xref: /linux/net/ipv6/ip6_fib.c (revision c0e297dc61f8d4453e07afbea1fa8d0e67cd4a34)
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 
36 #include <net/ip6_fib.h>
37 #include <net/ip6_route.h>
38 
39 #define RT6_DEBUG 2
40 
41 #if RT6_DEBUG >= 3
42 #define RT6_TRACE(x...) pr_debug(x)
43 #else
44 #define RT6_TRACE(x...) do { ; } while (0)
45 #endif
46 
47 static struct kmem_cache *fib6_node_kmem __read_mostly;
48 
49 struct fib6_cleaner {
50 	struct fib6_walker w;
51 	struct net *net;
52 	int (*func)(struct rt6_info *, void *arg);
53 	int sernum;
54 	void *arg;
55 };
56 
57 static DEFINE_RWLOCK(fib6_walker_lock);
58 
59 #ifdef CONFIG_IPV6_SUBTREES
60 #define FWS_INIT FWS_S
61 #else
62 #define FWS_INIT FWS_L
63 #endif
64 
65 static void fib6_prune_clones(struct net *net, struct fib6_node *fn);
66 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
67 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
68 static int fib6_walk(struct fib6_walker *w);
69 static int fib6_walk_continue(struct fib6_walker *w);
70 
71 /*
72  *	A routing update causes an increase of the serial number on the
73  *	affected subtree. This allows for cached routes to be asynchronously
74  *	tested when modifications are made to the destination cache as a
75  *	result of redirects, path MTU changes, etc.
76  */
77 
78 static void fib6_gc_timer_cb(unsigned long arg);
79 
80 static LIST_HEAD(fib6_walkers);
81 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
82 
83 static void fib6_walker_link(struct fib6_walker *w)
84 {
85 	write_lock_bh(&fib6_walker_lock);
86 	list_add(&w->lh, &fib6_walkers);
87 	write_unlock_bh(&fib6_walker_lock);
88 }
89 
90 static void fib6_walker_unlink(struct fib6_walker *w)
91 {
92 	write_lock_bh(&fib6_walker_lock);
93 	list_del(&w->lh);
94 	write_unlock_bh(&fib6_walker_lock);
95 }
96 
97 static int fib6_new_sernum(struct net *net)
98 {
99 	int new, old;
100 
101 	do {
102 		old = atomic_read(&net->ipv6.fib6_sernum);
103 		new = old < INT_MAX ? old + 1 : 1;
104 	} while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
105 				old, new) != old);
106 	return new;
107 }
108 
109 enum {
110 	FIB6_NO_SERNUM_CHANGE = 0,
111 };
112 
113 /*
114  *	Auxiliary address test functions for the radix tree.
115  *
116  *	These assume a 32bit processor (although it will work on
117  *	64bit processors)
118  */
119 
120 /*
121  *	test bit
122  */
123 #if defined(__LITTLE_ENDIAN)
124 # define BITOP_BE32_SWIZZLE	(0x1F & ~7)
125 #else
126 # define BITOP_BE32_SWIZZLE	0
127 #endif
128 
129 static __be32 addr_bit_set(const void *token, int fn_bit)
130 {
131 	const __be32 *addr = token;
132 	/*
133 	 * Here,
134 	 *	1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
135 	 * is optimized version of
136 	 *	htonl(1 << ((~fn_bit)&0x1F))
137 	 * See include/asm-generic/bitops/le.h.
138 	 */
139 	return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
140 	       addr[fn_bit >> 5];
141 }
142 
143 static struct fib6_node *node_alloc(void)
144 {
145 	struct fib6_node *fn;
146 
147 	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
148 
149 	return fn;
150 }
151 
152 static void node_free(struct fib6_node *fn)
153 {
154 	kmem_cache_free(fib6_node_kmem, fn);
155 }
156 
157 static void rt6_free_pcpu(struct rt6_info *non_pcpu_rt)
158 {
159 	int cpu;
160 
161 	if (!non_pcpu_rt->rt6i_pcpu)
162 		return;
163 
164 	for_each_possible_cpu(cpu) {
165 		struct rt6_info **ppcpu_rt;
166 		struct rt6_info *pcpu_rt;
167 
168 		ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
169 		pcpu_rt = *ppcpu_rt;
170 		if (pcpu_rt) {
171 			dst_free(&pcpu_rt->dst);
172 			*ppcpu_rt = NULL;
173 		}
174 	}
175 
176 	non_pcpu_rt->rt6i_pcpu = NULL;
177 }
178 
179 static void rt6_release(struct rt6_info *rt)
180 {
181 	if (atomic_dec_and_test(&rt->rt6i_ref)) {
182 		rt6_free_pcpu(rt);
183 		dst_free(&rt->dst);
184 	}
185 }
186 
187 static void fib6_link_table(struct net *net, struct fib6_table *tb)
188 {
189 	unsigned int h;
190 
191 	/*
192 	 * Initialize table lock at a single place to give lockdep a key,
193 	 * tables aren't visible prior to being linked to the list.
194 	 */
195 	rwlock_init(&tb->tb6_lock);
196 
197 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
198 
199 	/*
200 	 * No protection necessary, this is the only list mutatation
201 	 * operation, tables never disappear once they exist.
202 	 */
203 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
204 }
205 
206 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
207 
208 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
209 {
210 	struct fib6_table *table;
211 
212 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
213 	if (table) {
214 		table->tb6_id = id;
215 		table->tb6_root.leaf = net->ipv6.ip6_null_entry;
216 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
217 		inet_peer_base_init(&table->tb6_peers);
218 	}
219 
220 	return table;
221 }
222 
223 struct fib6_table *fib6_new_table(struct net *net, u32 id)
224 {
225 	struct fib6_table *tb;
226 
227 	if (id == 0)
228 		id = RT6_TABLE_MAIN;
229 	tb = fib6_get_table(net, id);
230 	if (tb)
231 		return tb;
232 
233 	tb = fib6_alloc_table(net, id);
234 	if (tb)
235 		fib6_link_table(net, tb);
236 
237 	return tb;
238 }
239 
240 struct fib6_table *fib6_get_table(struct net *net, u32 id)
241 {
242 	struct fib6_table *tb;
243 	struct hlist_head *head;
244 	unsigned int h;
245 
246 	if (id == 0)
247 		id = RT6_TABLE_MAIN;
248 	h = id & (FIB6_TABLE_HASHSZ - 1);
249 	rcu_read_lock();
250 	head = &net->ipv6.fib_table_hash[h];
251 	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
252 		if (tb->tb6_id == id) {
253 			rcu_read_unlock();
254 			return tb;
255 		}
256 	}
257 	rcu_read_unlock();
258 
259 	return NULL;
260 }
261 
262 static void __net_init fib6_tables_init(struct net *net)
263 {
264 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
265 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
266 }
267 #else
268 
269 struct fib6_table *fib6_new_table(struct net *net, u32 id)
270 {
271 	return fib6_get_table(net, id);
272 }
273 
274 struct fib6_table *fib6_get_table(struct net *net, u32 id)
275 {
276 	  return net->ipv6.fib6_main_tbl;
277 }
278 
279 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
280 				   int flags, pol_lookup_t lookup)
281 {
282 	return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
283 }
284 
285 static void __net_init fib6_tables_init(struct net *net)
286 {
287 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
288 }
289 
290 #endif
291 
292 static int fib6_dump_node(struct fib6_walker *w)
293 {
294 	int res;
295 	struct rt6_info *rt;
296 
297 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
298 		res = rt6_dump_route(rt, w->args);
