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) {
268 static void __net_init fib6_tables_init(struct net *net)
270 fib6_link_table(net, net->ipv6.fib6_main_tbl);
271 fib6_link_table(net, net->ipv6.fib6_local_tbl);
275 struct fib6_table *fib6_new_table(struct net *net, u32 id)
277 return fib6_get_table(net, id);
280 struct fib6_table *fib6_get_table(struct net *net, u32 id)
282 return net->ipv6.fib6_main_tbl;
285 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
286 int flags, pol_lookup_t lookup)
288 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
291 static void __net_init fib6_tables_init(struct net *net)
293 fib6_link_table(net, net->ipv6.fib6_main_tbl);
298 static int fib6_dump_node(struct fib6_walker *w)
303 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
304 res = rt6_dump_route(rt, w->args);
306 /* Frame is full, suspend walking */
315 static void fib6_dump_end(struct netlink_callback *cb)
317 struct fib6_walker *w = (void *)cb->args[2];
322 fib6_walker_unlink(w);
327 cb->done = (void *)cb->args[3];
331 static int fib6_dump_done(struct netlink_callback *cb)
334 return cb->done ? cb->done(cb) : 0;
337 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
338 struct netlink_callback *cb)
340 struct fib6_walker *w;
343 w = (void *)cb->args[2];
344 w->root = &table->tb6_root;
346 if (cb->args[4] == 0) {
350 read_lock_bh(&table->tb6_lock);
352 read_unlock_bh(&table->tb6_lock);
355 cb->args[5] = w->root->fn_sernum;
358 if (cb->args[5] != w->root->fn_sernum) {
359 /* Begin at the root if the tree changed */
360 cb->args[5] = w->root->fn_sernum;
367 read_lock_bh(&table->tb6_lock);
368 res = fib6_walk_continue(w);
369 read_unlock_bh(&table->tb6_lock);
371 fib6_walker_unlink(w);
379 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
381 struct net *net = sock_net(skb->sk);
383 unsigned int e = 0, s_e;
384 struct rt6_rtnl_dump_arg arg;
385 struct fib6_walker *w;
386 struct fib6_table *tb;
387 struct hlist_head *head;
393 w = (void *)cb->args[2];
397 * 1. hook callback destructor.
399 cb->args[3] = (long)cb->done;
400 cb->done = fib6_dump_done;
403 * 2. allocate and initialize walker.
405 w = kzalloc(sizeof(*w), GFP_ATOMIC);
408 w->func = fib6_dump_node;
409 cb->args[2] = (long)w;
418 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
420 head = &net->ipv6.fib_table_hash[h];
421 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
424 res = fib6_dump_table(tb, skb, cb);
436 res = res < 0 ? res : skb->len;
445 * return the appropriate node for a routing tree "add" operation
446 * by either creating and inserting or by returning an existing
450 static struct fib6_node *fib6_add_1(struct fib6_node *root,
451 struct in6_addr *addr, int plen,
452 int offset, int allow_create,
453 int replace_required, int sernum)
455 struct fib6_node *fn, *in, *ln;
456 struct fib6_node *pn = NULL;
461 RT6_TRACE("fib6_add_1\n");
463 /* insert node in tree */
468 key = (struct rt6key *)((u8 *)fn->leaf + offset);
473 if (plen < fn->fn_bit ||
474 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
476 if (replace_required) {
477 pr_warn("Can't replace route, no match found\n");
478 return ERR_PTR(-ENOENT);
480 pr_warn("NLM_F_CREATE should be set when creating new route\n");
489 if (plen == fn->fn_bit) {
490 /* clean up an intermediate node */
491 if (!(fn->fn_flags & RTN_RTINFO)) {
492 rt6_release(fn->leaf);
496 fn->fn_sernum = sernum;
502 * We have more bits to go
505 /* Try to walk down on tree. */
506 fn->fn_sernum = sernum;
507 dir = addr_bit_set(addr, fn->fn_bit);
509 fn = dir ? fn->right : fn->left;
513 /* We should not create new node because
514 * NLM_F_REPLACE was specified without NLM_F_CREATE
515 * I assume it is safe to require NLM_F_CREATE when
516 * REPLACE flag is used! Later we may want to remove the
517 * check for replace_required, because according
518 * to netlink specification, NLM_F_CREATE
519 * MUST be specified if new route is created.
