xref: /kernel/linux/linux-5.10/net/ipv6/ip6_fib.c (revision 8c2ecf20)
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 *	Linux INET6 implementation
4 *	Forwarding Information Database
5 *
6 *	Authors:
7 *	Pedro Roque		<roque@di.fc.ul.pt>
8 *
9 *	Changes:
10 *	Yuji SEKIYA @USAGI:	Support default route on router node;
11 *				remove ip6_null_entry from the top of
12 *				routing table.
13 *	Ville Nuorvala:		Fixed routing subtrees.
14 */
15
16#define pr_fmt(fmt) "IPv6: " fmt
17
18#include <linux/errno.h>
19#include <linux/types.h>
20#include <linux/net.h>
21#include <linux/route.h>
22#include <linux/netdevice.h>
23#include <linux/in6.h>
24#include <linux/init.h>
25#include <linux/list.h>
26#include <linux/slab.h>
27
28#include <net/ip.h>
29#include <net/ipv6.h>
30#include <net/ndisc.h>
31#include <net/addrconf.h>
32#include <net/lwtunnel.h>
33#include <net/fib_notifier.h>
34
35#include <net/ip6_fib.h>
36#include <net/ip6_route.h>
37
38static struct kmem_cache *fib6_node_kmem __read_mostly;
39
40struct fib6_cleaner {
41	struct fib6_walker w;
42	struct net *net;
43	int (*func)(struct fib6_info *, void *arg);
44	int sernum;
45	void *arg;
46	bool skip_notify;
47};
48
49#ifdef CONFIG_IPV6_SUBTREES
50#define FWS_INIT FWS_S
51#else
52#define FWS_INIT FWS_L
53#endif
54
55static struct fib6_info *fib6_find_prefix(struct net *net,
56					 struct fib6_table *table,
57					 struct fib6_node *fn);
58static struct fib6_node *fib6_repair_tree(struct net *net,
59					  struct fib6_table *table,
60					  struct fib6_node *fn);
61static int fib6_walk(struct net *net, struct fib6_walker *w);
62static int fib6_walk_continue(struct fib6_walker *w);
63
64/*
65 *	A routing update causes an increase of the serial number on the
66 *	affected subtree. This allows for cached routes to be asynchronously
67 *	tested when modifications are made to the destination cache as a
68 *	result of redirects, path MTU changes, etc.
69 */
70
71static void fib6_gc_timer_cb(struct timer_list *t);
72
73#define FOR_WALKERS(net, w) \
74	list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
75
76static void fib6_walker_link(struct net *net, struct fib6_walker *w)
77{
78	write_lock_bh(&net->ipv6.fib6_walker_lock);
79	list_add(&w->lh, &net->ipv6.fib6_walkers);
80	write_unlock_bh(&net->ipv6.fib6_walker_lock);
81}
82
83static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
84{
85	write_lock_bh(&net->ipv6.fib6_walker_lock);
86	list_del(&w->lh);
87	write_unlock_bh(&net->ipv6.fib6_walker_lock);
88}
89
90static int fib6_new_sernum(struct net *net)
91{
92	int new, old;
93
94	do {
95		old = atomic_read(&net->ipv6.fib6_sernum);
96		new = old < INT_MAX ? old + 1 : 1;
97	} while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
98				old, new) != old);
99	return new;
100}
101
102enum {
103	FIB6_NO_SERNUM_CHANGE = 0,
104};
105
106void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
107{
108	struct fib6_node *fn;
109
110	fn = rcu_dereference_protected(f6i->fib6_node,
111			lockdep_is_held(&f6i->fib6_table->tb6_lock));
112	if (fn)
113		WRITE_ONCE(fn->fn_sernum, fib6_new_sernum(net));
114}
115
116/*
117 *	Auxiliary address test functions for the radix tree.
118 *
119 *	These assume a 32bit processor (although it will work on
120 *	64bit processors)
121 */
122
123/*
124 *	test bit
125 */
126#if defined(__LITTLE_ENDIAN)
127# define BITOP_BE32_SWIZZLE	(0x1F & ~7)
128#else
129# define BITOP_BE32_SWIZZLE	0
130#endif
131
132static __be32 addr_bit_set(const void *token, int fn_bit)
133{
134	const __be32 *addr = token;
135	/*
136	 * Here,
137	 *	1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
138	 * is optimized version of
139	 *	htonl(1 << ((~fn_bit)&0x1F))
140	 * See include/asm-generic/bitops/le.h.
