]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - net/ipv6/ip6_fib.c
Merge tag 'trace-fixes-3.14' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt...
[mirror_ubuntu-artful-kernel.git] / net / ipv6 / ip6_fib.c
1 /*
2 * Linux INET6 implementation
3 * Forwarding Information Database
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 */
13
14 /*
15 * Changes:
16 * Yuji SEKIYA @USAGI: Support default route on router node;
17 * remove ip6_null_entry from the top of
18 * routing table.
19 * Ville Nuorvala: Fixed routing subtrees.
20 */
21
22 #define pr_fmt(fmt) "IPv6: " fmt
23
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/net.h>
27 #include <linux/route.h>
28 #include <linux/netdevice.h>
29 #include <linux/in6.h>
30 #include <linux/init.h>
31 #include <linux/list.h>
32 #include <linux/slab.h>
33
34 #include <net/ipv6.h>
35 #include <net/ndisc.h>
36 #include <net/addrconf.h>
37
38 #include <net/ip6_fib.h>
39 #include <net/ip6_route.h>
40
41 #define RT6_DEBUG 2
42
43 #if RT6_DEBUG >= 3
44 #define RT6_TRACE(x...) pr_debug(x)
45 #else
46 #define RT6_TRACE(x...) do { ; } while (0)
47 #endif
48
49 static struct kmem_cache * fib6_node_kmem __read_mostly;
50
51 enum fib_walk_state_t
52 {
53 #ifdef CONFIG_IPV6_SUBTREES
54 FWS_S,
55 #endif
56 FWS_L,
57 FWS_R,
58 FWS_C,
59 FWS_U
60 };
61
62 struct fib6_cleaner_t
63 {
64 struct fib6_walker_t w;
65 struct net *net;
66 int (*func)(struct rt6_info *, void *arg);
67 void *arg;
68 };
69
70 static DEFINE_RWLOCK(fib6_walker_lock);
71
72 #ifdef CONFIG_IPV6_SUBTREES
73 #define FWS_INIT FWS_S
74 #else
75 #define FWS_INIT FWS_L
76 #endif
77
78 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
79 struct rt6_info *rt);
80 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
81 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
82 static int fib6_walk(struct fib6_walker_t *w);
83 static int fib6_walk_continue(struct fib6_walker_t *w);
84
85 /*
86 * A routing update causes an increase of the serial number on the
87 * affected subtree. This allows for cached routes to be asynchronously
88 * tested when modifications are made to the destination cache as a
89 * result of redirects, path MTU changes, etc.
90 */
91
92 static __u32 rt_sernum;
93
94 static void fib6_gc_timer_cb(unsigned long arg);
95
96 static LIST_HEAD(fib6_walkers);
97 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
98
99 static inline void fib6_walker_link(struct fib6_walker_t *w)
100 {
101 write_lock_bh(&fib6_walker_lock);
102 list_add(&w->lh, &fib6_walkers);
103 write_unlock_bh(&fib6_walker_lock);
104 }
105
106 static inline void fib6_walker_unlink(struct fib6_walker_t *w)
107 {
108 write_lock_bh(&fib6_walker_lock);
109 list_del(&w->lh);
110 write_unlock_bh(&fib6_walker_lock);
111 }
112 static __inline__ u32 fib6_new_sernum(void)
113 {
114 u32 n = ++rt_sernum;
115 if ((__s32)n <= 0)
116 rt_sernum = n = 1;
117 return n;
118 }
119
120 /*
121 * Auxiliary address test functions for the radix tree.
122 *
123 * These assume a 32bit processor (although it will work on
124 * 64bit processors)
125 */
126
127 /*
128 * test bit
129 */
130 #if defined(__LITTLE_ENDIAN)
131 # define BITOP_BE32_SWIZZLE (0x1F & ~7)
132 #else
133 # define BITOP_BE32_SWIZZLE 0
134 #endif
135
136 static __inline__ __be32 addr_bit_set(const void *token, int fn_bit)
137 {
138 const __be32 *addr = token;
139 /*
140 * Here,
141 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
142 * is optimized version of
143 * htonl(1 << ((~fn_bit)&0x1F))
144 * See include/asm-generic/bitops/le.h.
145 */
146 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
147 addr[fn_bit >> 5];
148 }
149
150 static __inline__ struct fib6_node * node_alloc(void)
151 {
152 struct fib6_node *fn;
153
154 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
155
156 return fn;
157 }
158
159 static __inline__ void node_free(struct fib6_node * fn)
160 {
161 kmem_cache_free(fib6_node_kmem, fn);
162 }
163
164 static __inline__ void rt6_release(struct rt6_info *rt)
165 {
166 if (atomic_dec_and_test(&rt->rt6i_ref))
167 dst_free(&rt->dst);
168 }
169
170 static void fib6_link_table(struct net *net, struct fib6_table *tb)
171 {
172 unsigned int h;
173
174 /*
175 * Initialize table lock at a single place to give lockdep a key,
176 * tables aren't visible prior to being linked to the list.
177 */
178 rwlock_init(&tb->tb6_lock);
179
180 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
181
182 /*
183 * No protection necessary, this is the only list mutatation
184 * operation, tables never disappear once they exist.
