?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
RT-AC66 3.0.0.4.374.130 core
[tomato.git] / release / src-rt-6.x / linux / linux-2.6 / net / ipv6 / ip6_fib.c
blob662a7d9681fdb03b4c2a9c089ec4dffdcfb2f4d2
1 /*
2 * Linux INET6 implementation
3 * Forwarding Information Database
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
8 * $Id: ip6_fib.c,v 1.25 2001/10/31 21:55:55 davem Exp $
9 *
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.
14 */
15
16 /*
17 * Changes:
18 * Yuji SEKIYA @USAGI: Support default route on router node;
19 * remove ip6_null_entry from the top of
20 * routing table.
21 * Ville Nuorvala: Fixed routing subtrees.
22 */
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>
31
32 #ifdef CONFIG_PROC_FS
33 #include <linux/proc_fs.h>
34 #endif
35
36 #include <net/ipv6.h>
37 #include <net/ndisc.h>
38 #include <net/addrconf.h>
39
40 #include <net/ip6_fib.h>
41 #include <net/ip6_route.h>
42
43 #define RT6_DEBUG 2
44
45 #if RT6_DEBUG >= 3
46 #define RT6_TRACE(x...) printk(KERN_DEBUG x)
47 #else
48 #define RT6_TRACE(x...) do { ; } while (0)
49 #endif
50
51 struct rt6_statistics rt6_stats;
52
53 static struct kmem_cache * fib6_node_kmem __read_mostly;
54
55 enum fib_walk_state_t
56 {
57 #ifdef CONFIG_IPV6_SUBTREES
58 FWS_S,
59 #endif
60 FWS_L,
61 FWS_R,
62 FWS_C,
63 FWS_U
64 };
65
66 struct fib6_cleaner_t
67 {
68 struct fib6_walker_t w;
69 int (*func)(struct rt6_info *, void *arg);
70 void *arg;
71 };
72
73 static DEFINE_RWLOCK(fib6_walker_lock);
74
75 #ifdef CONFIG_IPV6_SUBTREES
76 #define FWS_INIT FWS_S
77 #else
78 #define FWS_INIT FWS_L
79 #endif
80
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);
86
87 /*
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.
92 */
93
94 static __u32 rt_sernum;
95
96 static DEFINE_TIMER(ip6_fib_timer, fib6_run_gc, 0, 0);
97
98 static struct fib6_walker_t fib6_walker_list = {
99 .prev = &fib6_walker_list,
100 .next = &fib6_walker_list,
101 };
102
103 #define FOR_WALKERS(w) for ((w)=fib6_walker_list.next; (w) != &fib6_walker_list; (w)=(w)->next)
104
105 static inline void fib6_walker_link(struct fib6_walker_t *w)
106 {
107 write_lock_bh(&fib6_walker_lock);
108 w->next = fib6_walker_list.next;
109 w->prev = &fib6_walker_list;
110 w->next->prev = w;
111 w->prev->next = w;
112 write_unlock_bh(&fib6_walker_lock);
113 }
114
115 static inline void fib6_walker_unlink(struct fib6_walker_t *w)
116 {
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);
122 }
123 static __inline__ u32 fib6_new_sernum(void)
124 {
125 u32 n = ++rt_sernum;
126 if ((__s32)n <= 0)
127 rt_sernum = n = 1;
128 return n;
129 }
130
131 /*
132 * Auxiliary address test functions for the radix tree.
133 *
134 * These assume a 32bit processor (although it will work on
135 * 64bit processors)
136 */
137
138 /*
139 * test bit
140 */
141
142 static __inline__ __be32 addr_bit_set(void *token, int fn_bit)
143 {
144 __be32 *addr = token;
145
146 return htonl(1 << ((~fn_bit)&0x1F)) & addr[fn_bit>>5];
147 }
148
149 static __inline__ struct fib6_node * node_alloc(void)
150 {
151 struct fib6_node *fn;
152
153 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
154
155 return fn;
156 }
157
158 static __inline__ void node_free(struct fib6_node * fn)
159 {
160 kmem_cache_free(fib6_node_kmem, fn);
161 }
162
163 static __inline__ void rt6_release(struct rt6_info *rt)
164 {
165 if (atomic_dec_and_test(&rt->rt6i_ref))
166 dst_free(&rt->u.dst);
167 }
168
169 static struct fib6_table fib6_main_tbl = {
170 .tb6_id = RT6_TABLE_MAIN,
171 .tb6_root = {
172 .leaf = &ip6_null_entry,
173 .fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO,
174 },
175 };
176
177 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
178 #define FIB_TABLE_HASHSZ 256
179 #else
180 #define FIB_TABLE_HASHSZ 1
181 #endif
182 static struct hlist_head fib_table_hash[FIB_TABLE_HASHSZ];
183
184 static void fib6_link_table(struct fib6_table *tb)
185 {
186 unsigned int h;
187
188 /*
189 * Initialize table lock at a single place to give lockdep a key,
190 * tables aren't visible prior to being linked to the list.
191 */
192 rwlock_init(&tb->tb6_lock);
193
194 h = tb->tb6_id & (FIB_TABLE_HASHSZ - 1);
195
196 /*
197 * No protection necessary, this is the only list mutatation
198 * operation, tables never disappear once they exist.
