt->dst.rt6_next;
1413 read_unlock(&fib6_walker_lock);
1415 rt->dst.rt6_next = NULL;
1417 /* If it was last route, expunge its radix tree node */
1419 fn->fn_flags &= ~RTN_RTINFO;
1420 net->ipv6.rt6_stats->fib_route_nodes--;
1421 fn = fib6_repair_tree(net, fn);
1424 fib6_purge_rt(rt, fn, net);
1426 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1430 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1432 struct net *net = info->nl_net;
1433 struct fib6_node *fn = rt->rt6i_node;
1434 struct rt6_info **rtp;
1437 if (rt->dst.obsolete > 0) {
1442 if (!fn || rt == net->ipv6.ip6_null_entry)
1445 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1447 if (!(rt->rt6i_flags & RTF_CACHE)) {
1448 struct fib6_node *pn = fn;
1449 #ifdef CONFIG_IPV6_SUBTREES
1450 /* clones of this route might be in another subtree */
1451 if (rt->rt6i_src.plen) {
1452 while (!(pn->fn_flags & RTN_ROOT))
1457 fib6_prune_clones(info->nl_net, pn);
1461 * Walk the leaf entries looking for ourself
1464 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1466 fib6_del_route(fn, rtp, info);
1474 * Tree traversal function.
1476 * Certainly, it is not interrupt safe.
1477 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1478 * It means, that we can modify tree during walking
1479 * and use this function for garbage collection, clone pruning,
1480 * cleaning tree when a device goes down etc. etc.
1482 * It guarantees that every node will be traversed,
1483 * and that it will be traversed only once.
1485 * Callback function w->func may return:
1486 * 0 -> continue walking.
1487 * positive value -> walking is suspended (used by tree dumps,
1488 * and probably by gc, if it will be split to several slices)
1489 * negative value -> terminate walking.
1491 * The function itself returns:
1492 * 0 -> walk is complete.
1493 * >0 -> walk is incomplete (i.e. suspended)
1494 * <0 -> walk is terminated by an error.
1497 static int fib6_walk_continue(struct fib6_walker *w)
1499 struct fib6_node *fn, *pn;
1506 if (w->prune && fn != w->root &&
1507 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1512 #ifdef CONFIG_IPV6_SUBTREES
1514 if (FIB6_SUBTREE(fn)) {
1515 w->node = FIB6_SUBTREE(fn);
1523 w->state = FWS_INIT;
1529 w->node = fn->right;
1530 w->state = FWS_INIT;
1536 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1558 #ifdef CONFIG_IPV6_SUBTREES
1559 if (FIB6_SUBTREE(pn) == fn) {
1560 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1565 if (pn->left == fn) {
1569 if (pn->right == fn) {
1571 w->leaf = w->node->leaf;
1581 static int fib6_walk(struct fib6_walker *w)
1585 w->state = FWS_INIT;
1588 fib6_walker_link(w);
1589 res = fib6_walk_continue(w);
1591 fib6_walker_unlink(w);
1595 static int fib6_clean_node(struct fib6_walker *w)
1598 struct rt6_info *rt;
1599 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1600 struct nl_info info = {
1604 if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1605 w->node->fn_sernum != c->sernum)
1606 w->node->fn_sernum = c->sernum;
1609 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1614 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1615 res = c->func(rt, c->arg);
1618 res = fib6_del(rt, &info);
1621 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1622 __func__, rt, rt->rt6i_node, res);
1635 * Convenient frontend to tree walker.
1637 * func is called on each route.
1638 * It may return -1 -> delete this route.
1639 * 0 -> continue walking
1641 * prune==1 -> only immediate children of node (certainly,
1642 * ignoring pure split nodes) will be scanned.
1645 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1646 int (*func)(struct rt6_info *, void *arg),
1647 bool prune, int sernum, void *arg)
1649 struct fib6_cleaner c;
1652 c.w.func = fib6_clean_node;
1664 static void __fib6_clean_all(struct net *net,
1665 int (*func)(struct rt6_info *, void *),
1666 int sernum, void *arg)
1668 struct fib6_table *table;
1669 struct hlist_head *head;
1673 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1674 head = &net->ipv6.fib_table_hash[h];
1675 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1676 write_lock_bh(&table->tb6_lock);
1677 fib6_clean_tree(net, &table->tb6_root,
1678 func, false, sernum, arg);
1679 write_unlock_bh(&table->tb6_lock);
1685 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1688 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1691 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1693 if (rt->rt6i_flags & RTF_CACHE) {
1694 RT6_TRACE("pruning clone %p\n", rt);
1701 static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1703 fib6_clean_tree(net, fn, fib6_prune_clone, true,
1704 FIB6_NO_SERNUM_CHANGE, NULL);
1707 static void fib6_flush_trees(struct net *net)
1709 int new_sernum = fib6_new_sernum(net);
1711 __fib6_clean_all(net, NULL, new_sernum, NULL);
1715 * Garbage collection
1718 static struct fib6_gc_args
1724 static int fib6_age(struct rt6_info *rt, void *arg)
1726 unsigned long now = jiffies;
1729 * check addrconf expiration here.
