Merge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux
[zen-stable.git] / net / ipv6 / ip6_fib.c
blob93718f3db79b3bcacd7bef609978175c68211db3
1 /*
2 * Linux INET6 implementation
3 * Forwarding Information Database
5 * Authors:
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.
15 * Changes:
16 * Yuji SEKIYA @USAGI: Support default route on router node;
17 * remove ip6_null_entry from the top of
18 * routing table.
19 * Ville Nuorvala: Fixed routing subtrees.
21 #include <linux/errno.h>
22 #include <linux/types.h>
23 #include <linux/net.h>
24 #include <linux/route.h>
25 #include <linux/netdevice.h>
26 #include <linux/in6.h>
27 #include <linux/init.h>
28 #include <linux/list.h>
29 #include <linux/slab.h>
31 #include <net/ipv6.h>
32 #include <net/ndisc.h>
33 #include <net/addrconf.h>
35 #include <net/ip6_fib.h>
36 #include <net/ip6_route.h>
38 #define RT6_DEBUG 2
40 #if RT6_DEBUG >= 3
41 #define RT6_TRACE(x...) printk(KERN_DEBUG x)
42 #else
43 #define RT6_TRACE(x...) do { ; } while (0)
44 #endif
46 static struct kmem_cache * fib6_node_kmem __read_mostly;
48 enum fib_walk_state_t
50 #ifdef CONFIG_IPV6_SUBTREES
51 FWS_S,
52 #endif
53 FWS_L,
54 FWS_R,
55 FWS_C,
56 FWS_U
59 struct fib6_cleaner_t
61 struct fib6_walker_t w;
62 struct net *net;
63 int (*func)(struct rt6_info *, void *arg);
64 void *arg;
67 static DEFINE_RWLOCK(fib6_walker_lock);
69 #ifdef CONFIG_IPV6_SUBTREES
70 #define FWS_INIT FWS_S
71 #else
72 #define FWS_INIT FWS_L
73 #endif
75 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
76 struct rt6_info *rt);
77 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
78 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
79 static int fib6_walk(struct fib6_walker_t *w);
80 static int fib6_walk_continue(struct fib6_walker_t *w);
83 * A routing update causes an increase of the serial number on the
84 * affected subtree. This allows for cached routes to be asynchronously
85 * tested when modifications are made to the destination cache as a
86 * result of redirects, path MTU changes, etc.
89 static __u32 rt_sernum;
91 static void fib6_gc_timer_cb(unsigned long arg);
93 static LIST_HEAD(fib6_walkers);
94 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
96 static inline void fib6_walker_link(struct fib6_walker_t *w)
98 write_lock_bh(&fib6_walker_lock);
99 list_add(&w->lh, &fib6_walkers);
100 write_unlock_bh(&fib6_walker_lock);
103 static inline void fib6_walker_unlink(struct fib6_walker_t *w)
105 write_lock_bh(&fib6_walker_lock);
106 list_del(&w->lh);
107 write_unlock_bh(&fib6_walker_lock);
109 static __inline__ u32 fib6_new_sernum(void)
111 u32 n = ++rt_sernum;
112 if ((__s32)n <= 0)
113 rt_sernum = n = 1;
114 return n;
118 * Auxiliary address test functions for the radix tree.
120 * These assume a 32bit processor (although it will work on
121 * 64bit processors)
125 * test bit
127 #if defined(__LITTLE_ENDIAN)
128 # define BITOP_BE32_SWIZZLE (0x1F & ~7)
129 #else
130 # define BITOP_BE32_SWIZZLE 0
131 #endif
133 static __inline__ __be32 addr_bit_set(const void *token, int fn_bit)
135 const __be32 *addr = token;
137 * Here,
138 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
139 * is optimized version of
140 * htonl(1 << ((~fn_bit)&0x1F))
141 * See include/asm-generic/bitops/le.h.
143 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
144 addr[fn_bit >> 5];
147 static __inline__ struct fib6_node * node_alloc(void)
149 struct fib6_node *fn;
151 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
153 return fn;
156 static __inline__ void node_free(struct fib6_node * fn)
158 kmem_cache_free(fib6_node_kmem, fn);
161 static __inline__ void rt6_release(struct rt6_info *rt)
163 if (atomic_dec_and_test(&rt->rt6i_ref))
164 dst_free(&rt->dst);
167 static void fib6_link_table(struct net *net, struct fib6_table *tb)
169 unsigned int h;
172 * Initialize table lock at a single place to give lockdep a key,
173 * tables aren't visible prior to being linked to the list.
175 rwlock_init(&tb->tb6_lock);
177 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
180 * No protection necessary, this is the only list mutatation
181 * operation, tables never disappear once they exist.