299 		if (res < 0) {
300 			/* Frame is full, suspend walking */
301 			w->leaf = rt;
302 			return 1;
303 		}
304 	}
305 	w->leaf = NULL;
306 	return 0;
307 }
308 
309 static void fib6_dump_end(struct netlink_callback *cb)
310 {
311 	struct fib6_walker *w = (void *)cb->args[2];
312 
313 	if (w) {
314 		if (cb->args[4]) {
315 			cb->args[4] = 0;
316 			fib6_walker_unlink(w);
317 		}
318 		cb->args[2] = 0;
319 		kfree(w);
320 	}
321 	cb->done = (void *)cb->args[3];
322 	cb->args[1] = 3;
323 }
324 
325 static int fib6_dump_done(struct netlink_callback *cb)
326 {
327 	fib6_dump_end(cb);
328 	return cb->done ? cb->done(cb) : 0;
329 }
330 
331 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
332 			   struct netlink_callback *cb)
333 {
334 	struct fib6_walker *w;
335 	int res;
336 
337 	w = (void *)cb->args[2];
338 	w->root = &table->tb6_root;
339 
340 	if (cb->args[4] == 0) {
341 		w->count = 0;
342 		w->skip = 0;
343 
344 		read_lock_bh(&table->tb6_lock);
345 		res = fib6_walk(w);
346 		read_unlock_bh(&table->tb6_lock);
347 		if (res > 0) {
348 			cb->args[4] = 1;
349 			cb->args[5] = w->root->fn_sernum;
350 		}
351 	} else {
352 		if (cb->args[5] != w->root->fn_sernum) {
353 			/* Begin at the root if the tree changed */
354 			cb->args[5] = w->root->fn_sernum;
355 			w->state = FWS_INIT;
356 			w->node = w->root;
357 			w->skip = w->count;
358 		} else
359 			w->skip = 0;
360 
361 		read_lock_bh(&table->tb6_lock);
362 		res = fib6_walk_continue(w);
363 		read_unlock_bh(&table->tb6_lock);
364 		if (res <= 0) {
365 			fib6_walker_unlink(w);
366 			cb->args[4] = 0;
367 		}
368 	}
369 
370 	return res;
371 }
372 
373 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
374 {
375 	struct net *net = sock_net(skb->sk);
376 	unsigned int h, s_h;
377 	unsigned int e = 0, s_e;
378 	struct rt6_rtnl_dump_arg arg;
379 	struct fib6_walker *w;
380 	struct fib6_table *tb;
381 	struct hlist_head *head;
382 	int res = 0;
383 
384 	s_h = cb->args[0];
385 	s_e = cb->args[1];
386 
387 	w = (void *)cb->args[2];
388 	if (!w) {
389 		/* New dump:
390 		 *
391 		 * 1. hook callback destructor.
392 		 */
393 		cb->args[3] = (long)cb->done;
394 		cb->done = fib6_dump_done;
395 
396 		/*
397 		 * 2. allocate and initialize walker.
398 		 */
399 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
400 		if (!w)
401 			return -ENOMEM;
402 		w->func = fib6_dump_node;
403 		cb->args[2] = (long)w;
404 	}
405 
406 	arg.skb = skb;
407 	arg.cb = cb;
408 	arg.net = net;
409 	w->args = &arg;
410 
411 	rcu_read_lock();
412 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
413 		e = 0;
414 		head = &net->ipv6.fib_table_hash[h];
415 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
416 			if (e < s_e)
417 				goto next;
418 			res = fib6_dump_table(tb, skb, cb);
419 			if (res != 0)
420 				goto out;
421 next:
422 			e++;
423 		}
424 	}
425 out:
426 	rcu_read_unlock();
427 	cb->args[1] = e;
428 	cb->args[0] = h;
429 
430 	res = res < 0 ? res : skb->len;
431 	if (res <= 0)
432 		fib6_dump_end(cb);
433 	return res;
434 }
435 
436 /*
437  *	Routing Table
438  *
439  *	return the appropriate node for a routing tree "add" operation
440  *	by either creating and inserting or by returning an existing
441  *	node.
442  */
443 
444 static struct fib6_node *fib6_add_1(struct fib6_node *root,
445 				     struct in6_addr *addr, int plen,
446 				     int offset, int allow_create,
447 				     int replace_required, int sernum)
448 {
449 	struct fib6_node *fn, *in, *ln;
450 	struct fib6_node *pn = NULL;
451 	struct rt6key *key;
452 	int	bit;
453 	__be32	dir = 0;
454 
455 	RT6_TRACE("fib6_add_1\n");
456 
457 	/* insert node in tree */
458 
459 	fn = root;
460 
461 	do {
462 		key = (struct rt6key *)((u8 *)fn->leaf + offset);
463 
464 		/*
465 		 *	Prefix match
466 		 */
467 		if (plen < fn->fn_bit ||
468 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
469 			if (!allow_create) {
470 				if (replace_required) {
471 					pr_warn("Can't replace route, no match found\n");
472 					return ERR_PTR(-ENOENT);
473 				}
474 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
475 			}
476 			goto insert_above;
477 		}
478 
479 		/*
480 		 *	Exact match ?
481 		 */
482 
483 		if (plen == fn->fn_bit) {
484 			/* clean up an intermediate node */
485 			if (!(fn->fn_flags & RTN_RTINFO)) {
486 				rt6_release(fn->leaf);
487 				fn->leaf = NULL;
488 			}
489 
490 			fn->fn_sernum = sernum;
491 
492 			return fn;
493 		}
494 
495 		/*
496 		 *	We have more bits to go
497 		 */
498 
499 		/* Try to walk down on tree. */
500 		fn->fn_sernum = sernum;
501 		dir = addr_bit_set(addr, fn->fn_bit);
502 		pn = fn;
503 		fn = dir ? fn->right : fn->left;
504 	} while (fn);
505 
506 	if (!allow_create) {
507 		/* We should not create new node because
508 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
509 		 * I assume it is safe to require NLM_F_CREATE when
510 		 * REPLACE flag is used! Later we may want to remove the
511 		 * check for replace_required, because according
512 		 * to netlink specification, NLM_F_CREATE
513 		 * MUST be specified if new route is created.
514 		 * That would keep IPv6 consistent with IPv4
515 		 */
516 		if (replace_required) {
517 			pr_warn("Can't replace route, no match found\n");
518 			return ERR_PTR(-ENOENT);
519 		}
520 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
521 	}
522 	/*
523 	 *	We walked to the bottom of tree.
524 	 *	Create new leaf node without children.
525 	 */
526 
527 	ln = node_alloc();
528 
529 	if (!ln)
530 		return ERR_PTR(-ENOMEM);
531 	ln->fn_bit = plen;
532 
533 	ln->parent = pn;
534 	ln->fn_sernum = sernum;
535 
536 	if (dir)
537 		pn->right = ln;
538 	else
539 		pn->left  = ln;
540 
541 	return ln;
542 
543 
544 insert_above:
545 	/*
546 	 * split since we don't have a common prefix anymore or
547 	 * we have a less significant route.
548 	 * we've to insert an intermediate node on the list
549 	 * this new node will point to the one we need to create
550 	 * and the current
551 	 */
552 
553 	pn = fn->parent;
554 
555 	/* find 1st bit in difference between the 2 addrs.
556 
557 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
558 	   but if it is >= plen, the value is ignored in any case.
559 	 */
560 
561 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
562 
563 	/*
564 	 *		(intermediate)[in]
565 	 *	          /	   \
566 	 *	(new leaf node)[ln] (old node)[fn]
567 	 */
568 	if (plen > bit) {
569 		in = node_alloc();
570 		ln = node_alloc();
571 
572 		if (!in || !ln) {
573 			if (in)
574 				node_free(in);
575 			if (ln)
576 				node_free(ln);
577 			return ERR_PTR(-ENOMEM);
578 		}
579 
580 		/*
581 		 * new intermediate node.
