2 * Linux INET6 implementation
3 * Forwarding Information Database
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * $Id: ip6_fib.c,v 1.25 2001/10/31 21:55:55 davem Exp $
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
18 * Yuji SEKIYA @USAGI: Support default route on router node;
19 * remove ip6_null_entry from the top of
21 * Ville Nuorvala: Fixed routing subtrees.
23 #include <linux/errno.h>
24 #include <linux/types.h>
25 #include <linux/net.h>
26 #include <linux/route.h>
27 #include <linux/netdevice.h>
28 #include <linux/in6.h>
29 #include <linux/init.h>
30 #include <linux/list.h>
33 #include <linux/proc_fs.h>
37 #include <net/ndisc.h>
38 #include <net/addrconf.h>
40 #include <net/ip6_fib.h>
41 #include <net/ip6_route.h>
46 #define RT6_TRACE(x...) printk(KERN_DEBUG x)
48 #define RT6_TRACE(x...) do { ; } while (0)
51 static struct kmem_cache * fib6_node_kmem __read_mostly;
55 #ifdef CONFIG_IPV6_SUBTREES
66 struct fib6_walker_t w;
68 int (*func)(struct rt6_info *, void *arg);
72 static DEFINE_RWLOCK(fib6_walker_lock);
74 #ifdef CONFIG_IPV6_SUBTREES
75 #define FWS_INIT FWS_S
77 #define FWS_INIT FWS_L
80 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
82 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
83 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
84 static int fib6_walk(struct fib6_walker_t *w);
85 static int fib6_walk_continue(struct fib6_walker_t *w);
88 * A routing update causes an increase of the serial number on the
89 * affected subtree. This allows for cached routes to be asynchronously
90 * tested when modifications are made to the destination cache as a
91 * result of redirects, path MTU changes, etc.
94 static __u32 rt_sernum;
96 static void fib6_gc_timer_cb(unsigned long arg);
98 static struct fib6_walker_t fib6_walker_list = {
99 .prev = &fib6_walker_list,
100 .next = &fib6_walker_list,
103 #define FOR_WALKERS(w) for ((w)=fib6_walker_list.next; (w) != &fib6_walker_list; (w)=(w)->next)
105 static inline void fib6_walker_link(struct fib6_walker_t *w)
107 write_lock_bh(&fib6_walker_lock);
108 w->next = fib6_walker_list.next;
109 w->prev = &fib6_walker_list;
112 write_unlock_bh(&fib6_walker_lock);
115 static inline void fib6_walker_unlink(struct fib6_walker_t *w)
117 write_lock_bh(&fib6_walker_lock);
118 w->next->prev = w->prev;
119 w->prev->next = w->next;
120 w->prev = w->next = w;
121 write_unlock_bh(&fib6_walker_lock);
123 static __inline__ u32 fib6_new_sernum(void)
132 * Auxiliary address test functions for the radix tree.
134 * These assume a 32bit processor (although it will work on
142 static __inline__ __be32 addr_bit_set(void *token, int fn_bit)
144 __be32 *addr = token;
146 return htonl(1 << ((~fn_bit)&0x1F)) & addr[fn_bit>>5];
149 static __inline__ struct fib6_node * node_alloc(void)
151 struct fib6_node *fn;
153 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
158 static __inline__ void node_free(struct fib6_node * fn)
160 kmem_cache_free(fib6_node_kmem, fn);
163 static __inline__ void rt6_release(struct rt6_info *rt)
165 if (atomic_dec_and_test(&rt->rt6i_ref))
166 dst_free(&rt->u.dst);
169 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
170 #define FIB_TABLE_HASHSZ 256
172 #define FIB_TABLE_HASHSZ 1
175 static void fib6_link_table(struct net *net, struct fib6_table *tb)
180 * Initialize table lock at a single place to give lockdep a key,
181 * tables aren't visible prior to being linked to the list.
183 rwlock_init(&tb->tb6_lock);
185 h = tb->tb6_id & (FIB_TABLE_HASHSZ - 1);
188 * No protection necessary, this is the only list mutatation
189 * operation, tables never disappear once they exist.
