xref: /openbmc/linux/net/ipv6/ip6_fib.c (revision eb3fcf00)
1 /*
2  *	Linux INET6 implementation
3  *	Forwarding Information Database
4  *
5  *	Authors:
6  *	Pedro Roque		<roque@di.fc.ul.pt>
7  *
8  *	This program is free software; you can redistribute it and/or
9  *      modify it under the terms of the GNU General Public License
10  *      as published by the Free Software Foundation; either version
11  *      2 of the License, or (at your option) any later version.
12  *
13  *	Changes:
14  *	Yuji SEKIYA @USAGI:	Support default route on router node;
15  *				remove ip6_null_entry from the top of
16  *				routing table.
17  *	Ville Nuorvala:		Fixed routing subtrees.
18  */
19 
20 #define pr_fmt(fmt) "IPv6: " fmt
21 
22 #include <linux/errno.h>
23 #include <linux/types.h>
24 #include <linux/net.h>
25 #include <linux/route.h>
26 #include <linux/netdevice.h>
27 #include <linux/in6.h>
28 #include <linux/init.h>
29 #include <linux/list.h>
30 #include <linux/slab.h>
31 
32 #include <net/ipv6.h>
33 #include <net/ndisc.h>
34 #include <net/addrconf.h>
35 #include <net/lwtunnel.h>
36 
37 #include <net/ip6_fib.h>
38 #include <net/ip6_route.h>
39 
40 #define RT6_DEBUG 2
41 
42 #if RT6_DEBUG >= 3
43 #define RT6_TRACE(x...) pr_debug(x)
44 #else
45 #define RT6_TRACE(x...) do { ; } while (0)
46 #endif
47 
48 static struct kmem_cache *fib6_node_kmem __read_mostly;
49 
50 struct fib6_cleaner {
51 	struct fib6_walker w;
52 	struct net *net;
53 	int (*func)(struct rt6_info *, void *arg);
54 	int sernum;
55 	void *arg;
56 };
57 
58 static DEFINE_RWLOCK(fib6_walker_lock);
59 
60 #ifdef CONFIG_IPV6_SUBTREES
61 #define FWS_INIT FWS_S
62 #else
63 #define FWS_INIT FWS_L
64 #endif
65 
66 static void fib6_prune_clones(struct net *net, struct fib6_node *fn);
67 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
68 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
69 static int fib6_walk(struct fib6_walker *w);
70 static int fib6_walk_continue(struct fib6_walker *w);
71 
72 /*
73  *	A routing update causes an increase of the serial number on the
74  *	affected subtree. This allows for cached routes to be asynchronously
75  *	tested when modifications are made to the destination cache as a
76  *	result of redirects, path MTU changes, etc.
77  */
78 
79 static void fib6_gc_timer_cb(unsigned long arg);
80 
81 static LIST_HEAD(fib6_walkers);
82 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
83 
84 static void fib6_walker_link(struct fib6_walker *w)
85 {
86 	write_lock_bh(&fib6_walker_lock);
87 	list_add(&w->lh, &fib6_walkers);
88 	write_unlock_bh(&fib6_walker_lock);
89 }
90 
91 static void fib6_walker_unlink(struct fib6_walker *w)
92 {
93 	write_lock_bh(&fib6_walker_lock);
94 	list_del(&w->lh);
95 	write_unlock_bh(&fib6_walker_lock);
96 }
97 
98 static int fib6_new_sernum(struct net *net)
99 {
100 	int new, old;
101 
102 	do {
103 		old = atomic_read(&net->ipv6.fib6_sernum);
104 		new = old < INT_MAX ? old + 1 : 1;
105 	} while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
106 				old, new) != old);
107 	return new;
108 }
109 
110 enum {
111 	FIB6_NO_SERNUM_CHANGE = 0,
112 };
113 
114 /*
115  *	Auxiliary address test functions for the radix tree.
116  *
117  *	These assume a 32bit processor (although it will work on
118  *	64bit processors)
119  */
120 
121 /*
122  *	test bit
123  */
124 #if defined(__LITTLE_ENDIAN)
125 # define BITOP_BE32_SWIZZLE	(0x1F & ~7)
126 #else
127 # define BITOP_BE32_SWIZZLE	0
128 #endif
129 
130 static __be32 addr_bit_set(const void *token, int fn_bit)
131 {
132 	const __be32 *addr = token;
133 	/*
134 	 * Here,
135 	 *	1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
136 	 * is optimized version of
137 	 *	htonl(1 << ((~fn_bit)&0x1F))
138 	 * See include/asm-generic/bitops/le.h.
139 	 */
140 	return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
141 	       addr[fn_bit >> 5];
142 }
143 
144 static struct fib6_node *node_alloc(void)
145 {
146 	struct fib6_node *fn;
147 
148 	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
149 
150 	return fn;
151 }
152 
153 static void node_free(struct fib6_node *fn)
154 {
155 	kmem_cache_free(fib6_node_kmem, fn);
156 }
157 
158 static void rt6_rcu_free(struct rt6_info *rt)
159 {
160 	call_rcu(&rt->dst.rcu_head, dst_rcu_free);
161 }
162 
163 static void rt6_free_pcpu(struct rt6_info *non_pcpu_rt)
164 {
165 	int cpu;
166 
167 	if (!non_pcpu_rt->rt6i_pcpu)
168 		return;
169 
170 	for_each_possible_cpu(cpu) {
171 		struct rt6_info **ppcpu_rt;
172 		struct rt6_info *pcpu_rt;
173 
174 		ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
175 		pcpu_rt = *ppcpu_rt;
176 		if (pcpu_rt) {
177 			rt6_rcu_free(pcpu_rt);
178 			*ppcpu_rt = NULL;
179 		}
180 	}
181 
182 	non_pcpu_rt->rt6i_pcpu = NULL;
183 }
184 
185 static void rt6_release(struct rt6_info *rt)
186 {
187 	if (atomic_dec_and_test(&rt->rt6i_ref)) {
188 		rt6_free_pcpu(rt);
189 		rt6_rcu_free(rt);
190 	}
191 }
192 
193 static void fib6_link_table(struct net *net, struct fib6_table *tb)
194 {
195 	unsigned int h;
196 
197 	/*
198 	 * Initialize table lock at a single place to give lockdep a key,
199 	 * tables aren't visible prior to being linked to the list.
200 	 */
201 	rwlock_init(&tb->tb6_lock);
202 
203 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
204 
205 	/*
206 	 * No protection necessary, this is the only list mutatation
207 	 * operation, tables never disappear once they exist.