183 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
186 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
188 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
190 struct fib6_table *table;
192 table = kzalloc(sizeof(*table), GFP_ATOMIC);
193 if (table != NULL) {
194 table->tb6_id = id;
195 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
196 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
199 return table;
202 struct fib6_table *fib6_new_table(struct net *net, u32 id)
204 struct fib6_table *tb;
206 if (id == 0)
207 id = RT6_TABLE_MAIN;
208 tb = fib6_get_table(net, id);
209 if (tb)
210 return tb;
212 tb = fib6_alloc_table(net, id);
213 if (tb != NULL)
214 fib6_link_table(net, tb);
216 return tb;
219 struct fib6_table *fib6_get_table(struct net *net, u32 id)
221 struct fib6_table *tb;
222 struct hlist_head *head;
223 struct hlist_node *node;
224 unsigned int h;
226 if (id == 0)
227 id = RT6_TABLE_MAIN;
228 h = id & (FIB6_TABLE_HASHSZ - 1);
229 rcu_read_lock();
230 head = &net->ipv6.fib_table_hash[h];
231 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
232 if (tb->tb6_id == id) {
233 rcu_read_unlock();
234 return tb;
237 rcu_read_unlock();
239 return NULL;
242 static void __net_init fib6_tables_init(struct net *net)
244 fib6_link_table(net, net->ipv6.fib6_main_tbl);
245 fib6_link_table(net, net->ipv6.fib6_local_tbl);
247 #else
249 struct fib6_table *fib6_new_table(struct net *net, u32 id)
251 return fib6_get_table(net, id);
254 struct fib6_table *fib6_get_table(struct net *net, u32 id)
256 return net->ipv6.fib6_main_tbl;
259 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
260 int flags, pol_lookup_t lookup)
262 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
265 static void __net_init fib6_tables_init(struct net *net)
267 fib6_link_table(net, net->ipv6.fib6_main_tbl);
270 #endif
272 static int fib6_dump_node(struct fib6_walker_t *w)
274 int res;
275 struct rt6_info *rt;
277 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
278 res = rt6_dump_route(rt, w->args);
279 if (res < 0) {
280 /* Frame is full, suspend walking */
281 w->leaf = rt;
282 return 1;
284 WARN_ON(res == 0);
286 w->leaf = NULL;
287 return 0;
290 static void fib6_dump_end(struct netlink_callback *cb)
292 struct fib6_walker_t *w = (void*)cb->args[2];
294 if (w) {
295 if (cb->args[4]) {
296 cb->args[4] = 0;
297 fib6_walker_unlink(w);
299 cb->args[2] = 0;
300 kfree(w);
302 cb->done = (void*)cb->args[3];
303 cb->args[1] = 3;
306 static int fib6_dump_done(struct netlink_callback *cb)
308 fib6_dump_end(cb);
309 return cb->done ? cb->done(cb) : 0;
312 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
313 struct netlink_callback *cb)
315 struct fib6_walker_t *w;
316 int res;
318 w = (void *)cb->args[2];
319 w->root = &table->tb6_root;
321 if (cb->args[4] == 0) {
322 w->count = 0;
323 w->skip = 0;
325 read_lock_bh(&table->tb6_lock);
326 res = fib6_walk(w);
327 read_unlock_bh(&table->tb6_lock);
328 if (res > 0) {
329 cb->args[4] = 1;
330 cb->args[5] = w->root->fn_sernum;
332 } else {
333 if (cb->args[5] != w->root->fn_sernum) {
334 /* Begin at the root if the tree changed */
335 cb->args[5] = w->root->fn_sernum;
336 w->state = FWS_INIT;
337 w->node = w->root;
338 w->skip = w->count;
339 } else
340 w->skip = 0;
342 read_lock_bh(&table->tb6_lock);
343 res = fib6_walk_continue(w);
344 read_unlock_bh(&table->tb6_lock);
345 if (res <= 0) {
346 fib6_walker_unlink(w);
347 cb->args[4] = 0;
351 return res;
354 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
356 struct net *net = sock_net(skb->sk);
357 unsigned int h, s_h;
358 unsigned int e = 0, s_e;
359 struct rt6_rtnl_dump_arg arg;
360 struct fib6_walker_t *w;
361 struct fib6_table *tb;
362 struct hlist_node *node;
363 struct hlist_head *head;
364 int res = 0;
366 s_h = cb->args[0];
367 s_e = cb->args[1];
369 w = (void *)cb->args[2];
370 if (w == NULL) {
371 /* New dump:
373 * 1. hook callback destructor.
375 cb->args[3] = (long)cb->done;
376 cb->done = fib6_dump_done;
379 * 2. allocate and initialize walker.