191 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
194 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
196 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
198 struct fib6_table *table;
200 table = kzalloc(sizeof(*table), GFP_ATOMIC);
203 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
204 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
210 struct fib6_table *fib6_new_table(struct net *net, u32 id)
212 struct fib6_table *tb;
216 tb = fib6_get_table(net, id);
220 tb = fib6_alloc_table(net, id);
222 fib6_link_table(net, tb);
227 struct fib6_table *fib6_get_table(struct net *net, u32 id)
229 struct fib6_table *tb;
230 struct hlist_head *head;
231 struct hlist_node *node;
236 h = id & (FIB_TABLE_HASHSZ - 1);
238 head = &net->ipv6.fib_table_hash[h];
239 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
240 if (tb->tb6_id == id) {
250 static void fib6_tables_init(struct net *net)
252 fib6_link_table(net, net->ipv6.fib6_main_tbl);
253 fib6_link_table(net, net->ipv6.fib6_local_tbl);
257 struct fib6_table *fib6_new_table(struct net *net, u32 id)
259 return fib6_get_table(net, id);
262 struct fib6_table *fib6_get_table(struct net *net, u32 id)
264 return net->ipv6.fib6_main_tbl;
267 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi *fl,
268 int flags, pol_lookup_t lookup)
270 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl, flags);
273 static void fib6_tables_init(struct net *net)
275 fib6_link_table(net, net->ipv6.fib6_main_tbl);
280 static int fib6_dump_node(struct fib6_walker_t *w)
285 for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) {
286 res = rt6_dump_route(rt, w->args);
288 /* Frame is full, suspend walking */
298 static void fib6_dump_end(struct netlink_callback *cb)
300 struct fib6_walker_t *w = (void*)cb->args[2];
306 cb->done = (void*)cb->args[3];
310 static int fib6_dump_done(struct netlink_callback *cb)
313 return cb->done ? cb->done(cb) : 0;
316 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
317 struct netlink_callback *cb)
319 struct fib6_walker_t *w;
322 w = (void *)cb->args[2];
323 w->root = &table->tb6_root;
325 if (cb->args[4] == 0) {
326 read_lock_bh(&table->tb6_lock);
328 read_unlock_bh(&table->tb6_lock);
332 read_lock_bh(&table->tb6_lock);
333 res = fib6_walk_continue(w);
334 read_unlock_bh(&table->tb6_lock);
337 fib6_walker_unlink(w);
340 fib6_walker_unlink(w);
347 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
349 struct net *net = sock_net(skb->sk);
351 unsigned int e = 0, s_e;
352 struct rt6_rtnl_dump_arg arg;
353 struct fib6_walker_t *w;
354 struct fib6_table *tb;
355 struct hlist_node *node;
356 struct hlist_head *head;
362 w = (void *)cb->args[2];
366 * 1. hook callback destructor.
368 cb->args[3] = (long)cb->done;
369 cb->done = fib6_dump_done;
372 * 2. allocate and initialize walker.
374 w = kzalloc(sizeof(*w), GFP_ATOMIC);
377 w->func = fib6_dump_node;
378 cb->args[2] = (long)w;
385 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
387 head = &net->ipv6.fib_table_hash[h];
388 hlist_for_each_entry(tb, node, head, tb6_hlist) {
391 res = fib6_dump_table(tb, skb, cb);
402 res = res < 0 ? res : skb->len;
411 * return the appropriate node for a routing tree "add" operation
412 * by either creating and inserting or by returning an existing
416 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
417 int addrlen, int plen,
420 struct fib6_node *fn, *in, *ln;
421 struct fib6_node *pn = NULL;
425 __u32 sernum = fib6_new_sernum();
427 RT6_TRACE("fib6_add_1\n");
429 /* insert node in tree */
434 key = (struct rt6key *)((u8 *)fn->leaf + offset);
439 if (plen < fn->fn_bit ||
440 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
447 if (plen == fn->fn_bit) {
448 /* clean up an intermediate node */
449 if ((fn->fn_flags & RTN_RTINFO) == 0) {
450 rt6_release(fn->leaf);
454 fn->fn_sernum = sernum;
460 * We have more bits to go
463 /* Try to walk down on tree. */
464 fn->fn_sernum = sernum;
465 dir = addr_bit_set(addr, fn->fn_bit);
467 fn = dir ? fn->right: fn->left;
471 * We walked to the bottom of tree.
