]>
Commit | Line | Data |
---|---|---|
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 | * Changes: | |
14 | * Yuji SEKIYA @USAGI: Support default route on router node; | |
15 | * remove ip6_null_entry from the top of | |
16 | * routing table. | |
17 | * Ville Nuorvala: Fixed routing subtrees. | |
18 | */ | |
19 | ||
20 | #define pr_fmt(fmt) "IPv6: " fmt | |
21 | ||
22 | #include <linux/errno.h> | |
23 | #include <linux/types.h> | |
24 | #include <linux/net.h> | |
25 | #include <linux/route.h> | |
26 | #include <linux/netdevice.h> | |
27 | #include <linux/in6.h> | |
28 | #include <linux/init.h> | |
29 | #include <linux/list.h> | |
30 | #include <linux/slab.h> | |
31 | ||
32 | #include <net/ipv6.h> | |
33 | #include <net/ndisc.h> | |
34 | #include <net/addrconf.h> | |
35 | #include <net/lwtunnel.h> | |
36 | ||
37 | #include <net/ip6_fib.h> | |
38 | #include <net/ip6_route.h> | |
39 | ||
40 | #define RT6_DEBUG 2 | |
41 | ||
42 | #if RT6_DEBUG >= 3 | |
43 | #define RT6_TRACE(x...) pr_debug(x) | |
44 | #else | |
45 | #define RT6_TRACE(x...) do { ; } while (0) | |
46 | #endif | |
47 | ||
48 | static struct kmem_cache *fib6_node_kmem __read_mostly; | |
49 | ||
50 | struct fib6_cleaner { | |
51 | struct fib6_walker w; | |
52 | struct net *net; | |
53 | int (*func)(struct rt6_info *, void *arg); | |
54 | int sernum; | |
55 | void *arg; | |
56 | }; | |
57 | ||
58 | #ifdef CONFIG_IPV6_SUBTREES | |
59 | #define FWS_INIT FWS_S | |
60 | #else | |
61 | #define FWS_INIT FWS_L | |
62 | #endif | |
63 | ||
64 | static void fib6_prune_clones(struct net *net, struct fib6_node *fn); | |
65 | static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn); | |
66 | static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn); | |
67 | static int fib6_walk(struct net *net, struct fib6_walker *w); | |
68 | static int fib6_walk_continue(struct fib6_walker *w); | |
69 | ||
70 | /* | |
71 | * A routing update causes an increase of the serial number on the | |
72 | * affected subtree. This allows for cached routes to be asynchronously | |
73 | * tested when modifications are made to the destination cache as a | |
74 | * result of redirects, path MTU changes, etc. | |
75 | */ | |
76 | ||
77 | static void fib6_gc_timer_cb(unsigned long arg); | |
78 | ||
79 | #define FOR_WALKERS(net, w) \ | |
80 | list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh) | |
81 | ||
82 | static void fib6_walker_link(struct net *net, struct fib6_walker *w) | |
83 | { | |
84 | write_lock_bh(&net->ipv6.fib6_walker_lock); | |
85 | list_add(&w->lh, &net->ipv6.fib6_walkers); | |
86 | write_unlock_bh(&net->ipv6.fib6_walker_lock); | |
87 | } | |
88 | ||
89 | static void fib6_walker_unlink(struct net *net, struct fib6_walker *w) | |
90 | { | |
91 | write_lock_bh(&net->ipv6.fib6_walker_lock); | |
92 | list_del(&w->lh); | |
93 | write_unlock_bh(&net->ipv6.fib6_walker_lock); | |
94 | } | |
95 | ||
96 | static int fib6_new_sernum(struct net *net) | |
97 | { | |
98 | int new, old; | |
99 | ||
100 | do { | |
101 | old = atomic_read(&net->ipv6.fib6_sernum); | |
102 | new = old < INT_MAX ? old + 1 : 1; | |
103 | } while (atomic_cmpxchg(&net->ipv6.fib6_sernum, | |
104 | old, new) != old); | |
105 | return new; | |
106 | } | |
107 | ||
108 | enum { | |
109 | FIB6_NO_SERNUM_CHANGE = 0, | |
110 | }; | |
111 | ||
112 | /* | |
113 | * Auxiliary address test functions for the radix tree. | |
114 | * | |
115 | * These assume a 32bit processor (although it will work on | |
116 | * 64bit processors) | |
117 | */ | |
118 | ||
119 | /* | |
120 | * test bit | |
121 | */ | |
122 | #if defined(__LITTLE_ENDIAN) | |
123 | # define BITOP_BE32_SWIZZLE (0x1F & ~7) | |
124 | #else | |
125 | # define BITOP_BE32_SWIZZLE 0 | |
126 | #endif | |
127 | ||
128 | static __be32 addr_bit_set(const void *token, int fn_bit) | |
129 | { | |
130 | const __be32 *addr = token; | |
131 | /* | |
132 | * Here, | |
133 | * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f) | |
134 | * is optimized version of | |
135 | * htonl(1 << ((~fn_bit)&0x1F)) | |
136 | * See include/asm-generic/bitops/le.h. | |
137 | */ | |
138 | return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) & | |
139 | addr[fn_bit >> 5]; | |
140 | } | |
141 | ||
142 | static struct fib6_node *node_alloc(void) | |
143 | { | |
144 | struct fib6_node *fn; | |
145 | ||
146 | fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC); | |
147 | ||
148 | return fn; | |
149 | } | |
150 | ||
151 | static void node_free(struct fib6_node *fn) | |
152 | { | |
153 | kmem_cache_free(fib6_node_kmem, fn); | |
154 | } | |
155 | ||
156 | static void rt6_rcu_free(struct rt6_info *rt) | |
157 | { | |
158 | call_rcu(&rt->dst.rcu_head, dst_rcu_free); | |
159 | } | |
160 | ||
161 | static void rt6_free_pcpu(struct rt6_info *non_pcpu_rt) | |
162 | { | |
163 | int cpu; | |
164 | ||
165 | if (!non_pcpu_rt->rt6i_pcpu) | |
166 | return; | |
167 | ||
168 | for_each_possible_cpu(cpu) { | |
169 | struct rt6_info **ppcpu_rt; | |
170 | struct rt6_info *pcpu_rt; | |
171 | ||
172 | ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu); | |
173 | pcpu_rt = *ppcpu_rt; | |
174 | if (pcpu_rt) { | |
175 | rt6_rcu_free(pcpu_rt); | |
176 | *ppcpu_rt = NULL; | |
177 | } | |
178 | } | |
179 | ||
180 | free_percpu(non_pcpu_rt->rt6i_pcpu); | |
181 | non_pcpu_rt->rt6i_pcpu = NULL; | |
182 | } | |
183 | ||
184 | static void rt6_release(struct rt6_info *rt) | |
185 | { | |
186 | if (atomic_dec_and_test(&rt->rt6i_ref)) { | |
187 | rt6_free_pcpu(rt); | |
188 | rt6_rcu_free(rt); | |
189 | } | |
190 | } | |
191 | ||
192 | static void fib6_link_table(struct net *net, struct fib6_table *tb) | |
193 | { | |
194 | unsigned int h; | |
195 | ||
196 | /* | |
197 | * Initialize table lock at a single place to give lockdep a key, | |
198 | * tables aren't visible prior to being linked to the list. | |
199 | */ | |
200 | rwlock_init(&tb->tb6_lock); | |
201 | ||
202 | h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1); | |
203 | ||
204 | /* | |
205 | * No protection necessary, this is the only list mutatation | |
206 | * operation, tables never disappear once they exist. | |
207 | */ | |
208 | hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]); | |
209 | } | |
210 | ||
211 | #ifdef CONFIG_IPV6_MULTIPLE_TABLES | |
212 | ||
213 | static struct fib6_table *fib6_alloc_table(struct net *net, u32 id) | |
214 | { | |
215 | struct fib6_table *table; | |
216 | ||
217 | table = kzalloc(sizeof(*table), GFP_ATOMIC); | |
218 | if (table) { | |
219 | table->tb6_id = id; | |
220 | table->tb6_root.leaf = net->ipv6.ip6_null_entry; | |
221 | table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; | |
222 | inet_peer_base_init(&table->tb6_peers); | |
223 | } | |
224 | ||
225 | return table; | |
226 | } | |
227 | ||
228 | struct fib6_table *fib6_new_table(struct net *net, u32 id) | |
229 | { | |
230 | struct fib6_table *tb; | |
231 | ||
232 | if (id == 0) | |
233 | id = RT6_TABLE_MAIN; | |
234 | tb = fib6_get_table(net, id); | |
235 | if (tb) | |
236 | return tb; | |
237 | ||
238 | tb = fib6_alloc_table(net, id); | |
239 | if (tb) | |
240 | fib6_link_table(net, tb); | |
241 | ||
242 | return tb; | |
243 | } | |
244 | EXPORT_SYMBOL_GPL(fib6_new_table); | |
245 | ||
246 | struct fib6_table *fib6_get_table(struct net *net, u32 id) | |
247 | { | |
248 | struct fib6_table *tb; | |
249 | struct hlist_head *head; | |
250 | unsigned int h; | |
251 | ||
252 | if (id == 0) | |
253 | id = RT6_TABLE_MAIN; | |
254 | h = id & (FIB6_TABLE_HASHSZ - 1); | |
255 | rcu_read_lock(); | |
256 | head = &net->ipv6.fib_table_hash[h]; | |
257 | hlist_for_each_entry_rcu(tb, head, tb6_hlist) { | |
258 | if (tb->tb6_id == id) { | |
259 | rcu_read_unlock(); | |
260 | return tb; | |
261 | } | |
262 | } | |
263 | rcu_read_unlock(); | |
264 | ||
265 | return NULL; | |
266 | } | |
267 | EXPORT_SYMBOL_GPL(fib6_get_table); | |
268 | ||
269 | static void __net_init fib6_tables_init(struct net *net) | |
270 | { | |
271 | fib6_link_table(net, net->ipv6.fib6_main_tbl); | |
272 | fib6_link_table(net, net->ipv6.fib6_local_tbl); | |
273 | } | |