185 */
186 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
187 }
188
189 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
190
191 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
192 {
193 struct fib6_table *table;
194
195 table = kzalloc(sizeof(*table), GFP_ATOMIC);
196 if (table) {
197 table->tb6_id = id;
198 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
199 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
200 inet_peer_base_init(&table->tb6_peers);
201 }
202
203 return table;
204 }
205
206 struct fib6_table *fib6_new_table(struct net *net, u32 id)
207 {
208 struct fib6_table *tb;
209
210 if (id == 0)
211 id = RT6_TABLE_MAIN;
212 tb = fib6_get_table(net, id);
213 if (tb)
214 return tb;
215
216 tb = fib6_alloc_table(net, id);
217 if (tb)
218 fib6_link_table(net, tb);
219
220 return tb;
221 }
222
223 struct fib6_table *fib6_get_table(struct net *net, u32 id)
224 {
225 struct fib6_table *tb;
226 struct hlist_head *head;
227 unsigned int h;
228
229 if (id == 0)
230 id = RT6_TABLE_MAIN;
231 h = id & (FIB6_TABLE_HASHSZ - 1);
232 rcu_read_lock();
233 head = &net->ipv6.fib_table_hash[h];
234 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
235 if (tb->tb6_id == id) {
236 rcu_read_unlock();
237 return tb;
238 }
239 }
240 rcu_read_unlock();
241
242 return NULL;
243 }
244
245 static void __net_init fib6_tables_init(struct net *net)
246 {
247 fib6_link_table(net, net->ipv6.fib6_main_tbl);
248 fib6_link_table(net, net->ipv6.fib6_local_tbl);
249 }
250 #else
251
252 struct fib6_table *fib6_new_table(struct net *net, u32 id)
253 {
254 return fib6_get_table(net, id);
255 }
256
257 struct fib6_table *fib6_get_table(struct net *net, u32 id)
258 {
259 return net->ipv6.fib6_main_tbl;
260 }
261
262 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
263 int flags, pol_lookup_t lookup)
264 {
265 return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
266 }
267
268 static void __net_init fib6_tables_init(struct net *net)
269 {
270 fib6_link_table(net, net->ipv6.fib6_main_tbl);
271 }
272
273 #endif
274
275 static int fib6_dump_node(struct fib6_walker_t *w)
276 {
277 int res;
278 struct rt6_info *rt;
279
280 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
281 res = rt6_dump_route(rt, w->args);
282 if (res < 0) {
283 /* Frame is full, suspend walking */
284 w->leaf = rt;
285 return 1;
286 }
287 WARN_ON(res == 0);
288 }
289 w->leaf = NULL;
290 return 0;
291 }
292
293 static void fib6_dump_end(struct netlink_callback *cb)
294 {
295 struct fib6_walker_t *w = (void*)cb->args[2];
296
297 if (w) {
298 if (cb->args[4]) {
299 cb->args[4] = 0;
300 fib6_walker_unlink(w);
301 }
302 cb->args[2] = 0;
303 kfree(w);
304 }
305 cb->done = (void*)cb->args[3];
306 cb->args[1] = 3;
307 }
308
309 static int fib6_dump_done(struct netlink_callback *cb)
310 {
311 fib6_dump_end(cb);
312 return cb->done ? cb->done(cb) : 0;
313 }
314
315 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
316 struct netlink_callback *cb)
317 {
318 struct fib6_walker_t *w;
319 int res;
320
321 w = (void *)cb->args[2];
322 w->root = &table->tb6_root;
323
324 if (cb->args[4] == 0) {
325 w->count = 0;
326 w->skip = 0;
327
328 read_lock_bh(&table->tb6_lock);
329 res = fib6_walk(w);
330 read_unlock_bh(&table->tb6_lock);
331 if (res > 0) {
332 cb->args[4] = 1;
333 cb->args[5] = w->root->fn_sernum;
334 }
335 } else {
336 if (cb->args[5] != w->root->fn_sernum) {
337 /* Begin at the root if the tree changed */
338 cb->args[5] = w->root->fn_sernum;
339 w->state = FWS_INIT;
340 w->node = w->root;
341 w->skip = w->count;
342 } else
343 w->skip = 0;
344
345 read_lock_bh(&table->tb6_lock);
346 res = fib6_walk_continue(w);
347 read_unlock_bh(&table->tb6_lock);
348 if (res <= 0) {
349 fib6_walker_unlink(w);
350 cb->args[4] = 0;
351 }
352 }
353
354 return res;
355 }
356
357 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
358 {
359 struct net *net = sock_net(skb->sk);
360 unsigned int h, s_h;
361 unsigned int e = 0, s_e;
362 struct rt6_rtnl_dump_arg arg;
363 struct fib6_walker_t *w;
364 struct fib6_table *tb;
365 struct hlist_head *head;
366 int res = 0;
367
368 s_h = cb->args[0];
369 s_e = cb->args[1];
370
371 w = (void *)cb->args[2];
372 if (!w) {
373 /* New dump:
374 *
375 * 1. hook callback destructor.
376 */
377 cb->args[3] = (long)cb->done;
378 cb->done = fib6_dump_done;
379
380 /*
381 * 2. allocate and initialize walker.
382 */
383 w = kzalloc(sizeof(*w), GFP_ATOMIC);
384 if (!w)
385 return -ENOMEM;
386 w->func = fib6_dump_node;
387 cb->args[2] = (long)w;
388 }
389
390 arg.skb = skb;
391 arg.cb = cb;
392 arg.net = net;
393 w->args = &arg;
394
395 rcu_read_lock();
396 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
397 e = 0;
398 head = &net->ipv6.fib_table_hash[h];
399 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
400 if (e < s_e)
401 goto next;
402 res = fib6_dump_table(tb, skb, cb);
403 if (res != 0)
404 goto out;
405 next:
406 e++;
407 }
408 }
409 out:
410 rcu_read_unlock();
411 cb->args[1] = e;
412 cb->args[0] = h;
413
414 res = res < 0 ? res : skb->len;
415 if (res <= 0)
416 fib6_dump_end(cb);
417 return res;
418 }
419
420 /*
421 * Routing Table
422 *
423 * return the appropriate node for a routing tree "add" operation
424 * by either creating and inserting or by returning an existing
425 * node.
426 */
427
428 static struct fib6_node *fib6_add_1(struct fib6_node *root,
429 struct in6_addr *addr, int plen,
430 int offset, int allow_create,
431 int replace_required)
432 {
433 struct fib6_node *fn, *in, *ln;
434 struct fib6_node *pn = NULL;
435 struct rt6key *key;
436 int bit;
437 __be32 dir = 0;
438 __u32 sernum = fib6_new_sernum();
439
440 RT6_TRACE("fib6_add_1\n");
441
442 /* insert node in tree */
443
444 fn = root;
445
446 do {
447 key = (struct rt6key *)((u8 *)fn->leaf + offset);
448
449 /*
450 * Prefix match
451 */
452 if (plen < fn->fn_bit ||
453 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
454 if (!allow_create) {
455 if (replace_required) {
456 pr_warn("Can't replace route, no match found\n");
457 return ERR_PTR(-ENOENT);
458 }
459 pr_warn("NLM_F_CREATE should be set when creating new route\n");
460 }
461 goto insert_above;
462 }
463
464 /*
465 * Exact match ?
466 */
467
468 if (plen == fn->fn_bit) {
469 /* clean up an intermediate node */
470 if (!(fn->fn_flags & RTN_RTINFO)) {
471 rt6_release(fn->leaf);
472 fn->leaf = NULL;
473 }
474
475 fn->fn_sernum = sernum;
476
477 return fn;
478 }
479
480 /*
481 * We have more bits to go
482 */
483
484 /* Try to walk down on tree. */
485 fn->fn_sernum = sernum;
486 dir = addr_bit_set(addr, fn->fn_bit);
487 pn = fn;
488 fn = dir ? fn->right: fn->left;
489 } while (fn);
490
491 if (!allow_create) {
492 /* We should not create new node because
493 * NLM_F_REPLACE was specified without NLM_F_CREATE
494 * I assume it is safe to require NLM_F_CREATE when
495 * REPLACE flag is used! Later we may want to remove the
496 * check for replace_required, because according
497 * to netlink specification, NLM_F_CREATE
498 * MUST be specified if new route is created.
499 * That would keep IPv6 consistent with IPv4
500 */
501 if (replace_required) {
502 pr_warn("Can't replace route, no match found\n");
503 return ERR_PTR(-ENOENT);
504 }
505 pr_warn("NLM_F_CREATE should be set when creating new route\n");
506 }
507 /*
508 * We walked to the bottom of tree.
