blob: 320d91d20ad7ddfd981cee92161482f218c225b4 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09002 * Linux INET6 implementation
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 * Forwarding Information Database
4 *
5 * Authors:
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09006 * Pedro Roque <roque@di.fc.ul.pt>
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * 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.
YOSHIFUJI Hideakic0bece92006-08-23 17:23:25 -070019 * Ville Nuorvala: Fixed routing subtrees.
Linus Torvalds1da177e2005-04-16 15:20:36 -070020 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#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>
Thomas Grafc71099a2006-08-04 23:20:06 -070028#include <linux/list.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090029#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
31#ifdef CONFIG_PROC_FS
32#include <linux/proc_fs.h>
33#endif
34
35#include <net/ipv6.h>
36#include <net/ndisc.h>
37#include <net/addrconf.h>
38
39#include <net/ip6_fib.h>
40#include <net/ip6_route.h>
41
42#define RT6_DEBUG 2
43
44#if RT6_DEBUG >= 3
45#define RT6_TRACE(x...) printk(KERN_DEBUG x)
46#else
47#define RT6_TRACE(x...) do { ; } while (0)
48#endif
49
Christoph Lametere18b8902006-12-06 20:33:20 -080050static struct kmem_cache * fib6_node_kmem __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
52enum fib_walk_state_t
53{
54#ifdef CONFIG_IPV6_SUBTREES
55 FWS_S,
56#endif
57 FWS_L,
58 FWS_R,
59 FWS_C,
60 FWS_U
61};
62
63struct fib6_cleaner_t
64{
65 struct fib6_walker_t w;
Benjamin Theryec7d43c2008-03-03 23:31:57 -080066 struct net *net;
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 int (*func)(struct rt6_info *, void *arg);
68 void *arg;
69};
70
Adrian Bunk90d41122006-08-14 23:49:16 -070071static DEFINE_RWLOCK(fib6_walker_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
73#ifdef CONFIG_IPV6_SUBTREES
74#define FWS_INIT FWS_S
Linus Torvalds1da177e2005-04-16 15:20:36 -070075#else
76#define FWS_INIT FWS_L
Linus Torvalds1da177e2005-04-16 15:20:36 -070077#endif
78
Benjamin Theryec7d43c2008-03-03 23:31:57 -080079static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
80 struct rt6_info *rt);
Daniel Lezcano8ed67782008-03-04 13:48:30 -080081static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
82static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
Adrian Bunk90d41122006-08-14 23:49:16 -070083static int fib6_walk(struct fib6_walker_t *w);
84static int fib6_walk_continue(struct fib6_walker_t *w);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
86/*
87 * A routing update causes an increase of the serial number on the
88 * affected subtree. This allows for cached routes to be asynchronously
89 * tested when modifications are made to the destination cache as a
90 * result of redirects, path MTU changes, etc.
91 */
92
93static __u32 rt_sernum;
94
Daniel Lezcano5b7c9312008-03-03 23:28:58 -080095static void fib6_gc_timer_cb(unsigned long arg);
96
Alexey Dobriyanbbef49d2010-02-18 08:13:30 +000097static LIST_HEAD(fib6_walkers);
98#define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
Adrian Bunk90d41122006-08-14 23:49:16 -0700100static inline void fib6_walker_link(struct fib6_walker_t *w)
101{
102 write_lock_bh(&fib6_walker_lock);
Alexey Dobriyanbbef49d2010-02-18 08:13:30 +0000103 list_add(&w->lh, &fib6_walkers);
Adrian Bunk90d41122006-08-14 23:49:16 -0700104 write_unlock_bh(&fib6_walker_lock);
105}
106
107static inline void fib6_walker_unlink(struct fib6_walker_t *w)
108{
109 write_lock_bh(&fib6_walker_lock);
Alexey Dobriyanbbef49d2010-02-18 08:13:30 +0000110 list_del(&w->lh);
Adrian Bunk90d41122006-08-14 23:49:16 -0700111 write_unlock_bh(&fib6_walker_lock);
112}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113static __inline__ u32 fib6_new_sernum(void)
114{
115 u32 n = ++rt_sernum;
116 if ((__s32)n <= 0)
117 rt_sernum = n = 1;
118 return n;
119}
120
121/*
122 * Auxiliary address test functions for the radix tree.
123 *
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900124 * These assume a 32bit processor (although it will work on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 * 64bit processors)
126 */
127
128/*
129 * test bit
130 */
YOSHIFUJI Hideaki / 吉藤英明02cdce52010-03-27 01:24:16 +0000131#if defined(__LITTLE_ENDIAN)
132# define BITOP_BE32_SWIZZLE (0x1F & ~7)
133#else
134# define BITOP_BE32_SWIZZLE 0
135#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136
Eric Dumazetb71d1d42011-04-22 04:53:02 +0000137static __inline__ __be32 addr_bit_set(const void *token, int fn_bit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138{
Eric Dumazetb71d1d42011-04-22 04:53:02 +0000139 const __be32 *addr = token;
YOSHIFUJI Hideaki / 吉藤英明02cdce52010-03-27 01:24:16 +0000140 /*
141 * Here,
142 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
143 * is optimized version of
144 * htonl(1 << ((~fn_bit)&0x1F))
145 * See include/asm-generic/bitops/le.h.
146 */
Eric Dumazet0eae88f2010-04-20 19:06:52 -0700147 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
148 addr[fn_bit >> 5];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149}
150
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151static __inline__ struct fib6_node * node_alloc(void)
152{
153 struct fib6_node *fn;
154
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800155 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
157 return fn;
158}
159
160static __inline__ void node_free(struct fib6_node * fn)
161{
162 kmem_cache_free(fib6_node_kmem, fn);
163}
164
165static __inline__ void rt6_release(struct rt6_info *rt)
166{
167 if (atomic_dec_and_test(&rt->rt6i_ref))
Changli Gaod8d1f302010-06-10 23:31:35 -0700168 dst_free(&rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169}
170
Daniel Lezcano58f09b72008-03-03 23:25:27 -0800171static void fib6_link_table(struct net *net, struct fib6_table *tb)
Patrick McHardy1b43af52006-08-10 23:11:17 -0700172{
173 unsigned int h;
174
Thomas Graf375216a2006-10-21 20:20:54 -0700175 /*
176 * Initialize table lock at a single place to give lockdep a key,
177 * tables aren't visible prior to being linked to the list.
178 */
179 rwlock_init(&tb->tb6_lock);
180
Neil Hormana33bc5c2009-07-30 18:52:15 -0700181 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
Patrick McHardy1b43af52006-08-10 23:11:17 -0700182
183 /*
184 * No protection necessary, this is the only list mutatation
185 * operation, tables never disappear once they exist.
