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 struct rt6_statistics rt6_stats;
53 static struct kmem_cache * fib6_node_kmem __read_mostly;
57 #ifdef CONFIG_IPV6_SUBTREES
68 struct fib6_walker_t w;
69 int (*func)(struct rt6_info *, void *arg);
73 static DEFINE_RWLOCK(fib6_walker_lock);
75 #ifdef CONFIG_IPV6_SUBTREES
76 #define FWS_INIT FWS_S
78 #define FWS_INIT FWS_L
81 static void fib6_prune_clones(struct fib6_node *fn, struct rt6_info *rt);
82 static struct rt6_info * fib6_find_prefix(struct fib6_node *fn);
83 static struct fib6_node * fib6_repair_tree(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 timer_list *ip6_fib_timer;
100 static struct fib6_walker_t fib6_walker_list = {
101 .prev = &fib6_walker_list,
102 .next = &fib6_walker_list,
105 #define FOR_WALKERS(w) for ((w)=fib6_walker_list.next; (w) != &fib6_walker_list; (w)=(w)->next)
107 static inline void fib6_walker_link(struct fib6_walker_t *w)
109 write_lock_bh(&fib6_walker_lock);
110 w->next = fib6_walker_list.next;
111 w->prev = &fib6_walker_list;
114 write_unlock_bh(&fib6_walker_lock);
117 static inline void fib6_walker_unlink(struct fib6_walker_t *w)
119 write_lock_bh(&fib6_walker_lock);
120 w->next->prev = w->prev;
121 w->prev->next = w->next;
122 w->prev = w->next = w;
123 write_unlock_bh(&fib6_walker_lock);
125 static __inline__ u32 fib6_new_sernum(void)
134 * Auxiliary address test functions for the radix tree.
136 * These assume a 32bit processor (although it will work on
144 static __inline__ __be32 addr_bit_set(void *token, int fn_bit)
146 __be32 *addr = token;
148 return htonl(1 << ((~fn_bit)&0x1F)) & addr[fn_bit>>5];
151 static __inline__ struct fib6_node * node_alloc(void)
153 struct fib6_node *fn;
155 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
160 static __inline__ void node_free(struct fib6_node * fn)
162 kmem_cache_free(fib6_node_kmem, fn);
165 static __inline__ void rt6_release(struct rt6_info *rt)
167 if (atomic_dec_and_test(&rt->rt6i_ref))
168 dst_free(&rt->u.dst);
171 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
172 #define FIB_TABLE_HASHSZ 256
174 #define FIB_TABLE_HASHSZ 1
177 static void fib6_link_table(struct net *net, struct fib6_table *tb)
182 * Initialize table lock at a single place to give lockdep a key,
183 * tables aren't visible prior to being linked to the list.
185 rwlock_init(&tb->tb6_lock);
187 h = tb->tb6_id & (FIB_TABLE_HASHSZ - 1);
190 * No protection necessary, this is the only list mutatation
191 * operation, tables never disappear once they exist.