274 | #else | |
275 | ||
276 | struct fib6_table *fib6_new_table(struct net *net, u32 id) | |
277 | { | |
278 | return fib6_get_table(net, id); | |
279 | } | |
280 | ||
281 | struct fib6_table *fib6_get_table(struct net *net, u32 id) | |
282 | { | |
283 | return net->ipv6.fib6_main_tbl; | |
284 | } | |
285 | ||
286 | struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6, | |
287 | int flags, pol_lookup_t lookup) | |
288 | { | |
289 | struct rt6_info *rt; | |
290 | ||
291 | rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, flags); | |
292 | if (rt->rt6i_flags & RTF_REJECT && | |
293 | rt->dst.error == -EAGAIN) { | |
294 | ip6_rt_put(rt); | |
295 | rt = net->ipv6.ip6_null_entry; | |
296 | dst_hold(&rt->dst); | |
297 | } | |
298 | ||
299 | return &rt->dst; | |
300 | } | |
301 | ||
302 | static void __net_init fib6_tables_init(struct net *net) | |
303 | { | |
304 | fib6_link_table(net, net->ipv6.fib6_main_tbl); | |
305 | } | |
306 | ||
307 | #endif | |
308 | ||
309 | static int fib6_dump_node(struct fib6_walker *w) | |
310 | { | |
311 | int res; | |
312 | struct rt6_info *rt; | |
313 | ||
314 | for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { | |
315 | res = rt6_dump_route(rt, w->args); | |
316 | if (res < 0) { | |
317 | /* Frame is full, suspend walking */ | |
318 | w->leaf = rt; | |
319 | return 1; | |
320 | } | |
321 | ||
322 | /* Multipath routes are dumped in one route with the | |
323 | * RTA_MULTIPATH attribute. Jump 'rt' to point to the | |
324 | * last sibling of this route (no need to dump the | |
325 | * sibling routes again) | |
326 | */ | |
327 | if (rt->rt6i_nsiblings) | |
328 | rt = list_last_entry(&rt->rt6i_siblings, | |
329 | struct rt6_info, | |
330 | rt6i_siblings); | |
331 | } | |
332 | w->leaf = NULL; | |
333 | return 0; | |
334 | } | |
335 | ||
336 | static void fib6_dump_end(struct netlink_callback *cb) | |
337 | { | |
338 | struct net *net = sock_net(cb->skb->sk); | |
339 | struct fib6_walker *w = (void *)cb->args[2]; | |
340 | ||
341 | if (w) { | |
342 | if (cb->args[4]) { | |
343 | cb->args[4] = 0; | |
344 | fib6_walker_unlink(net, w); | |
345 | } | |
346 | cb->args[2] = 0; | |
347 | kfree(w); | |
348 | } | |
349 | cb->done = (void *)cb->args[3]; | |
350 | cb->args[1] = 3; | |
351 | } | |
352 | ||
353 | static int fib6_dump_done(struct netlink_callback *cb) | |
354 | { | |
355 | fib6_dump_end(cb); | |
356 | return cb->done ? cb->done(cb) : 0; | |
357 | } | |
358 | ||
359 | static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb, | |
360 | struct netlink_callback *cb) | |
361 | { | |
362 | struct net *net = sock_net(skb->sk); | |
363 | struct fib6_walker *w; | |
364 | int res; | |
365 | ||
366 | w = (void *)cb->args[2]; | |
367 | w->root = &table->tb6_root; | |
368 | ||
369 | if (cb->args[4] == 0) { | |
370 | w->count = 0; | |
371 | w->skip = 0; | |
372 | ||
373 | read_lock_bh(&table->tb6_lock); | |
374 | res = fib6_walk(net, w); | |
375 | read_unlock_bh(&table->tb6_lock); | |
376 | if (res > 0) { | |
377 | cb->args[4] = 1; | |
378 | cb->args[5] = w->root->fn_sernum; | |
379 | } | |
380 | } else { | |
381 | if (cb->args[5] != w->root->fn_sernum) { | |
382 | /* Begin at the root if the tree changed */ | |
383 | cb->args[5] = w->root->fn_sernum; | |
384 | w->state = FWS_INIT; | |
385 | w->node = w->root; | |
386 | w->skip = w->count; | |
387 | } else | |
388 | w->skip = 0; | |
389 | ||
390 | read_lock_bh(&table->tb6_lock); | |
391 | res = fib6_walk_continue(w); | |
392 | read_unlock_bh(&table->tb6_lock); | |
393 | if (res <= 0) { | |
394 | fib6_walker_unlink(net, w); | |
395 | cb->args[4] = 0; | |
396 | } | |
397 | } | |
398 | ||
399 | return res; | |
400 | } | |
401 | ||
402 | static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) | |
403 | { | |
404 | struct net *net = sock_net(skb->sk); | |
405 | unsigned int h, s_h; | |
406 | unsigned int e = 0, s_e; | |
407 | struct rt6_rtnl_dump_arg arg; | |
408 | struct fib6_walker *w; | |
409 | struct fib6_table *tb; | |
410 | struct hlist_head *head; | |
411 | int res = 0; | |
412 | ||
413 | s_h = cb->args[0]; | |
414 | s_e = cb->args[1]; | |
415 | ||
416 | w = (void *)cb->args[2]; | |
417 | if (!w) { | |
418 | /* New dump: | |
419 | * | |
420 | * 1. hook callback destructor. | |
421 | */ | |
422 | cb->args[3] = (long)cb->done; | |
423 | cb->done = fib6_dump_done; | |
424 | ||
425 | /* | |
426 | * 2. allocate and initialize walker. | |
427 | */ | |
428 | w = kzalloc(sizeof(*w), GFP_ATOMIC); | |
429 | if (!w) | |
430 | return -ENOMEM; | |
431 | w->func = fib6_dump_node; | |
432 | cb->args[2] = (long)w; | |
433 | } | |
434 | ||
435 | arg.skb = skb; | |
436 | arg.cb = cb; | |
437 | arg.net = net; | |
438 | w->args = &arg; | |
439 | ||
440 | rcu_read_lock(); | |
441 | for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) { | |
442 | e = 0; | |
443 | head = &net->ipv6.fib_table_hash[h]; | |
444 | hlist_for_each_entry_rcu(tb, head, tb6_hlist) { | |
445 | if (e < s_e) | |
446 | goto next; | |
447 | res = fib6_dump_table(tb, skb, cb); | |
448 | if (res != 0) | |
449 | goto out; | |
450 | next: | |
451 | e++; | |
452 | } | |
453 | } | |
454 | out: | |
455 | rcu_read_unlock(); | |
456 | cb->args[1] = e; | |
457 | cb->args[0] = h; | |
458 | ||
459 | res = res < 0 ? res : skb->len; | |
460 | if (res <= 0) | |
461 | fib6_dump_end(cb); | |
462 | return res; | |
463 | } | |
464 | ||
465 | /* | |
466 | * Routing Table | |
467 | * | |
468 | * return the appropriate node for a routing tree "add" operation | |
469 | * by either creating and inserting or by returning an existing | |
470 | * node. | |
471 | */ | |
472 | ||
473 | static struct fib6_node *fib6_add_1(struct fib6_node *root, | |
474 | struct in6_addr *addr, int plen, | |
475 | int offset, int allow_create, | |
476 | int replace_required, int sernum) | |
477 | { | |
478 | struct fib6_node *fn, *in, *ln; | |
479 | struct fib6_node *pn = NULL; | |
480 | struct rt6key *key; | |
481 | int bit; | |
482 | __be32 dir = 0; | |
483 | ||
484 | RT6_TRACE("fib6_add_1\n"); | |
485 | ||
486 | /* insert node in tree */ | |
487 | ||
488 | fn = root; | |
489 | ||
490 | do { | |
491 | key = (struct rt6key *)((u8 *)fn->leaf + offset); | |
492 | ||
493 | /* | |
494 | * Prefix match | |
495 | */ | |
496 | if (plen < fn->fn_bit || | |
497 | !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) { | |
498 | if (!allow_create) { | |
499 | if (replace_required) { | |
500 | pr_warn("Can't replace route, no match found\n"); | |
501 | return ERR_PTR(-ENOENT); | |
502 | } | |
503 | pr_warn("NLM_F_CREATE should be set when creating new route\n"); | |
504 | } | |
505 | goto insert_above; | |
506 | } | |
507 | ||
508 | /* | |
509 | * Exact match ? | |
510 | */ | |
511 | ||
512 | if (plen == fn->fn_bit) { | |
513 | /* clean up an intermediate node */ | |
514 | if (!(fn->fn_flags & RTN_RTINFO)) { | |
515 | rt6_release(fn->leaf); | |
516 | fn->leaf = NULL; | |
517 | } | |
518 | ||
519 | fn->fn_sernum = sernum; | |
520 | ||
521 | return fn; | |
522 | } | |
523 | ||
524 | /* | |
525 | * We have more bits to go | |
526 | */ | |
527 | ||
528 | /* Try to walk down on tree. */ | |
529 | fn->fn_sernum = sernum; | |
530 | dir = addr_bit_set(addr, fn->fn_bit); | |
531 | pn = fn; | |
532 | fn = dir ? fn->right : fn->left; | |
533 | } while (fn); | |
534 | ||
535 | if (!allow_create) { | |
536 | /* We should not create new node because | |
537 | * NLM_F_REPLACE was specified without NLM_F_CREATE | |
538 | * I assume it is safe to require NLM_F_CREATE when | |
539 | * REPLACE flag is used! Later we may want to remove the | |
540 | * check for replace_required, because according | |
541 | * to netlink specification, NLM_F_CREATE | |
542 | * MUST be specified if new route is created. | |
543 | * That would keep IPv6 consistent with IPv4 | |
544 | */ | |
545 | if (replace_required) { | |
546 | pr_warn("Can't replace route, no match found\n"); | |
547 | return ERR_PTR(-ENOENT); | |
548 | } | |
549 | pr_warn("NLM_F_CREATE should be set when creating new route\n"); | |
550 | } | |
551 | /* | |
552 | * We walked to the bottom of tree. | |
