Merge branches 'x86/amd', 'x86/vt-d', 'arm/exynos', 'arm/mediatek' and 'arm/renesas...
[deliverable/linux.git] / net / ipv6 / ip6_fib.c
1 /*
2 * Linux INET6 implementation
3 * Forwarding Information Database
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 *
13 * 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 w->leaf = NULL;
323 return 0;
324 }
325
326 static void fib6_dump_end(struct netlink_callback *cb)
327 {
328 struct net *net = sock_net(cb->skb->sk);
329 struct fib6_walker *w = (void *)cb->args[2];
330
331 if (w) {
332 if (cb->args[4]) {
333 cb->args[4] = 0;
334 fib6_walker_unlink(net, w);
335 }
336 cb->args[2] = 0;
337 kfree(w);
338 }
339 cb->done = (void *)cb->args[3];
340 cb->args[1] = 3;
341 }
342
343 static int fib6_dump_done(struct netlink_callback *cb)
344 {
345 fib6_dump_end(cb);
346 return cb->done ? cb->done(cb) : 0;
347 }
348
349 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
350 struct netlink_callback *cb)
351 {
352 struct net *net = sock_net(skb->sk);
353 struct fib6_walker *w;
354 int res;
355
356 w = (void *)cb->args[2];
357 w->root = &table->tb6_root;
358
359 if (cb->args[4] == 0) {
360 w->count = 0;
361 w->skip = 0;
362
363 read_lock_bh(&table->tb6_lock);
364 res = fib6_walk(net, w);
365 read_unlock_bh(&table->tb6_lock);
366 if (res > 0) {
367 cb->args[4] = 1;
368 cb->args[5] = w->root->fn_sernum;
369 }
370 } else {
371 if (cb->args[5] != w->root->fn_sernum) {
372 /* Begin at the root if the tree changed */
373 cb->args[5] = w->root->fn_sernum;
374 w->state = FWS_INIT;
375 w->node = w->root;
376 w->skip = w->count;
377 } else
378 w->skip = 0;
379
380 read_lock_bh(&table->tb6_lock);
381 res = fib6_walk_continue(w);
382 read_unlock_bh(&table->tb6_lock);
383 if (res <= 0) {
384 fib6_walker_unlink(net, w);
385 cb->args[4] = 0;
386 }
387 }
388
389 return res;
390 }
391
392 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
393 {
394 struct net *net = sock_net(skb->sk);
395 unsigned int h, s_h;
396 unsigned int e = 0, s_e;
397 struct rt6_rtnl_dump_arg arg;
398 struct fib6_walker *w;
399 struct fib6_table *tb;
400 struct hlist_head *head;
401 int res = 0;
402
403 s_h = cb->args[0];
404 s_e = cb->args[1];
405
406 w = (void *)cb->args[2];
407 if (!w) {
408 /* New dump:
409 *
410 * 1. hook callback destructor.
411 */
412 cb->args[3] = (long)cb->done;
413 cb->done = fib6_dump_done;
414
415 /*
416 * 2. allocate and initialize walker.
417 */
418 w = kzalloc(sizeof(*w), GFP_ATOMIC);
419 if (!w)
420 return -ENOMEM;
421 w->func = fib6_dump_node;
422 cb->args[2] = (long)w;
423 }
424
425 arg.skb = skb;
426 arg.cb = cb;
427 arg.net = net;
428 w->args = &arg;
429
430 rcu_read_lock();
431 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
432 e = 0;
433 head = &net->ipv6.fib_table_hash[h];
434 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
435 if (e < s_e)
436 goto next;
437 res = fib6_dump_table(tb, skb, cb);
438 if (res != 0)
439 goto out;
440 next:
441 e++;
442 }
443 }
444 out:
445 rcu_read_unlock();
446 cb->args[1] = e;
447 cb->args[0] = h;
448
449 res = res < 0 ? res : skb->len;
450 if (res <= 0)
451 fib6_dump_end(cb);
452 return res;
453 }
454
455 /*
456 * Routing Table
457 *
458 * return the appropriate node for a routing tree "add" operation
459 * by either creating and inserting or by returning an existing
460 * node.
461 */
462
463 static struct fib6_node *fib6_add_1(struct fib6_node *root,
464 struct in6_addr *addr, int plen,
465 int offset, int allow_create,
466 int replace_required, int sernum)
467 {
468 struct fib6_node *fn, *in, *ln;
469 struct fib6_node *pn = NULL;
470 struct rt6key *key;
471 int bit;
472 __be32 dir = 0;
473
474 RT6_TRACE("fib6_add_1\n");
475
476 /* insert node in tree */
477
478 fn = root;
479
480 do {
481 key = (struct rt6key *)((u8 *)fn->leaf + offset);
482
483 /*
484 * Prefix match
485 */
486 if (plen < fn->fn_bit ||
487 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
488 if (!allow_create) {
489 if (replace_required) {
490 pr_warn("Can't replace route, no match found\n");
491 return ERR_PTR(-ENOENT);
492 }
493 pr_warn("NLM_F_CREATE should be set when creating new route\n");
494 }
495 goto insert_above;
496 }
497
498 /*
499 * Exact match ?
500 */
501
502 if (plen == fn->fn_bit) {
503 /* clean up an intermediate node */
504 if (!(fn->fn_flags & RTN_RTINFO)) {
505 rt6_release(fn->leaf);
506 fn->leaf = NULL;
507 }
508
509 fn->fn_sernum = sernum;
510
511 return fn;
512 }
513
514 /*
515 * We have more bits to go
516 */
517
518 /* Try to walk down on tree. */
519 fn->fn_sernum = sernum;
520 dir = addr_bit_set(addr, fn->fn_bit);
521 pn = fn;
522 fn = dir ? fn->right : fn->left;
523 } while (fn);
524
525 if (!allow_create) {
526 /* We should not create new node because
527 * NLM_F_REPLACE was specified without NLM_F_CREATE
528 * I assume it is safe to require NLM_F_CREATE when
529 * REPLACE flag is used! Later we may want to remove the
530 * check for replace_required, because according
531 * to netlink specification, NLM_F_CREATE
532 * MUST be specified if new route is created.
533 * That would keep IPv6 consistent with IPv4
534 */
535 if (replace_required) {
536 pr_warn("Can't replace route, no match found\n");
537 return ERR_PTR(-ENOENT);
538 }
539 pr_warn("NLM_F_CREATE should be set when creating new route\n");
540 }
541 /*
542 * We walked to the bottom of tree.
