2 * Linux INET6 implementation
3 * Forwarding Information Database
6 * Pedro Roque <pedro_m@yahoo.com>
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
22 #include <linux/config.h>
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>
32 #include <linux/proc_fs.h>
36 #include <net/ndisc.h>
37 #include <net/addrconf.h>
39 #include <net/ip6_fib.h>
40 #include <net/ip6_route.h>
43 #undef CONFIG_IPV6_SUBTREES
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 kmem_cache_t * fib6_node_kmem;
57 #ifdef CONFIG_IPV6_SUBTREES
68 struct fib6_walker_t w;
69 int (*func)(struct rt6_info *, void *arg);
73 rwlock_t fib6_walker_lock = RW_LOCK_UNLOCKED;
76 #ifdef CONFIG_IPV6_SUBTREES
77 #define FWS_INIT FWS_S
78 #define SUBTREE(fn) ((fn)->subtree)
80 #define FWS_INIT FWS_L
81 #define SUBTREE(fn) NULL
84 static void fib6_prune_clones(struct fib6_node *fn, struct rt6_info *rt);
85 static struct fib6_node * fib6_repair_tree(struct fib6_node *fn);
88 * A routing update causes an increase of the serial number on the
89 * afected 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 = 0;
96 static struct timer_list ip6_fib_timer = { function: fib6_run_gc };
98 struct fib6_walker_t fib6_walker_list = {
99 &fib6_walker_list, &fib6_walker_list,
102 #define FOR_WALKERS(w) for ((w)=fib6_walker_list.next; (w) != &fib6_walker_list; (w)=(w)->next)
104 static __inline__ u32 fib6_new_sernum(void)
113 * Auxiliary address test functions for the radix tree.
115 * These assume a 32bit processor (although it will work on
120 * compare "prefix length" bits of an address
123 static __inline__ int addr_match(void *token1, void *token2, int prefixlen)
130 pdw = prefixlen >> 5; /* num of whole __u32 in prefix */
131 pbi = prefixlen & 0x1f; /* num of bits in incomplete u32 in prefix */
134 if (memcmp(a1, a2, pdw << 2))
140 mask = htonl((0xffffffff) << (32 - pbi));
142 if ((a1[pdw] ^ a2[pdw]) & mask)
153 static __inline__ int addr_bit_set(void *token, int fn_bit)
157 return htonl(1 << ((~fn_bit)&0x1F)) & addr[fn_bit>>5];
161 * find the first different bit between two addresses
162 * length of address must be a multiple of 32bits
165 static __inline__ int addr_diff(void *token1, void *token2, int addrlen)
173 for (i = 0; i < addrlen; i++) {
183 while ((xb & (1 << j)) == 0)
186 return (i * 32 + 31 - j);
191 * we should *never* get to this point since that
192 * would mean the addrs are equal
194 * However, we do get to it 8) And exacly, when
195 * addresses are equal 8)
197 * ip route add 1111::/128 via ...
198 * ip route add 1111::/64 via ...
201 * Ideally, this function should stop comparison
202 * at prefix length. It does not, but it is still OK,
203 * if returned value is greater than prefix length.
210 static __inline__ struct fib6_node * node_alloc(void)
212 struct fib6_node *fn;
214 if ((fn = kmem_cache_alloc(fib6_node_kmem, SLAB_ATOMIC)) != NULL)
215 memset(fn, 0, sizeof(struct fib6_node));
220 static __inline__ void node_free(struct fib6_node * fn)
222 kmem_cache_free(fib6_node_kmem, fn);
225 static __inline__ void rt6_release(struct rt6_info *rt)
227 if (atomic_dec_and_test(&rt->rt6i_ref))
228 dst_free(&rt->u.dst);
235 * return the apropriate node for a routing tree "add" operation
236 * by either creating and inserting or by returning an existing
240 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
241 int addrlen, int plen,
244 struct fib6_node *fn, *in, *ln;
245 struct fib6_node *pn = NULL;
249 __u32 sernum = fib6_new_sernum();
251 RT6_TRACE("fib6_add_1\n");
253 /* insert node in tree */
258 key = (struct rt6key *)((u8 *)fn->leaf + offset);
263 if (plen < fn->fn_bit ||
264 !addr_match(&key->addr, addr, fn->fn_bit))
271 if (plen == fn->fn_bit) {
272 /* clean up an intermediate node */
273 if ((fn->fn_flags & RTN_RTINFO) == 0) {
274 rt6_release(fn->leaf);
278 fn->fn_sernum = sernum;
284 * We have more bits to go
287 /* Try to walk down on tree. */
288 fn->fn_sernum = sernum;
289 dir = addr_bit_set(addr, fn->fn_bit);
291 fn = dir ? fn->right: fn->left;
295 * We walked to the bottom of tree.