553 | * Create new leaf node without children. | |
554 | */ | |
555 | ||
556 | ln = node_alloc(); | |
557 | ||
558 | if (!ln) | |
559 | return ERR_PTR(-ENOMEM); | |
560 | ln->fn_bit = plen; | |
561 | ||
562 | ln->parent = pn; | |
563 | ln->fn_sernum = sernum; | |
564 | ||
565 | if (dir) | |
566 | pn->right = ln; | |
567 | else | |
568 | pn->left = ln; | |
569 | ||
570 | return ln; | |
571 | ||
572 | ||
573 | insert_above: | |
574 | /* | |
575 | * split since we don't have a common prefix anymore or | |
576 | * we have a less significant route. | |
577 | * we've to insert an intermediate node on the list | |
578 | * this new node will point to the one we need to create | |
579 | * and the current | |
580 | */ | |
581 | ||
582 | pn = fn->parent; | |
583 | ||
584 | /* find 1st bit in difference between the 2 addrs. | |
585 | ||
586 | See comment in __ipv6_addr_diff: bit may be an invalid value, | |
587 | but if it is >= plen, the value is ignored in any case. | |
588 | */ | |
589 | ||
590 | bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr)); | |
591 | ||
592 | /* | |
593 | * (intermediate)[in] | |
594 | * / \ | |
595 | * (new leaf node)[ln] (old node)[fn] | |
596 | */ | |
597 | if (plen > bit) { | |
598 | in = node_alloc(); | |
599 | ln = node_alloc(); | |
600 | ||
601 | if (!in || !ln) { | |
602 | if (in) | |
603 | node_free(in); | |
604 | if (ln) | |
605 | node_free(ln); | |
606 | return ERR_PTR(-ENOMEM); | |
607 | } | |
608 | ||
609 | /* | |
610 | * new intermediate node. | |
611 | * RTN_RTINFO will | |
612 | * be off since that an address that chooses one of | |
613 | * the branches would not match less specific routes | |
614 | * in the other branch | |
615 | */ | |
616 | ||
617 | in->fn_bit = bit; | |
618 | ||
619 | in->parent = pn; | |
620 | in->leaf = fn->leaf; | |
621 | atomic_inc(&in->leaf->rt6i_ref); | |
622 | ||
623 | in->fn_sernum = sernum; | |
624 | ||
625 | /* update parent pointer */ | |
626 | if (dir) | |
627 | pn->right = in; | |
628 | else | |
629 | pn->left = in; | |
630 | ||
631 | ln->fn_bit = plen; | |
632 | ||
633 | ln->parent = in; | |
634 | fn->parent = in; | |
635 | ||
636 | ln->fn_sernum = sernum; | |
637 | ||
638 | if (addr_bit_set(addr, bit)) { | |
639 | in->right = ln; | |
640 | in->left = fn; | |
641 | } else { | |
642 | in->left = ln; | |
643 | in->right = fn; | |
644 | } | |
645 | } else { /* plen <= bit */ | |
646 | ||
647 | /* | |
648 | * (new leaf node)[ln] | |
649 | * / \ | |
650 | * (old node)[fn] NULL | |
651 | */ | |
652 | ||
653 | ln = node_alloc(); | |
654 | ||
655 | if (!ln) | |
656 | return ERR_PTR(-ENOMEM); | |
657 | ||
658 | ln->fn_bit = plen; | |
659 | ||
660 | ln->parent = pn; | |
661 | ||
662 | ln->fn_sernum = sernum; | |
663 | ||
664 | if (dir) | |
665 | pn->right = ln; | |
666 | else | |
667 | pn->left = ln; | |
668 | ||
669 | if (addr_bit_set(&key->addr, plen)) | |
670 | ln->right = fn; | |
671 | else | |
672 | ln->left = fn; | |
673 | ||
674 | fn->parent = ln; | |
675 | } | |
676 | return ln; | |
677 | } | |
678 | ||
679 | static bool rt6_qualify_for_ecmp(struct rt6_info *rt) | |
680 | { | |
681 | return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) == | |
682 | RTF_GATEWAY; | |
683 | } | |
684 | ||
685 | static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc) | |
686 | { | |
687 | int i; | |
688 | ||
689 | for (i = 0; i < RTAX_MAX; i++) { | |
690 | if (test_bit(i, mxc->mx_valid)) | |
691 | mp[i] = mxc->mx[i]; | |
692 | } | |
693 | } | |
694 | ||
695 | static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc) | |
696 | { | |
697 | if (!mxc->mx) | |
698 | return 0; | |
699 | ||
700 | if (dst->flags & DST_HOST) { | |
701 | u32 *mp = dst_metrics_write_ptr(dst); | |
702 | ||
703 | if (unlikely(!mp)) | |
704 | return -ENOMEM; | |
705 | ||
706 | fib6_copy_metrics(mp, mxc); | |
707 | } else { | |
708 | dst_init_metrics(dst, mxc->mx, false); | |
709 | ||
710 | /* We've stolen mx now. */ | |
711 | mxc->mx = NULL; | |
712 | } | |
713 | ||
714 | return 0; | |
715 | } | |
716 | ||
717 | static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn, | |
718 | struct net *net) | |
719 | { | |
720 | if (atomic_read(&rt->rt6i_ref) != 1) { | |
721 | /* This route is used as dummy address holder in some split | |
722 | * nodes. It is not leaked, but it still holds other resources, | |
723 | * which must be released in time. So, scan ascendant nodes | |
724 | * and replace dummy references to this route with references | |
725 | * to still alive ones. | |
726 | */ | |
727 | while (fn) { | |
728 | if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) { | |
729 | fn->leaf = fib6_find_prefix(net, fn); | |
730 | atomic_inc(&fn->leaf->rt6i_ref); | |
731 | rt6_release(rt); | |
732 | } | |
733 | fn = fn->parent; | |
734 | } | |
735 | /* No more references are possible at this point. */ | |
736 | BUG_ON(atomic_read(&rt->rt6i_ref) != 1); | |
737 | } | |
738 | } | |
739 | ||
740 | /* | |
741 | * Insert routing information in a node. | |
742 | */ | |
743 | ||
744 | static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt, | |
745 | struct nl_info *info, struct mx6_config *mxc) | |
746 | { | |
747 | struct rt6_info *iter = NULL; | |
748 | struct rt6_info **ins; | |
749 | struct rt6_info **fallback_ins = NULL; | |
750 | int replace = (info->nlh && | |
751 | (info->nlh->nlmsg_flags & NLM_F_REPLACE)); | |
752 | int add = (!info->nlh || | |
753 | (info->nlh->nlmsg_flags & NLM_F_CREATE)); | |
754 | int found = 0; | |
755 | bool rt_can_ecmp = rt6_qualify_for_ecmp(rt); | |
756 | u16 nlflags = NLM_F_EXCL; | |
757 | int err; | |
758 | ||
759 | if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND)) | |
760 | nlflags |= NLM_F_APPEND; | |
761 | ||
762 | ins = &fn->leaf; | |
763 | ||
764 | for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) { | |
765 | /* | |
766 | * Search for duplicates | |
767 | */ | |
768 | ||
769 | if (iter->rt6i_metric == rt->rt6i_metric) { | |
770 | /* | |
771 | * Same priority level | |
772 | */ | |
773 | if (info->nlh && | |
774 | (info->nlh->nlmsg_flags & NLM_F_EXCL)) | |
775 | return -EEXIST; | |
776 | ||
777 | nlflags &= ~NLM_F_EXCL; | |
778 | if (replace) { | |
779 | if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) { | |
780 | found++; | |
781 | break; | |
782 | } | |
783 | if (rt_can_ecmp) | |
784 | fallback_ins = fallback_ins ?: ins; | |
785 | goto next_iter; | |
786 | } | |
787 | ||
788 | if (iter->dst.dev == rt->dst.dev && | |
789 | iter->rt6i_idev == rt->rt6i_idev && | |
790 | ipv6_addr_equal(&iter->rt6i_gateway, | |
791 | &rt->rt6i_gateway)) { | |
792 | if (rt->rt6i_nsiblings) | |
793 | rt->rt6i_nsiblings = 0; | |
794 | if (!(iter->rt6i_flags & RTF_EXPIRES)) | |
795 | return -EEXIST; | |
796 | if (!(rt->rt6i_flags & RTF_EXPIRES)) | |
797 | rt6_clean_expires(iter); | |
798 | else | |
799 | rt6_set_expires(iter, rt->dst.expires); | |
800 | iter->rt6i_pmtu = rt->rt6i_pmtu; | |
801 | return -EEXIST; | |
802 | } | |
803 | /* If we have the same destination and the same metric, | |
804 | * but not the same gateway, then the route we try to | |
805 | * add is sibling to this route, increment our counter | |
806 | * of siblings, and later we will add our route to the | |
807 | * list. | |
808 | * Only static routes (which don't have flag | |
809 | * RTF_EXPIRES) are used for ECMPv6. | |
810 | * | |
811 | * To avoid long list, we only had siblings if the | |
812 | * route have a gateway. | |
813 | */ | |
814 | if (rt_can_ecmp && | |
815 | rt6_qualify_for_ecmp(iter)) | |
816 | rt->rt6i_nsiblings++; | |
817 | } | |
818 | ||
819 | if (iter->rt6i_metric > rt->rt6i_metric) | |
820 | break; | |
821 | ||
822 | next_iter: | |
823 | ins = &iter->dst.rt6_next; | |
824 | } | |
825 | ||
826 | if (fallback_ins && !found) { | |
827 | /* No ECMP-able route found, replace first non-ECMP one */ | |
828 | ins = fallback_ins; | |
829 | iter = *ins; | |
830 | found++; | |
831 | } | |
832 | ||
833 | /* Reset round-robin state, if necessary */ | |
834 | if (ins == &fn->leaf) | |
835 | fn->rr_ptr = NULL; | |
836 | ||
