Merge tag 'scsi-misc' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi...
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / net / ipv6 / ip6_fib.c
1 /*
2 * Linux INET6 implementation
3 * Forwarding Information Database
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
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.
12 */
13
14 /*
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.
20 */
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>
30
31 #include <net/ipv6.h>
32 #include <net/ndisc.h>
33 #include <net/addrconf.h>
34
35 #include <net/ip6_fib.h>
36 #include <net/ip6_route.h>
37
38 #define RT6_DEBUG 2
39
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
45
46 static struct kmem_cache * fib6_node_kmem __read_mostly;
47
48 enum fib_walk_state_t
49 {
50 #ifdef CONFIG_IPV6_SUBTREES
51 FWS_S,
52 #endif
53 FWS_L,
54 FWS_R,
55 FWS_C,
56 FWS_U
57 };
58
59 struct fib6_cleaner_t
60 {
61 struct fib6_walker_t w;
62 struct net *net;
63 int (*func)(struct rt6_info *, void *arg);
64 void *arg;
65 };
66
67 static DEFINE_RWLOCK(fib6_walker_lock);
68
69 #ifdef CONFIG_IPV6_SUBTREES
70 #define FWS_INIT FWS_S
71 #else
72 #define FWS_INIT FWS_L
73 #endif
74
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);
81
82 /*
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.
87 */
88
89 static __u32 rt_sernum;
90
91 static void fib6_gc_timer_cb(unsigned long arg);
92
93 static LIST_HEAD(fib6_walkers);
94 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
95
96 static inline void fib6_walker_link(struct fib6_walker_t *w)
97 {
98 write_lock_bh(&fib6_walker_lock);
99 list_add(&w->lh, &fib6_walkers);
100 write_unlock_bh(&fib6_walker_lock);
101 }
102
103 static inline void fib6_walker_unlink(struct fib6_walker_t *w)
104 {
105 write_lock_bh(&fib6_walker_lock);
106 list_del(&w->lh);
107 write_unlock_bh(&fib6_walker_lock);
108 }
109 static __inline__ u32 fib6_new_sernum(void)
110 {
111 u32 n = ++rt_sernum;
112 if ((__s32)n <= 0)
113 rt_sernum = n = 1;
114 return n;
115 }
116
117 /*
118 * Auxiliary address test functions for the radix tree.
119 *
120 * These assume a 32bit processor (although it will work on
121 * 64bit processors)
122 */
123
124 /*
125 * test bit
126 */
127 #if defined(__LITTLE_ENDIAN)
128 # define BITOP_BE32_SWIZZLE (0x1F & ~7)
129 #else
130 # define BITOP_BE32_SWIZZLE 0
131 #endif
132
133 static __inline__ __be32 addr_bit_set(const void *token, int fn_bit)
134 {
135 const __be32 *addr = token;
136 /*
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.
142 */
143 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
144 addr[fn_bit >> 5];
145 }
146
147 static __inline__ struct fib6_node * node_alloc(void)
148 {
149 struct fib6_node *fn;
150
151 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
152
153 return fn;
154 }
155
156 static __inline__ void node_free(struct fib6_node * fn)
157 {
158 kmem_cache_free(fib6_node_kmem, fn);
159 }
160
161 static __inline__ void rt6_release(struct rt6_info *rt)
162 {
163 if (atomic_dec_and_test(&rt->rt6i_ref))
164 dst_free(&rt->dst);
165 }
166
167 static void fib6_link_table(struct net *net, struct fib6_table *tb)
168 {
169 unsigned int h;
170
171 /*
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.
174 */
175 rwlock_init(&tb->tb6_lock);
176
177 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
178
179 /*
180 * No protection necessary, this is the only list mutatation
181 * operation, tables never disappear once they exist.
182 */
183 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
184 }
185
186 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
187
188 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
189 {
190 struct fib6_table *table;
191
192 table = kzalloc(sizeof(*table), GFP_ATOMIC);
193 if (table) {
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;
197 }
198
199 return table;
200 }
201
202 struct fib6_table *fib6_new_table(struct net *net, u32 id)
203 {
204 struct fib6_table *tb;
205
206 if (id == 0)
207 id = RT6_TABLE_MAIN;
208 tb = fib6_get_table(net, id);
209 if (tb)
210 return tb;
211
212 tb = fib6_alloc_table(net, id);
213 if (tb)
214 fib6_link_table(net, tb);
215
216 return tb;
217 }
218
219 struct fib6_table *fib6_get_table(struct net *net, u32 id)
220 {
221 struct fib6_table *tb;
222 struct hlist_head *head;
223 struct hlist_node *node;
224 unsigned int h;
225
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;
235 }
236 }
237 rcu_read_unlock();
238
239 return NULL;
240 }
241
242 static void __net_init fib6_tables_init(struct net *net)
243 {
244 fib6_link_table(net, net->ipv6.fib6_main_tbl);
245 fib6_link_table(net, net->ipv6.fib6_local_tbl);
246 }
247 #else
248
249 struct fib6_table *fib6_new_table(struct net *net, u32 id)
250 {
251 return fib6_get_table(net, id);
252 }
253
254 struct fib6_table *fib6_get_table(struct net *net, u32 id)
255 {
256 return net->ipv6.fib6_main_tbl;
257 }
258
259 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
260 int flags, pol_lookup_t lookup)
261 {
262 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
263 }
264
265 static void __net_init fib6_tables_init(struct net *net)
266 {
267 fib6_link_table(net, net->ipv6.fib6_main_tbl);
268 }
269
270 #endif
271
272 static int fib6_dump_node(struct fib6_walker_t *w)
273 {
274 int res;
275 struct rt6_info *rt;
276
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;
283 }
284 WARN_ON(res == 0);
285 }
286 w->leaf = NULL;
287 return 0;
288 }
289
290 static void fib6_dump_end(struct netlink_callback *cb)
291 {
292 struct fib6_walker_t *w = (void*)cb->args[2];
293
294 if (w) {
295 if (cb->args[4]) {
296 cb->args[4] = 0;
297 fib6_walker_unlink(w);
298 }
299 cb->args[2] = 0;
300 kfree(w);
301 }
302 cb->done = (void*)cb->args[3];
303 cb->args[1] = 3;
304 }
305
306 static int fib6_dump_done(struct netlink_callback *cb)
307 {
308 fib6_dump_end(cb);
309 return cb->done ? cb->done(cb) : 0;
310 }
311
312 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
313 struct netlink_callback *cb)
314 {
315 struct fib6_walker_t *w;
316 int res;
317
318 w = (void *)cb->args[2];
319 w->root = &table->tb6_root;
320
321 if (cb->args[4] == 0) {
322 w->count = 0;
323 w->skip = 0;
324
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;
331 }
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;
341
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;
348 }
349 }
350
351 return res;
352 }
353
354 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
355 {
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;
365
366 s_h = cb->args[0];
367 s_e = cb->args[1];
368
369 w = (void *)cb->args[2];
370 if (!w) {
371 /* New dump:
372 *
373 * 1. hook callback destructor.