582 		 * RTN_RTINFO will
583 		 * be off since that an address that chooses one of
584 		 * the branches would not match less specific routes
585 		 * in the other branch
586 		 */
587 
588 		in->fn_bit = bit;
589 
590 		in->parent = pn;
591 		in->leaf = fn->leaf;
592 		atomic_inc(&in->leaf->rt6i_ref);
593 
594 		in->fn_sernum = sernum;
595 
596 		/* update parent pointer */
597 		if (dir)
598 			pn->right = in;
599 		else
600 			pn->left  = in;
601 
602 		ln->fn_bit = plen;
603 
604 		ln->parent = in;
605 		fn->parent = in;
606 
607 		ln->fn_sernum = sernum;
608 
609 		if (addr_bit_set(addr, bit)) {
610 			in->right = ln;
611 			in->left  = fn;
612 		} else {
613 			in->left  = ln;
614 			in->right = fn;
615 		}
616 	} else { /* plen <= bit */
617 
618 		/*
619 		 *		(new leaf node)[ln]
620 		 *	          /	   \
621 		 *	     (old node)[fn] NULL
622 		 */
623 
624 		ln = node_alloc();
625 
626 		if (!ln)
627 			return ERR_PTR(-ENOMEM);
628 
629 		ln->fn_bit = plen;
630 
631 		ln->parent = pn;
632 
633 		ln->fn_sernum = sernum;
634 
635 		if (dir)
636 			pn->right = ln;
637 		else
638 			pn->left  = ln;
639 
640 		if (addr_bit_set(&key->addr, plen))
641 			ln->right = fn;
642 		else
643 			ln->left  = fn;
644 
645 		fn->parent = ln;
646 	}
647 	return ln;
648 }
649 
650 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
651 {
652 	return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
653 	       RTF_GATEWAY;
654 }
655 
656 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
657 {
658 	int i;
659 
660 	for (i = 0; i < RTAX_MAX; i++) {
661 		if (test_bit(i, mxc->mx_valid))
662 			mp[i] = mxc->mx[i];
663 	}
664 }
665 
666 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
667 {
668 	if (!mxc->mx)
669 		return 0;
670 
671 	if (dst->flags & DST_HOST) {
672 		u32 *mp = dst_metrics_write_ptr(dst);
673 
674 		if (unlikely(!mp))
675 			return -ENOMEM;
676 
677 		fib6_copy_metrics(mp, mxc);
678 	} else {
679 		dst_init_metrics(dst, mxc->mx, false);
680 
681 		/* We've stolen mx now. */
682 		mxc->mx = NULL;
683 	}
684 
685 	return 0;
686 }
687 
688 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
689 			  struct net *net)
690 {
691 	if (atomic_read(&rt->rt6i_ref) != 1) {
692 		/* This route is used as dummy address holder in some split
693 		 * nodes. It is not leaked, but it still holds other resources,
694 		 * which must be released in time. So, scan ascendant nodes
695 		 * and replace dummy references to this route with references
696 		 * to still alive ones.
697 		 */
698 		while (fn) {
699 			if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
700 				fn->leaf = fib6_find_prefix(net, fn);
701 				atomic_inc(&fn->leaf->rt6i_ref);
702 				rt6_release(rt);
703 			}
704 			fn = fn->parent;
705 		}
706 		/* No more references are possible at this point. */
707 		BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
708 	}
709 }
710 
711 /*
712  *	Insert routing information in a node.
713  */
714 
715 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
716 			    struct nl_info *info, struct mx6_config *mxc)
717 {
718 	struct rt6_info *iter = NULL;
719 	struct rt6_info **ins;
720 	struct rt6_info **fallback_ins = NULL;
721 	int replace = (info->nlh &&
722 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
723 	int add = (!info->nlh ||
724 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
725 	int found = 0;
726 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
727 	int err;
728 
729 	ins = &fn->leaf;
730 
731 	for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
732 		/*
733 		 *	Search for duplicates
734 		 */
735 
736 		if (iter->rt6i_metric == rt->rt6i_metric) {
737 			/*
738 			 *	Same priority level
739 			 */
740 			if (info->nlh &&
741 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
742 				return -EEXIST;
743 			if (replace) {
744 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
745 					found++;
746 					break;
747 				}
748 				if (rt_can_ecmp)
749 					fallback_ins = fallback_ins ?: ins;
750 				goto next_iter;
751 			}
752 
753 			if (iter->dst.dev == rt->dst.dev &&
754 			    iter->rt6i_idev == rt->rt6i_idev &&
755 			    ipv6_addr_equal(&iter->rt6i_gateway,
756 					    &rt->rt6i_gateway)) {
757 				if (rt->rt6i_nsiblings)
758 					rt->rt6i_nsiblings = 0;
759 				if (!(iter->rt6i_flags & RTF_EXPIRES))
760 					return -EEXIST;
761 				if (!(rt->rt6i_flags & RTF_EXPIRES))
762 					rt6_clean_expires(iter);
763 				else
764 					rt6_set_expires(iter, rt->dst.expires);
765 				iter->rt6i_pmtu = rt->rt6i_pmtu;
766 				return -EEXIST;
767 			}
768 			/* If we have the same destination and the same metric,
769 			 * but not the same gateway, then the route we try to
770 			 * add is sibling to this route, increment our counter
771 			 * of siblings, and later we will add our route to the
772 			 * list.
773 			 * Only static routes (which don't have flag
774 			 * RTF_EXPIRES) are used for ECMPv6.
775 			 *
776 			 * To avoid long list, we only had siblings if the
777 			 * route have a gateway.
778 			 */
779 			if (rt_can_ecmp &&
780 			    rt6_qualify_for_ecmp(iter))
781 				rt->rt6i_nsiblings++;
782 		}
783 
784 		if (iter->rt6i_metric > rt->rt6i_metric)
785 			break;
786 
787 next_iter:
788 		ins = &iter->dst.rt6_next;
789 	}
790 
791 	if (fallback_ins && !found) {
792 		/* No ECMP-able route found, replace first non-ECMP one */
793 		ins = fallback_ins;
794 		iter = *ins;
795 		found++;
796 	}
797 
798 	/* Reset round-robin state, if necessary */
799 	if (ins == &fn->leaf)
800 		fn->rr_ptr = NULL;
801 
802 	/* Link this route to others same route. */
803 	if (rt->rt6i_nsiblings) {
804 		unsigned int rt6i_nsiblings;
805 		struct rt6_info *sibling, *temp_sibling;
806 
807 		/* Find the first route that have the same metric */
808 		sibling = fn->leaf;
809 		while (sibling) {
810 			if (sibling->rt6i_metric == rt->rt6i_metric &&
811 			    rt6_qualify_for_ecmp(sibling)) {
812 				list_add_tail(&rt->rt6i_siblings,
813 					      &sibling->rt6i_siblings);
814 				break;
815 			}
816 			sibling = sibling->dst.rt6_next;
817 		}
818 		/* For each sibling in the list, increment the counter of
819 		 * siblings. BUG() if counters does not match, list of siblings
820 		 * is broken!