543 * Create new leaf node without children.
544 */
545
546 ln = node_alloc();
547
548 if (!ln)
549 return ERR_PTR(-ENOMEM);
550 ln->fn_bit = plen;
551
552 ln->parent = pn;
553 ln->fn_sernum = sernum;
554
555 if (dir)
556 pn->right = ln;
557 else
558 pn->left = ln;
559
560 return ln;
561
562
563 insert_above:
564 /*
565 * split since we don't have a common prefix anymore or
566 * we have a less significant route.
567 * we've to insert an intermediate node on the list
568 * this new node will point to the one we need to create
569 * and the current
570 */
571
572 pn = fn->parent;
573
574 /* find 1st bit in difference between the 2 addrs.
575
576 See comment in __ipv6_addr_diff: bit may be an invalid value,
577 but if it is >= plen, the value is ignored in any case.
578 */
579
580 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
581
582 /*
583 * (intermediate)[in]
584 * / \
585 * (new leaf node)[ln] (old node)[fn]
586 */
587 if (plen > bit) {
588 in = node_alloc();
589 ln = node_alloc();
590
591 if (!in || !ln) {
592 if (in)
593 node_free(in);
594 if (ln)
595 node_free(ln);
596 return ERR_PTR(-ENOMEM);
597 }
598
599 /*
600 * new intermediate node.
601 * RTN_RTINFO will
602 * be off since that an address that chooses one of
603 * the branches would not match less specific routes
604 * in the other branch
605 */
606
607 in->fn_bit = bit;
608
609 in->parent = pn;
610 in->leaf = fn->leaf;
611 atomic_inc(&in->leaf->rt6i_ref);
612
613 in->fn_sernum = sernum;
614
615 /* update parent pointer */
616 if (dir)
617 pn->right = in;
618 else
619 pn->left = in;
620
621 ln->fn_bit = plen;
622
623 ln->parent = in;
624 fn->parent = in;
625
626 ln->fn_sernum = sernum;
627
628 if (addr_bit_set(addr, bit)) {
629 in->right = ln;
630 in->left = fn;
631 } else {
632 in->left = ln;
633 in->right = fn;
634 }
635 } else { /* plen <= bit */
636
637 /*
638 * (new leaf node)[ln]
639 * / \
640 * (old node)[fn] NULL
641 */
642
643 ln = node_alloc();
644
645 if (!ln)
646 return ERR_PTR(-ENOMEM);
647
648 ln->fn_bit = plen;
649
650 ln->parent = pn;
651
652 ln->fn_sernum = sernum;
653
654 if (dir)
655 pn->right = ln;
656 else
657 pn->left = ln;
658
659 if (addr_bit_set(&key->addr, plen))
660 ln->right = fn;
661 else
662 ln->left = fn;
663
664 fn->parent = ln;
665 }
666 return ln;
667 }
668
669 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
670 {
671 return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
672 RTF_GATEWAY;
673 }
674
675 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
676 {
677 int i;
678
679 for (i = 0; i < RTAX_MAX; i++) {
680 if (test_bit(i, mxc->mx_valid))
681 mp[i] = mxc->mx[i];
682 }
683 }
684
685 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
686 {
687 if (!mxc->mx)
688 return 0;
689
690 if (dst->flags & DST_HOST) {
691 u32 *mp = dst_metrics_write_ptr(dst);
692
693 if (unlikely(!mp))
694 return -ENOMEM;
695
696 fib6_copy_metrics(mp, mxc);
697 } else {
698 dst_init_metrics(dst, mxc->mx, false);
699
700 /* We've stolen mx now. */
701 mxc->mx = NULL;
702 }
703
704 return 0;
705 }
706
707 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
708 struct net *net)
709 {
710 if (atomic_read(&rt->rt6i_ref) != 1) {
711 /* This route is used as dummy address holder in some split
712 * nodes. It is not leaked, but it still holds other resources,
713 * which must be released in time. So, scan ascendant nodes
714 * and replace dummy references to this route with references
715 * to still alive ones.
716 */
717 while (fn) {
718 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
719 fn->leaf = fib6_find_prefix(net, fn);
720 atomic_inc(&fn->leaf->rt6i_ref);
721 rt6_release(rt);
722 }
723 fn = fn->parent;
724 }
725 /* No more references are possible at this point. */
726 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
727 }
728 }
729
730 /*
731 * Insert routing information in a node.
732 */
733
734 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
735 struct nl_info *info, struct mx6_config *mxc)
736 {
737 struct rt6_info *iter = NULL;
738 struct rt6_info **ins;
739 struct rt6_info **fallback_ins = NULL;
740 int replace = (info->nlh &&
741 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
742 int add = (!info->nlh ||
743 (info->nlh->nlmsg_flags & NLM_F_CREATE));
744 int found = 0;
745 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
746 int err;
747
748 ins = &fn->leaf;
749
750 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
751 /*
752 * Search for duplicates
753 */
754
755 if (iter->rt6i_metric == rt->rt6i_metric) {
756 /*
757 * Same priority level
758 */
759 if (info->nlh &&
760 (info->nlh->nlmsg_flags & NLM_F_EXCL))
761 return -EEXIST;
762 if (replace) {
763 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
764 found++;
765 break;
766 }
767 if (rt_can_ecmp)
768 fallback_ins = fallback_ins ?: ins;
769 goto next_iter;
770 }
771
772 if (iter->dst.dev == rt->dst.dev &&
773 iter->rt6i_idev == rt->rt6i_idev &&
774 ipv6_addr_equal(&iter->rt6i_gateway,
775 &rt->rt6i_gateway)) {
776 if (rt->rt6i_nsiblings)
777 rt->rt6i_nsiblings = 0;
778 if (!(iter->rt6i_flags & RTF_EXPIRES))
779 return -EEXIST;
780 if (!(rt->rt6i_flags & RTF_EXPIRES))
781 rt6_clean_expires(iter);
782 else
783 rt6_set_expires(iter, rt->dst.expires);
784 iter->rt6i_pmtu = rt->rt6i_pmtu;
785 return -EEXIST;
786 }
787 /* If we have the same destination and the same metric,
788 * but not the same gateway, then the route we try to
789 * add is sibling to this route, increment our counter
790 * of siblings, and later we will add our route to the
791 * list.
792 * Only static routes (which don't have flag
793 * RTF_EXPIRES) are used for ECMPv6.
794 *
795 * To avoid long list, we only had siblings if the
796 * route have a gateway.
