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.
16 * Yuji SEKIYA @USAGI: Support default route on router node;
17 * remove ip6_null_entry from the top of
19 * Ville Nuorvala: Fixed routing subtrees.
22 #define pr_fmt(fmt) "IPv6: " fmt
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/net.h>
27 #include <linux/route.h>
28 #include <linux/netdevice.h>
29 #include <linux/in6.h>
30 #include <linux/init.h>
31 #include <linux/list.h>
32 #include <linux/slab.h>
35 #include <net/ndisc.h>
36 #include <net/addrconf.h>
38 #include <net/ip6_fib.h>
39 #include <net/ip6_route.h>
44 #define RT6_TRACE(x...) pr_debug(x)
46 #define RT6_TRACE(x...) do { ; } while (0)
49 static struct kmem_cache * fib6_node_kmem __read_mostly;
53 #ifdef CONFIG_IPV6_SUBTREES
64 struct fib6_walker_t w;
66 int (*func)(struct rt6_info *, void *arg);
70 static DEFINE_RWLOCK(fib6_walker_lock);
72 #ifdef CONFIG_IPV6_SUBTREES
73 #define FWS_INIT FWS_S
75 #define FWS_INIT FWS_L
78 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
80 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
81 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
82 static int fib6_walk(struct fib6_walker_t *w);
83 static int fib6_walk_continue(struct fib6_walker_t *w);
86 * A routing update causes an increase of the serial number on the
87 * affected subtree. This allows for cached routes to be asynchronously
88 * tested when modifications are made to the destination cache as a
89 * result of redirects, path MTU changes, etc.
92 static __u32 rt_sernum;
94 static void fib6_gc_timer_cb(unsigned long arg);
96 static LIST_HEAD(fib6_walkers);
97 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
99 static inline void fib6_walker_link(struct fib6_walker_t *w)
101 write_lock_bh(&fib6_walker_lock);
102 list_add(&w->lh, &fib6_walkers);
103 write_unlock_bh(&fib6_walker_lock);
106 static inline void fib6_walker_unlink(struct fib6_walker_t *w)
108 write_lock_bh(&fib6_walker_lock);
110 write_unlock_bh(&fib6_walker_lock);
112 static __inline__ u32 fib6_new_sernum(void)
121 * Auxiliary address test functions for the radix tree.
123 * These assume a 32bit processor (although it will work on
130 #if defined(__LITTLE_ENDIAN)
131 # define BITOP_BE32_SWIZZLE (0x1F & ~7)
133 # define BITOP_BE32_SWIZZLE 0
136 static __inline__ __be32 addr_bit_set(const void *token, int fn_bit)
138 const __be32 *addr = token;
141 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
142 * is optimized version of
143 * htonl(1 << ((~fn_bit)&0x1F))
144 * See include/asm-generic/bitops/le.h.
146 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
150 static __inline__ struct fib6_node * node_alloc(void)
152 struct fib6_node *fn;
154 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
159 static __inline__ void node_free(struct fib6_node * fn)
161 kmem_cache_free(fib6_node_kmem, fn);
164 static __inline__ void rt6_release(struct rt6_info *rt)
166 if (atomic_dec_and_test(&rt->rt6i_ref))
170 static void fib6_link_table(struct net *net, struct fib6_table *tb)
175 * Initialize table lock at a single place to give lockdep a key,
176 * tables aren't visible prior to being linked to the list.
178 rwlock_init(&tb->tb6_lock);
180 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
183 * No protection necessary, this is the only list mutatation
184 * operation, tables never disappear once they exist.