186 */
Daniel Lezcano58f09b72008-03-03 23:25:27 -0800187 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
Patrick McHardy1b43af52006-08-10 23:11:17 -0700188}
189
190#ifdef CONFIG_IPV6_MULTIPLE_TABLES
Daniel Lezcanoe0b855902008-03-03 23:24:31 -0800191
Daniel Lezcano8ed67782008-03-04 13:48:30 -0800192static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
Thomas Grafc71099a2006-08-04 23:20:06 -0700193{
194 struct fib6_table *table;
195
196 table = kzalloc(sizeof(*table), GFP_ATOMIC);
197 if (table != NULL) {
198 table->tb6_id = id;
Daniel Lezcano8ed67782008-03-04 13:48:30 -0800199 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
Thomas Grafc71099a2006-08-04 23:20:06 -0700200 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
201 }
202
203 return table;
204}
205
Daniel Lezcano58f09b72008-03-03 23:25:27 -0800206struct fib6_table *fib6_new_table(struct net *net, u32 id)
Thomas Grafc71099a2006-08-04 23:20:06 -0700207{
208 struct fib6_table *tb;
209
210 if (id == 0)
211 id = RT6_TABLE_MAIN;
Daniel Lezcano58f09b72008-03-03 23:25:27 -0800212 tb = fib6_get_table(net, id);
Thomas Grafc71099a2006-08-04 23:20:06 -0700213 if (tb)
214 return tb;
215
Daniel Lezcano8ed67782008-03-04 13:48:30 -0800216 tb = fib6_alloc_table(net, id);
Thomas Grafc71099a2006-08-04 23:20:06 -0700217 if (tb != NULL)
Daniel Lezcano58f09b72008-03-03 23:25:27 -0800218 fib6_link_table(net, tb);
Thomas Grafc71099a2006-08-04 23:20:06 -0700219
220 return tb;
221}
222
Daniel Lezcano58f09b72008-03-03 23:25:27 -0800223struct fib6_table *fib6_get_table(struct net *net, u32 id)
Thomas Grafc71099a2006-08-04 23:20:06 -0700224{
225 struct fib6_table *tb;
Daniel Lezcano58f09b72008-03-03 23:25:27 -0800226 struct hlist_head *head;
Thomas Grafc71099a2006-08-04 23:20:06 -0700227 struct hlist_node *node;
228 unsigned int h;
229
230 if (id == 0)
231 id = RT6_TABLE_MAIN;
Neil Hormana33bc5c2009-07-30 18:52:15 -0700232 h = id & (FIB6_TABLE_HASHSZ - 1);
Thomas Grafc71099a2006-08-04 23:20:06 -0700233 rcu_read_lock();
Daniel Lezcano58f09b72008-03-03 23:25:27 -0800234 head = &net->ipv6.fib_table_hash[h];
235 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
Thomas Grafc71099a2006-08-04 23:20:06 -0700236 if (tb->tb6_id == id) {
237 rcu_read_unlock();
238 return tb;
239 }
240 }
241 rcu_read_unlock();
242
243 return NULL;
244}
245
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000246static void __net_init fib6_tables_init(struct net *net)
Thomas Grafc71099a2006-08-04 23:20:06 -0700247{
Daniel Lezcano58f09b72008-03-03 23:25:27 -0800248 fib6_link_table(net, net->ipv6.fib6_main_tbl);
249 fib6_link_table(net, net->ipv6.fib6_local_tbl);
Thomas Grafc71099a2006-08-04 23:20:06 -0700250}
Thomas Grafc71099a2006-08-04 23:20:06 -0700251#else
252
Daniel Lezcano58f09b72008-03-03 23:25:27 -0800253struct fib6_table *fib6_new_table(struct net *net, u32 id)
Thomas Grafc71099a2006-08-04 23:20:06 -0700254{
Daniel Lezcano58f09b72008-03-03 23:25:27 -0800255 return fib6_get_table(net, id);
Thomas Grafc71099a2006-08-04 23:20:06 -0700256}
257
Daniel Lezcano58f09b72008-03-03 23:25:27 -0800258struct fib6_table *fib6_get_table(struct net *net, u32 id)
Thomas Grafc71099a2006-08-04 23:20:06 -0700259{
Daniel Lezcano58f09b72008-03-03 23:25:27 -0800260 return net->ipv6.fib6_main_tbl;
Thomas Grafc71099a2006-08-04 23:20:06 -0700261}
262
David S. Miller4c9483b2011-03-12 16:22:43 -0500263struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
Daniel Lezcano58f09b72008-03-03 23:25:27 -0800264 int flags, pol_lookup_t lookup)
Thomas Grafc71099a2006-08-04 23:20:06 -0700265{
David S. Miller4c9483b2011-03-12 16:22:43 -0500266 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
Thomas Grafc71099a2006-08-04 23:20:06 -0700267}
268
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000269static void __net_init fib6_tables_init(struct net *net)
Thomas Grafc71099a2006-08-04 23:20:06 -0700270{
Daniel Lezcano58f09b72008-03-03 23:25:27 -0800271 fib6_link_table(net, net->ipv6.fib6_main_tbl);
Thomas Grafc71099a2006-08-04 23:20:06 -0700272}
273
274#endif
275
Patrick McHardy1b43af52006-08-10 23:11:17 -0700276static int fib6_dump_node(struct fib6_walker_t *w)
277{
278 int res;
279 struct rt6_info *rt;
280
Changli Gaod8d1f302010-06-10 23:31:35 -0700281 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
Patrick McHardy1b43af52006-08-10 23:11:17 -0700282 res = rt6_dump_route(rt, w->args);
283 if (res < 0) {
284 /* Frame is full, suspend walking */
285 w->leaf = rt;
286 return 1;
287 }
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700288 WARN_ON(res == 0);
Patrick McHardy1b43af52006-08-10 23:11:17 -0700289 }
290 w->leaf = NULL;
291 return 0;
292}
293
294static void fib6_dump_end(struct netlink_callback *cb)
295{
296 struct fib6_walker_t *w = (void*)cb->args[2];
297
298 if (w) {
Herbert Xu7891cc82009-01-13 22:17:51 -0800299 if (cb->args[4]) {
300 cb->args[4] = 0;
301 fib6_walker_unlink(w);
302 }
Patrick McHardy1b43af52006-08-10 23:11:17 -0700303 cb->args[2] = 0;
304 kfree(w);
305 }
306 cb->done = (void*)cb->args[3];
307 cb->args[1] = 3;
308}
309
310static int fib6_dump_done(struct netlink_callback *cb)
311{
312 fib6_dump_end(cb);
313 return cb->done ? cb->done(cb) : 0;
314}
315
316static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
317 struct netlink_callback *cb)
318{
319 struct fib6_walker_t *w;
320 int res;
321
322 w = (void *)cb->args[2];
323 w->root = &table->tb6_root;
324
325 if (cb->args[4] == 0) {
Patrick McHardy2bec5a32010-02-08 05:19:03 +0000326 w->count = 0;
327 w->skip = 0;
328
Patrick McHardy1b43af52006-08-10 23:11:17 -0700329 read_lock_bh(&table->tb6_lock);
330 res = fib6_walk(w);
331 read_unlock_bh(&table->tb6_lock);
Patrick McHardy2bec5a32010-02-08 05:19:03 +0000332 if (res > 0) {
Patrick McHardy1b43af52006-08-10 23:11:17 -0700333 cb->args[4] = 1;
Patrick McHardy2bec5a32010-02-08 05:19:03 +0000334 cb->args[5] = w->root->fn_sernum;
335 }
Patrick McHardy1b43af52006-08-10 23:11:17 -0700336 } else {
Patrick McHardy2bec5a32010-02-08 05:19:03 +0000337 if (cb->args[5] != w->root->fn_sernum) {
338 /* Begin at the root if the tree changed */
339 cb->args[5] = w->root->fn_sernum;
340 w->state = FWS_INIT;
341 w->node = w->root;
342 w->skip = w->count;
343 } else
344 w->skip = 0;
345
Patrick McHardy1b43af52006-08-10 23:11:17 -0700346 read_lock_bh(&table->tb6_lock);
347 res = fib6_walk_continue(w);
348 read_unlock_bh(&table->tb6_lock);
Herbert Xu7891cc82009-01-13 22:17:51 -0800349 if (res <= 0) {
350 fib6_walker_unlink(w);
351 cb->args[4] = 0;
Patrick McHardy1b43af52006-08-10 23:11:17 -0700352 }
Patrick McHardy1b43af52006-08-10 23:11:17 -0700353 }
Herbert Xu7891cc82009-01-13 22:17:51 -0800354
Patrick McHardy1b43af52006-08-10 23:11:17 -0700355 return res;
356}
357
Thomas Grafc127ea22007-03-22 11:58:32 -0700358static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
Patrick McHardy1b43af52006-08-10 23:11:17 -0700359{
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900360 struct net *net = sock_net(skb->sk);
Patrick McHardy1b43af52006-08-10 23:11:17 -0700361 unsigned int h, s_h;
362 unsigned int e = 0, s_e;
363 struct rt6_rtnl_dump_arg arg;
364 struct fib6_walker_t *w;
365 struct fib6_table *tb;
366 struct hlist_node *node;
Daniel Lezcano58f09b72008-03-03 23:25:27 -0800367 struct hlist_head *head;
Patrick McHardy1b43af52006-08-10 23:11:17 -0700368 int res = 0;
369
370 s_h = cb->args[0];
371 s_e = cb->args[1];
372
373 w = (void *)cb->args[2];
374 if (w == NULL) {
375 /* New dump:
376 *
377 * 1. hook callback destructor.