821 		 */
822 		rt6i_nsiblings = 0;
823 		list_for_each_entry_safe(sibling, temp_sibling,
824 					 &rt->rt6i_siblings, rt6i_siblings) {
825 			sibling->rt6i_nsiblings++;
826 			BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
827 			rt6i_nsiblings++;
828 		}
829 		BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
830 	}
831 
832 	/*
833 	 *	insert node
834 	 */
835 	if (!replace) {
836 		if (!add)
837 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
838 
839 add:
840 		err = fib6_commit_metrics(&rt->dst, mxc);
841 		if (err)
842 			return err;
843 
844 		rt->dst.rt6_next = iter;
845 		*ins = rt;
846 		rt->rt6i_node = fn;
847 		atomic_inc(&rt->rt6i_ref);
848 		inet6_rt_notify(RTM_NEWROUTE, rt, info);
849 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
850 
851 		if (!(fn->fn_flags & RTN_RTINFO)) {
852 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
853 			fn->fn_flags |= RTN_RTINFO;
854 		}
855 
856 	} else {
857 		int nsiblings;
858 
859 		if (!found) {
860 			if (add)
861 				goto add;
862 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
863 			return -ENOENT;
864 		}
865 
866 		err = fib6_commit_metrics(&rt->dst, mxc);
867 		if (err)
868 			return err;
869 
870 		*ins = rt;
871 		rt->rt6i_node = fn;
872 		rt->dst.rt6_next = iter->dst.rt6_next;
873 		atomic_inc(&rt->rt6i_ref);
874 		inet6_rt_notify(RTM_NEWROUTE, rt, info);
875 		if (!(fn->fn_flags & RTN_RTINFO)) {
876 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
877 			fn->fn_flags |= RTN_RTINFO;
878 		}
879 		nsiblings = iter->rt6i_nsiblings;
880 		fib6_purge_rt(iter, fn, info->nl_net);
881 		rt6_release(iter);
882 
883 		if (nsiblings) {
884 			/* Replacing an ECMP route, remove all siblings */
885 			ins = &rt->dst.rt6_next;
886 			iter = *ins;
887 			while (iter) {
888 				if (rt6_qualify_for_ecmp(iter)) {
889 					*ins = iter->dst.rt6_next;
890 					fib6_purge_rt(iter, fn, info->nl_net);
891 					rt6_release(iter);
892 					nsiblings--;
893 				} else {
894 					ins = &iter->dst.rt6_next;
895 				}
896 				iter = *ins;
897 			}
898 			WARN_ON(nsiblings != 0);
899 		}
900 	}
901 
902 	return 0;
903 }
904 
905 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
906 {
907 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
908 	    (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
909 		mod_timer(&net->ipv6.ip6_fib_timer,
910 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
911 }
912 
913 void fib6_force_start_gc(struct net *net)
914 {
915 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
916 		mod_timer(&net->ipv6.ip6_fib_timer,
917 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
918 }
919 
920 /*
921  *	Add routing information to the routing tree.
922  *	<destination addr>/<source addr>
923  *	with source addr info in sub-trees
924  */
925 
926 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
927 	     struct nl_info *info, struct mx6_config *mxc)
928 {
929 	struct fib6_node *fn, *pn = NULL;
930 	int err = -ENOMEM;
931 	int allow_create = 1;
932 	int replace_required = 0;
933 	int sernum = fib6_new_sernum(info->nl_net);
934 
935 	if (info->nlh) {
936 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
937 			allow_create = 0;
938 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
939 			replace_required = 1;
940 	}
941 	if (!allow_create && !replace_required)
942 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
943 
944 	fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
945 			offsetof(struct rt6_info, rt6i_dst), allow_create,
946 			replace_required, sernum);
947 	if (IS_ERR(fn)) {
948 		err = PTR_ERR(fn);
949 		fn = NULL;
950 		goto out;
951 	}
952 
953 	pn = fn;
954 
955 #ifdef CONFIG_IPV6_SUBTREES
956 	if (rt->rt6i_src.plen) {
957 		struct fib6_node *sn;
958 
959 		if (!fn->subtree) {
960 			struct fib6_node *sfn;
961 
962 			/*
963 			 * Create subtree.
964 			 *
965 			 *		fn[main tree]
966 			 *		|
967 			 *		sfn[subtree root]
968 			 *		   \
969 			 *		    sn[new leaf node]
970 			 */
971 
972 			/* Create subtree root node */
973 			sfn = node_alloc();
974 			if (!sfn)
975 				goto st_failure;
976 
977 			sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
978 			atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
979 			sfn->fn_flags = RTN_ROOT;
980 			sfn->fn_sernum = sernum;
981 
982 			/* Now add the first leaf node to new subtree */
983 
984 			sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
985 					rt->rt6i_src.plen,
986 					offsetof(struct rt6_info, rt6i_src),
987 					allow_create, replace_required, sernum);
988 
989 			if (IS_ERR(sn)) {
990 				/* If it is failed, discard just allocated
991 				   root, and then (in st_failure) stale node
992 				   in main tree.
993 				 */
994 				node_free(sfn);
995 				err = PTR_ERR(sn);
996 				goto st_failure;
997 			}
998 
999 			/* Now link new subtree to main tree */
1000 			sfn->parent = fn;
1001 			fn->subtree = sfn;
1002 		} else {
1003 			sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
1004 					rt->rt6i_src.plen,
1005 					offsetof(struct rt6_info, rt6i_src),
1006 					allow_create, replace_required, sernum);
1007 
1008 			if (IS_ERR(sn)) {
1009 				err = PTR_ERR(sn);
1010 				goto st_failure;
1011 			}
1012 		}
1013 
1014 		if (!fn->leaf) {
1015 			fn->leaf = rt;
1016 			atomic_inc(&rt->rt6i_ref);
1017 		}
1018 		fn = sn;
1019 	}
1020 #endif
1021 
1022 	err = fib6_add_rt2node(fn, rt, info, mxc);
1023 	if (!err) {
1024 		fib6_start_gc(info->nl_net, rt);
1025 		if (!(rt->rt6i_flags & RTF_CACHE))
1026 			fib6_prune_clones(info->nl_net, pn);
1027 	}
1028 
1029 out:
1030 	if (err) {
1031 #ifdef CONFIG_IPV6_SUBTREES
1032 		/*
1033 		 * If fib6_add_1 has cleared the old leaf pointer in the
1034 		 * super-tree leaf node we have to find a new one for it.
1035 		 */
1036 		if (pn != fn && pn->leaf == rt) {
1037 			pn->leaf = NULL;
1038 			atomic_dec(&rt->rt6i_ref);
1039 		}
1040 		if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
1041 			pn->leaf = fib6_find_prefix(info->nl_net, pn);
1042 #if RT6_DEBUG >= 2
1043 			if (!pn->leaf) {
1044 				WARN_ON(pn->leaf == NULL);
1045 				pn->leaf = info->nl_net->ipv6.ip6_null_entry;
1046 			}
1047 #endif
1048 			atomic_inc(&pn->leaf->rt6i_ref);
1049 		}
1050 #endif
1051 		dst_free(&rt->dst);
1052 	}
1053 	return err;
1054 
1055 #ifdef CONFIG_IPV6_SUBTREES
1056 	/* Subtree creation failed, probably main tree node
1057 	   is orphan. If it is, shoot it.