381 w = kzalloc(sizeof(*w), GFP_ATOMIC);
382 if (w == NULL)
383 return -ENOMEM;
384 w->func = fib6_dump_node;
385 cb->args[2] = (long)w;
388 arg.skb = skb;
389 arg.cb = cb;
390 arg.net = net;
391 w->args = &arg;
393 rcu_read_lock();
394 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
395 e = 0;
396 head = &net->ipv6.fib_table_hash[h];
397 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
398 if (e < s_e)
399 goto next;
400 res = fib6_dump_table(tb, skb, cb);
401 if (res != 0)
402 goto out;
403 next:
404 e++;
407 out:
408 rcu_read_unlock();
409 cb->args[1] = e;
410 cb->args[0] = h;
412 res = res < 0 ? res : skb->len;
413 if (res <= 0)
414 fib6_dump_end(cb);
415 return res;
419 * Routing Table
421 * return the appropriate node for a routing tree "add" operation
422 * by either creating and inserting or by returning an existing
423 * node.
426 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
427 int addrlen, int plen,
428 int offset)
430 struct fib6_node *fn, *in, *ln;
431 struct fib6_node *pn = NULL;
432 struct rt6key *key;
433 int bit;
434 __be32 dir = 0;
435 __u32 sernum = fib6_new_sernum();
437 RT6_TRACE("fib6_add_1\n");
439 /* insert node in tree */
441 fn = root;
443 do {
444 key = (struct rt6key *)((u8 *)fn->leaf + offset);
447 * Prefix match
449 if (plen < fn->fn_bit ||
450 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
451 goto insert_above;
454 * Exact match ?
457 if (plen == fn->fn_bit) {
458 /* clean up an intermediate node */
459 if ((fn->fn_flags & RTN_RTINFO) == 0) {
460 rt6_release(fn->leaf);
461 fn->leaf = NULL;
464 fn->fn_sernum = sernum;
466 return fn;
470 * We have more bits to go
473 /* Try to walk down on tree. */
474 fn->fn_sernum = sernum;
475 dir = addr_bit_set(addr, fn->fn_bit);
476 pn = fn;
477 fn = dir ? fn->right: fn->left;
478 } while (fn);
481 * We walked to the bottom of tree.
482 * Create new leaf node without children.
485 ln = node_alloc();
487 if (ln == NULL)
488 return NULL;
489 ln->fn_bit = plen;
491 ln->parent = pn;
492 ln->fn_sernum = sernum;
494 if (dir)
495 pn->right = ln;
496 else
497 pn->left = ln;
499 return ln;
502 insert_above:
504 * split since we don't have a common prefix anymore or
505 * we have a less significant route.
506 * we've to insert an intermediate node on the list
507 * this new node will point to the one we need to create
508 * and the current
511 pn = fn->parent;
513 /* find 1st bit in difference between the 2 addrs.
515 See comment in __ipv6_addr_diff: bit may be an invalid value,
516 but if it is >= plen, the value is ignored in any case.
519 bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
522 * (intermediate)[in]
523 * / \
524 * (new leaf node)[ln] (old node)[fn]
526 if (plen > bit) {
527 in = node_alloc();
528 ln = node_alloc();
530 if (in == NULL || ln == NULL) {
531 if (in)
532 node_free(in);
533 if (ln)
534 node_free(ln);
535 return NULL;
539 * new intermediate node.
540 * RTN_RTINFO will
541 * be off since that an address that chooses one of
542 * the branches would not match less specific routes
543 * in the other branch
546 in->fn_bit = bit;
548 in->parent = pn;
549 in->leaf = fn->leaf;
550 atomic_inc(&in->leaf->rt6i_ref);
552 in->fn_sernum = sernum;
554 /* update parent pointer */
555 if (dir)
556 pn->right = in;
557 else
558 pn->left = in;
560 ln->fn_bit = plen;
562 ln->parent = in;
563 fn->parent = in;
565 ln->fn_sernum = sernum;
567 if (addr_bit_set(addr, bit)) {
568 in->right = ln;
569 in->left = fn;
570 } else {
571 in->left = ln;
572 in->right = fn;
574 } else { /* plen <= bit */
577 * (new leaf node)[ln]
578 * / \
579 * (old node)[fn] NULL
582 ln = node_alloc();
584 if (ln == NULL)
585 return NULL;
587 ln->fn_bit = plen;
589 ln->parent = pn;
591 ln->fn_sernum = sernum;
593 if (dir)
594 pn->right = ln;
595 else
596 pn->left = ln;
598 if (addr_bit_set(&key->addr, plen))
599 ln->right = fn;
600 else
601 ln->left = fn;
603 fn->parent = ln;
605 return ln;
609 * Insert routing information in a node.