141	 */
142	return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
143	       addr[fn_bit >> 5];
144}
145
146struct fib6_info *fib6_info_alloc(gfp_t gfp_flags, bool with_fib6_nh)
147{
148	struct fib6_info *f6i;
149	size_t sz = sizeof(*f6i);
150
151	if (with_fib6_nh)
152		sz += sizeof(struct fib6_nh);
153
154	f6i = kzalloc(sz, gfp_flags);
155	if (!f6i)
156		return NULL;
157
158	/* fib6_siblings is a union with nh_list, so this initializes both */
159	INIT_LIST_HEAD(&f6i->fib6_siblings);
160	refcount_set(&f6i->fib6_ref, 1);
161
162	return f6i;
163}
164
165void fib6_info_destroy_rcu(struct rcu_head *head)
166{
167	struct fib6_info *f6i = container_of(head, struct fib6_info, rcu);
168
169	WARN_ON(f6i->fib6_node);
170
171	if (f6i->nh)
172		nexthop_put(f6i->nh);
173	else
174		fib6_nh_release(f6i->fib6_nh);
175
176	ip_fib_metrics_put(f6i->fib6_metrics);
177	kfree(f6i);
178}
179EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
180
181static struct fib6_node *node_alloc(struct net *net)
182{
183	struct fib6_node *fn;
184
185	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
186	if (fn)
187		net->ipv6.rt6_stats->fib_nodes++;
188
189	return fn;
190}
191
192static void node_free_immediate(struct net *net, struct fib6_node *fn)
193{
194	kmem_cache_free(fib6_node_kmem, fn);
195	net->ipv6.rt6_stats->fib_nodes--;
196}
197
198static void node_free_rcu(struct rcu_head *head)
199{
200	struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
201
202	kmem_cache_free(fib6_node_kmem, fn);
203}
204
205static void node_free(struct net *net, struct fib6_node *fn)
206{
207	call_rcu(&fn->rcu, node_free_rcu);
208	net->ipv6.rt6_stats->fib_nodes--;
209}
210
211static void fib6_free_table(struct fib6_table *table)
212{
213	inetpeer_invalidate_tree(&table->tb6_peers);
214	kfree(table);
215}
216
217static void fib6_link_table(struct net *net, struct fib6_table *tb)
218{
219	unsigned int h;
220
221	/*
222	 * Initialize table lock at a single place to give lockdep a key,
223	 * tables aren't visible prior to being linked to the list.
224	 */
225	spin_lock_init(&tb->tb6_lock);
226	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
227
228	/*
229	 * No protection necessary, this is the only list mutatation
230	 * operation, tables never disappear once they exist.
231	 */
232	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
233}
234
235#ifdef CONFIG_IPV6_MULTIPLE_TABLES
236
237static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
238{
239	struct fib6_table *table;
240
241	table = kzalloc(sizeof(*table), GFP_ATOMIC);
242	if (table) {
243		table->tb6_id = id;
244		rcu_assign_pointer(table->tb6_root.leaf,
245				   net->ipv6.fib6_null_entry);
246		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
247		inet_peer_base_init(&table->tb6_peers);
248	}
249
250	return table;
251}
252
253struct fib6_table *fib6_new_table(struct net *net, u32 id)
254{
255	struct fib6_table *tb;
256
257	if (id == 0)
258		id = RT6_TABLE_MAIN;
259	tb = fib6_get_table(net, id);
260	if (tb)
261		return tb;
262
263	tb = fib6_alloc_table(net, id);
264	if (tb)
265		fib6_link_table(net, tb);
266
267	return tb;
268}
269EXPORT_SYMBOL_GPL(fib6_new_table);
270
271struct fib6_table *fib6_get_table(struct net *net, u32 id)
272{
273	struct fib6_table *tb;
274	struct hlist_head *head;
275	unsigned int h;
276
277	if (id == 0)
278		id = RT6_TABLE_MAIN;
279	h = id & (FIB6_TABLE_HASHSZ - 1);
280	rcu_read_lock();
281	head = &net->ipv6.fib_table_hash[h];
282	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
283		if (tb->tb6_id == id) {
284			rcu_read_unlock();
285			return tb;
286		}
287	}
288	rcu_read_unlock();
289
290	return NULL;
291}
292EXPORT_SYMBOL_GPL(fib6_get_table);
293
294static void __net_init fib6_tables_init(struct net *net)
295{
296	fib6_link_table(net, net->ipv6.fib6_main_tbl);
297	fib6_link_table(net, net->ipv6.fib6_local_tbl);
298}
299#else
300
301struct fib6_table *fib6_new_table(struct net *net, u32 id)
302{
303	return fib6_get_table(net, id);
304}
305
306struct fib6_table *fib6_get_table(struct net *net, u32 id)
307{
308	  return net->ipv6.fib6_main_tbl;
309}
310
311struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
312				   const struct sk_buff *skb,