472 * Create new leaf node without children.
482 ln->fn_sernum = sernum;
494 * split since we don't have a common prefix anymore or
495 * we have a less significant route.
496 * we've to insert an intermediate node on the list
497 * this new node will point to the one we need to create
503 /* find 1st bit in difference between the 2 addrs.
505 See comment in __ipv6_addr_diff: bit may be an invalid value,
506 but if it is >= plen, the value is ignored in any case.
509 bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
514 * (new leaf node)[ln] (old node)[fn]
520 if (in == NULL || ln == NULL) {
529 * new intermediate node.
531 * be off since that an address that chooses one of
532 * the branches would not match less specific routes
533 * in the other branch
540 atomic_inc(&in->leaf->rt6i_ref);
542 in->fn_sernum = sernum;
544 /* update parent pointer */
555 ln->fn_sernum = sernum;
557 if (addr_bit_set(addr, bit)) {
564 } else { /* plen <= bit */
567 * (new leaf node)[ln]
569 * (old node)[fn] NULL
581 ln->fn_sernum = sernum;
588 if (addr_bit_set(&key->addr, plen))
599 * Insert routing information in a node.
602 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
603 struct nl_info *info)
605 struct rt6_info *iter = NULL;
606 struct rt6_info **ins;
610 for (iter = fn->leaf; iter; iter=iter->u.dst.rt6_next) {
612 * Search for duplicates
615 if (iter->rt6i_metric == rt->rt6i_metric) {
617 * Same priority level
620 if (iter->rt6i_dev == rt->rt6i_dev &&
621 iter->rt6i_idev == rt->rt6i_idev &&
622 ipv6_addr_equal(&iter->rt6i_gateway,
623 &rt->rt6i_gateway)) {
624 if (!(iter->rt6i_flags&RTF_EXPIRES))
626 iter->rt6i_expires = rt->rt6i_expires;
627 if (!(rt->rt6i_flags&RTF_EXPIRES)) {
628 iter->rt6i_flags &= ~RTF_EXPIRES;
629 iter->rt6i_expires = 0;
635 if (iter->rt6i_metric > rt->rt6i_metric)
638 ins = &iter->u.dst.rt6_next;
641 /* Reset round-robin state, if necessary */
642 if (ins == &fn->leaf)
649 rt->u.dst.rt6_next = iter;
652 atomic_inc(&rt->rt6i_ref);
653 inet6_rt_notify(RTM_NEWROUTE, rt, info);
654 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
656 if ((fn->fn_flags & RTN_RTINFO) == 0) {
657 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
658 fn->fn_flags |= RTN_RTINFO;
664 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
666 if (net->ipv6.ip6_fib_timer->expires == 0 &&
667 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE)))
668 mod_timer(net->ipv6.ip6_fib_timer, jiffies +
669 net->ipv6.sysctl.ip6_rt_gc_interval);
672 void fib6_force_start_gc(struct net *net)
674 if (net->ipv6.ip6_fib_timer->expires == 0)
675 mod_timer(net->ipv6.ip6_fib_timer, jiffies +
676 net->ipv6.sysctl.ip6_rt_gc_interval);
680 * Add routing information to the routing tree.
681 * <destination addr>/<source addr>
682 * with source addr info in sub-trees
685 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
687 struct fib6_node *fn, *pn = NULL;
690 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
691 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst));
698 #ifdef CONFIG_IPV6_SUBTREES
699 if (rt->rt6i_src.plen) {
700 struct fib6_node *sn;
702 if (fn->subtree == NULL) {
703 struct fib6_node *sfn;
715 /* Create subtree root node */
720 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
721 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
722 sfn->fn_flags = RTN_ROOT;
723 sfn->fn_sernum = fib6_new_sernum();
725 /* Now add the first leaf node to new subtree */
727 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
728 sizeof(struct in6_addr), rt->rt6i_src.plen,
729 offsetof(struct rt6_info, rt6i_src));
732 /* If it is failed, discard just allocated
733 root, and then (in st_failure) stale node
740 /* Now link new subtree to main tree */
744 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
745 sizeof(struct in6_addr), rt->rt6i_src.plen,
746 offsetof(struct rt6_info, rt6i_src));
752 if (fn->leaf == NULL) {
754 atomic_inc(&rt->rt6i_ref);
760 err = fib6_add_rt2node(fn, rt, info);
763 fib6_start_gc(info->nl_net, rt);
764 if (!(rt->rt6i_flags&RTF_CACHE))
765 fib6_prune_clones(info->nl_net, pn, rt);
770 #ifdef CONFIG_IPV6_SUBTREES
772 * If fib6_add_1 has cleared the old leaf pointer in the
773 * super-tree leaf node we have to find a new one for it.