374 */
375 cb->args[3] = (long)cb->done;
376 cb->done = fib6_dump_done;
377
378 /*
379 * 2. allocate and initialize walker.
380 */
381 w = kzalloc(sizeof(*w), GFP_ATOMIC);
382 if (!w)
383 return -ENOMEM;
384 w->func = fib6_dump_node;
385 cb->args[2] = (long)w;
386 }
387
388 arg.skb = skb;
389 arg.cb = cb;
390 arg.net = net;
391 w->args = &arg;
392
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++;
405 }
406 }
407 out:
408 rcu_read_unlock();
409 cb->args[1] = e;
410 cb->args[0] = h;
411
412 res = res < 0 ? res : skb->len;
413 if (res <= 0)
414 fib6_dump_end(cb);
415 return res;
416 }
417
418 /*
419 * Routing Table
420 *
421 * return the appropriate node for a routing tree "add" operation
422 * by either creating and inserting or by returning an existing
423 * node.
424 */
425
426 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
427 int addrlen, int plen,
428 int offset, int allow_create,
429 int replace_required)
430 {
431 struct fib6_node *fn, *in, *ln;
432 struct fib6_node *pn = NULL;
433 struct rt6key *key;
434 int bit;
435 __be32 dir = 0;
436 __u32 sernum = fib6_new_sernum();
437
438 RT6_TRACE("fib6_add_1\n");
439
440 /* insert node in tree */
441
442 fn = root;
443
444 do {
445 key = (struct rt6key *)((u8 *)fn->leaf + offset);
446
447 /*
448 * Prefix match
449 */
450 if (plen < fn->fn_bit ||
451 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
452 if (!allow_create) {
453 if (replace_required) {
454 pr_warn("IPv6: Can't replace route, "
455 "no match found\n");
456 return ERR_PTR(-ENOENT);
457 }
458 pr_warn("IPv6: NLM_F_CREATE should be set "
459 "when creating new route\n");
460 }
461 goto insert_above;
462 }
463
464 /*
465 * Exact match ?
466 */
467
468 if (plen == fn->fn_bit) {
469 /* clean up an intermediate node */
470 if (!(fn->fn_flags & RTN_RTINFO)) {
471 rt6_release(fn->leaf);
472 fn->leaf = NULL;
473 }
474
475 fn->fn_sernum = sernum;
476
477 return fn;
478 }
479
480 /*
481 * We have more bits to go
482 */
483
484 /* Try to walk down on tree. */
485 fn->fn_sernum = sernum;
486 dir = addr_bit_set(addr, fn->fn_bit);
487 pn = fn;
488 fn = dir ? fn->right: fn->left;
489 } while (fn);
490
491 if (!allow_create) {
492 /* We should not create new node because
493 * NLM_F_REPLACE was specified without NLM_F_CREATE
494 * I assume it is safe to require NLM_F_CREATE when
495 * REPLACE flag is used! Later we may want to remove the
496 * check for replace_required, because according
497 * to netlink specification, NLM_F_CREATE
498 * MUST be specified if new route is created.
499 * That would keep IPv6 consistent with IPv4
500 */
501 if (replace_required) {
502 pr_warn("IPv6: Can't replace route, no match found\n");
503 return ERR_PTR(-ENOENT);
504 }
505 pr_warn("IPv6: NLM_F_CREATE should be set "
506 "when creating new route\n");
507 }
508 /*
509 * We walked to the bottom of tree.
510 * Create new leaf node without children.
511 */
512
513 ln = node_alloc();
514
515 if (!ln)
516 return NULL;
517 ln->fn_bit = plen;
518
519 ln->parent = pn;
520 ln->fn_sernum = sernum;
521
522 if (dir)
523 pn->right = ln;
524 else
525 pn->left = ln;
526
527 return ln;
528
529
530 insert_above:
531 /*
532 * split since we don't have a common prefix anymore or
533 * we have a less significant route.