186 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
189 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
191 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
193 struct fib6_table *table;
195 table = kzalloc(sizeof(*table), GFP_ATOMIC);
198 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
199 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
205 struct fib6_table *fib6_new_table(struct net *net, u32 id)
207 struct fib6_table *tb;
211 tb = fib6_get_table(net, id);
215 tb = fib6_alloc_table(net, id);
217 fib6_link_table(net, tb);
222 struct fib6_table *fib6_get_table(struct net *net, u32 id)
224 struct fib6_table *tb;
225 struct hlist_head *head;
226 struct hlist_node *node;
231 h = id & (FIB6_TABLE_HASHSZ - 1);
233 head = &net->ipv6.fib_table_hash[h];
234 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
235 if (tb->tb6_id == id) {
245 static void __net_init fib6_tables_init(struct net *net)
247 fib6_link_table(net, net->ipv6.fib6_main_tbl);
248 fib6_link_table(net, net->ipv6.fib6_local_tbl);
252 struct fib6_table *fib6_new_table(struct net *net, u32 id)
254 return fib6_get_table(net, id);
257 struct fib6_table *fib6_get_table(struct net *net, u32 id)
259 return net->ipv6.fib6_main_tbl;
262 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
263 int flags, pol_lookup_t lookup)
265 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
268 static void __net_init fib6_tables_init(struct net *net)
270 fib6_link_table(net, net->ipv6.fib6_main_tbl);
275 static int fib6_dump_node(struct fib6_walker_t *w)
280 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
281 res = rt6_dump_route(rt, w->args);
283 /* Frame is full, suspend walking */
293 static void fib6_dump_end(struct netlink_callback *cb)
295 struct fib6_walker_t *w = (void*)cb->args[2];
300 fib6_walker_unlink(w);
305 cb->done = (void*)cb->args[3];
309 static int fib6_dump_done(struct netlink_callback *cb)
312 return cb->done ? cb->done(cb) : 0;
315 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
316 struct netlink_callback *cb)
318 struct fib6_walker_t *w;
321 w = (void *)cb->args[2];
322 w->root = &table->tb6_root;
324 if (cb->args[4] == 0) {
328 read_lock_bh(&table->tb6_lock);
330 read_unlock_bh(&table->tb6_lock);
333 cb->args[5] = w->root->fn_sernum;
336 if (cb->args[5] != w->root->fn_sernum) {
337 /* Begin at the root if the tree changed */
338 cb->args[5] = w->root->fn_sernum;
345 read_lock_bh(&table->tb6_lock);
346 res = fib6_walk_continue(w);
347 read_unlock_bh(&table->tb6_lock);
349 fib6_walker_unlink(w);
357 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
359 struct net *net = sock_net(skb->sk);
361 unsigned int e = 0, s_e;
362 struct rt6_rtnl_dump_arg arg;
363 struct fib6_walker_t *w;
364 struct fib6_table *tb;
365 struct hlist_node *node;
366 struct hlist_head *head;
372 w = (void *)cb->args[2];
376 * 1. hook callback destructor.
378 cb->args[3] = (long)cb->done;
379 cb->done = fib6_dump_done;
382 * 2. allocate and initialize walker.
384 w = kzalloc(sizeof(*w), GFP_ATOMIC);
387 w->func = fib6_dump_node;
388 cb->args[2] = (long)w;
397 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
399 head = &net->ipv6.fib_table_hash[h];
400 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
403 res = fib6_dump_table(tb, skb, cb);
415 res = res < 0 ? res : skb->len;
424 * return the appropriate node for a routing tree "add" operation
425 * by either creating and inserting or by returning an existing
429 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
430 int addrlen, int plen,
431 int offset, int allow_create,
432 int replace_required)
434 struct fib6_node *fn, *in, *ln;
435 struct fib6_node *pn = NULL;
439 __u32 sernum = fib6_new_sernum();
441 RT6_TRACE("fib6_add_1\n");
443 /* insert node in tree */
448 key = (struct rt6key *)((u8 *)fn->leaf + offset);
453 if (plen < fn->fn_bit ||
454 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
456 if (replace_required) {
457 pr_warn("Can't replace route, no match found\n");
458 return ERR_PTR(-ENOENT);
460 pr_warn("NLM_F_CREATE should be set when creating new route\n");
469 if (plen == fn->fn_bit) {
470 /* clean up an intermediate node */
471 if (!(fn->fn_flags & RTN_RTINFO)) {
472 rt6_release(fn->leaf);
476 fn->fn_sernum = sernum;
482 * We have more bits to go
485 /* Try to walk down on tree. */
486 fn->fn_sernum = sernum;
487 dir = addr_bit_set(addr, fn->fn_bit);
489 fn = dir ? fn->right: fn->left;
493 /* We should not create new node because
494 * NLM_F_REPLACE was specified without NLM_F_CREATE
495 * I assume it is safe to require NLM_F_CREATE when
496 * REPLACE flag is used! Later we may want to remove the
497 * check for replace_required, because according
498 * to netlink specification, NLM_F_CREATE
499 * MUST be specified if new route is created.