797 */
798 if (rt_can_ecmp &&
799 rt6_qualify_for_ecmp(iter))
800 rt->rt6i_nsiblings++;
801 }
802
803 if (iter->rt6i_metric > rt->rt6i_metric)
804 break;
805
806 next_iter:
807 ins = &iter->dst.rt6_next;
808 }
809
810 if (fallback_ins && !found) {
811 /* No ECMP-able route found, replace first non-ECMP one */
812 ins = fallback_ins;
813 iter = *ins;
814 found++;
815 }
816
817 /* Reset round-robin state, if necessary */
818 if (ins == &fn->leaf)
819 fn->rr_ptr = NULL;
820
821 /* Link this route to others same route. */
822 if (rt->rt6i_nsiblings) {
823 unsigned int rt6i_nsiblings;
824 struct rt6_info *sibling, *temp_sibling;
825
826 /* Find the first route that have the same metric */
827 sibling = fn->leaf;
828 while (sibling) {
829 if (sibling->rt6i_metric == rt->rt6i_metric &&
830 rt6_qualify_for_ecmp(sibling)) {
831 list_add_tail(&rt->rt6i_siblings,
832 &sibling->rt6i_siblings);
833 break;
834 }
835 sibling = sibling->dst.rt6_next;
836 }
837 /* For each sibling in the list, increment the counter of
838 * siblings. BUG() if counters does not match, list of siblings
839 * is broken!
840 */
841 rt6i_nsiblings = 0;
842 list_for_each_entry_safe(sibling, temp_sibling,
843 &rt->rt6i_siblings, rt6i_siblings) {
844 sibling->rt6i_nsiblings++;
845 BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
846 rt6i_nsiblings++;
847 }
848 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
849 }
850
851 /*
852 * insert node
853 */
854 if (!replace) {
855 if (!add)
856 pr_warn("NLM_F_CREATE should be set when creating new route\n");
857
858 add:
859 err = fib6_commit_metrics(&rt->dst, mxc);
860 if (err)
861 return err;
862
863 rt->dst.rt6_next = iter;
864 *ins = rt;
865 rt->rt6i_node = fn;
866 atomic_inc(&rt->rt6i_ref);
867 inet6_rt_notify(RTM_NEWROUTE, rt, info, 0);
868 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
869
870 if (!(fn->fn_flags & RTN_RTINFO)) {
871 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
872 fn->fn_flags |= RTN_RTINFO;
873 }
874
875 } else {
876 int nsiblings;
877
878 if (!found) {
879 if (add)
880 goto add;
881 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
882 return -ENOENT;
883 }
884
885 err = fib6_commit_metrics(&rt->dst, mxc);
886 if (err)
887 return err;
888
889 *ins = rt;
890 rt->rt6i_node = fn;
891 rt->dst.rt6_next = iter->dst.rt6_next;
892 atomic_inc(&rt->rt6i_ref);
893 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
894 if (!(fn->fn_flags & RTN_RTINFO)) {
895 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
896 fn->fn_flags |= RTN_RTINFO;
897 }
898 nsiblings = iter->rt6i_nsiblings;
899 fib6_purge_rt(iter, fn, info->nl_net);
900 rt6_release(iter);
901
902 if (nsiblings) {
903 /* Replacing an ECMP route, remove all siblings */
904 ins = &rt->dst.rt6_next;
905 iter = *ins;
906 while (iter) {
907 if (rt6_qualify_for_ecmp(iter)) {
908 *ins = iter->dst.rt6_next;
909 fib6_purge_rt(iter, fn, info->nl_net);
910 rt6_release(iter);
911 nsiblings--;
912 } else {
913 ins = &iter->dst.rt6_next;
914 }
915 iter = *ins;
916 }
917 WARN_ON(nsiblings != 0);
918 }
919 }
920
921 return 0;
922 }
923
924 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
925 {
926 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
927 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
928 mod_timer(&net->ipv6.ip6_fib_timer,
929 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
930 }
931
932 void fib6_force_start_gc(struct net *net)
933 {
934 if (!timer_pending(&net->ipv6.ip6_fib_timer))
935 mod_timer(&net->ipv6.ip6_fib_timer,
936 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
937 }
938
939 /*
940 * Add routing information to the routing tree.
941 * <destination addr>/<source addr>
942 * with source addr info in sub-trees
943 */
944
945 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
946 struct nl_info *info, struct mx6_config *mxc)
947 {
948 struct fib6_node *fn, *pn = NULL;
949 int err = -ENOMEM;
950 int allow_create = 1;
951 int replace_required = 0;
952 int sernum = fib6_new_sernum(info->nl_net);
953
954 if (WARN_ON_ONCE((rt->dst.flags & DST_NOCACHE) &&
955 !atomic_read(&rt->dst.__refcnt)))
956 return -EINVAL;
957
958 if (info->nlh) {
959 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
960 allow_create = 0;
961 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
962 replace_required = 1;
963 }
964 if (!allow_create && !replace_required)
965 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
966
967 fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
968 offsetof(struct rt6_info, rt6i_dst), allow_create,
969 replace_required, sernum);
970 if (IS_ERR(fn)) {
971 err = PTR_ERR(fn);
972 fn = NULL;
973 goto out;
974 }
975
976 pn = fn;
977
978 #ifdef CONFIG_IPV6_SUBTREES
979 if (rt->rt6i_src.plen) {
980 struct fib6_node *sn;
981
982 if (!fn->subtree) {
983 struct fib6_node *sfn;
984
985 /*
986 * Create subtree.
987 *
988 * fn[main tree]
989 * |
990 * sfn[subtree root]
991 * \
992 * sn[new leaf node]
993 */
994
995 /* Create subtree root node */
996 sfn = node_alloc();
997 if (!sfn)
998 goto st_failure;
999
1000 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
1001 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
1002 sfn->fn_flags = RTN_ROOT;
1003 sfn->fn_sernum = sernum;
1004
1005 /* Now add the first leaf node to new subtree */
1006
1007 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
1008 rt->rt6i_src.plen,
1009 offsetof(struct rt6_info, rt6i_src),
1010 allow_create, replace_required, sernum);
1011
1012 if (IS_ERR(sn)) {
1013 /* If it is failed, discard just allocated
1014 root, and then (in st_failure) stale node
1015 in main tree.