1058 	 */
1059 st_failure:
1060 	if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1061 		fib6_repair_tree(info->nl_net, fn);
1062 	dst_free(&rt->dst);
1063 	return err;
1064 #endif
1065 }
1066 
1067 /*
1068  *	Routing tree lookup
1069  *
1070  */
1071 
1072 struct lookup_args {
1073 	int			offset;		/* key offset on rt6_info	*/
1074 	const struct in6_addr	*addr;		/* search key			*/
1075 };
1076 
1077 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1078 				       struct lookup_args *args)
1079 {
1080 	struct fib6_node *fn;
1081 	__be32 dir;
1082 
1083 	if (unlikely(args->offset == 0))
1084 		return NULL;
1085 
1086 	/*
1087 	 *	Descend on a tree
1088 	 */
1089 
1090 	fn = root;
1091 
1092 	for (;;) {
1093 		struct fib6_node *next;
1094 
1095 		dir = addr_bit_set(args->addr, fn->fn_bit);
1096 
1097 		next = dir ? fn->right : fn->left;
1098 
1099 		if (next) {
1100 			fn = next;
1101 			continue;
1102 		}
1103 		break;
1104 	}
1105 
1106 	while (fn) {
1107 		if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1108 			struct rt6key *key;
1109 
1110 			key = (struct rt6key *) ((u8 *) fn->leaf +
1111 						 args->offset);
1112 
1113 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1114 #ifdef CONFIG_IPV6_SUBTREES
1115 				if (fn->subtree) {
1116 					struct fib6_node *sfn;
1117 					sfn = fib6_lookup_1(fn->subtree,
1118 							    args + 1);
1119 					if (!sfn)
1120 						goto backtrack;
1121 					fn = sfn;
1122 				}
1123 #endif
1124 				if (fn->fn_flags & RTN_RTINFO)
1125 					return fn;
1126 			}
1127 		}
1128 #ifdef CONFIG_IPV6_SUBTREES
1129 backtrack:
1130 #endif
1131 		if (fn->fn_flags & RTN_ROOT)
1132 			break;
1133 
1134 		fn = fn->parent;
1135 	}
1136 
1137 	return NULL;
1138 }
1139 
1140 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1141 			      const struct in6_addr *saddr)
1142 {
1143 	struct fib6_node *fn;
1144 	struct lookup_args args[] = {
1145 		{
1146 			.offset = offsetof(struct rt6_info, rt6i_dst),
1147 			.addr = daddr,
1148 		},
1149 #ifdef CONFIG_IPV6_SUBTREES
1150 		{
1151 			.offset = offsetof(struct rt6_info, rt6i_src),
1152 			.addr = saddr,
1153 		},
1154 #endif
1155 		{
1156 			.offset = 0,	/* sentinel */
1157 		}
1158 	};
1159 
1160 	fn = fib6_lookup_1(root, daddr ? args : args + 1);
1161 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1162 		fn = root;
1163 
1164 	return fn;
1165 }
1166 
1167 /*
1168  *	Get node with specified destination prefix (and source prefix,
1169  *	if subtrees are used)
1170  */
1171 
1172 
1173 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1174 				       const struct in6_addr *addr,
1175 				       int plen, int offset)
1176 {
1177 	struct fib6_node *fn;
1178 
1179 	for (fn = root; fn ; ) {
1180 		struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1181 
1182 		/*
1183 		 *	Prefix match
1184 		 */
1185 		if (plen < fn->fn_bit ||
1186 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1187 			return NULL;
1188 
1189 		if (plen == fn->fn_bit)
1190 			return fn;
1191 
1192 		/*
1193 		 *	We have more bits to go
1194 		 */
1195 		if (addr_bit_set(addr, fn->fn_bit))
1196 			fn = fn->right;
1197 		else
1198 			fn = fn->left;
1199 	}
1200 	return NULL;
1201 }
1202 
1203 struct fib6_node *fib6_locate(struct fib6_node *root,
1204 			      const struct in6_addr *daddr, int dst_len,
1205 			      const struct in6_addr *saddr, int src_len)
1206 {
1207 	struct fib6_node *fn;
1208 
1209 	fn = fib6_locate_1(root, daddr, dst_len,
1210 			   offsetof(struct rt6_info, rt6i_dst));
1211 
1212 #ifdef CONFIG_IPV6_SUBTREES
1213 	if (src_len) {
1214 		WARN_ON(saddr == NULL);
1215 		if (fn && fn->subtree)
1216 			fn = fib6_locate_1(fn->subtree, saddr, src_len,
1217 					   offsetof(struct rt6_info, rt6i_src));
1218 	}
1219 #endif
1220 
1221 	if (fn && fn->fn_flags & RTN_RTINFO)
1222 		return fn;
1223 
1224 	return NULL;
1225 }
1226 
1227 
1228 /*
1229  *	Deletion
1230  *
1231  */
1232 
1233 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1234 {
1235 	if (fn->fn_flags & RTN_ROOT)
1236 		return net->ipv6.ip6_null_entry;
1237 
1238 	while (fn) {
1239 		if (fn->left)
1240 			return fn->left->leaf;
1241 		if (fn->right)
1242 			return fn->right->leaf;
1243 
1244 		fn = FIB6_SUBTREE(fn);
1245 	}
1246 	return NULL;
1247 }
1248 
1249 /*
1250  *	Called to trim the tree of intermediate nodes when possible. "fn"
1251  *	is the node we want to try and remove.
1252  */
1253 
1254 static struct fib6_node *fib6_repair_tree(struct net *net,
1255 					   struct fib6_node *fn)
1256 {
1257 	int children;
1258 	int nstate;
1259 	struct fib6_node *child, *pn;
1260 	struct fib6_walker *w;
1261 	int iter = 0;
1262 
1263 	for (;;) {
1264 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1265 		iter++;
1266 
1267 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1268 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1269 		WARN_ON(fn->leaf);
1270 
1271 		children = 0;
1272 		child = NULL;
1273 		if (fn->right)
1274 			child = fn->right, children |= 1;
1275 		if (fn->left)
1276 			child = fn->left, children |= 2;
1277 
1278 		if (children == 3 || FIB6_SUBTREE(fn)
1279 #ifdef CONFIG_IPV6_SUBTREES
1280 		    /* Subtree root (i.e. fn) may have one child */
1281 		    || (children && fn->fn_flags & RTN_ROOT)
1282 #endif
1283 		    ) {
1284 			fn->leaf = fib6_find_prefix(net, fn);
1285 #if RT6_DEBUG >= 2
1286 			if (!fn->leaf) {
1287 				WARN_ON(!fn->leaf);
1288 				fn->leaf = net->ipv6.ip6_null_entry;
1289 			}
1290 #endif
1291 			atomic_inc(&fn->leaf->rt6i_ref);
1292 			return fn->parent;
1293 		}
1294 
1295 		pn = fn->parent;
1296 #ifdef CONFIG_IPV6_SUBTREES
1297 		if (FIB6_SUBTREE(pn) == fn) {
1298 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1299 			FIB6_SUBTREE(pn) = NULL;
1300 			nstate = FWS_L;
1301 		} else {
1302 			WARN_ON(fn->fn_flags & RTN_ROOT);
1303 #endif
1304 			if (pn->right == fn)
1305 				pn->right = child;
1306 			else if (pn->left == fn)
1307 				pn->left = child;
1308 #if RT6_DEBUG >= 2
1309 			else
1310 				WARN_ON(1);
1311 #endif
1312 			if (child)
1313 				child->parent = pn;
1314 			nstate = FWS_R;
1315 #ifdef CONFIG_IPV6_SUBTREES
1316 		}
1317 #endif
1318 
1319 		read_lock(&fib6_walker_lock);
1320 		FOR_WALKERS(w) {
1321 			if (!child) {
1322 				if (w->root == fn) {
1323 					w->root = w->node = NULL;
1324 					RT6_TRACE("W %p adjusted by delroot 1\n", w);
1325 				} else if (w->node == fn) {
1326 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1327 					w->node = pn;
1328 					w->state = nstate;
1329 				}
1330 			} else {
1331 				if (w->root == fn) {
1332 					w->root = child;
1333 					RT6_TRACE("W %p adjusted by delroot 2\n", w);
1334 				}
1335 				if (w->node == fn) {
1336 					w->node = child;
1337 					if (children&2) {
1338 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1339 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1340 					} else {
1341 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1342 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1343 					}
1344 				}
1345 			}