500 * That would keep IPv6 consistent with IPv4
502 if (replace_required) {
503 pr_warn("Can't replace route, no match found\n");
504 return ERR_PTR(-ENOENT);
506 pr_warn("NLM_F_CREATE should be set when creating new route\n");
509 * We walked to the bottom of tree.
510 * Create new leaf node without children.
520 ln->fn_sernum = sernum;
532 * split since we don't have a common prefix anymore or
533 * we have a less significant route.
534 * we've to insert an intermediate node on the list
535 * this new node will point to the one we need to create
541 /* find 1st bit in difference between the 2 addrs.
543 See comment in __ipv6_addr_diff: bit may be an invalid value,
544 but if it is >= plen, the value is ignored in any case.
547 bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
552 * (new leaf node)[ln] (old node)[fn]
567 * new intermediate node.
569 * be off since that an address that chooses one of
570 * the branches would not match less specific routes
571 * in the other branch
578 atomic_inc(&in->leaf->rt6i_ref);
580 in->fn_sernum = sernum;
582 /* update parent pointer */
593 ln->fn_sernum = sernum;
595 if (addr_bit_set(addr, bit)) {
602 } else { /* plen <= bit */
605 * (new leaf node)[ln]
607 * (old node)[fn] NULL
619 ln->fn_sernum = sernum;
626 if (addr_bit_set(&key->addr, plen))
637 * Insert routing information in a node.
640 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
641 struct nl_info *info)
643 struct rt6_info *iter = NULL;
644 struct rt6_info **ins;
645 int replace = (info->nlh &&
646 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
647 int add = (!info->nlh ||
648 (info->nlh->nlmsg_flags & NLM_F_CREATE));
653 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
655 * Search for duplicates
658 if (iter->rt6i_metric == rt->rt6i_metric) {
660 * Same priority level
663 (info->nlh->nlmsg_flags & NLM_F_EXCL))
670 if (iter->dst.dev == rt->dst.dev &&
671 iter->rt6i_idev == rt->rt6i_idev &&
672 ipv6_addr_equal(&iter->rt6i_gateway,
673 &rt->rt6i_gateway)) {
674 if (!(iter->rt6i_flags & RTF_EXPIRES))
676 if (!(rt->rt6i_flags & RTF_EXPIRES))
677 rt6_clean_expires(iter);
679 rt6_set_expires(iter, rt->dst.expires);
684 if (iter->rt6i_metric > rt->rt6i_metric)
687 ins = &iter->dst.rt6_next;
690 /* Reset round-robin state, if necessary */
691 if (ins == &fn->leaf)
699 pr_warn("NLM_F_CREATE should be set when creating new route\n");
702 rt->dst.rt6_next = iter;
705 atomic_inc(&rt->rt6i_ref);
706 inet6_rt_notify(RTM_NEWROUTE, rt, info);
707 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
709 if (!(fn->fn_flags & RTN_RTINFO)) {
710 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
711 fn->fn_flags |= RTN_RTINFO;
718 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
723 rt->dst.rt6_next = iter->dst.rt6_next;
724 atomic_inc(&rt->rt6i_ref);
725 inet6_rt_notify(RTM_NEWROUTE, rt, info);
727 if (!(fn->fn_flags & RTN_RTINFO)) {
728 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
729 fn->fn_flags |= RTN_RTINFO;
736 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
738 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
739 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
740 mod_timer(&net->ipv6.ip6_fib_timer,
741 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
744 void fib6_force_start_gc(struct net *net)
746 if (!timer_pending(&net->ipv6.ip6_fib_timer))
747 mod_timer(&net->ipv6.ip6_fib_timer,
748 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
752 * Add routing information to the routing tree.