1016 */
1017 node_free(sfn);
1018 err = PTR_ERR(sn);
1019 goto st_failure;
1020 }
1021
1022 /* Now link new subtree to main tree */
1023 sfn->parent = fn;
1024 fn->subtree = sfn;
1025 } else {
1026 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
1027 rt->rt6i_src.plen,
1028 offsetof(struct rt6_info, rt6i_src),
1029 allow_create, replace_required, sernum);
1030
1031 if (IS_ERR(sn)) {
1032 err = PTR_ERR(sn);
1033 goto st_failure;
1034 }
1035 }
1036
1037 if (!fn->leaf) {
1038 fn->leaf = rt;
1039 atomic_inc(&rt->rt6i_ref);
1040 }
1041 fn = sn;
1042 }
1043 #endif
1044
1045 err = fib6_add_rt2node(fn, rt, info, mxc);
1046 if (!err) {
1047 fib6_start_gc(info->nl_net, rt);
1048 if (!(rt->rt6i_flags & RTF_CACHE))
1049 fib6_prune_clones(info->nl_net, pn);
1050 rt->dst.flags &= ~DST_NOCACHE;
1051 }
1052
1053 out:
1054 if (err) {
1055 #ifdef CONFIG_IPV6_SUBTREES
1056 /*
1057 * If fib6_add_1 has cleared the old leaf pointer in the
1058 * super-tree leaf node we have to find a new one for it.
1059 */
1060 if (pn != fn && pn->leaf == rt) {
1061 pn->leaf = NULL;
1062 atomic_dec(&rt->rt6i_ref);
1063 }
1064 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
1065 pn->leaf = fib6_find_prefix(info->nl_net, pn);
1066 #if RT6_DEBUG >= 2
1067 if (!pn->leaf) {
1068 WARN_ON(pn->leaf == NULL);
1069 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
1070 }
1071 #endif
1072 atomic_inc(&pn->leaf->rt6i_ref);
1073 }
1074 #endif
1075 if (!(rt->dst.flags & DST_NOCACHE))
1076 dst_free(&rt->dst);
1077 }
1078 return err;
1079
1080 #ifdef CONFIG_IPV6_SUBTREES
1081 /* Subtree creation failed, probably main tree node
1082 is orphan. If it is, shoot it.
1083 */
1084 st_failure:
1085 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1086 fib6_repair_tree(info->nl_net, fn);
1087 if (!(rt->dst.flags & DST_NOCACHE))
1088 dst_free(&rt->dst);
1089 return err;
1090 #endif
1091 }
1092
1093 /*
1094 * Routing tree lookup
1095 *
1096 */
1097
1098 struct lookup_args {
1099 int offset; /* key offset on rt6_info */
1100 const struct in6_addr *addr; /* search key */
1101 };
1102
1103 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1104 struct lookup_args *args)
1105 {
1106 struct fib6_node *fn;
1107 __be32 dir;
1108
1109 if (unlikely(args->offset == 0))
1110 return NULL;
1111
1112 /*
1113 * Descend on a tree
1114 */
1115
1116 fn = root;
1117
1118 for (;;) {
1119 struct fib6_node *next;
1120
1121 dir = addr_bit_set(args->addr, fn->fn_bit);
1122
1123 next = dir ? fn->right : fn->left;
1124
1125 if (next) {
1126 fn = next;
1127 continue;
1128 }
1129 break;
1130 }
1131
1132 while (fn) {
1133 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1134 struct rt6key *key;
1135
1136 key = (struct rt6key *) ((u8 *) fn->leaf +
1137 args->offset);
1138
1139 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1140 #ifdef CONFIG_IPV6_SUBTREES
1141 if (fn->subtree) {
1142 struct fib6_node *sfn;
1143 sfn = fib6_lookup_1(fn->subtree,
1144 args + 1);
1145 if (!sfn)
1146 goto backtrack;
1147 fn = sfn;
1148 }
1149 #endif
1150 if (fn->fn_flags & RTN_RTINFO)
1151 return fn;
1152 }
1153 }
1154 #ifdef CONFIG_IPV6_SUBTREES
1155 backtrack:
1156 #endif
1157 if (fn->fn_flags & RTN_ROOT)
1158 break;
1159
1160 fn = fn->parent;
1161 }
1162
1163 return NULL;
1164 }
1165
1166 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1167 const struct in6_addr *saddr)
1168 {
1169 struct fib6_node *fn;
1170 struct lookup_args args[] = {
1171 {
1172 .offset = offsetof(struct rt6_info, rt6i_dst),
1173 .addr = daddr,
1174 },
1175 #ifdef CONFIG_IPV6_SUBTREES
1176 {
1177 .offset = offsetof(struct rt6_info, rt6i_src),
1178 .addr = saddr,
1179 },
1180 #endif
1181 {
1182 .offset = 0, /* sentinel */
1183 }
1184 };
1185
1186 fn = fib6_lookup_1(root, daddr ? args : args + 1);
1187 if (!fn || fn->fn_flags & RTN_TL_ROOT)
1188 fn = root;
1189
1190 return fn;
1191 }
1192
1193 /*
1194 * Get node with specified destination prefix (and source prefix,
1195 * if subtrees are used)
1196 */
1197
1198
1199 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1200 const struct in6_addr *addr,
1201 int plen, int offset)
1202 {
1203 struct fib6_node *fn;
1204
1205 for (fn = root; fn ; ) {
1206 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1207
1208 /*
1209 * Prefix match
1210 */
1211 if (plen < fn->fn_bit ||
1212 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1213 return NULL;
1214
1215 if (plen == fn->fn_bit)
1216 return fn;
1217
1218 /*
1219 * We have more bits to go
1220 */
1221 if (addr_bit_set(addr, fn->fn_bit))
1222 fn = fn->right;
1223 else
1224 fn = fn->left;
1225 }
1226 return NULL;
1227 }
1228
1229 struct fib6_node *fib6_locate(struct fib6_node *root,
1230 const struct in6_addr *daddr, int dst_len,
1231 const struct in6_addr *saddr, int src_len)
1232 {
1233 struct fib6_node *fn;
1234
1235 fn = fib6_locate_1(root, daddr, dst_len,
1236 offsetof(struct rt6_info, rt6i_dst));
1237
1238 #ifdef CONFIG_IPV6_SUBTREES
1239 if (src_len) {
1240 WARN_ON(saddr == NULL);
1241 if (fn && fn->subtree)
1242 fn = fib6_locate_1(fn->subtree, saddr, src_len,
1243 offsetof(struct rt6_info, rt6i_src));
1244 }
1245 #endif
1246
1247 if (fn && fn->fn_flags & RTN_RTINFO)
1248 return fn;
1249
1250 return NULL;
1251 }
1252
1253
1254 /*
1255 * Deletion
1256 *
1257 */
1258
1259 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1260 {
1261 if (fn->fn_flags & RTN_ROOT)
1262 return net->ipv6.ip6_null_entry;
1263
1264 while (fn) {
1265 if (fn->left)
1266 return fn->left->leaf;
1267 if (fn->right)
1268 return fn->right->leaf;
1269
1270 fn = FIB6_SUBTREE(fn);
1271 }
1272 return NULL;
1273 }
1274
1275 /*
1276 * Called to trim the tree of intermediate nodes when possible. "fn"
1277 * is the node we want to try and remove.