378 */
379 cb->args[3] = (long)cb->done;
380 cb->done = fib6_dump_done;
381
382 /*
383 * 2. allocate and initialize walker.
384 */
385 w = kzalloc(sizeof(*w), GFP_ATOMIC);
386 if (w == NULL)
387 return -ENOMEM;
388 w->func = fib6_dump_node;
389 cb->args[2] = (long)w;
390 }
391
392 arg.skb = skb;
393 arg.cb = cb;
Brian Haley191cd582008-08-14 15:33:21 -0700394 arg.net = net;
Patrick McHardy1b43af52006-08-10 23:11:17 -0700395 w->args = &arg;
396
Eric Dumazete67f88d2011-04-27 22:56:07 +0000397 rcu_read_lock();
Neil Hormana33bc5c2009-07-30 18:52:15 -0700398 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
Patrick McHardy1b43af52006-08-10 23:11:17 -0700399 e = 0;
Daniel Lezcano58f09b72008-03-03 23:25:27 -0800400 head = &net->ipv6.fib_table_hash[h];
Eric Dumazete67f88d2011-04-27 22:56:07 +0000401 hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
Patrick McHardy1b43af52006-08-10 23:11:17 -0700402 if (e < s_e)
403 goto next;
404 res = fib6_dump_table(tb, skb, cb);
405 if (res != 0)
406 goto out;
407next:
408 e++;
409 }
410 }
411out:
Eric Dumazete67f88d2011-04-27 22:56:07 +0000412 rcu_read_unlock();
Patrick McHardy1b43af52006-08-10 23:11:17 -0700413 cb->args[1] = e;
414 cb->args[0] = h;
415
416 res = res < 0 ? res : skb->len;
417 if (res <= 0)
418 fib6_dump_end(cb);
419 return res;
420}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
422/*
423 * Routing Table
424 *
425 * return the appropriate node for a routing tree "add" operation
426 * by either creating and inserting or by returning an existing
427 * node.
428 */
429
430static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
431 int addrlen, int plen,
432 int offset)
433{
434 struct fib6_node *fn, *in, *ln;
435 struct fib6_node *pn = NULL;
436 struct rt6key *key;
437 int bit;
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900438 __be32 dir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 __u32 sernum = fib6_new_sernum();
440
441 RT6_TRACE("fib6_add_1\n");
442
443 /* insert node in tree */
444
445 fn = root;
446
447 do {
448 key = (struct rt6key *)((u8 *)fn->leaf + offset);
449
450 /*
451 * Prefix match
452 */
453 if (plen < fn->fn_bit ||
454 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
455 goto insert_above;
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900456
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 /*
458 * Exact match ?
459 */
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900460
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 if (plen == fn->fn_bit) {
462 /* clean up an intermediate node */
463 if ((fn->fn_flags & RTN_RTINFO) == 0) {
464 rt6_release(fn->leaf);
465 fn->leaf = NULL;
466 }
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900467
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 fn->fn_sernum = sernum;
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900469
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 return fn;
471 }
472
473 /*
474 * We have more bits to go
475 */
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900476
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 /* Try to walk down on tree. */
478 fn->fn_sernum = sernum;
479 dir = addr_bit_set(addr, fn->fn_bit);
480 pn = fn;
481 fn = dir ? fn->right: fn->left;
482 } while (fn);
483
484 /*
485 * We walked to the bottom of tree.
486 * Create new leaf node without children.
487 */
488
489 ln = node_alloc();
490
491 if (ln == NULL)
492 return NULL;
493 ln->fn_bit = plen;
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900494
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 ln->parent = pn;
496 ln->fn_sernum = sernum;
497
498 if (dir)
499 pn->right = ln;
500 else
501 pn->left = ln;
502
503 return ln;
504
505
506insert_above:
507 /*
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900508 * split since we don't have a common prefix anymore or
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 * we have a less significant route.
510 * we've to insert an intermediate node on the list
511 * this new node will point to the one we need to create
512 * and the current
513 */
514
515 pn = fn->parent;
516
517 /* find 1st bit in difference between the 2 addrs.
518
YOSHIFUJI Hideaki971f3592005-11-08 09:37:56 -0800519 See comment in __ipv6_addr_diff: bit may be an invalid value,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 but if it is >= plen, the value is ignored in any case.
521 */
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900522
YOSHIFUJI Hideaki971f3592005-11-08 09:37:56 -0800523 bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900525 /*
526 * (intermediate)[in]
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 * / \
528 * (new leaf node)[ln] (old node)[fn]
529 */
530 if (plen > bit) {
531 in = node_alloc();
532 ln = node_alloc();
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900533
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 if (in == NULL || ln == NULL) {
535 if (in)
536 node_free(in);
537 if (ln)
538 node_free(ln);
539 return NULL;
540 }
541
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900542 /*
543 * new intermediate node.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 * RTN_RTINFO will
545 * be off since that an address that chooses one of
546 * the branches would not match less specific routes
547 * in the other branch
548 */
549
550 in->fn_bit = bit;
551
552 in->parent = pn;
553 in->leaf = fn->leaf;
554 atomic_inc(&in->leaf->rt6i_ref);
555
556 in->fn_sernum = sernum;
557
558 /* update parent pointer */
559 if (dir)
560 pn->right = in;
561 else
562 pn->left = in;
563
564 ln->fn_bit = plen;
565
566 ln->parent = in;
567 fn->parent = in;
568
569 ln->fn_sernum = sernum;
570
571 if (addr_bit_set(addr, bit)) {
572 in->right = ln;
573 in->left = fn;
574 } else {
575 in->left = ln;
576 in->right = fn;
577 }
578 } else { /* plen <= bit */
579
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900580 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 * (new leaf node)[ln]
582 * / \
583 * (old node)[fn] NULL
584 */
585
586 ln = node_alloc();
587
588 if (ln == NULL)
589 return NULL;
590
591 ln->fn_bit = plen;
592
593 ln->parent = pn;
594
595 ln->fn_sernum = sernum;
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900596
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 if (dir)
598 pn->right = ln;
599 else
600 pn->left = ln;
601
602 if (addr_bit_set(&key->addr, plen))
603 ln->right = fn;
604 else
605 ln->left = fn;
606
607 fn->parent = ln;
608 }
609 return ln;
610}
611
612/*
613 * Insert routing information in a node.