296 * Create new leaf node without children.
306 ln->fn_sernum = sernum;
318 * split since we don't have a common prefix anymore or
319 * we have a less significant route.
320 * we've to insert an intermediate node on the list
321 * this new node will point to the one we need to create
327 /* find 1st bit in difference between the 2 addrs.
329 See comment in addr_diff: bit may be an invalid value,
330 but if it is >= plen, the value is ignored in any case.
333 bit = addr_diff(addr, &key->addr, addrlen);
338 * (new leaf node)[ln] (old node)[fn]
344 if (in == NULL || ln == NULL) {
353 * new intermediate node.
355 * be off since that an address that chooses one of
356 * the branches would not match less specific routes
357 * in the other branch
364 atomic_inc(&in->leaf->rt6i_ref);
366 in->fn_sernum = sernum;
368 /* update parent pointer */
379 ln->fn_sernum = sernum;
381 if (addr_bit_set(addr, bit)) {
388 } else { /* plen <= bit */
391 * (new leaf node)[ln]
393 * (old node)[fn] NULL
405 ln->fn_sernum = sernum;
412 if (addr_bit_set(&key->addr, plen))
423 * Insert routing information in a node.
426 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
427 struct nlmsghdr *nlh, struct netlink_skb_parms *req)
429 struct rt6_info *iter = NULL;
430 struct rt6_info **ins;
434 if (fn->fn_flags&RTN_TL_ROOT &&
435 fn->leaf == &ip6_null_entry &&
436 !(rt->rt6i_flags & (RTF_DEFAULT | RTF_ADDRCONF | RTF_ALLONLINK)) ){
438 * The top fib of ip6 routing table includes ip6_null_entry.
445 for (iter = fn->leaf; iter; iter=iter->u.next) {
447 * Search for duplicates
450 if (iter->rt6i_metric == rt->rt6i_metric) {
452 * Same priority level
455 if ((iter->rt6i_dev == rt->rt6i_dev) &&
456 (iter->rt6i_flowr == rt->rt6i_flowr) &&
457 (ipv6_addr_cmp(&iter->rt6i_gateway,
458 &rt->rt6i_gateway) == 0)) {
459 if (!(iter->rt6i_flags&RTF_EXPIRES))
461 iter->rt6i_expires = rt->rt6i_expires;
462 if (!(rt->rt6i_flags&RTF_EXPIRES)) {
463 iter->rt6i_flags &= ~RTF_EXPIRES;
464 iter->rt6i_expires = 0;
470 if (iter->rt6i_metric > rt->rt6i_metric)
484 atomic_inc(&rt->rt6i_ref);
485 inet6_rt_notify(RTM_NEWROUTE, rt, nlh, req);
486 rt6_stats.fib_rt_entries++;
488 if ((fn->fn_flags & RTN_RTINFO) == 0) {
489 rt6_stats.fib_route_nodes++;
490 fn->fn_flags |= RTN_RTINFO;
496 static __inline__ void fib6_start_gc(struct rt6_info *rt)
498 if (ip6_fib_timer.expires == 0 &&
499 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE)))
500 mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval);
504 * Add routing information to the routing tree.
505 * <destination addr>/<source addr>
506 * with source addr info in sub-trees
509 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nlmsghdr *nlh,
510 struct netlink_skb_parms *req)
512 struct fib6_node *fn;
515 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
516 rt->rt6i_dst.plen, (u8*) &rt->rt6i_dst - (u8*) rt);
521 #ifdef CONFIG_IPV6_SUBTREES
522 if (rt->rt6i_src.plen) {
523 struct fib6_node *sn;
525 if (fn->subtree == NULL) {
526 struct fib6_node *sfn;
538 /* Create subtree root node */
543 sfn->leaf = &ip6_null_entry;
544 atomic_inc(&ip6_null_entry.rt6i_ref);
545 sfn->fn_flags = RTN_ROOT;
546 sfn->fn_sernum = fib6_new_sernum();
548 /* Now add the first leaf node to new subtree */
550 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
551 sizeof(struct in6_addr), rt->rt6i_src.plen,
552 (u8*) &rt->rt6i_src - (u8*) rt);
555 /* If it is failed, discard just allocated
556 root, and then (in st_failure) stale node
563 /* Now link new subtree to main tree */
566 if (fn->leaf == NULL) {
568 atomic_inc(&rt->rt6i_ref);
571 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
572 sizeof(struct in6_addr), rt->rt6i_src.plen,
573 (u8*) &rt->rt6i_src - (u8*) rt);
583 err = fib6_add_rt2node(fn, rt, nlh, req);
587 if (!(rt->rt6i_flags&RTF_CACHE))
588 fib6_prune_clones(fn, rt);
593 dst_free(&rt->u.dst);
596 #ifdef CONFIG_IPV6_SUBTREES
597 /* Subtree creation failed, probably main tree node
598 is orphan. If it is, shoot it.