1278 */
1279
1280 static struct fib6_node *fib6_repair_tree(struct net *net,
1281 struct fib6_node *fn)
1282 {
1283 int children;
1284 int nstate;
1285 struct fib6_node *child, *pn;
1286 struct fib6_walker *w;
1287 int iter = 0;
1288
1289 for (;;) {
1290 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1291 iter++;
1292
1293 WARN_ON(fn->fn_flags & RTN_RTINFO);
1294 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1295 WARN_ON(fn->leaf);
1296
1297 children = 0;
1298 child = NULL;
1299 if (fn->right)
1300 child = fn->right, children |= 1;
1301 if (fn->left)
1302 child = fn->left, children |= 2;
1303
1304 if (children == 3 || FIB6_SUBTREE(fn)
1305 #ifdef CONFIG_IPV6_SUBTREES
1306 /* Subtree root (i.e. fn) may have one child */
1307 || (children && fn->fn_flags & RTN_ROOT)
1308 #endif
1309 ) {
1310 fn->leaf = fib6_find_prefix(net, fn);
1311 #if RT6_DEBUG >= 2
1312 if (!fn->leaf) {
1313 WARN_ON(!fn->leaf);
1314 fn->leaf = net->ipv6.ip6_null_entry;
1315 }
1316 #endif
1317 atomic_inc(&fn->leaf->rt6i_ref);
1318 return fn->parent;
1319 }
1320
1321 pn = fn->parent;
1322 #ifdef CONFIG_IPV6_SUBTREES
1323 if (FIB6_SUBTREE(pn) == fn) {
1324 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1325 FIB6_SUBTREE(pn) = NULL;
1326 nstate = FWS_L;
1327 } else {
1328 WARN_ON(fn->fn_flags & RTN_ROOT);
1329 #endif
1330 if (pn->right == fn)
1331 pn->right = child;
1332 else if (pn->left == fn)
1333 pn->left = child;
1334 #if RT6_DEBUG >= 2
1335 else
1336 WARN_ON(1);
1337 #endif
1338 if (child)
1339 child->parent = pn;
1340 nstate = FWS_R;
1341 #ifdef CONFIG_IPV6_SUBTREES
1342 }
1343 #endif
1344
1345 read_lock(&net->ipv6.fib6_walker_lock);
1346 FOR_WALKERS(net, w) {
1347 if (!child) {
1348 if (w->root == fn) {
1349 w->root = w->node = NULL;
1350 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1351 } else if (w->node == fn) {
1352 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1353 w->node = pn;
1354 w->state = nstate;
1355 }
1356 } else {
1357 if (w->root == fn) {
1358 w->root = child;
1359 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1360 }
1361 if (w->node == fn) {
1362 w->node = child;
1363 if (children&2) {
1364 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1365 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1366 } else {
1367 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1368 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1369 }
1370 }
1371 }
1372 }
1373 read_unlock(&net->ipv6.fib6_walker_lock);
1374
1375 node_free(fn);
1376 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1377 return pn;
1378
1379 rt6_release(pn->leaf);
1380 pn->leaf = NULL;
1381 fn = pn;
1382 }
1383 }
1384
1385 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1386 struct nl_info *info)
1387 {
1388 struct fib6_walker *w;
1389 struct rt6_info *rt = *rtp;
1390 struct net *net = info->nl_net;
1391
1392 RT6_TRACE("fib6_del_route\n");
1393
1394 /* Unlink it */
1395 *rtp = rt->dst.rt6_next;
1396 rt->rt6i_node = NULL;
1397 net->ipv6.rt6_stats->fib_rt_entries--;
1398 net->ipv6.rt6_stats->fib_discarded_routes++;
1399
1400 /* Reset round-robin state, if necessary */
1401 if (fn->rr_ptr == rt)
1402 fn->rr_ptr = NULL;
1403
1404 /* Remove this entry from other siblings */
1405 if (rt->rt6i_nsiblings) {
1406 struct rt6_info *sibling, *next_sibling;
1407
1408 list_for_each_entry_safe(sibling, next_sibling,
1409 &rt->rt6i_siblings, rt6i_siblings)
1410 sibling->rt6i_nsiblings--;
1411 rt->rt6i_nsiblings = 0;
1412 list_del_init(&rt->rt6i_siblings);
1413 }
1414
1415 /* Adjust walkers */
1416 read_lock(&net->ipv6.fib6_walker_lock);
1417 FOR_WALKERS(net, w) {
1418 if (w->state == FWS_C && w->leaf == rt) {
1419 RT6_TRACE("walker %p adjusted by delroute\n", w);
1420 w->leaf = rt->dst.rt6_next;
1421 if (!w->leaf)
1422 w->state = FWS_U;
1423 }
1424 }
1425 read_unlock(&net->ipv6.fib6_walker_lock);
1426
1427 rt->dst.rt6_next = NULL;
1428
1429 /* If it was last route, expunge its radix tree node */
1430 if (!fn->leaf) {
1431 fn->fn_flags &= ~RTN_RTINFO;
1432 net->ipv6.rt6_stats->fib_route_nodes--;
1433 fn = fib6_repair_tree(net, fn);
1434 }
1435
1436 fib6_purge_rt(rt, fn, net);
1437
1438 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1439 rt6_release(rt);
1440 }
1441
1442 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1443 {
1444 struct net *net = info->nl_net;
1445 struct fib6_node *fn = rt->rt6i_node;
1446 struct rt6_info **rtp;
1447
1448 #if RT6_DEBUG >= 2
1449 if (rt->dst.obsolete > 0) {
1450 WARN_ON(fn);
1451 return -ENOENT;
1452 }
1453 #endif
1454 if (!fn || rt == net->ipv6.ip6_null_entry)
1455 return -ENOENT;
1456
1457 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1458
1459 if (!(rt->rt6i_flags & RTF_CACHE)) {
1460 struct fib6_node *pn = fn;
1461 #ifdef CONFIG_IPV6_SUBTREES
1462 /* clones of this route might be in another subtree */
1463 if (rt->rt6i_src.plen) {
1464 while (!(pn->fn_flags & RTN_ROOT))
1465 pn = pn->parent;
1466 pn = pn->parent;
1467 }
1468 #endif
1469 fib6_prune_clones(info->nl_net, pn);
1470 }
1471
1472 /*
1473 * Walk the leaf entries looking for ourself
1474 */
1475
1476 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1477 if (*rtp == rt) {
1478 fib6_del_route(fn, rtp, info);
1479 return 0;
1480 }
1481 }
1482 return -ENOENT;
1483 }
1484
1485 /*
1486 * Tree traversal function.