837 | /* Link this route to others same route. */ | |
838 | if (rt->rt6i_nsiblings) { | |
839 | unsigned int rt6i_nsiblings; | |
840 | struct rt6_info *sibling, *temp_sibling; | |
841 | ||
842 | /* Find the first route that have the same metric */ | |
843 | sibling = fn->leaf; | |
844 | while (sibling) { | |
845 | if (sibling->rt6i_metric == rt->rt6i_metric && | |
846 | rt6_qualify_for_ecmp(sibling)) { | |
847 | list_add_tail(&rt->rt6i_siblings, | |
848 | &sibling->rt6i_siblings); | |
849 | break; | |
850 | } | |
851 | sibling = sibling->dst.rt6_next; | |
852 | } | |
853 | /* For each sibling in the list, increment the counter of | |
854 | * siblings. BUG() if counters does not match, list of siblings | |
855 | * is broken! | |
856 | */ | |
857 | rt6i_nsiblings = 0; | |
858 | list_for_each_entry_safe(sibling, temp_sibling, | |
859 | &rt->rt6i_siblings, rt6i_siblings) { | |
860 | sibling->rt6i_nsiblings++; | |
861 | BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings); | |
862 | rt6i_nsiblings++; | |
863 | } | |
864 | BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings); | |
865 | } | |
866 | ||
867 | /* | |
868 | * insert node | |
869 | */ | |
870 | if (!replace) { | |
871 | if (!add) | |
872 | pr_warn("NLM_F_CREATE should be set when creating new route\n"); | |
873 | ||
874 | add: | |
875 | nlflags |= NLM_F_CREATE; | |
876 | err = fib6_commit_metrics(&rt->dst, mxc); | |
877 | if (err) | |
878 | return err; | |
879 | ||
880 | rt->dst.rt6_next = iter; | |
881 | *ins = rt; | |
882 | rt->rt6i_node = fn; | |
883 | atomic_inc(&rt->rt6i_ref); | |
884 | inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags); | |
885 | info->nl_net->ipv6.rt6_stats->fib_rt_entries++; | |
886 | ||
887 | if (!(fn->fn_flags & RTN_RTINFO)) { | |
888 | info->nl_net->ipv6.rt6_stats->fib_route_nodes++; | |
889 | fn->fn_flags |= RTN_RTINFO; | |
890 | } | |
891 | ||
892 | } else { | |
893 | int nsiblings; | |
894 | ||
895 | if (!found) { | |
896 | if (add) | |
897 | goto add; | |
898 | pr_warn("NLM_F_REPLACE set, but no existing node found!\n"); | |
899 | return -ENOENT; | |
900 | } | |
901 | ||
902 | err = fib6_commit_metrics(&rt->dst, mxc); | |
903 | if (err) | |
904 | return err; | |
905 | ||
906 | *ins = rt; | |
907 | rt->rt6i_node = fn; | |
908 | rt->dst.rt6_next = iter->dst.rt6_next; | |
909 | atomic_inc(&rt->rt6i_ref); | |
910 | inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE); | |
911 | if (!(fn->fn_flags & RTN_RTINFO)) { | |
912 | info->nl_net->ipv6.rt6_stats->fib_route_nodes++; | |
913 | fn->fn_flags |= RTN_RTINFO; | |
914 | } | |
915 | nsiblings = iter->rt6i_nsiblings; | |
916 | fib6_purge_rt(iter, fn, info->nl_net); | |
917 | rt6_release(iter); | |
918 | ||
919 | if (nsiblings) { | |
920 | /* Replacing an ECMP route, remove all siblings */ | |
921 | ins = &rt->dst.rt6_next; | |
922 | iter = *ins; | |
923 | while (iter) { | |
924 | if (rt6_qualify_for_ecmp(iter)) { | |
925 | *ins = iter->dst.rt6_next; | |
926 | fib6_purge_rt(iter, fn, info->nl_net); | |
927 | rt6_release(iter); | |
928 | nsiblings--; | |
929 | } else { | |
930 | ins = &iter->dst.rt6_next; | |
931 | } | |
932 | iter = *ins; | |
933 | } | |
934 | WARN_ON(nsiblings != 0); | |
935 | } | |
936 | } | |
937 | ||
938 | return 0; | |
939 | } | |
940 | ||
941 | static void fib6_start_gc(struct net *net, struct rt6_info *rt) | |
942 | { | |
943 | if (!timer_pending(&net->ipv6.ip6_fib_timer) && | |
944 | (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE))) | |
945 | mod_timer(&net->ipv6.ip6_fib_timer, | |
946 | jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); | |
947 | } | |
948 | ||
949 | void fib6_force_start_gc(struct net *net) | |
950 | { | |
951 | if (!timer_pending(&net->ipv6.ip6_fib_timer)) | |
952 | mod_timer(&net->ipv6.ip6_fib_timer, | |
953 | jiffies + net->ipv6.sysctl.ip6_rt_gc_interval); | |
954 | } | |
955 | ||
956 | /* | |
957 | * Add routing information to the routing tree. | |
958 | * <destination addr>/<source addr> | |
959 | * with source addr info in sub-trees | |
960 | */ | |
961 | ||
962 | int fib6_add(struct fib6_node *root, struct rt6_info *rt, | |
963 | struct nl_info *info, struct mx6_config *mxc) | |
964 | { | |
965 | struct fib6_node *fn, *pn = NULL; | |
966 | int err = -ENOMEM; | |
967 | int allow_create = 1; | |
968 | int replace_required = 0; | |
969 | int sernum = fib6_new_sernum(info->nl_net); | |
970 | ||
971 | if (WARN_ON_ONCE((rt->dst.flags & DST_NOCACHE) && | |
972 | !atomic_read(&rt->dst.__refcnt))) | |
973 | return -EINVAL; | |
974 | ||
975 | if (info->nlh) { | |
976 | if (!(info->nlh->nlmsg_flags & NLM_F_CREATE)) | |
977 | allow_create = 0; | |
978 | if (info->nlh->nlmsg_flags & NLM_F_REPLACE) | |
979 | replace_required = 1; | |
980 | } | |
981 | if (!allow_create && !replace_required) | |
982 | pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n"); | |
983 | ||
984 | fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen, | |
985 | offsetof(struct rt6_info, rt6i_dst), allow_create, | |
986 | replace_required, sernum); | |
987 | if (IS_ERR(fn)) { | |
988 | err = PTR_ERR(fn); | |
989 | fn = NULL; | |
990 | goto out; | |
991 | } | |
992 | ||
993 | pn = fn; | |
994 | ||
995 | #ifdef CONFIG_IPV6_SUBTREES | |
996 | if (rt->rt6i_src.plen) { | |
997 | struct fib6_node *sn; | |
998 | ||
999 | if (!fn->subtree) { | |
1000 | struct fib6_node *sfn; | |
1001 | ||
1002 | /* | |
1003 | * Create subtree. | |
1004 | * | |
1005 | * fn[main tree] | |
1006 | * | | |
1007 | * sfn[subtree root] | |
1008 | * \ | |
1009 | * sn[new leaf node] | |
1010 | */ | |
1011 | ||
1012 | /* Create subtree root node */ | |
1013 | sfn = node_alloc(); | |
1014 | if (!sfn) | |
1015 | goto st_failure; | |
1016 | ||
1017 | sfn->leaf = info->nl_net->ipv6.ip6_null_entry; | |
1018 | atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref); | |
1019 | sfn->fn_flags = RTN_ROOT; | |
1020 | sfn->fn_sernum = sernum; | |
1021 | ||
1022 | /* Now add the first leaf node to new subtree */ | |
1023 | ||
1024 | sn = fib6_add_1(sfn, &rt->rt6i_src.addr, | |
1025 | rt->rt6i_src.plen, | |
1026 | offsetof(struct rt6_info, rt6i_src), | |
1027 | allow_create, replace_required, sernum); | |
1028 | ||
1029 | if (IS_ERR(sn)) { | |
1030 | /* If it is failed, discard just allocated | |
1031 | root, and then (in st_failure) stale node | |
1032 | in main tree. | |
1033 | */ | |
1034 | node_free(sfn); | |
1035 | err = PTR_ERR(sn); | |
1036 | goto st_failure; | |
1037 | } | |
1038 | ||
1039 | /* Now link new subtree to main tree */ | |
1040 | sfn->parent = fn; | |
1041 | fn->subtree = sfn; | |
1042 | } else { | |
1043 | sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr, | |
1044 | rt->rt6i_src.plen, | |
1045 | offsetof(struct rt6_info, rt6i_src), | |
1046 | allow_create, replace_required, sernum); | |
1047 | ||
1048 | if (IS_ERR(sn)) { | |
1049 | err = PTR_ERR(sn); | |
1050 | goto st_failure; | |
1051 | } | |
1052 | } | |
1053 | ||
1054 | if (!fn->leaf) { | |
1055 | fn->leaf = rt; | |
1056 | atomic_inc(&rt->rt6i_ref); | |
1057 | } | |
1058 | fn = sn; | |
1059 | } | |
1060 | #endif | |
1061 | ||
1062 | err = fib6_add_rt2node(fn, rt, info, mxc); | |
1063 | if (!err) { | |
1064 | fib6_start_gc(info->nl_net, rt); | |
1065 | if (!(rt->rt6i_flags & RTF_CACHE)) | |
1066 | fib6_prune_clones(info->nl_net, pn); | |
1067 | rt->dst.flags &= ~DST_NOCACHE; | |
1068 | } | |
1069 | ||
1070 | out: | |
1071 | if (err) { | |
1072 | #ifdef CONFIG_IPV6_SUBTREES | |
1073 | /* | |
1074 | * If fib6_add_1 has cleared the old leaf pointer in the | |
1075 | * super-tree leaf node we have to find a new one for it. | |
1076 | */ | |
1077 | if (pn != fn && pn->leaf == rt) { | |
1078 | pn->leaf = NULL; | |
1079 | atomic_dec(&rt->rt6i_ref); | |
1080 | } | |
1081 | if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) { | |
1082 | pn->leaf = fib6_find_prefix(info->nl_net, pn); | |
1083 | #if RT6_DEBUG >= 2 | |
1084 | if (!pn->leaf) { | |
1085 | WARN_ON(pn->leaf == NULL); | |
1086 | pn->leaf = info->nl_net->ipv6.ip6_null_entry; | |
1087 | } | |
1088 | #endif | |
1089 | atomic_inc(&pn->leaf->rt6i_ref); | |