509 * Create new leaf node without children.
510 */
511
512 ln = node_alloc();
513
514 if (!ln)
515 return ERR_PTR(-ENOMEM);
516 ln->fn_bit = plen;
517
518 ln->parent = pn;
519 ln->fn_sernum = sernum;
520
521 if (dir)
522 pn->right = ln;
523 else
524 pn->left = ln;
525
526 return ln;
527
528
529 insert_above:
530 /*
531 * split since we don't have a common prefix anymore or
532 * we have a less significant route.
533 * we've to insert an intermediate node on the list
534 * this new node will point to the one we need to create
535 * and the current
536 */
537
538 pn = fn->parent;
539
540 /* find 1st bit in difference between the 2 addrs.
541
542 See comment in __ipv6_addr_diff: bit may be an invalid value,
543 but if it is >= plen, the value is ignored in any case.
544 */
545
546 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
547
548 /*
549 * (intermediate)[in]
550 * / \
551 * (new leaf node)[ln] (old node)[fn]
552 */
553 if (plen > bit) {
554 in = node_alloc();
555 ln = node_alloc();
556
557 if (!in || !ln) {
558 if (in)
559 node_free(in);
560 if (ln)
561 node_free(ln);
562 return ERR_PTR(-ENOMEM);
563 }
564
565 /*
566 * new intermediate node.
567 * RTN_RTINFO will
568 * be off since that an address that chooses one of
569 * the branches would not match less specific routes
570 * in the other branch
571 */
572
573 in->fn_bit = bit;
574
575 in->parent = pn;
576 in->leaf = fn->leaf;
577 atomic_inc(&in->leaf->rt6i_ref);
578
579 in->fn_sernum = sernum;
580
581 /* update parent pointer */
582 if (dir)
583 pn->right = in;
584 else
585 pn->left = in;
586
587 ln->fn_bit = plen;
588
589 ln->parent = in;
590 fn->parent = in;
591
592 ln->fn_sernum = sernum;
593
594 if (addr_bit_set(addr, bit)) {
595 in->right = ln;
596 in->left = fn;
597 } else {
598 in->left = ln;
599 in->right = fn;
600 }
601 } else { /* plen <= bit */
602
603 /*
604 * (new leaf node)[ln]
605 * / \
606 * (old node)[fn] NULL
607 */
608
609 ln = node_alloc();
610
611 if (!ln)
612 return ERR_PTR(-ENOMEM);
613
614 ln->fn_bit = plen;
615
616 ln->parent = pn;
617
618 ln->fn_sernum = sernum;
619
620 if (dir)
621 pn->right = ln;
622 else
623 pn->left = ln;
624
625 if (addr_bit_set(&key->addr, plen))
626 ln->right = fn;
627 else
628 ln->left = fn;
629
630 fn->parent = ln;
631 }
632 return ln;
633 }
634
635 static inline bool rt6_qualify_for_ecmp(struct rt6_info *rt)
636 {
637 return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
638 RTF_GATEWAY;
639 }
640
641 /*
642 * Insert routing information in a node.
643 */
644
645 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
646 struct nl_info *info)
647 {
648 struct rt6_info *iter = NULL;
649 struct rt6_info **ins;
650 int replace = (info->nlh &&
651 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
652 int add = (!info->nlh ||
653 (info->nlh->nlmsg_flags & NLM_F_CREATE));
654 int found = 0;
655 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
656
657 ins = &fn->leaf;
658
659 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
660 /*
661 * Search for duplicates
662 */
663
664 if (iter->rt6i_metric == rt->rt6i_metric) {
665 /*
666 * Same priority level
667 */
668 if (info->nlh &&
669 (info->nlh->nlmsg_flags & NLM_F_EXCL))
670 return -EEXIST;
671 if (replace) {
672 found++;
673 break;
674 }
675
676 if (iter->dst.dev == rt->dst.dev &&
677 iter->rt6i_idev == rt->rt6i_idev &&
678 ipv6_addr_equal(&iter->rt6i_gateway,
679 &rt->rt6i_gateway)) {
680 if (rt->rt6i_nsiblings)
681 rt->rt6i_nsiblings = 0;
682 if (!(iter->rt6i_flags & RTF_EXPIRES))
683 return -EEXIST;
684 if (!(rt->rt6i_flags & RTF_EXPIRES))
685 rt6_clean_expires(iter);
686 else
687 rt6_set_expires(iter, rt->dst.expires);
688 return -EEXIST;
689 }
690 /* If we have the same destination and the same metric,
691 * but not the same gateway, then the route we try to
692 * add is sibling to this route, increment our counter
693 * of siblings, and later we will add our route to the
694 * list.
695 * Only static routes (which don't have flag
696 * RTF_EXPIRES) are used for ECMPv6.
697 *
698 * To avoid long list, we only had siblings if the
699 * route have a gateway.
700 */
701 if (rt_can_ecmp &&
702 rt6_qualify_for_ecmp(iter))
703 rt->rt6i_nsiblings++;
704 }
705
706 if (iter->rt6i_metric > rt->rt6i_metric)
707 break;
708
709 ins = &iter->dst.rt6_next;
710 }
711
712 /* Reset round-robin state, if necessary */
713 if (ins == &fn->leaf)
714 fn->rr_ptr = NULL;
715
716 /* Link this route to others same route. */
717 if (rt->rt6i_nsiblings) {
718 unsigned int rt6i_nsiblings;
719 struct rt6_info *sibling, *temp_sibling;
720
721 /* Find the first route that have the same metric */
722 sibling = fn->leaf;
723 while (sibling) {
724 if (sibling->rt6i_metric == rt->rt6i_metric &&
725 rt6_qualify_for_ecmp(sibling)) {
726 list_add_tail(&rt->rt6i_siblings,
727 &sibling->rt6i_siblings);
728 break;
729 }
730 sibling = sibling->dst.rt6_next;
731 }
732 /* For each sibling in the list, increment the counter of
733 * siblings. BUG() if counters does not match, list of siblings
734 * is broken!