534 * we've to insert an intermediate node on the list
535 * this new node will point to the one we need to create
536 * and the current
537 */
538
539 pn = fn->parent;
540
541 /* find 1st bit in difference between the 2 addrs.
542
543 See comment in __ipv6_addr_diff: bit may be an invalid value,
544 but if it is >= plen, the value is ignored in any case.
545 */
546
547 bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
548
549 /*
550 * (intermediate)[in]
551 * / \
552 * (new leaf node)[ln] (old node)[fn]
553 */
554 if (plen > bit) {
555 in = node_alloc();
556 ln = node_alloc();
557
558 if (!in || !ln) {
559 if (in)
560 node_free(in);
561 if (ln)
562 node_free(ln);
563 return NULL;
564 }
565
566 /*
567 * new intermediate node.
568 * RTN_RTINFO will
569 * be off since that an address that chooses one of
570 * the branches would not match less specific routes
571 * in the other branch
572 */
573
574 in->fn_bit = bit;
575
576 in->parent = pn;
577 in->leaf = fn->leaf;
578 atomic_inc(&in->leaf->rt6i_ref);
579
580 in->fn_sernum = sernum;
581
582 /* update parent pointer */
583 if (dir)
584 pn->right = in;
585 else
586 pn->left = in;
587
588 ln->fn_bit = plen;
589
590 ln->parent = in;
591 fn->parent = in;
592
593 ln->fn_sernum = sernum;
594
595 if (addr_bit_set(addr, bit)) {
596 in->right = ln;
597 in->left = fn;
598 } else {
599 in->left = ln;
600 in->right = fn;
601 }
602 } else { /* plen <= bit */
603
604 /*
605 * (new leaf node)[ln]
606 * / \
607 * (old node)[fn] NULL
608 */
609
610 ln = node_alloc();
611
612 if (!ln)
613 return NULL;
614
615 ln->fn_bit = plen;
616
617 ln->parent = pn;
618
619 ln->fn_sernum = sernum;
620
621 if (dir)
622 pn->right = ln;
623 else
624 pn->left = ln;
625
626 if (addr_bit_set(&key->addr, plen))
627 ln->right = fn;
628 else
629 ln->left = fn;
630
631 fn->parent = ln;
632 }
633 return ln;
634 }
635
636 /*
637 * Insert routing information in a node.
638 */
639
640 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
641 struct nl_info *info)
642 {
643 struct rt6_info *iter = NULL;
644 struct rt6_info **ins;
645 int replace = (info->nlh &&
646 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
647 int add = (!info->nlh ||
648 (info->nlh->nlmsg_flags & NLM_F_CREATE));
649 int found = 0;
650
651 ins = &fn->leaf;
652
653 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
654 /*
655 * Search for duplicates
656 */
657
658 if (iter->rt6i_metric == rt->rt6i_metric) {
659 /*
660 * Same priority level
661 */
662 if (info->nlh &&
663 (info->nlh->nlmsg_flags & NLM_F_EXCL))
664 return -EEXIST;
665 if (replace) {
666 found++;
667 break;
668 }
669
670 if (iter->dst.dev == rt->dst.dev &&
671 iter->rt6i_idev == rt->rt6i_idev &&
672 ipv6_addr_equal(&iter->rt6i_gateway,
673 &rt->rt6i_gateway)) {
674 if (!(iter->rt6i_flags & RTF_EXPIRES))
675 return -EEXIST;
676 iter->dst.expires = rt->dst.expires;
677 if (!(rt->rt6i_flags & RTF_EXPIRES)) {
678 iter->rt6i_flags &= ~RTF_EXPIRES;
679 iter->dst.expires = 0;
680 }
681 return -EEXIST;
682 }
683 }
684
685 if (iter->rt6i_metric > rt->rt6i_metric)
686 break;
687
688 ins = &iter->dst.rt6_next;
689 }
690
691 /* Reset round-robin state, if necessary */
692 if (ins == &fn->leaf)
693 fn->rr_ptr = NULL;
694
695 /*
696 * insert node
697 */
698 if (!replace) {
699 if (!add)
700 pr_warn("IPv6: NLM_F_CREATE should be set when creating new route\n");
701
702 add:
703 rt->dst.rt6_next = iter;
704 *ins = rt;
705 rt->rt6i_node = fn;
706 atomic_inc(&rt->rt6i_ref);
707 inet6_rt_notify(RTM_NEWROUTE, rt, info);
708 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
709
710 if (!(fn->fn_flags & RTN_RTINFO)) {
711 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
712 fn->fn_flags |= RTN_RTINFO;
713 }
714
715 } else {
716 if (!found) {
717 if (add)
718 goto add;
719 pr_warn("IPv6: NLM_F_REPLACE set, but no existing node found!\n");
720 return -ENOENT;
721 }
722 *ins = rt;
723 rt->rt6i_node = fn;
724 rt->dst.rt6_next = iter->dst.rt6_next;
725 atomic_inc(&rt->rt6i_ref);
726 inet6_rt_notify(RTM_NEWROUTE, rt, info);
727 rt6_release(iter);
728 if (!(fn->fn_flags & RTN_RTINFO)) {
729 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
730 fn->fn_flags |= RTN_RTINFO;
731 }
732 }
733
734 return 0;
735 }
736
737 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
738 {
739 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
740 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
741 mod_timer(&net->ipv6.ip6_fib_timer,
742 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
743 }
744
745 void fib6_force_start_gc(struct net *net)
746 {
747 if (!timer_pending(&net->ipv6.ip6_fib_timer))
748 mod_timer(&net->ipv6.ip6_fib_timer,
749 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
750 }
751
752 /*
753 * Add routing information to the routing tree.