520 * That would keep IPv6 consistent with IPv4
522 if (replace_required) {
523 pr_warn("Can't replace route, no match found\n");
524 return ERR_PTR(-ENOENT);
526 pr_warn("NLM_F_CREATE should be set when creating new route\n");
529 * We walked to the bottom of tree.
530 * Create new leaf node without children.
536 return ERR_PTR(-ENOMEM);
540 ln->fn_sernum = sernum;
552 * split since we don't have a common prefix anymore or
553 * we have a less significant route.
554 * we've to insert an intermediate node on the list
555 * this new node will point to the one we need to create
561 /* find 1st bit in difference between the 2 addrs.
563 See comment in __ipv6_addr_diff: bit may be an invalid value,
564 but if it is >= plen, the value is ignored in any case.
567 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
572 * (new leaf node)[ln] (old node)[fn]
583 return ERR_PTR(-ENOMEM);
587 * new intermediate node.
589 * be off since that an address that chooses one of
590 * the branches would not match less specific routes
591 * in the other branch
598 atomic_inc(&in->leaf->rt6i_ref);
600 in->fn_sernum = sernum;
602 /* update parent pointer */
613 ln->fn_sernum = sernum;
615 if (addr_bit_set(addr, bit)) {
622 } else { /* plen <= bit */
625 * (new leaf node)[ln]
627 * (old node)[fn] NULL
633 return ERR_PTR(-ENOMEM);
639 ln->fn_sernum = sernum;
646 if (addr_bit_set(&key->addr, plen))
656 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
658 return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
662 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
666 for (i = 0; i < RTAX_MAX; i++) {
667 if (test_bit(i, mxc->mx_valid))
672 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
677 if (dst->flags & DST_HOST) {
678 u32 *mp = dst_metrics_write_ptr(dst);
683 fib6_copy_metrics(mp, mxc);
685 dst_init_metrics(dst, mxc->mx, false);
687 /* We've stolen mx now. */
694 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
697 if (atomic_read(&rt->rt6i_ref) != 1) {
698 /* This route is used as dummy address holder in some split
699 * nodes. It is not leaked, but it still holds other resources,
700 * which must be released in time. So, scan ascendant nodes
701 * and replace dummy references to this route with references
702 * to still alive ones.
705 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
706 fn->leaf = fib6_find_prefix(net, fn);
707 atomic_inc(&fn->leaf->rt6i_ref);
712 /* No more references are possible at this point. */
713 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
718 * Insert routing information in a node.
721 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
722 struct nl_info *info, struct mx6_config *mxc)
724 struct rt6_info *iter = NULL;
725 struct rt6_info **ins;
726 struct rt6_info **fallback_ins = NULL;
727 int replace = (info->nlh &&
728 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
729 int add = (!info->nlh ||
730 (info->nlh->nlmsg_flags & NLM_F_CREATE));
732 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
737 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
739 * Search for duplicates
742 if (iter->rt6i_metric == rt->rt6i_metric) {
744 * Same priority level
747 (info->nlh->nlmsg_flags & NLM_F_EXCL))
750 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
755 fallback_ins = fallback_ins ?: ins;
759 if (iter->dst.dev == rt->dst.dev &&
760 iter->rt6i_idev == rt->rt6i_idev &&
761 ipv6_addr_equal(&iter->rt6i_gateway,
762 &rt->rt6i_gateway)) {
763 if (rt->rt6i_nsiblings)
764 rt->rt6i_nsiblings = 0;
765 if (!(iter->rt6i_flags & RTF_EXPIRES))
767 if (!(rt->rt6i_flags & RTF_EXPIRES))
768 rt6_clean_expires(iter);
770 rt6_set_expires(iter, rt->dst.expires);
771 iter->rt6i_pmtu = rt->rt6i_pmtu;
774 /* If we have the same destination and the same metric,
775 * but not the same gateway, then the route we try to
776 * add is sibling to this route, increment our counter
777 * of siblings, and later we will add our route to the
779 * Only static routes (which don't have flag
780 * RTF_EXPIRES) are used for ECMPv6.