208 	 */
209 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
210 }
211 
212 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
213 
214 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
215 {
216 	struct fib6_table *table;
217 
218 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
219 	if (table) {
220 		table->tb6_id = id;
221 		table->tb6_root.leaf = net->ipv6.ip6_null_entry;
222 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
223 		inet_peer_base_init(&table->tb6_peers);
224 	}
225 
226 	return table;
227 }
228 
229 struct fib6_table *fib6_new_table(struct net *net, u32 id)
230 {
231 	struct fib6_table *tb;
232 
233 	if (id == 0)
234 		id = RT6_TABLE_MAIN;
235 	tb = fib6_get_table(net, id);
236 	if (tb)
237 		return tb;
238 
239 	tb = fib6_alloc_table(net, id);
240 	if (tb)
241 		fib6_link_table(net, tb);
242 
243 	return tb;
244 }
245 
246 struct fib6_table *fib6_get_table(struct net *net, u32 id)
247 {
248 	struct fib6_table *tb;
249 	struct hlist_head *head;
250 	unsigned int h;
251 
252 	if (id == 0)
253 		id = RT6_TABLE_MAIN;
254 	h = id & (FIB6_TABLE_HASHSZ - 1);
255 	rcu_read_lock();
256 	head = &net->ipv6.fib_table_hash[h];
257 	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
258 		if (tb->tb6_id == id) {
259 			rcu_read_unlock();
260 			return tb;
261 		}
262 	}
263 	rcu_read_unlock();
264 
265 	return NULL;
266 }
267 
268 static void __net_init fib6_tables_init(struct net *net)
269 {
270 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
271 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
272 }
273 #else
274 
275 struct fib6_table *fib6_new_table(struct net *net, u32 id)
276 {
277 	return fib6_get_table(net, id);
278 }
279 
280 struct fib6_table *fib6_get_table(struct net *net, u32 id)
281 {
282 	  return net->ipv6.fib6_main_tbl;
283 }
284 
285 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
286 				   int flags, pol_lookup_t lookup)
287 {
288 	return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
289 }
290 
291 static void __net_init fib6_tables_init(struct net *net)
292 {
293 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
294 }
295 
296 #endif
297 
298 static int fib6_dump_node(struct fib6_walker *w)
299 {
300 	int res;
301 	struct rt6_info *rt;
302 
303 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
304 		res = rt6_dump_route(rt, w->args);
305 		if (res < 0) {
306 			/* Frame is full, suspend walking */
307 			w->leaf = rt;
308 			return 1;
309 		}
310 	}
311 	w->leaf = NULL;
312 	return 0;
313 }
314 
315 static void fib6_dump_end(struct netlink_callback *cb)
316 {
317 	struct fib6_walker *w = (void *)cb->args[2];
318 
319 	if (w) {
320 		if (cb->args[4]) {
321 			cb->args[4] = 0;
322 			fib6_walker_unlink(w);
323 		}
324 		cb->args[2] = 0;
325 		kfree(w);
326 	}
327 	cb->done = (void *)cb->args[3];
328 	cb->args[1] = 3;
329 }
330 
331 static int fib6_dump_done(struct netlink_callback *cb)
332 {
333 	fib6_dump_end(cb);
334 	return cb->done ? cb->done(cb) : 0;
335 }
336 
337 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
338 			   struct netlink_callback *cb)
339 {
340 	struct fib6_walker *w;
341 	int res;
342 
343 	w = (void *)cb->args[2];
344 	w->root = &table->tb6_root;
345 
346 	if (cb->args[4] == 0) {
347 		w->count = 0;
348 		w->skip = 0;
349 
350 		read_lock_bh(&table->tb6_lock);
351 		res = fib6_walk(w);
352 		read_unlock_bh(&table->tb6_lock);
353 		if (res > 0) {
354 			cb->args[4] = 1;
355 			cb->args[5] = w->root->fn_sernum;
356 		}
357 	} else {
358 		if (cb->args[5] != w->root->fn_sernum) {
359 			/* Begin at the root if the tree changed */
360 			cb->args[5] = w->root->fn_sernum;
361 			w->state = FWS_INIT;
362 			w->node = w->root;
363 			w->skip = w->count;
364 		} else
365 			w->skip = 0;
366 
367 		read_lock_bh(&table->tb6_lock);
368 		res = fib6_walk_continue(w);
369 		read_unlock_bh(&table->tb6_lock);
370 		if (res <= 0) {
371 			fib6_walker_unlink(w);
372 			cb->args[4] = 0;
373 		}
374 	}
375 
376 	return res;
377 }
378 
379 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
380 {
381 	struct net *net = sock_net(skb->sk);
382 	unsigned int h, s_h;
383 	unsigned int e = 0, s_e;
384 	struct rt6_rtnl_dump_arg arg;
385 	struct fib6_walker *w;
386 	struct fib6_table *tb;
387 	struct hlist_head *head;
388 	int res = 0;
389 
390 	s_h = cb->args[0];
391 	s_e = cb->args[1];
392 
393 	w = (void *)cb->args[2];
394 	if (!w) {
395 		/* New dump:
396 		 *
397 		 * 1. hook callback destructor.
398 		 */
399 		cb->args[3] = (long)cb->done;
400 		cb->done = fib6_dump_done;
401 
402 		/*
403 		 * 2. allocate and initialize walker.
404 		 */
405 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
406 		if (!w)
407 			return -ENOMEM;
408 		w->func = fib6_dump_node;
409 		cb->args[2] = (long)w;
410 	}
411 
412 	arg.skb = skb;
413 	arg.cb = cb;
414 	arg.net = net;
415 	w->args = &arg;
416 
417 	rcu_read_lock();
418 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
419 		e = 0;
420 		head = &net->ipv6.fib_table_hash[h];
421 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
422 			if (e < s_e)
423 				goto next;
424 			res = fib6_dump_table(tb, skb, cb);
425 			if (res != 0)
426 				goto out;
427 next:
428 			e++;
429 		}
430 	}
431 out:
432 	rcu_read_unlock();
433 	cb->args[1] = e;
434 	cb->args[0] = h;
435 
436 	res = res < 0 ? res : skb->len;
437 	if (res <= 0)
438 		fib6_dump_end(cb);
439 	return res;
440 }
441 
442 /*
443  *	Routing Table
444  *
445  *	return the appropriate node for a routing tree "add" operation
446  *	by either creating and inserting or by returning an existing
447  *	node.
448  */
449 
450 static struct fib6_node *fib6_add_1(struct fib6_node *root,
451 				     struct in6_addr *addr, int plen,
452 				     int offset, int allow_create,
453 				     int replace_required, int sernum)
454 {
455 	struct fib6_node *fn, *in, *ln;
456 	struct fib6_node *pn = NULL;
457 	struct rt6key *key;
458 	int	bit;
459 	__be32	dir = 0;
460 
461 	RT6_TRACE("fib6_add_1\n");
462 
463 	/* insert node in tree */
464 
465 	fn = root;
466 
467 	do {
468 		key = (struct rt6key *)((u8 *)fn->leaf + offset);
469 
470 		/*
471 		 *	Prefix match
472 		 */
473 		if (plen < fn->fn_bit ||
474 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
475 			if (!allow_create) {
476 				if (replace_required) {
477 					pr_warn("Can't replace route, no match found\n");
478 					return ERR_PTR(-ENOENT);
479 				}
480 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
481 			}
482 			goto insert_above;
483 		}
484 
485 		/*
486 		 *	Exact match ?
487 		 */
488 
489 		if (plen == fn->fn_bit) {
490 			/* clean up an intermediate node */
491 			if (!(fn->fn_flags & RTN_RTINFO)) {
492 				rt6_release(fn->leaf);
493 				fn->leaf = NULL;
494 			}
495 
496 			fn->fn_sernum = sernum;
497 
498 			return fn;
499 		}
500 
501 		/*
502 		 *	We have more bits to go
503 		 */
504 
505 		/* Try to walk down on tree. */
506 		fn->fn_sernum = sernum;
507 		dir = addr_bit_set(addr, fn->fn_bit);
508 		pn = fn;
509 		fn = dir ? fn->right : fn->left;
510 	} while (fn);
511 
512 	if (!allow_create) {
513 		/* We should not create new node because
514 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
515 		 * I assume it is safe to require NLM_F_CREATE when
516 		 * REPLACE flag is used! Later we may want to remove the
517 		 * check for replace_required, because according
518 		 * to netlink specification, NLM_F_CREATE
519 		 * MUST be specified if new route is created.
520 		 * That would keep IPv6 consistent with IPv4
521 		 */
522 		if (replace_required) {
523 			pr_warn("Can't replace route, no match found\n");
524 			return ERR_PTR(-ENOENT);
525 		}
526 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
527 	}
528 	/*
529 	 *	We walked to the bottom of tree.
530 	 *	Create new leaf node without children.
531 	 */
532 
533 	ln = node_alloc();
534 
535 	if (!ln)
536 		return ERR_PTR(-ENOMEM);
537 	ln->fn_bit = plen;
538 
539 	ln->parent = pn;
540 	ln->fn_sernum = sernum;
541 
542 	if (dir)
543 		pn->right = ln;
544 	else
545 		pn->left  = ln;
546 
547 	return ln;
548 
549 
550 insert_above:
551 	/*
552 	 * split since we don't have a common prefix anymore or
553 	 * we have a less significant route.