1730 * Routes are expired even if they are in use.
1732 * Also age clones. Note, that clones are aged out
1733 * only if they are not in use now.
1736 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1737 if (time_after(now, rt->dst.expires)) {
1738 RT6_TRACE("expiring %p\n", rt);
1742 } else if (rt->rt6i_flags & RTF_CACHE) {
1743 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1744 time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1745 RT6_TRACE("aging clone %p\n", rt);
1747 } else if (rt->rt6i_flags & RTF_GATEWAY) {
1748 struct neighbour *neigh;
1749 __u8 neigh_flags = 0;
1751 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1753 neigh_flags = neigh->flags;
1754 neigh_release(neigh);
1756 if (!(neigh_flags & NTF_ROUTER)) {
1757 RT6_TRACE("purging route %p via non-router but gateway\n",
1768 static DEFINE_SPINLOCK(fib6_gc_lock);
1770 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1775 spin_lock_bh(&fib6_gc_lock);
1776 } else if (!spin_trylock_bh(&fib6_gc_lock)) {
1777 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1780 gc_args.timeout = expires ? (int)expires :
1781 net->ipv6.sysctl.ip6_rt_gc_interval;
1783 gc_args.more = icmp6_dst_gc();
1785 fib6_clean_all(net, fib6_age, NULL);
1787 net->ipv6.ip6_rt_last_gc = now;
1790 mod_timer(&net->ipv6.ip6_fib_timer,
1792 + net->ipv6.sysctl.ip6_rt_gc_interval));
1794 del_timer(&net->ipv6.ip6_fib_timer);
1795 spin_unlock_bh(&fib6_gc_lock);
1798 static void fib6_gc_timer_cb(unsigned long arg)
1800 fib6_run_gc(0, (struct net *)arg, true);
1803 static int __net_init fib6_net_init(struct net *net)
1805 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1807 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1809 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1810 if (!net->ipv6.rt6_stats)
1813 /* Avoid false sharing : Use at least a full cache line */
1814 size = max_t(size_t, size, L1_CACHE_BYTES);
1816 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1817 if (!net->ipv6.fib_table_hash)
1820 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1822 if (!net->ipv6.fib6_main_tbl)
1823 goto out_fib_table_hash;
1825 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1826 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1827 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1828 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1829 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1831 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1832 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1834 if (!net->ipv6.fib6_local_tbl)
1835 goto out_fib6_main_tbl;
1836 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1837 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1838 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1839 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1840 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1842 fib6_tables_init(net);
1846 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1848 kfree(net->ipv6.fib6_main_tbl);
1851 kfree(net->ipv6.fib_table_hash);
1853 kfree(net->ipv6.rt6_stats);
1858 static void fib6_net_exit(struct net *net)
1860 rt6_ifdown(net, NULL);
1861 del_timer_sync(&net->ipv6.ip6_fib_timer);
1863 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1864 inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1865 kfree(net->ipv6.fib6_local_tbl);
1867 inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1868 kfree(net->ipv6.fib6_main_tbl);
1869 kfree(net->ipv6.fib_table_hash);
1870 kfree(net->ipv6.rt6_stats);
1873 static struct pernet_operations fib6_net_ops = {
1874 .init = fib6_net_init,
1875 .exit = fib6_net_exit,
1878 int __init fib6_init(void)
1882 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1883 sizeof(struct fib6_node),
1884 0, SLAB_HWCACHE_ALIGN,
1886 if (!fib6_node_kmem)
1889 ret = register_pernet_subsys(&fib6_net_ops);
1891 goto out_kmem_cache_create;
1893 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1896 goto out_unregister_subsys;
1898 __fib6_flush_trees = fib6_flush_trees;