614 */
615
616static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
Thomas Graf86872cb2006-08-22 00:01:08 -0700617 struct nl_info *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618{
619 struct rt6_info *iter = NULL;
620 struct rt6_info **ins;
621
622 ins = &fn->leaf;
623
Changli Gaod8d1f302010-06-10 23:31:35 -0700624 for (iter = fn->leaf; iter; iter=iter->dst.rt6_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 /*
626 * Search for duplicates
627 */
628
629 if (iter->rt6i_metric == rt->rt6i_metric) {
630 /*
631 * Same priority level
632 */
633
634 if (iter->rt6i_dev == rt->rt6i_dev &&
635 iter->rt6i_idev == rt->rt6i_idev &&
636 ipv6_addr_equal(&iter->rt6i_gateway,
637 &rt->rt6i_gateway)) {
638 if (!(iter->rt6i_flags&RTF_EXPIRES))
639 return -EEXIST;
640 iter->rt6i_expires = rt->rt6i_expires;
641 if (!(rt->rt6i_flags&RTF_EXPIRES)) {
642 iter->rt6i_flags &= ~RTF_EXPIRES;
643 iter->rt6i_expires = 0;
644 }
645 return -EEXIST;
646 }
647 }
648
649 if (iter->rt6i_metric > rt->rt6i_metric)
650 break;
651
Changli Gaod8d1f302010-06-10 23:31:35 -0700652 ins = &iter->dst.rt6_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 }
654
David S. Millerf11e6652007-03-24 20:36:25 -0700655 /* Reset round-robin state, if necessary */
656 if (ins == &fn->leaf)
657 fn->rr_ptr = NULL;
658
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 /*
660 * insert node
661 */
662
Changli Gaod8d1f302010-06-10 23:31:35 -0700663 rt->dst.rt6_next = iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 *ins = rt;
665 rt->rt6i_node = fn;
666 atomic_inc(&rt->rt6i_ref);
Thomas Graf86872cb2006-08-22 00:01:08 -0700667 inet6_rt_notify(RTM_NEWROUTE, rt, info);
Benjamin Theryc5728722008-03-03 23:34:17 -0800668 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669
670 if ((fn->fn_flags & RTN_RTINFO) == 0) {
Benjamin Theryc5728722008-03-03 23:34:17 -0800671 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 fn->fn_flags |= RTN_RTINFO;
673 }
674
675 return 0;
676}
677
Daniel Lezcano63152fc2008-03-03 23:31:11 -0800678static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
Stephen Hemminger417f28b2008-07-22 14:33:45 -0700680 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE)))
Stephen Hemminger417f28b2008-07-22 14:33:45 -0700682 mod_timer(&net->ipv6.ip6_fib_timer,
Stephen Hemminger847499c2008-07-21 13:21:35 -0700683 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684}
685
Daniel Lezcano63152fc2008-03-03 23:31:11 -0800686void fib6_force_start_gc(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687{
Stephen Hemminger417f28b2008-07-22 14:33:45 -0700688 if (!timer_pending(&net->ipv6.ip6_fib_timer))
689 mod_timer(&net->ipv6.ip6_fib_timer,
Stephen Hemminger847499c2008-07-21 13:21:35 -0700690 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691}
692
693/*
694 * Add routing information to the routing tree.
695 * <destination addr>/<source addr>
696 * with source addr info in sub-trees
697 */
698
Thomas Graf86872cb2006-08-22 00:01:08 -0700699int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700{
YOSHIFUJI Hideaki66729e12006-08-23 17:20:34 -0700701 struct fib6_node *fn, *pn = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 int err = -ENOMEM;
703
704 fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
705 rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst));
706
707 if (fn == NULL)
708 goto out;
709
YOSHIFUJI Hideaki66729e12006-08-23 17:20:34 -0700710 pn = fn;
711
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712#ifdef CONFIG_IPV6_SUBTREES
713 if (rt->rt6i_src.plen) {
714 struct fib6_node *sn;
715
716 if (fn->subtree == NULL) {
717 struct fib6_node *sfn;
718
719 /*
720 * Create subtree.
721 *
722 * fn[main tree]
723 * |
724 * sfn[subtree root]
725 * \
726 * sn[new leaf node]
727 */
728
729 /* Create subtree root node */
730 sfn = node_alloc();
731 if (sfn == NULL)
732 goto st_failure;
733
Daniel Lezcano8ed67782008-03-04 13:48:30 -0800734 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
735 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 sfn->fn_flags = RTN_ROOT;
737 sfn->fn_sernum = fib6_new_sernum();
738
739 /* Now add the first leaf node to new subtree */
740
741 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
742 sizeof(struct in6_addr), rt->rt6i_src.plen,
743 offsetof(struct rt6_info, rt6i_src));
744
745 if (sn == NULL) {
746 /* If it is failed, discard just allocated
747 root, and then (in st_failure) stale node
748 in main tree.
749 */
750 node_free(sfn);
751 goto st_failure;
752 }
753
754 /* Now link new subtree to main tree */
755 sfn->parent = fn;
756 fn->subtree = sfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 } else {
758 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
759 sizeof(struct in6_addr), rt->rt6i_src.plen,
760 offsetof(struct rt6_info, rt6i_src));
761
762 if (sn == NULL)
763 goto st_failure;
764 }
765
YOSHIFUJI Hideaki66729e12006-08-23 17:20:34 -0700766 if (fn->leaf == NULL) {
767 fn->leaf = rt;
768 atomic_inc(&rt->rt6i_ref);
769 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 fn = sn;
771 }
772#endif
773
Thomas Graf86872cb2006-08-22 00:01:08 -0700774 err = fib6_add_rt2node(fn, rt, info);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
776 if (err == 0) {
Daniel Lezcano63152fc2008-03-03 23:31:11 -0800777 fib6_start_gc(info->nl_net, rt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 if (!(rt->rt6i_flags&RTF_CACHE))
Benjamin Theryec7d43c2008-03-03 23:31:57 -0800779 fib6_prune_clones(info->nl_net, pn, rt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 }
781
782out:
YOSHIFUJI Hideaki66729e12006-08-23 17:20:34 -0700783 if (err) {
784#ifdef CONFIG_IPV6_SUBTREES
785 /*
786 * If fib6_add_1 has cleared the old leaf pointer in the
787 * super-tree leaf node we have to find a new one for it.
788 */
David S. Miller3c051232008-04-18 01:46:19 -0700789 if (pn != fn && pn->leaf == rt) {
790 pn->leaf = NULL;
791 atomic_dec(&rt->rt6i_ref);
792 }
YOSHIFUJI Hideaki66729e12006-08-23 17:20:34 -0700793 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
Daniel Lezcano8ed67782008-03-04 13:48:30 -0800794 pn->leaf = fib6_find_prefix(info->nl_net, pn);
YOSHIFUJI Hideaki66729e12006-08-23 17:20:34 -0700795#if RT6_DEBUG >= 2
796 if (!pn->leaf) {
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700797 WARN_ON(pn->leaf == NULL);
Daniel Lezcano8ed67782008-03-04 13:48:30 -0800798 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
YOSHIFUJI Hideaki66729e12006-08-23 17:20:34 -0700799 }
800#endif
801 atomic_inc(&pn->leaf->rt6i_ref);
802 }
803#endif
Changli Gaod8d1f302010-06-10 23:31:35 -0700804 dst_free(&rt->dst);
YOSHIFUJI Hideaki66729e12006-08-23 17:20:34 -0700805 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 return err;
807
808#ifdef CONFIG_IPV6_SUBTREES
809 /* Subtree creation failed, probably main tree node
810 is orphan. If it is, shoot it.