554 	 * we've to insert an intermediate node on the list
555 	 * this new node will point to the one we need to create
556 	 * and the current
557 	 */
558 
559 	pn = fn->parent;
560 
561 	/* find 1st bit in difference between the 2 addrs.
562 
563 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
564 	   but if it is >= plen, the value is ignored in any case.
565 	 */
566 
567 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
568 
569 	/*
570 	 *		(intermediate)[in]
571 	 *	          /	   \
572 	 *	(new leaf node)[ln] (old node)[fn]
573 	 */
574 	if (plen > bit) {
575 		in = node_alloc();
576 		ln = node_alloc();
577 
578 		if (!in || !ln) {
579 			if (in)
580 				node_free(in);
581 			if (ln)
582 				node_free(ln);
583 			return ERR_PTR(-ENOMEM);
584 		}
585 
586 		/*
587 		 * new intermediate node.
588 		 * RTN_RTINFO will
589 		 * be off since that an address that chooses one of
590 		 * the branches would not match less specific routes
591 		 * in the other branch
592 		 */
593 
594 		in->fn_bit = bit;
595 
596 		in->parent = pn;
597 		in->leaf = fn->leaf;
598 		atomic_inc(&in->leaf->rt6i_ref);
599 
600 		in->fn_sernum = sernum;
601 
602 		/* update parent pointer */
603 		if (dir)
604 			pn->right = in;
605 		else
606 			pn->left  = in;
607 
608 		ln->fn_bit = plen;
609 
610 		ln->parent = in;
611 		fn->parent = in;
612 
613 		ln->fn_sernum = sernum;
614 
615 		if (addr_bit_set(addr, bit)) {
616 			in->right = ln;
617 			in->left  = fn;
618 		} else {
619 			in->left  = ln;
620 			in->right = fn;
621 		}
622 	} else { /* plen <= bit */
623 
624 		/*
625 		 *		(new leaf node)[ln]
626 		 *	          /	   \
627 		 *	     (old node)[fn] NULL
628 		 */
629 
630 		ln = node_alloc();
631 
632 		if (!ln)
633 			return ERR_PTR(-ENOMEM);
634 
635 		ln->fn_bit = plen;
636 
637 		ln->parent = pn;
638 
639 		ln->fn_sernum = sernum;
640 
641 		if (dir)
642 			pn->right = ln;
643 		else
644 			pn->left  = ln;
645 
646 		if (addr_bit_set(&key->addr, plen))
647 			ln->right = fn;
648 		else
649 			ln->left  = fn;
650 
651 		fn->parent = ln;
652 	}
653 	return ln;
654 }
655 
656 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
657 {
658 	return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
659 	       RTF_GATEWAY;
660 }
661 
662 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
663 {
664 	int i;
665 
666 	for (i = 0; i < RTAX_MAX; i++) {
667 		if (test_bit(i, mxc->mx_valid))
668 			mp[i] = mxc->mx[i];
669 	}
670 }
671 
672 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
673 {
674 	if (!mxc->mx)
675 		return 0;
676 
677 	if (dst->flags & DST_HOST) {
678 		u32 *mp = dst_metrics_write_ptr(dst);
679 
680 		if (unlikely(!mp))
681 			return -ENOMEM;
682 
683 		fib6_copy_metrics(mp, mxc);
684 	} else {
685 		dst_init_metrics(dst, mxc->mx, false);
686 
687 		/* We've stolen mx now. */
688 		mxc->mx = NULL;
689 	}
690 
691 	return 0;
692 }
693 
694 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
695 			  struct net *net)
696 {
697 	if (atomic_read(&rt->rt6i_ref) != 1) {
698 		/* This route is used as dummy address holder in some split
699 		 * nodes. It is not leaked, but it still holds other resources,
700 		 * which must be released in time. So, scan ascendant nodes
701 		 * and replace dummy references to this route with references
702 		 * to still alive ones.
703 		 */
704 		while (fn) {
705 			if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
706 				fn->leaf = fib6_find_prefix(net, fn);
707 				atomic_inc(&fn->leaf->rt6i_ref);
708 				rt6_release(rt);
709 			}
710 			fn = fn->parent;
711 		}
712 		/* No more references are possible at this point. */
713 		BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
714 	}
715 }
716 
717 /*
718  *	Insert routing information in a node.
719  */
720 
721 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
722 			    struct nl_info *info, struct mx6_config *mxc)
723 {
724 	struct rt6_info *iter = NULL;
725 	struct rt6_info **ins;
726 	struct rt6_info **fallback_ins = NULL;
727 	int replace = (info->nlh &&
728 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
729 	int add = (!info->nlh ||
730 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
731 	int found = 0;
732 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
733 	int err;
734 
735 	ins = &fn->leaf;
736 
737 	for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
738 		/*
739 		 *	Search for duplicates
740 		 */
741 
742 		if (iter->rt6i_metric == rt->rt6i_metric) {
743 			/*
744 			 *	Same priority level
745 			 */
746 			if (info->nlh &&
747 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
748 				return -EEXIST;
749 			if (replace) {
750 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
751 					found++;
752 					break;
753 				}
754 				if (rt_can_ecmp)
755 					fallback_ins = fallback_ins ?: ins;
756 				goto next_iter;
757 			}
758 
759 			if (iter->dst.dev == rt->dst.dev &&
760 			    iter->rt6i_idev == rt->rt6i_idev &&
761 			    ipv6_addr_equal(&iter->rt6i_gateway,
762 					    &rt->rt6i_gateway)) {
763 				if (rt->rt6i_nsiblings)
764 					rt->rt6i_nsiblings = 0;
765 				if (!(iter->rt6i_flags & RTF_EXPIRES))
766 					return -EEXIST;
767 				if (!(rt->rt6i_flags & RTF_EXPIRES))
768 					rt6_clean_expires(iter);
769 				else
770 					rt6_set_expires(iter, rt->dst.expires);
771 				iter->rt6i_pmtu = rt->rt6i_pmtu;
772 				return -EEXIST;
773 			}
774 			/* If we have the same destination and the same metric,
775 			 * but not the same gateway, then the route we try to
776 			 * add is sibling to this route, increment our counter
777 			 * of siblings, and later we will add our route to the
778 			 * list.
779 			 * Only static routes (which don't have flag
780 			 * RTF_EXPIRES) are used for ECMPv6.
781 			 *
782 			 * To avoid long list, we only had siblings if the
783 			 * route have a gateway.
784 			 */
785 			if (rt_can_ecmp &&
786 			    rt6_qualify_for_ecmp(iter))
787 				rt->rt6i_nsiblings++;
788 		}
789 
790 		if (iter->rt6i_metric > rt->rt6i_metric)
791 			break;
792 
793 next_iter:
794 		ins = &iter->dst.rt6_next;
795 	}
796 
797 	if (fallback_ins && !found) {
798 		/* No ECMP-able route found, replace first non-ECMP one */
799 		ins = fallback_ins;
800 		iter = *ins;
801 		found++;
802 	}
803 
804 	/* Reset round-robin state, if necessary */
805 	if (ins == &fn->leaf)
806 		fn->rr_ptr = NULL;
807 
808 	/* Link this route to others same route. */
809 	if (rt->rt6i_nsiblings) {
810 		unsigned int rt6i_nsiblings;
811 		struct rt6_info *sibling, *temp_sibling;
812 
813 		/* Find the first route that have the same metric */
814 		sibling = fn->leaf;
815 		while (sibling) {
816 			if (sibling->rt6i_metric == rt->rt6i_metric &&
817 			    rt6_qualify_for_ecmp(sibling)) {
818 				list_add_tail(&rt->rt6i_siblings,
819 					      &sibling->rt6i_siblings);
820 				break;
821 			}
822 			sibling = sibling->dst.rt6_next;
823 		}
824 		/* For each sibling in the list, increment the counter of
825 		 * siblings. BUG() if counters does not match, list of siblings
826 		 * is broken!