199 */
200 hlist_add_head_rcu(&tb->tb6_hlist, &fib_table_hash[h]);
201 }
202
203 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
204 static struct fib6_table fib6_local_tbl = {
205 .tb6_id = RT6_TABLE_LOCAL,
206 .tb6_root = {
207 .leaf = &ip6_null_entry,
208 .fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO,
209 },
210 };
211
212 static struct fib6_table *fib6_alloc_table(u32 id)
213 {
214 struct fib6_table *table;
215
216 table = kzalloc(sizeof(*table), GFP_ATOMIC);
217 if (table != NULL) {
218 table->tb6_id = id;
219 table->tb6_root.leaf = &ip6_null_entry;
220 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
221 }
222
223 return table;
224 }
225
226 struct fib6_table *fib6_new_table(u32 id)
227 {
228 struct fib6_table *tb;
229
230 if (id == 0)
231 id = RT6_TABLE_MAIN;
232 tb = fib6_get_table(id);
233 if (tb)
234 return tb;
235
236 tb = fib6_alloc_table(id);
237 if (tb != NULL)
238 fib6_link_table(tb);
239
240 return tb;
241 }
242
243 struct fib6_table *fib6_get_table(u32 id)
244 {
245 struct fib6_table *tb;
246 struct hlist_node *node;
247 unsigned int h;
248
249 if (id == 0)
250 id = RT6_TABLE_MAIN;
251 h = id & (FIB_TABLE_HASHSZ - 1);
252 rcu_read_lock();
253 hlist_for_each_entry_rcu(tb, node, &fib_table_hash[h], tb6_hlist) {
254 if (tb->tb6_id == id) {
255 rcu_read_unlock();
256 return tb;
257 }
258 }
259 rcu_read_unlock();
260
261 return NULL;
262 }
263
264 static void __init fib6_tables_init(void)
265 {
266 fib6_link_table(&fib6_main_tbl);
267 fib6_link_table(&fib6_local_tbl);
268 }
269
270 #else
271
272 struct fib6_table *fib6_new_table(u32 id)
273 {
274 return fib6_get_table(id);
275 }
276
277 struct fib6_table *fib6_get_table(u32 id)
278 {
279 return &fib6_main_tbl;
280 }
281
282 struct dst_entry *fib6_rule_lookup(struct flowi *fl, int flags,
283 pol_lookup_t lookup)
284 {
285 return (struct dst_entry *) lookup(&fib6_main_tbl, fl, flags);
286 }
287
288 static void __init fib6_tables_init(void)
289 {
290 fib6_link_table(&fib6_main_tbl);
291 }
292
293 #endif
294
295 static int fib6_dump_node(struct fib6_walker_t *w)
296 {
297 int res;
298 struct rt6_info *rt;
299
300 for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) {
301 res = rt6_dump_route(rt, w->args);
302 if (res < 0) {
303 /* Frame is full, suspend walking */
304 w->leaf = rt;
305 return 1;
306 }
307 BUG_TRAP(res!=0);
308 }
309 w->leaf = NULL;
310 return 0;
311 }
312
313 static void fib6_dump_end(struct netlink_callback *cb)
314 {
315 struct fib6_walker_t *w = (void*)cb->args[2];
316
317 if (w) {
318 cb->args[2] = 0;
319 kfree(w);
320 }
321 cb->done = (void*)cb->args[3];
322 cb->args[1] = 3;
323 }
324
325 static int fib6_dump_done(struct netlink_callback *cb)
326 {
327 fib6_dump_end(cb);
328 return cb->done ? cb->done(cb) : 0;
329 }
330
331 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
332 struct netlink_callback *cb)
333 {
334 struct fib6_walker_t *w;
335 int res;
336
337 w = (void *)cb->args[2];
338 w->root = &table->tb6_root;
339
340 if (cb->args[4] == 0) {
341 read_lock_bh(&table->tb6_lock);
342 res = fib6_walk(w);
343 read_unlock_bh(&table->tb6_lock);
344 if (res > 0)
345 cb->args[4] = 1;
346 } else {
347 read_lock_bh(&table->tb6_lock);
348 res = fib6_walk_continue(w);
349 read_unlock_bh(&table->tb6_lock);
350 if (res != 0) {
351 if (res < 0)
352 fib6_walker_unlink(w);
353 goto end;
354 }
355 fib6_walker_unlink(w);
356 cb->args[4] = 0;
357 }
358 end:
359 return res;
360 }
361
362 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
363 {
364 unsigned int h, s_h;
365 unsigned int e = 0, s_e;
366 struct rt6_rtnl_dump_arg arg;
367 struct fib6_walker_t *w;
368 struct fib6_table *tb;
369 struct hlist_node *node;
370 int res = 0;
371
372 s_h = cb->args[0];
373 s_e = cb->args[1];
374
375 w = (void *)cb->args[2];
376 if (w == NULL) {
377 /* New dump:
378 *
379 * 1. hook callback destructor.
380 */
381 cb->args[3] = (long)cb->done;
382 cb->done = fib6_dump_done;
383
384 /*
385 * 2. allocate and initialize walker.