782 * To avoid long list, we only had siblings if the
783 * route have a gateway.
786 rt6_qualify_for_ecmp(iter))
787 rt->rt6i_nsiblings++;
790 if (iter->rt6i_metric > rt->rt6i_metric)
794 ins = &iter->dst.rt6_next;
797 if (fallback_ins && !found) {
798 /* No ECMP-able route found, replace first non-ECMP one */
804 /* Reset round-robin state, if necessary */
805 if (ins == &fn->leaf)
808 /* Link this route to others same route. */
809 if (rt->rt6i_nsiblings) {
810 unsigned int rt6i_nsiblings;
811 struct rt6_info *sibling, *temp_sibling;
813 /* Find the first route that have the same metric */
816 if (sibling->rt6i_metric == rt->rt6i_metric &&
817 rt6_qualify_for_ecmp(sibling)) {
818 list_add_tail(&rt->rt6i_siblings,
819 &sibling->rt6i_siblings);
822 sibling = sibling->dst.rt6_next;
824 /* For each sibling in the list, increment the counter of
825 * siblings. BUG() if counters does not match, list of siblings
829 list_for_each_entry_safe(sibling, temp_sibling,
830 &rt->rt6i_siblings, rt6i_siblings) {
831 sibling->rt6i_nsiblings++;
832 BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
835 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
843 pr_warn("NLM_F_CREATE should be set when creating new route\n");
846 err = fib6_commit_metrics(&rt->dst, mxc);
850 rt->dst.rt6_next = iter;
853 atomic_inc(&rt->rt6i_ref);
854 inet6_rt_notify(RTM_NEWROUTE, rt, info, 0);
855 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
857 if (!(fn->fn_flags & RTN_RTINFO)) {
858 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
859 fn->fn_flags |= RTN_RTINFO;
868 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
872 err = fib6_commit_metrics(&rt->dst, mxc);
878 rt->dst.rt6_next = iter->dst.rt6_next;
879 atomic_inc(&rt->rt6i_ref);
880 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
881 if (!(fn->fn_flags & RTN_RTINFO)) {
882 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
883 fn->fn_flags |= RTN_RTINFO;
885 nsiblings = iter->rt6i_nsiblings;
886 fib6_purge_rt(iter, fn, info->nl_net);
890 /* Replacing an ECMP route, remove all siblings */
891 ins = &rt->dst.rt6_next;
894 if (rt6_qualify_for_ecmp(iter)) {
895 *ins = iter->dst.rt6_next;
896 fib6_purge_rt(iter, fn, info->nl_net);
900 ins = &iter->dst.rt6_next;
904 WARN_ON(nsiblings != 0);
911 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
913 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
914 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
915 mod_timer(&net->ipv6.ip6_fib_timer,
916 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
919 void fib6_force_start_gc(struct net *net)
921 if (!timer_pending(&net->ipv6.ip6_fib_timer))
922 mod_timer(&net->ipv6.ip6_fib_timer,
923 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
927 * Add routing information to the routing tree.
928 * <destination addr>/<source addr>
929 * with source addr info in sub-trees
932 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
933 struct nl_info *info, struct mx6_config *mxc)
935 struct fib6_node *fn, *pn = NULL;
937 int allow_create = 1;
938 int replace_required = 0;
939 int sernum = fib6_new_sernum(info->nl_net);
941 if (WARN_ON_ONCE((rt->dst.flags & DST_NOCACHE) &&
942 !atomic_read(&rt->dst.__refcnt)))
946 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
948 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
949 replace_required = 1;
951 if (!allow_create && !replace_required)
952 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
954 fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
955 offsetof(struct rt6_info, rt6i_dst), allow_create,
956 replace_required, sernum);
965 #ifdef CONFIG_IPV6_SUBTREES
966 if (rt->rt6i_src.plen) {
967 struct fib6_node *sn;
970 struct fib6_node *sfn;
982 /* Create subtree root node */
987 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
988 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
989 sfn->fn_flags = RTN_ROOT;
990 sfn->fn_sernum = sernum;
992 /* Now add the first leaf node to new subtree */
994 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
996 offsetof(struct rt6_info, rt6i_src),
997 allow_create, replace_required, sernum);
1000 /* If it is failed, discard just allocated
1001 root, and then (in st_failure) stale node
1009 /* Now link new subtree to main tree */
1013 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
1015 offsetof(struct rt6_info, rt6i_src),
1016 allow_create, replace_required, sernum);
1026 atomic_inc(&rt->rt6i_ref);
1032 err = fib6_add_rt2node(fn, rt, info, mxc);
1034 fib6_start_gc(info->nl_net, rt);
1035 if (!(rt->rt6i_flags & RTF_CACHE))
1036 fib6_prune_clones(info->nl_net, pn);
1037 rt->dst.flags &= ~DST_NOCACHE;
1042 #ifdef CONFIG_IPV6_SUBTREES
1044 * If fib6_add_1 has cleared the old leaf pointer in the
1045 * super-tree leaf node we have to find a new one for it.