735 */
736 rt6i_nsiblings = 0;
737 list_for_each_entry_safe(sibling, temp_sibling,
738 &rt->rt6i_siblings, rt6i_siblings) {
739 sibling->rt6i_nsiblings++;
740 BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
741 rt6i_nsiblings++;
742 }
743 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
744 }
745
746 /*
747 * insert node
748 */
749 if (!replace) {
750 if (!add)
751 pr_warn("NLM_F_CREATE should be set when creating new route\n");
752
753 add:
754 rt->dst.rt6_next = iter;
755 *ins = rt;
756 rt->rt6i_node = fn;
757 atomic_inc(&rt->rt6i_ref);
758 inet6_rt_notify(RTM_NEWROUTE, rt, info);
759 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
760
761 if (!(fn->fn_flags & RTN_RTINFO)) {
762 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
763 fn->fn_flags |= RTN_RTINFO;
764 }
765
766 } else {
767 if (!found) {
768 if (add)
769 goto add;
770 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
771 return -ENOENT;
772 }
773 *ins = rt;
774 rt->rt6i_node = fn;
775 rt->dst.rt6_next = iter->dst.rt6_next;
776 atomic_inc(&rt->rt6i_ref);
777 inet6_rt_notify(RTM_NEWROUTE, rt, info);
778 rt6_release(iter);
779 if (!(fn->fn_flags & RTN_RTINFO)) {
780 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
781 fn->fn_flags |= RTN_RTINFO;
782 }
783 }
784
785 return 0;
786 }
787
788 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
789 {
790 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
791 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
792 mod_timer(&net->ipv6.ip6_fib_timer,
793 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
794 }
795
796 void fib6_force_start_gc(struct net *net)
797 {
798 if (!timer_pending(&net->ipv6.ip6_fib_timer))
799 mod_timer(&net->ipv6.ip6_fib_timer,
800 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
801 }
802
803 /*
804 * Add routing information to the routing tree.
805 * <destination addr>/<source addr>
806 * with source addr info in sub-trees
807 */
808
809 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
810 {
811 struct fib6_node *fn, *pn = NULL;
812 int err = -ENOMEM;
813 int allow_create = 1;
814 int replace_required = 0;
815
816 if (info->nlh) {
817 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
818 allow_create = 0;
819 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
820 replace_required = 1;
821 }
822 if (!allow_create && !replace_required)
823 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
824
825 fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
826 offsetof(struct rt6_info, rt6i_dst), allow_create,
827 replace_required);
828 if (IS_ERR(fn)) {
829 err = PTR_ERR(fn);
830 fn = NULL;
831 goto out;
832 }
833
834 pn = fn;
835
836 #ifdef CONFIG_IPV6_SUBTREES
837 if (rt->rt6i_src.plen) {
838 struct fib6_node *sn;
839
840 if (!fn->subtree) {
841 struct fib6_node *sfn;
842
843 /*
844 * Create subtree.
845 *
846 * fn[main tree]
847 * |
848 * sfn[subtree root]
849 * \
850 * sn[new leaf node]
851 */
852
853 /* Create subtree root node */
854 sfn = node_alloc();
855 if (!sfn)
856 goto st_failure;
857
858 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
859 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
860 sfn->fn_flags = RTN_ROOT;
861 sfn->fn_sernum = fib6_new_sernum();
862
863 /* Now add the first leaf node to new subtree */
864
865 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
866 rt->rt6i_src.plen,
867 offsetof(struct rt6_info, rt6i_src),
868 allow_create, replace_required);
869
870 if (IS_ERR(sn)) {
871 /* If it is failed, discard just allocated
872 root, and then (in st_failure) stale node
873 in main tree.
874 */
875 node_free(sfn);
876 err = PTR_ERR(sn);
877 goto st_failure;
878 }
879
880 /* Now link new subtree to main tree */
881 sfn->parent = fn;
882 fn->subtree = sfn;
883 } else {
884 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
885 rt->rt6i_src.plen,
886 offsetof(struct rt6_info, rt6i_src),
887 allow_create, replace_required);
888
889 if (IS_ERR(sn)) {
890 err = PTR_ERR(sn);
891 goto st_failure;
892 }
893 }
894
895 if (!fn->leaf) {
896 fn->leaf = rt;
897 atomic_inc(&rt->rt6i_ref);
898 }
899 fn = sn;
900 }
901 #endif
902
903 err = fib6_add_rt2node(fn, rt, info);
904 if (!err) {
905 fib6_start_gc(info->nl_net, rt);
906 if (!(rt->rt6i_flags & RTF_CACHE))
907 fib6_prune_clones(info->nl_net, pn, rt);
908 }
909
910 out:
911 if (err) {
912 #ifdef CONFIG_IPV6_SUBTREES
913 /*
914 * If fib6_add_1 has cleared the old leaf pointer in the
915 * super-tree leaf node we have to find a new one for it.
916 */
917 if (pn != fn && pn->leaf == rt) {
918 pn->leaf = NULL;
919 atomic_dec(&rt->rt6i_ref);
920 }
921 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
922 pn->leaf = fib6_find_prefix(info->nl_net, pn);
923 #if RT6_DEBUG >= 2
924 if (!pn->leaf) {
925 WARN_ON(pn->leaf == NULL);
926 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
927 }
928 #endif
929 atomic_inc(&pn->leaf->rt6i_ref);
930 }
931 #endif
932 dst_free(&rt->dst);
933 }
934 return err;
935
936 #ifdef CONFIG_IPV6_SUBTREES
937 /* Subtree creation failed, probably main tree node
938 is orphan. If it is, shoot it.