612 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
613 struct nl_info *info)
615 struct rt6_info *iter = NULL;
616 struct rt6_info **ins;
618 ins = &fn->leaf;
620 for (iter = fn->leaf; iter; iter=iter->dst.rt6_next) {
622 * Search for duplicates
625 if (iter->rt6i_metric == rt->rt6i_metric) {
627 * Same priority level
630 if (iter->rt6i_dev == rt->rt6i_dev &&
631 iter->rt6i_idev == rt->rt6i_idev &&
632 ipv6_addr_equal(&iter->rt6i_gateway,
633 &rt->rt6i_gateway)) {
634 if (!(iter->rt6i_flags&RTF_EXPIRES))
635 return -EEXIST;
636 iter->rt6i_expires = rt->rt6i_expires;
637 if (!(rt->rt6i_flags&RTF_EXPIRES)) {
638 iter->rt6i_flags &= ~RTF_EXPIRES;
639 iter->rt6i_expires = 0;
641 return -EEXIST;
645 if (iter->rt6i_metric > rt->rt6i_metric)
646 break;
648 ins = &iter->dst.rt6_next;
651 /* Reset round-robin state, if necessary */
652 if (ins == &fn->leaf)
653 fn->rr_ptr = NULL;
656 * insert node
659 rt->dst.rt6_next = iter;
660 *ins = rt;
661 rt->rt6i_node = fn;
662 atomic_inc(&rt->rt6i_ref);
663 inet6_rt_notify(RTM_NEWROUTE, rt, info);
664 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
666 if ((fn->fn_flags & RTN_RTINFO) == 0) {
667 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
668 fn->fn_flags |= RTN_RTINFO;
671 return 0;
674 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
676 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
677 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE)))
678 mod_timer(&net->ipv6.ip6_fib_timer,
679 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
682 void fib6_force_start_gc(struct net *net)
684 if (!timer_pending(&net->ipv6.ip6_fib_timer))
685 mod_timer(&net->ipv6.ip6_fib_timer,
686 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
690 * Add routing information to the routing tree.
691 * <destination addr>/<source addr>
692 * with source addr info in sub-trees
695 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
697 struct fib6_node *fn, *pn = NULL;
698 int err = -ENOMEM;
700 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
701 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst));
703 if (fn == NULL)
704 goto out;
706 pn = fn;
708 #ifdef CONFIG_IPV6_SUBTREES
709 if (rt->rt6i_src.plen) {
710 struct fib6_node *sn;
712 if (fn->subtree == NULL) {
713 struct fib6_node *sfn;
716 * Create subtree.
718 * fn[main tree]
720 * sfn[subtree root]
722 * sn[new leaf node]
725 /* Create subtree root node */
726 sfn = node_alloc();
727 if (sfn == NULL)
728 goto st_failure;
730 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
731 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
732 sfn->fn_flags = RTN_ROOT;
733 sfn->fn_sernum = fib6_new_sernum();
735 /* Now add the first leaf node to new subtree */
737 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
738 sizeof(struct in6_addr), rt->rt6i_src.plen,
739 offsetof(struct rt6_info, rt6i_src));
741 if (sn == NULL) {
742 /* If it is failed, discard just allocated
743 root, and then (in st_failure) stale node
744 in main tree.
746 node_free(sfn);
747 goto st_failure;
750 /* Now link new subtree to main tree */
751 sfn->parent = fn;
752 fn->subtree = sfn;
753 } else {
754 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
755 sizeof(struct in6_addr), rt->rt6i_src.plen,
756 offsetof(struct rt6_info, rt6i_src));
758 if (sn == NULL)
759 goto st_failure;
762 if (fn->leaf == NULL) {
763 fn->leaf = rt;
764 atomic_inc(&rt->rt6i_ref);
766 fn = sn;
768 #endif
770 err = fib6_add_rt2node(fn, rt, info);
772 if (err == 0) {
773 fib6_start_gc(info->nl_net, rt);
774 if (!(rt->rt6i_flags&RTF_CACHE))
775 fib6_prune_clones(info->nl_net, pn, rt);
778 out:
779 if (err) {
780 #ifdef CONFIG_IPV6_SUBTREES
782 * If fib6_add_1 has cleared the old leaf pointer in the
783 * super-tree leaf node we have to find a new one for it.
785 if (pn != fn && pn->leaf == rt) {
786 pn->leaf = NULL;
787 atomic_dec(&rt->rt6i_ref);
789 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
790 pn->leaf = fib6_find_prefix(info->nl_net, pn);
791 #if RT6_DEBUG >= 2
792 if (!pn->leaf) {
793 WARN_ON(pn->leaf == NULL);
794 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
796 #endif
797 atomic_inc(&pn->leaf->rt6i_ref);
799 #endif
800 dst_free(&rt->dst);
802 return err;
804 #ifdef CONFIG_IPV6_SUBTREES
805 /* Subtree creation failed, probably main tree node
806 is orphan. If it is, shoot it.