1090 | } | |
1091 | #endif | |
1092 | if (!(rt->dst.flags & DST_NOCACHE)) | |
1093 | dst_free(&rt->dst); | |
1094 | } | |
1095 | return err; | |
1096 | ||
1097 | #ifdef CONFIG_IPV6_SUBTREES | |
1098 | /* Subtree creation failed, probably main tree node | |
1099 | is orphan. If it is, shoot it. | |
1100 | */ | |
1101 | st_failure: | |
1102 | if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT))) | |
1103 | fib6_repair_tree(info->nl_net, fn); | |
1104 | if (!(rt->dst.flags & DST_NOCACHE)) | |
1105 | dst_free(&rt->dst); | |
1106 | return err; | |
1107 | #endif | |
1108 | } | |
1109 | ||
1110 | /* | |
1111 | * Routing tree lookup | |
1112 | * | |
1113 | */ | |
1114 | ||
1115 | struct lookup_args { | |
1116 | int offset; /* key offset on rt6_info */ | |
1117 | const struct in6_addr *addr; /* search key */ | |
1118 | }; | |
1119 | ||
1120 | static struct fib6_node *fib6_lookup_1(struct fib6_node *root, | |
1121 | struct lookup_args *args) | |
1122 | { | |
1123 | struct fib6_node *fn; | |
1124 | __be32 dir; | |
1125 | ||
1126 | if (unlikely(args->offset == 0)) | |
1127 | return NULL; | |
1128 | ||
1129 | /* | |
1130 | * Descend on a tree | |
1131 | */ | |
1132 | ||
1133 | fn = root; | |
1134 | ||
1135 | for (;;) { | |
1136 | struct fib6_node *next; | |
1137 | ||
1138 | dir = addr_bit_set(args->addr, fn->fn_bit); | |
1139 | ||
1140 | next = dir ? fn->right : fn->left; | |
1141 | ||
1142 | if (next) { | |
1143 | fn = next; | |
1144 | continue; | |
1145 | } | |
1146 | break; | |
1147 | } | |
1148 | ||
1149 | while (fn) { | |
1150 | if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) { | |
1151 | struct rt6key *key; | |
1152 | ||
1153 | key = (struct rt6key *) ((u8 *) fn->leaf + | |
1154 | args->offset); | |
1155 | ||
1156 | if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) { | |
1157 | #ifdef CONFIG_IPV6_SUBTREES | |
1158 | if (fn->subtree) { | |
1159 | struct fib6_node *sfn; | |
1160 | sfn = fib6_lookup_1(fn->subtree, | |
1161 | args + 1); | |
1162 | if (!sfn) | |
1163 | goto backtrack; | |
1164 | fn = sfn; | |
1165 | } | |
1166 | #endif | |
1167 | if (fn->fn_flags & RTN_RTINFO) | |
1168 | return fn; | |
1169 | } | |
1170 | } | |
1171 | #ifdef CONFIG_IPV6_SUBTREES | |
1172 | backtrack: | |
1173 | #endif | |
1174 | if (fn->fn_flags & RTN_ROOT) | |
1175 | break; | |
1176 | ||
1177 | fn = fn->parent; | |
1178 | } | |
1179 | ||
1180 | return NULL; | |
1181 | } | |
1182 | ||
1183 | struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr, | |
1184 | const struct in6_addr *saddr) | |
1185 | { | |
1186 | struct fib6_node *fn; | |
1187 | struct lookup_args args[] = { | |
1188 | { | |
1189 | .offset = offsetof(struct rt6_info, rt6i_dst), | |
1190 | .addr = daddr, | |
1191 | }, | |
1192 | #ifdef CONFIG_IPV6_SUBTREES | |
1193 | { | |
1194 | .offset = offsetof(struct rt6_info, rt6i_src), | |
1195 | .addr = saddr, | |
1196 | }, | |
1197 | #endif | |
1198 | { | |
1199 | .offset = 0, /* sentinel */ | |
1200 | } | |
1201 | }; | |
1202 | ||
1203 | fn = fib6_lookup_1(root, daddr ? args : args + 1); | |
1204 | if (!fn || fn->fn_flags & RTN_TL_ROOT) | |
1205 | fn = root; | |
1206 | ||
1207 | return fn; | |
1208 | } | |
1209 | ||
1210 | /* | |
1211 | * Get node with specified destination prefix (and source prefix, | |
1212 | * if subtrees are used) | |
1213 | */ | |
1214 | ||
1215 | ||
1216 | static struct fib6_node *fib6_locate_1(struct fib6_node *root, | |
1217 | const struct in6_addr *addr, | |
1218 | int plen, int offset) | |
1219 | { | |
1220 | struct fib6_node *fn; | |
1221 | ||
1222 | for (fn = root; fn ; ) { | |
1223 | struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset); | |
1224 | ||
1225 | /* | |
1226 | * Prefix match | |
1227 | */ | |
1228 | if (plen < fn->fn_bit || | |
1229 | !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) | |
1230 | return NULL; | |
1231 | ||
1232 | if (plen == fn->fn_bit) | |
1233 | return fn; | |
1234 | ||
1235 | /* | |
1236 | * We have more bits to go | |
1237 | */ | |
1238 | if (addr_bit_set(addr, fn->fn_bit)) | |
1239 | fn = fn->right; | |
1240 | else | |
1241 | fn = fn->left; | |
1242 | } | |
1243 | return NULL; | |
1244 | } | |
1245 | ||
1246 | struct fib6_node *fib6_locate(struct fib6_node *root, | |
1247 | const struct in6_addr *daddr, int dst_len, | |
1248 | const struct in6_addr *saddr, int src_len) | |
1249 | { | |
1250 | struct fib6_node *fn; | |
1251 | ||
1252 | fn = fib6_locate_1(root, daddr, dst_len, | |
1253 | offsetof(struct rt6_info, rt6i_dst)); | |
1254 | ||
1255 | #ifdef CONFIG_IPV6_SUBTREES | |
1256 | if (src_len) { | |
1257 | WARN_ON(saddr == NULL); | |
1258 | if (fn && fn->subtree) | |
1259 | fn = fib6_locate_1(fn->subtree, saddr, src_len, | |
1260 | offsetof(struct rt6_info, rt6i_src)); | |
1261 | } | |
1262 | #endif | |
1263 | ||
1264 | if (fn && fn->fn_flags & RTN_RTINFO) | |
1265 | return fn; | |
1266 | ||
1267 | return NULL; | |
1268 | } | |
1269 | ||
1270 | ||
1271 | /* | |
1272 | * Deletion | |
1273 | * | |
1274 | */ | |
1275 | ||
1276 | static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn) | |
1277 | { | |
1278 | if (fn->fn_flags & RTN_ROOT) | |
1279 | return net->ipv6.ip6_null_entry; | |
1280 | ||
1281 | while (fn) { | |
1282 | if (fn->left) | |
1283 | return fn->left->leaf; | |
1284 | if (fn->right) | |
1285 | return fn->right->leaf; | |
1286 | ||
1287 | fn = FIB6_SUBTREE(fn); | |
1288 | } | |
1289 | return NULL; | |
1290 | } | |
1291 | ||
1292 | /* | |
1293 | * Called to trim the tree of intermediate nodes when possible. "fn" | |
1294 | * is the node we want to try and remove. | |
1295 | */ | |
1296 | ||
1297 | static struct fib6_node *fib6_repair_tree(struct net *net, | |
1298 | struct fib6_node *fn) | |
1299 | { | |
1300 | int children; | |
1301 | int nstate; | |
1302 | struct fib6_node *child, *pn; | |
1303 | struct fib6_walker *w; | |
1304 | int iter = 0; | |
1305 | ||
1306 | for (;;) { | |
1307 | RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); | |
1308 | iter++; | |
1309 | ||
1310 | WARN_ON(fn->fn_flags & RTN_RTINFO); | |
1311 | WARN_ON(fn->fn_flags & RTN_TL_ROOT); | |
1312 | WARN_ON(fn->leaf); | |
1313 | ||
1314 | children = 0; | |
1315 | child = NULL; | |
1316 | if (fn->right) | |
1317 | child = fn->right, children |= 1; | |
1318 | if (fn->left) | |
1319 | child = fn->left, children |= 2; | |
1320 | ||
1321 | if (children == 3 || FIB6_SUBTREE(fn) | |
1322 | #ifdef CONFIG_IPV6_SUBTREES | |
1323 | /* Subtree root (i.e. fn) may have one child */ | |
1324 | || (children && fn->fn_flags & RTN_ROOT) | |
1325 | #endif | |
1326 | ) { | |
1327 | fn->leaf = fib6_find_prefix(net, fn); | |
1328 | #if RT6_DEBUG >= 2 | |
1329 | if (!fn->leaf) { | |
1330 | WARN_ON(!fn->leaf); | |
1331 | fn->leaf = net->ipv6.ip6_null_entry; | |
1332 | } | |
1333 | #endif | |
1334 | atomic_inc(&fn->leaf->rt6i_ref); | |
1335 | return fn->parent; | |
1336 | } | |
1337 | ||
1338 | pn = fn->parent; | |
1339 | #ifdef CONFIG_IPV6_SUBTREES | |
1340 | if (FIB6_SUBTREE(pn) == fn) { | |
1341 | WARN_ON(!(fn->fn_flags & RTN_ROOT)); | |
1342 | FIB6_SUBTREE(pn) = NULL; | |
1343 | nstate = FWS_L; | |
1344 | } else { | |
1345 | WARN_ON(fn->fn_flags & RTN_ROOT); | |
1346 | #endif | |
1347 | if (pn->right == fn) | |
1348 | pn->right = child; | |
1349 | else if (pn->left == fn) | |
1350 | pn->left = child; | |
1351 | #if RT6_DEBUG >= 2 | |
1352 | else | |
1353 | WARN_ON(1); | |
1354 | #endif | |
1355 | if (child) | |
1356 | child->parent = pn; | |
1357 | nstate = FWS_R; | |
1358 | #ifdef CONFIG_IPV6_SUBTREES | |
1359 | } | |
1360 | #endif | |
1361 | ||
1362 | read_lock(&net->ipv6.fib6_walker_lock); | |
1363 | FOR_WALKERS(net, w) { | |
1364 | if (!child) { | |
1365 | if (w->root == fn) { | |
1366 | w->root = w->node = NULL; | |
1367 | RT6_TRACE("W %p adjusted by delroot 1\n", w); | |
1368 | } else if (w->node == fn) { | |
1369 | RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate); | |
1370 | w->node = pn; | |
1371 | w->state = nstate; | |
1372 | } | |
1373 | } else { | |
1374 | if (w->root == fn) { | |
1375 | w->root = child; | |
1376 | RT6_TRACE("W %p adjusted by delroot 2\n", w); | |