1047 if (pn != fn && pn->leaf == rt) {
1049 atomic_dec(&rt->rt6i_ref);
1051 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
1052 pn->leaf = fib6_find_prefix(info->nl_net, pn);
1055 WARN_ON(pn->leaf == NULL);
1056 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
1059 atomic_inc(&pn->leaf->rt6i_ref);
1062 if (!(rt->dst.flags & DST_NOCACHE))
1067 #ifdef CONFIG_IPV6_SUBTREES
1068 /* Subtree creation failed, probably main tree node
1069 is orphan. If it is, shoot it.
1072 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1073 fib6_repair_tree(info->nl_net, fn);
1074 if (!(rt->dst.flags & DST_NOCACHE))
1081 * Routing tree lookup
1085 struct lookup_args {
1086 int offset; /* key offset on rt6_info */
1087 const struct in6_addr *addr; /* search key */
1090 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1091 struct lookup_args *args)
1093 struct fib6_node *fn;
1096 if (unlikely(args->offset == 0))
1106 struct fib6_node *next;
1108 dir = addr_bit_set(args->addr, fn->fn_bit);
1110 next = dir ? fn->right : fn->left;
1120 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1123 key = (struct rt6key *) ((u8 *) fn->leaf +
1126 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1127 #ifdef CONFIG_IPV6_SUBTREES
1129 struct fib6_node *sfn;
1130 sfn = fib6_lookup_1(fn->subtree,
1137 if (fn->fn_flags & RTN_RTINFO)
1141 #ifdef CONFIG_IPV6_SUBTREES
1144 if (fn->fn_flags & RTN_ROOT)
1153 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1154 const struct in6_addr *saddr)
1156 struct fib6_node *fn;
1157 struct lookup_args args[] = {
1159 .offset = offsetof(struct rt6_info, rt6i_dst),
1162 #ifdef CONFIG_IPV6_SUBTREES
1164 .offset = offsetof(struct rt6_info, rt6i_src),
1169 .offset = 0, /* sentinel */
1173 fn = fib6_lookup_1(root, daddr ? args : args + 1);
1174 if (!fn || fn->fn_flags & RTN_TL_ROOT)
1181 * Get node with specified destination prefix (and source prefix,
1182 * if subtrees are used)
1186 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1187 const struct in6_addr *addr,
1188 int plen, int offset)
1190 struct fib6_node *fn;
1192 for (fn = root; fn ; ) {
1193 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1198 if (plen < fn->fn_bit ||
1199 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1202 if (plen == fn->fn_bit)
1206 * We have more bits to go
1208 if (addr_bit_set(addr, fn->fn_bit))
1216 struct fib6_node *fib6_locate(struct fib6_node *root,
1217 const struct in6_addr *daddr, int dst_len,
1218 const struct in6_addr *saddr, int src_len)
1220 struct fib6_node *fn;
1222 fn = fib6_locate_1(root, daddr, dst_len,
1223 offsetof(struct rt6_info, rt6i_dst));
1225 #ifdef CONFIG_IPV6_SUBTREES
1227 WARN_ON(saddr == NULL);
1228 if (fn && fn->subtree)
1229 fn = fib6_locate_1(fn->subtree, saddr, src_len,
1230 offsetof(struct rt6_info, rt6i_src));
1234 if (fn && fn->fn_flags & RTN_RTINFO)
1246 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1248 if (fn->fn_flags & RTN_ROOT)
1249 return net->ipv6.ip6_null_entry;
1253 return fn->left->leaf;
1255 return fn->right->leaf;
1257 fn = FIB6_SUBTREE(fn);
1263 * Called to trim the tree of intermediate nodes when possible. "fn"
1264 * is the node we want to try and remove.