193 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
196 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
198 static struct fib6_table *fib6_alloc_table(u32 id)
200 struct fib6_table *table;
202 table = kzalloc(sizeof(*table), GFP_ATOMIC);
205 table->tb6_root.leaf = &ip6_null_entry;
206 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
212 struct fib6_table *fib6_new_table(struct net *net, u32 id)
214 struct fib6_table *tb;
218 tb = fib6_get_table(net, id);
222 tb = fib6_alloc_table(id);
224 fib6_link_table(net, tb);
229 struct fib6_table *fib6_get_table(struct net *net, u32 id)
231 struct fib6_table *tb;
232 struct hlist_head *head;
233 struct hlist_node *node;
238 h = id & (FIB_TABLE_HASHSZ - 1);
240 head = &net->ipv6.fib_table_hash[h];
241 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
242 if (tb->tb6_id == id) {
252 static void fib6_tables_init(struct net *net)
254 fib6_link_table(net, net->ipv6.fib6_main_tbl);
255 fib6_link_table(net, net->ipv6.fib6_local_tbl);
259 struct fib6_table *fib6_new_table(struct net *net, u32 id)
261 return fib6_get_table(net, id);
264 struct fib6_table *fib6_get_table(struct net *net, u32 id)
266 return net->ipv6.fib6_main_tbl;
269 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi *fl,
270 int flags, pol_lookup_t lookup)
272 return (struct dst_entry *) lookup(net->ipv6.fib6_main_tbl, fl, flags);
275 static void fib6_tables_init(struct net *net)
277 fib6_link_table(net, net->ipv6.fib6_main_tbl);
282 static int fib6_dump_node(struct fib6_walker_t *w)
287 for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) {
288 res = rt6_dump_route(rt, w->args);
290 /* Frame is full, suspend walking */
300 static void fib6_dump_end(struct netlink_callback *cb)
302 struct fib6_walker_t *w = (void*)cb->args[2];
308 cb->done = (void*)cb->args[3];
312 static int fib6_dump_done(struct netlink_callback *cb)
315 return cb->done ? cb->done(cb) : 0;
318 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
319 struct netlink_callback *cb)
321 struct fib6_walker_t *w;
324 w = (void *)cb->args[2];
325 w->root = &table->tb6_root;
327 if (cb->args[4] == 0) {
328 read_lock_bh(&table->tb6_lock);
330 read_unlock_bh(&table->tb6_lock);
334 read_lock_bh(&table->tb6_lock);
335 res = fib6_walk_continue(w);
336 read_unlock_bh(&table->tb6_lock);
339 fib6_walker_unlink(w);
342 fib6_walker_unlink(w);
349 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
351 struct net *net = skb->sk->sk_net;
353 unsigned int e = 0, s_e;
354 struct rt6_rtnl_dump_arg arg;
355 struct fib6_walker_t *w;
356 struct fib6_table *tb;
357 struct hlist_node *node;
358 struct hlist_head *head;
364 w = (void *)cb->args[2];
368 * 1. hook callback destructor.
370 cb->args[3] = (long)cb->done;
371 cb->done = fib6_dump_done;
374 * 2. allocate and initialize walker.
376 w = kzalloc(sizeof(*w), GFP_ATOMIC);
379 w->func = fib6_dump_node;
380 cb->args[2] = (long)w;
387 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
389 head = &net->ipv6.fib_table_hash[h];
390 hlist_for_each_entry(tb, node, head, tb6_hlist) {
393 res = fib6_dump_table(tb, skb, cb);
404 res = res < 0 ? res : skb->len;
413 * return the appropriate node for a routing tree "add" operation
414 * by either creating and inserting or by returning an existing
418 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
419 int addrlen, int plen,
422 struct fib6_node *fn, *in, *ln;
423 struct fib6_node *pn = NULL;
427 __u32 sernum = fib6_new_sernum();
429 RT6_TRACE("fib6_add_1\n");
431 /* insert node in tree */
436 key = (struct rt6key *)((u8 *)fn->leaf + offset);
441 if (plen < fn->fn_bit ||
442 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
449 if (plen == fn->fn_bit) {
450 /* clean up an intermediate node */
451 if ((fn->fn_flags & RTN_RTINFO) == 0) {
452 rt6_release(fn->leaf);
456 fn->fn_sernum = sernum;
462 * We have more bits to go
465 /* Try to walk down on tree. */
466 fn->fn_sernum = sernum;
467 dir = addr_bit_set(addr, fn->fn_bit);
469 fn = dir ? fn->right: fn->left;
473 * We walked to the bottom of tree.
474 * Create new leaf node without children.
484 ln->fn_sernum = sernum;
496 * split since we don't have a common prefix anymore or
497 * we have a less significant route.