1902 out_unregister_subsys:
1903 unregister_pernet_subsys(&fib6_net_ops);
1904 out_kmem_cache_create:
1905 kmem_cache_destroy(fib6_node_kmem);
1909 void fib6_gc_cleanup(void)
1911 unregister_pernet_subsys(&fib6_net_ops);
1912 kmem_cache_destroy(fib6_node_kmem);
1915 #ifdef CONFIG_PROC_FS
1917 struct ipv6_route_iter {
1918 struct seq_net_private p;
1919 struct fib6_walker w;
1921 struct fib6_table *tbl;
1925 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
1927 struct rt6_info *rt = v;
1928 struct ipv6_route_iter *iter = seq->private;
1930 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
1932 #ifdef CONFIG_IPV6_SUBTREES
1933 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
1935 seq_puts(seq, "00000000000000000000000000000000 00 ");
1937 if (rt->rt6i_flags & RTF_GATEWAY)
1938 seq_printf(seq, "%pi6", &rt->rt6i_gateway);
1940 seq_puts(seq, "00000000000000000000000000000000");
1942 seq_printf(seq, " %08x %08x %08x %08x %8s\n",
1943 rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
1944 rt->dst.__use, rt->rt6i_flags,
1945 rt->dst.dev ? rt->dst.dev->name : "");
1946 iter->w.leaf = NULL;
1950 static int ipv6_route_yield(struct fib6_walker *w)
1952 struct ipv6_route_iter *iter = w->args;
1958 iter->w.leaf = iter->w.leaf->dst.rt6_next;
1960 if (!iter->skip && iter->w.leaf)
1962 } while (iter->w.leaf);
1967 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter)
1969 memset(&iter->w, 0, sizeof(iter->w));
1970 iter->w.func = ipv6_route_yield;
1971 iter->w.root = &iter->tbl->tb6_root;
1972 iter->w.state = FWS_INIT;
1973 iter->w.node = iter->w.root;
1974 iter->w.args = iter;
1975 iter->sernum = iter->w.root->fn_sernum;
1976 INIT_LIST_HEAD(&iter->w.lh);
1977 fib6_walker_link(&iter->w);
1980 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
1984 struct hlist_node *node;
1987 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
1988 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
1994 while (!node && h < FIB6_TABLE_HASHSZ) {
1995 node = rcu_dereference_bh(
1996 hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
1998 return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2001 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2003 if (iter->sernum != iter->w.root->fn_sernum) {
2004 iter->sernum = iter->w.root->fn_sernum;
2005 iter->w.state = FWS_INIT;
2006 iter->w.node = iter->w.root;
2007 WARN_ON(iter->w.skip);
2008 iter->w.skip = iter->w.count;
2012 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2016 struct net *net = seq_file_net(seq);
2017 struct ipv6_route_iter *iter = seq->private;
2022 n = ((struct rt6_info *)v)->dst.rt6_next;
2029 ipv6_route_check_sernum(iter);
2030 read_lock(&iter->tbl->tb6_lock);
2031 r = fib6_walk_continue(&iter->w);
2032 read_unlock(&iter->tbl->tb6_lock);
2036 return iter->w.leaf;
2038 fib6_walker_unlink(&iter->w);
2041 fib6_walker_unlink(&iter->w);
2043 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2047 ipv6_route_seq_setup_walk(iter);
2051 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2054 struct net *net = seq_file_net(seq);
2055 struct ipv6_route_iter *iter = seq->private;
2058 iter->tbl = ipv6_route_seq_next_table(NULL, net);
2062 ipv6_route_seq_setup_walk(iter);
2063 return ipv6_route_seq_next(seq, NULL, pos);
2069 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2071 struct fib6_walker *w = &iter->w;
2072 return w->node && !(w->state == FWS_U && w->node == w->root);
2075 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2078 struct ipv6_route_iter *iter = seq->private;
2080 if (ipv6_route_iter_active(iter))
2081 fib6_walker_unlink(&iter->w);
2083 rcu_read_unlock_bh();
2086 static const struct seq_operations ipv6_route_seq_ops = {
2087 .start = ipv6_route_seq_start,
2088 .next = ipv6_route_seq_next,
2089 .stop = ipv6_route_seq_stop,
2090 .show = ipv6_route_seq_show
2093 int ipv6_route_open(struct inode *inode, struct file *file)
2095 return seq_open_net(inode, file, &ipv6_route_seq_ops,
2096 sizeof(struct ipv6_route_iter));
2099 #endif /* CONFIG_PROC_FS */