1267 static struct fib6_node *fib6_repair_tree(struct net *net,
1268 struct fib6_node *fn)
1272 struct fib6_node *child, *pn;
1273 struct fib6_walker *w;
1277 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1280 WARN_ON(fn->fn_flags & RTN_RTINFO);
1281 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1287 child = fn->right, children |= 1;
1289 child = fn->left, children |= 2;
1291 if (children == 3 || FIB6_SUBTREE(fn)
1292 #ifdef CONFIG_IPV6_SUBTREES
1293 /* Subtree root (i.e. fn) may have one child */
1294 || (children && fn->fn_flags & RTN_ROOT)
1297 fn->leaf = fib6_find_prefix(net, fn);
1301 fn->leaf = net->ipv6.ip6_null_entry;
1304 atomic_inc(&fn->leaf->rt6i_ref);
1309 #ifdef CONFIG_IPV6_SUBTREES
1310 if (FIB6_SUBTREE(pn) == fn) {
1311 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1312 FIB6_SUBTREE(pn) = NULL;
1315 WARN_ON(fn->fn_flags & RTN_ROOT);
1317 if (pn->right == fn)
1319 else if (pn->left == fn)
1328 #ifdef CONFIG_IPV6_SUBTREES
1332 read_lock(&fib6_walker_lock);
1335 if (w->root == fn) {
1336 w->root = w->node = NULL;
1337 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1338 } else if (w->node == fn) {
1339 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1344 if (w->root == fn) {
1346 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1348 if (w->node == fn) {
1351 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1352 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1354 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1355 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1360 read_unlock(&fib6_walker_lock);
1363 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1366 rt6_release(pn->leaf);
1372 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1373 struct nl_info *info)
1375 struct fib6_walker *w;
1376 struct rt6_info *rt = *rtp;
1377 struct net *net = info->nl_net;
1379 RT6_TRACE("fib6_del_route\n");
1382 *rtp = rt->dst.rt6_next;
1383 rt->rt6i_node = NULL;
1384 net->ipv6.rt6_stats->fib_rt_entries--;
1385 net->ipv6.rt6_stats->fib_discarded_routes++;
1387 /* Reset round-robin state, if necessary */
1388 if (fn->rr_ptr == rt)
1391 /* Remove this entry from other siblings */
1392 if (rt->rt6i_nsiblings) {
1393 struct rt6_info *sibling, *next_sibling;
1395 list_for_each_entry_safe(sibling, next_sibling,
1396 &rt->rt6i_siblings, rt6i_siblings)
1397 sibling->rt6i_nsiblings--;
1398 rt->rt6i_nsiblings = 0;
1399 list_del_init(&rt->rt6i_siblings);
1402 /* Adjust walkers */
1403 read_lock(&fib6_walker_lock);
1405 if (w->state == FWS_C && w->leaf == rt) {
1406 RT6_TRACE("walker %p adjusted by delroute\n", w);
1407 w->leaf = rt->dst.rt6_next;
1412 read_unlock(&fib6_walker_lock);
1414 rt->dst.rt6_next = NULL;
1416 /* If it was last route, expunge its radix tree node */
1418 fn->fn_flags &= ~RTN_RTINFO;
1419 net->ipv6.rt6_stats->fib_route_nodes--;
1420 fn = fib6_repair_tree(net, fn);
1423 fib6_purge_rt(rt, fn, net);
1425 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1429 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1431 struct net *net = info->nl_net;
1432 struct fib6_node *fn = rt->rt6i_node;
1433 struct rt6_info **rtp;
1436 if (rt->dst.obsolete > 0) {
1441 if (!fn || rt == net->ipv6.ip6_null_entry)
1444 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1446 if (!(rt->rt6i_flags & RTF_CACHE)) {
1447 struct fib6_node *pn = fn;
1448 #ifdef CONFIG_IPV6_SUBTREES
1449 /* clones of this route might be in another subtree */
1450 if (rt->rt6i_src.plen) {
1451 while (!(pn->fn_flags & RTN_ROOT))
1456 fib6_prune_clones(info->nl_net, pn);
1460 * Walk the leaf entries looking for ourself
1463 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1465 fib6_del_route(fn, rtp, info);
1473 * Tree traversal function.