754 * <destination addr>/<source addr>
755 * with source addr info in sub-trees
756 */
757
758 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
759 {
760 struct fib6_node *fn, *pn = NULL;
761 int err = -ENOMEM;
762 int allow_create = 1;
763 int replace_required = 0;
764
765 if (info->nlh) {
766 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
767 allow_create = 0;
768 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
769 replace_required = 1;
770 }
771 if (!allow_create && !replace_required)
772 pr_warn("IPv6: RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
773
774 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
775 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst),
776 allow_create, replace_required);
777
778 if (IS_ERR(fn)) {
779 err = PTR_ERR(fn);
780 fn = NULL;
781 }
782
783 if (!fn)
784 goto out;
785
786 pn = fn;
787
788 #ifdef CONFIG_IPV6_SUBTREES
789 if (rt->rt6i_src.plen) {
790 struct fib6_node *sn;
791
792 if (!fn->subtree) {
793 struct fib6_node *sfn;
794
795 /*
796 * Create subtree.
797 *
798 * fn[main tree]
799 * |
800 * sfn[subtree root]
801 * \
802 * sn[new leaf node]
803 */
804
805 /* Create subtree root node */
806 sfn = node_alloc();
807 if (!sfn)
808 goto st_failure;
809
810 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
811 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
812 sfn->fn_flags = RTN_ROOT;
813 sfn->fn_sernum = fib6_new_sernum();
814
815 /* Now add the first leaf node to new subtree */
816
817 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
818 sizeof(struct in6_addr), rt->rt6i_src.plen,
819 offsetof(struct rt6_info, rt6i_src),
820 allow_create, replace_required);
821
822 if (!sn) {
823 /* If it is failed, discard just allocated
824 root, and then (in st_failure) stale node
825 in main tree.
826 */
827 node_free(sfn);
828 goto st_failure;
829 }
830
831 /* Now link new subtree to main tree */
832 sfn->parent = fn;
833 fn->subtree = sfn;
834 } else {
835 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
836 sizeof(struct in6_addr), rt->rt6i_src.plen,
837 offsetof(struct rt6_info, rt6i_src),
838 allow_create, replace_required);
839
840 if (IS_ERR(sn)) {
841 err = PTR_ERR(sn);
842 sn = NULL;
843 }
844 if (!sn)
845 goto st_failure;
846 }
847
848 if (!fn->leaf) {
849 fn->leaf = rt;
850 atomic_inc(&rt->rt6i_ref);
851 }
852 fn = sn;
853 }
854 #endif
855
856 err = fib6_add_rt2node(fn, rt, info);
857 if (!err) {
858 fib6_start_gc(info->nl_net, rt);
859 if (!(rt->rt6i_flags & RTF_CACHE))
860 fib6_prune_clones(info->nl_net, pn, rt);
861 }
862
863 out:
864 if (err) {
865 #ifdef CONFIG_IPV6_SUBTREES
866 /*
867 * If fib6_add_1 has cleared the old leaf pointer in the
868 * super-tree leaf node we have to find a new one for it.
869 */
870 if (pn != fn && pn->leaf == rt) {
871 pn->leaf = NULL;
872 atomic_dec(&rt->rt6i_ref);
873 }
874 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
875 pn->leaf = fib6_find_prefix(info->nl_net, pn);
876 #if RT6_DEBUG >= 2
877 if (!pn->leaf) {
878 WARN_ON(pn->leaf == NULL);
879 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
880 }
881 #endif
882 atomic_inc(&pn->leaf->rt6i_ref);
883 }
884 #endif
885 dst_free(&rt->dst);
886 }
887 return err;
888
889 #ifdef CONFIG_IPV6_SUBTREES
890 /* Subtree creation failed, probably main tree node
891 is orphan. If it is, shoot it.