313				   int flags, pol_lookup_t lookup)
314{
315	struct rt6_info *rt;
316
317	rt = pol_lookup_func(lookup,
318			net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
319	if (rt->dst.error == -EAGAIN) {
320		ip6_rt_put_flags(rt, flags);
321		rt = net->ipv6.ip6_null_entry;
322		if (!(flags & RT6_LOOKUP_F_DST_NOREF))
323			dst_hold(&rt->dst);
324	}
325
326	return &rt->dst;
327}
328
329/* called with rcu lock held; no reference taken on fib6_info */
330int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
331		struct fib6_result *res, int flags)
332{
333	return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6,
334				 res, flags);
335}
336
337static void __net_init fib6_tables_init(struct net *net)
338{
339	fib6_link_table(net, net->ipv6.fib6_main_tbl);
340}
341
342#endif
343
344unsigned int fib6_tables_seq_read(struct net *net)
345{
346	unsigned int h, fib_seq = 0;
347
348	rcu_read_lock();
349	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
350		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
351		struct fib6_table *tb;
352
353		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
354			fib_seq += tb->fib_seq;
355	}
356	rcu_read_unlock();
357
358	return fib_seq;
359}
360
361static int call_fib6_entry_notifier(struct notifier_block *nb,
362				    enum fib_event_type event_type,
363				    struct fib6_info *rt,
364				    struct netlink_ext_ack *extack)
365{
366	struct fib6_entry_notifier_info info = {
367		.info.extack = extack,
368		.rt = rt,
369	};
370
371	return call_fib6_notifier(nb, event_type, &info.info);
372}
373
374static int call_fib6_multipath_entry_notifier(struct notifier_block *nb,
375					      enum fib_event_type event_type,
376					      struct fib6_info *rt,
377					      unsigned int nsiblings,
378					      struct netlink_ext_ack *extack)
379{
380	struct fib6_entry_notifier_info info = {
381		.info.extack = extack,
382		.rt = rt,
383		.nsiblings = nsiblings,
384	};
385
386	return call_fib6_notifier(nb, event_type, &info.info);
387}
388
389int call_fib6_entry_notifiers(struct net *net,
390			      enum fib_event_type event_type,
391			      struct fib6_info *rt,
392			      struct netlink_ext_ack *extack)
393{
394	struct fib6_entry_notifier_info info = {
395		.info.extack = extack,
396		.rt = rt,
397	};
398
399	rt->fib6_table->fib_seq++;
400	return call_fib6_notifiers(net, event_type, &info.info);
401}
402
403int call_fib6_multipath_entry_notifiers(struct net *net,
404					enum fib_event_type event_type,
405					struct fib6_info *rt,
406					unsigned int nsiblings,
407					struct netlink_ext_ack *extack)
408{
409	struct fib6_entry_notifier_info info = {
410		.info.extack = extack,
411		.rt = rt,
412		.nsiblings = nsiblings,
413	};
414
415	rt->fib6_table->fib_seq++;
416	return call_fib6_notifiers(net, event_type, &info.info);
417}
418
419int call_fib6_entry_notifiers_replace(struct net *net, struct fib6_info *rt)
420{
421	struct fib6_entry_notifier_info info = {
422		.rt = rt,
423		.nsiblings = rt->fib6_nsiblings,
424	};
425
426	rt->fib6_table->fib_seq++;
427	return call_fib6_notifiers(net, FIB_EVENT_ENTRY_REPLACE, &info.info);
428}
429
430struct fib6_dump_arg {
431	struct net *net;
432	struct notifier_block *nb;
433	struct netlink_ext_ack *extack;
434};
435
436static int fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
437{
438	enum fib_event_type fib_event = FIB_EVENT_ENTRY_REPLACE;
439	int err;
440
441	if (!rt || rt == arg->net->ipv6.fib6_null_entry)
442		return 0;
443
444	if (rt->fib6_nsiblings)
445		err = call_fib6_multipath_entry_notifier(arg->nb, fib_event,
446							 rt,
447							 rt->fib6_nsiblings,
448							 arg->extack);
449	else
450		err = call_fib6_entry_notifier(arg->nb, fib_event, rt,
451					       arg->extack);
452
453	return err;
454}
455
456static int fib6_node_dump(struct fib6_walker *w)
457{
458	int err;
459
460	err = fib6_rt_dump(w->leaf, w->args);
461	w->leaf = NULL;
462	return err;
463}
464
465static int fib6_table_dump(struct net *net, struct fib6_table *tb,
466			   struct fib6_walker *w)
467{
468	int err;
469
470	w->root = &tb->tb6_root;
471	spin_lock_bh(&tb->tb6_lock);
472	err = fib6_walk(net, w);
473	spin_unlock_bh(&tb->tb6_lock);
474	return err;
475}
476
477/* Called with rcu_read_lock() */