811 */
812st_failure:
813 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
Daniel Lezcano8ed67782008-03-04 13:48:30 -0800814 fib6_repair_tree(info->nl_net, fn);
Changli Gaod8d1f302010-06-10 23:31:35 -0700815 dst_free(&rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 return err;
817#endif
818}
819
820/*
821 * Routing tree lookup
822 *
823 */
824
825struct lookup_args {
826 int offset; /* key offset on rt6_info */
Eric Dumazetb71d1d42011-04-22 04:53:02 +0000827 const struct in6_addr *addr; /* search key */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828};
829
830static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
831 struct lookup_args *args)
832{
833 struct fib6_node *fn;
Al Viroe69a4adc2006-11-14 20:56:00 -0800834 __be32 dir;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
YOSHIFUJI Hideaki825e2882006-08-23 17:21:29 -0700836 if (unlikely(args->offset == 0))
837 return NULL;
838
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 /*
840 * Descend on a tree
841 */
842
843 fn = root;
844
845 for (;;) {
846 struct fib6_node *next;
847
848 dir = addr_bit_set(args->addr, fn->fn_bit);
849
850 next = dir ? fn->right : fn->left;
851
852 if (next) {
853 fn = next;
854 continue;
855 }
856
857 break;
858 }
859
YOSHIFUJI Hideaki3fc5e042006-08-23 17:21:12 -0700860 while(fn) {
YOSHIFUJI Hideaki7fc33162006-08-23 17:22:24 -0700861 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 struct rt6key *key;
863
864 key = (struct rt6key *) ((u8 *) fn->leaf +
865 args->offset);
866
YOSHIFUJI Hideaki3fc5e042006-08-23 17:21:12 -0700867 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
868#ifdef CONFIG_IPV6_SUBTREES
869 if (fn->subtree)
870 fn = fib6_lookup_1(fn->subtree, args + 1);
871#endif
872 if (!fn || fn->fn_flags & RTN_RTINFO)
873 return fn;
874 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 }
876
YOSHIFUJI Hideaki3fc5e042006-08-23 17:21:12 -0700877 if (fn->fn_flags & RTN_ROOT)
878 break;
879
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 fn = fn->parent;
881 }
882
883 return NULL;
884}
885
Eric Dumazetb71d1d42011-04-22 04:53:02 +0000886struct fib6_node * fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
887 const struct in6_addr *saddr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 struct fib6_node *fn;
YOSHIFUJI Hideaki825e2882006-08-23 17:21:29 -0700890 struct lookup_args args[] = {
891 {
892 .offset = offsetof(struct rt6_info, rt6i_dst),
893 .addr = daddr,
894 },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895#ifdef CONFIG_IPV6_SUBTREES
YOSHIFUJI Hideaki825e2882006-08-23 17:21:29 -0700896 {
897 .offset = offsetof(struct rt6_info, rt6i_src),
898 .addr = saddr,
899 },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900#endif
YOSHIFUJI Hideaki825e2882006-08-23 17:21:29 -0700901 {
902 .offset = 0, /* sentinel */
903 }
904 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
YOSHIFUJI Hideakifefc2a62006-08-23 17:21:50 -0700906 fn = fib6_lookup_1(root, daddr ? args : args + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
908 if (fn == NULL || fn->fn_flags & RTN_TL_ROOT)
909 fn = root;
910
911 return fn;
912}
913
914/*
915 * Get node with specified destination prefix (and source prefix,
916 * if subtrees are used)
917 */
918
919
920static struct fib6_node * fib6_locate_1(struct fib6_node *root,
Eric Dumazetb71d1d42011-04-22 04:53:02 +0000921 const struct in6_addr *addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 int plen, int offset)
923{
924 struct fib6_node *fn;
925
926 for (fn = root; fn ; ) {
927 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
928
929 /*
930 * Prefix match
931 */
932 if (plen < fn->fn_bit ||
933 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
934 return NULL;
935
936 if (plen == fn->fn_bit)
937 return fn;
938
939 /*
940 * We have more bits to go
941 */
942 if (addr_bit_set(addr, fn->fn_bit))
943 fn = fn->right;
944 else
945 fn = fn->left;
946 }
947 return NULL;
948}
949
950struct fib6_node * fib6_locate(struct fib6_node *root,
Eric Dumazetb71d1d42011-04-22 04:53:02 +0000951 const struct in6_addr *daddr, int dst_len,
952 const struct in6_addr *saddr, int src_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953{
954 struct fib6_node *fn;
955
956 fn = fib6_locate_1(root, daddr, dst_len,
957 offsetof(struct rt6_info, rt6i_dst));
958
959#ifdef CONFIG_IPV6_SUBTREES
960 if (src_len) {
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700961 WARN_ON(saddr == NULL);
YOSHIFUJI Hideaki3fc5e042006-08-23 17:21:12 -0700962 if (fn && fn->subtree)
963 fn = fib6_locate_1(fn->subtree, saddr, src_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 offsetof(struct rt6_info, rt6i_src));
965 }
966#endif
967
968 if (fn && fn->fn_flags&RTN_RTINFO)
969 return fn;
970
971 return NULL;
972}
973
974
975/*
976 * Deletion
977 *
978 */
979
Daniel Lezcano8ed67782008-03-04 13:48:30 -0800980static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981{
982 if (fn->fn_flags&RTN_ROOT)
Daniel Lezcano8ed67782008-03-04 13:48:30 -0800983 return net->ipv6.ip6_null_entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984
985 while(fn) {
986 if(fn->left)
987 return fn->left->leaf;
988
989 if(fn->right)
990 return fn->right->leaf;
991
YOSHIFUJI Hideaki7fc33162006-08-23 17:22:24 -0700992 fn = FIB6_SUBTREE(fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 }
994 return NULL;
995}
996
997/*
998 * Called to trim the tree of intermediate nodes when possible. "fn"
999 * is the node we want to try and remove.
1000 */
1001
Daniel Lezcano8ed67782008-03-04 13:48:30 -08001002static struct fib6_node *fib6_repair_tree(struct net *net,
1003 struct fib6_node *fn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004{
1005 int children;
1006 int nstate;
1007 struct fib6_node *child, *pn;
1008 struct fib6_walker_t *w;
1009 int iter = 0;
1010
1011 for (;;) {
1012 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1013 iter++;
1014
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001015 WARN_ON(fn->fn_flags & RTN_RTINFO);
1016 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1017 WARN_ON(fn->leaf != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018
1019 children = 0;
1020 child = NULL;
1021 if (fn->right) child = fn->right, children |= 1;
1022 if (fn->left) child = fn->left, children |= 2;
1023
YOSHIFUJI Hideaki7fc33162006-08-23 17:22:24 -07001024 if (children == 3 || FIB6_SUBTREE(fn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025#ifdef CONFIG_IPV6_SUBTREES
1026 /* Subtree root (i.e. fn) may have one child */
1027 || (children && fn->fn_flags&RTN_ROOT)
1028#endif
1029 ) {
Daniel Lezcano8ed67782008-03-04 13:48:30 -08001030 fn->leaf = fib6_find_prefix(net, fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031#if RT6_DEBUG >= 2
1032 if (fn->leaf==NULL) {
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001033 WARN_ON(!fn->leaf);
Daniel Lezcano8ed67782008-03-04 13:48:30 -08001034 fn->leaf = net->ipv6.ip6_null_entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 }
1036#endif
1037 atomic_inc(&fn->leaf->rt6i_ref);
1038 return fn->parent;
1039 }
1040
1041 pn = fn->parent;
1042#ifdef CONFIG_IPV6_SUBTREES
YOSHIFUJI Hideaki7fc33162006-08-23 17:22:24 -07001043 if (FIB6_SUBTREE(pn) == fn) {
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001044 WARN_ON(!(fn->fn_flags & RTN_ROOT));
YOSHIFUJI Hideaki7fc33162006-08-23 17:22:24 -07001045 FIB6_SUBTREE(pn) = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 nstate = FWS_L;