827 		 */
828 		rt6i_nsiblings = 0;
829 		list_for_each_entry_safe(sibling, temp_sibling,
830 					 &rt->rt6i_siblings, rt6i_siblings) {
831 			sibling->rt6i_nsiblings++;
832 			BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
833 			rt6i_nsiblings++;
834 		}
835 		BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
836 	}
837 
838 	/*
839 	 *	insert node
840 	 */
841 	if (!replace) {
842 		if (!add)
843 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
844 
845 add:
846 		err = fib6_commit_metrics(&rt->dst, mxc);
847 		if (err)
848 			return err;
849 
850 		rt->dst.rt6_next = iter;
851 		*ins = rt;
852 		rt->rt6i_node = fn;
853 		atomic_inc(&rt->rt6i_ref);
854 		inet6_rt_notify(RTM_NEWROUTE, rt, info, 0);
855 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
856 
857 		if (!(fn->fn_flags & RTN_RTINFO)) {
858 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
859 			fn->fn_flags |= RTN_RTINFO;
860 		}
861 
862 	} else {
863 		int nsiblings;
864 
865 		if (!found) {
866 			if (add)
867 				goto add;
868 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
869 			return -ENOENT;
870 		}
871 
872 		err = fib6_commit_metrics(&rt->dst, mxc);
873 		if (err)
874 			return err;
875 
876 		*ins = rt;
877 		rt->rt6i_node = fn;
878 		rt->dst.rt6_next = iter->dst.rt6_next;
879 		atomic_inc(&rt->rt6i_ref);
880 		inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
881 		if (!(fn->fn_flags & RTN_RTINFO)) {
882 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
883 			fn->fn_flags |= RTN_RTINFO;
884 		}
885 		nsiblings = iter->rt6i_nsiblings;
886 		fib6_purge_rt(iter, fn, info->nl_net);
887 		rt6_release(iter);
888 
889 		if (nsiblings) {
890 			/* Replacing an ECMP route, remove all siblings */
891 			ins = &rt->dst.rt6_next;
892 			iter = *ins;
893 			while (iter) {
894 				if (rt6_qualify_for_ecmp(iter)) {
895 					*ins = iter->dst.rt6_next;
896 					fib6_purge_rt(iter, fn, info->nl_net);
897 					rt6_release(iter);
898 					nsiblings--;
899 				} else {
900 					ins = &iter->dst.rt6_next;
901 				}
902 				iter = *ins;
903 			}
904 			WARN_ON(nsiblings != 0);
905 		}
906 	}
907 
908 	return 0;
909 }
910 
911 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
912 {
913 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
914 	    (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
915 		mod_timer(&net->ipv6.ip6_fib_timer,
916 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
917 }
918 
919 void fib6_force_start_gc(struct net *net)
920 {
921 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
922 		mod_timer(&net->ipv6.ip6_fib_timer,
923 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
924 }
925 
926 /*
927  *	Add routing information to the routing tree.
928  *	<destination addr>/<source addr>
929  *	with source addr info in sub-trees
930  */
931 
932 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
933 	     struct nl_info *info, struct mx6_config *mxc)
934 {
935 	struct fib6_node *fn, *pn = NULL;
936 	int err = -ENOMEM;
937 	int allow_create = 1;
938 	int replace_required = 0;
939 	int sernum = fib6_new_sernum(info->nl_net);
940 
941 	if (WARN_ON_ONCE((rt->dst.flags & DST_NOCACHE) &&
942 			 !atomic_read(&rt->dst.__refcnt)))
943 		return -EINVAL;
944 
945 	if (info->nlh) {
946 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
947 			allow_create = 0;
948 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
949 			replace_required = 1;
950 	}
951 	if (!allow_create && !replace_required)
952 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
953 
954 	fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
955 			offsetof(struct rt6_info, rt6i_dst), allow_create,
956 			replace_required, sernum);
957 	if (IS_ERR(fn)) {
958 		err = PTR_ERR(fn);
959 		fn = NULL;
960 		goto out;
961 	}
962 
963 	pn = fn;
964 
965 #ifdef CONFIG_IPV6_SUBTREES
966 	if (rt->rt6i_src.plen) {
967 		struct fib6_node *sn;
968 
969 		if (!fn->subtree) {
970 			struct fib6_node *sfn;
971 
972 			/*
973 			 * Create subtree.
974 			 *
975 			 *		fn[main tree]
976 			 *		|
977 			 *		sfn[subtree root]
978 			 *		   \
979 			 *		    sn[new leaf node]
980 			 */
981 
982 			/* Create subtree root node */
983 			sfn = node_alloc();
984 			if (!sfn)
985 				goto st_failure;
986 
987 			sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
988 			atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
989 			sfn->fn_flags = RTN_ROOT;
990 			sfn->fn_sernum = sernum;
991 
992 			/* Now add the first leaf node to new subtree */
993 
994 			sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
995 					rt->rt6i_src.plen,
996 					offsetof(struct rt6_info, rt6i_src),
997 					allow_create, replace_required, sernum);
998 
999 			if (IS_ERR(sn)) {
1000 				/* If it is failed, discard just allocated
1001 				   root, and then (in st_failure) stale node
1002 				   in main tree.
1003 				 */
1004 				node_free(sfn);
1005 				err = PTR_ERR(sn);
1006 				goto st_failure;
1007 			}
1008 
1009 			/* Now link new subtree to main tree */
1010 			sfn->parent = fn;
1011 			fn->subtree = sfn;
1012 		} else {
1013 			sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
1014 					rt->rt6i_src.plen,
1015 					offsetof(struct rt6_info, rt6i_src),
1016 					allow_create, replace_required, sernum);
1017 
1018 			if (IS_ERR(sn)) {
1019 				err = PTR_ERR(sn);
1020 				goto st_failure;
1021 			}
1022 		}
1023 
1024 		if (!fn->leaf) {
1025 			fn->leaf = rt;
1026 			atomic_inc(&rt->rt6i_ref);
1027 		}
1028 		fn = sn;
1029 	}
1030 #endif
1031 
1032 	err = fib6_add_rt2node(fn, rt, info, mxc);
1033 	if (!err) {
1034 		fib6_start_gc(info->nl_net, rt);
1035 		if (!(rt->rt6i_flags & RTF_CACHE))
1036 			fib6_prune_clones(info->nl_net, pn);
1037 		rt->dst.flags &= ~DST_NOCACHE;
1038 	}
1039 
1040 out:
1041 	if (err) {
1042 #ifdef CONFIG_IPV6_SUBTREES
1043 		/*
1044 		 * If fib6_add_1 has cleared the old leaf pointer in the
1045 		 * super-tree leaf node we have to find a new one for it.
1046 		 */
1047 		if (pn != fn && pn->leaf == rt) {
1048 			pn->leaf = NULL;
1049 			atomic_dec(&rt->rt6i_ref);
1050 		}
1051 		if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
1052 			pn->leaf = fib6_find_prefix(info->nl_net, pn);
1053 #if RT6_DEBUG >= 2
1054 			if (!pn->leaf) {
1055 				WARN_ON(pn->leaf == NULL);
1056 				pn->leaf = info->nl_net->ipv6.ip6_null_entry;
1057 			}
1058 #endif
1059 			atomic_inc(&pn->leaf->rt6i_ref);
1060 		}
1061 #endif
1062 		if (!(rt->dst.flags & DST_NOCACHE))
1063 			dst_free(&rt->dst);
1064 	}
1065 	return err;
1066 
1067 #ifdef CONFIG_IPV6_SUBTREES
1068 	/* Subtree creation failed, probably main tree node
1069 	   is orphan. If it is, shoot it.