478int fib6_tables_dump(struct net *net, struct notifier_block *nb,
479		     struct netlink_ext_ack *extack)
480{
481	struct fib6_dump_arg arg;
482	struct fib6_walker *w;
483	unsigned int h;
484	int err = 0;
485
486	w = kzalloc(sizeof(*w), GFP_ATOMIC);
487	if (!w)
488		return -ENOMEM;
489
490	w->func = fib6_node_dump;
491	arg.net = net;
492	arg.nb = nb;
493	arg.extack = extack;
494	w->args = &arg;
495
496	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
497		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
498		struct fib6_table *tb;
499
500		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
501			err = fib6_table_dump(net, tb, w);
502			if (err)
503				goto out;
504		}
505	}
506
507out:
508	kfree(w);
509
510	/* The tree traversal function should never return a positive value. */
511	return err > 0 ? -EINVAL : err;
512}
513
514static int fib6_dump_node(struct fib6_walker *w)
515{
516	int res;
517	struct fib6_info *rt;
518
519	for_each_fib6_walker_rt(w) {
520		res = rt6_dump_route(rt, w->args, w->skip_in_node);
521		if (res >= 0) {
522			/* Frame is full, suspend walking */
523			w->leaf = rt;
524
525			/* We'll restart from this node, so if some routes were
526			 * already dumped, skip them next time.
527			 */
528			w->skip_in_node += res;
529
530			return 1;
531		}
532		w->skip_in_node = 0;
533
534		/* Multipath routes are dumped in one route with the
535		 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
536		 * last sibling of this route (no need to dump the
537		 * sibling routes again)
538		 */
539		if (rt->fib6_nsiblings)
540			rt = list_last_entry(&rt->fib6_siblings,
541					     struct fib6_info,
542					     fib6_siblings);
543	}
544	w->leaf = NULL;
545	return 0;
546}
547
548static void fib6_dump_end(struct netlink_callback *cb)
549{
550	struct net *net = sock_net(cb->skb->sk);
551	struct fib6_walker *w = (void *)cb->args[2];
552
553	if (w) {
554		if (cb->args[4]) {
555			cb->args[4] = 0;
556			fib6_walker_unlink(net, w);
557		}
558		cb->args[2] = 0;
559		kfree(w);
560	}
561	cb->done = (void *)cb->args[3];
562	cb->args[1] = 3;
563}
564
565static int fib6_dump_done(struct netlink_callback *cb)
566{
567	fib6_dump_end(cb);
568	return cb->done ? cb->done(cb) : 0;
569}
570
571static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
572			   struct netlink_callback *cb)
573{
574	struct net *net = sock_net(skb->sk);
575	struct fib6_walker *w;
576	int res;
577
578	w = (void *)cb->args[2];
579	w->root = &table->tb6_root;
580
581	if (cb->args[4] == 0) {
582		w->count = 0;
583		w->skip = 0;
584		w->skip_in_node = 0;
585
586		spin_lock_bh(&table->tb6_lock);
587		res = fib6_walk(net, w);
588		spin_unlock_bh(&table->tb6_lock);
589		if (res > 0) {
590			cb->args[4] = 1;
591			cb->args[5] = READ_ONCE(w->root->fn_sernum);
592		}
593	} else {
594		int sernum = READ_ONCE(w->root->fn_sernum);
595		if (cb->args[5] != sernum) {
596			/* Begin at the root if the tree changed */
597			cb->args[5] = sernum;
598			w->state = FWS_INIT;
599			w->node = w->root;
600			w->skip = w->count;
601			w->skip_in_node = 0;
602		} else
603			w->skip = 0;
604
605		spin_lock_bh(&table->tb6_lock);
606		res = fib6_walk_continue(w);
607		spin_unlock_bh(&table->tb6_lock);
608		if (res <= 0) {
609			fib6_walker_unlink(net, w);
610			cb->args[4] = 0;
611		}
612	}
613
614	return res;
615}
616
617static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
618{
619	struct rt6_rtnl_dump_arg arg = { .filter.dump_exceptions = true,
620					 .filter.dump_routes = true };
621	const struct nlmsghdr *nlh = cb->nlh;
622	struct net *net = sock_net(skb->sk);
623	unsigned int h, s_h;
624	unsigned int e = 0, s_e;
625	struct fib6_walker *w;
626	struct fib6_table *tb;
627	struct hlist_head *head;
628	int res = 0;
629
630	if (cb->strict_check) {
631		int err;
632
633		err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
634		if (err < 0)
635			return err;
636	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
637		struct rtmsg *rtm = nlmsg_data(nlh);
638
639		if (rtm->rtm_flags & RTM_F_PREFIX)
640			arg.filter.flags = RTM_F_PREFIX;
641	}
642
643	w = (void *)cb->args[2];
644	if (!w) {
645		/* New dump:
646		 *
647		 * 1. allocate and initialize walker.