939 */
940 st_failure:
941 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
942 fib6_repair_tree(info->nl_net, fn);
943 dst_free(&rt->dst);
944 return err;
945 #endif
946 }
947
948 /*
949 * Routing tree lookup
950 *
951 */
952
953 struct lookup_args {
954 int offset; /* key offset on rt6_info */
955 const struct in6_addr *addr; /* search key */
956 };
957
958 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
959 struct lookup_args *args)
960 {
961 struct fib6_node *fn;
962 __be32 dir;
963
964 if (unlikely(args->offset == 0))
965 return NULL;
966
967 /*
968 * Descend on a tree
969 */
970
971 fn = root;
972
973 for (;;) {
974 struct fib6_node *next;
975
976 dir = addr_bit_set(args->addr, fn->fn_bit);
977
978 next = dir ? fn->right : fn->left;
979
980 if (next) {
981 fn = next;
982 continue;
983 }
984 break;
985 }
986
987 while (fn) {
988 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
989 struct rt6key *key;
990
991 key = (struct rt6key *) ((u8 *) fn->leaf +
992 args->offset);
993
994 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
995 #ifdef CONFIG_IPV6_SUBTREES
996 if (fn->subtree) {
997 struct fib6_node *sfn;
998 sfn = fib6_lookup_1(fn->subtree,
999 args + 1);
1000 if (!sfn)
1001 goto backtrack;
1002 fn = sfn;
1003 }
1004 #endif
1005 if (fn->fn_flags & RTN_RTINFO)
1006 return fn;
1007 }
1008 }
1009 #ifdef CONFIG_IPV6_SUBTREES
1010 backtrack:
1011 #endif
1012 if (fn->fn_flags & RTN_ROOT)
1013 break;
1014
1015 fn = fn->parent;
1016 }
1017
1018 return NULL;
1019 }
1020
1021 struct fib6_node * fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1022 const struct in6_addr *saddr)
1023 {
1024 struct fib6_node *fn;
1025 struct lookup_args args[] = {
1026 {
1027 .offset = offsetof(struct rt6_info, rt6i_dst),
1028 .addr = daddr,
1029 },
1030 #ifdef CONFIG_IPV6_SUBTREES
1031 {
1032 .offset = offsetof(struct rt6_info, rt6i_src),
1033 .addr = saddr,
1034 },
1035 #endif
1036 {
1037 .offset = 0, /* sentinel */
1038 }
1039 };
1040
1041 fn = fib6_lookup_1(root, daddr ? args : args + 1);
1042 if (!fn || fn->fn_flags & RTN_TL_ROOT)
1043 fn = root;
1044
1045 return fn;
1046 }
1047
1048 /*
1049 * Get node with specified destination prefix (and source prefix,
1050 * if subtrees are used)
1051 */
1052
1053
1054 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
1055 const struct in6_addr *addr,
1056 int plen, int offset)
1057 {
1058 struct fib6_node *fn;
1059
1060 for (fn = root; fn ; ) {
1061 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1062
1063 /*
1064 * Prefix match
1065 */
1066 if (plen < fn->fn_bit ||
1067 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1068 return NULL;
1069
1070 if (plen == fn->fn_bit)
1071 return fn;
1072
1073 /*
1074 * We have more bits to go
1075 */
1076 if (addr_bit_set(addr, fn->fn_bit))
1077 fn = fn->right;
1078 else
1079 fn = fn->left;
1080 }
1081 return NULL;
1082 }
1083
1084 struct fib6_node * fib6_locate(struct fib6_node *root,
1085 const struct in6_addr *daddr, int dst_len,
1086 const struct in6_addr *saddr, int src_len)
1087 {
1088 struct fib6_node *fn;
1089
1090 fn = fib6_locate_1(root, daddr, dst_len,
1091 offsetof(struct rt6_info, rt6i_dst));
1092
1093 #ifdef CONFIG_IPV6_SUBTREES
1094 if (src_len) {
1095 WARN_ON(saddr == NULL);
1096 if (fn && fn->subtree)
1097 fn = fib6_locate_1(fn->subtree, saddr, src_len,
1098 offsetof(struct rt6_info, rt6i_src));
1099 }
1100 #endif
1101
1102 if (fn && fn->fn_flags & RTN_RTINFO)
1103 return fn;
1104
1105 return NULL;
1106 }
1107
1108
1109 /*
1110 * Deletion
1111 *
1112 */
1113
1114 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1115 {
1116 if (fn->fn_flags & RTN_ROOT)
1117 return net->ipv6.ip6_null_entry;
1118
1119 while (fn) {
1120 if (fn->left)
1121 return fn->left->leaf;
1122 if (fn->right)
1123 return fn->right->leaf;
1124
1125 fn = FIB6_SUBTREE(fn);
1126 }
1127 return NULL;
1128 }
1129
1130 /*
1131 * Called to trim the tree of intermediate nodes when possible. "fn"
1132 * is the node we want to try and remove.
1133 */
1134
1135 static struct fib6_node *fib6_repair_tree(struct net *net,
1136 struct fib6_node *fn)
1137 {
1138 int children;
1139 int nstate;
1140 struct fib6_node *child, *pn;
1141 struct fib6_walker_t *w;
1142 int iter = 0;
1143
1144 for (;;) {
1145 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1146 iter++;
1147
1148 WARN_ON(fn->fn_flags & RTN_RTINFO);
1149 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1150 WARN_ON(fn->leaf != NULL);
1151
1152 children = 0;
1153 child = NULL;
1154 if (fn->right) child = fn->right, children |= 1;
1155 if (fn->left) child = fn->left, children |= 2;
1156
1157 if (children == 3 || FIB6_SUBTREE(fn)
1158 #ifdef CONFIG_IPV6_SUBTREES
1159 /* Subtree root (i.e. fn) may have one child */
1160 || (children && fn->fn_flags & RTN_ROOT)
1161 #endif
1162 ) {
1163 fn->leaf = fib6_find_prefix(net, fn);
1164 #if RT6_DEBUG >= 2
1165 if (!fn->leaf) {
1166 WARN_ON(!fn->leaf);
1167 fn->leaf = net->ipv6.ip6_null_entry;
1168 }
1169 #endif
1170 atomic_inc(&fn->leaf->rt6i_ref);
1171 return fn->parent;
1172 }
1173
1174 pn = fn->parent;
1175 #ifdef CONFIG_IPV6_SUBTREES
1176 if (FIB6_SUBTREE(pn) == fn) {
1177 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1178 FIB6_SUBTREE(pn) = NULL;
1179 nstate = FWS_L;
1180 } else {
1181 WARN_ON(fn->fn_flags & RTN_ROOT);
1182 #endif
1183 if (pn->right == fn) pn->right = child;
1184 else if (pn->left == fn) pn->left = child;
1185 #if RT6_DEBUG >= 2
1186 else
1187 WARN_ON(1);
1188 #endif
1189 if (child)
1190 child->parent = pn;
1191 nstate = FWS_R;
1192 #ifdef CONFIG_IPV6_SUBTREES
1193 }
1194 #endif
1195
1196 read_lock(&fib6_walker_lock);
1197 FOR_WALKERS(w) {
1198 if (!child) {
1199 if (w->root == fn) {
1200 w->root = w->node = NULL;
1201 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1202 } else if (w->node == fn) {
1203 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1204 w->node = pn;
1205 w->state = nstate;
1206 }
1207 } else {
1208 if (w->root == fn) {
1209 w->root = child;
1210 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1211 }
1212 if (w->node == fn) {
1213 w->node = child;
1214 if (children&2) {
1215 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1216 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1217 } else {
1218 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1219 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1220 }
1221 }
1222 }
1223 }
1224 read_unlock(&fib6_walker_lock);
1225
1226 node_free(fn);
1227 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1228 return pn;
1229
1230 rt6_release(pn->leaf);
1231 pn->leaf = NULL;
1232 fn = pn;
1233 }
1234 }
1235
1236 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1237 struct nl_info *info)
1238 {
1239 struct fib6_walker_t *w;
1240 struct rt6_info *rt = *rtp;
1241 struct net *net = info->nl_net;
1242
1243 RT6_TRACE("fib6_del_route\n");
1244
1245 /* Unlink it */
1246 *rtp = rt->dst.rt6_next;
1247 rt->rt6i_node = NULL;
1248 net->ipv6.rt6_stats->fib_rt_entries--;
1249 net->ipv6.rt6_stats->fib_discarded_routes++;
1250
1251 /* Reset round-robin state, if necessary */
1252 if (fn->rr_ptr == rt)
1253 fn->rr_ptr = NULL;
1254
1255 /* Remove this entry from other siblings */
1256 if (rt->rt6i_nsiblings) {
1257 struct rt6_info *sibling, *next_sibling;
1258
1259 list_for_each_entry_safe(sibling, next_sibling,
1260 &rt->rt6i_siblings, rt6i_siblings)
1261 sibling->rt6i_nsiblings--;
1262 rt->rt6i_nsiblings = 0;
1263 list_del_init(&rt->rt6i_siblings);
1264 }
1265
1266 /* Adjust walkers */
1267 read_lock(&fib6_walker_lock);
1268 FOR_WALKERS(w) {
1269 if (w->state == FWS_C && w->leaf == rt) {
1270 RT6_TRACE("walker %p adjusted by delroute\n", w);
1271 w->leaf = rt->dst.rt6_next;
1272 if (!w->leaf)
1273 w->state = FWS_U;
1274 }
1275 }
1276 read_unlock(&fib6_walker_lock);
1277
1278 rt->dst.rt6_next = NULL;
1279
1280 /* If it was last route, expunge its radix tree node */
1281 if (!fn->leaf) {
1282 fn->fn_flags &= ~RTN_RTINFO;
1283 net->ipv6.rt6_stats->fib_route_nodes--;
1284 fn = fib6_repair_tree(net, fn);
1285 }
1286
1287 if (atomic_read(&rt->rt6i_ref) != 1) {
1288 /* This route is used as dummy address holder in some split
1289 * nodes. It is not leaked, but it still holds other resources,
1290 * which must be released in time. So, scan ascendant nodes
1291 * and replace dummy references to this route with references
1292 * to still alive ones.