775 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
776 pn->leaf = fib6_find_prefix(info->nl_net, pn);
779 BUG_TRAP(pn->leaf != NULL);
780 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
783 atomic_inc(&pn->leaf->rt6i_ref);
786 dst_free(&rt->u.dst);
790 #ifdef CONFIG_IPV6_SUBTREES
791 /* Subtree creation failed, probably main tree node
792 is orphan. If it is, shoot it.
795 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
796 fib6_repair_tree(info->nl_net, fn);
797 dst_free(&rt->u.dst);
803 * Routing tree lookup
808 int offset; /* key offset on rt6_info */
809 struct in6_addr *addr; /* search key */
812 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
813 struct lookup_args *args)
815 struct fib6_node *fn;
818 if (unlikely(args->offset == 0))
828 struct fib6_node *next;
830 dir = addr_bit_set(args->addr, fn->fn_bit);
832 next = dir ? fn->right : fn->left;
843 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
846 key = (struct rt6key *) ((u8 *) fn->leaf +
849 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
850 #ifdef CONFIG_IPV6_SUBTREES
852 fn = fib6_lookup_1(fn->subtree, args + 1);
854 if (!fn || fn->fn_flags & RTN_RTINFO)
859 if (fn->fn_flags & RTN_ROOT)
868 struct fib6_node * fib6_lookup(struct fib6_node *root, struct in6_addr *daddr,
869 struct in6_addr *saddr)
871 struct fib6_node *fn;
872 struct lookup_args args[] = {
874 .offset = offsetof(struct rt6_info, rt6i_dst),
877 #ifdef CONFIG_IPV6_SUBTREES
879 .offset = offsetof(struct rt6_info, rt6i_src),
884 .offset = 0, /* sentinel */
888 fn = fib6_lookup_1(root, daddr ? args : args + 1);
890 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT)
897 * Get node with specified destination prefix (and source prefix,
898 * if subtrees are used)
902 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
903 struct in6_addr *addr,
904 int plen, int offset)
906 struct fib6_node *fn;
908 for (fn = root; fn ; ) {
909 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
914 if (plen < fn->fn_bit ||
915 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
918 if (plen == fn->fn_bit)
922 * We have more bits to go
924 if (addr_bit_set(addr, fn->fn_bit))
932 struct fib6_node * fib6_locate(struct fib6_node *root,
933 struct in6_addr *daddr, int dst_len,
934 struct in6_addr *saddr, int src_len)
936 struct fib6_node *fn;
938 fn = fib6_locate_1(root, daddr, dst_len,
939 offsetof(struct rt6_info, rt6i_dst));
941 #ifdef CONFIG_IPV6_SUBTREES
943 BUG_TRAP(saddr!=NULL);
944 if (fn && fn->subtree)
945 fn = fib6_locate_1(fn->subtree, saddr, src_len,
946 offsetof(struct rt6_info, rt6i_src));
950 if (fn && fn->fn_flags&RTN_RTINFO)
962 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
964 if (fn->fn_flags&RTN_ROOT)
965 return net->ipv6.ip6_null_entry;
969 return fn->left->leaf;
972 return fn->right->leaf;
974 fn = FIB6_SUBTREE(fn);
980 * Called to trim the tree of intermediate nodes when possible. "fn"
981 * is the node we want to try and remove.