1346 		}
1347 		read_unlock(&fib6_walker_lock);
1348 
1349 		node_free(fn);
1350 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1351 			return pn;
1352 
1353 		rt6_release(pn->leaf);
1354 		pn->leaf = NULL;
1355 		fn = pn;
1356 	}
1357 }
1358 
1359 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1360 			   struct nl_info *info)
1361 {
1362 	struct fib6_walker *w;
1363 	struct rt6_info *rt = *rtp;
1364 	struct net *net = info->nl_net;
1365 
1366 	RT6_TRACE("fib6_del_route\n");
1367 
1368 	/* Unlink it */
1369 	*rtp = rt->dst.rt6_next;
1370 	rt->rt6i_node = NULL;
1371 	net->ipv6.rt6_stats->fib_rt_entries--;
1372 	net->ipv6.rt6_stats->fib_discarded_routes++;
1373 
1374 	/* Reset round-robin state, if necessary */
1375 	if (fn->rr_ptr == rt)
1376 		fn->rr_ptr = NULL;
1377 
1378 	/* Remove this entry from other siblings */
1379 	if (rt->rt6i_nsiblings) {
1380 		struct rt6_info *sibling, *next_sibling;
1381 
1382 		list_for_each_entry_safe(sibling, next_sibling,
1383 					 &rt->rt6i_siblings, rt6i_siblings)
1384 			sibling->rt6i_nsiblings--;
1385 		rt->rt6i_nsiblings = 0;
1386 		list_del_init(&rt->rt6i_siblings);
1387 	}
1388 
1389 	/* Adjust walkers */
1390 	read_lock(&fib6_walker_lock);
1391 	FOR_WALKERS(w) {
1392 		if (w->state == FWS_C && w->leaf == rt) {
1393 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1394 			w->leaf = rt->dst.rt6_next;
1395 			if (!w->leaf)
1396 				w->state = FWS_U;
1397 		}
1398 	}
1399 	read_unlock(&fib6_walker_lock);
1400 
1401 	rt->dst.rt6_next = NULL;
1402 
1403 	/* If it was last route, expunge its radix tree node */
1404 	if (!fn->leaf) {
1405 		fn->fn_flags &= ~RTN_RTINFO;
1406 		net->ipv6.rt6_stats->fib_route_nodes--;
1407 		fn = fib6_repair_tree(net, fn);
1408 	}
1409 
1410 	fib6_purge_rt(rt, fn, net);
1411 
1412 	inet6_rt_notify(RTM_DELROUTE, rt, info);
1413 	rt6_release(rt);
1414 }
1415 
1416 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1417 {
1418 	struct net *net = info->nl_net;
1419 	struct fib6_node *fn = rt->rt6i_node;
1420 	struct rt6_info **rtp;
1421 
1422 #if RT6_DEBUG >= 2
1423 	if (rt->dst.obsolete > 0) {
1424 		WARN_ON(fn);
1425 		return -ENOENT;
1426 	}
1427 #endif
1428 	if (!fn || rt == net->ipv6.ip6_null_entry)
1429 		return -ENOENT;
1430 
1431 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1432 
1433 	if (!(rt->rt6i_flags & RTF_CACHE)) {
1434 		struct fib6_node *pn = fn;
1435 #ifdef CONFIG_IPV6_SUBTREES
1436 		/* clones of this route might be in another subtree */
1437 		if (rt->rt6i_src.plen) {
1438 			while (!(pn->fn_flags & RTN_ROOT))
1439 				pn = pn->parent;
1440 			pn = pn->parent;
1441 		}
1442 #endif
1443 		fib6_prune_clones(info->nl_net, pn);
1444 	}
1445 
1446 	/*
1447 	 *	Walk the leaf entries looking for ourself
1448 	 */
1449 
1450 	for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1451 		if (*rtp == rt) {
1452 			fib6_del_route(fn, rtp, info);
1453 			return 0;
1454 		}
1455 	}
1456 	return -ENOENT;
1457 }
1458 
1459 /*
1460  *	Tree traversal function.
1461  *
1462  *	Certainly, it is not interrupt safe.
1463  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1464  *	It means, that we can modify tree during walking
1465  *	and use this function for garbage collection, clone pruning,
1466  *	cleaning tree when a device goes down etc. etc.
1467  *
1468  *	It guarantees that every node will be traversed,
1469  *	and that it will be traversed only once.
1470  *
1471  *	Callback function w->func may return:
1472  *	0 -> continue walking.
1473  *	positive value -> walking is suspended (used by tree dumps,
1474  *	and probably by gc, if it will be split to several slices)
1475  *	negative value -> terminate walking.
1476  *
1477  *	The function itself returns:
1478  *	0   -> walk is complete.
1479  *	>0  -> walk is incomplete (i.e. suspended)
1480  *	<0  -> walk is terminated by an error.
1481  */
1482 
1483 static int fib6_walk_continue(struct fib6_walker *w)
1484 {
1485 	struct fib6_node *fn, *pn;
1486 
1487 	for (;;) {
1488 		fn = w->node;
1489 		if (!fn)
1490 			return 0;
1491 
1492 		if (w->prune && fn != w->root &&
1493 		    fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1494 			w->state = FWS_C;
1495 			w->leaf = fn->leaf;
1496 		}
1497 		switch (w->state) {
1498 #ifdef CONFIG_IPV6_SUBTREES
1499 		case FWS_S:
1500 			if (FIB6_SUBTREE(fn)) {
1501 				w->node = FIB6_SUBTREE(fn);
1502 				continue;
1503 			}
1504 			w->state = FWS_L;
1505 #endif
1506 		case FWS_L:
1507 			if (fn->left) {
1508 				w->node = fn->left;
1509 				w->state = FWS_INIT;
1510 				continue;
1511 			}
1512 			w->state = FWS_R;
1513 		case FWS_R:
1514 			if (fn->right) {
1515 				w->node = fn->right;
1516 				w->state = FWS_INIT;
1517 				continue;
1518 			}
1519 			w->state = FWS_C;
1520 			w->leaf = fn->leaf;
1521 		case FWS_C:
1522 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1523 				int err;
1524 
1525 				if (w->skip) {
1526 					w->skip--;
1527 					goto skip;
1528 				}
1529 
1530 				err = w->func(w);
1531 				if (err)
1532 					return err;
1533 
1534 				w->count++;
1535 				continue;
1536 			}
1537 skip:
1538 			w->state = FWS_U;
1539 		case FWS_U:
1540 			if (fn == w->root)
1541 				return 0;
1542 			pn = fn->parent;
1543 			w->node = pn;
1544 #ifdef CONFIG_IPV6_SUBTREES
1545 			if (FIB6_SUBTREE(pn) == fn) {
1546 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1547 				w->state = FWS_L;
1548 				continue;
1549 			}
1550 #endif
1551 			if (pn->left == fn) {
1552 				w->state = FWS_R;
1553 				continue;
1554 			}
1555 			if (pn->right == fn) {
1556 				w->state = FWS_C;
1557 				w->leaf = w->node->leaf;
1558 				continue;
1559 			}
1560 #if RT6_DEBUG >= 2
1561 			WARN_ON(1);
1562 #endif
1563 		}
1564 	}
1565 }
1566 
1567 static int fib6_walk(struct fib6_walker *w)
1568 {
1569 	int res;
1570 
1571 	w->state = FWS_INIT;
1572 	w->node = w->root;
1573 
1574 	fib6_walker_link(w);
1575 	res = fib6_walk_continue(w);
1576 	if (res <= 0)
1577 		fib6_walker_unlink(w);
1578 	return res;
1579 }
1580 
1581 static int fib6_clean_node(struct fib6_walker *w)
1582 {
1583 	int res;
1584 	struct rt6_info *rt;
1585 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1586 	struct nl_info info = {
1587 		.nl_net = c->net,
1588 	};
1589 
1590 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1591 	    w->node->fn_sernum != c->sernum)
1592 		w->node->fn_sernum = c->sernum;
1593 
1594 	if (!c->func) {
1595 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1596 		w->leaf = NULL;
1597 		return 0;
1598 	}
1599 
1600 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1601 		res = c->func(rt, c->arg);
1602 		if (res < 0) {
1603 			w->leaf = rt;
1604 			res = fib6_del(rt, &info);
1605 			if (res) {
1606 #if RT6_DEBUG >= 2
1607 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1608 					 __func__, rt, rt->rt6i_node, res);
1609 #endif
1610 				continue;
1611 			}
1612 			return 0;
1613 		}
1614 		WARN_ON(res != 0);
1615 	}
1616 	w->leaf = rt;
1617 	return 0;
1618 }
1619 
1620 /*
1621  *	Convenient frontend to tree walker.