1047 } else {
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001048 WARN_ON(fn->fn_flags & RTN_ROOT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049#endif
1050 if (pn->right == fn) pn->right = child;
1051 else if (pn->left == fn) pn->left = child;
1052#if RT6_DEBUG >= 2
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001053 else
1054 WARN_ON(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055#endif
1056 if (child)
1057 child->parent = pn;
1058 nstate = FWS_R;
1059#ifdef CONFIG_IPV6_SUBTREES
1060 }
1061#endif
1062
1063 read_lock(&fib6_walker_lock);
1064 FOR_WALKERS(w) {
1065 if (child == NULL) {
1066 if (w->root == fn) {
1067 w->root = w->node = NULL;
1068 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1069 } else if (w->node == fn) {
1070 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1071 w->node = pn;
1072 w->state = nstate;
1073 }
1074 } else {
1075 if (w->root == fn) {
1076 w->root = child;
1077 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1078 }
1079 if (w->node == fn) {
1080 w->node = child;
1081 if (children&2) {
1082 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1083 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1084 } else {
1085 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1086 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1087 }
1088 }
1089 }
1090 }
1091 read_unlock(&fib6_walker_lock);
1092
1093 node_free(fn);
YOSHIFUJI Hideaki7fc33162006-08-23 17:22:24 -07001094 if (pn->fn_flags&RTN_RTINFO || FIB6_SUBTREE(pn))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 return pn;
1096
1097 rt6_release(pn->leaf);
1098 pn->leaf = NULL;
1099 fn = pn;
1100 }
1101}
1102
1103static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
Thomas Graf86872cb2006-08-22 00:01:08 -07001104 struct nl_info *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105{
1106 struct fib6_walker_t *w;
1107 struct rt6_info *rt = *rtp;
Benjamin Theryc5728722008-03-03 23:34:17 -08001108 struct net *net = info->nl_net;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109
1110 RT6_TRACE("fib6_del_route\n");
1111
1112 /* Unlink it */
Changli Gaod8d1f302010-06-10 23:31:35 -07001113 *rtp = rt->dst.rt6_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 rt->rt6i_node = NULL;
Benjamin Theryc5728722008-03-03 23:34:17 -08001115 net->ipv6.rt6_stats->fib_rt_entries--;
1116 net->ipv6.rt6_stats->fib_discarded_routes++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117
David S. Millerf11e6652007-03-24 20:36:25 -07001118 /* Reset round-robin state, if necessary */
1119 if (fn->rr_ptr == rt)
1120 fn->rr_ptr = NULL;
1121
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 /* Adjust walkers */
1123 read_lock(&fib6_walker_lock);
1124 FOR_WALKERS(w) {
1125 if (w->state == FWS_C && w->leaf == rt) {
1126 RT6_TRACE("walker %p adjusted by delroute\n", w);
Changli Gaod8d1f302010-06-10 23:31:35 -07001127 w->leaf = rt->dst.rt6_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 if (w->leaf == NULL)
1129 w->state = FWS_U;
1130 }
1131 }
1132 read_unlock(&fib6_walker_lock);
1133
Changli Gaod8d1f302010-06-10 23:31:35 -07001134 rt->dst.rt6_next = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 /* If it was last route, expunge its radix tree node */
1137 if (fn->leaf == NULL) {
1138 fn->fn_flags &= ~RTN_RTINFO;
Benjamin Theryc5728722008-03-03 23:34:17 -08001139 net->ipv6.rt6_stats->fib_route_nodes--;
Daniel Lezcano8ed67782008-03-04 13:48:30 -08001140 fn = fib6_repair_tree(net, fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 }
1142
1143 if (atomic_read(&rt->rt6i_ref) != 1) {
1144 /* This route is used as dummy address holder in some split
1145 * nodes. It is not leaked, but it still holds other resources,
1146 * which must be released in time. So, scan ascendant nodes
1147 * and replace dummy references to this route with references
1148 * to still alive ones.
1149 */
1150 while (fn) {
1151 if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) {
Daniel Lezcano8ed67782008-03-04 13:48:30 -08001152 fn->leaf = fib6_find_prefix(net, fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 atomic_inc(&fn->leaf->rt6i_ref);
1154 rt6_release(rt);
1155 }
1156 fn = fn->parent;
1157 }
1158 /* No more references are possible at this point. */
Pavel Emelyanov2df96af2008-02-18 20:50:42 -08001159 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 }
1161
Thomas Graf86872cb2006-08-22 00:01:08 -07001162 inet6_rt_notify(RTM_DELROUTE, rt, info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 rt6_release(rt);
1164}
1165
Thomas Graf86872cb2006-08-22 00:01:08 -07001166int fib6_del(struct rt6_info *rt, struct nl_info *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167{
Daniel Lezcano8ed67782008-03-04 13:48:30 -08001168 struct net *net = info->nl_net;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 struct fib6_node *fn = rt->rt6i_node;
1170 struct rt6_info **rtp;
1171
1172#if RT6_DEBUG >= 2
Changli Gaod8d1f302010-06-10 23:31:35 -07001173 if (rt->dst.obsolete>0) {
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001174 WARN_ON(fn != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 return -ENOENT;
1176 }
1177#endif
Daniel Lezcano8ed67782008-03-04 13:48:30 -08001178 if (fn == NULL || rt == net->ipv6.ip6_null_entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 return -ENOENT;
1180
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001181 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182
YOSHIFUJI Hideaki150730d2006-08-23 17:22:55 -07001183 if (!(rt->rt6i_flags&RTF_CACHE)) {
1184 struct fib6_node *pn = fn;
1185#ifdef CONFIG_IPV6_SUBTREES
1186 /* clones of this route might be in another subtree */
1187 if (rt->rt6i_src.plen) {
1188 while (!(pn->fn_flags&RTN_ROOT))
1189 pn = pn->parent;
1190 pn = pn->parent;
1191 }
1192#endif
Benjamin Theryec7d43c2008-03-03 23:31:57 -08001193 fib6_prune_clones(info->nl_net, pn, rt);
YOSHIFUJI Hideaki150730d2006-08-23 17:22:55 -07001194 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195
1196 /*
1197 * Walk the leaf entries looking for ourself
1198 */
1199
Changli Gaod8d1f302010-06-10 23:31:35 -07001200 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 if (*rtp == rt) {
Thomas Graf86872cb2006-08-22 00:01:08 -07001202 fib6_del_route(fn, rtp, info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 return 0;
1204 }
1205 }
1206 return -ENOENT;
1207}
1208
1209/*
1210 * Tree traversal function.
1211 *
1212 * Certainly, it is not interrupt safe.
1213 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1214 * It means, that we can modify tree during walking
1215 * and use this function for garbage collection, clone pruning,
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09001216 * cleaning tree when a device goes down etc. etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 *
1218 * It guarantees that every node will be traversed,
1219 * and that it will be traversed only once.
1220 *
1221 * Callback function w->func may return:
1222 * 0 -> continue walking.
1223 * positive value -> walking is suspended (used by tree dumps,
1224 * and probably by gc, if it will be split to several slices)
1225 * negative value -> terminate walking.
1226 *
1227 * The function itself returns:
1228 * 0 -> walk is complete.
1229 * >0 -> walk is incomplete (i.e. suspended)
1230 * <0 -> walk is terminated by an error.