1070 	 */
1071 st_failure:
1072 	if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1073 		fib6_repair_tree(info->nl_net, fn);
1074 	if (!(rt->dst.flags & DST_NOCACHE))
1075 		dst_free(&rt->dst);
1076 	return err;
1077 #endif
1078 }
1079 
1080 /*
1081  *	Routing tree lookup
1082  *
1083  */
1084 
1085 struct lookup_args {
1086 	int			offset;		/* key offset on rt6_info	*/
1087 	const struct in6_addr	*addr;		/* search key			*/
1088 };
1089 
1090 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1091 				       struct lookup_args *args)
1092 {
1093 	struct fib6_node *fn;
1094 	__be32 dir;
1095 
1096 	if (unlikely(args->offset == 0))
1097 		return NULL;
1098 
1099 	/*
1100 	 *	Descend on a tree
1101 	 */
1102 
1103 	fn = root;
1104 
1105 	for (;;) {
1106 		struct fib6_node *next;
1107 
1108 		dir = addr_bit_set(args->addr, fn->fn_bit);
1109 
1110 		next = dir ? fn->right : fn->left;
1111 
1112 		if (next) {
1113 			fn = next;
1114 			continue;
1115 		}
1116 		break;
1117 	}
1118 
1119 	while (fn) {
1120 		if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1121 			struct rt6key *key;
1122 
1123 			key = (struct rt6key *) ((u8 *) fn->leaf +
1124 						 args->offset);
1125 
1126 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1127 #ifdef CONFIG_IPV6_SUBTREES
1128 				if (fn->subtree) {
1129 					struct fib6_node *sfn;
1130 					sfn = fib6_lookup_1(fn->subtree,
1131 							    args + 1);
1132 					if (!sfn)
1133 						goto backtrack;
1134 					fn = sfn;
1135 				}
1136 #endif
1137 				if (fn->fn_flags & RTN_RTINFO)
1138 					return fn;
1139 			}
1140 		}
1141 #ifdef CONFIG_IPV6_SUBTREES
1142 backtrack:
1143 #endif
1144 		if (fn->fn_flags & RTN_ROOT)
1145 			break;
1146 
1147 		fn = fn->parent;
1148 	}
1149 
1150 	return NULL;
1151 }
1152 
1153 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1154 			      const struct in6_addr *saddr)
1155 {
1156 	struct fib6_node *fn;
1157 	struct lookup_args args[] = {
1158 		{
1159 			.offset = offsetof(struct rt6_info, rt6i_dst),
1160 			.addr = daddr,
1161 		},
1162 #ifdef CONFIG_IPV6_SUBTREES
1163 		{
1164 			.offset = offsetof(struct rt6_info, rt6i_src),
1165 			.addr = saddr,
1166 		},
1167 #endif
1168 		{
1169 			.offset = 0,	/* sentinel */
1170 		}
1171 	};
1172 
1173 	fn = fib6_lookup_1(root, daddr ? args : args + 1);
1174 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1175 		fn = root;
1176 
1177 	return fn;
1178 }
1179 
1180 /*
1181  *	Get node with specified destination prefix (and source prefix,
1182  *	if subtrees are used)
1183  */
1184 
1185 
1186 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1187 				       const struct in6_addr *addr,
1188 				       int plen, int offset)
1189 {
1190 	struct fib6_node *fn;
1191 
1192 	for (fn = root; fn ; ) {
1193 		struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1194 
1195 		/*
1196 		 *	Prefix match
1197 		 */
1198 		if (plen < fn->fn_bit ||
1199 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1200 			return NULL;
1201 
1202 		if (plen == fn->fn_bit)
1203 			return fn;
1204 
1205 		/*
1206 		 *	We have more bits to go
1207 		 */
1208 		if (addr_bit_set(addr, fn->fn_bit))
1209 			fn = fn->right;
1210 		else
1211 			fn = fn->left;
1212 	}
1213 	return NULL;
1214 }
1215 
1216 struct fib6_node *fib6_locate(struct fib6_node *root,
1217 			      const struct in6_addr *daddr, int dst_len,
1218 			      const struct in6_addr *saddr, int src_len)
1219 {
1220 	struct fib6_node *fn;
1221 
1222 	fn = fib6_locate_1(root, daddr, dst_len,
1223 			   offsetof(struct rt6_info, rt6i_dst));
1224 
1225 #ifdef CONFIG_IPV6_SUBTREES
1226 	if (src_len) {
1227 		WARN_ON(saddr == NULL);
1228 		if (fn && fn->subtree)
1229 			fn = fib6_locate_1(fn->subtree, saddr, src_len,
1230 					   offsetof(struct rt6_info, rt6i_src));
1231 	}
1232 #endif
1233 
1234 	if (fn && fn->fn_flags & RTN_RTINFO)
1235 		return fn;
1236 
1237 	return NULL;
1238 }
1239 
1240 
1241 /*
1242  *	Deletion
1243  *
1244  */
1245 
1246 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1247 {
1248 	if (fn->fn_flags & RTN_ROOT)
1249 		return net->ipv6.ip6_null_entry;
1250 
1251 	while (fn) {
1252 		if (fn->left)
1253 			return fn->left->leaf;
1254 		if (fn->right)
1255 			return fn->right->leaf;
1256 
1257 		fn = FIB6_SUBTREE(fn);
1258 	}
1259 	return NULL;
1260 }
1261 
1262 /*
1263  *	Called to trim the tree of intermediate nodes when possible. "fn"
1264  *	is the node we want to try and remove.
1265  */
1266 
1267 static struct fib6_node *fib6_repair_tree(struct net *net,
1268 					   struct fib6_node *fn)
1269 {
1270 	int children;
1271 	int nstate;
1272 	struct fib6_node *child, *pn;
1273 	struct fib6_walker *w;
1274 	int iter = 0;
1275 
1276 	for (;;) {
1277 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1278 		iter++;
1279 
1280 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1281 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1282 		WARN_ON(fn->leaf);
1283 
1284 		children = 0;
1285 		child = NULL;
1286 		if (fn->right)
1287 			child = fn->right, children |= 1;
1288 		if (fn->left)
1289 			child = fn->left, children |= 2;
1290 
1291 		if (children == 3 || FIB6_SUBTREE(fn)
1292 #ifdef CONFIG_IPV6_SUBTREES
1293 		    /* Subtree root (i.e. fn) may have one child */
1294 		    || (children && fn->fn_flags & RTN_ROOT)
1295 #endif
1296 		    ) {
1297 			fn->leaf = fib6_find_prefix(net, fn);
1298 #if RT6_DEBUG >= 2
1299 			if (!fn->leaf) {
1300 				WARN_ON(!fn->leaf);
1301 				fn->leaf = net->ipv6.ip6_null_entry;
1302 			}
1303 #endif
1304 			atomic_inc(&fn->leaf->rt6i_ref);
1305 			return fn->parent;
1306 		}
1307 
1308 		pn = fn->parent;
1309 #ifdef CONFIG_IPV6_SUBTREES
1310 		if (FIB6_SUBTREE(pn) == fn) {
1311 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1312 			FIB6_SUBTREE(pn) = NULL;
1313 			nstate = FWS_L;
1314 		} else {
1315 			WARN_ON(fn->fn_flags & RTN_ROOT);
1316 #endif
1317 			if (pn->right == fn)
1318 				pn->right = child;
1319 			else if (pn->left == fn)
1320 				pn->left = child;
1321 #if RT6_DEBUG >= 2
1322 			else
1323 				WARN_ON(1);
1324 #endif
1325 			if (child)
1326 				child->parent = pn;
1327 			nstate = FWS_R;
1328 #ifdef CONFIG_IPV6_SUBTREES
1329 		}
1330 #endif
1331 
1332 		read_lock(&fib6_walker_lock);
1333 		FOR_WALKERS(w) {
1334 			if (!child) {
1335 				if (w->root == fn) {
1336 					w->root = w->node = NULL;
1337 					RT6_TRACE("W %p adjusted by delroot 1\n", w);
1338 				} else if (w->node == fn) {
1339 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1340 					w->node = pn;
1341 					w->state = nstate;
1342 				}
1343 			} else {
1344 				if (w->root == fn) {
1345 					w->root = child;
1346 					RT6_TRACE("W %p adjusted by delroot 2\n", w);
1347 				}
1348 				if (w->node == fn) {
1349 					w->node = child;
1350 					if (children&2) {
1351 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1352 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1353 					} else {
1354 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1355 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1356 					}
1357 				}