892 */
893 st_failure:
894 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
895 fib6_repair_tree(info->nl_net, fn);
896 dst_free(&rt->dst);
897 return err;
898 #endif
899 }
900
901 /*
902 * Routing tree lookup
903 *
904 */
905
906 struct lookup_args {
907 int offset; /* key offset on rt6_info */
908 const struct in6_addr *addr; /* search key */
909 };
910
911 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
912 struct lookup_args *args)
913 {
914 struct fib6_node *fn;
915 __be32 dir;
916
917 if (unlikely(args->offset == 0))
918 return NULL;
919
920 /*
921 * Descend on a tree
922 */
923
924 fn = root;
925
926 for (;;) {
927 struct fib6_node *next;
928
929 dir = addr_bit_set(args->addr, fn->fn_bit);
930
931 next = dir ? fn->right : fn->left;
932
933 if (next) {
934 fn = next;
935 continue;
936 }
937 break;
938 }
939
940 while (fn) {
941 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
942 struct rt6key *key;
943
944 key = (struct rt6key *) ((u8 *) fn->leaf +
945 args->offset);
946
947 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
948 #ifdef CONFIG_IPV6_SUBTREES
949 if (fn->subtree)
950 fn = fib6_lookup_1(fn->subtree, args + 1);
951 #endif
952 if (!fn || fn->fn_flags & RTN_RTINFO)
953 return fn;
954 }
955 }
956
957 if (fn->fn_flags & RTN_ROOT)
958 break;
959
960 fn = fn->parent;
961 }
962
963 return NULL;
964 }
965
966 struct fib6_node * fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
967 const struct in6_addr *saddr)
968 {
969 struct fib6_node *fn;
970 struct lookup_args args[] = {
971 {
972 .offset = offsetof(struct rt6_info, rt6i_dst),
973 .addr = daddr,
974 },
975 #ifdef CONFIG_IPV6_SUBTREES
976 {
977 .offset = offsetof(struct rt6_info, rt6i_src),
978 .addr = saddr,
979 },
980 #endif
981 {
982 .offset = 0, /* sentinel */
983 }
984 };
985
986 fn = fib6_lookup_1(root, daddr ? args : args + 1);
987 if (!fn || fn->fn_flags & RTN_TL_ROOT)
988 fn = root;
989
990 return fn;
991 }
992
993 /*
994 * Get node with specified destination prefix (and source prefix,
995 * if subtrees are used)
996 */
997
998
999 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
1000 const struct in6_addr *addr,
1001 int plen, int offset)
1002 {
1003 struct fib6_node *fn;
1004
1005 for (fn = root; fn ; ) {
1006 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1007
1008 /*
1009 * Prefix match
1010 */
1011 if (plen < fn->fn_bit ||
1012 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1013 return NULL;
1014
1015 if (plen == fn->fn_bit)
1016 return fn;
1017
1018 /*
1019 * We have more bits to go
1020 */
1021 if (addr_bit_set(addr, fn->fn_bit))
1022 fn = fn->right;
1023 else
1024 fn = fn->left;
1025 }
1026 return NULL;
1027 }
1028
1029 struct fib6_node * fib6_locate(struct fib6_node *root,
1030 const struct in6_addr *daddr, int dst_len,
1031 const struct in6_addr *saddr, int src_len)
1032 {
1033 struct fib6_node *fn;
1034
1035 fn = fib6_locate_1(root, daddr, dst_len,
1036 offsetof(struct rt6_info, rt6i_dst));
1037
1038 #ifdef CONFIG_IPV6_SUBTREES
1039 if (src_len) {
1040 WARN_ON(saddr == NULL);
1041 if (fn && fn->subtree)
1042 fn = fib6_locate_1(fn->subtree, saddr, src_len,
1043 offsetof(struct rt6_info, rt6i_src));
1044 }
1045 #endif
1046
1047 if (fn && fn->fn_flags & RTN_RTINFO)
1048 return fn;
1049
1050 return NULL;
1051 }
1052
1053
1054 /*
1055 * Deletion
1056 *
1057 */
1058
1059 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1060 {
1061 if (fn->fn_flags & RTN_ROOT)
1062 return net->ipv6.ip6_null_entry;
1063
1064 while (fn) {
1065 if (fn->left)
1066 return fn->left->leaf;
1067 if (fn->right)
1068 return fn->right->leaf;
1069
1070 fn = FIB6_SUBTREE(fn);
1071 }
1072 return NULL;
1073 }
1074
1075 /*
1076 * Called to trim the tree of intermediate nodes when possible. "fn"
1077 * is the node we want to try and remove.
1078 */
1079
1080 static struct fib6_node *fib6_repair_tree(struct net *net,
1081 struct fib6_node *fn)
1082 {
1083 int children;
1084 int nstate;
1085 struct fib6_node *child, *pn;
1086 struct fib6_walker_t *w;
1087 int iter = 0;
1088
1089 for (;;) {
1090 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1091 iter++;
1092
1093 WARN_ON(fn->fn_flags & RTN_RTINFO);
1094 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1095 WARN_ON(fn->leaf != NULL);
1096
1097 children = 0;
1098 child = NULL;
1099 if (fn->right) child = fn->right, children |= 1;
1100 if (fn->left) child = fn->left, children |= 2;
1101
1102 if (children == 3 || FIB6_SUBTREE(fn)
1103 #ifdef CONFIG_IPV6_SUBTREES
1104 /* Subtree root (i.e. fn) may have one child */
1105 || (children && fn->fn_flags & RTN_ROOT)
1106 #endif
1107 ) {
1108 fn->leaf = fib6_find_prefix(net, fn);
1109 #if RT6_DEBUG >= 2
1110 if (!fn->leaf) {
1111 WARN_ON(!fn->leaf);
1112 fn->leaf = net->ipv6.ip6_null_entry;
1113 }
1114 #endif
1115 atomic_inc(&fn->leaf->rt6i_ref);
1116 return fn->parent;
1117 }
1118
1119 pn = fn->parent;
1120 #ifdef CONFIG_IPV6_SUBTREES
1121 if (FIB6_SUBTREE(pn) == fn) {
1122 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1123 FIB6_SUBTREE(pn) = NULL;
1124 nstate = FWS_L;
1125 } else {
1126 WARN_ON(fn->fn_flags & RTN_ROOT);