498 * we've to insert an intermediate node on the list
499 * this new node will point to the one we need to create
505 /* find 1st bit in difference between the 2 addrs.
507 See comment in __ipv6_addr_diff: bit may be an invalid value,
508 but if it is >= plen, the value is ignored in any case.
511 bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
516 * (new leaf node)[ln] (old node)[fn]
522 if (in == NULL || ln == NULL) {
531 * new intermediate node.
533 * be off since that an address that chooses one of
534 * the branches would not match less specific routes
535 * in the other branch
542 atomic_inc(&in->leaf->rt6i_ref);
544 in->fn_sernum = sernum;
546 /* update parent pointer */
557 ln->fn_sernum = sernum;
559 if (addr_bit_set(addr, bit)) {
566 } else { /* plen <= bit */
569 * (new leaf node)[ln]
571 * (old node)[fn] NULL
583 ln->fn_sernum = sernum;
590 if (addr_bit_set(&key->addr, plen))
601 * Insert routing information in a node.
604 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
605 struct nl_info *info)
607 struct rt6_info *iter = NULL;
608 struct rt6_info **ins;
612 for (iter = fn->leaf; iter; iter=iter->u.dst.rt6_next) {
614 * Search for duplicates
617 if (iter->rt6i_metric == rt->rt6i_metric) {
619 * Same priority level
622 if (iter->rt6i_dev == rt->rt6i_dev &&
623 iter->rt6i_idev == rt->rt6i_idev &&
624 ipv6_addr_equal(&iter->rt6i_gateway,
625 &rt->rt6i_gateway)) {
626 if (!(iter->rt6i_flags&RTF_EXPIRES))
628 iter->rt6i_expires = rt->rt6i_expires;
629 if (!(rt->rt6i_flags&RTF_EXPIRES)) {
630 iter->rt6i_flags &= ~RTF_EXPIRES;
631 iter->rt6i_expires = 0;
637 if (iter->rt6i_metric > rt->rt6i_metric)
640 ins = &iter->u.dst.rt6_next;
643 /* Reset round-robin state, if necessary */
644 if (ins == &fn->leaf)
651 rt->u.dst.rt6_next = iter;
654 atomic_inc(&rt->rt6i_ref);
655 inet6_rt_notify(RTM_NEWROUTE, rt, info);
656 rt6_stats.fib_rt_entries++;
658 if ((fn->fn_flags & RTN_RTINFO) == 0) {
659 rt6_stats.fib_route_nodes++;
660 fn->fn_flags |= RTN_RTINFO;
666 static __inline__ void fib6_start_gc(struct rt6_info *rt)
668 if (ip6_fib_timer->expires == 0 &&
669 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE)))
670 mod_timer(ip6_fib_timer, jiffies +
671 init_net.ipv6.sysctl.ip6_rt_gc_interval);
674 void fib6_force_start_gc(void)
676 if (ip6_fib_timer->expires == 0)
677 mod_timer(ip6_fib_timer, jiffies +
678 init_net.ipv6.sysctl.ip6_rt_gc_interval);
682 * Add routing information to the routing tree.
683 * <destination addr>/<source addr>
684 * with source addr info in sub-trees
687 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
689 struct fib6_node *fn, *pn = NULL;
692 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
693 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst));
700 #ifdef CONFIG_IPV6_SUBTREES
701 if (rt->rt6i_src.plen) {
702 struct fib6_node *sn;
704 if (fn->subtree == NULL) {
705 struct fib6_node *sfn;
717 /* Create subtree root node */
722 sfn->leaf = &ip6_null_entry;
723 atomic_inc(&ip6_null_entry.rt6i_ref);
724 sfn->fn_flags = RTN_ROOT;
725 sfn->fn_sernum = fib6_new_sernum();
727 /* Now add the first leaf node to new subtree */
729 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
730 sizeof(struct in6_addr), rt->rt6i_src.plen,
731 offsetof(struct rt6_info, rt6i_src));
734 /* If it is failed, discard just allocated
735 root, and then (in st_failure) stale node
742 /* Now link new subtree to main tree */
746 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
747 sizeof(struct in6_addr), rt->rt6i_src.plen,
748 offsetof(struct rt6_info, rt6i_src));
754 if (fn->leaf == NULL) {
756 atomic_inc(&rt->rt6i_ref);
762 err = fib6_add_rt2node(fn, rt, info);
766 if (!(rt->rt6i_flags&RTF_CACHE))
767 fib6_prune_clones(pn, rt);
772 #ifdef CONFIG_IPV6_SUBTREES
774 * If fib6_add_1 has cleared the old leaf pointer in the
775 * super-tree leaf node we have to find a new one for it.