808 st_failure:
809 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
810 fib6_repair_tree(info->nl_net, fn);
811 dst_free(&rt->dst);
812 return err;
813 #endif
817 * Routing tree lookup
821 struct lookup_args {
822 int offset; /* key offset on rt6_info */
823 const struct in6_addr *addr; /* search key */
826 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
827 struct lookup_args *args)
829 struct fib6_node *fn;
830 __be32 dir;
832 if (unlikely(args->offset == 0))
833 return NULL;
836 * Descend on a tree
839 fn = root;
841 for (;;) {
842 struct fib6_node *next;
844 dir = addr_bit_set(args->addr, fn->fn_bit);
846 next = dir ? fn->right : fn->left;
848 if (next) {
849 fn = next;
850 continue;
853 break;
856 while(fn) {
857 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
858 struct rt6key *key;
860 key = (struct rt6key *) ((u8 *) fn->leaf +
861 args->offset);
863 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
864 #ifdef CONFIG_IPV6_SUBTREES
865 if (fn->subtree)
866 fn = fib6_lookup_1(fn->subtree, args + 1);
867 #endif
868 if (!fn || fn->fn_flags & RTN_RTINFO)
869 return fn;
873 if (fn->fn_flags & RTN_ROOT)
874 break;
876 fn = fn->parent;
879 return NULL;
882 struct fib6_node * fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
883 const struct in6_addr *saddr)
885 struct fib6_node *fn;
886 struct lookup_args args[] = {
888 .offset = offsetof(struct rt6_info, rt6i_dst),
889 .addr = daddr,
891 #ifdef CONFIG_IPV6_SUBTREES
893 .offset = offsetof(struct rt6_info, rt6i_src),
894 .addr = saddr,
896 #endif
898 .offset = 0, /* sentinel */
902 fn = fib6_lookup_1(root, daddr ? args : args + 1);
904 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT)
905 fn = root;
907 return fn;
911 * Get node with specified destination prefix (and source prefix,
912 * if subtrees are used)
916 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
917 const struct in6_addr *addr,
918 int plen, int offset)
920 struct fib6_node *fn;
922 for (fn = root; fn ; ) {
923 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
926 * Prefix match
928 if (plen < fn->fn_bit ||
929 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
930 return NULL;
932 if (plen == fn->fn_bit)
933 return fn;
936 * We have more bits to go
938 if (addr_bit_set(addr, fn->fn_bit))
939 fn = fn->right;
940 else
941 fn = fn->left;
943 return NULL;
946 struct fib6_node * fib6_locate(struct fib6_node *root,
947 const struct in6_addr *daddr, int dst_len,
948 const struct in6_addr *saddr, int src_len)
950 struct fib6_node *fn;
952 fn = fib6_locate_1(root, daddr, dst_len,
953 offsetof(struct rt6_info, rt6i_dst));
955 #ifdef CONFIG_IPV6_SUBTREES
956 if (src_len) {
957 WARN_ON(saddr == NULL);
958 if (fn && fn->subtree)
959 fn = fib6_locate_1(fn->subtree, saddr, src_len,
960 offsetof(struct rt6_info, rt6i_src));
962 #endif
964 if (fn && fn->fn_flags&RTN_RTINFO)
965 return fn;
967 return NULL;
972 * Deletion
976 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
978 if (fn->fn_flags&RTN_ROOT)
979 return net->ipv6.ip6_null_entry;
981 while(fn) {
982 if(fn->left)
983 return fn->left->leaf;
985 if(fn->right)
986 return fn->right->leaf;
988 fn = FIB6_SUBTREE(fn);
990 return NULL;
994 * Called to trim the tree of intermediate nodes when possible. "fn"
995 * is the node we want to try and remove.
998 static struct fib6_node *fib6_repair_tree(struct net *net,
999 struct fib6_node *fn)
1001 int children;
1002 int nstate;
1003 struct fib6_node *child, *pn;
1004 struct fib6_walker_t *w;
1005 int iter = 0;
1007 for (;;) {
1008 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1009 iter++;
1011 WARN_ON(fn->fn_flags & RTN_RTINFO);
1012 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1013 WARN_ON(fn->leaf != NULL);
1015 children = 0;
1016 child = NULL;
1017 if (fn->right) child = fn->right, children |= 1;
1018 if (fn->left) child = fn->left, children |= 2;
1020 if (children == 3 || FIB6_SUBTREE(fn)
1021 #ifdef CONFIG_IPV6_SUBTREES
1022 /* Subtree root (i.e. fn) may have one child */
1023 || (children && fn->fn_flags&RTN_ROOT)
1024 #endif
1026 fn->leaf = fib6_find_prefix(net, fn);
1027 #if RT6_DEBUG >= 2
1028 if (fn->leaf==NULL) {
1029 WARN_ON(!fn->leaf);
1030 fn->leaf = net->ipv6.ip6_null_entry;
1032 #endif
1033 atomic_inc(&fn->leaf->rt6i_ref);
1034 return fn->parent;
1037 pn = fn->parent;
1038 #ifdef CONFIG_IPV6_SUBTREES
1039 if (FIB6_SUBTREE(pn) == fn) {
1040 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1041 FIB6_SUBTREE(pn) = NULL;
1042 nstate = FWS_L;