984 static struct fib6_node *fib6_repair_tree(struct net *net,
985 struct fib6_node *fn)
989 struct fib6_node *child, *pn;
990 struct fib6_walker_t *w;
994 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
997 BUG_TRAP(!(fn->fn_flags&RTN_RTINFO));
998 BUG_TRAP(!(fn->fn_flags&RTN_TL_ROOT));
999 BUG_TRAP(fn->leaf==NULL);
1003 if (fn->right) child = fn->right, children |= 1;
1004 if (fn->left) child = fn->left, children |= 2;
1006 if (children == 3 || FIB6_SUBTREE(fn)
1007 #ifdef CONFIG_IPV6_SUBTREES
1008 /* Subtree root (i.e. fn) may have one child */
1009 || (children && fn->fn_flags&RTN_ROOT)
1012 fn->leaf = fib6_find_prefix(net, fn);
1014 if (fn->leaf==NULL) {
1016 fn->leaf = net->ipv6.ip6_null_entry;
1019 atomic_inc(&fn->leaf->rt6i_ref);
1024 #ifdef CONFIG_IPV6_SUBTREES
1025 if (FIB6_SUBTREE(pn) == fn) {
1026 BUG_TRAP(fn->fn_flags&RTN_ROOT);
1027 FIB6_SUBTREE(pn) = NULL;
1030 BUG_TRAP(!(fn->fn_flags&RTN_ROOT));
1032 if (pn->right == fn) pn->right = child;
1033 else if (pn->left == fn) pn->left = child;
1040 #ifdef CONFIG_IPV6_SUBTREES
1044 read_lock(&fib6_walker_lock);
1046 if (child == NULL) {
1047 if (w->root == fn) {
1048 w->root = w->node = NULL;
1049 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1050 } else if (w->node == fn) {
1051 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1056 if (w->root == fn) {
1058 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1060 if (w->node == fn) {
1063 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1064 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1066 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1067 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1072 read_unlock(&fib6_walker_lock);
1075 if (pn->fn_flags&RTN_RTINFO || FIB6_SUBTREE(pn))
1078 rt6_release(pn->leaf);
1084 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1085 struct nl_info *info)
1087 struct fib6_walker_t *w;
1088 struct rt6_info *rt = *rtp;
1089 struct net *net = info->nl_net;
1091 RT6_TRACE("fib6_del_route\n");
1094 *rtp = rt->u.dst.rt6_next;
1095 rt->rt6i_node = NULL;
1096 net->ipv6.rt6_stats->fib_rt_entries--;
1097 net->ipv6.rt6_stats->fib_discarded_routes++;
1099 /* Reset round-robin state, if necessary */
1100 if (fn->rr_ptr == rt)
1103 /* Adjust walkers */
1104 read_lock(&fib6_walker_lock);
1106 if (w->state == FWS_C && w->leaf == rt) {
1107 RT6_TRACE("walker %p adjusted by delroute\n", w);
1108 w->leaf = rt->u.dst.rt6_next;
1109 if (w->leaf == NULL)
1113 read_unlock(&fib6_walker_lock);
1115 rt->u.dst.rt6_next = NULL;
1117 /* If it was last route, expunge its radix tree node */
1118 if (fn->leaf == NULL) {
1119 fn->fn_flags &= ~RTN_RTINFO;
1120 net->ipv6.rt6_stats->fib_route_nodes--;
1121 fn = fib6_repair_tree(net, fn);
1124 if (atomic_read(&rt->rt6i_ref) != 1) {
1125 /* This route is used as dummy address holder in some split
1126 * nodes. It is not leaked, but it still holds other resources,
1127 * which must be released in time. So, scan ascendant nodes
1128 * and replace dummy references to this route with references
1129 * to still alive ones.
1132 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) {
1133 fn->leaf = fib6_find_prefix(net, fn);
1134 atomic_inc(&fn->leaf->rt6i_ref);
1139 /* No more references are possible at this point. */
1140 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1143 inet6_rt_notify(RTM_DELROUTE, rt, info);
1147 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1149 struct net *net = info->nl_net;
1150 struct fib6_node *fn = rt->rt6i_node;
1151 struct rt6_info **rtp;
1154 if (rt->u.dst.obsolete>0) {
1159 if (fn == NULL || rt == net->ipv6.ip6_null_entry)
1162 BUG_TRAP(fn->fn_flags&RTN_RTINFO);
1164 if (!(rt->rt6i_flags&RTF_CACHE)) {
1165 struct fib6_node *pn = fn;
1166 #ifdef CONFIG_IPV6_SUBTREES
1167 /* clones of this route might be in another subtree */
1168 if (rt->rt6i_src.plen) {
1169 while (!(pn->fn_flags&RTN_ROOT))
1174 fib6_prune_clones(info->nl_net, pn, rt);
1178 * Walk the leaf entries looking for ourself
1181 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->u.dst.rt6_next) {
1183 fib6_del_route(fn, rtp, info);
1191 * Tree traversal function.