1293 */
1294 while (fn) {
1295 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
1296 fn->leaf = fib6_find_prefix(net, fn);
1297 atomic_inc(&fn->leaf->rt6i_ref);
1298 rt6_release(rt);
1299 }
1300 fn = fn->parent;
1301 }
1302 /* No more references are possible at this point. */
1303 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1304 }
1305
1306 inet6_rt_notify(RTM_DELROUTE, rt, info);
1307 rt6_release(rt);
1308 }
1309
1310 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1311 {
1312 struct net *net = info->nl_net;
1313 struct fib6_node *fn = rt->rt6i_node;
1314 struct rt6_info **rtp;
1315
1316 #if RT6_DEBUG >= 2
1317 if (rt->dst.obsolete>0) {
1318 WARN_ON(fn != NULL);
1319 return -ENOENT;
1320 }
1321 #endif
1322 if (!fn || rt == net->ipv6.ip6_null_entry)
1323 return -ENOENT;
1324
1325 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1326
1327 if (!(rt->rt6i_flags & RTF_CACHE)) {
1328 struct fib6_node *pn = fn;
1329 #ifdef CONFIG_IPV6_SUBTREES
1330 /* clones of this route might be in another subtree */
1331 if (rt->rt6i_src.plen) {
1332 while (!(pn->fn_flags & RTN_ROOT))
1333 pn = pn->parent;
1334 pn = pn->parent;
1335 }
1336 #endif
1337 fib6_prune_clones(info->nl_net, pn, rt);
1338 }
1339
1340 /*
1341 * Walk the leaf entries looking for ourself
1342 */
1343
1344 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1345 if (*rtp == rt) {
1346 fib6_del_route(fn, rtp, info);
1347 return 0;
1348 }
1349 }
1350 return -ENOENT;
1351 }
1352
1353 /*
1354 * Tree traversal function.
1355 *
1356 * Certainly, it is not interrupt safe.
1357 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1358 * It means, that we can modify tree during walking
1359 * and use this function for garbage collection, clone pruning,
1360 * cleaning tree when a device goes down etc. etc.
1361 *
1362 * It guarantees that every node will be traversed,
1363 * and that it will be traversed only once.
1364 *
1365 * Callback function w->func may return:
1366 * 0 -> continue walking.
1367 * positive value -> walking is suspended (used by tree dumps,
1368 * and probably by gc, if it will be split to several slices)
1369 * negative value -> terminate walking.
1370 *
1371 * The function itself returns:
1372 * 0 -> walk is complete.
1373 * >0 -> walk is incomplete (i.e. suspended)
1374 * <0 -> walk is terminated by an error.
1375 */
1376
1377 static int fib6_walk_continue(struct fib6_walker_t *w)
1378 {
1379 struct fib6_node *fn, *pn;
1380
1381 for (;;) {
1382 fn = w->node;
1383 if (!fn)
1384 return 0;
1385
1386 if (w->prune && fn != w->root &&
1387 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1388 w->state = FWS_C;
1389 w->leaf = fn->leaf;
1390 }
1391 switch (w->state) {
1392 #ifdef CONFIG_IPV6_SUBTREES
1393 case FWS_S:
1394 if (FIB6_SUBTREE(fn)) {
1395 w->node = FIB6_SUBTREE(fn);
1396 continue;
1397 }
1398 w->state = FWS_L;
1399 #endif
1400 case FWS_L:
1401 if (fn->left) {
1402 w->node = fn->left;
1403 w->state = FWS_INIT;
1404 continue;
1405 }
1406 w->state = FWS_R;
1407 case FWS_R:
1408 if (fn->right) {
1409 w->node = fn->right;
1410 w->state = FWS_INIT;
1411 continue;
1412 }
1413 w->state = FWS_C;
1414 w->leaf = fn->leaf;
1415 case FWS_C:
1416 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1417 int err;
1418
1419 if (w->skip) {
1420 w->skip--;
1421 continue;
1422 }
1423
1424 err = w->func(w);
1425 if (err)
1426 return err;
1427
1428 w->count++;
1429 continue;
1430 }
1431 w->state = FWS_U;
1432 case FWS_U:
1433 if (fn == w->root)
1434 return 0;
1435 pn = fn->parent;
1436 w->node = pn;
1437 #ifdef CONFIG_IPV6_SUBTREES
1438 if (FIB6_SUBTREE(pn) == fn) {
1439 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1440 w->state = FWS_L;
1441 continue;
1442 }
1443 #endif
1444 if (pn->left == fn) {
1445 w->state = FWS_R;
1446 continue;
1447 }
1448 if (pn->right == fn) {
1449 w->state = FWS_C;
1450 w->leaf = w->node->leaf;
1451 continue;
1452 }
1453 #if RT6_DEBUG >= 2
1454 WARN_ON(1);
1455 #endif
1456 }
1457 }
1458 }
1459
1460 static int fib6_walk(struct fib6_walker_t *w)
1461 {
1462 int res;
1463
1464 w->state = FWS_INIT;
1465 w->node = w->root;
1466
1467 fib6_walker_link(w);
1468 res = fib6_walk_continue(w);
1469 if (res <= 0)
1470 fib6_walker_unlink(w);
1471 return res;
1472 }
1473
1474 static int fib6_clean_node(struct fib6_walker_t *w)
1475 {
1476 int res;
1477 struct rt6_info *rt;
1478 struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1479 struct nl_info info = {
1480 .nl_net = c->net,
1481 };
1482
1483 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1484 res = c->func(rt, c->arg);
1485 if (res < 0) {
1486 w->leaf = rt;
1487 res = fib6_del(rt, &info);
1488 if (res) {
1489 #if RT6_DEBUG >= 2
1490 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1491 __func__, rt, rt->rt6i_node, res);
1492 #endif
1493 continue;
1494 }
1495 return 0;
1496 }
1497 WARN_ON(res != 0);
1498 }
1499 w->leaf = rt;
1500 return 0;
1501 }
1502
1503 /*
1504 * Convenient frontend to tree walker.