386 */
387 w = kzalloc(sizeof(*w), GFP_ATOMIC);
388 if (w == NULL)
389 return -ENOMEM;
390 w->func = fib6_dump_node;
391 cb->args[2] = (long)w;
392 }
393
394 arg.skb = skb;
395 arg.cb = cb;
396 w->args = &arg;
397
398 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
399 e = 0;
400 hlist_for_each_entry(tb, node, &fib_table_hash[h], tb6_hlist) {
401 if (e < s_e)
402 goto next;
403 res = fib6_dump_table(tb, skb, cb);
404 if (res != 0)
405 goto out;
406 next:
407 e++;
408 }
409 }
410 out:
411 cb->args[1] = e;
412 cb->args[0] = h;
413
414 res = res < 0 ? res : skb->len;
415 if (res <= 0)
416 fib6_dump_end(cb);
417 return res;
418 }
419
420 /*
421 * Routing Table
422 *
423 * return the appropriate node for a routing tree "add" operation
424 * by either creating and inserting or by returning an existing
425 * node.
426 */
427
428 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
429 int addrlen, int plen,
430 int offset)
431 {
432 struct fib6_node *fn, *in, *ln;
433 struct fib6_node *pn = NULL;
434 struct rt6key *key;
435 int bit;
436 __be32 dir = 0;
437 __u32 sernum = fib6_new_sernum();
438
439 RT6_TRACE("fib6_add_1\n");
440
441 /* insert node in tree */
442
443 fn = root;
444
445 do {
446 key = (struct rt6key *)((u8 *)fn->leaf + offset);
447
448 /*
449 * Prefix match
450 */
451 if (plen < fn->fn_bit ||
452 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
453 goto insert_above;
454
455 /*
456 * Exact match ?
457 */
458
459 if (plen == fn->fn_bit) {
460 /* clean up an intermediate node */
461 if ((fn->fn_flags & RTN_RTINFO) == 0) {
462 rt6_release(fn->leaf);
463 fn->leaf = NULL;
464 }
465
466 fn->fn_sernum = sernum;
467
468 return fn;
469 }
470
471 /*
472 * We have more bits to go
473 */
474
475 /* Try to walk down on tree. */
476 fn->fn_sernum = sernum;
477 dir = addr_bit_set(addr, fn->fn_bit);
478 pn = fn;
479 fn = dir ? fn->right: fn->left;
480 } while (fn);
481
482 /*
483 * We walked to the bottom of tree.
484 * Create new leaf node without children.
485 */
486
487 ln = node_alloc();
488
489 if (ln == NULL)
490 return NULL;
491 ln->fn_bit = plen;
492
493 ln->parent = pn;
494 ln->fn_sernum = sernum;
495
496 if (dir)
497 pn->right = ln;
498 else
499 pn->left = ln;
500
501 return ln;
502
503
504 insert_above:
505 /*
506 * split since we don't have a common prefix anymore or
507 * we have a less significant route.
508 * we've to insert an intermediate node on the list
509 * this new node will point to the one we need to create
510 * and the current
511 */
512
513 pn = fn->parent;
514
515 /* find 1st bit in difference between the 2 addrs.
516
517 See comment in __ipv6_addr_diff: bit may be an invalid value,
518 but if it is >= plen, the value is ignored in any case.
519 */
520
521 bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
522
523 /*
524 * (intermediate)[in]
525 * / \
526 * (new leaf node)[ln] (old node)[fn]
527 */
528 if (plen > bit) {
529 in = node_alloc();
530 ln = node_alloc();
531
532 if (in == NULL || ln == NULL) {
533 if (in)
534 node_free(in);
535 if (ln)
536 node_free(ln);
537 return NULL;
538 }
539
540 /*
541 * new intermediate node.
542 * RTN_RTINFO will
543 * be off since that an address that chooses one of
544 * the branches would not match less specific routes
545 * in the other branch
546 */
547
548 in->fn_bit = bit;
549
550 in->parent = pn;
551 in->leaf = fn->leaf;
552 atomic_inc(&in->leaf->rt6i_ref);
553
554 in->fn_sernum = sernum;
555
556 /* update parent pointer */
557 if (dir)
558 pn->right = in;
559 else
560 pn->left = in;
561
562 ln->fn_bit = plen;
563
564 ln->parent = in;
565 fn->parent = in;
566
567 ln->fn_sernum = sernum;
568
569 if (addr_bit_set(addr, bit)) {
570 in->right = ln;
571 in->left = fn;
572 } else {
573 in->left = ln;
574 in->right = fn;
575 }
576 } else { /* plen <= bit */
577
578 /*
579 * (new leaf node)[ln]
580 * / \
581 * (old node)[fn] NULL
582 */
583
584 ln = node_alloc();
585
586 if (ln == NULL)
587 return NULL;
588
589 ln->fn_bit = plen;
590
591 ln->parent = pn;
592
593 ln->fn_sernum = sernum;
594
595 if (dir)
596 pn->right = ln;
597 else
598 pn->left = ln;
599
600 if (addr_bit_set(&key->addr, plen))
601 ln->right = fn;
602 else
603 ln->left = fn;
604
605 fn->parent = ln;
606 }
607 return ln;
608 }
609
610 /*
611 * Insert routing information in a node.