1358 			}
1359 		}
1360 		read_unlock(&fib6_walker_lock);
1361 
1362 		node_free(fn);
1363 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1364 			return pn;
1365 
1366 		rt6_release(pn->leaf);
1367 		pn->leaf = NULL;
1368 		fn = pn;
1369 	}
1370 }
1371 
1372 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1373 			   struct nl_info *info)
1374 {
1375 	struct fib6_walker *w;
1376 	struct rt6_info *rt = *rtp;
1377 	struct net *net = info->nl_net;
1378 
1379 	RT6_TRACE("fib6_del_route\n");
1380 
1381 	/* Unlink it */
1382 	*rtp = rt->dst.rt6_next;
1383 	rt->rt6i_node = NULL;
1384 	net->ipv6.rt6_stats->fib_rt_entries--;
1385 	net->ipv6.rt6_stats->fib_discarded_routes++;
1386 
1387 	/* Reset round-robin state, if necessary */
1388 	if (fn->rr_ptr == rt)
1389 		fn->rr_ptr = NULL;
1390 
1391 	/* Remove this entry from other siblings */
1392 	if (rt->rt6i_nsiblings) {
1393 		struct rt6_info *sibling, *next_sibling;
1394 
1395 		list_for_each_entry_safe(sibling, next_sibling,
1396 					 &rt->rt6i_siblings, rt6i_siblings)
1397 			sibling->rt6i_nsiblings--;
1398 		rt->rt6i_nsiblings = 0;
1399 		list_del_init(&rt->rt6i_siblings);
1400 	}
1401 
1402 	/* Adjust walkers */
1403 	read_lock(&fib6_walker_lock);
1404 	FOR_WALKERS(w) {
1405 		if (w->state == FWS_C && w->leaf == rt) {
1406 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1407 			w->leaf = rt->dst.rt6_next;
1408 			if (!w->leaf)
1409 				w->state = FWS_U;
1410 		}
1411 	}
1412 	read_unlock(&fib6_walker_lock);
1413 
1414 	rt->dst.rt6_next = NULL;
1415 
1416 	/* If it was last route, expunge its radix tree node */
1417 	if (!fn->leaf) {
1418 		fn->fn_flags &= ~RTN_RTINFO;
1419 		net->ipv6.rt6_stats->fib_route_nodes--;
1420 		fn = fib6_repair_tree(net, fn);
1421 	}
1422 
1423 	fib6_purge_rt(rt, fn, net);
1424 
1425 	inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1426 	rt6_release(rt);
1427 }
1428 
1429 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1430 {
1431 	struct net *net = info->nl_net;
1432 	struct fib6_node *fn = rt->rt6i_node;
1433 	struct rt6_info **rtp;
1434 
1435 #if RT6_DEBUG >= 2
1436 	if (rt->dst.obsolete > 0) {
1437 		WARN_ON(fn);
1438 		return -ENOENT;
1439 	}
1440 #endif
1441 	if (!fn || rt == net->ipv6.ip6_null_entry)
1442 		return -ENOENT;
1443 
1444 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1445 
1446 	if (!(rt->rt6i_flags & RTF_CACHE)) {
1447 		struct fib6_node *pn = fn;
1448 #ifdef CONFIG_IPV6_SUBTREES
1449 		/* clones of this route might be in another subtree */
1450 		if (rt->rt6i_src.plen) {
1451 			while (!(pn->fn_flags & RTN_ROOT))
1452 				pn = pn->parent;
1453 			pn = pn->parent;
1454 		}
1455 #endif
1456 		fib6_prune_clones(info->nl_net, pn);
1457 	}
1458 
1459 	/*
1460 	 *	Walk the leaf entries looking for ourself
1461 	 */
1462 
1463 	for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1464 		if (*rtp == rt) {
1465 			fib6_del_route(fn, rtp, info);
1466 			return 0;
1467 		}
1468 	}
1469 	return -ENOENT;
1470 }
1471 
1472 /*
1473  *	Tree traversal function.
1474  *
1475  *	Certainly, it is not interrupt safe.
1476  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1477  *	It means, that we can modify tree during walking
1478  *	and use this function for garbage collection, clone pruning,
1479  *	cleaning tree when a device goes down etc. etc.
1480  *
1481  *	It guarantees that every node will be traversed,
1482  *	and that it will be traversed only once.
1483  *
1484  *	Callback function w->func may return:
1485  *	0 -> continue walking.
1486  *	positive value -> walking is suspended (used by tree dumps,
1487  *	and probably by gc, if it will be split to several slices)
1488  *	negative value -> terminate walking.
1489  *
1490  *	The function itself returns:
1491  *	0   -> walk is complete.
1492  *	>0  -> walk is incomplete (i.e. suspended)
1493  *	<0  -> walk is terminated by an error.
1494  */
1495 
1496 static int fib6_walk_continue(struct fib6_walker *w)
1497 {
1498 	struct fib6_node *fn, *pn;
1499 
1500 	for (;;) {
1501 		fn = w->node;
1502 		if (!fn)
1503 			return 0;
1504 
1505 		if (w->prune && fn != w->root &&
1506 		    fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1507 			w->state = FWS_C;
1508 			w->leaf = fn->leaf;
1509 		}
1510 		switch (w->state) {
1511 #ifdef CONFIG_IPV6_SUBTREES
1512 		case FWS_S:
1513 			if (FIB6_SUBTREE(fn)) {
1514 				w->node = FIB6_SUBTREE(fn);
1515 				continue;
1516 			}
1517 			w->state = FWS_L;
1518 #endif
1519 		case FWS_L:
1520 			if (fn->left) {
1521 				w->node = fn->left;
1522 				w->state = FWS_INIT;
1523 				continue;
1524 			}
1525 			w->state = FWS_R;
1526 		case FWS_R:
1527 			if (fn->right) {
1528 				w->node = fn->right;
1529 				w->state = FWS_INIT;
1530 				continue;
1531 			}
1532 			w->state = FWS_C;
1533 			w->leaf = fn->leaf;
1534 		case FWS_C:
1535 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1536 				int err;
1537 
1538 				if (w->skip) {
1539 					w->skip--;
1540 					goto skip;
1541 				}
1542 
1543 				err = w->func(w);
1544 				if (err)
1545 					return err;
1546 
1547 				w->count++;
1548 				continue;
1549 			}
1550 skip:
1551 			w->state = FWS_U;
1552 		case FWS_U:
1553 			if (fn == w->root)
1554 				return 0;
1555 			pn = fn->parent;
1556 			w->node = pn;
1557 #ifdef CONFIG_IPV6_SUBTREES
1558 			if (FIB6_SUBTREE(pn) == fn) {
1559 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1560 				w->state = FWS_L;
1561 				continue;
1562 			}
1563 #endif
1564 			if (pn->left == fn) {
1565 				w->state = FWS_R;
1566 				continue;
1567 			}
1568 			if (pn->right == fn) {
1569 				w->state = FWS_C;
1570 				w->leaf = w->node->leaf;
1571 				continue;
1572 			}
1573 #if RT6_DEBUG >= 2
1574 			WARN_ON(1);
1575 #endif
1576 		}
1577 	}
1578 }
1579 
1580 static int fib6_walk(struct fib6_walker *w)
1581 {
1582 	int res;
1583 
1584 	w->state = FWS_INIT;
1585 	w->node = w->root;
1586 
1587 	fib6_walker_link(w);
1588 	res = fib6_walk_continue(w);
1589 	if (res <= 0)
1590 		fib6_walker_unlink(w);
1591 	return res;
1592 }
1593 
1594 static int fib6_clean_node(struct fib6_walker *w)
1595 {
1596 	int res;
1597 	struct rt6_info *rt;
1598 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1599 	struct nl_info info = {
1600 		.nl_net = c->net,
1601 	};
1602 
1603 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1604 	    w->node->fn_sernum != c->sernum)
1605 		w->node->fn_sernum = c->sernum;
1606 
1607 	if (!c->func) {
1608 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1609 		w->leaf = NULL;
1610 		return 0;
1611 	}
1612 
1613 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1614 		res = c->func(rt, c->arg);
1615 		if (res < 0) {
1616 			w->leaf = rt;
1617 			res = fib6_del(rt, &info);
1618 			if (res) {
1619 #if RT6_DEBUG >= 2
1620 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1621 					 __func__, rt, rt->rt6i_node, res);
1622 #endif
1623 				continue;
1624 			}
1625 			return 0;
1626 		}
1627 		WARN_ON(res != 0);
1628 	}
1629 	w->leaf = rt;
1630 	return 0;
1631 }
1632 
1633 /*
1634  *	Convenient frontend to tree walker.