601 if (fn && !(fn->fn_flags&RTN_RTINFO|RTN_ROOT))
603 dst_free(&rt->u.dst);
609 * Routing tree lookup
614 int offset; /* key offset on rt6_info */
615 struct in6_addr *addr; /* search key */
618 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
619 struct lookup_args *args)
621 struct fib6_node *fn;
631 struct fib6_node *next;
633 dir = addr_bit_set(args->addr, fn->fn_bit);
635 next = dir ? fn->right : fn->left;
645 while ((fn->fn_flags & RTN_ROOT) == 0) {
646 #ifdef CONFIG_IPV6_SUBTREES
648 struct fib6_node *st;
649 struct lookup_args *narg;
654 st = fib6_lookup_1(fn->subtree, narg);
656 if (st && !(st->fn_flags & RTN_ROOT))
662 if (fn->fn_flags & RTN_RTINFO) {
665 key = (struct rt6key *) ((u8 *) fn->leaf +
668 if (addr_match(&key->addr, args->addr, key->plen))
678 struct fib6_node * fib6_lookup(struct fib6_node *root, struct in6_addr *daddr,
679 struct in6_addr *saddr)
681 struct lookup_args args[2];
682 struct rt6_info *rt = NULL;
683 struct fib6_node *fn;
685 args[0].offset = (u8*) &rt->rt6i_dst - (u8*) rt;
686 args[0].addr = daddr;
688 #ifdef CONFIG_IPV6_SUBTREES
689 args[1].offset = (u8*) &rt->rt6i_src - (u8*) rt;
690 args[1].addr = saddr;
693 fn = fib6_lookup_1(root, args);
695 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT)
702 * Get node with sepciafied destination prefix (and source prefix,
703 * if subtrees are used)
707 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
708 struct in6_addr *addr,
709 int plen, int offset)
711 struct fib6_node *fn;
713 for (fn = root; fn ; ) {
714 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
719 if (plen < fn->fn_bit ||
720 !addr_match(&key->addr, addr, fn->fn_bit))
723 if (plen == fn->fn_bit)
727 * We have more bits to go
729 if (addr_bit_set(addr, fn->fn_bit))
737 struct fib6_node * fib6_locate(struct fib6_node *root,
738 struct in6_addr *daddr, int dst_len,
739 struct in6_addr *saddr, int src_len)
741 struct rt6_info *rt = NULL;
742 struct fib6_node *fn;
744 fn = fib6_locate_1(root, daddr, dst_len,
745 (u8*) &rt->rt6i_dst - (u8*) rt);
747 #ifdef CONFIG_IPV6_SUBTREES
749 BUG_TRAP(saddr!=NULL);
753 fn = fib6_locate_1(fn, saddr, src_len,
754 (u8*) &rt->rt6i_src - (u8*) rt);
758 if (fn && fn->fn_flags&RTN_RTINFO)
770 static struct rt6_info * fib6_find_prefix(struct fib6_node *fn)
772 if (fn->fn_flags&RTN_ROOT)
773 return &ip6_null_entry;
777 return fn->left->leaf;
780 return fn->right->leaf;
788 * Called to trim the tree of intermediate nodes when possible. "fn"
789 * is the node we want to try and remove.