612 */
613
614 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
615 struct nl_info *info)
616 {
617 struct rt6_info *iter = NULL;
618 struct rt6_info **ins;
619
620 ins = &fn->leaf;
621
622 for (iter = fn->leaf; iter; iter=iter->u.dst.rt6_next) {
623 /*
624 * Search for duplicates
625 */
626
627 if (iter->rt6i_metric == rt->rt6i_metric) {
628 /*
629 * Same priority level
630 */
631
632 if (iter->rt6i_dev == rt->rt6i_dev &&
633 iter->rt6i_idev == rt->rt6i_idev &&
634 ipv6_addr_equal(&iter->rt6i_gateway,
635 &rt->rt6i_gateway)) {
636 if (!(iter->rt6i_flags&RTF_EXPIRES))
637 return -EEXIST;
638 iter->rt6i_expires = rt->rt6i_expires;
639 if (!(rt->rt6i_flags&RTF_EXPIRES)) {
640 iter->rt6i_flags &= ~RTF_EXPIRES;
641 iter->rt6i_expires = 0;
642 }
643 return -EEXIST;
644 }
645 }
646
647 if (iter->rt6i_metric > rt->rt6i_metric)
648 break;
649
650 ins = &iter->u.dst.rt6_next;
651 }
652
653 /* Reset round-robin state, if necessary */
654 if (ins == &fn->leaf)
655 fn->rr_ptr = NULL;
656
657 /*
658 * insert node
659 */
660
661 rt->u.dst.rt6_next = iter;
662 *ins = rt;
663 rt->rt6i_node = fn;
664 atomic_inc(&rt->rt6i_ref);
665 inet6_rt_notify(RTM_NEWROUTE, rt, info);
666 rt6_stats.fib_rt_entries++;
667
668 if ((fn->fn_flags & RTN_RTINFO) == 0) {
669 rt6_stats.fib_route_nodes++;
670 fn->fn_flags |= RTN_RTINFO;
671 }
672
673 return 0;
674 }
675
676 static __inline__ void fib6_start_gc(struct rt6_info *rt)
677 {
678 if (ip6_fib_timer.expires == 0 &&
679 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE)))
680 mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval);
681 }
682
683 void fib6_force_start_gc(void)
684 {
685 if (ip6_fib_timer.expires == 0)
686 mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval);
687 }
688
689 /*
690 * Add routing information to the routing tree.
691 * <destination addr>/<source addr>
692 * with source addr info in sub-trees
693 */
694
695 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
696 {
697 struct fib6_node *fn, *pn = NULL;
698 int err = -ENOMEM;
699
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));
702
703 if (fn == NULL)
704 goto out;
705
706 pn = fn;
707
708 #ifdef CONFIG_IPV6_SUBTREES
709 if (rt->rt6i_src.plen) {
710 struct fib6_node *sn;
711
712 if (fn->subtree == NULL) {
713 struct fib6_node *sfn;
714
715 /*
716 * Create subtree.
717 *
718 * fn[main tree]
719 * |
720 * sfn[subtree root]
721 * \
722 * sn[new leaf node]
723 */
724
725 /* Create subtree root node */
726 sfn = node_alloc();
727 if (sfn == NULL)
728 goto st_failure;
729
730 sfn->leaf = &ip6_null_entry;
731 atomic_inc(&ip6_null_entry.rt6i_ref);
732 sfn->fn_flags = RTN_ROOT;
733 sfn->fn_sernum = fib6_new_sernum();
734
735 /* Now add the first leaf node to new subtree */
736
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));
740
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.
745 */
746 node_free(sfn);
747 goto st_failure;
748 }
749
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));
757
758 if (sn == NULL)
759 goto st_failure;
760 }
761
762 if (fn->leaf == NULL) {
763 fn->leaf = rt;
764 atomic_inc(&rt->rt6i_ref);
765 }
766 fn = sn;
767 }
768 #endif
769
770 err = fib6_add_rt2node(fn, rt, info);
771
772 if (err == 0) {
773 fib6_start_gc(rt);
774 if (!(rt->rt6i_flags&RTF_CACHE))
775 fib6_prune_clones(pn, rt);
776 }
777
778 out:
779 if (err) {
780 #ifdef CONFIG_IPV6_SUBTREES
781 /*
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.
784 */
785 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
786 pn->leaf = fib6_find_prefix(pn);
787 #if RT6_DEBUG >= 2
788 if (!pn->leaf) {
789 BUG_TRAP(pn->leaf != NULL);
790 pn->leaf = &ip6_null_entry;
791 }
792 #endif
793 atomic_inc(&pn->leaf->rt6i_ref);
794 }
795 #endif
796 dst_free(&rt->u.dst);
797 }
798 return err;
799
800 #ifdef CONFIG_IPV6_SUBTREES
801 /* Subtree creation failed, probably main tree node
802 is orphan. If it is, shoot it.