1043 } else {
1044 WARN_ON(fn->fn_flags & RTN_ROOT);
1045 #endif
1046 if (pn->right == fn) pn->right = child;
1047 else if (pn->left == fn) pn->left = child;
1048 #if RT6_DEBUG >= 2
1049 else
1050 WARN_ON(1);
1051 #endif
1052 if (child)
1053 child->parent = pn;
1054 nstate = FWS_R;
1055 #ifdef CONFIG_IPV6_SUBTREES
1057 #endif
1059 read_lock(&fib6_walker_lock);
1060 FOR_WALKERS(w) {
1061 if (child == NULL) {
1062 if (w->root == fn) {
1063 w->root = w->node = NULL;
1064 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1065 } else if (w->node == fn) {
1066 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1067 w->node = pn;
1068 w->state = nstate;
1070 } else {
1071 if (w->root == fn) {
1072 w->root = child;
1073 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1075 if (w->node == fn) {
1076 w->node = child;
1077 if (children&2) {
1078 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1079 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1080 } else {
1081 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1082 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1087 read_unlock(&fib6_walker_lock);
1089 node_free(fn);
1090 if (pn->fn_flags&RTN_RTINFO || FIB6_SUBTREE(pn))
1091 return pn;
1093 rt6_release(pn->leaf);
1094 pn->leaf = NULL;
1095 fn = pn;
1099 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1100 struct nl_info *info)
1102 struct fib6_walker_t *w;
1103 struct rt6_info *rt = *rtp;
1104 struct net *net = info->nl_net;
1106 RT6_TRACE("fib6_del_route\n");
1108 /* Unlink it */
1109 *rtp = rt->dst.rt6_next;
1110 rt->rt6i_node = NULL;
1111 net->ipv6.rt6_stats->fib_rt_entries--;
1112 net->ipv6.rt6_stats->fib_discarded_routes++;
1114 /* Reset round-robin state, if necessary */
1115 if (fn->rr_ptr == rt)
1116 fn->rr_ptr = NULL;
1118 /* Adjust walkers */
1119 read_lock(&fib6_walker_lock);
1120 FOR_WALKERS(w) {
1121 if (w->state == FWS_C && w->leaf == rt) {
1122 RT6_TRACE("walker %p adjusted by delroute\n", w);
1123 w->leaf = rt->dst.rt6_next;
1124 if (w->leaf == NULL)
1125 w->state = FWS_U;
1128 read_unlock(&fib6_walker_lock);
1130 rt->dst.rt6_next = NULL;
1132 /* If it was last route, expunge its radix tree node */
1133 if (fn->leaf == NULL) {
1134 fn->fn_flags &= ~RTN_RTINFO;
1135 net->ipv6.rt6_stats->fib_route_nodes--;
1136 fn = fib6_repair_tree(net, fn);
1139 if (atomic_read(&rt->rt6i_ref) != 1) {
1140 /* This route is used as dummy address holder in some split
1141 * nodes. It is not leaked, but it still holds other resources,
1142 * which must be released in time. So, scan ascendant nodes
1143 * and replace dummy references to this route with references
1144 * to still alive ones.
1146 while (fn) {
1147 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) {
1148 fn->leaf = fib6_find_prefix(net, fn);
1149 atomic_inc(&fn->leaf->rt6i_ref);
1150 rt6_release(rt);
1152 fn = fn->parent;
1154 /* No more references are possible at this point. */
1155 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1158 inet6_rt_notify(RTM_DELROUTE, rt, info);
1159 rt6_release(rt);
1162 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1164 struct net *net = info->nl_net;
1165 struct fib6_node *fn = rt->rt6i_node;
1166 struct rt6_info **rtp;
1168 #if RT6_DEBUG >= 2
1169 if (rt->dst.obsolete>0) {
1170 WARN_ON(fn != NULL);
1171 return -ENOENT;
1173 #endif
1174 if (fn == NULL || rt == net->ipv6.ip6_null_entry)
1175 return -ENOENT;
1177 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1179 if (!(rt->rt6i_flags&RTF_CACHE)) {
1180 struct fib6_node *pn = fn;
1181 #ifdef CONFIG_IPV6_SUBTREES
1182 /* clones of this route might be in another subtree */
1183 if (rt->rt6i_src.plen) {
1184 while (!(pn->fn_flags&RTN_ROOT))
1185 pn = pn->parent;
1186 pn = pn->parent;
1188 #endif
1189 fib6_prune_clones(info->nl_net, pn, rt);
1193 * Walk the leaf entries looking for ourself
1196 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1197 if (*rtp == rt) {
1198 fib6_del_route(fn, rtp, info);
1199 return 0;
1202 return -ENOENT;
1206 * Tree traversal function.
1208 * Certainly, it is not interrupt safe.
1209 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1210 * It means, that we can modify tree during walking
1211 * and use this function for garbage collection, clone pruning,
1212 * cleaning tree when a device goes down etc. etc.
1214 * It guarantees that every node will be traversed,
1215 * and that it will be traversed only once.
1217 * Callback function w->func may return:
1218 * 0 -> continue walking.