777 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
778 pn->leaf = fib6_find_prefix(pn);
781 BUG_TRAP(pn->leaf != NULL);
782 pn->leaf = &ip6_null_entry;
785 atomic_inc(&pn->leaf->rt6i_ref);
788 dst_free(&rt->u.dst);
792 #ifdef CONFIG_IPV6_SUBTREES
793 /* Subtree creation failed, probably main tree node
794 is orphan. If it is, shoot it.
797 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
798 fib6_repair_tree(fn);
799 dst_free(&rt->u.dst);
805 * Routing tree lookup
810 int offset; /* key offset on rt6_info */
811 struct in6_addr *addr; /* search key */
814 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
815 struct lookup_args *args)
817 struct fib6_node *fn;
820 if (unlikely(args->offset == 0))
830 struct fib6_node *next;
832 dir = addr_bit_set(args->addr, fn->fn_bit);
834 next = dir ? fn->right : fn->left;
845 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
848 key = (struct rt6key *) ((u8 *) fn->leaf +
851 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
852 #ifdef CONFIG_IPV6_SUBTREES
854 fn = fib6_lookup_1(fn->subtree, args + 1);
856 if (!fn || fn->fn_flags & RTN_RTINFO)
861 if (fn->fn_flags & RTN_ROOT)
870 struct fib6_node * fib6_lookup(struct fib6_node *root, struct in6_addr *daddr,
871 struct in6_addr *saddr)
873 struct fib6_node *fn;
874 struct lookup_args args[] = {
876 .offset = offsetof(struct rt6_info, rt6i_dst),
879 #ifdef CONFIG_IPV6_SUBTREES
881 .offset = offsetof(struct rt6_info, rt6i_src),
886 .offset = 0, /* sentinel */
890 fn = fib6_lookup_1(root, daddr ? args : args + 1);
892 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT)
899 * Get node with specified destination prefix (and source prefix,
900 * if subtrees are used)
904 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
905 struct in6_addr *addr,
906 int plen, int offset)
908 struct fib6_node *fn;
910 for (fn = root; fn ; ) {
911 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
916 if (plen < fn->fn_bit ||
917 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
920 if (plen == fn->fn_bit)
924 * We have more bits to go
926 if (addr_bit_set(addr, fn->fn_bit))
934 struct fib6_node * fib6_locate(struct fib6_node *root,
935 struct in6_addr *daddr, int dst_len,
936 struct in6_addr *saddr, int src_len)
938 struct fib6_node *fn;
940 fn = fib6_locate_1(root, daddr, dst_len,
941 offsetof(struct rt6_info, rt6i_dst));
943 #ifdef CONFIG_IPV6_SUBTREES
945 BUG_TRAP(saddr!=NULL);
946 if (fn && fn->subtree)
947 fn = fib6_locate_1(fn->subtree, saddr, src_len,
948 offsetof(struct rt6_info, rt6i_src));
952 if (fn && fn->fn_flags&RTN_RTINFO)
964 static struct rt6_info * fib6_find_prefix(struct fib6_node *fn)
966 if (fn->fn_flags&RTN_ROOT)
967 return &ip6_null_entry;
971 return fn->left->leaf;
974 return fn->right->leaf;
976 fn = FIB6_SUBTREE(fn);
982 * Called to trim the tree of intermediate nodes when possible. "fn"
983 * is the node we want to try and remove.