1622  *
1623  *	func is called on each route.
1624  *		It may return -1 -> delete this route.
1625  *		              0  -> continue walking
1626  *
1627  *	prune==1 -> only immediate children of node (certainly,
1628  *	ignoring pure split nodes) will be scanned.
1629  */
1630 
1631 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1632 			    int (*func)(struct rt6_info *, void *arg),
1633 			    bool prune, int sernum, void *arg)
1634 {
1635 	struct fib6_cleaner c;
1636 
1637 	c.w.root = root;
1638 	c.w.func = fib6_clean_node;
1639 	c.w.prune = prune;
1640 	c.w.count = 0;
1641 	c.w.skip = 0;
1642 	c.func = func;
1643 	c.sernum = sernum;
1644 	c.arg = arg;
1645 	c.net = net;
1646 
1647 	fib6_walk(&c.w);
1648 }
1649 
1650 static void __fib6_clean_all(struct net *net,
1651 			     int (*func)(struct rt6_info *, void *),
1652 			     int sernum, void *arg)
1653 {
1654 	struct fib6_table *table;
1655 	struct hlist_head *head;
1656 	unsigned int h;
1657 
1658 	rcu_read_lock();
1659 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1660 		head = &net->ipv6.fib_table_hash[h];
1661 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1662 			write_lock_bh(&table->tb6_lock);
1663 			fib6_clean_tree(net, &table->tb6_root,
1664 					func, false, sernum, arg);
1665 			write_unlock_bh(&table->tb6_lock);
1666 		}
1667 	}
1668 	rcu_read_unlock();
1669 }
1670 
1671 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1672 		    void *arg)
1673 {
1674 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1675 }
1676 
1677 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1678 {
1679 	if (rt->rt6i_flags & RTF_CACHE) {
1680 		RT6_TRACE("pruning clone %p\n", rt);
1681 		return -1;
1682 	}
1683 
1684 	return 0;
1685 }
1686 
1687 static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1688 {
1689 	fib6_clean_tree(net, fn, fib6_prune_clone, true,
1690 			FIB6_NO_SERNUM_CHANGE, NULL);
1691 }
1692 
1693 static void fib6_flush_trees(struct net *net)
1694 {
1695 	int new_sernum = fib6_new_sernum(net);
1696 
1697 	__fib6_clean_all(net, NULL, new_sernum, NULL);
1698 }
1699 
1700 /*
1701  *	Garbage collection
1702  */
1703 
1704 static struct fib6_gc_args
1705 {
1706 	int			timeout;
1707 	int			more;
1708 } gc_args;
1709 
1710 static int fib6_age(struct rt6_info *rt, void *arg)
1711 {
1712 	unsigned long now = jiffies;
1713 
1714 	/*
1715 	 *	check addrconf expiration here.
1716 	 *	Routes are expired even if they are in use.
1717 	 *
1718 	 *	Also age clones. Note, that clones are aged out
1719 	 *	only if they are not in use now.
1720 	 */
1721 
1722 	if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1723 		if (time_after(now, rt->dst.expires)) {
1724 			RT6_TRACE("expiring %p\n", rt);
1725 			return -1;
1726 		}
1727 		gc_args.more++;
1728 	} else if (rt->rt6i_flags & RTF_CACHE) {
1729 		if (atomic_read(&rt->dst.__refcnt) == 0 &&
1730 		    time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1731 			RT6_TRACE("aging clone %p\n", rt);
1732 			return -1;
1733 		} else if (rt->rt6i_flags & RTF_GATEWAY) {
1734 			struct neighbour *neigh;
1735 			__u8 neigh_flags = 0;
1736 
1737 			neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1738 			if (neigh) {
1739 				neigh_flags = neigh->flags;
1740 				neigh_release(neigh);
1741 			}
1742 			if (!(neigh_flags & NTF_ROUTER)) {
1743 				RT6_TRACE("purging route %p via non-router but gateway\n",
1744 					  rt);
1745 				return -1;
1746 			}
1747 		}
1748 		gc_args.more++;
1749 	}
1750 
1751 	return 0;
1752 }
1753 
1754 static DEFINE_SPINLOCK(fib6_gc_lock);
1755 
1756 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1757 {
1758 	unsigned long now;
1759 
1760 	if (force) {
1761 		spin_lock_bh(&fib6_gc_lock);
1762 	} else if (!spin_trylock_bh(&fib6_gc_lock)) {
1763 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1764 		return;
1765 	}
1766 	gc_args.timeout = expires ? (int)expires :
1767 			  net->ipv6.sysctl.ip6_rt_gc_interval;
1768 
1769 	gc_args.more = icmp6_dst_gc();
1770 
1771 	fib6_clean_all(net, fib6_age, NULL);
1772 	now = jiffies;
1773 	net->ipv6.ip6_rt_last_gc = now;
1774 
1775 	if (gc_args.more)
1776 		mod_timer(&net->ipv6.ip6_fib_timer,
1777 			  round_jiffies(now
1778 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
1779 	else
1780 		del_timer(&net->ipv6.ip6_fib_timer);
1781 	spin_unlock_bh(&fib6_gc_lock);
1782 }
1783 
1784 static void fib6_gc_timer_cb(unsigned long arg)
1785 {
1786 	fib6_run_gc(0, (struct net *)arg, true);
1787 }
1788 
1789 static int __net_init fib6_net_init(struct net *net)
1790 {
1791 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1792 
1793 	setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1794 
1795 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1796 	if (!net->ipv6.rt6_stats)
1797 		goto out_timer;
1798 
1799 	/* Avoid false sharing : Use at least a full cache line */
1800 	size = max_t(size_t, size, L1_CACHE_BYTES);
1801 
1802 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1803 	if (!net->ipv6.fib_table_hash)
1804 		goto out_rt6_stats;
1805 
1806 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1807 					  GFP_KERNEL);
1808 	if (!net->ipv6.fib6_main_tbl)
1809 		goto out_fib_table_hash;
1810 
1811 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1812 	net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1813 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1814 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1815 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1816 
1817 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1818 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1819 					   GFP_KERNEL);
1820 	if (!net->ipv6.fib6_local_tbl)
1821 		goto out_fib6_main_tbl;
1822 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1823 	net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1824 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1825 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1826 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1827 #endif
1828 	fib6_tables_init(net);
1829 
1830 	return 0;
1831 
1832 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1833 out_fib6_main_tbl:
1834 	kfree(net->ipv6.fib6_main_tbl);
1835 #endif
1836 out_fib_table_hash:
1837 	kfree(net->ipv6.fib_table_hash);
1838 out_rt6_stats:
1839 	kfree(net->ipv6.rt6_stats);
1840 out_timer:
1841 	return -ENOMEM;
1842 }
1843 
1844 static void fib6_net_exit(struct net *net)
1845 {
1846 	rt6_ifdown(net, NULL);