648		 */
649		w = kzalloc(sizeof(*w), GFP_ATOMIC);
650		if (!w)
651			return -ENOMEM;
652		w->func = fib6_dump_node;
653		cb->args[2] = (long)w;
654
655		/* 2. hook callback destructor.
656		 */
657		cb->args[3] = (long)cb->done;
658		cb->done = fib6_dump_done;
659
660	}
661
662	arg.skb = skb;
663	arg.cb = cb;
664	arg.net = net;
665	w->args = &arg;
666
667	if (arg.filter.table_id) {
668		tb = fib6_get_table(net, arg.filter.table_id);
669		if (!tb) {
670			if (rtnl_msg_family(cb->nlh) != PF_INET6)
671				goto out;
672
673			NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
674			return -ENOENT;
675		}
676
677		if (!cb->args[0]) {
678			res = fib6_dump_table(tb, skb, cb);
679			if (!res)
680				cb->args[0] = 1;
681		}
682		goto out;
683	}
684
685	s_h = cb->args[0];
686	s_e = cb->args[1];
687
688	rcu_read_lock();
689	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
690		e = 0;
691		head = &net->ipv6.fib_table_hash[h];
692		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
693			if (e < s_e)
694				goto next;
695			res = fib6_dump_table(tb, skb, cb);
696			if (res != 0)
697				goto out_unlock;
698next:
699			e++;
700		}
701	}
702out_unlock:
703	rcu_read_unlock();
704	cb->args[1] = e;
705	cb->args[0] = h;
706out:
707	res = res < 0 ? res : skb->len;
708	if (res <= 0)
709		fib6_dump_end(cb);
710	return res;
711}
712
713void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
714{
715	if (!f6i)
716		return;
717
718	if (f6i->fib6_metrics == &dst_default_metrics) {
719		struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
720
721		if (!p)
722			return;
723
724		refcount_set(&p->refcnt, 1);
725		f6i->fib6_metrics = p;
726	}
727
728	f6i->fib6_metrics->metrics[metric - 1] = val;
729}
730
731/*
732 *	Routing Table
733 *
734 *	return the appropriate node for a routing tree "add" operation
735 *	by either creating and inserting or by returning an existing
736 *	node.
737 */
738
739static struct fib6_node *fib6_add_1(struct net *net,
740				    struct fib6_table *table,
741				    struct fib6_node *root,
742				    struct in6_addr *addr, int plen,
743				    int offset, int allow_create,
744				    int replace_required,
745				    struct netlink_ext_ack *extack)
746{
747	struct fib6_node *fn, *in, *ln;
748	struct fib6_node *pn = NULL;
749	struct rt6key *key;
750	int	bit;
751	__be32	dir = 0;
752
753	RT6_TRACE("fib6_add_1\n");
754
755	/* insert node in tree */
756
757	fn = root;
758
759	do {
760		struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
761					    lockdep_is_held(&table->tb6_lock));
762		key = (struct rt6key *)((u8 *)leaf + offset);
763
764		/*
765		 *	Prefix match
766		 */
767		if (plen < fn->fn_bit ||
768		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
769			if (!allow_create) {
770				if (replace_required) {
771					NL_SET_ERR_MSG(extack,
772						       "Can not replace route - no match found");
773					pr_warn("Can't replace route, no match found\n");
774					return ERR_PTR(-ENOENT);
775				}
776				pr_warn("NLM_F_CREATE should be set when creating new route\n");
777			}
778			goto insert_above;
779		}
780
781		/*
782		 *	Exact match ?
783		 */
784
785		if (plen == fn->fn_bit) {
786			/* clean up an intermediate node */
787			if (!(fn->fn_flags & RTN_RTINFO)) {
788				RCU_INIT_POINTER(fn->leaf, NULL);
789				fib6_info_release(leaf);
790			/* remove null_entry in the root node */
791			} else if (fn->fn_flags & RTN_TL_ROOT &&
792				   rcu_access_pointer(fn->leaf) ==
793				   net->ipv6.fib6_null_entry) {
794				RCU_INIT_POINTER(fn->leaf, NULL);
795			}
796
797			return fn;
798		}
799
800		/*
801		 *	We have more bits to go
802		 */
803
804		/* Try to walk down on tree. */
805		dir = addr_bit_set(addr, fn->fn_bit);
806		pn = fn;
807		fn = dir ?