1219 * positive value -> walking is suspended (used by tree dumps,
1220 * and probably by gc, if it will be split to several slices)
1221 * negative value -> terminate walking.
1223 * The function itself returns:
1224 * 0 -> walk is complete.
1225 * >0 -> walk is incomplete (i.e. suspended)
1226 * <0 -> walk is terminated by an error.
1229 static int fib6_walk_continue(struct fib6_walker_t *w)
1231 struct fib6_node *fn, *pn;
1233 for (;;) {
1234 fn = w->node;
1235 if (fn == NULL)
1236 return 0;
1238 if (w->prune && fn != w->root &&
1239 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) {
1240 w->state = FWS_C;
1241 w->leaf = fn->leaf;
1243 switch (w->state) {
1244 #ifdef CONFIG_IPV6_SUBTREES
1245 case FWS_S:
1246 if (FIB6_SUBTREE(fn)) {
1247 w->node = FIB6_SUBTREE(fn);
1248 continue;
1250 w->state = FWS_L;
1251 #endif
1252 case FWS_L:
1253 if (fn->left) {
1254 w->node = fn->left;
1255 w->state = FWS_INIT;
1256 continue;
1258 w->state = FWS_R;
1259 case FWS_R:
1260 if (fn->right) {
1261 w->node = fn->right;
1262 w->state = FWS_INIT;
1263 continue;
1265 w->state = FWS_C;
1266 w->leaf = fn->leaf;
1267 case FWS_C:
1268 if (w->leaf && fn->fn_flags&RTN_RTINFO) {
1269 int err;
1271 if (w->count < w->skip) {
1272 w->count++;
1273 continue;
1276 err = w->func(w);
1277 if (err)
1278 return err;
1280 w->count++;
1281 continue;
1283 w->state = FWS_U;
1284 case FWS_U:
1285 if (fn == w->root)
1286 return 0;
1287 pn = fn->parent;
1288 w->node = pn;
1289 #ifdef CONFIG_IPV6_SUBTREES
1290 if (FIB6_SUBTREE(pn) == fn) {
1291 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1292 w->state = FWS_L;
1293 continue;
1295 #endif
1296 if (pn->left == fn) {
1297 w->state = FWS_R;
1298 continue;
1300 if (pn->right == fn) {
1301 w->state = FWS_C;
1302 w->leaf = w->node->leaf;
1303 continue;
1305 #if RT6_DEBUG >= 2
1306 WARN_ON(1);
1307 #endif
1312 static int fib6_walk(struct fib6_walker_t *w)
1314 int res;
1316 w->state = FWS_INIT;
1317 w->node = w->root;
1319 fib6_walker_link(w);
1320 res = fib6_walk_continue(w);
1321 if (res <= 0)
1322 fib6_walker_unlink(w);
1323 return res;
1326 static int fib6_clean_node(struct fib6_walker_t *w)
1328 int res;
1329 struct rt6_info *rt;
1330 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1331 struct nl_info info = {
1332 .nl_net = c->net,
1335 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1336 res = c->func(rt, c->arg);
1337 if (res < 0) {
1338 w->leaf = rt;
1339 res = fib6_del(rt, &info);
1340 if (res) {
1341 #if RT6_DEBUG >= 2
1342 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res);
1343 #endif
1344 continue;
1346 return 0;
1348 WARN_ON(res != 0);
1350 w->leaf = rt;
1351 return 0;
1355 * Convenient frontend to tree walker.
1357 * func is called on each route.
1358 * It may return -1 -> delete this route.
1359 * 0 -> continue walking
1361 * prune==1 -> only immediate children of node (certainly,
1362 * ignoring pure split nodes) will be scanned.
1365 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1366 int (*func)(struct rt6_info *, void *arg),
1367 int prune, void *arg)
1369 struct fib6_cleaner_t c;
1371 c.w.root = root;
1372 c.w.func = fib6_clean_node;
1373 c.w.prune = prune;
1374 c.w.count = 0;
1375 c.w.skip = 0;
1376 c.func = func;
1377 c.arg = arg;
1378 c.net = net;
1380 fib6_walk(&c.w);
1383 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1384 int prune, void *arg)
1386 struct fib6_table *table;
1387 struct hlist_node *node;
1388 struct hlist_head *head;
1389 unsigned int h;
1391 rcu_read_lock();
1392 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1393 head = &net->ipv6.fib_table_hash[h];
1394 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1395 write_lock_bh(&table->tb6_lock);
1396 fib6_clean_tree(net, &table->tb6_root,
1397 func, prune, arg);
1398 write_unlock_bh(&table->tb6_lock);
1401 rcu_read_unlock();
1404 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1406 if (rt->rt6i_flags & RTF_CACHE) {
1407 RT6_TRACE("pruning clone %p\n", rt);
1408 return -1;
1411 return 0;
1414 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1415 struct rt6_info *rt)
1417 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1421 * Garbage collection
1424 static struct fib6_gc_args
1426 int timeout;
1427 int more;
1428 } gc_args;
1430 static int fib6_age(struct rt6_info *rt, void *arg)
1432 unsigned long now = jiffies;
1435 * check addrconf expiration here.