792 static struct fib6_node * fib6_repair_tree(struct fib6_node *fn)
796 struct fib6_node *child, *pn;
797 struct fib6_walker_t *w;
801 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
804 BUG_TRAP(!(fn->fn_flags&RTN_RTINFO));
805 BUG_TRAP(!(fn->fn_flags&RTN_TL_ROOT));
806 BUG_TRAP(fn->leaf==NULL);
810 if (fn->right) child = fn->right, children |= 1;
811 if (fn->left) child = fn->left, children |= 2;
813 if (children == 3 || SUBTREE(fn)
814 #ifdef CONFIG_IPV6_SUBTREES
815 /* Subtree root (i.e. fn) may have one child */
816 || (children && fn->fn_flags&RTN_ROOT)
819 fn->leaf = fib6_find_prefix(fn);
821 if (fn->leaf==NULL) {
823 fn->leaf = &ip6_null_entry;
826 atomic_inc(&fn->leaf->rt6i_ref);
831 #ifdef CONFIG_IPV6_SUBTREES
832 if (SUBTREE(pn) == fn) {
833 BUG_TRAP(fn->fn_flags&RTN_ROOT);
837 BUG_TRAP(!(fn->fn_flags&RTN_ROOT));
839 if (pn->right == fn) pn->right = child;
840 else if (pn->left == fn) pn->left = child;
847 #ifdef CONFIG_IPV6_SUBTREES
851 read_lock(&fib6_walker_lock);
855 w->root = w->node = NULL;
856 RT6_TRACE("W %p adjusted by delroot 1\n", w);
857 } else if (w->node == fn) {
858 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
865 RT6_TRACE("W %p adjusted by delroot 2\n", w);
870 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
871 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
873 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
874 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
879 read_unlock(&fib6_walker_lock);
882 if (pn->fn_flags&RTN_RTINFO || SUBTREE(pn))
885 rt6_release(pn->leaf);
891 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
892 struct nlmsghdr *nlh, struct netlink_skb_parms *req)
894 struct fib6_walker_t *w;
895 struct rt6_info *rt = *rtp;
897 RT6_TRACE("fib6_del_route\n");
901 rt->rt6i_node = NULL;
902 rt6_stats.fib_rt_entries--;
905 read_lock(&fib6_walker_lock);
907 if (w->state == FWS_C && w->leaf == rt) {
908 RT6_TRACE("walker %p adjusted by delroute\n", w);
909 w->leaf = rt->u.next;
914 read_unlock(&fib6_walker_lock);
918 if (fn->leaf == NULL && fn->fn_flags&RTN_TL_ROOT)
919 fn->leaf = &ip6_null_entry;
921 /* If it was last route, expunge its radix tree node */
922 if (fn->leaf == NULL) {
923 fn->fn_flags &= ~RTN_RTINFO;
924 rt6_stats.fib_route_nodes--;
925 fn = fib6_repair_tree(fn);
928 if (atomic_read(&rt->rt6i_ref) != 1) {
929 /* This route is used as dummy address holder in some split
930 * nodes. It is not leaked, but it still holds other resources,
931 * which must be released in time. So, scan ascendant nodes
932 * and replace dummy references to this route with references
933 * to still alive ones.
936 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) {
937 fn->leaf = fib6_find_prefix(fn);
938 atomic_inc(&fn->leaf->rt6i_ref);
943 /* No more references are possiible at this point. */
944 if (atomic_read(&rt->rt6i_ref) != 1) BUG();
947 inet6_rt_notify(RTM_DELROUTE, rt, nlh, req);
951 int fib6_del(struct rt6_info *rt, struct nlmsghdr *nlh, struct netlink_skb_parms *req)
953 struct fib6_node *fn = rt->rt6i_node;
954 struct rt6_info **rtp;
957 if (rt->u.dst.obsolete>0) {
962 if (fn == NULL || rt == &ip6_null_entry)
965 BUG_TRAP(fn->fn_flags&RTN_RTINFO);
967 if (!(rt->rt6i_flags&RTF_CACHE))
968 fib6_prune_clones(fn, rt);
971 * Walk the leaf entries looking for ourself
974 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->u.next) {
976 fib6_del_route(fn, rtp, nlh, req);
984 * Tree traversal function.
986 * Certainly, it is not interrupt safe.
987 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
988 * It means, that we can modify tree during walking
989 * and use this function for garbage collection, clone pruning,
990 * cleaning tree when a device goes down etc. etc.
992 * It guarantees that every node will be traversed,
993 * and that it will be traversed only once.
995 * Callback function w->func may return:
996 * 0 -> continue walking.
997 * positive value -> walking is suspended (used by tree dumps,
998 * and probably by gc, if it will be split to several slices)
999 * negative value -> terminate walking.
1001 * The function itself returns:
1002 * 0 -> walk is complete.