808		     rcu_dereference_protected(fn->right,
809					lockdep_is_held(&table->tb6_lock)) :
810		     rcu_dereference_protected(fn->left,
811					lockdep_is_held(&table->tb6_lock));
812	} while (fn);
813
814	if (!allow_create) {
815		/* We should not create new node because
816		 * NLM_F_REPLACE was specified without NLM_F_CREATE
817		 * I assume it is safe to require NLM_F_CREATE when
818		 * REPLACE flag is used! Later we may want to remove the
819		 * check for replace_required, because according
820		 * to netlink specification, NLM_F_CREATE
821		 * MUST be specified if new route is created.
822		 * That would keep IPv6 consistent with IPv4
823		 */
824		if (replace_required) {
825			NL_SET_ERR_MSG(extack,
826				       "Can not replace route - no match found");
827			pr_warn("Can't replace route, no match found\n");
828			return ERR_PTR(-ENOENT);
829		}
830		pr_warn("NLM_F_CREATE should be set when creating new route\n");
831	}
832	/*
833	 *	We walked to the bottom of tree.
834	 *	Create new leaf node without children.
835	 */
836
837	ln = node_alloc(net);
838
839	if (!ln)
840		return ERR_PTR(-ENOMEM);
841	ln->fn_bit = plen;
842	RCU_INIT_POINTER(ln->parent, pn);
843
844	if (dir)
845		rcu_assign_pointer(pn->right, ln);
846	else
847		rcu_assign_pointer(pn->left, ln);
848
849	return ln;
850
851
852insert_above:
853	/*
854	 * split since we don't have a common prefix anymore or
855	 * we have a less significant route.
856	 * we've to insert an intermediate node on the list
857	 * this new node will point to the one we need to create
858	 * and the current
859	 */
860
861	pn = rcu_dereference_protected(fn->parent,
862				       lockdep_is_held(&table->tb6_lock));
863
864	/* find 1st bit in difference between the 2 addrs.
865
866	   See comment in __ipv6_addr_diff: bit may be an invalid value,
867	   but if it is >= plen, the value is ignored in any case.
868	 */
869
870	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
871
872	/*
873	 *		(intermediate)[in]
874	 *	          /	   \
875	 *	(new leaf node)[ln] (old node)[fn]
876	 */
877	if (plen > bit) {
878		in = node_alloc(net);
879		ln = node_alloc(net);
880
881		if (!in || !ln) {
882			if (in)
883				node_free_immediate(net, in);
884			if (ln)
885				node_free_immediate(net, ln);
886			return ERR_PTR(-ENOMEM);
887		}
888
889		/*
890		 * new intermediate node.
891		 * RTN_RTINFO will
892		 * be off since that an address that chooses one of
893		 * the branches would not match less specific routes
894		 * in the other branch
895		 */
896
897		in->fn_bit = bit;
898
899		RCU_INIT_POINTER(in->parent, pn);
900		in->leaf = fn->leaf;
901		fib6_info_hold(rcu_dereference_protected(in->leaf,
902				lockdep_is_held(&table->tb6_lock)));
903
904		/* update parent pointer */
905		if (dir)
906			rcu_assign_pointer(pn->right, in);
907		else
908			rcu_assign_pointer(pn->left, in);
909
910		ln->fn_bit = plen;
911
912		RCU_INIT_POINTER(ln->parent, in);
913		rcu_assign_pointer(fn->parent, in);
914
915		if (addr_bit_set(addr, bit)) {
916			rcu_assign_pointer(in->right, ln);
917			rcu_assign_pointer(in->left, fn);
918		} else {
919			rcu_assign_pointer(in->left, ln);
920			rcu_assign_pointer(in->right, fn);
921		}
922	} else { /* plen <= bit */
923
924		/*
925		 *		(new leaf node)[ln]
926		 *	          /	   \
927		 *	     (old node)[fn] NULL
928		 */
929
930		ln = node_alloc(net);
931
932		if (!ln)
933			return ERR_PTR(-ENOMEM);
934
935		ln->fn_bit = plen;
936
937		RCU_INIT_POINTER(ln->parent, pn);
938
939		if (addr_bit_set(&key->addr, plen))
940			RCU_INIT_POINTER(ln->right, fn);
941		else
942			RCU_INIT_POINTER(ln->left, fn);
943
944		rcu_assign_pointer(fn->parent, ln);
945
946		if (dir)
947			rcu_assign_pointer(pn->right, ln);
948		else
949			rcu_assign_pointer(pn->left, ln);
950	}
951	return ln;
952}
953
954static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
955				  const struct fib6_info *match,
956				  const struct fib6_table *table)
957{
958	int cpu;
959
960	if (!fib6_nh->rt6i_pcpu)
961		return;
962
963	rcu_read_lock();
964	/* release the reference to this fib entry from
965	 * all of its cached pcpu routes
966	 */
967	for_each_possible_cpu(cpu) {
968		struct rt6_info **ppcpu_rt;
969		struct rt6_info *pcpu_rt;
970
971		ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
972
973		/* Paired with xchg() in rt6_get_pcpu_route() */
974		pcpu_rt = READ_ONCE(*ppcpu_rt);
975
976		/* only dropping the 'from' reference if the cached route
977		 * is using 'match'. The cached pcpu_rt->from only changes
978		 * from a fib6_info to NULL (ip6_dst_destroy); it can never
979		 * change from one fib6_info reference to another
980		 */
981		if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