803 */
804 st_failure:
805 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
806 fib6_repair_tree(fn);
807 dst_free(&rt->u.dst);
808 return err;
809 #endif
810 }
811
812 /*
813 * Routing tree lookup
814 *
815 */
816
817 struct lookup_args {
818 int offset; /* key offset on rt6_info */
819 struct in6_addr *addr; /* search key */
820 };
821
822 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
823 struct lookup_args *args)
824 {
825 struct fib6_node *fn;
826 __be32 dir;
827
828 if (unlikely(args->offset == 0))
829 return NULL;
830
831 /*
832 * Descend on a tree
833 */
834
835 fn = root;
836
837 for (;;) {
838 struct fib6_node *next;
839
840 dir = addr_bit_set(args->addr, fn->fn_bit);
841
842 next = dir ? fn->right : fn->left;
843
844 if (next) {
845 fn = next;
846 continue;
847 }
848
849 break;
850 }
851
852 while(fn) {
853 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
854 struct rt6key *key;
855
856 key = (struct rt6key *) ((u8 *) fn->leaf +
857 args->offset);
858
859 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
860 #ifdef CONFIG_IPV6_SUBTREES
861 if (fn->subtree)
862 fn = fib6_lookup_1(fn->subtree, args + 1);
863 #endif
864 if (!fn || fn->fn_flags & RTN_RTINFO)
865 return fn;
866 }
867 }
868
869 if (fn->fn_flags & RTN_ROOT)
870 break;
871
872 fn = fn->parent;
873 }
874
875 return NULL;
876 }
877
878 struct fib6_node * fib6_lookup(struct fib6_node *root, struct in6_addr *daddr,
879 struct in6_addr *saddr)
880 {
881 struct fib6_node *fn;
882 struct lookup_args args[] = {
883 {
884 .offset = offsetof(struct rt6_info, rt6i_dst),
885 .addr = daddr,
886 },
887 #ifdef CONFIG_IPV6_SUBTREES
888 {
889 .offset = offsetof(struct rt6_info, rt6i_src),
890 .addr = saddr,
891 },
892 #endif
893 {
894 .offset = 0, /* sentinel */
895 }
896 };
897
898 fn = fib6_lookup_1(root, daddr ? args : args + 1);
899
900 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT)
901 fn = root;
902
903 return fn;
904 }
905
906 /*
907 * Get node with specified destination prefix (and source prefix,
908 * if subtrees are used)
909 */
910
911
912 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
913 struct in6_addr *addr,
914 int plen, int offset)
915 {
916 struct fib6_node *fn;
917
918 for (fn = root; fn ; ) {
919 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
920
921 /*
922 * Prefix match
923 */
924 if (plen < fn->fn_bit ||
925 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
926 return NULL;
927
928 if (plen == fn->fn_bit)
929 return fn;
930
931 /*
932 * We have more bits to go
933 */
934 if (addr_bit_set(addr, fn->fn_bit))
935 fn = fn->right;
936 else
937 fn = fn->left;
938 }
939 return NULL;
940 }
941
942 struct fib6_node * fib6_locate(struct fib6_node *root,
943 struct in6_addr *daddr, int dst_len,
944 struct in6_addr *saddr, int src_len)
945 {
946 struct fib6_node *fn;
947
948 fn = fib6_locate_1(root, daddr, dst_len,
949 offsetof(struct rt6_info, rt6i_dst));
950
951 #ifdef CONFIG_IPV6_SUBTREES
952 if (src_len) {
953 BUG_TRAP(saddr!=NULL);
954 if (fn && fn->subtree)
955 fn = fib6_locate_1(fn->subtree, saddr, src_len,
956 offsetof(struct rt6_info, rt6i_src));
957 }
958 #endif
959
960 if (fn && fn->fn_flags&RTN_RTINFO)
961 return fn;
962
963 return NULL;
964 }
965
966
967 /*
968 * Deletion
969 *
970 */
971
972 static struct rt6_info * fib6_find_prefix(struct fib6_node *fn)
973 {
974 if (fn->fn_flags&RTN_ROOT)
975 return &ip6_null_entry;
976
977 while(fn) {
978 if(fn->left)
979 return fn->left->leaf;
980
981 if(fn->right)
982 return fn->right->leaf;
983
984 fn = FIB6_SUBTREE(fn);
985 }
986 return NULL;
987 }
988
989 /*
990 * Called to trim the tree of intermediate nodes when possible. "fn"
991 * is the node we want to try and remove.