1475 * Certainly, it is not interrupt safe.
1476 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1477 * It means, that we can modify tree during walking
1478 * and use this function for garbage collection, clone pruning,
1479 * cleaning tree when a device goes down etc. etc.
1481 * It guarantees that every node will be traversed,
1482 * and that it will be traversed only once.
1484 * Callback function w->func may return:
1485 * 0 -> continue walking.
1486 * positive value -> walking is suspended (used by tree dumps,
1487 * and probably by gc, if it will be split to several slices)
1488 * negative value -> terminate walking.
1490 * The function itself returns:
1491 * 0 -> walk is complete.
1492 * >0 -> walk is incomplete (i.e. suspended)
1493 * <0 -> walk is terminated by an error.
1496 static int fib6_walk_continue(struct fib6_walker *w)
1498 struct fib6_node *fn, *pn;
1505 if (w->prune && fn != w->root &&
1506 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1511 #ifdef CONFIG_IPV6_SUBTREES
1513 if (FIB6_SUBTREE(fn)) {
1514 w->node = FIB6_SUBTREE(fn);
1522 w->state = FWS_INIT;
1528 w->node = fn->right;
1529 w->state = FWS_INIT;
1535 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1557 #ifdef CONFIG_IPV6_SUBTREES
1558 if (FIB6_SUBTREE(pn) == fn) {
1559 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1564 if (pn->left == fn) {
1568 if (pn->right == fn) {
1570 w->leaf = w->node->leaf;
1580 static int fib6_walk(struct fib6_walker *w)
1584 w->state = FWS_INIT;
1587 fib6_walker_link(w);
1588 res = fib6_walk_continue(w);
1590 fib6_walker_unlink(w);
1594 static int fib6_clean_node(struct fib6_walker *w)
1597 struct rt6_info *rt;
1598 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1599 struct nl_info info = {
1603 if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1604 w->node->fn_sernum != c->sernum)
1605 w->node->fn_sernum = c->sernum;
1608 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1613 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1614 res = c->func(rt, c->arg);
1617 res = fib6_del(rt, &info);
1620 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1621 __func__, rt, rt->rt6i_node, res);
1634 * Convenient frontend to tree walker.
1636 * func is called on each route.
1637 * It may return -1 -> delete this route.
1638 * 0 -> continue walking
1640 * prune==1 -> only immediate children of node (certainly,
1641 * ignoring pure split nodes) will be scanned.
1644 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1645 int (*func)(struct rt6_info *, void *arg),
1646 bool prune, int sernum, void *arg)
1648 struct fib6_cleaner c;
1651 c.w.func = fib6_clean_node;
1663 static void __fib6_clean_all(struct net *net,
1664 int (*func)(struct rt6_info *, void *),
1665 int sernum, void *arg)
1667 struct fib6_table *table;
1668 struct hlist_head *head;
1672 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1673 head = &net->ipv6.fib_table_hash[h];
1674 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1675 write_lock_bh(&table->tb6_lock);
1676 fib6_clean_tree(net, &table->tb6_root,
1677 func, false, sernum, arg);
1678 write_unlock_bh(&table->tb6_lock);
1684 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1687 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1690 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1692 if (rt->rt6i_flags & RTF_CACHE) {
1693 RT6_TRACE("pruning clone %p\n", rt);
1700 static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1702 fib6_clean_tree(net, fn, fib6_prune_clone, true,
1703 FIB6_NO_SERNUM_CHANGE, NULL);
1706 static void fib6_flush_trees(struct net *net)
1708 int new_sernum = fib6_new_sernum(net);
1710 __fib6_clean_all(net, NULL, new_sernum, NULL);
1714 * Garbage collection
1717 static struct fib6_gc_args
1723 static int fib6_age(struct rt6_info *rt, void *arg)
1725 unsigned long now = jiffies;
1728 * check addrconf expiration here.