982			struct fib6_info *from;
983
984			from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
985			fib6_info_release(from);
986		}
987	}
988	rcu_read_unlock();
989}
990
991struct fib6_nh_pcpu_arg {
992	struct fib6_info	*from;
993	const struct fib6_table *table;
994};
995
996static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg)
997{
998	struct fib6_nh_pcpu_arg *arg = _arg;
999
1000	__fib6_drop_pcpu_from(nh, arg->from, arg->table);
1001	return 0;
1002}
1003
1004static void fib6_drop_pcpu_from(struct fib6_info *f6i,
1005				const struct fib6_table *table)
1006{
1007	/* Make sure rt6_make_pcpu_route() wont add other percpu routes
1008	 * while we are cleaning them here.
1009	 */
1010	f6i->fib6_destroying = 1;
1011	mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
1012
1013	if (f6i->nh) {
1014		struct fib6_nh_pcpu_arg arg = {
1015			.from = f6i,
1016			.table = table
1017		};
1018
1019		nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from,
1020					 &arg);
1021	} else {
1022		struct fib6_nh *fib6_nh;
1023
1024		fib6_nh = f6i->fib6_nh;
1025		__fib6_drop_pcpu_from(fib6_nh, f6i, table);
1026	}
1027}
1028
1029static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
1030			  struct net *net)
1031{
1032	struct fib6_table *table = rt->fib6_table;
1033
1034	/* Flush all cached dst in exception table */
1035	rt6_flush_exceptions(rt);
1036	fib6_drop_pcpu_from(rt, table);
1037
1038	if (rt->nh && !list_empty(&rt->nh_list))
1039		list_del_init(&rt->nh_list);
1040
1041	if (refcount_read(&rt->fib6_ref) != 1) {
1042		/* This route is used as dummy address holder in some split
1043		 * nodes. It is not leaked, but it still holds other resources,
1044		 * which must be released in time. So, scan ascendant nodes
1045		 * and replace dummy references to this route with references
1046		 * to still alive ones.
1047		 */
1048		while (fn) {
1049			struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1050					    lockdep_is_held(&table->tb6_lock));
1051			struct fib6_info *new_leaf;
1052			if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
1053				new_leaf = fib6_find_prefix(net, table, fn);
1054				fib6_info_hold(new_leaf);
1055
1056				rcu_assign_pointer(fn->leaf, new_leaf);
1057				fib6_info_release(rt);
1058			}
1059			fn = rcu_dereference_protected(fn->parent,
1060				    lockdep_is_held(&table->tb6_lock));
1061		}
1062	}
1063}
1064
1065/*
1066 *	Insert routing information in a node.
1067 */
1068
1069static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
1070			    struct nl_info *info,
1071			    struct netlink_ext_ack *extack)
1072{
1073	struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1074				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1075	struct fib6_info *iter = NULL;
1076	struct fib6_info __rcu **ins;
1077	struct fib6_info __rcu **fallback_ins = NULL;
1078	int replace = (info->nlh &&
1079		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
1080	int add = (!info->nlh ||
1081		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
1082	int found = 0;
1083	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
1084	bool notify_sibling_rt = false;
1085	u16 nlflags = NLM_F_EXCL;
1086	int err;
1087
1088	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
1089		nlflags |= NLM_F_APPEND;
1090
1091	ins = &fn->leaf;
1092
1093	for (iter = leaf; iter;
1094	     iter = rcu_dereference_protected(iter->fib6_next,
1095				lockdep_is_held(&rt->fib6_table->tb6_lock))) {
1096		/*
1097		 *	Search for duplicates
1098		 */
1099
1100		if (iter->fib6_metric == rt->fib6_metric) {
1101			/*
1102			 *	Same priority level
1103			 */
1104			if (info->nlh &&
1105			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
1106				return -EEXIST;
1107
1108			nlflags &= ~NLM_F_EXCL;
1109			if (replace) {
1110				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1111					found++;
1112					break;
1113				}
1114				fallback_ins = fallback_ins ?: ins;
1115				goto next_iter;
1116			}
1117
1118			if (rt6_duplicate_nexthop(iter, rt)) {
1119				if (rt->fib6_nsiblings)
1120					rt->fib6_nsiblings = 0;
1121				if (!(iter->fib6_flags & RTF_EXPIRES))
1122					return -EEXIST;
1123				if (!(rt->fib6_flags & RTF_EXPIRES))
1124					fib6_clean_expires(iter);
1125				else
1126					fib6_set_expires(iter, rt->expires);
1127
1128				if (rt->fib6_pmtu)
1129					fib6_metric_set(iter, RTAX_MTU,
1130							rt->fib6_pmtu);
1131				return -EEXIST;
1132			}
1133			/* If we have the same destination and the same metric,
1134			 * but not the same gateway, then the route we try to
1135			 * add is sibling to this route, increment our counter
1136			 * of siblings, and later we will add our route to the
1137			 * list.