1193 * Certainly, it is not interrupt safe.
1194 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1195 * It means, that we can modify tree during walking
1196 * and use this function for garbage collection, clone pruning,
1197 * cleaning tree when a device goes down etc. etc.
1199 * It guarantees that every node will be traversed,
1200 * and that it will be traversed only once.
1202 * Callback function w->func may return:
1203 * 0 -> continue walking.
1204 * positive value -> walking is suspended (used by tree dumps,
1205 * and probably by gc, if it will be split to several slices)
1206 * negative value -> terminate walking.
1208 * The function itself returns:
1209 * 0 -> walk is complete.
1210 * >0 -> walk is incomplete (i.e. suspended)
1211 * <0 -> walk is terminated by an error.
1214 static int fib6_walk_continue(struct fib6_walker_t *w)
1216 struct fib6_node *fn, *pn;
1223 if (w->prune && fn != w->root &&
1224 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) {
1229 #ifdef CONFIG_IPV6_SUBTREES
1231 if (FIB6_SUBTREE(fn)) {
1232 w->node = FIB6_SUBTREE(fn);
1240 w->state = FWS_INIT;
1246 w->node = fn->right;
1247 w->state = FWS_INIT;
1253 if (w->leaf && fn->fn_flags&RTN_RTINFO) {
1254 int err = w->func(w);
1265 #ifdef CONFIG_IPV6_SUBTREES
1266 if (FIB6_SUBTREE(pn) == fn) {
1267 BUG_TRAP(fn->fn_flags&RTN_ROOT);
1272 if (pn->left == fn) {
1276 if (pn->right == fn) {
1278 w->leaf = w->node->leaf;
1288 static int fib6_walk(struct fib6_walker_t *w)
1292 w->state = FWS_INIT;
1295 fib6_walker_link(w);
1296 res = fib6_walk_continue(w);
1298 fib6_walker_unlink(w);
1302 static int fib6_clean_node(struct fib6_walker_t *w)
1305 struct rt6_info *rt;
1306 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1307 struct nl_info info = {
1311 for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) {
1312 res = c->func(rt, c->arg);
1315 res = fib6_del(rt, &info);
1318 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res);
1331 * Convenient frontend to tree walker.
1333 * func is called on each route.
1334 * It may return -1 -> delete this route.
1335 * 0 -> continue walking
1337 * prune==1 -> only immediate children of node (certainly,
1338 * ignoring pure split nodes) will be scanned.
1341 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1342 int (*func)(struct rt6_info *, void *arg),
1343 int prune, void *arg)
1345 struct fib6_cleaner_t c;
1348 c.w.func = fib6_clean_node;
1357 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1358 int prune, void *arg)
1360 struct fib6_table *table;
1361 struct hlist_node *node;
1362 struct hlist_head *head;
1366 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
1367 head = &net->ipv6.fib_table_hash[h];
1368 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1369 write_lock_bh(&table->tb6_lock);
1370 fib6_clean_tree(net, &table->tb6_root,
1372 write_unlock_bh(&table->tb6_lock);
1378 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1380 if (rt->rt6i_flags & RTF_CACHE) {
1381 RT6_TRACE("pruning clone %p\n", rt);
1388 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1389 struct rt6_info *rt)
1391 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1395 * Garbage collection
1398 static struct fib6_gc_args
1404 static int fib6_age(struct rt6_info *rt, void *arg)
1406 unsigned long now = jiffies;
1409 * check addrconf expiration here.
1410 * Routes are expired even if they are in use.
1412 * Also age clones. Note, that clones are aged out
1413 * only if they are not in use now.