1127 #endif
1128 if (pn->right == fn) pn->right = child;
1129 else if (pn->left == fn) pn->left = child;
1130 #if RT6_DEBUG >= 2
1131 else
1132 WARN_ON(1);
1133 #endif
1134 if (child)
1135 child->parent = pn;
1136 nstate = FWS_R;
1137 #ifdef CONFIG_IPV6_SUBTREES
1138 }
1139 #endif
1140
1141 read_lock(&fib6_walker_lock);
1142 FOR_WALKERS(w) {
1143 if (!child) {
1144 if (w->root == fn) {
1145 w->root = w->node = NULL;
1146 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1147 } else if (w->node == fn) {
1148 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1149 w->node = pn;
1150 w->state = nstate;
1151 }
1152 } else {
1153 if (w->root == fn) {
1154 w->root = child;
1155 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1156 }
1157 if (w->node == fn) {
1158 w->node = child;
1159 if (children&2) {
1160 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1161 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1162 } else {
1163 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1164 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1165 }
1166 }
1167 }
1168 }
1169 read_unlock(&fib6_walker_lock);
1170
1171 node_free(fn);
1172 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1173 return pn;
1174
1175 rt6_release(pn->leaf);
1176 pn->leaf = NULL;
1177 fn = pn;
1178 }
1179 }
1180
1181 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1182 struct nl_info *info)
1183 {
1184 struct fib6_walker_t *w;
1185 struct rt6_info *rt = *rtp;
1186 struct net *net = info->nl_net;
1187
1188 RT6_TRACE("fib6_del_route\n");
1189
1190 /* Unlink it */
1191 *rtp = rt->dst.rt6_next;
1192 rt->rt6i_node = NULL;
1193 net->ipv6.rt6_stats->fib_rt_entries--;
1194 net->ipv6.rt6_stats->fib_discarded_routes++;
1195
1196 /* Reset round-robin state, if necessary */
1197 if (fn->rr_ptr == rt)
1198 fn->rr_ptr = NULL;
1199
1200 /* Adjust walkers */
1201 read_lock(&fib6_walker_lock);
1202 FOR_WALKERS(w) {
1203 if (w->state == FWS_C && w->leaf == rt) {
1204 RT6_TRACE("walker %p adjusted by delroute\n", w);
1205 w->leaf = rt->dst.rt6_next;
1206 if (!w->leaf)
1207 w->state = FWS_U;
1208 }
1209 }
1210 read_unlock(&fib6_walker_lock);
1211
1212 rt->dst.rt6_next = NULL;
1213
1214 /* If it was last route, expunge its radix tree node */
1215 if (!fn->leaf) {
1216 fn->fn_flags &= ~RTN_RTINFO;
1217 net->ipv6.rt6_stats->fib_route_nodes--;
1218 fn = fib6_repair_tree(net, fn);
1219 }
1220
1221 if (atomic_read(&rt->rt6i_ref) != 1) {
1222 /* This route is used as dummy address holder in some split
1223 * nodes. It is not leaked, but it still holds other resources,
1224 * which must be released in time. So, scan ascendant nodes
1225 * and replace dummy references to this route with references
1226 * to still alive ones.
1227 */
1228 while (fn) {
1229 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
1230 fn->leaf = fib6_find_prefix(net, fn);
1231 atomic_inc(&fn->leaf->rt6i_ref);
1232 rt6_release(rt);
1233 }
1234 fn = fn->parent;
1235 }
1236 /* No more references are possible at this point. */
1237 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1238 }
1239
1240 inet6_rt_notify(RTM_DELROUTE, rt, info);
1241 rt6_release(rt);
1242 }
1243
1244 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1245 {
1246 struct net *net = info->nl_net;
1247 struct fib6_node *fn = rt->rt6i_node;
1248 struct rt6_info **rtp;
1249
1250 #if RT6_DEBUG >= 2
1251 if (rt->dst.obsolete>0) {
1252 WARN_ON(fn != NULL);
1253 return -ENOENT;
1254 }
1255 #endif
1256 if (!fn || rt == net->ipv6.ip6_null_entry)
1257 return -ENOENT;
1258
1259 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1260
1261 if (!(rt->rt6i_flags & RTF_CACHE)) {
1262 struct fib6_node *pn = fn;
1263 #ifdef CONFIG_IPV6_SUBTREES
1264 /* clones of this route might be in another subtree */
1265 if (rt->rt6i_src.plen) {
1266 while (!(pn->fn_flags & RTN_ROOT))
1267 pn = pn->parent;
1268 pn = pn->parent;
1269 }
1270 #endif
1271 fib6_prune_clones(info->nl_net, pn, rt);
1272 }
1273
1274 /*
1275 * Walk the leaf entries looking for ourself
1276 */
1277
1278 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1279 if (*rtp == rt) {
1280 fib6_del_route(fn, rtp, info);
1281 return 0;
1282 }
1283 }
1284 return -ENOENT;
1285 }
1286
1287 /*
1288 * Tree traversal function.
1289 *
1290 * Certainly, it is not interrupt safe.
1291 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1292 * It means, that we can modify tree during walking
1293 * and use this function for garbage collection, clone pruning,
1294 * cleaning tree when a device goes down etc. etc.
1295 *
1296 * It guarantees that every node will be traversed,
1297 * and that it will be traversed only once.
1298 *
1299 * Callback function w->func may return:
1300 * 0 -> continue walking.
1301 * positive value -> walking is suspended (used by tree dumps,
1302 * and probably by gc, if it will be split to several slices)
1303 * negative value -> terminate walking.
1304 *
1305 * The function itself returns:
1306 * 0 -> walk is complete.
1307 * >0 -> walk is incomplete (i.e. suspended)
1308 * <0 -> walk is terminated by an error.