1377 | } | |
1378 | if (w->node == fn) { | |
1379 | w->node = child; | |
1380 | if (children&2) { | |
1381 | RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); | |
1382 | w->state = w->state >= FWS_R ? FWS_U : FWS_INIT; | |
1383 | } else { | |
1384 | RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); | |
1385 | w->state = w->state >= FWS_C ? FWS_U : FWS_INIT; | |
1386 | } | |
1387 | } | |
1388 | } | |
1389 | } | |
1390 | read_unlock(&net->ipv6.fib6_walker_lock); | |
1391 | ||
1392 | node_free(fn); | |
1393 | if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn)) | |
1394 | return pn; | |
1395 | ||
1396 | rt6_release(pn->leaf); | |
1397 | pn->leaf = NULL; | |
1398 | fn = pn; | |
1399 | } | |
1400 | } | |
1401 | ||
1402 | static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp, | |
1403 | struct nl_info *info) | |
1404 | { | |
1405 | struct fib6_walker *w; | |
1406 | struct rt6_info *rt = *rtp; | |
1407 | struct net *net = info->nl_net; | |
1408 | ||
1409 | RT6_TRACE("fib6_del_route\n"); | |
1410 | ||
1411 | /* Unlink it */ | |
1412 | *rtp = rt->dst.rt6_next; | |
1413 | rt->rt6i_node = NULL; | |
1414 | net->ipv6.rt6_stats->fib_rt_entries--; | |
1415 | net->ipv6.rt6_stats->fib_discarded_routes++; | |
1416 | ||
1417 | /* Reset round-robin state, if necessary */ | |
1418 | if (fn->rr_ptr == rt) | |
1419 | fn->rr_ptr = NULL; | |
1420 | ||
1421 | /* Remove this entry from other siblings */ | |
1422 | if (rt->rt6i_nsiblings) { | |
1423 | struct rt6_info *sibling, *next_sibling; | |
1424 | ||
1425 | list_for_each_entry_safe(sibling, next_sibling, | |
1426 | &rt->rt6i_siblings, rt6i_siblings) | |
1427 | sibling->rt6i_nsiblings--; | |
1428 | rt->rt6i_nsiblings = 0; | |
1429 | list_del_init(&rt->rt6i_siblings); | |
1430 | } | |
1431 | ||
1432 | /* Adjust walkers */ | |
1433 | read_lock(&net->ipv6.fib6_walker_lock); | |
1434 | FOR_WALKERS(net, w) { | |
1435 | if (w->state == FWS_C && w->leaf == rt) { | |
1436 | RT6_TRACE("walker %p adjusted by delroute\n", w); | |
1437 | w->leaf = rt->dst.rt6_next; | |
1438 | if (!w->leaf) | |
1439 | w->state = FWS_U; | |
1440 | } | |
1441 | } | |
1442 | read_unlock(&net->ipv6.fib6_walker_lock); | |
1443 | ||
1444 | rt->dst.rt6_next = NULL; | |
1445 | ||
1446 | /* If it was last route, expunge its radix tree node */ | |
1447 | if (!fn->leaf) { | |
1448 | fn->fn_flags &= ~RTN_RTINFO; | |
1449 | net->ipv6.rt6_stats->fib_route_nodes--; | |
1450 | fn = fib6_repair_tree(net, fn); | |
1451 | } | |
1452 | ||
1453 | fib6_purge_rt(rt, fn, net); | |
1454 | ||
1455 | inet6_rt_notify(RTM_DELROUTE, rt, info, 0); | |
1456 | rt6_release(rt); | |
1457 | } | |
1458 | ||
1459 | int fib6_del(struct rt6_info *rt, struct nl_info *info) | |
1460 | { | |
1461 | struct net *net = info->nl_net; | |
1462 | struct fib6_node *fn = rt->rt6i_node; | |
1463 | struct rt6_info **rtp; | |
1464 | ||
1465 | #if RT6_DEBUG >= 2 | |
1466 | if (rt->dst.obsolete > 0) { | |
1467 | WARN_ON(fn); | |
1468 | return -ENOENT; | |
1469 | } | |
1470 | #endif | |
1471 | if (!fn || rt == net->ipv6.ip6_null_entry) | |
1472 | return -ENOENT; | |
1473 | ||
1474 | WARN_ON(!(fn->fn_flags & RTN_RTINFO)); | |
1475 | ||
1476 | if (!(rt->rt6i_flags & RTF_CACHE)) { | |
1477 | struct fib6_node *pn = fn; | |
1478 | #ifdef CONFIG_IPV6_SUBTREES | |
1479 | /* clones of this route might be in another subtree */ | |
1480 | if (rt->rt6i_src.plen) { | |
1481 | while (!(pn->fn_flags & RTN_ROOT)) | |
1482 | pn = pn->parent; | |
1483 | pn = pn->parent; | |
1484 | } | |
1485 | #endif | |
1486 | fib6_prune_clones(info->nl_net, pn); | |
1487 | } | |
1488 | ||
1489 | /* | |
1490 | * Walk the leaf entries looking for ourself | |
1491 | */ | |
1492 | ||
1493 | for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) { | |
1494 | if (*rtp == rt) { | |
1495 | fib6_del_route(fn, rtp, info); | |
1496 | return 0; | |
1497 | } | |
1498 | } | |
1499 | return -ENOENT; | |
1500 | } | |
1501 | ||
1502 | /* | |
1503 | * Tree traversal function. | |
1504 | * | |
1505 | * Certainly, it is not interrupt safe. | |
1506 | * However, it is internally reenterable wrt itself and fib6_add/fib6_del. | |
1507 | * It means, that we can modify tree during walking | |
1508 | * and use this function for garbage collection, clone pruning, | |
1509 | * cleaning tree when a device goes down etc. etc. | |
1510 | * | |
1511 | * It guarantees that every node will be traversed, | |
1512 | * and that it will be traversed only once. | |
1513 | * | |
1514 | * Callback function w->func may return: | |
1515 | * 0 -> continue walking. | |
1516 | * positive value -> walking is suspended (used by tree dumps, | |
1517 | * and probably by gc, if it will be split to several slices) | |
1518 | * negative value -> terminate walking. | |
1519 | * | |
1520 | * The function itself returns: | |
1521 | * 0 -> walk is complete. | |
1522 | * >0 -> walk is incomplete (i.e. suspended) | |
1523 | * <0 -> walk is terminated by an error. | |
1524 | */ | |
1525 | ||
1526 | static int fib6_walk_continue(struct fib6_walker *w) | |
1527 | { | |
1528 | struct fib6_node *fn, *pn; | |
1529 | ||
1530 | for (;;) { | |
1531 | fn = w->node; | |
1532 | if (!fn) | |
1533 | return 0; | |
1534 | ||
1535 | if (w->prune && fn != w->root && | |
1536 | fn->fn_flags & RTN_RTINFO && w->state < FWS_C) { | |
1537 | w->state = FWS_C; | |
1538 | w->leaf = fn->leaf; | |
1539 | } | |
1540 | switch (w->state) { | |
1541 | #ifdef CONFIG_IPV6_SUBTREES | |
1542 | case FWS_S: | |
1543 | if (FIB6_SUBTREE(fn)) { | |
1544 | w->node = FIB6_SUBTREE(fn); | |
1545 | continue; | |
1546 | } | |
1547 | w->state = FWS_L; | |
1548 | #endif | |
1549 | case FWS_L: | |
1550 | if (fn->left) { | |
1551 | w->node = fn->left; | |
1552 | w->state = FWS_INIT; | |
1553 | continue; | |
1554 | } | |
1555 | w->state = FWS_R; | |
1556 | case FWS_R: | |
1557 | if (fn->right) { | |
1558 | w->node = fn->right; | |
1559 | w->state = FWS_INIT; | |
1560 | continue; | |
1561 | } | |
1562 | w->state = FWS_C; | |
1563 | w->leaf = fn->leaf; | |
1564 | case FWS_C: | |
1565 | if (w->leaf && fn->fn_flags & RTN_RTINFO) { | |
1566 | int err; | |
1567 | ||
1568 | if (w->skip) { | |
1569 | w->skip--; | |
1570 | goto skip; | |
1571 | } | |
1572 | ||
1573 | err = w->func(w); | |
1574 | if (err) | |
1575 | return err; | |
1576 | ||
1577 | w->count++; | |
1578 | continue; | |
1579 | } | |
1580 | skip: | |
1581 | w->state = FWS_U; | |
1582 | case FWS_U: | |
1583 | if (fn == w->root) | |
1584 | return 0; | |
1585 | pn = fn->parent; | |
1586 | w->node = pn; | |
1587 | #ifdef CONFIG_IPV6_SUBTREES | |
1588 | if (FIB6_SUBTREE(pn) == fn) { | |
1589 | WARN_ON(!(fn->fn_flags & RTN_ROOT)); | |
1590 | w->state = FWS_L; | |
1591 | continue; | |
1592 | } | |
1593 | #endif | |
1594 | if (pn->left == fn) { | |
1595 | w->state = FWS_R; | |
1596 | continue; | |
1597 | } | |
1598 | if (pn->right == fn) { | |
1599 | w->state = FWS_C; | |
1600 | w->leaf = w->node->leaf; | |
1601 | continue; | |
1602 | } | |
1603 | #if RT6_DEBUG >= 2 | |
1604 | WARN_ON(1); | |
1605 | #endif | |
1606 | } | |
1607 | } | |
1608 | } | |
1609 | ||
1610 | static int fib6_walk(struct net *net, struct fib6_walker *w) | |
1611 | { | |
1612 | int res; | |
1613 | ||
1614 | w->state = FWS_INIT; | |
1615 | w->node = w->root; | |
1616 | ||
1617 | fib6_walker_link(net, w); | |
1618 | res = fib6_walk_continue(w); | |
1619 | if (res <= 0) | |
1620 | fib6_walker_unlink(net, w); | |
1621 | return res; | |
1622 | } | |
1623 | ||
1624 | static int fib6_clean_node(struct fib6_walker *w) | |
1625 | { | |
1626 | int res; | |
1627 | struct rt6_info *rt; | |
1628 | struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w); | |
1629 | struct nl_info info = { | |
1630 | .nl_net = c->net, | |
1631 | }; | |
1632 | ||
1633 | if (c->sernum != FIB6_NO_SERNUM_CHANGE && | |
1634 | w->node->fn_sernum != c->sernum) | |
1635 | w->node->fn_sernum = c->sernum; | |
1636 | ||
1637 | if (!c->func) { | |
1638 | WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE); | |
1639 | w->leaf = NULL; | |
1640 | return 0; | |
1641 | } | |
1642 | ||