1416 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) {
1417 if (time_after(now, rt->rt6i_expires)) {
1418 RT6_TRACE("expiring %p\n", rt);
1422 } else if (rt->rt6i_flags & RTF_CACHE) {
1423 if (atomic_read(&rt->u.dst.__refcnt) == 0 &&
1424 time_after_eq(now, rt->u.dst.lastuse + gc_args.timeout)) {
1425 RT6_TRACE("aging clone %p\n", rt);
1427 } else if ((rt->rt6i_flags & RTF_GATEWAY) &&
1428 (!(rt->rt6i_nexthop->flags & NTF_ROUTER))) {
1429 RT6_TRACE("purging route %p via non-router but gateway\n",
1439 static DEFINE_SPINLOCK(fib6_gc_lock);
1441 void fib6_run_gc(unsigned long expires, struct net *net)
1443 if (expires != ~0UL) {
1444 spin_lock_bh(&fib6_gc_lock);
1445 gc_args.timeout = expires ? (int)expires :
1446 net->ipv6.sysctl.ip6_rt_gc_interval;
1449 if (!spin_trylock(&fib6_gc_lock)) {
1450 mod_timer(net->ipv6.ip6_fib_timer, jiffies + HZ);
1454 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1458 icmp6_dst_gc(&gc_args.more);
1460 fib6_clean_all(net, fib6_age, 0, NULL);
1463 mod_timer(net->ipv6.ip6_fib_timer, jiffies +
1464 net->ipv6.sysctl.ip6_rt_gc_interval);
1466 del_timer(net->ipv6.ip6_fib_timer);
1467 net->ipv6.ip6_fib_timer->expires = 0;
1469 spin_unlock_bh(&fib6_gc_lock);
1472 static void fib6_gc_timer_cb(unsigned long arg)
1474 fib6_run_gc(0, (struct net *)arg);
1477 static int fib6_net_init(struct net *net)
1480 struct timer_list *timer;
1483 timer = kzalloc(sizeof(*timer), GFP_KERNEL);
1487 setup_timer(timer, fib6_gc_timer_cb, (unsigned long)net);
1488 net->ipv6.ip6_fib_timer = timer;
1490 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1491 if (!net->ipv6.rt6_stats)
1494 net->ipv6.fib_table_hash =
1495 kzalloc(sizeof(*net->ipv6.fib_table_hash)*FIB_TABLE_HASHSZ,
1497 if (!net->ipv6.fib_table_hash)
1500 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1502 if (!net->ipv6.fib6_main_tbl)
1503 goto out_fib_table_hash;
1505 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1506 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1507 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1508 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1510 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1511 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1513 if (!net->ipv6.fib6_local_tbl)
1514 goto out_fib6_main_tbl;
1515 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1516 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1517 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1518 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1520 fib6_tables_init(net);
1526 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1528 kfree(net->ipv6.fib6_main_tbl);
1531 kfree(net->ipv6.fib_table_hash);
1533 kfree(net->ipv6.rt6_stats);
1539 static void fib6_net_exit(struct net *net)
1541 rt6_ifdown(net, NULL);
1542 del_timer(net->ipv6.ip6_fib_timer);
1543 kfree(net->ipv6.ip6_fib_timer);
1544 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1545 kfree(net->ipv6.fib6_local_tbl);
1547 kfree(net->ipv6.fib6_main_tbl);
1548 kfree(net->ipv6.fib_table_hash);
1549 kfree(net->ipv6.rt6_stats);
1552 static struct pernet_operations fib6_net_ops = {
1553 .init = fib6_net_init,
1554 .exit = fib6_net_exit,
1557 int __init fib6_init(void)
1561 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1562 sizeof(struct fib6_node),
1563 0, SLAB_HWCACHE_ALIGN,
1565 if (!fib6_node_kmem)
1568 ret = register_pernet_subsys(&fib6_net_ops);
1570 goto out_kmem_cache_create;
1572 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib);
1574 goto out_unregister_subsys;
1578 out_unregister_subsys:
1579 unregister_pernet_subsys(&fib6_net_ops);
1580 out_kmem_cache_create:
1581 kmem_cache_destroy(fib6_node_kmem);
1585 void fib6_gc_cleanup(void)
1587 unregister_pernet_subsys(&fib6_net_ops);
1588 kmem_cache_destroy(fib6_node_kmem);