1309 */
1310
1311 static int fib6_walk_continue(struct fib6_walker_t *w)
1312 {
1313 struct fib6_node *fn, *pn;
1314
1315 for (;;) {
1316 fn = w->node;
1317 if (!fn)
1318 return 0;
1319
1320 if (w->prune && fn != w->root &&
1321 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1322 w->state = FWS_C;
1323 w->leaf = fn->leaf;
1324 }
1325 switch (w->state) {
1326 #ifdef CONFIG_IPV6_SUBTREES
1327 case FWS_S:
1328 if (FIB6_SUBTREE(fn)) {
1329 w->node = FIB6_SUBTREE(fn);
1330 continue;
1331 }
1332 w->state = FWS_L;
1333 #endif
1334 case FWS_L:
1335 if (fn->left) {
1336 w->node = fn->left;
1337 w->state = FWS_INIT;
1338 continue;
1339 }
1340 w->state = FWS_R;
1341 case FWS_R:
1342 if (fn->right) {
1343 w->node = fn->right;
1344 w->state = FWS_INIT;
1345 continue;
1346 }
1347 w->state = FWS_C;
1348 w->leaf = fn->leaf;
1349 case FWS_C:
1350 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1351 int err;
1352
1353 if (w->count < w->skip) {
1354 w->count++;
1355 continue;
1356 }
1357
1358 err = w->func(w);
1359 if (err)
1360 return err;
1361
1362 w->count++;
1363 continue;
1364 }
1365 w->state = FWS_U;
1366 case FWS_U:
1367 if (fn == w->root)
1368 return 0;
1369 pn = fn->parent;
1370 w->node = pn;
1371 #ifdef CONFIG_IPV6_SUBTREES
1372 if (FIB6_SUBTREE(pn) == fn) {
1373 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1374 w->state = FWS_L;
1375 continue;
1376 }
1377 #endif
1378 if (pn->left == fn) {
1379 w->state = FWS_R;
1380 continue;
1381 }
1382 if (pn->right == fn) {
1383 w->state = FWS_C;
1384 w->leaf = w->node->leaf;
1385 continue;
1386 }
1387 #if RT6_DEBUG >= 2
1388 WARN_ON(1);
1389 #endif
1390 }
1391 }
1392 }
1393
1394 static int fib6_walk(struct fib6_walker_t *w)
1395 {
1396 int res;
1397
1398 w->state = FWS_INIT;
1399 w->node = w->root;
1400
1401 fib6_walker_link(w);
1402 res = fib6_walk_continue(w);
1403 if (res <= 0)
1404 fib6_walker_unlink(w);
1405 return res;
1406 }
1407
1408 static int fib6_clean_node(struct fib6_walker_t *w)
1409 {
1410 int res;
1411 struct rt6_info *rt;
1412 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1413 struct nl_info info = {
1414 .nl_net = c->net,
1415 };
1416
1417 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1418 res = c->func(rt, c->arg);
1419 if (res < 0) {
1420 w->leaf = rt;
1421 res = fib6_del(rt, &info);
1422 if (res) {
1423 #if RT6_DEBUG >= 2
1424 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res);
1425 #endif
1426 continue;
1427 }
1428 return 0;
1429 }
1430 WARN_ON(res != 0);
1431 }
1432 w->leaf = rt;
1433 return 0;
1434 }
1435
1436 /*
1437 * Convenient frontend to tree walker.
1438 *
1439 * func is called on each route.
1440 * It may return -1 -> delete this route.
1441 * 0 -> continue walking
1442 *
1443 * prune==1 -> only immediate children of node (certainly,
1444 * ignoring pure split nodes) will be scanned.
1445 */
1446
1447 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1448 int (*func)(struct rt6_info *, void *arg),
1449 int prune, void *arg)
1450 {
1451 struct fib6_cleaner_t c;
1452
1453 c.w.root = root;
1454 c.w.func = fib6_clean_node;
1455 c.w.prune = prune;
1456 c.w.count = 0;
1457 c.w.skip = 0;
1458 c.func = func;
1459 c.arg = arg;
1460 c.net = net;
1461
1462 fib6_walk(&c.w);
1463 }
1464
1465 void fib6_clean_all_ro(struct net *net, int (*func)(struct rt6_info *, void *arg),
1466 int prune, void *arg)
1467 {
1468 struct fib6_table *table;
1469 struct hlist_node *node;
1470 struct hlist_head *head;
1471 unsigned int h;
1472
1473 rcu_read_lock();
1474 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1475 head = &net->ipv6.fib_table_hash[h];
1476 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1477 read_lock_bh(&table->tb6_lock);
1478 fib6_clean_tree(net, &table->tb6_root,
1479 func, prune, arg);
1480 read_unlock_bh(&table->tb6_lock);
1481 }
1482 }
1483 rcu_read_unlock();
1484 }
1485 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1486 int prune, void *arg)
1487 {
1488 struct fib6_table *table;
1489 struct hlist_node *node;
1490 struct hlist_head *head;
1491 unsigned int h;
1492
1493 rcu_read_lock();
1494 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1495 head = &net->ipv6.fib_table_hash[h];
1496 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1497 write_lock_bh(&table->tb6_lock);
1498 fib6_clean_tree(net, &table->tb6_root,
1499 func, prune, arg);
1500 write_unlock_bh(&table->tb6_lock);
1501 }
1502 }
1503 rcu_read_unlock();
1504 }
1505
1506 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1507 {
1508 if (rt->rt6i_flags & RTF_CACHE) {
1509 RT6_TRACE("pruning clone %p\n", rt);
1510 return -1;
1511 }
1512
1513 return 0;
1514 }
1515
1516 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1517 struct rt6_info *rt)
1518 {
1519 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1520 }
1521
1522 /*
1523 * Garbage collection
1524 */
1525
1526 static struct fib6_gc_args
1527 {
1528 int timeout;
1529 int more;
1530 } gc_args;
1531
1532 static int fib6_age(struct rt6_info *rt, void *arg)
1533 {
1534 unsigned long now = jiffies;
1535
1536 /*
1537 * check addrconf expiration here.