1436 * Routes are expired even if they are in use.
1438 * Also age clones. Note, that clones are aged out
1439 * only if they are not in use now.
1442 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) {
1443 if (time_after(now, rt->rt6i_expires)) {
1444 RT6_TRACE("expiring %p\n", rt);
1445 return -1;
1447 gc_args.more++;
1448 } else if (rt->rt6i_flags & RTF_CACHE) {
1449 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1450 time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1451 RT6_TRACE("aging clone %p\n", rt);
1452 return -1;
1453 } else if ((rt->rt6i_flags & RTF_GATEWAY) &&
1454 (!(dst_get_neighbour_raw(&rt->dst)->flags & NTF_ROUTER))) {
1455 RT6_TRACE("purging route %p via non-router but gateway\n",
1456 rt);
1457 return -1;
1459 gc_args.more++;
1462 return 0;
1465 static DEFINE_SPINLOCK(fib6_gc_lock);
1467 void fib6_run_gc(unsigned long expires, struct net *net)
1469 if (expires != ~0UL) {
1470 spin_lock_bh(&fib6_gc_lock);
1471 gc_args.timeout = expires ? (int)expires :
1472 net->ipv6.sysctl.ip6_rt_gc_interval;
1473 } else {
1474 if (!spin_trylock_bh(&fib6_gc_lock)) {
1475 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1476 return;
1478 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1481 gc_args.more = icmp6_dst_gc();
1483 fib6_clean_all(net, fib6_age, 0, NULL);
1485 if (gc_args.more)
1486 mod_timer(&net->ipv6.ip6_fib_timer,
1487 round_jiffies(jiffies
1488 + net->ipv6.sysctl.ip6_rt_gc_interval));
1489 else
1490 del_timer(&net->ipv6.ip6_fib_timer);
1491 spin_unlock_bh(&fib6_gc_lock);
1494 static void fib6_gc_timer_cb(unsigned long arg)
1496 fib6_run_gc(0, (struct net *)arg);
1499 static int __net_init fib6_net_init(struct net *net)
1501 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1503 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1505 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1506 if (!net->ipv6.rt6_stats)
1507 goto out_timer;
1509 /* Avoid false sharing : Use at least a full cache line */
1510 size = max_t(size_t, size, L1_CACHE_BYTES);
1512 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1513 if (!net->ipv6.fib_table_hash)
1514 goto out_rt6_stats;
1516 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1517 GFP_KERNEL);
1518 if (!net->ipv6.fib6_main_tbl)
1519 goto out_fib_table_hash;
1521 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1522 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1523 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1524 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1526 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1527 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1528 GFP_KERNEL);
1529 if (!net->ipv6.fib6_local_tbl)
1530 goto out_fib6_main_tbl;
1531 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1532 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1533 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1534 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1535 #endif
1536 fib6_tables_init(net);
1538 return 0;
1540 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1541 out_fib6_main_tbl:
1542 kfree(net->ipv6.fib6_main_tbl);
1543 #endif
1544 out_fib_table_hash:
1545 kfree(net->ipv6.fib_table_hash);
1546 out_rt6_stats:
1547 kfree(net->ipv6.rt6_stats);
1548 out_timer:
1549 return -ENOMEM;
1552 static void fib6_net_exit(struct net *net)
1554 rt6_ifdown(net, NULL);
1555 del_timer_sync(&net->ipv6.ip6_fib_timer);
1557 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1558 kfree(net->ipv6.fib6_local_tbl);
1559 #endif
1560 kfree(net->ipv6.fib6_main_tbl);
1561 kfree(net->ipv6.fib_table_hash);
1562 kfree(net->ipv6.rt6_stats);
1565 static struct pernet_operations fib6_net_ops = {
1566 .init = fib6_net_init,
1567 .exit = fib6_net_exit,
1570 int __init fib6_init(void)
1572 int ret = -ENOMEM;
1574 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1575 sizeof(struct fib6_node),
1576 0, SLAB_HWCACHE_ALIGN,
1577 NULL);
1578 if (!fib6_node_kmem)
1579 goto out;
1581 ret = register_pernet_subsys(&fib6_net_ops);
1582 if (ret)
1583 goto out_kmem_cache_create;
1585 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1586 NULL);
1587 if (ret)
1588 goto out_unregister_subsys;
1589 out:
1590 return ret;
1592 out_unregister_subsys:
1593 unregister_pernet_subsys(&fib6_net_ops);
1594 out_kmem_cache_create:
1595 kmem_cache_destroy(fib6_node_kmem);
1596 goto out;
1599 void fib6_gc_cleanup(void)
1601 unregister_pernet_subsys(&fib6_net_ops);
1602 kmem_cache_destroy(fib6_node_kmem);