1635  *
1636  *	func is called on each route.
1637  *		It may return -1 -> delete this route.
1638  *		              0  -> continue walking
1639  *
1640  *	prune==1 -> only immediate children of node (certainly,
1641  *	ignoring pure split nodes) will be scanned.
1642  */
1643 
1644 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1645 			    int (*func)(struct rt6_info *, void *arg),
1646 			    bool prune, int sernum, void *arg)
1647 {
1648 	struct fib6_cleaner c;
1649 
1650 	c.w.root = root;
1651 	c.w.func = fib6_clean_node;
1652 	c.w.prune = prune;
1653 	c.w.count = 0;
1654 	c.w.skip = 0;
1655 	c.func = func;
1656 	c.sernum = sernum;
1657 	c.arg = arg;
1658 	c.net = net;
1659 
1660 	fib6_walk(&c.w);
1661 }
1662 
1663 static void __fib6_clean_all(struct net *net,
1664 			     int (*func)(struct rt6_info *, void *),
1665 			     int sernum, void *arg)
1666 {
1667 	struct fib6_table *table;
1668 	struct hlist_head *head;
1669 	unsigned int h;
1670 
1671 	rcu_read_lock();
1672 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1673 		head = &net->ipv6.fib_table_hash[h];
1674 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1675 			write_lock_bh(&table->tb6_lock);
1676 			fib6_clean_tree(net, &table->tb6_root,
1677 					func, false, sernum, arg);
1678 			write_unlock_bh(&table->tb6_lock);
1679 		}
1680 	}
1681 	rcu_read_unlock();
1682 }
1683 
1684 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1685 		    void *arg)
1686 {
1687 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1688 }
1689 
1690 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1691 {
1692 	if (rt->rt6i_flags & RTF_CACHE) {
1693 		RT6_TRACE("pruning clone %p\n", rt);
1694 		return -1;
1695 	}
1696 
1697 	return 0;
1698 }
1699 
1700 static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1701 {
1702 	fib6_clean_tree(net, fn, fib6_prune_clone, true,
1703 			FIB6_NO_SERNUM_CHANGE, NULL);
1704 }
1705 
1706 static void fib6_flush_trees(struct net *net)
1707 {
1708 	int new_sernum = fib6_new_sernum(net);
1709 
1710 	__fib6_clean_all(net, NULL, new_sernum, NULL);
1711 }
1712 
1713 /*
1714  *	Garbage collection
1715  */
1716 
1717 static struct fib6_gc_args
1718 {
1719 	int			timeout;
1720 	int			more;
1721 } gc_args;
1722 
1723 static int fib6_age(struct rt6_info *rt, void *arg)
1724 {
1725 	unsigned long now = jiffies;
1726 
1727 	/*
1728 	 *	check addrconf expiration here.
1729 	 *	Routes are expired even if they are in use.
1730 	 *
1731 	 *	Also age clones. Note, that clones are aged out
1732 	 *	only if they are not in use now.
1733 	 */
1734 
1735 	if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1736 		if (time_after(now, rt->dst.expires)) {
1737 			RT6_TRACE("expiring %p\n", rt);
1738 			return -1;
1739 		}
1740 		gc_args.more++;
1741 	} else if (rt->rt6i_flags & RTF_CACHE) {
1742 		if (atomic_read(&rt->dst.__refcnt) == 0 &&
1743 		    time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1744 			RT6_TRACE("aging clone %p\n", rt);
1745 			return -1;
1746 		} else if (rt->rt6i_flags & RTF_GATEWAY) {
1747 			struct neighbour *neigh;
1748 			__u8 neigh_flags = 0;
1749 
1750 			neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1751 			if (neigh) {
1752 				neigh_flags = neigh->flags;
1753 				neigh_release(neigh);
1754 			}
1755 			if (!(neigh_flags & NTF_ROUTER)) {
1756 				RT6_TRACE("purging route %p via non-router but gateway\n",
1757 					  rt);
1758 				return -1;
1759 			}
1760 		}
1761 		gc_args.more++;
1762 	}
1763 
1764 	return 0;
1765 }
1766 
1767 static DEFINE_SPINLOCK(fib6_gc_lock);
1768 
1769 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1770 {
1771 	unsigned long now;
1772 
1773 	if (force) {
1774 		spin_lock_bh(&fib6_gc_lock);
1775 	} else if (!spin_trylock_bh(&fib6_gc_lock)) {
1776 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1777 		return;
1778 	}
1779 	gc_args.timeout = expires ? (int)expires :
1780 			  net->ipv6.sysctl.ip6_rt_gc_interval;
1781 
1782 	gc_args.more = icmp6_dst_gc();
1783 
1784 	fib6_clean_all(net, fib6_age, NULL);
1785 	now = jiffies;
1786 	net->ipv6.ip6_rt_last_gc = now;
1787 
1788 	if (gc_args.more)
1789 		mod_timer(&net->ipv6.ip6_fib_timer,
1790 			  round_jiffies(now
1791 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
1792 	else
1793 		del_timer(&net->ipv6.ip6_fib_timer);
1794 	spin_unlock_bh(&fib6_gc_lock);
1795 }
1796 
1797 static void fib6_gc_timer_cb(unsigned long arg)
1798 {
1799 	fib6_run_gc(0, (struct net *)arg, true);
1800 }
1801 
1802 static int __net_init fib6_net_init(struct net *net)
1803 {
1804 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1805 
1806 	setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1807 
1808 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1809 	if (!net->ipv6.rt6_stats)
1810 		goto out_timer;
1811 
1812 	/* Avoid false sharing : Use at least a full cache line */
1813 	size = max_t(size_t, size, L1_CACHE_BYTES);
1814 
1815 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1816 	if (!net->ipv6.fib_table_hash)
1817 		goto out_rt6_stats;
1818 
1819 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1820 					  GFP_KERNEL);
1821 	if (!net->ipv6.fib6_main_tbl)
1822 		goto out_fib_table_hash;
1823 
1824 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1825 	net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1826 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1827 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1828 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1829 
1830 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1831 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1832 					   GFP_KERNEL);
1833 	if (!net->ipv6.fib6_local_tbl)
1834 		goto out_fib6_main_tbl;
1835 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1836 	net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1837 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1838 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1839 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1840 #endif
1841 	fib6_tables_init(net);
1842 
1843 	return 0;
1844 
1845 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1846 out_fib6_main_tbl:
1847 	kfree(net->ipv6.fib6_main_tbl);
1848 #endif
1849 out_fib_table_hash:
1850 	kfree(net->ipv6.fib_table_hash);
1851 out_rt6_stats:
1852 	kfree(net->ipv6.rt6_stats);
1853 out_timer:
1854 	return -ENOMEM;
1855 }
1856 
1857 static void fib6_net_exit(struct net *net)
1858 {
1859 	rt6_ifdown(net, NULL);