1847 	del_timer_sync(&net->ipv6.ip6_fib_timer);
1848 
1849 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1850 	inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1851 	kfree(net->ipv6.fib6_local_tbl);
1852 #endif
1853 	inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1854 	kfree(net->ipv6.fib6_main_tbl);
1855 	kfree(net->ipv6.fib_table_hash);
1856 	kfree(net->ipv6.rt6_stats);
1857 }
1858 
1859 static struct pernet_operations fib6_net_ops = {
1860 	.init = fib6_net_init,
1861 	.exit = fib6_net_exit,
1862 };
1863 
1864 int __init fib6_init(void)
1865 {
1866 	int ret = -ENOMEM;
1867 
1868 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
1869 					   sizeof(struct fib6_node),
1870 					   0, SLAB_HWCACHE_ALIGN,
1871 					   NULL);
1872 	if (!fib6_node_kmem)
1873 		goto out;
1874 
1875 	ret = register_pernet_subsys(&fib6_net_ops);
1876 	if (ret)
1877 		goto out_kmem_cache_create;
1878 
1879 	ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1880 			      NULL);
1881 	if (ret)
1882 		goto out_unregister_subsys;
1883 
1884 	__fib6_flush_trees = fib6_flush_trees;
1885 out:
1886 	return ret;
1887 
1888 out_unregister_subsys:
1889 	unregister_pernet_subsys(&fib6_net_ops);
1890 out_kmem_cache_create:
1891 	kmem_cache_destroy(fib6_node_kmem);
1892 	goto out;
1893 }
1894 
1895 void fib6_gc_cleanup(void)
1896 {
1897 	unregister_pernet_subsys(&fib6_net_ops);
1898 	kmem_cache_destroy(fib6_node_kmem);
1899 }
1900 
1901 #ifdef CONFIG_PROC_FS
1902 
1903 struct ipv6_route_iter {
1904 	struct seq_net_private p;
1905 	struct fib6_walker w;
1906 	loff_t skip;
1907 	struct fib6_table *tbl;
1908 	int sernum;
1909 };
1910 
1911 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
1912 {
1913 	struct rt6_info *rt = v;
1914 	struct ipv6_route_iter *iter = seq->private;
1915 
1916 	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
1917 
1918 #ifdef CONFIG_IPV6_SUBTREES
1919 	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
1920 #else
1921 	seq_puts(seq, "00000000000000000000000000000000 00 ");
1922 #endif
1923 	if (rt->rt6i_flags & RTF_GATEWAY)
1924 		seq_printf(seq, "%pi6", &rt->rt6i_gateway);
1925 	else
1926 		seq_puts(seq, "00000000000000000000000000000000");
1927 
1928 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
1929 		   rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
1930 		   rt->dst.__use, rt->rt6i_flags,
1931 		   rt->dst.dev ? rt->dst.dev->name : "");
1932 	iter->w.leaf = NULL;
1933 	return 0;
1934 }
1935 
1936 static int ipv6_route_yield(struct fib6_walker *w)
1937 {
1938 	struct ipv6_route_iter *iter = w->args;
1939 
1940 	if (!iter->skip)
1941 		return 1;
1942 
1943 	do {
1944 		iter->w.leaf = iter->w.leaf->dst.rt6_next;
1945 		iter->skip--;
1946 		if (!iter->skip && iter->w.leaf)
1947 			return 1;
1948 	} while (iter->w.leaf);
1949 
1950 	return 0;
1951 }
1952 
1953 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter)
1954 {
1955 	memset(&iter->w, 0, sizeof(iter->w));
1956 	iter->w.func = ipv6_route_yield;
1957 	iter->w.root = &iter->tbl->tb6_root;
1958 	iter->w.state = FWS_INIT;
1959 	iter->w.node = iter->w.root;
1960 	iter->w.args = iter;
1961 	iter->sernum = iter->w.root->fn_sernum;
1962 	INIT_LIST_HEAD(&iter->w.lh);
1963 	fib6_walker_link(&iter->w);
1964 }
1965 
1966 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
1967 						    struct net *net)
1968 {
1969 	unsigned int h;
1970 	struct hlist_node *node;
1971 
1972 	if (tbl) {
1973 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
1974 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
1975 	} else {
1976 		h = 0;
1977 		node = NULL;
1978 	}
1979 
1980 	while (!node && h < FIB6_TABLE_HASHSZ) {
1981 		node = rcu_dereference_bh(
1982 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
1983 	}
1984 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
1985 }
1986 
1987 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
1988 {
1989 	if (iter->sernum != iter->w.root->fn_sernum) {
1990 		iter->sernum = iter->w.root->fn_sernum;
1991 		iter->w.state = FWS_INIT;
1992 		iter->w.node = iter->w.root;
1993 		WARN_ON(iter->w.skip);
1994 		iter->w.skip = iter->w.count;
1995 	}
1996 }
1997 
1998 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1999 {
2000 	int r;
2001 	struct rt6_info *n;
2002 	struct net *net = seq_file_net(seq);
2003 	struct ipv6_route_iter *iter = seq->private;
2004 
2005 	if (!v)
2006 		goto iter_table;
2007 
2008 	n = ((struct rt6_info *)v)->dst.rt6_next;
2009 	if (n) {
2010 		++*pos;
2011 		return n;
2012 	}
2013 
2014 iter_table:
2015 	ipv6_route_check_sernum(iter);
2016 	read_lock(&iter->tbl->tb6_lock);
2017 	r = fib6_walk_continue(&iter->w);
2018 	read_unlock(&iter->tbl->tb6_lock);
2019 	if (r > 0) {
2020 		if (v)
2021 			++*pos;
2022 		return iter->w.leaf;
2023 	} else if (r < 0) {
2024 		fib6_walker_unlink(&iter->w);
2025 		return NULL;
2026 	}
2027 	fib6_walker_unlink(&iter->w);
2028 
2029 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2030 	if (!iter->tbl)
2031 		return NULL;
2032 
2033 	ipv6_route_seq_setup_walk(iter);
2034 	goto iter_table;
2035 }
2036 
2037 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2038 	__acquires(RCU_BH)
2039 {
2040 	struct net *net = seq_file_net(seq);
2041 	struct ipv6_route_iter *iter = seq->private;
2042 
2043 	rcu_read_lock_bh();
2044 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2045 	iter->skip = *pos;
2046 
2047 	if (iter->tbl) {
2048 		ipv6_route_seq_setup_walk(iter);
2049 		return ipv6_route_seq_next(seq, NULL, pos);
2050 	} else {
2051 		return NULL;
2052 	}
2053 }
2054 
2055 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2056 {
2057 	struct fib6_walker *w = &iter->w;
2058 	return w->node && !(w->state == FWS_U && w->node == w->root);
2059 }
2060 
2061 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2062 	__releases(RCU_BH)
2063 {
2064 	struct ipv6_route_iter *iter = seq->private;
2065 
2066 	if (ipv6_route_iter_active(iter))
2067 		fib6_walker_unlink(&iter->w);
2068 
2069 	rcu_read_unlock_bh();
2070 }
2071 
2072 static const struct seq_operations ipv6_route_seq_ops = {
2073 	.start	= ipv6_route_seq_start,
2074 	.next	= ipv6_route_seq_next,
2075 	.stop	= ipv6_route_seq_stop,
2076 	.show	= ipv6_route_seq_show
2077 };
2078 
2079 int ipv6_route_open(struct inode *inode, struct file *file)
2080 {
2081 	return seq_open_net(inode, file, &ipv6_route_seq_ops,
2082 			    sizeof(struct ipv6_route_iter));
2083 }
2084 
2085 #endif /* CONFIG_PROC_FS */
2086