1643 | for (rt = w->leaf; rt; rt = rt->dst.rt6_next) { | |
1644 | res = c->func(rt, c->arg); | |
1645 | if (res < 0) { | |
1646 | w->leaf = rt; | |
1647 | res = fib6_del(rt, &info); | |
1648 | if (res) { | |
1649 | #if RT6_DEBUG >= 2 | |
1650 | pr_debug("%s: del failed: rt=%p@%p err=%d\n", | |
1651 | __func__, rt, rt->rt6i_node, res); | |
1652 | #endif | |
1653 | continue; | |
1654 | } | |
1655 | return 0; | |
1656 | } | |
1657 | WARN_ON(res != 0); | |
1658 | } | |
1659 | w->leaf = rt; | |
1660 | return 0; | |
1661 | } | |
1662 | ||
1663 | /* | |
1664 | * Convenient frontend to tree walker. | |
1665 | * | |
1666 | * func is called on each route. | |
1667 | * It may return -1 -> delete this route. | |
1668 | * 0 -> continue walking | |
1669 | * | |
1670 | * prune==1 -> only immediate children of node (certainly, | |
1671 | * ignoring pure split nodes) will be scanned. | |
1672 | */ | |
1673 | ||
1674 | static void fib6_clean_tree(struct net *net, struct fib6_node *root, | |
1675 | int (*func)(struct rt6_info *, void *arg), | |
1676 | bool prune, int sernum, void *arg) | |
1677 | { | |
1678 | struct fib6_cleaner c; | |
1679 | ||
1680 | c.w.root = root; | |
1681 | c.w.func = fib6_clean_node; | |
1682 | c.w.prune = prune; | |
1683 | c.w.count = 0; | |
1684 | c.w.skip = 0; | |
1685 | c.func = func; | |
1686 | c.sernum = sernum; | |
1687 | c.arg = arg; | |
1688 | c.net = net; | |
1689 | ||
1690 | fib6_walk(net, &c.w); | |
1691 | } | |
1692 | ||
1693 | static void __fib6_clean_all(struct net *net, | |
1694 | int (*func)(struct rt6_info *, void *), | |
1695 | int sernum, void *arg) | |
1696 | { | |
1697 | struct fib6_table *table; | |
1698 | struct hlist_head *head; | |
1699 | unsigned int h; | |
1700 | ||
1701 | rcu_read_lock(); | |
1702 | for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { | |
1703 | head = &net->ipv6.fib_table_hash[h]; | |
1704 | hlist_for_each_entry_rcu(table, head, tb6_hlist) { | |
1705 | write_lock_bh(&table->tb6_lock); | |
1706 | fib6_clean_tree(net, &table->tb6_root, | |
1707 | func, false, sernum, arg); | |
1708 | write_unlock_bh(&table->tb6_lock); | |
1709 | } | |
1710 | } | |
1711 | rcu_read_unlock(); | |
1712 | } | |
1713 | ||
1714 | void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *), | |
1715 | void *arg) | |
1716 | { | |
1717 | __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg); | |
1718 | } | |
1719 | ||
1720 | static int fib6_prune_clone(struct rt6_info *rt, void *arg) | |
1721 | { | |
1722 | if (rt->rt6i_flags & RTF_CACHE) { | |
1723 | RT6_TRACE("pruning clone %p\n", rt); | |
1724 | return -1; | |
1725 | } | |
1726 | ||
1727 | return 0; | |
1728 | } | |
1729 | ||
1730 | static void fib6_prune_clones(struct net *net, struct fib6_node *fn) | |
1731 | { | |
1732 | fib6_clean_tree(net, fn, fib6_prune_clone, true, | |
1733 | FIB6_NO_SERNUM_CHANGE, NULL); | |
1734 | } | |
1735 | ||
1736 | static void fib6_flush_trees(struct net *net) | |
1737 | { | |
1738 | int new_sernum = fib6_new_sernum(net); | |
1739 | ||
1740 | __fib6_clean_all(net, NULL, new_sernum, NULL); | |
1741 | } | |
1742 | ||
1743 | /* | |
1744 | * Garbage collection | |
1745 | */ | |
1746 | ||
1747 | struct fib6_gc_args | |
1748 | { | |
1749 | int timeout; | |
1750 | int more; | |
1751 | }; | |
1752 | ||
1753 | static int fib6_age(struct rt6_info *rt, void *arg) | |
1754 | { | |
1755 | struct fib6_gc_args *gc_args = arg; | |
1756 | unsigned long now = jiffies; | |
1757 | ||
1758 | /* | |
1759 | * check addrconf expiration here. | |
1760 | * Routes are expired even if they are in use. | |
1761 | * | |
1762 | * Also age clones. Note, that clones are aged out | |
1763 | * only if they are not in use now. | |
1764 | */ | |
1765 | ||
1766 | if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) { | |
1767 | if (time_after(now, rt->dst.expires)) { | |
1768 | RT6_TRACE("expiring %p\n", rt); | |
1769 | return -1; | |
1770 | } | |
1771 | gc_args->more++; | |
1772 | } else if (rt->rt6i_flags & RTF_CACHE) { | |
1773 | if (atomic_read(&rt->dst.__refcnt) == 0 && | |
1774 | time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) { | |
1775 | RT6_TRACE("aging clone %p\n", rt); | |
1776 | return -1; | |
1777 | } else if (rt->rt6i_flags & RTF_GATEWAY) { | |
1778 | struct neighbour *neigh; | |
1779 | __u8 neigh_flags = 0; | |
1780 | ||
1781 | neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway); | |
1782 | if (neigh) { | |
1783 | neigh_flags = neigh->flags; | |
1784 | neigh_release(neigh); | |
1785 | } | |
1786 | if (!(neigh_flags & NTF_ROUTER)) { | |
1787 | RT6_TRACE("purging route %p via non-router but gateway\n", | |
1788 | rt); | |
1789 | return -1; | |
1790 | } | |
1791 | } | |
1792 | gc_args->more++; | |
1793 | } | |
1794 | ||
1795 | return 0; | |
1796 | } | |
1797 | ||
1798 | void fib6_run_gc(unsigned long expires, struct net *net, bool force) | |
1799 | { | |
1800 | struct fib6_gc_args gc_args; | |
1801 | unsigned long now; | |
1802 | ||
1803 | if (force) { | |
1804 | spin_lock_bh(&net->ipv6.fib6_gc_lock); | |
1805 | } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) { | |
1806 | mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ); | |
1807 | return; | |
1808 | } | |
1809 | gc_args.timeout = expires ? (int)expires : | |
1810 | net->ipv6.sysctl.ip6_rt_gc_interval; | |
1811 | ||
1812 | gc_args.more = icmp6_dst_gc(); | |
1813 | ||
1814 | fib6_clean_all(net, fib6_age, &gc_args); | |
1815 | now = jiffies; | |
1816 | net->ipv6.ip6_rt_last_gc = now; | |
1817 | ||
1818 | if (gc_args.more) | |
1819 | mod_timer(&net->ipv6.ip6_fib_timer, | |
1820 | round_jiffies(now | |
1821 | + net->ipv6.sysctl.ip6_rt_gc_interval)); | |
1822 | else | |
1823 | del_timer(&net->ipv6.ip6_fib_timer); | |
1824 | spin_unlock_bh(&net->ipv6.fib6_gc_lock); | |
1825 | } | |
1826 | ||
1827 | static void fib6_gc_timer_cb(unsigned long arg) | |
1828 | { | |
1829 | fib6_run_gc(0, (struct net *)arg, true); | |
1830 | } | |
1831 | ||
1832 | static int __net_init fib6_net_init(struct net *net) | |
1833 | { | |
1834 | size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ; | |
1835 | ||
1836 | spin_lock_init(&net->ipv6.fib6_gc_lock); | |
1837 | rwlock_init(&net->ipv6.fib6_walker_lock); | |
1838 | INIT_LIST_HEAD(&net->ipv6.fib6_walkers); | |
1839 | setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net); | |
1840 | ||
1841 | net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL); | |
1842 | if (!net->ipv6.rt6_stats) | |
1843 | goto out_timer; | |
1844 | ||
1845 | /* Avoid false sharing : Use at least a full cache line */ | |
1846 | size = max_t(size_t, size, L1_CACHE_BYTES); | |
1847 | ||
1848 | net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL); | |
1849 | if (!net->ipv6.fib_table_hash) | |
1850 | goto out_rt6_stats; | |
1851 | ||
1852 | net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl), | |
1853 | GFP_KERNEL); | |
1854 | if (!net->ipv6.fib6_main_tbl) | |
1855 | goto out_fib_table_hash; | |
1856 | ||
1857 | net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN; | |
1858 | net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; | |
1859 | net->ipv6.fib6_main_tbl->tb6_root.fn_flags = | |
1860 | RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; | |
1861 | inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers); | |
1862 | ||
1863 | #ifdef CONFIG_IPV6_MULTIPLE_TABLES | |
1864 | net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl), | |
1865 | GFP_KERNEL); | |
1866 | if (!net->ipv6.fib6_local_tbl) | |
1867 | goto out_fib6_main_tbl; | |
1868 | net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL; | |
1869 | net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry; | |
1870 | net->ipv6.fib6_local_tbl->tb6_root.fn_flags = | |
1871 | RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; | |
1872 | inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers); | |
1873 | #endif | |
1874 | fib6_tables_init(net); | |
1875 | ||
1876 | return 0; | |
1877 | ||
1878 | #ifdef CONFIG_IPV6_MULTIPLE_TABLES | |
1879 | out_fib6_main_tbl: | |
1880 | kfree(net->ipv6.fib6_main_tbl); | |
1881 | #endif | |