986 static struct fib6_node * fib6_repair_tree(struct fib6_node *fn)
990 struct fib6_node *child, *pn;
991 struct fib6_walker_t *w;
995 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
998 BUG_TRAP(!(fn->fn_flags&RTN_RTINFO));
999 BUG_TRAP(!(fn->fn_flags&RTN_TL_ROOT));
1000 BUG_TRAP(fn->leaf==NULL);
1004 if (fn->right) child = fn->right, children |= 1;
1005 if (fn->left) child = fn->left, children |= 2;
1007 if (children == 3 || FIB6_SUBTREE(fn)
1008 #ifdef CONFIG_IPV6_SUBTREES
1009 /* Subtree root (i.e. fn) may have one child */
1010 || (children && fn->fn_flags&RTN_ROOT)
1013 fn->leaf = fib6_find_prefix(fn);
1015 if (fn->leaf==NULL) {
1017 fn->leaf = &ip6_null_entry;
1020 atomic_inc(&fn->leaf->rt6i_ref);
1025 #ifdef CONFIG_IPV6_SUBTREES
1026 if (FIB6_SUBTREE(pn) == fn) {
1027 BUG_TRAP(fn->fn_flags&RTN_ROOT);
1028 FIB6_SUBTREE(pn) = NULL;
1031 BUG_TRAP(!(fn->fn_flags&RTN_ROOT));
1033 if (pn->right == fn) pn->right = child;
1034 else if (pn->left == fn) pn->left = child;
1041 #ifdef CONFIG_IPV6_SUBTREES
1045 read_lock(&fib6_walker_lock);
1047 if (child == NULL) {
1048 if (w->root == fn) {
1049 w->root = w->node = NULL;
1050 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1051 } else if (w->node == fn) {
1052 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1057 if (w->root == fn) {
1059 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1061 if (w->node == fn) {
1064 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1065 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1067 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1068 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1073 read_unlock(&fib6_walker_lock);
1076 if (pn->fn_flags&RTN_RTINFO || FIB6_SUBTREE(pn))
1079 rt6_release(pn->leaf);
1085 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1086 struct nl_info *info)
1088 struct fib6_walker_t *w;
1089 struct rt6_info *rt = *rtp;
1091 RT6_TRACE("fib6_del_route\n");
1094 *rtp = rt->u.dst.rt6_next;
1095 rt->rt6i_node = NULL;
1096 rt6_stats.fib_rt_entries--;
1097 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 rt6_stats.fib_route_nodes--;
1121 fn = fib6_repair_tree(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(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 fib6_node *fn = rt->rt6i_node;
1150 struct rt6_info **rtp;
1153 if (rt->u.dst.obsolete>0) {
1158 if (fn == NULL || rt == &ip6_null_entry)
1161 BUG_TRAP(fn->fn_flags&RTN_RTINFO);
1163 if (!(rt->rt6i_flags&RTF_CACHE)) {
1164 struct fib6_node *pn = fn;
1165 #ifdef CONFIG_IPV6_SUBTREES
1166 /* clones of this route might be in another subtree */
1167 if (rt->rt6i_src.plen) {
1168 while (!(pn->fn_flags&RTN_ROOT))
1173 fib6_prune_clones(pn, rt);
1177 * Walk the leaf entries looking for ourself
1180 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->u.dst.rt6_next) {
1182 fib6_del_route(fn, rtp, info);
1190 * Tree traversal function.
1192 * Certainly, it is not interrupt safe.
1193 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1194 * It means, that we can modify tree during walking
1195 * and use this function for garbage collection, clone pruning,
1196 * cleaning tree when a device goes down etc. etc.