1487 *
1488 * Certainly, it is not interrupt safe.
1489 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1490 * It means, that we can modify tree during walking
1491 * and use this function for garbage collection, clone pruning,
1492 * cleaning tree when a device goes down etc. etc.
1493 *
1494 * It guarantees that every node will be traversed,
1495 * and that it will be traversed only once.
1496 *
1497 * Callback function w->func may return:
1498 * 0 -> continue walking.
1499 * positive value -> walking is suspended (used by tree dumps,
1500 * and probably by gc, if it will be split to several slices)
1501 * negative value -> terminate walking.
1502 *
1503 * The function itself returns:
1504 * 0 -> walk is complete.
1505 * >0 -> walk is incomplete (i.e. suspended)
1506 * <0 -> walk is terminated by an error.
1507 */
1508
1509 static int fib6_walk_continue(struct fib6_walker *w)
1510 {
1511 struct fib6_node *fn, *pn;
1512
1513 for (;;) {
1514 fn = w->node;
1515 if (!fn)
1516 return 0;
1517
1518 if (w->prune && fn != w->root &&
1519 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1520 w->state = FWS_C;
1521 w->leaf = fn->leaf;
1522 }
1523 switch (w->state) {
1524 #ifdef CONFIG_IPV6_SUBTREES
1525 case FWS_S:
1526 if (FIB6_SUBTREE(fn)) {
1527 w->node = FIB6_SUBTREE(fn);
1528 continue;
1529 }
1530 w->state = FWS_L;
1531 #endif
1532 case FWS_L:
1533 if (fn->left) {
1534 w->node = fn->left;
1535 w->state = FWS_INIT;
1536 continue;
1537 }
1538 w->state = FWS_R;
1539 case FWS_R:
1540 if (fn->right) {
1541 w->node = fn->right;
1542 w->state = FWS_INIT;
1543 continue;
1544 }
1545 w->state = FWS_C;
1546 w->leaf = fn->leaf;
1547 case FWS_C:
1548 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1549 int err;
1550
1551 if (w->skip) {
1552 w->skip--;
1553 goto skip;
1554 }
1555
1556 err = w->func(w);
1557 if (err)
1558 return err;
1559
1560 w->count++;
1561 continue;
1562 }
1563 skip:
1564 w->state = FWS_U;
1565 case FWS_U:
1566 if (fn == w->root)
1567 return 0;
1568 pn = fn->parent;
1569 w->node = pn;
1570 #ifdef CONFIG_IPV6_SUBTREES
1571 if (FIB6_SUBTREE(pn) == fn) {
1572 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1573 w->state = FWS_L;
1574 continue;
1575 }
1576 #endif
1577 if (pn->left == fn) {
1578 w->state = FWS_R;
1579 continue;
1580 }
1581 if (pn->right == fn) {
1582 w->state = FWS_C;
1583 w->leaf = w->node->leaf;
1584 continue;
1585 }
1586 #if RT6_DEBUG >= 2
1587 WARN_ON(1);
1588 #endif
1589 }
1590 }
1591 }
1592
1593 static int fib6_walk(struct net *net, struct fib6_walker *w)
1594 {
1595 int res;
1596
1597 w->state = FWS_INIT;
1598 w->node = w->root;
1599
1600 fib6_walker_link(net, w);
1601 res = fib6_walk_continue(w);
1602 if (res <= 0)
1603 fib6_walker_unlink(net, w);
1604 return res;
1605 }
1606
1607 static int fib6_clean_node(struct fib6_walker *w)
1608 {
1609 int res;
1610 struct rt6_info *rt;
1611 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1612 struct nl_info info = {
1613 .nl_net = c->net,
1614 };
1615
1616 if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1617 w->node->fn_sernum != c->sernum)
1618 w->node->fn_sernum = c->sernum;
1619
1620 if (!c->func) {
1621 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1622 w->leaf = NULL;
1623 return 0;
1624 }
1625
1626 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1627 res = c->func(rt, c->arg);
1628 if (res < 0) {
1629 w->leaf = rt;
1630 res = fib6_del(rt, &info);
1631 if (res) {
1632 #if RT6_DEBUG >= 2
1633 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1634 __func__, rt, rt->rt6i_node, res);
1635 #endif
1636 continue;
1637 }
1638 return 0;
1639 }
1640 WARN_ON(res != 0);
1641 }
1642 w->leaf = rt;
1643 return 0;
1644 }
1645
1646 /*
1647 * Convenient frontend to tree walker.
1648 *
1649 * func is called on each route.
1650 * It may return -1 -> delete this route.
1651 * 0 -> continue walking
1652 *
1653 * prune==1 -> only immediate children of node (certainly,
1654 * ignoring pure split nodes) will be scanned.