1505 *
1506 * func is called on each route.
1507 * It may return -1 -> delete this route.
1508 * 0 -> continue walking
1509 *
1510 * prune==1 -> only immediate children of node (certainly,
1511 * ignoring pure split nodes) will be scanned.
1512 */
1513
1514 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1515 int (*func)(struct rt6_info *, void *arg),
1516 int prune, void *arg)
1517 {
1518 struct fib6_cleaner_t c;
1519
1520 c.w.root = root;
1521 c.w.func = fib6_clean_node;
1522 c.w.prune = prune;
1523 c.w.count = 0;
1524 c.w.skip = 0;
1525 c.func = func;
1526 c.arg = arg;
1527 c.net = net;
1528
1529 fib6_walk(&c.w);
1530 }
1531
1532 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1533 void *arg)
1534 {
1535 struct fib6_table *table;
1536 struct hlist_head *head;
1537 unsigned int h;
1538
1539 rcu_read_lock();
1540 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1541 head = &net->ipv6.fib_table_hash[h];
1542 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1543 write_lock_bh(&table->tb6_lock);
1544 fib6_clean_tree(net, &table->tb6_root,
1545 func, 0, arg);
1546 write_unlock_bh(&table->tb6_lock);
1547 }
1548 }
1549 rcu_read_unlock();
1550 }
1551
1552 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1553 {
1554 if (rt->rt6i_flags & RTF_CACHE) {
1555 RT6_TRACE("pruning clone %p\n", rt);
1556 return -1;
1557 }
1558
1559 return 0;
1560 }
1561
1562 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1563 struct rt6_info *rt)
1564 {
1565 fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1566 }
1567
1568 /*
1569 * Garbage collection
1570 */
1571
1572 static struct fib6_gc_args
1573 {
1574 int timeout;
1575 int more;
1576 } gc_args;
1577
1578 static int fib6_age(struct rt6_info *rt, void *arg)
1579 {
1580 unsigned long now = jiffies;
1581
1582 /*
1583 * check addrconf expiration here.
1584 * Routes are expired even if they are in use.
1585 *
1586 * Also age clones. Note, that clones are aged out
1587 * only if they are not in use now.
1588 */
1589
1590 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1591 if (time_after(now, rt->dst.expires)) {
1592 RT6_TRACE("expiring %p\n", rt);
1593 return -1;
1594 }
1595 gc_args.more++;
1596 } else if (rt->rt6i_flags & RTF_CACHE) {
1597 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1598 time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1599 RT6_TRACE("aging clone %p\n", rt);
1600 return -1;
1601 } else if (rt->rt6i_flags & RTF_GATEWAY) {
1602 struct neighbour *neigh;
1603 __u8 neigh_flags = 0;
1604
1605 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1606 if (neigh) {
1607 neigh_flags = neigh->flags;
1608 neigh_release(neigh);
1609 }
1610 if (!(neigh_flags & NTF_ROUTER)) {
1611 RT6_TRACE("purging route %p via non-router but gateway\n",
1612 rt);
1613 return -1;
1614 }
1615 }
1616 gc_args.more++;
1617 }
1618
1619 return 0;
1620 }
1621
1622 static DEFINE_SPINLOCK(fib6_gc_lock);
1623
1624 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1625 {
1626 unsigned long now;
1627
1628 if (force) {
1629 spin_lock_bh(&fib6_gc_lock);
1630 } else if (!spin_trylock_bh(&fib6_gc_lock)) {
1631 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1632 return;
1633 }
1634 gc_args.timeout = expires ? (int)expires :
1635 net->ipv6.sysctl.ip6_rt_gc_interval;
1636
1637 gc_args.more = icmp6_dst_gc();
1638
1639 fib6_clean_all(net, fib6_age, NULL);
1640 now = jiffies;
1641 net->ipv6.ip6_rt_last_gc = now;
1642
1643 if (gc_args.more)
1644 mod_timer(&net->ipv6.ip6_fib_timer,
1645 round_jiffies(now
1646 + net->ipv6.sysctl.ip6_rt_gc_interval));
1647 else
1648 del_timer(&net->ipv6.ip6_fib_timer);
1649 spin_unlock_bh(&fib6_gc_lock);
1650 }
1651
1652 static void fib6_gc_timer_cb(unsigned long arg)
1653 {
1654 fib6_run_gc(0, (struct net *)arg, true);
1655 }
1656
1657 static int __net_init fib6_net_init(struct net *net)
1658 {
1659 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1660
1661 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1662
1663 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1664 if (!net->ipv6.rt6_stats)
1665 goto out_timer;
1666
1667 /* Avoid false sharing : Use at least a full cache line */
1668 size = max_t(size_t, size, L1_CACHE_BYTES);
1669
1670 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1671 if (!net->ipv6.fib_table_hash)
1672 goto out_rt6_stats;
1673
1674 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1675 GFP_KERNEL);
1676 if (!net->ipv6.fib6_main_tbl)
1677 goto out_fib_table_hash;
1678
1679 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1680 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1681 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1682 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1683 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1684
1685 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1686 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1687 GFP_KERNEL);
1688 if (!net->ipv6.fib6_local_tbl)
1689 goto out_fib6_main_tbl;
1690 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1691 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1692 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1693 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1694 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1695 #endif
1696 fib6_tables_init(net);
1697
1698 return 0;
1699
1700 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1701 out_fib6_main_tbl:
1702 kfree(net->ipv6.fib6_main_tbl);
1703 #endif
1704 out_fib_table_hash:
1705 kfree(net->ipv6.fib_table_hash);
1706 out_rt6_stats:
1707 kfree(net->ipv6.rt6_stats);
1708 out_timer:
1709 return -ENOMEM;
1710 }
1711
1712 static void fib6_net_exit(struct net *net)
1713 {
1714 rt6_ifdown(net, NULL);
1715 del_timer_sync(&net->ipv6.ip6_fib_timer);
1716
1717 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1718 inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1719 kfree(net->ipv6.fib6_local_tbl);
1720 #endif