992 */
993
994 static struct fib6_node * fib6_repair_tree(struct fib6_node *fn)
995 {
996 int children;
997 int nstate;
998 struct fib6_node *child, *pn;
999 struct fib6_walker_t *w;
1000 int iter = 0;
1001
1002 for (;;) {
1003 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1004 iter++;
1005
1006 BUG_TRAP(!(fn->fn_flags&RTN_RTINFO));
1007 BUG_TRAP(!(fn->fn_flags&RTN_TL_ROOT));
1008 BUG_TRAP(fn->leaf==NULL);
1009
1010 children = 0;
1011 child = NULL;
1012 if (fn->right) child = fn->right, children |= 1;
1013 if (fn->left) child = fn->left, children |= 2;
1014
1015 if (children == 3 || FIB6_SUBTREE(fn)
1016 #ifdef CONFIG_IPV6_SUBTREES
1017 /* Subtree root (i.e. fn) may have one child */
1018 || (children && fn->fn_flags&RTN_ROOT)
1019 #endif
1020 ) {
1021 fn->leaf = fib6_find_prefix(fn);
1022 #if RT6_DEBUG >= 2
1023 if (fn->leaf==NULL) {
1024 BUG_TRAP(fn->leaf);
1025 fn->leaf = &ip6_null_entry;
1026 }
1027 #endif
1028 atomic_inc(&fn->leaf->rt6i_ref);
1029 return fn->parent;
1030 }
1031
1032 pn = fn->parent;
1033 #ifdef CONFIG_IPV6_SUBTREES
1034 if (FIB6_SUBTREE(pn) == fn) {
1035 BUG_TRAP(fn->fn_flags&RTN_ROOT);
1036 FIB6_SUBTREE(pn) = NULL;
1037 nstate = FWS_L;
1038 } else {
1039 BUG_TRAP(!(fn->fn_flags&RTN_ROOT));
1040 #endif
1041 if (pn->right == fn) pn->right = child;
1042 else if (pn->left == fn) pn->left = child;
1043 #if RT6_DEBUG >= 2
1044 else BUG_TRAP(0);
1045 #endif
1046 if (child)
1047 child->parent = pn;
1048 nstate = FWS_R;
1049 #ifdef CONFIG_IPV6_SUBTREES
1050 }
1051 #endif
1052
1053 read_lock(&fib6_walker_lock);
1054 FOR_WALKERS(w) {
1055 if (child == NULL) {
1056 if (w->root == fn) {
1057 w->root = w->node = NULL;
1058 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1059 } else if (w->node == fn) {
1060 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1061 w->node = pn;
1062 w->state = nstate;
1063 }
1064 } else {
1065 if (w->root == fn) {
1066 w->root = child;
1067 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1068 }
1069 if (w->node == fn) {
1070 w->node = child;
1071 if (children&2) {
1072 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1073 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1074 } else {
1075 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1076 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1077 }
1078 }
1079 }
1080 }
1081 read_unlock(&fib6_walker_lock);
1082
1083 node_free(fn);
1084 if (pn->fn_flags&RTN_RTINFO || FIB6_SUBTREE(pn))
1085 return pn;
1086
1087 rt6_release(pn->leaf);
1088 pn->leaf = NULL;
1089 fn = pn;
1090 }
1091 }
1092
1093 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1094 struct nl_info *info)
1095 {
1096 struct fib6_walker_t *w;
1097 struct rt6_info *rt = *rtp;
1098
1099 RT6_TRACE("fib6_del_route\n");
1100
1101 /* Unlink it */
1102 *rtp = rt->u.dst.rt6_next;
1103 rt->rt6i_node = NULL;
1104 rt6_stats.fib_rt_entries--;
1105 rt6_stats.fib_discarded_routes++;
1106
1107 /* Reset round-robin state, if necessary */
1108 if (fn->rr_ptr == rt)
1109 fn->rr_ptr = NULL;
1110
1111 /* Adjust walkers */
1112 read_lock(&fib6_walker_lock);
1113 FOR_WALKERS(w) {
1114 if (w->state == FWS_C && w->leaf == rt) {
1115 RT6_TRACE("walker %p adjusted by delroute\n", w);
1116 w->leaf = rt->u.dst.rt6_next;
1117 if (w->leaf == NULL)
1118 w->state = FWS_U;
1119 }
1120 }
1121 read_unlock(&fib6_walker_lock);
1122
1123 rt->u.dst.rt6_next = NULL;
1124
1125 if (fn->leaf == NULL && fn->fn_flags&RTN_TL_ROOT)
1126 fn->leaf = &ip6_null_entry;
1127
1128 /* If it was last route, expunge its radix tree node */
1129 if (fn->leaf == NULL) {
1130 fn->fn_flags &= ~RTN_RTINFO;
1131 rt6_stats.fib_route_nodes--;
1132 fn = fib6_repair_tree(fn);
1133 }
1134
1135 if (atomic_read(&rt->rt6i_ref) != 1) {
1136 /* This route is used as dummy address holder in some split
1137 * nodes. It is not leaked, but it still holds other resources,
1138 * which must be released in time. So, scan ascendant nodes
1139 * and replace dummy references to this route with references
1140 * to still alive ones.