1198 * It guarantees that every node will be traversed,
1199 * and that it will be traversed only once.
1201 * Callback function w->func may return:
1202 * 0 -> continue walking.
1203 * positive value -> walking is suspended (used by tree dumps,
1204 * and probably by gc, if it will be split to several slices)
1205 * negative value -> terminate walking.
1207 * The function itself returns:
1208 * 0 -> walk is complete.
1209 * >0 -> walk is incomplete (i.e. suspended)
1210 * <0 -> walk is terminated by an error.
1213 static int fib6_walk_continue(struct fib6_walker_t *w)
1215 struct fib6_node *fn, *pn;
1222 if (w->prune && fn != w->root &&
1223 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) {
1228 #ifdef CONFIG_IPV6_SUBTREES
1230 if (FIB6_SUBTREE(fn)) {
1231 w->node = FIB6_SUBTREE(fn);
1239 w->state = FWS_INIT;
1245 w->node = fn->right;
1246 w->state = FWS_INIT;
1252 if (w->leaf && fn->fn_flags&RTN_RTINFO) {
1253 int err = w->func(w);
1264 #ifdef CONFIG_IPV6_SUBTREES
1265 if (FIB6_SUBTREE(pn) == fn) {
1266 BUG_TRAP(fn->fn_flags&RTN_ROOT);
1271 if (pn->left == fn) {
1275 if (pn->right == fn) {
1277 w->leaf = w->node->leaf;
1287 static int fib6_walk(struct fib6_walker_t *w)
1291 w->state = FWS_INIT;
1294 fib6_walker_link(w);
1295 res = fib6_walk_continue(w);
1297 fib6_walker_unlink(w);
1301 static int fib6_clean_node(struct fib6_walker_t *w)
1303 struct nl_info info = {
1304 .nl_net = &init_net,
1307 struct rt6_info *rt;
1308 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1310 for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) {
1311 res = c->func(rt, c->arg);
1314 res = fib6_del(rt, &info);
1317 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res);
1330 * Convenient frontend to tree walker.
1332 * func is called on each route.
1333 * It may return -1 -> delete this route.
1334 * 0 -> continue walking
1336 * prune==1 -> only immediate children of node (certainly,
1337 * ignoring pure split nodes) will be scanned.
1340 static void fib6_clean_tree(struct fib6_node *root,
1341 int (*func)(struct rt6_info *, void *arg),
1342 int prune, void *arg)
1344 struct fib6_cleaner_t c;
1347 c.w.func = fib6_clean_node;
1355 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1356 int prune, void *arg)
1358 struct fib6_table *table;
1359 struct hlist_node *node;
1360 struct hlist_head *head;
1364 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
1365 head = &net->ipv6.fib_table_hash[h];
1366 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1367 write_lock_bh(&table->tb6_lock);
1368 fib6_clean_tree(&table->tb6_root, func, prune, arg);
1369 write_unlock_bh(&table->tb6_lock);
1375 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1377 if (rt->rt6i_flags & RTF_CACHE) {
1378 RT6_TRACE("pruning clone %p\n", rt);
1385 static void fib6_prune_clones(struct fib6_node *fn, struct rt6_info *rt)
1387 fib6_clean_tree(fn, fib6_prune_clone, 1, rt);
1391 * Garbage collection
1394 static struct fib6_gc_args
1400 static int fib6_age(struct rt6_info *rt, void *arg)
1402 unsigned long now = jiffies;
1405 * check addrconf expiration here.
1406 * Routes are expired even if they are in use.
1408 * Also age clones. Note, that clones are aged out
1409 * only if they are not in use now.