1721 inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1722 kfree(net->ipv6.fib6_main_tbl);
1723 kfree(net->ipv6.fib_table_hash);
1724 kfree(net->ipv6.rt6_stats);
1725 }
1726
1727 static struct pernet_operations fib6_net_ops = {
1728 .init = fib6_net_init,
1729 .exit = fib6_net_exit,
1730 };
1731
1732 int __init fib6_init(void)
1733 {
1734 int ret = -ENOMEM;
1735
1736 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1737 sizeof(struct fib6_node),
1738 0, SLAB_HWCACHE_ALIGN,
1739 NULL);
1740 if (!fib6_node_kmem)
1741 goto out;
1742
1743 ret = register_pernet_subsys(&fib6_net_ops);
1744 if (ret)
1745 goto out_kmem_cache_create;
1746
1747 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1748 NULL);
1749 if (ret)
1750 goto out_unregister_subsys;
1751 out:
1752 return ret;
1753
1754 out_unregister_subsys:
1755 unregister_pernet_subsys(&fib6_net_ops);
1756 out_kmem_cache_create:
1757 kmem_cache_destroy(fib6_node_kmem);
1758 goto out;
1759 }
1760
1761 void fib6_gc_cleanup(void)
1762 {
1763 unregister_pernet_subsys(&fib6_net_ops);
1764 kmem_cache_destroy(fib6_node_kmem);
1765 }
1766
1767 #ifdef CONFIG_PROC_FS
1768
1769 struct ipv6_route_iter {
1770 struct seq_net_private p;
1771 struct fib6_walker_t w;
1772 loff_t skip;
1773 struct fib6_table *tbl;
1774 __u32 sernum;
1775 };
1776
1777 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
1778 {
1779 struct rt6_info *rt = v;
1780 struct ipv6_route_iter *iter = seq->private;
1781
1782 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
1783
1784 #ifdef CONFIG_IPV6_SUBTREES
1785 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
1786 #else
1787 seq_puts(seq, "00000000000000000000000000000000 00 ");
1788 #endif
1789 if (rt->rt6i_flags & RTF_GATEWAY)
1790 seq_printf(seq, "%pi6", &rt->rt6i_gateway);
1791 else
1792 seq_puts(seq, "00000000000000000000000000000000");
1793
1794 seq_printf(seq, " %08x %08x %08x %08x %8s\n",
1795 rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
1796 rt->dst.__use, rt->rt6i_flags,
1797 rt->dst.dev ? rt->dst.dev->name : "");
1798 iter->w.leaf = NULL;
1799 return 0;
1800 }
1801
1802 static int ipv6_route_yield(struct fib6_walker_t *w)
1803 {
1804 struct ipv6_route_iter *iter = w->args;
1805
1806 if (!iter->skip)
1807 return 1;
1808
1809 do {
1810 iter->w.leaf = iter->w.leaf->dst.rt6_next;
1811 iter->skip--;
1812 if (!iter->skip && iter->w.leaf)
1813 return 1;
1814 } while (iter->w.leaf);
1815
1816 return 0;
1817 }
1818
1819 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter)
1820 {
1821 memset(&iter->w, 0, sizeof(iter->w));
1822 iter->w.func = ipv6_route_yield;
1823 iter->w.root = &iter->tbl->tb6_root;
1824 iter->w.state = FWS_INIT;
1825 iter->w.node = iter->w.root;
1826 iter->w.args = iter;
1827 iter->sernum = iter->w.root->fn_sernum;
1828 INIT_LIST_HEAD(&iter->w.lh);
1829 fib6_walker_link(&iter->w);
1830 }
1831
1832 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
1833 struct net *net)
1834 {
1835 unsigned int h;
1836 struct hlist_node *node;
1837
1838 if (tbl) {
1839 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
1840 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
1841 } else {
1842 h = 0;
1843 node = NULL;
1844 }
1845
1846 while (!node && h < FIB6_TABLE_HASHSZ) {
1847 node = rcu_dereference_bh(
1848 hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
1849 }
1850 return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
1851 }
1852
1853 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
1854 {
1855 if (iter->sernum != iter->w.root->fn_sernum) {
1856 iter->sernum = iter->w.root->fn_sernum;
1857 iter->w.state = FWS_INIT;
1858 iter->w.node = iter->w.root;
1859 WARN_ON(iter->w.skip);
1860 iter->w.skip = iter->w.count;
1861 }
1862 }
1863
1864 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1865 {
1866 int r;
1867 struct rt6_info *n;
1868 struct net *net = seq_file_net(seq);
1869 struct ipv6_route_iter *iter = seq->private;
1870
1871 if (!v)
1872 goto iter_table;
1873
1874 n = ((struct rt6_info *)v)->dst.rt6_next;
1875 if (n) {
1876 ++*pos;
1877 return n;
1878 }
1879
1880 iter_table:
1881 ipv6_route_check_sernum(iter);
1882 read_lock(&iter->tbl->tb6_lock);
1883 r = fib6_walk_continue(&iter->w);
1884 read_unlock(&iter->tbl->tb6_lock);
1885 if (r > 0) {
1886 if (v)
1887 ++*pos;
1888 return iter->w.leaf;
1889 } else if (r < 0) {
1890 fib6_walker_unlink(&iter->w);
1891 return NULL;
1892 }
1893 fib6_walker_unlink(&iter->w);
1894
1895 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
1896 if (!iter->tbl)
1897 return NULL;
1898
1899 ipv6_route_seq_setup_walk(iter);
1900 goto iter_table;
1901 }
1902
1903 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
1904 __acquires(RCU_BH)
1905 {
1906 struct net *net = seq_file_net(seq);
1907 struct ipv6_route_iter *iter = seq->private;
1908
1909 rcu_read_lock_bh();
1910 iter->tbl = ipv6_route_seq_next_table(NULL, net);
1911 iter->skip = *pos;
1912
1913 if (iter->tbl) {
1914 ipv6_route_seq_setup_walk(iter);
1915 return ipv6_route_seq_next(seq, NULL, pos);
1916 } else {
1917 return NULL;
1918 }
1919 }
1920
1921 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
1922 {
1923 struct fib6_walker_t *w = &iter->w;
1924 return w->node && !(w->state == FWS_U && w->node == w->root);
1925 }
1926
1927 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
1928 __releases(RCU_BH)
1929 {
1930 struct ipv6_route_iter *iter = seq->private;
1931
1932 if (ipv6_route_iter_active(iter))
1933 fib6_walker_unlink(&iter->w);
1934
1935 rcu_read_unlock_bh();
1936 }
1937
1938 static const struct seq_operations ipv6_route_seq_ops = {
1939 .start = ipv6_route_seq_start,
1940 .next = ipv6_route_seq_next,
1941 .stop = ipv6_route_seq_stop,
1942 .show = ipv6_route_seq_show
1943 };
1944
1945 int ipv6_route_open(struct inode *inode, struct file *file)
1946 {
1947 return seq_open_net(inode, file, &ipv6_route_seq_ops,
1948 sizeof(struct ipv6_route_iter));
1949 }
1950
1951 #endif /* CONFIG_PROC_FS */