1141 */
1142 while (fn) {
1143 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) {
1144 fn->leaf = fib6_find_prefix(fn);
1145 atomic_inc(&fn->leaf->rt6i_ref);
1146 rt6_release(rt);
1147 }
1148 fn = fn->parent;
1149 }
1150 /* No more references are possible at this point. */
1151 if (atomic_read(&rt->rt6i_ref) != 1) BUG();
1152 }
1153
1154 inet6_rt_notify(RTM_DELROUTE, rt, info);
1155 rt6_release(rt);
1156 }
1157
1158 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1159 {
1160 struct fib6_node *fn = rt->rt6i_node;
1161 struct rt6_info **rtp;
1162
1163 #if RT6_DEBUG >= 2
1164 if (rt->u.dst.obsolete>0) {
1165 BUG_TRAP(fn==NULL);
1166 return -ENOENT;
1167 }
1168 #endif
1169 if (fn == NULL || rt == &ip6_null_entry)
1170 return -ENOENT;
1171
1172 BUG_TRAP(fn->fn_flags&RTN_RTINFO);
1173
1174 if (!(rt->rt6i_flags&RTF_CACHE)) {
1175 struct fib6_node *pn = fn;
1176 #ifdef CONFIG_IPV6_SUBTREES
1177 /* clones of this route might be in another subtree */
1178 if (rt->rt6i_src.plen) {
1179 while (!(pn->fn_flags&RTN_ROOT))
1180 pn = pn->parent;
1181 pn = pn->parent;
1182 }
1183 #endif
1184 fib6_prune_clones(pn, rt);
1185 }
1186
1187 /*
1188 * Walk the leaf entries looking for ourself
1189 */
1190
1191 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->u.dst.rt6_next) {
1192 if (*rtp == rt) {
1193 fib6_del_route(fn, rtp, info);
1194 return 0;
1195 }
1196 }
1197 return -ENOENT;
1198 }
1199
1200 /*
1201 * Tree traversal function.
1202 *
1203 * Certainly, it is not interrupt safe.
1204 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1205 * It means, that we can modify tree during walking
1206 * and use this function for garbage collection, clone pruning,
1207 * cleaning tree when a device goes down etc. etc.
1208 *
1209 * It guarantees that every node will be traversed,
1210 * and that it will be traversed only once.
1211 *
1212 * Callback function w->func may return:
1213 * 0 -> continue walking.
1214 * positive value -> walking is suspended (used by tree dumps,
1215 * and probably by gc, if it will be split to several slices)
1216 * negative value -> terminate walking.
1217 *
1218 * The function itself returns:
1219 * 0 -> walk is complete.
1220 * >0 -> walk is incomplete (i.e. suspended)
1221 * <0 -> walk is terminated by an error.
1222 */
1223
1224 static int fib6_walk_continue(struct fib6_walker_t *w)
1225 {
1226 struct fib6_node *fn, *pn;
1227
1228 for (;;) {
1229 fn = w->node;
1230 if (fn == NULL)
1231 return 0;
1232
1233 if (w->prune && fn != w->root &&
1234 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) {
1235 w->state = FWS_C;
1236 w->leaf = fn->leaf;
1237 }
1238 switch (w->state) {
1239 #ifdef CONFIG_IPV6_SUBTREES
1240 case FWS_S:
1241 if (FIB6_SUBTREE(fn)) {
1242 w->node = FIB6_SUBTREE(fn);
1243 continue;
1244 }
1245 w->state = FWS_L;
1246 #endif
1247 case FWS_L:
1248 if (fn->left) {
1249 w->node = fn->left;
1250 w->state = FWS_INIT;
1251 continue;
1252 }
1253 w->state = FWS_R;
1254 case FWS_R:
1255 if (fn->right) {
1256 w->node = fn->right;
1257 w->state = FWS_INIT;
1258 continue;
1259 }
1260 w->state = FWS_C;
1261 w->leaf = fn->leaf;
1262 case FWS_C:
1263 if (w->leaf && fn->fn_flags&RTN_RTINFO) {
1264 int err = w->func(w);
1265 if (err)
1266 return err;
1267 continue;
1268 }
1269 w->state = FWS_U;
1270 case FWS_U:
1271 if (fn == w->root)
1272 return 0;
1273 pn = fn->parent;
1274 w->node = pn;
1275 #ifdef CONFIG_IPV6_SUBTREES
1276 if (FIB6_SUBTREE(pn) == fn) {
1277 BUG_TRAP(fn->fn_flags&RTN_ROOT);
1278 w->state = FWS_L;
1279 continue;
1280 }
1281 #endif
1282 if (pn->left == fn) {
1283 w->state = FWS_R;
1284 continue;
1285 }
1286 if (pn->right == fn) {
1287 w->state = FWS_C;
1288 w->leaf = w->node->leaf;
1289 continue;
1290 }
1291 #if RT6_DEBUG >= 2
1292 BUG_TRAP(0);
1293 #endif
1294 }
1295 }
1296 }
1297
1298 static int fib6_walk(struct fib6_walker_t *w)
1299 {
1300 int res;
1301
1302 w->state = FWS_INIT;
1303 w->node = w->root;
1304
1305 fib6_walker_link(w);
1306 res = fib6_walk_continue(w);
1307 if (res <= 0)
1308 fib6_walker_unlink(w);
1309 return res;
1310 }
1311
1312 static int fib6_clean_node(struct fib6_walker_t *w)
1313 {
1314 int res;
1315 struct rt6_info *rt;
1316 struct fib6_cleaner_t *c = (struct fib6_cleaner_t*)w;
1317
1318 for (rt = w->leaf; rt; rt = rt->u.dst.rt6_next) {
1319 res = c->func(rt, c->arg);
1320 if (res < 0) {
1321 w->leaf = rt;
1322 res = fib6_del(rt, NULL);
1323 if (res) {
1324 #if RT6_DEBUG >= 2
1325 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res);
1326 #endif
1327 continue;
1328 }
1329 return 0;
1330 }
1331 BUG_TRAP(res==0);
1332 }
1333 w->leaf = rt;
1334 return 0;
1335 }
1336
1337 /*
1338 * Convenient frontend to tree walker.