1003 * >0 -> walk is incomplete (i.e. suspended)
1004 * <0 -> walk is terminated by an error.
1007 int fib6_walk_continue(struct fib6_walker_t *w)
1009 struct fib6_node *fn, *pn;
1016 if (w->prune && fn != w->root &&
1017 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) {
1022 #ifdef CONFIG_IPV6_SUBTREES
1025 w->node = SUBTREE(fn);
1033 w->state = FWS_INIT;
1039 w->node = fn->right;
1040 w->state = FWS_INIT;
1046 if (w->leaf && fn->fn_flags&RTN_RTINFO) {
1047 int err = w->func(w);
1058 #ifdef CONFIG_IPV6_SUBTREES
1059 if (SUBTREE(pn) == fn) {
1060 BUG_TRAP(fn->fn_flags&RTN_ROOT);
1065 if (pn->left == fn) {
1069 if (pn->right == fn) {
1071 w->leaf = w->node->leaf;
1081 int fib6_walk(struct fib6_walker_t *w)
1085 w->state = FWS_INIT;
1088 fib6_walker_link(w);
1089 res = fib6_walk_continue(w);
1091 fib6_walker_unlink(w);
1095 static int fib6_clean_node(struct fib6_walker_t *w)
1098 struct rt6_info *rt;
1099 struct fib6_cleaner_t *c = (struct fib6_cleaner_t*)w;
1101 for (rt = w->leaf; rt; rt = rt->u.next) {
1102 res = c->func(rt, c->arg);
1105 res = fib6_del(rt, NULL, NULL);
1108 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res);
1121 * Convenient frontend to tree walker.
1123 * func is called on each route.
1124 * It may return -1 -> delete this route.
1125 * 0 -> continue walking
1127 * prune==1 -> only immediate children of node (certainly,
1128 * ignoring pure split nodes) will be scanned.
1131 void fib6_clean_tree(struct fib6_node *root,
1132 int (*func)(struct rt6_info *, void *arg),
1133 int prune, void *arg)
1135 struct fib6_cleaner_t c;
1138 c.w.func = fib6_clean_node;
1146 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1148 if (rt->rt6i_flags & RTF_CACHE) {
1149 RT6_TRACE("pruning clone %p\n", rt);
1156 static void fib6_prune_clones(struct fib6_node *fn, struct rt6_info *rt)
1158 fib6_clean_tree(fn, fib6_prune_clone, 1, rt);
1162 * Garbage collection
1165 static struct fib6_gc_args
1171 static int fib6_age(struct rt6_info *rt, void *arg)
1173 unsigned long now = jiffies;
1175 /* Age clones. Note, that clones are aged out
1176 only if they are not in use now.
1180 * check addrconf expiration here.
1181 * They are expired even if they are in use.
1184 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) {
1185 if (time_after(now, rt->rt6i_expires)) {
1186 RT6_TRACE("expiring %p\n", rt);
1190 } else if (rt->rt6i_flags & RTF_CACHE) {
1191 if (atomic_read(&rt->u.dst.__refcnt) == 0 &&
1192 time_after_eq(now, rt->u.dst.lastuse + gc_args.timeout)) {
1193 RT6_TRACE("aging clone %p\n", rt);
1202 static spinlock_t fib6_gc_lock = SPIN_LOCK_UNLOCKED;
1204 void fib6_run_gc(unsigned long dummy)
1206 if (dummy != ~0UL) {
1207 spin_lock_bh(&fib6_gc_lock);
1208 gc_args.timeout = (int)dummy;
1211 if (!spin_trylock(&fib6_gc_lock)) {
1212 mod_timer(&ip6_fib_timer, jiffies + HZ);
1216 gc_args.timeout = ip6_rt_gc_interval;
1221 write_lock_bh(&rt6_lock);
1222 fib6_clean_tree(&ip6_routing_table, fib6_age, 0, NULL);
1223 write_unlock_bh(&rt6_lock);
1226 mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval);
1228 del_timer(&ip6_fib_timer);
1229 ip6_fib_timer.expires = 0;
1231 spin_unlock_bh(&fib6_gc_lock);
1234 void __init fib6_init(void)
1236 if (!fib6_node_kmem)
1237 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1238 sizeof(struct fib6_node),
1239 0, SLAB_HWCACHE_ALIGN,
1244 void fib6_gc_cleanup(void)
1246 del_timer(&ip6_fib_timer);