753 * <destination addr>/<source addr>
754 * with source addr info in sub-trees
757 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
759 struct fib6_node *fn, *pn = NULL;
761 int allow_create = 1;
762 int replace_required = 0;
765 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
767 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
768 replace_required = 1;
770 if (!allow_create && !replace_required)
771 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
773 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
774 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst),
775 allow_create, replace_required);
787 #ifdef CONFIG_IPV6_SUBTREES
788 if (rt->rt6i_src.plen) {
789 struct fib6_node *sn;
792 struct fib6_node *sfn;
804 /* Create subtree root node */
809 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
810 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
811 sfn->fn_flags = RTN_ROOT;
812 sfn->fn_sernum = fib6_new_sernum();
814 /* Now add the first leaf node to new subtree */
816 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
817 sizeof(struct in6_addr), rt->rt6i_src.plen,
818 offsetof(struct rt6_info, rt6i_src),
819 allow_create, replace_required);
822 /* If it is failed, discard just allocated
823 root, and then (in st_failure) stale node
830 /* Now link new subtree to main tree */
834 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
835 sizeof(struct in6_addr), rt->rt6i_src.plen,
836 offsetof(struct rt6_info, rt6i_src),
837 allow_create, replace_required);
849 atomic_inc(&rt->rt6i_ref);
855 err = fib6_add_rt2node(fn, rt, info);
857 fib6_start_gc(info->nl_net, rt);
858 if (!(rt->rt6i_flags & RTF_CACHE))
859 fib6_prune_clones(info->nl_net, pn, rt);
864 #ifdef CONFIG_IPV6_SUBTREES
866 * If fib6_add_1 has cleared the old leaf pointer in the
867 * super-tree leaf node we have to find a new one for it.
869 if (pn != fn && pn->leaf == rt) {
871 atomic_dec(&rt->rt6i_ref);
873 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
874 pn->leaf = fib6_find_prefix(info->nl_net, pn);
877 WARN_ON(pn->leaf == NULL);
878 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
881 atomic_inc(&pn->leaf->rt6i_ref);
888 #ifdef CONFIG_IPV6_SUBTREES
889 /* Subtree creation failed, probably main tree node
890 is orphan. If it is, shoot it.
893 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
894 fib6_repair_tree(info->nl_net, fn);
901 * Routing tree lookup
906 int offset; /* key offset on rt6_info */
907 const struct in6_addr *addr; /* search key */
910 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
911 struct lookup_args *args)
913 struct fib6_node *fn;
916 if (unlikely(args->offset == 0))
926 struct fib6_node *next;
928 dir = addr_bit_set(args->addr, fn->fn_bit);
930 next = dir ? fn->right : fn->left;
940 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
943 key = (struct rt6key *) ((u8 *) fn->leaf +
946 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
947 #ifdef CONFIG_IPV6_SUBTREES
949 fn = fib6_lookup_1(fn->subtree, args + 1);
951 if (!fn || fn->fn_flags & RTN_RTINFO)
956 if (fn->fn_flags & RTN_ROOT)
965 struct fib6_node * fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
966 const struct in6_addr *saddr)
968 struct fib6_node *fn;
969 struct lookup_args args[] = {
971 .offset = offsetof(struct rt6_info, rt6i_dst),
974 #ifdef CONFIG_IPV6_SUBTREES
976 .offset = offsetof(struct rt6_info, rt6i_src),
981 .offset = 0, /* sentinel */
985 fn = fib6_lookup_1(root, daddr ? args : args + 1);
986 if (!fn || fn->fn_flags & RTN_TL_ROOT)
993 * Get node with specified destination prefix (and source prefix,
994 * if subtrees are used)
998 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
999 const struct in6_addr *addr,
1000 int plen, int offset)
1002 struct fib6_node *fn;
1004 for (fn = root; fn ; ) {
1005 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1010 if (plen < fn->fn_bit ||
1011 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1014 if (plen == fn->fn_bit)
1018 * We have more bits to go
1020 if (addr_bit_set(addr, fn->fn_bit))
1028 struct fib6_node * fib6_locate(struct fib6_node *root,
1029 const struct in6_addr *daddr, int dst_len,
1030 const struct in6_addr *saddr, int src_len)
1032 struct fib6_node *fn;
1034 fn = fib6_locate_1(root, daddr, dst_len,
1035 offsetof(struct rt6_info, rt6i_dst));
1037 #ifdef CONFIG_IPV6_SUBTREES
1039 WARN_ON(saddr == NULL);
1040 if (fn && fn->subtree)
1041 fn = fib6_locate_1(fn->subtree, saddr, src_len,
1042 offsetof(struct rt6_info, rt6i_src));
1046 if (fn && fn->fn_flags & RTN_RTINFO)
1058 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1060 if (fn->fn_flags & RTN_ROOT)
1061 return net->ipv6.ip6_null_entry;
1065 return fn->left->leaf;
1067 return fn->right->leaf;
1069 fn = FIB6_SUBTREE(fn);
1075 * Called to trim the tree of intermediate nodes when possible. "fn"
1076 * is the node we want to try and remove.