1231 */
1232
Adrian Bunk90d41122006-08-14 23:49:16 -07001233static int fib6_walk_continue(struct fib6_walker_t *w)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234{
1235 struct fib6_node *fn, *pn;
1236
1237 for (;;) {
1238 fn = w->node;
1239 if (fn == NULL)
1240 return 0;
1241
1242 if (w->prune && fn != w->root &&
1243 fn->fn_flags&RTN_RTINFO && w->state < FWS_C) {
1244 w->state = FWS_C;
1245 w->leaf = fn->leaf;
1246 }
1247 switch (w->state) {
1248#ifdef CONFIG_IPV6_SUBTREES
1249 case FWS_S:
YOSHIFUJI Hideaki7fc33162006-08-23 17:22:24 -07001250 if (FIB6_SUBTREE(fn)) {
1251 w->node = FIB6_SUBTREE(fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 continue;
1253 }
1254 w->state = FWS_L;
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09001255#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 case FWS_L:
1257 if (fn->left) {
1258 w->node = fn->left;
1259 w->state = FWS_INIT;
1260 continue;
1261 }
1262 w->state = FWS_R;
1263 case FWS_R:
1264 if (fn->right) {
1265 w->node = fn->right;
1266 w->state = FWS_INIT;
1267 continue;
1268 }
1269 w->state = FWS_C;
1270 w->leaf = fn->leaf;
1271 case FWS_C:
1272 if (w->leaf && fn->fn_flags&RTN_RTINFO) {
Patrick McHardy2bec5a32010-02-08 05:19:03 +00001273 int err;
1274
1275 if (w->count < w->skip) {
1276 w->count++;
1277 continue;
1278 }
1279
1280 err = w->func(w);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 if (err)
1282 return err;
Patrick McHardy2bec5a32010-02-08 05:19:03 +00001283
1284 w->count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 continue;
1286 }
1287 w->state = FWS_U;
1288 case FWS_U:
1289 if (fn == w->root)
1290 return 0;
1291 pn = fn->parent;
1292 w->node = pn;
1293#ifdef CONFIG_IPV6_SUBTREES
YOSHIFUJI Hideaki7fc33162006-08-23 17:22:24 -07001294 if (FIB6_SUBTREE(pn) == fn) {
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001295 WARN_ON(!(fn->fn_flags & RTN_ROOT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 w->state = FWS_L;
1297 continue;
1298 }
1299#endif
1300 if (pn->left == fn) {
1301 w->state = FWS_R;
1302 continue;
1303 }
1304 if (pn->right == fn) {
1305 w->state = FWS_C;
1306 w->leaf = w->node->leaf;
1307 continue;
1308 }
1309#if RT6_DEBUG >= 2
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001310 WARN_ON(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311#endif
1312 }
1313 }
1314}
1315
Adrian Bunk90d41122006-08-14 23:49:16 -07001316static int fib6_walk(struct fib6_walker_t *w)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317{
1318 int res;
1319
1320 w->state = FWS_INIT;
1321 w->node = w->root;
1322
1323 fib6_walker_link(w);
1324 res = fib6_walk_continue(w);
1325 if (res <= 0)
1326 fib6_walker_unlink(w);
1327 return res;
1328}
1329
1330static int fib6_clean_node(struct fib6_walker_t *w)
1331{
1332 int res;
1333 struct rt6_info *rt;
Benjamin Thery0a8891a2007-10-08 20:39:36 -07001334 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
Benjamin Theryec7d43c2008-03-03 23:31:57 -08001335 struct nl_info info = {
1336 .nl_net = c->net,
1337 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
Changli Gaod8d1f302010-06-10 23:31:35 -07001339 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 res = c->func(rt, c->arg);
1341 if (res < 0) {
1342 w->leaf = rt;
Denis V. Lunev528c4ce2007-12-13 09:45:12 -08001343 res = fib6_del(rt, &info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 if (res) {
1345#if RT6_DEBUG >= 2
1346 printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res);
1347#endif
1348 continue;
1349 }
1350 return 0;
1351 }
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001352 WARN_ON(res != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 }
1354 w->leaf = rt;
1355 return 0;
1356}
1357
1358/*
1359 * Convenient frontend to tree walker.
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09001360 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 * func is called on each route.
1362 * It may return -1 -> delete this route.
1363 * 0 -> continue walking
1364 *
1365 * prune==1 -> only immediate children of node (certainly,
1366 * ignoring pure split nodes) will be scanned.
1367 */
1368
Benjamin Theryec7d43c2008-03-03 23:31:57 -08001369static void fib6_clean_tree(struct net *net, struct fib6_node *root,
Adrian Bunk8ce11e62006-08-07 21:50:48 -07001370 int (*func)(struct rt6_info *, void *arg),
1371 int prune, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372{
1373 struct fib6_cleaner_t c;
1374
1375 c.w.root = root;
1376 c.w.func = fib6_clean_node;
1377 c.w.prune = prune;
Patrick McHardy2bec5a32010-02-08 05:19:03 +00001378 c.w.count = 0;
1379 c.w.skip = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 c.func = func;
1381 c.arg = arg;
Benjamin Theryec7d43c2008-03-03 23:31:57 -08001382 c.net = net;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
1384 fib6_walk(&c.w);
1385}
1386
Daniel Lezcanof3db4852008-03-03 23:27:06 -08001387void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
Thomas Grafc71099a2006-08-04 23:20:06 -07001388 int prune, void *arg)
1389{
Thomas Grafc71099a2006-08-04 23:20:06 -07001390 struct fib6_table *table;
Patrick McHardy1b43af52006-08-10 23:11:17 -07001391 struct hlist_node *node;
Daniel Lezcano58f09b72008-03-03 23:25:27 -08001392 struct hlist_head *head;
Patrick McHardy1b43af52006-08-10 23:11:17 -07001393 unsigned int h;
Thomas Grafc71099a2006-08-04 23:20:06 -07001394
Patrick McHardy1b43af52006-08-10 23:11:17 -07001395 rcu_read_lock();
Neil Hormana33bc5c2009-07-30 18:52:15 -07001396 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
Daniel Lezcanof3db4852008-03-03 23:27:06 -08001397 head = &net->ipv6.fib_table_hash[h];
Daniel Lezcano58f09b72008-03-03 23:25:27 -08001398 hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
Thomas Grafc71099a2006-08-04 23:20:06 -07001399 write_lock_bh(&table->tb6_lock);
Benjamin Theryec7d43c2008-03-03 23:31:57 -08001400 fib6_clean_tree(net, &table->tb6_root,
1401 func, prune, arg);
Thomas Grafc71099a2006-08-04 23:20:06 -07001402 write_unlock_bh(&table->tb6_lock);
1403 }
1404 }
Patrick McHardy1b43af52006-08-10 23:11:17 -07001405 rcu_read_unlock();
Thomas Grafc71099a2006-08-04 23:20:06 -07001406}
1407
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1409{
1410 if (rt->rt6i_flags & RTF_CACHE) {
1411 RT6_TRACE("pruning clone %p\n", rt);
1412 return -1;
1413 }
1414
1415 return 0;
1416}
1417
Benjamin Theryec7d43c2008-03-03 23:31:57 -08001418static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1419 struct rt6_info *rt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420{
Benjamin Theryec7d43c2008-03-03 23:31:57 -08001421 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422}
1423
1424/*
1425 * Garbage collection
1426 */
1427
1428static struct fib6_gc_args
1429{
1430 int timeout;
1431 int more;
1432} gc_args;
1433
1434static int fib6_age(struct rt6_info *rt, void *arg)
1435{
1436 unsigned long now = jiffies;
1437
1438 /*
1439 * check addrconf expiration here.
1440 * Routes are expired even if they are in use.
1441 *
1442 * Also age clones. Note, that clones are aged out
1443 * only if they are not in use now.