1339 *
1340 * func is called on each route.
1341 * It may return -1 -> delete this route.
1342 * 0 -> continue walking
1343 *
1344 * prune==1 -> only immediate children of node (certainly,
1345 * ignoring pure split nodes) will be scanned.
1346 */
1347
1348 static void fib6_clean_tree(struct fib6_node *root,
1349 int (*func)(struct rt6_info *, void *arg),
1350 int prune, void *arg)
1351 {
1352 struct fib6_cleaner_t c;
1353
1354 c.w.root = root;
1355 c.w.func = fib6_clean_node;
1356 c.w.prune = prune;
1357 c.func = func;
1358 c.arg = arg;
1359
1360 fib6_walk(&c.w);
1361 }
1362
1363 void fib6_clean_all(int (*func)(struct rt6_info *, void *arg),
1364 int prune, void *arg)
1365 {
1366 struct fib6_table *table;
1367 struct hlist_node *node;
1368 unsigned int h;
1369
1370 rcu_read_lock();
1371 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
1372 hlist_for_each_entry_rcu(table, node, &fib_table_hash[h],
1373 tb6_hlist) {
1374 write_lock_bh(&table->tb6_lock);
1375 fib6_clean_tree(&table->tb6_root, func, prune, arg);
1376 write_unlock_bh(&table->tb6_lock);
1377 }
1378 }
1379 rcu_read_unlock();
1380 }
1381
1382 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1383 {
1384 if (rt->rt6i_flags & RTF_CACHE) {
1385 RT6_TRACE("pruning clone %p\n", rt);
1386 return -1;
1387 }
1388
1389 return 0;
1390 }
1391
1392 static void fib6_prune_clones(struct fib6_node *fn, struct rt6_info *rt)
1393 {
1394 fib6_clean_tree(fn, fib6_prune_clone, 1, rt);
1395 }
1396
1397 /*
1398 * Garbage collection
1399 */
1400
1401 static struct fib6_gc_args
1402 {
1403 int timeout;
1404 int more;
1405 } gc_args;
1406
1407 static int fib6_age(struct rt6_info *rt, void *arg)
1408 {
1409 unsigned long now = jiffies;
1410
1411 /*
1412 * check addrconf expiration here.
1413 * Routes are expired even if they are in use.
1414 *
1415 * Also age clones. Note, that clones are aged out
1416 * only if they are not in use now.
1417 */
1418
1419 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) {
1420 if (time_after(now, rt->rt6i_expires)) {
1421 RT6_TRACE("expiring %p\n", rt);
1422 return -1;
1423 }
1424 gc_args.more++;
1425 } else if (rt->rt6i_flags & RTF_CACHE) {
1426 if (atomic_read(&rt->u.dst.__refcnt) == 0 &&
1427 time_after_eq(now, rt->u.dst.lastuse + gc_args.timeout)) {
1428 RT6_TRACE("aging clone %p\n", rt);
1429 return -1;
1430 } else if ((rt->rt6i_flags & RTF_GATEWAY) &&
1431 (!(rt->rt6i_nexthop->flags & NTF_ROUTER))) {
1432 RT6_TRACE("purging route %p via non-router but gateway\n",
1433 rt);
1434 return -1;
1435 }
1436 gc_args.more++;
1437 }
1438
1439 return 0;
1440 }
1441
1442 static DEFINE_SPINLOCK(fib6_gc_lock);
1443
1444 void fib6_run_gc(unsigned long dummy)
1445 {
1446 if (dummy != ~0UL) {
1447 spin_lock_bh(&fib6_gc_lock);
1448 gc_args.timeout = dummy ? (int)dummy : ip6_rt_gc_interval;
1449 } else {
1450 local_bh_disable();
1451 if (!spin_trylock(&fib6_gc_lock)) {
1452 mod_timer(&ip6_fib_timer, jiffies + HZ);
1453 local_bh_enable();
1454 return;
1455 }
1456 gc_args.timeout = ip6_rt_gc_interval;
1457 }
1458 gc_args.more = 0;
1459
1460 ndisc_dst_gc(&gc_args.more);
1461 fib6_clean_all(fib6_age, 0, NULL);
1462
1463 if (gc_args.more)
1464 mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval);
1465 else {
1466 del_timer(&ip6_fib_timer);
1467 ip6_fib_timer.expires = 0;
1468 }
1469 spin_unlock_bh(&fib6_gc_lock);
1470 }
1471
1472 void __init fib6_init(void)
1473 {
1474 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1475 sizeof(struct fib6_node),
1476 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1477 NULL, NULL);
1478
1479 fib6_tables_init();
1480
1481 __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib);
1482 }
1483
1484 void fib6_gc_cleanup(void)
1485 {
1486 del_timer(&ip6_fib_timer);
1487 kmem_cache_destroy(fib6_node_kmem);
1488 }
Tomato firmware and modifications.