1729 * Routes are expired even if they are in use.
1731 * Also age clones. Note, that clones are aged out
1732 * only if they are not in use now.
1735 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1736 if (time_after(now, rt->dst.expires)) {
1737 RT6_TRACE("expiring %p\n", rt);
1741 } else if (rt->rt6i_flags & RTF_CACHE) {
1742 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1743 time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1744 RT6_TRACE("aging clone %p\n", rt);
1746 } else if (rt->rt6i_flags & RTF_GATEWAY) {
1747 struct neighbour *neigh;
1748 __u8 neigh_flags = 0;
1750 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1752 neigh_flags = neigh->flags;
1753 neigh_release(neigh);
1755 if (!(neigh_flags & NTF_ROUTER)) {
1756 RT6_TRACE("purging route %p via non-router but gateway\n",
1767 static DEFINE_SPINLOCK(fib6_gc_lock);
1769 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1774 spin_lock_bh(&fib6_gc_lock);
1775 } else if (!spin_trylock_bh(&fib6_gc_lock)) {
1776 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1779 gc_args.timeout = expires ? (int)expires :
1780 net->ipv6.sysctl.ip6_rt_gc_interval;
1782 gc_args.more = icmp6_dst_gc();
1784 fib6_clean_all(net, fib6_age, NULL);
1786 net->ipv6.ip6_rt_last_gc = now;
1789 mod_timer(&net->ipv6.ip6_fib_timer,
1791 + net->ipv6.sysctl.ip6_rt_gc_interval));
1793 del_timer(&net->ipv6.ip6_fib_timer);
1794 spin_unlock_bh(&fib6_gc_lock);
1797 static void fib6_gc_timer_cb(unsigned long arg)
1799 fib6_run_gc(0, (struct net *)arg, true);
1802 static int __net_init fib6_net_init(struct net *net)
1804 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1806 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1808 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1809 if (!net->ipv6.rt6_stats)
1812 /* Avoid false sharing : Use at least a full cache line */
1813 size = max_t(size_t, size, L1_CACHE_BYTES);
1815 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1816 if (!net->ipv6.fib_table_hash)
1819 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1821 if (!net->ipv6.fib6_main_tbl)
1822 goto out_fib_table_hash;
1824 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1825 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1826 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1827 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1828 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1830 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1831 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1833 if (!net->ipv6.fib6_local_tbl)
1834 goto out_fib6_main_tbl;
1835 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1836 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1837 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1838 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1839 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1841 fib6_tables_init(net);
1845 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1847 kfree(net->ipv6.fib6_main_tbl);
1850 kfree(net->ipv6.fib_table_hash);
1852 kfree(net->ipv6.rt6_stats);
1857 static void fib6_net_exit(struct net *net)
1859 rt6_ifdown(net, NULL);
1860 del_timer_sync(&net->ipv6.ip6_fib_timer);
1862 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1863 inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1864 kfree(net->ipv6.fib6_local_tbl);
1866 inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1867 kfree(net->ipv6.fib6_main_tbl);
1868 kfree(net->ipv6.fib_table_hash);
1869 kfree(net->ipv6.rt6_stats);
1872 static struct pernet_operations fib6_net_ops = {
1873 .init = fib6_net_init,
1874 .exit = fib6_net_exit,
1877 int __init fib6_init(void)
1881 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1882 sizeof(struct fib6_node),
1883 0, SLAB_HWCACHE_ALIGN,
1885 if (!fib6_node_kmem)
1888 ret = register_pernet_subsys(&fib6_net_ops);
1890 goto out_kmem_cache_create;
1892 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1895 goto out_unregister_subsys;
1897 __fib6_flush_trees = fib6_flush_trees;