1444 */
1445
1446 if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) {
1447 if (time_after(now, rt->rt6i_expires)) {
1448 RT6_TRACE("expiring %p\n", rt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 return -1;
1450 }
1451 gc_args.more++;
1452 } else if (rt->rt6i_flags & RTF_CACHE) {
Changli Gaod8d1f302010-06-10 23:31:35 -07001453 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1454 time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 RT6_TRACE("aging clone %p\n", rt);
1456 return -1;
1457 } else if ((rt->rt6i_flags & RTF_GATEWAY) &&
Eric Dumazetf2c31e32011-07-29 19:00:53 +00001458 (!(dst_get_neighbour_raw(&rt->dst)->flags & NTF_ROUTER))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 RT6_TRACE("purging route %p via non-router but gateway\n",
1460 rt);
1461 return -1;
1462 }
1463 gc_args.more++;
1464 }
1465
1466 return 0;
1467}
1468
1469static DEFINE_SPINLOCK(fib6_gc_lock);
1470
Daniel Lezcano5b7c9312008-03-03 23:28:58 -08001471void fib6_run_gc(unsigned long expires, struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472{
Daniel Lezcano5b7c9312008-03-03 23:28:58 -08001473 if (expires != ~0UL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 spin_lock_bh(&fib6_gc_lock);
Daniel Lezcano5b7c9312008-03-03 23:28:58 -08001475 gc_args.timeout = expires ? (int)expires :
1476 net->ipv6.sysctl.ip6_rt_gc_interval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 } else {
Stephen Hemmingera76d7342008-07-22 14:34:35 -07001478 if (!spin_trylock_bh(&fib6_gc_lock)) {
Stephen Hemminger417f28b2008-07-22 14:33:45 -07001479 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 return;
1481 }
Daniel Lezcano5b7c9312008-03-03 23:28:58 -08001482 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484
Stephen Hemminger3d0f24a2008-07-22 14:35:50 -07001485 gc_args.more = icmp6_dst_gc();
Daniel Lezcanof3db4852008-03-03 23:27:06 -08001486
Daniel Lezcano5b7c9312008-03-03 23:28:58 -08001487 fib6_clean_all(net, fib6_age, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488
1489 if (gc_args.more)
Stephen Hemmingerc8a45222008-07-22 14:34:09 -07001490 mod_timer(&net->ipv6.ip6_fib_timer,
1491 round_jiffies(jiffies
1492 + net->ipv6.sysctl.ip6_rt_gc_interval));
Stephen Hemminger417f28b2008-07-22 14:33:45 -07001493 else
1494 del_timer(&net->ipv6.ip6_fib_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 spin_unlock_bh(&fib6_gc_lock);
1496}
1497
Daniel Lezcano5b7c9312008-03-03 23:28:58 -08001498static void fib6_gc_timer_cb(unsigned long arg)
1499{
1500 fib6_run_gc(0, (struct net *)arg);
1501}
1502
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00001503static int __net_init fib6_net_init(struct net *net)
Daniel Lezcano58f09b72008-03-03 23:25:27 -08001504{
Eric Dumazet10da66f2010-10-13 08:22:03 +00001505 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1506
Stephen Hemminger417f28b2008-07-22 14:33:45 -07001507 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
Daniel Lezcano63152fc2008-03-03 23:31:11 -08001508
Benjamin Theryc5728722008-03-03 23:34:17 -08001509 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1510 if (!net->ipv6.rt6_stats)
1511 goto out_timer;
1512
Eric Dumazet10da66f2010-10-13 08:22:03 +00001513 /* Avoid false sharing : Use at least a full cache line */
1514 size = max_t(size_t, size, L1_CACHE_BYTES);
1515
1516 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
Daniel Lezcano58f09b72008-03-03 23:25:27 -08001517 if (!net->ipv6.fib_table_hash)
Benjamin Theryc5728722008-03-03 23:34:17 -08001518 goto out_rt6_stats;
Daniel Lezcano58f09b72008-03-03 23:25:27 -08001519
1520 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1521 GFP_KERNEL);
1522 if (!net->ipv6.fib6_main_tbl)
1523 goto out_fib_table_hash;
1524
1525 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
Daniel Lezcano8ed67782008-03-04 13:48:30 -08001526 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
Daniel Lezcano58f09b72008-03-03 23:25:27 -08001527 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1528 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1529
1530#ifdef CONFIG_IPV6_MULTIPLE_TABLES
1531 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1532 GFP_KERNEL);
1533 if (!net->ipv6.fib6_local_tbl)
1534 goto out_fib6_main_tbl;
1535 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
Daniel Lezcano8ed67782008-03-04 13:48:30 -08001536 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
Daniel Lezcano58f09b72008-03-03 23:25:27 -08001537 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1538 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1539#endif
1540 fib6_tables_init(net);
1541
Stephen Hemminger417f28b2008-07-22 14:33:45 -07001542 return 0;
Daniel Lezcano58f09b72008-03-03 23:25:27 -08001543
1544#ifdef CONFIG_IPV6_MULTIPLE_TABLES
1545out_fib6_main_tbl:
1546 kfree(net->ipv6.fib6_main_tbl);
1547#endif
1548out_fib_table_hash:
1549 kfree(net->ipv6.fib_table_hash);
Benjamin Theryc5728722008-03-03 23:34:17 -08001550out_rt6_stats:
1551 kfree(net->ipv6.rt6_stats);
Daniel Lezcano63152fc2008-03-03 23:31:11 -08001552out_timer:
Stephen Hemminger417f28b2008-07-22 14:33:45 -07001553 return -ENOMEM;
Daniel Lezcano58f09b72008-03-03 23:25:27 -08001554 }
1555
1556static void fib6_net_exit(struct net *net)
1557{
Daniel Lezcano8ed67782008-03-04 13:48:30 -08001558 rt6_ifdown(net, NULL);
Stephen Hemminger417f28b2008-07-22 14:33:45 -07001559 del_timer_sync(&net->ipv6.ip6_fib_timer);
1560
Daniel Lezcano58f09b72008-03-03 23:25:27 -08001561#ifdef CONFIG_IPV6_MULTIPLE_TABLES
1562 kfree(net->ipv6.fib6_local_tbl);
1563#endif
1564 kfree(net->ipv6.fib6_main_tbl);
1565 kfree(net->ipv6.fib_table_hash);
Benjamin Theryc5728722008-03-03 23:34:17 -08001566 kfree(net->ipv6.rt6_stats);
Daniel Lezcano58f09b72008-03-03 23:25:27 -08001567}
1568
1569static struct pernet_operations fib6_net_ops = {
1570 .init = fib6_net_init,
1571 .exit = fib6_net_exit,
1572};
1573
Daniel Lezcanod63bddb2007-12-07 00:40:34 -08001574int __init fib6_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575{
Daniel Lezcanoe0b855902008-03-03 23:24:31 -08001576 int ret = -ENOMEM;
Daniel Lezcano63152fc2008-03-03 23:31:11 -08001577
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1579 sizeof(struct fib6_node),
Daniel Lezcanof845ab62007-12-07 00:45:16 -08001580 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09001581 NULL);
Daniel Lezcanof845ab62007-12-07 00:45:16 -08001582 if (!fib6_node_kmem)
Daniel Lezcanoe0b855902008-03-03 23:24:31 -08001583 goto out;
1584
Daniel Lezcano58f09b72008-03-03 23:25:27 -08001585 ret = register_pernet_subsys(&fib6_net_ops);
1586 if (ret)
Benjamin Theryc5728722008-03-03 23:34:17 -08001587 goto out_kmem_cache_create;
Daniel Lezcanoe0b855902008-03-03 23:24:31 -08001588
Greg Rosec7ac8672011-06-10 01:27:09 +00001589 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1590 NULL);
Daniel Lezcanod63bddb2007-12-07 00:40:34 -08001591 if (ret)
Daniel Lezcano58f09b72008-03-03 23:25:27 -08001592 goto out_unregister_subsys;
Daniel Lezcanod63bddb2007-12-07 00:40:34 -08001593out:
1594 return ret;
1595
Daniel Lezcano58f09b72008-03-03 23:25:27 -08001596out_unregister_subsys:
1597 unregister_pernet_subsys(&fib6_net_ops);
Daniel Lezcanod63bddb2007-12-07 00:40:34 -08001598out_kmem_cache_create:
1599 kmem_cache_destroy(fib6_node_kmem);
1600 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601}
1602
1603void fib6_gc_cleanup(void)
1604{
Daniel Lezcano58f09b72008-03-03 23:25:27 -08001605 unregister_pernet_subsys(&fib6_net_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 kmem_cache_destroy(fib6_node_kmem);
1607}