1412 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) {
1413 if (time_after(now, rt->rt6i_expires)) {
1414 RT6_TRACE("expiring %p\n", rt);
1418 } else if (rt->rt6i_flags & RTF_CACHE) {
1419 if (atomic_read(&rt->u.dst.__refcnt) == 0 &&
1420 time_after_eq(now, rt->u.dst.lastuse + gc_args.timeout)) {
1421 RT6_TRACE("aging clone %p\n", rt);
1423 } else if ((rt->rt6i_flags & RTF_GATEWAY) &&
1424 (!(rt->rt6i_nexthop->flags & NTF_ROUTER))) {
1425 RT6_TRACE("purging route %p via non-router but gateway\n",
1435 static DEFINE_SPINLOCK(fib6_gc_lock);
1437 void fib6_run_gc(unsigned long expires, struct net *net)
1439 if (expires != ~0UL) {
1440 spin_lock_bh(&fib6_gc_lock);
1441 gc_args.timeout = expires ? (int)expires :
1442 net->ipv6.sysctl.ip6_rt_gc_interval;
1445 if (!spin_trylock(&fib6_gc_lock)) {
1446 mod_timer(ip6_fib_timer, jiffies + HZ);
1450 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1454 icmp6_dst_gc(&gc_args.more);
1456 fib6_clean_all(net, fib6_age, 0, NULL);
1459 mod_timer(ip6_fib_timer, jiffies +
1460 net->ipv6.sysctl.ip6_rt_gc_interval);
1462 del_timer(ip6_fib_timer);
1463 ip6_fib_timer->expires = 0;
1465 spin_unlock_bh(&fib6_gc_lock);
1468 static void fib6_gc_timer_cb(unsigned long arg)
1470 fib6_run_gc(0, (struct net *)arg);
1473 static int fib6_net_init(struct net *net)
1478 net->ipv6.fib_table_hash =
1479 kzalloc(sizeof(*net->ipv6.fib_table_hash)*FIB_TABLE_HASHSZ,
1481 if (!net->ipv6.fib_table_hash)
1484 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1486 if (!net->ipv6.fib6_main_tbl)
1487 goto out_fib_table_hash;
1489 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1490 net->ipv6.fib6_main_tbl->tb6_root.leaf = &ip6_null_entry;
1491 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1492 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1494 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1495 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1497 if (!net->ipv6.fib6_local_tbl)
1498 goto out_fib6_main_tbl;
1499 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1500 net->ipv6.fib6_local_tbl->tb6_root.leaf = &ip6_null_entry;
1501 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1502 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1504 fib6_tables_init(net);
1510 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1512 kfree(net->ipv6.fib6_main_tbl);
1515 kfree(net->ipv6.fib_table_hash);
1519 static void fib6_net_exit(struct net *net)
1521 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1522 kfree(net->ipv6.fib6_local_tbl);
1524 kfree(net->ipv6.fib6_main_tbl);
1525 kfree(net->ipv6.fib_table_hash);
1528 static struct pernet_operations fib6_net_ops = {
1529 .init = fib6_net_init,
1530 .exit = fib6_net_exit,
1533 int __init fib6_init(void)
1536 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1537 sizeof(struct fib6_node),
1538 0, SLAB_HWCACHE_ALIGN,
1540 if (!fib6_node_kmem)
1544 ip6_fib_timer = kzalloc(sizeof(*ip6_fib_timer), GFP_KERNEL);
1546 goto out_kmem_cache_create;
1548 setup_timer(ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)&init_net);
1550 ret = register_pernet_subsys(&fib6_net_ops);
1554 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib);
1556 goto out_unregister_subsys;
1560 out_unregister_subsys:
1561 unregister_pernet_subsys(&fib6_net_ops);
1563 kfree(ip6_fib_timer);
1564 out_kmem_cache_create:
1565 kmem_cache_destroy(fib6_node_kmem);
1569 void fib6_gc_cleanup(void)
1571 del_timer(ip6_fib_timer);
1572 kfree(ip6_fib_timer);
1573 unregister_pernet_subsys(&fib6_net_ops);
1574 kmem_cache_destroy(fib6_node_kmem);