1538 * Routes are expired even if they are in use.
1539 *
1540 * Also age clones. Note, that clones are aged out
1541 * only if they are not in use now.
1542 */
1543
1544 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1545 if (time_after(now, rt->dst.expires)) {
1546 RT6_TRACE("expiring %p\n", rt);
1547 return -1;
1548 }
1549 gc_args.more++;
1550 } else if (rt->rt6i_flags & RTF_CACHE) {
1551 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1552 time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1553 RT6_TRACE("aging clone %p\n", rt);
1554 return -1;
1555 } else if (rt->rt6i_flags & RTF_GATEWAY) {
1556 struct neighbour *neigh;
1557 __u8 neigh_flags = 0;
1558
1559 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1560 if (neigh) {
1561 neigh_flags = neigh->flags;
1562 neigh_release(neigh);
1563 }
1564 if (neigh_flags & NTF_ROUTER) {
1565 RT6_TRACE("purging route %p via non-router but gateway\n",
1566 rt);
1567 return -1;
1568 }
1569 }
1570 gc_args.more++;
1571 }
1572
1573 return 0;
1574 }
1575
1576 static DEFINE_SPINLOCK(fib6_gc_lock);
1577
1578 void fib6_run_gc(unsigned long expires, struct net *net)
1579 {
1580 if (expires != ~0UL) {
1581 spin_lock_bh(&fib6_gc_lock);
1582 gc_args.timeout = expires ? (int)expires :
1583 net->ipv6.sysctl.ip6_rt_gc_interval;
1584 } else {
1585 if (!spin_trylock_bh(&fib6_gc_lock)) {
1586 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1587 return;
1588 }
1589 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1590 }
1591
1592 gc_args.more = icmp6_dst_gc();
1593
1594 fib6_clean_all(net, fib6_age, 0, NULL);
1595
1596 if (gc_args.more)
1597 mod_timer(&net->ipv6.ip6_fib_timer,
1598 round_jiffies(jiffies
1599 + net->ipv6.sysctl.ip6_rt_gc_interval));
1600 else
1601 del_timer(&net->ipv6.ip6_fib_timer);
1602 spin_unlock_bh(&fib6_gc_lock);
1603 }
1604
1605 static void fib6_gc_timer_cb(unsigned long arg)
1606 {
1607 fib6_run_gc(0, (struct net *)arg);
1608 }
1609
1610 static int __net_init fib6_net_init(struct net *net)
1611 {
1612 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1613
1614 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1615
1616 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1617 if (!net->ipv6.rt6_stats)
1618 goto out_timer;
1619
1620 /* Avoid false sharing : Use at least a full cache line */
1621 size = max_t(size_t, size, L1_CACHE_BYTES);
1622
1623 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1624 if (!net->ipv6.fib_table_hash)
1625 goto out_rt6_stats;
1626
1627 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1628 GFP_KERNEL);
1629 if (!net->ipv6.fib6_main_tbl)
1630 goto out_fib_table_hash;
1631
1632 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1633 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1634 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1635 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1636
1637 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1638 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1639 GFP_KERNEL);
1640 if (!net->ipv6.fib6_local_tbl)
1641 goto out_fib6_main_tbl;
1642 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1643 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1644 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1645 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1646 #endif
1647 fib6_tables_init(net);
1648
1649 return 0;
1650
1651 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1652 out_fib6_main_tbl:
1653 kfree(net->ipv6.fib6_main_tbl);
1654 #endif
1655 out_fib_table_hash:
1656 kfree(net->ipv6.fib_table_hash);
1657 out_rt6_stats:
1658 kfree(net->ipv6.rt6_stats);
1659 out_timer:
1660 return -ENOMEM;
1661 }
1662
1663 static void fib6_net_exit(struct net *net)
1664 {
1665 rt6_ifdown(net, NULL);
1666 del_timer_sync(&net->ipv6.ip6_fib_timer);
1667
1668 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1669 kfree(net->ipv6.fib6_local_tbl);
1670 #endif
1671 kfree(net->ipv6.fib6_main_tbl);
1672 kfree(net->ipv6.fib_table_hash);
1673 kfree(net->ipv6.rt6_stats);
1674 }
1675
1676 static struct pernet_operations fib6_net_ops = {
1677 .init = fib6_net_init,
1678 .exit = fib6_net_exit,
1679 };
1680
1681 int __init fib6_init(void)
1682 {
1683 int ret = -ENOMEM;
1684
1685 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1686 sizeof(struct fib6_node),
1687 0, SLAB_HWCACHE_ALIGN,
1688 NULL);
1689 if (!fib6_node_kmem)
1690 goto out;
1691
1692 ret = register_pernet_subsys(&fib6_net_ops);
1693 if (ret)
1694 goto out_kmem_cache_create;
1695
1696 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1697 NULL);
1698 if (ret)
1699 goto out_unregister_subsys;
1700 out:
1701 return ret;
1702
1703 out_unregister_subsys:
1704 unregister_pernet_subsys(&fib6_net_ops);
1705 out_kmem_cache_create:
1706 kmem_cache_destroy(fib6_node_kmem);
1707 goto out;
1708 }
1709
1710 void fib6_gc_cleanup(void)
1711 {
1712 unregister_pernet_subsys(&fib6_net_ops);
1713 kmem_cache_destroy(fib6_node_kmem);
1714 }