1655 */
1656
1657 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1658 int (*func)(struct rt6_info *, void *arg),
1659 bool prune, int sernum, void *arg)
1660 {
1661 struct fib6_cleaner c;
1662
1663 c.w.root = root;
1664 c.w.func = fib6_clean_node;
1665 c.w.prune = prune;
1666 c.w.count = 0;
1667 c.w.skip = 0;
1668 c.func = func;
1669 c.sernum = sernum;
1670 c.arg = arg;
1671 c.net = net;
1672
1673 fib6_walk(net, &c.w);
1674 }
1675
1676 static void __fib6_clean_all(struct net *net,
1677 int (*func)(struct rt6_info *, void *),
1678 int sernum, void *arg)
1679 {
1680 struct fib6_table *table;
1681 struct hlist_head *head;
1682 unsigned int h;
1683
1684 rcu_read_lock();
1685 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1686 head = &net->ipv6.fib_table_hash[h];
1687 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1688 write_lock_bh(&table->tb6_lock);
1689 fib6_clean_tree(net, &table->tb6_root,
1690 func, false, sernum, arg);
1691 write_unlock_bh(&table->tb6_lock);
1692 }
1693 }
1694 rcu_read_unlock();
1695 }
1696
1697 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1698 void *arg)
1699 {
1700 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1701 }
1702
1703 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1704 {
1705 if (rt->rt6i_flags & RTF_CACHE) {
1706 RT6_TRACE("pruning clone %p\n", rt);
1707 return -1;
1708 }
1709
1710 return 0;
1711 }
1712
1713 static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1714 {
1715 fib6_clean_tree(net, fn, fib6_prune_clone, true,
1716 FIB6_NO_SERNUM_CHANGE, NULL);
1717 }
1718
1719 static void fib6_flush_trees(struct net *net)
1720 {
1721 int new_sernum = fib6_new_sernum(net);
1722
1723 __fib6_clean_all(net, NULL, new_sernum, NULL);
1724 }
1725
1726 /*
1727 * Garbage collection
1728 */
1729
1730 struct fib6_gc_args
1731 {
1732 int timeout;
1733 int more;
1734 };
1735
1736 static int fib6_age(struct rt6_info *rt, void *arg)
1737 {
1738 struct fib6_gc_args *gc_args = arg;
1739 unsigned long now = jiffies;
1740
1741 /*
1742 * check addrconf expiration here.
1743 * Routes are expired even if they are in use.
1744 *
1745 * Also age clones. Note, that clones are aged out
1746 * only if they are not in use now.
1747 */
1748
1749 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1750 if (time_after(now, rt->dst.expires)) {
1751 RT6_TRACE("expiring %p\n", rt);
1752 return -1;
1753 }
1754 gc_args->more++;
1755 } else if (rt->rt6i_flags & RTF_CACHE) {
1756 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1757 time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
1758 RT6_TRACE("aging clone %p\n", rt);
1759 return -1;
1760 } else if (rt->rt6i_flags & RTF_GATEWAY) {
1761 struct neighbour *neigh;
1762 __u8 neigh_flags = 0;
1763
1764 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1765 if (neigh) {
1766 neigh_flags = neigh->flags;
1767 neigh_release(neigh);
1768 }
1769 if (!(neigh_flags & NTF_ROUTER)) {
1770 RT6_TRACE("purging route %p via non-router but gateway\n",
1771 rt);
1772 return -1;
1773 }
1774 }
1775 gc_args->more++;
1776 }
1777
1778 return 0;
1779 }
1780
1781 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1782 {
1783 struct fib6_gc_args gc_args;
1784 unsigned long now;
1785
1786 if (force) {
1787 spin_lock_bh(&net->ipv6.fib6_gc_lock);
1788 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
1789 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1790 return;
1791 }
1792 gc_args.timeout = expires ? (int)expires :
1793 net->ipv6.sysctl.ip6_rt_gc_interval;
1794
1795 gc_args.more = icmp6_dst_gc();
1796
1797 fib6_clean_all(net, fib6_age, &gc_args);
1798 now = jiffies;
1799 net->ipv6.ip6_rt_last_gc = now;
1800
1801 if (gc_args.more)
1802 mod_timer(&net->ipv6.ip6_fib_timer,
1803 round_jiffies(now
1804 + net->ipv6.sysctl.ip6_rt_gc_interval));
1805 else
1806 del_timer(&net->ipv6.ip6_fib_timer);
1807 spin_unlock_bh(&net->ipv6.fib6_gc_lock);
1808 }
1809
1810 static void fib6_gc_timer_cb(unsigned long arg)
1811 {
1812 fib6_run_gc(0, (struct net *)arg, true);
1813 }
1814
1815 static int __net_init fib6_net_init(struct net *net)
1816 {
1817 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1818
1819 spin_lock_init(&net->ipv6.fib6_gc_lock);
1820 rwlock_init(&net->ipv6.fib6_walker_lock);
1821 INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
1822 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1823
1824 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1825 if (!net->ipv6.rt6_stats)
1826 goto out_timer;
1827
1828 /* Avoid false sharing : Use at least a full cache line */
1829 size = max_t(size_t, size, L1_CACHE_BYTES);
1830
1831 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1832 if (!net->ipv6.fib_table_hash)
1833 goto out_rt6_stats;
1834
1835 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1836 GFP_KERNEL);
1837 if (!net->ipv6.fib6_main_tbl)
1838 goto out_fib_table_hash;
1839
1840 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1841 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1842 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1843 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1844 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1845
1846 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1847 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1848 GFP_KERNEL);
1849 if (!net->ipv6.fib6_local_tbl)
1850 goto out_fib6_main_tbl;
1851 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1852 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1853 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1854 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1855 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1856 #endif
1857 fib6_tables_init(net);
1858
1859 return 0;
1860
1861 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1862 out_fib6_main_tbl:
1863 kfree(net->ipv6.fib6_main_tbl);
1864 #endif
1865 out_fib_table_hash:
1866 kfree(net->ipv6.fib_table_hash);
1867 out_rt6_stats:
1868 kfree(net->ipv6.rt6_stats);
1869 out_timer:
1870 return -ENOMEM;
1871 }
1872
1873 static void fib6_net_exit(struct net *net)
1874 {
1875 rt6_ifdown(net, NULL);
1876 del_timer_sync(&net->ipv6.ip6_fib_timer);
1877