1882 | out_fib_table_hash: | |
1883 | kfree(net->ipv6.fib_table_hash); | |
1884 | out_rt6_stats: | |
1885 | kfree(net->ipv6.rt6_stats); | |
1886 | out_timer: | |
1887 | return -ENOMEM; | |
1888 | } | |
1889 | ||
1890 | static void fib6_net_exit(struct net *net) | |
1891 | { | |
1892 | rt6_ifdown(net, NULL); | |
1893 | del_timer_sync(&net->ipv6.ip6_fib_timer); | |
1894 | ||
1895 | #ifdef CONFIG_IPV6_MULTIPLE_TABLES | |
1896 | inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers); | |
1897 | kfree(net->ipv6.fib6_local_tbl); | |
1898 | #endif | |
1899 | inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers); | |
1900 | kfree(net->ipv6.fib6_main_tbl); | |
1901 | kfree(net->ipv6.fib_table_hash); | |
1902 | kfree(net->ipv6.rt6_stats); | |
1903 | } | |
1904 | ||
1905 | static struct pernet_operations fib6_net_ops = { | |
1906 | .init = fib6_net_init, | |
1907 | .exit = fib6_net_exit, | |
1908 | }; | |
1909 | ||
1910 | int __init fib6_init(void) | |
1911 | { | |
1912 | int ret = -ENOMEM; | |
1913 | ||
1914 | fib6_node_kmem = kmem_cache_create("fib6_nodes", | |
1915 | sizeof(struct fib6_node), | |
1916 | 0, SLAB_HWCACHE_ALIGN, | |
1917 | NULL); | |
1918 | if (!fib6_node_kmem) | |
1919 | goto out; | |
1920 | ||
1921 | ret = register_pernet_subsys(&fib6_net_ops); | |
1922 | if (ret) | |
1923 | goto out_kmem_cache_create; | |
1924 | ||
1925 | ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib, | |
1926 | NULL); | |
1927 | if (ret) | |
1928 | goto out_unregister_subsys; | |
1929 | ||
1930 | __fib6_flush_trees = fib6_flush_trees; | |
1931 | out: | |
1932 | return ret; | |
1933 | ||
1934 | out_unregister_subsys: | |
1935 | unregister_pernet_subsys(&fib6_net_ops); | |
1936 | out_kmem_cache_create: | |
1937 | kmem_cache_destroy(fib6_node_kmem); | |
1938 | goto out; | |
1939 | } | |
1940 | ||
1941 | void fib6_gc_cleanup(void) | |
1942 | { | |
1943 | unregister_pernet_subsys(&fib6_net_ops); | |
1944 | kmem_cache_destroy(fib6_node_kmem); | |
1945 | } | |
1946 | ||
1947 | #ifdef CONFIG_PROC_FS | |
1948 | ||
1949 | struct ipv6_route_iter { | |
1950 | struct seq_net_private p; | |
1951 | struct fib6_walker w; | |
1952 | loff_t skip; | |
1953 | struct fib6_table *tbl; | |
1954 | int sernum; | |
1955 | }; | |
1956 | ||
1957 | static int ipv6_route_seq_show(struct seq_file *seq, void *v) | |
1958 | { | |
1959 | struct rt6_info *rt = v; | |
1960 | struct ipv6_route_iter *iter = seq->private; | |
1961 | ||
1962 | seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen); | |
1963 | ||
1964 | #ifdef CONFIG_IPV6_SUBTREES | |
1965 | seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen); | |
1966 | #else | |
1967 | seq_puts(seq, "00000000000000000000000000000000 00 "); | |
1968 | #endif | |
1969 | if (rt->rt6i_flags & RTF_GATEWAY) | |
1970 | seq_printf(seq, "%pi6", &rt->rt6i_gateway); | |
1971 | else | |
1972 | seq_puts(seq, "00000000000000000000000000000000"); | |
1973 | ||
1974 | seq_printf(seq, " %08x %08x %08x %08x %8s\n", | |
1975 | rt->rt6i_metric, atomic_read(&rt->dst.__refcnt), | |
1976 | rt->dst.__use, rt->rt6i_flags, | |
1977 | rt->dst.dev ? rt->dst.dev->name : ""); | |
1978 | iter->w.leaf = NULL; | |
1979 | return 0; | |
1980 | } | |
1981 | ||
1982 | static int ipv6_route_yield(struct fib6_walker *w) | |
1983 | { | |
1984 | struct ipv6_route_iter *iter = w->args; | |
1985 | ||
1986 | if (!iter->skip) | |
1987 | return 1; | |
1988 | ||
1989 | do { | |
1990 | iter->w.leaf = iter->w.leaf->dst.rt6_next; | |
1991 | iter->skip--; | |
1992 | if (!iter->skip && iter->w.leaf) | |
1993 | return 1; | |
1994 | } while (iter->w.leaf); | |
1995 | ||
1996 | return 0; | |
1997 | } | |
1998 | ||
1999 | static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter, | |
2000 | struct net *net) | |
2001 | { | |
2002 | memset(&iter->w, 0, sizeof(iter->w)); | |
2003 | iter->w.func = ipv6_route_yield; | |
2004 | iter->w.root = &iter->tbl->tb6_root; | |
2005 | iter->w.state = FWS_INIT; | |
2006 | iter->w.node = iter->w.root; | |
2007 | iter->w.args = iter; | |
2008 | iter->sernum = iter->w.root->fn_sernum; | |
2009 | INIT_LIST_HEAD(&iter->w.lh); | |
2010 | fib6_walker_link(net, &iter->w); | |
2011 | } | |
2012 | ||
2013 | static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl, | |
2014 | struct net *net) | |
2015 | { | |
2016 | unsigned int h; | |
2017 | struct hlist_node *node; | |
2018 | ||
2019 | if (tbl) { | |
2020 | h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1; | |
2021 | node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist)); | |
2022 | } else { | |
2023 | h = 0; | |
2024 | node = NULL; | |
2025 | } | |
2026 | ||
2027 | while (!node && h < FIB6_TABLE_HASHSZ) { | |
2028 | node = rcu_dereference_bh( | |
2029 | hlist_first_rcu(&net->ipv6.fib_table_hash[h++])); | |
2030 | } | |
2031 | return hlist_entry_safe(node, struct fib6_table, tb6_hlist); | |
2032 | } | |
2033 | ||
2034 | static void ipv6_route_check_sernum(struct ipv6_route_iter *iter) | |
2035 | { | |
2036 | if (iter->sernum != iter->w.root->fn_sernum) { | |
2037 | iter->sernum = iter->w.root->fn_sernum; | |
2038 | iter->w.state = FWS_INIT; | |
2039 | iter->w.node = iter->w.root; | |
2040 | WARN_ON(iter->w.skip); | |
2041 | iter->w.skip = iter->w.count; | |
2042 | } | |
2043 | } | |
2044 | ||
2045 | static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2046 | { | |
2047 | int r; | |
2048 | struct rt6_info *n; | |
2049 | struct net *net = seq_file_net(seq); | |
2050 | struct ipv6_route_iter *iter = seq->private; | |
2051 | ||
2052 | if (!v) | |
2053 | goto iter_table; | |
2054 | ||
2055 | n = ((struct rt6_info *)v)->dst.rt6_next; | |
2056 | if (n) { | |
2057 | ++*pos; | |
2058 | return n; | |
2059 | } | |
2060 | ||
2061 | iter_table: | |
2062 | ipv6_route_check_sernum(iter); | |
2063 | read_lock(&iter->tbl->tb6_lock); | |
2064 | r = fib6_walk_continue(&iter->w); | |
2065 | read_unlock(&iter->tbl->tb6_lock); | |
2066 | if (r > 0) { | |
2067 | if (v) | |
2068 | ++*pos; | |
2069 | return iter->w.leaf; | |
2070 | } else if (r < 0) { | |
2071 | fib6_walker_unlink(net, &iter->w); | |
2072 | return NULL; | |
2073 | } | |
2074 | fib6_walker_unlink(net, &iter->w); | |
2075 | ||
2076 | iter->tbl = ipv6_route_seq_next_table(iter->tbl, net); | |
2077 | if (!iter->tbl) | |
2078 | return NULL; | |
2079 | ||
2080 | ipv6_route_seq_setup_walk(iter, net); | |
2081 | goto iter_table; | |
2082 | } | |
2083 | ||
2084 | static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos) | |
2085 | __acquires(RCU_BH) | |
2086 | { | |
2087 | struct net *net = seq_file_net(seq); | |
2088 | struct ipv6_route_iter *iter = seq->private; | |
2089 | ||
2090 | rcu_read_lock_bh(); | |
2091 | iter->tbl = ipv6_route_seq_next_table(NULL, net); | |
2092 | iter->skip = *pos; | |
2093 | ||
2094 | if (iter->tbl) { | |
2095 | ipv6_route_seq_setup_walk(iter, net); | |
2096 | return ipv6_route_seq_next(seq, NULL, pos); | |
2097 | } else { | |
2098 | return NULL; | |
2099 | } | |
2100 | } | |
2101 | ||
2102 | static bool ipv6_route_iter_active(struct ipv6_route_iter *iter) | |
2103 | { | |
2104 | struct fib6_walker *w = &iter->w; | |
2105 | return w->node && !(w->state == FWS_U && w->node == w->root); | |
2106 | } | |
2107 | ||
2108 | static void ipv6_route_seq_stop(struct seq_file *seq, void *v) | |
2109 | __releases(RCU_BH) | |
2110 | { | |
2111 | struct net *net = seq_file_net(seq); | |
2112 | struct ipv6_route_iter *iter = seq->private; | |
2113 | ||
2114 | if (ipv6_route_iter_active(iter)) | |
2115 | fib6_walker_unlink(net, &iter->w); | |
2116 | ||
2117 | rcu_read_unlock_bh(); | |
2118 | } | |
2119 | ||
2120 | static const struct seq_operations ipv6_route_seq_ops = { | |
2121 | .start = ipv6_route_seq_start, | |
2122 | .next = ipv6_route_seq_next, | |
2123 | .stop = ipv6_route_seq_stop, | |
2124 | .show = ipv6_route_seq_show | |
2125 | }; | |
2126 | ||
2127 | int ipv6_route_open(struct inode *inode, struct file *file) | |
2128 | { | |
2129 | return seq_open_net(inode, file, &ipv6_route_seq_ops, | |
2130 | sizeof(struct ipv6_route_iter)); | |
2131 | } | |
2132 | ||
2133 | #endif /* CONFIG_PROC_FS */ |