1860 	del_timer_sync(&net->ipv6.ip6_fib_timer);
1861 
1862 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1863 	inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1864 	kfree(net->ipv6.fib6_local_tbl);
1865 #endif
1866 	inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1867 	kfree(net->ipv6.fib6_main_tbl);
1868 	kfree(net->ipv6.fib_table_hash);
1869 	kfree(net->ipv6.rt6_stats);
1870 }
1871 
1872 static struct pernet_operations fib6_net_ops = {
1873 	.init = fib6_net_init,
1874 	.exit = fib6_net_exit,
1875 };
1876 
1877 int __init fib6_init(void)
1878 {
1879 	int ret = -ENOMEM;
1880 
1881 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
1882 					   sizeof(struct fib6_node),
1883 					   0, SLAB_HWCACHE_ALIGN,
1884 					   NULL);
1885 	if (!fib6_node_kmem)
1886 		goto out;
1887 
1888 	ret = register_pernet_subsys(&fib6_net_ops);
1889 	if (ret)
1890 		goto out_kmem_cache_create;
1891 
1892 	ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1893 			      NULL);
1894 	if (ret)
1895 		goto out_unregister_subsys;
1896 
1897 	__fib6_flush_trees = fib6_flush_trees;
1898 out:
1899 	return ret;
1900 
1901 out_unregister_subsys:
1902 	unregister_pernet_subsys(&fib6_net_ops);
1903 out_kmem_cache_create:
1904 	kmem_cache_destroy(fib6_node_kmem);
1905 	goto out;
1906 }
1907 
1908 void fib6_gc_cleanup(void)
1909 {
1910 	unregister_pernet_subsys(&fib6_net_ops);
1911 	kmem_cache_destroy(fib6_node_kmem);
1912 }
1913 
1914 #ifdef CONFIG_PROC_FS
1915 
1916 struct ipv6_route_iter {
1917 	struct seq_net_private p;
1918 	struct fib6_walker w;
1919 	loff_t skip;
1920 	struct fib6_table *tbl;
1921 	int sernum;
1922 };
1923 
1924 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
1925 {
1926 	struct rt6_info *rt = v;
1927 	struct ipv6_route_iter *iter = seq->private;
1928 
1929 	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
1930 
1931 #ifdef CONFIG_IPV6_SUBTREES
1932 	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
1933 #else
1934 	seq_puts(seq, "00000000000000000000000000000000 00 ");
1935 #endif
1936 	if (rt->rt6i_flags & RTF_GATEWAY)
1937 		seq_printf(seq, "%pi6", &rt->rt6i_gateway);
1938 	else
1939 		seq_puts(seq, "00000000000000000000000000000000");
1940 
1941 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
1942 		   rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
1943 		   rt->dst.__use, rt->rt6i_flags,
1944 		   rt->dst.dev ? rt->dst.dev->name : "");
1945 	iter->w.leaf = NULL;
1946 	return 0;
1947 }
1948 
1949 static int ipv6_route_yield(struct fib6_walker *w)
1950 {
1951 	struct ipv6_route_iter *iter = w->args;
1952 
1953 	if (!iter->skip)
1954 		return 1;
1955 
1956 	do {
1957 		iter->w.leaf = iter->w.leaf->dst.rt6_next;
1958 		iter->skip--;
1959 		if (!iter->skip && iter->w.leaf)
1960 			return 1;
1961 	} while (iter->w.leaf);
1962 
1963 	return 0;
1964 }
1965 
1966 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter)
1967 {
1968 	memset(&iter->w, 0, sizeof(iter->w));
1969 	iter->w.func = ipv6_route_yield;
1970 	iter->w.root = &iter->tbl->tb6_root;
1971 	iter->w.state = FWS_INIT;
1972 	iter->w.node = iter->w.root;
1973 	iter->w.args = iter;
1974 	iter->sernum = iter->w.root->fn_sernum;
1975 	INIT_LIST_HEAD(&iter->w.lh);
1976 	fib6_walker_link(&iter->w);
1977 }
1978 
1979 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
1980 						    struct net *net)
1981 {
1982 	unsigned int h;
1983 	struct hlist_node *node;
1984 
1985 	if (tbl) {
1986 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
1987 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
1988 	} else {
1989 		h = 0;
1990 		node = NULL;
1991 	}
1992 
1993 	while (!node && h < FIB6_TABLE_HASHSZ) {
1994 		node = rcu_dereference_bh(
1995 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
1996 	}
1997 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
1998 }
1999 
2000 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2001 {
2002 	if (iter->sernum != iter->w.root->fn_sernum) {
2003 		iter->sernum = iter->w.root->fn_sernum;
2004 		iter->w.state = FWS_INIT;
2005 		iter->w.node = iter->w.root;
2006 		WARN_ON(iter->w.skip);
2007 		iter->w.skip = iter->w.count;
2008 	}
2009 }
2010 
2011 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2012 {
2013 	int r;
2014 	struct rt6_info *n;
2015 	struct net *net = seq_file_net(seq);
2016 	struct ipv6_route_iter *iter = seq->private;
2017 
2018 	if (!v)
2019 		goto iter_table;
2020 
2021 	n = ((struct rt6_info *)v)->dst.rt6_next;
2022 	if (n) {
2023 		++*pos;
2024 		return n;
2025 	}
2026 
2027 iter_table:
2028 	ipv6_route_check_sernum(iter);
2029 	read_lock(&iter->tbl->tb6_lock);
2030 	r = fib6_walk_continue(&iter->w);
2031 	read_unlock(&iter->tbl->tb6_lock);
2032 	if (r > 0) {
2033 		if (v)
2034 			++*pos;
2035 		return iter->w.leaf;
2036 	} else if (r < 0) {
2037 		fib6_walker_unlink(&iter->w);
2038 		return NULL;
2039 	}
2040 	fib6_walker_unlink(&iter->w);
2041 
2042 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2043 	if (!iter->tbl)
2044 		return NULL;
2045 
2046 	ipv6_route_seq_setup_walk(iter);
2047 	goto iter_table;
2048 }
2049 
2050 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2051 	__acquires(RCU_BH)
2052 {
2053 	struct net *net = seq_file_net(seq);
2054 	struct ipv6_route_iter *iter = seq->private;
2055 
2056 	rcu_read_lock_bh();
2057 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2058 	iter->skip = *pos;
2059 
2060 	if (iter->tbl) {
2061 		ipv6_route_seq_setup_walk(iter);
2062 		return ipv6_route_seq_next(seq, NULL, pos);
2063 	} else {
2064 		return NULL;
2065 	}
2066 }
2067 
2068 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2069 {
2070 	struct fib6_walker *w = &iter->w;
2071 	return w->node && !(w->state == FWS_U && w->node == w->root);
2072 }
2073 
2074 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2075 	__releases(RCU_BH)
2076 {
2077 	struct ipv6_route_iter *iter = seq->private;
2078 
2079 	if (ipv6_route_iter_active(iter))
2080 		fib6_walker_unlink(&iter->w);
2081 
2082 	rcu_read_unlock_bh();
2083 }
2084 
2085 static const struct seq_operations ipv6_route_seq_ops = {
2086 	.start	= ipv6_route_seq_start,
2087 	.next	= ipv6_route_seq_next,
2088 	.stop	= ipv6_route_seq_stop,
2089 	.show	= ipv6_route_seq_show
2090 };
2091 
2092 int ipv6_route_open(struct inode *inode, struct file *file)
2093 {
2094 	return seq_open_net(inode, file, &ipv6_route_seq_ops,
2095 			    sizeof(struct ipv6_route_iter));
2096 }
2097 
2098 #endif /* CONFIG_PROC_FS */
2099