1079 static struct fib6_node *fib6_repair_tree(struct net *net,
1080 struct fib6_node *fn)
1084 struct fib6_node *child, *pn;
1085 struct fib6_walker_t *w;
1089 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1092 WARN_ON(fn->fn_flags & RTN_RTINFO);
1093 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1094 WARN_ON(fn->leaf != NULL);
1098 if (fn->right) child = fn->right, children |= 1;
1099 if (fn->left) child = fn->left, children |= 2;
1101 if (children == 3 || FIB6_SUBTREE(fn)
1102 #ifdef CONFIG_IPV6_SUBTREES
1103 /* Subtree root (i.e. fn) may have one child */
1104 || (children && fn->fn_flags & RTN_ROOT)
1107 fn->leaf = fib6_find_prefix(net, fn);
1111 fn->leaf = net->ipv6.ip6_null_entry;
1114 atomic_inc(&fn->leaf->rt6i_ref);
1119 #ifdef CONFIG_IPV6_SUBTREES
1120 if (FIB6_SUBTREE(pn) == fn) {
1121 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1122 FIB6_SUBTREE(pn) = NULL;
1125 WARN_ON(fn->fn_flags & RTN_ROOT);
1127 if (pn->right == fn) pn->right = child;
1128 else if (pn->left == fn) pn->left = child;
1136 #ifdef CONFIG_IPV6_SUBTREES
1140 read_lock(&fib6_walker_lock);
1143 if (w->root == fn) {
1144 w->root = w->node = NULL;
1145 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1146 } else if (w->node == fn) {
1147 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1152 if (w->root == fn) {
1154 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1156 if (w->node == fn) {
1159 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1160 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1162 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1163 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1168 read_unlock(&fib6_walker_lock);
1171 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1174 rt6_release(pn->leaf);
1180 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1181 struct nl_info *info)
1183 struct fib6_walker_t *w;
1184 struct rt6_info *rt = *rtp;
1185 struct net *net = info->nl_net;
1187 RT6_TRACE("fib6_del_route\n");
1190 *rtp = rt->dst.rt6_next;
1191 rt->rt6i_node = NULL;
1192 net->ipv6.rt6_stats->fib_rt_entries--;
1193 net->ipv6.rt6_stats->fib_discarded_routes++;
1195 /* Reset round-robin state, if necessary */
1196 if (fn->rr_ptr == rt)
1199 /* Adjust walkers */
1200 read_lock(&fib6_walker_lock);
1202 if (w->state == FWS_C && w->leaf == rt) {
1203 RT6_TRACE("walker %p adjusted by delroute\n", w);
1204 w->leaf = rt->dst.rt6_next;
1209 read_unlock(&fib6_walker_lock);
1211 rt->dst.rt6_next = NULL;
1213 /* If it was last route, expunge its radix tree node */
1215 fn->fn_flags &= ~RTN_RTINFO;
1216 net->ipv6.rt6_stats->fib_route_nodes--;
1217 fn = fib6_repair_tree(net, fn);
1220 if (atomic_read(&rt->rt6i_ref) != 1) {
1221 /* This route is used as dummy address holder in some split
1222 * nodes. It is not leaked, but it still holds other resources,
1223 * which must be released in time. So, scan ascendant nodes
1224 * and replace dummy references to this route with references
1225 * to still alive ones.