1878 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1879 inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1880 kfree(net->ipv6.fib6_local_tbl);
1881 #endif
1882 inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1883 kfree(net->ipv6.fib6_main_tbl);
1884 kfree(net->ipv6.fib_table_hash);
1885 kfree(net->ipv6.rt6_stats);
1886 }
1887
1888 static struct pernet_operations fib6_net_ops = {
1889 .init = fib6_net_init,
1890 .exit = fib6_net_exit,
1891 };
1892
1893 int __init fib6_init(void)
1894 {
1895 int ret = -ENOMEM;
1896
1897 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1898 sizeof(struct fib6_node),
1899 0, SLAB_HWCACHE_ALIGN,
1900 NULL);
1901 if (!fib6_node_kmem)
1902 goto out;
1903
1904 ret = register_pernet_subsys(&fib6_net_ops);
1905 if (ret)
1906 goto out_kmem_cache_create;
1907
1908 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1909 NULL);
1910 if (ret)
1911 goto out_unregister_subsys;
1912
1913 __fib6_flush_trees = fib6_flush_trees;
1914 out:
1915 return ret;
1916
1917 out_unregister_subsys:
1918 unregister_pernet_subsys(&fib6_net_ops);
1919 out_kmem_cache_create:
1920 kmem_cache_destroy(fib6_node_kmem);
1921 goto out;
1922 }
1923
1924 void fib6_gc_cleanup(void)
1925 {
1926 unregister_pernet_subsys(&fib6_net_ops);
1927 kmem_cache_destroy(fib6_node_kmem);
1928 }
1929
1930 #ifdef CONFIG_PROC_FS
1931
1932 struct ipv6_route_iter {
1933 struct seq_net_private p;
1934 struct fib6_walker w;
1935 loff_t skip;
1936 struct fib6_table *tbl;
1937 int sernum;
1938 };
1939
1940 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
1941 {
1942 struct rt6_info *rt = v;
1943 struct ipv6_route_iter *iter = seq->private;
1944
1945 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
1946
1947 #ifdef CONFIG_IPV6_SUBTREES
1948 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
1949 #else
1950 seq_puts(seq, "00000000000000000000000000000000 00 ");
1951 #endif
1952 if (rt->rt6i_flags & RTF_GATEWAY)
1953 seq_printf(seq, "%pi6", &rt->rt6i_gateway);
1954 else
1955 seq_puts(seq, "00000000000000000000000000000000");
1956
1957 seq_printf(seq, " %08x %08x %08x %08x %8s\n",
1958 rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
1959 rt->dst.__use, rt->rt6i_flags,
1960 rt->dst.dev ? rt->dst.dev->name : "");
1961 iter->w.leaf = NULL;
1962 return 0;
1963 }
1964
1965 static int ipv6_route_yield(struct fib6_walker *w)
1966 {
1967 struct ipv6_route_iter *iter = w->args;
1968
1969 if (!iter->skip)
1970 return 1;
1971
1972 do {
1973 iter->w.leaf = iter->w.leaf->dst.rt6_next;
1974 iter->skip--;
1975 if (!iter->skip && iter->w.leaf)
1976 return 1;
1977 } while (iter->w.leaf);
1978
1979 return 0;
1980 }
1981
1982 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
1983 struct net *net)
1984 {
1985 memset(&iter->w, 0, sizeof(iter->w));
1986 iter->w.func = ipv6_route_yield;
1987 iter->w.root = &iter->tbl->tb6_root;
1988 iter->w.state = FWS_INIT;
1989 iter->w.node = iter->w.root;
1990 iter->w.args = iter;
1991 iter->sernum = iter->w.root->fn_sernum;
1992 INIT_LIST_HEAD(&iter->w.lh);
1993 fib6_walker_link(net, &iter->w);
1994 }
1995
1996 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
1997 struct net *net)
1998 {
1999 unsigned int h;
2000 struct hlist_node *node;
2001
2002 if (tbl) {
2003 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2004 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2005 } else {
2006 h = 0;
2007 node = NULL;
2008 }
2009
2010 while (!node && h < FIB6_TABLE_HASHSZ) {
2011 node = rcu_dereference_bh(
2012 hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2013 }
2014 return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2015 }
2016
2017 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2018 {
2019 if (iter->sernum != iter->w.root->fn_sernum) {
2020 iter->sernum = iter->w.root->fn_sernum;
2021 iter->w.state = FWS_INIT;
2022 iter->w.node = iter->w.root;
2023 WARN_ON(iter->w.skip);
2024 iter->w.skip = iter->w.count;
2025 }
2026 }
2027
2028 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2029 {
2030 int r;
2031 struct rt6_info *n;
2032 struct net *net = seq_file_net(seq);
2033 struct ipv6_route_iter *iter = seq->private;
2034
2035 if (!v)
2036 goto iter_table;
2037
2038 n = ((struct rt6_info *)v)->dst.rt6_next;
2039 if (n) {
2040 ++*pos;
2041 return n;
2042 }
2043
2044 iter_table:
2045 ipv6_route_check_sernum(iter);
2046 read_lock(&iter->tbl->tb6_lock);
2047 r = fib6_walk_continue(&iter->w);
2048 read_unlock(&iter->tbl->tb6_lock);
2049 if (r > 0) {
2050 if (v)
2051 ++*pos;
2052 return iter->w.leaf;
2053 } else if (r < 0) {
2054 fib6_walker_unlink(net, &iter->w);
2055 return NULL;
2056 }
2057 fib6_walker_unlink(net, &iter->w);
2058
2059 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2060 if (!iter->tbl)
2061 return NULL;
2062
2063 ipv6_route_seq_setup_walk(iter, net);
2064 goto iter_table;
2065 }
2066
2067 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2068 __acquires(RCU_BH)
2069 {
2070 struct net *net = seq_file_net(seq);
2071 struct ipv6_route_iter *iter = seq->private;
2072
2073 rcu_read_lock_bh();
2074 iter->tbl = ipv6_route_seq_next_table(NULL, net);
2075 iter->skip = *pos;
2076
2077 if (iter->tbl) {
2078 ipv6_route_seq_setup_walk(iter, net);
2079 return ipv6_route_seq_next(seq, NULL, pos);
2080 } else {
2081 return NULL;
2082 }
2083 }
2084
2085 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2086 {
2087 struct fib6_walker *w = &iter->w;
2088 return w->node && !(w->state == FWS_U && w->node == w->root);
2089 }
2090
2091 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2092 __releases(RCU_BH)
2093 {
2094 struct net *net = seq_file_net(seq);
2095 struct ipv6_route_iter *iter = seq->private;
2096
2097 if (ipv6_route_iter_active(iter))
2098 fib6_walker_unlink(net, &iter->w);
2099
2100 rcu_read_unlock_bh();
2101 }
2102
2103 static const struct seq_operations ipv6_route_seq_ops = {
2104 .start = ipv6_route_seq_start,
2105 .next = ipv6_route_seq_next,
2106 .stop = ipv6_route_seq_stop,
2107 .show = ipv6_route_seq_show
2108 };
2109
2110 int ipv6_route_open(struct inode *inode, struct file *file)
2111 {
2112 return seq_open_net(inode, file, &ipv6_route_seq_ops,
2113 sizeof(struct ipv6_route_iter));
2114 }
2115
2116 #endif /* CONFIG_PROC_FS */
This page took 0.096317 seconds and 5 git commands to generate.