1901 out_unregister_subsys:
1902 unregister_pernet_subsys(&fib6_net_ops);
1903 out_kmem_cache_create:
1904 kmem_cache_destroy(fib6_node_kmem);
1908 void fib6_gc_cleanup(void)
1910 unregister_pernet_subsys(&fib6_net_ops);
1911 kmem_cache_destroy(fib6_node_kmem);
1914 #ifdef CONFIG_PROC_FS
1916 struct ipv6_route_iter {
1917 struct seq_net_private p;
1918 struct fib6_walker w;
1920 struct fib6_table *tbl;
1924 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
1926 struct rt6_info *rt = v;
1927 struct ipv6_route_iter *iter = seq->private;
1929 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
1931 #ifdef CONFIG_IPV6_SUBTREES
1932 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
1934 seq_puts(seq, "00000000000000000000000000000000 00 ");
1936 if (rt->rt6i_flags & RTF_GATEWAY)
1937 seq_printf(seq, "%pi6", &rt->rt6i_gateway);
1939 seq_puts(seq, "00000000000000000000000000000000");
1941 seq_printf(seq, " %08x %08x %08x %08x %8s\n",
1942 rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
1943 rt->dst.__use, rt->rt6i_flags,
1944 rt->dst.dev ? rt->dst.dev->name : "");
1945 iter->w.leaf = NULL;
1949 static int ipv6_route_yield(struct fib6_walker *w)
1951 struct ipv6_route_iter *iter = w->args;
1957 iter->w.leaf = iter->w.leaf->dst.rt6_next;
1959 if (!iter->skip && iter->w.leaf)
1961 } while (iter->w.leaf);
1966 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter)
1968 memset(&iter->w, 0, sizeof(iter->w));
1969 iter->w.func = ipv6_route_yield;
1970 iter->w.root = &iter->tbl->tb6_root;
1971 iter->w.state = FWS_INIT;
1972 iter->w.node = iter->w.root;
1973 iter->w.args = iter;
1974 iter->sernum = iter->w.root->fn_sernum;
1975 INIT_LIST_HEAD(&iter->w.lh);
1976 fib6_walker_link(&iter->w);
1979 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
1983 struct hlist_node *node;
1986 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
1987 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
1993 while (!node && h < FIB6_TABLE_HASHSZ) {
1994 node = rcu_dereference_bh(
1995 hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
1997 return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2000 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2002 if (iter->sernum != iter->w.root->fn_sernum) {
2003 iter->sernum = iter->w.root->fn_sernum;
2004 iter->w.state = FWS_INIT;
2005 iter->w.node = iter->w.root;
2006 WARN_ON(iter->w.skip);
2007 iter->w.skip = iter->w.count;
2011 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2015 struct net *net = seq_file_net(seq);
2016 struct ipv6_route_iter *iter = seq->private;
2021 n = ((struct rt6_info *)v)->dst.rt6_next;
2028 ipv6_route_check_sernum(iter);
2029 read_lock(&iter->tbl->tb6_lock);
2030 r = fib6_walk_continue(&iter->w);
2031 read_unlock(&iter->tbl->tb6_lock);
2035 return iter->w.leaf;
2037 fib6_walker_unlink(&iter->w);
2040 fib6_walker_unlink(&iter->w);
2042 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2046 ipv6_route_seq_setup_walk(iter);
2050 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2053 struct net *net = seq_file_net(seq);
2054 struct ipv6_route_iter *iter = seq->private;
2057 iter->tbl = ipv6_route_seq_next_table(NULL, net);
2061 ipv6_route_seq_setup_walk(iter);
2062 return ipv6_route_seq_next(seq, NULL, pos);
2068 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2070 struct fib6_walker *w = &iter->w;
2071 return w->node && !(w->state == FWS_U && w->node == w->root);
2074 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2077 struct ipv6_route_iter *iter = seq->private;
2079 if (ipv6_route_iter_active(iter))
2080 fib6_walker_unlink(&iter->w);
2082 rcu_read_unlock_bh();
2085 static const struct seq_operations ipv6_route_seq_ops = {
2086 .start = ipv6_route_seq_start,
2087 .next = ipv6_route_seq_next,
2088 .stop = ipv6_route_seq_stop,
2089 .show = ipv6_route_seq_show
2092 int ipv6_route_open(struct inode *inode, struct file *file)
2094 return seq_open_net(inode, file, &ipv6_route_seq_ops,
2095 sizeof(struct ipv6_route_iter));
2098 #endif /* CONFIG_PROC_FS */