1228 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
1229 fn->leaf = fib6_find_prefix(net, fn);
1230 atomic_inc(&fn->leaf->rt6i_ref);
1235 /* No more references are possible at this point. */
1236 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1239 inet6_rt_notify(RTM_DELROUTE, rt, info);
1243 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1245 struct net *net = info->nl_net;
1246 struct fib6_node *fn = rt->rt6i_node;
1247 struct rt6_info **rtp;
1250 if (rt->dst.obsolete>0) {
1251 WARN_ON(fn != NULL);
1255 if (!fn || rt == net->ipv6.ip6_null_entry)
1258 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1260 if (!(rt->rt6i_flags & RTF_CACHE)) {
1261 struct fib6_node *pn = fn;
1262 #ifdef CONFIG_IPV6_SUBTREES
1263 /* clones of this route might be in another subtree */
1264 if (rt->rt6i_src.plen) {
1265 while (!(pn->fn_flags & RTN_ROOT))
1270 fib6_prune_clones(info->nl_net, pn, rt);
1274 * Walk the leaf entries looking for ourself
1277 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1279 fib6_del_route(fn, rtp, info);
1287 * Tree traversal function.
1289 * Certainly, it is not interrupt safe.
1290 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1291 * It means, that we can modify tree during walking
1292 * and use this function for garbage collection, clone pruning,
1293 * cleaning tree when a device goes down etc. etc.
1295 * It guarantees that every node will be traversed,
1296 * and that it will be traversed only once.
1298 * Callback function w->func may return:
1299 * 0 -> continue walking.
1300 * positive value -> walking is suspended (used by tree dumps,
1301 * and probably by gc, if it will be split to several slices)
1302 * negative value -> terminate walking.
1304 * The function itself returns:
1305 * 0 -> walk is complete.
1306 * >0 -> walk is incomplete (i.e. suspended)
1307 * <0 -> walk is terminated by an error.
1310 static int fib6_walk_continue(struct fib6_walker_t *w)
1312 struct fib6_node *fn, *pn;
1319 if (w->prune && fn != w->root &&
1320 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1325 #ifdef CONFIG_IPV6_SUBTREES
1327 if (FIB6_SUBTREE(fn)) {
1328 w->node = FIB6_SUBTREE(fn);
1336 w->state = FWS_INIT;
1342 w->node = fn->right;
1343 w->state = FWS_INIT;
1349 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1370 #ifdef CONFIG_IPV6_SUBTREES
1371 if (FIB6_SUBTREE(pn) == fn) {
1372 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1377 if (pn->left == fn) {
1381 if (pn->right == fn) {
1383 w->leaf = w->node->leaf;
1393 static int fib6_walk(struct fib6_walker_t *w)
1397 w->state = FWS_INIT;
1400 fib6_walker_link(w);
1401 res = fib6_walk_continue(w);
1403 fib6_walker_unlink(w);
1407 static int fib6_clean_node(struct fib6_walker_t *w)
1410 struct rt6_info *rt;
1411 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1412 struct nl_info info = {
1416 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1417 res = c->func(rt, c->arg);
1420 res = fib6_del(rt, &info);
1423 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1424 __func__, rt, rt->rt6i_node, res);
1437 * Convenient frontend to tree walker.
1439 * func is called on each route.
1440 * It may return -1 -> delete this route.
1441 * 0 -> continue walking
1443 * prune==1 -> only immediate children of node (certainly,
1444 * ignoring pure split nodes) will be scanned.
1447 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1448 int (*func)(struct rt6_info *, void *arg),
1449 int prune, void *arg)
1451 struct fib6_cleaner_t c;
1454 c.w.func = fib6_clean_node;
1465 void fib6_clean_all_ro(struct net *net, int (*func)(struct rt6_info *, void *arg),
1466 int prune, void *arg)
1468 struct fib6_table *table;
1469 struct hlist_node *node;
1470 struct hlist_head *head;
1474 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1475 head = &net->ipv6.fib_table_hash[h];
1476 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1477 read_lock_bh(&table->tb6_lock);
1478 fib6_clean_tree(net, &table->tb6_root,
1480 read_unlock_bh(&table->tb6_lock);
1485 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1486 int prune, void *arg)
1488 struct fib6_table *table;
1489 struct hlist_node *node;
1490 struct hlist_head *head;
1494 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1495 head = &net->ipv6.fib_table_hash[h];
1496 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1497 write_lock_bh(&table->tb6_lock);
1498 fib6_clean_tree(net, &table->tb6_root,
1500 write_unlock_bh(&table->tb6_lock);
1506 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1508 if (rt->rt6i_flags & RTF_CACHE) {
1509 RT6_TRACE("pruning clone %p\n", rt);
1516 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1517 struct rt6_info *rt)
1519 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1523 * Garbage collection
1526 static struct fib6_gc_args
1532 static int fib6_age(struct rt6_info *rt, void *arg)
1534 unsigned long now = jiffies;
1537 * check addrconf expiration here.
1538 * Routes are expired even if they are in use.
1540 * Also age clones. Note, that clones are aged out
1541 * only if they are not in use now.
1544 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1545 if (time_after(now, rt->dst.expires)) {
1546 RT6_TRACE("expiring %p\n", rt);
1550 } else if (rt->rt6i_flags & RTF_CACHE) {
1551 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1552 time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1553 RT6_TRACE("aging clone %p\n", rt);
1555 } else if (rt->rt6i_flags & RTF_GATEWAY) {
1556 struct neighbour *neigh;
1557 __u8 neigh_flags = 0;
1559 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1561 neigh_flags = neigh->flags;
1562 neigh_release(neigh);
1564 if (!(neigh_flags & NTF_ROUTER)) {
1565 RT6_TRACE("purging route %p via non-router but gateway\n",
1576 static DEFINE_SPINLOCK(fib6_gc_lock);
1578 void fib6_run_gc(unsigned long expires, struct net *net)
1580 if (expires != ~0UL) {
1581 spin_lock_bh(&fib6_gc_lock);
1582 gc_args.timeout = expires ? (int)expires :
1583 net->ipv6.sysctl.ip6_rt_gc_interval;
1585 if (!spin_trylock_bh(&fib6_gc_lock)) {
1586 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1589 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1592 gc_args.more = icmp6_dst_gc();
1594 fib6_clean_all(net, fib6_age, 0, NULL);
1597 mod_timer(&net->ipv6.ip6_fib_timer,
1598 round_jiffies(jiffies
1599 + net->ipv6.sysctl.ip6_rt_gc_interval));
1601 del_timer(&net->ipv6.ip6_fib_timer);
1602 spin_unlock_bh(&fib6_gc_lock);
1605 static void fib6_gc_timer_cb(unsigned long arg)
1607 fib6_run_gc(0, (struct net *)arg);
1610 static int __net_init fib6_net_init(struct net *net)
1612 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1614 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1616 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1617 if (!net->ipv6.rt6_stats)
1620 /* Avoid false sharing : Use at least a full cache line */
1621 size = max_t(size_t, size, L1_CACHE_BYTES);
1623 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1624 if (!net->ipv6.fib_table_hash)
1627 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1629 if (!net->ipv6.fib6_main_tbl)
1630 goto out_fib_table_hash;
1632 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1633 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1634 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1635 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1637 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1638 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1640 if (!net->ipv6.fib6_local_tbl)
1641 goto out_fib6_main_tbl;
1642 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1643 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1644 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1645 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1647 fib6_tables_init(net);
1651 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1653 kfree(net->ipv6.fib6_main_tbl);
1656 kfree(net->ipv6.fib_table_hash);
1658 kfree(net->ipv6.rt6_stats);
1663 static void fib6_net_exit(struct net *net)
1665 rt6_ifdown(net, NULL);
1666 del_timer_sync(&net->ipv6.ip6_fib_timer);
1668 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1669 kfree(net->ipv6.fib6_local_tbl);
1671 kfree(net->ipv6.fib6_main_tbl);
1672 kfree(net->ipv6.fib_table_hash);
1673 kfree(net->ipv6.rt6_stats);
1676 static struct pernet_operations fib6_net_ops = {
1677 .init = fib6_net_init,
1678 .exit = fib6_net_exit,
1681 int __init fib6_init(void)
1685 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1686 sizeof(struct fib6_node),
1687 0, SLAB_HWCACHE_ALIGN,
1689 if (!fib6_node_kmem)
1692 ret = register_pernet_subsys(&fib6_net_ops);
1694 goto out_kmem_cache_create;
1696 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1699 goto out_unregister_subsys;
1703 out_unregister_subsys:
1704 unregister_pernet_subsys(&fib6_net_ops);
1705 out_kmem_cache_create:
1706 kmem_cache_destroy(fib6_node_kmem);
1710 void fib6_gc_cleanup(void)
1712 unregister_pernet_subsys(&fib6_net_ops);
1713 kmem_cache_destroy(fib6_node_kmem);