1138			 * Only static routes (which don't have flag
1139			 * RTF_EXPIRES) are used for ECMPv6.
1140			 *
1141			 * To avoid long list, we only had siblings if the
1142			 * route have a gateway.
1143			 */
1144			if (rt_can_ecmp &&
1145			    rt6_qualify_for_ecmp(iter))
1146				rt->fib6_nsiblings++;
1147		}
1148
1149		if (iter->fib6_metric > rt->fib6_metric)
1150			break;
1151
1152next_iter:
1153		ins = &iter->fib6_next;
1154	}
1155
1156	if (fallback_ins && !found) {
1157		/* No matching route with same ecmp-able-ness found, replace
1158		 * first matching route
1159		 */
1160		ins = fallback_ins;
1161		iter = rcu_dereference_protected(*ins,
1162				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1163		found++;
1164	}
1165
1166	/* Reset round-robin state, if necessary */
1167	if (ins == &fn->leaf)
1168		fn->rr_ptr = NULL;
1169
1170	/* Link this route to others same route. */
1171	if (rt->fib6_nsiblings) {
1172		unsigned int fib6_nsiblings;
1173		struct fib6_info *sibling, *temp_sibling;
1174
1175		/* Find the first route that have the same metric */
1176		sibling = leaf;
1177		notify_sibling_rt = true;
1178		while (sibling) {
1179			if (sibling->fib6_metric == rt->fib6_metric &&
1180			    rt6_qualify_for_ecmp(sibling)) {
1181				list_add_tail(&rt->fib6_siblings,
1182					      &sibling->fib6_siblings);
1183				break;
1184			}
1185			sibling = rcu_dereference_protected(sibling->fib6_next,
1186				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1187			notify_sibling_rt = false;
1188		}
1189		/* For each sibling in the list, increment the counter of
1190		 * siblings. BUG() if counters does not match, list of siblings
1191		 * is broken!
1192		 */
1193		fib6_nsiblings = 0;
1194		list_for_each_entry_safe(sibling, temp_sibling,
1195					 &rt->fib6_siblings, fib6_siblings) {
1196			sibling->fib6_nsiblings++;
1197			BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1198			fib6_nsiblings++;
1199		}
1200		BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1201		rt6_multipath_rebalance(temp_sibling);
1202	}
1203
1204	/*
1205	 *	insert node
1206	 */
1207	if (!replace) {
1208		if (!add)
1209			pr_warn("NLM_F_CREATE should be set when creating new route\n");
1210
1211add:
1212		nlflags |= NLM_F_CREATE;
1213
1214		/* The route should only be notified if it is the first
1215		 * route in the node or if it is added as a sibling
1216		 * route to the first route in the node.
1217		 */
1218		if (!info->skip_notify_kernel &&
1219		    (notify_sibling_rt || ins == &fn->leaf)) {
1220			enum fib_event_type fib_event;
1221
1222			if (notify_sibling_rt)
1223				fib_event = FIB_EVENT_ENTRY_APPEND;
1224			else
1225				fib_event = FIB_EVENT_ENTRY_REPLACE;
1226			err = call_fib6_entry_notifiers(info->nl_net,
1227							fib_event, rt,
1228							extack);
1229			if (err) {
1230				struct fib6_info *sibling, *next_sibling;
1231
1232				/* If the route has siblings, then it first
1233				 * needs to be unlinked from them.
1234				 */
1235				if (!rt->fib6_nsiblings)
1236					return err;
1237
1238				list_for_each_entry_safe(sibling, next_sibling,
1239							 &rt->fib6_siblings,
1240							 fib6_siblings)
1241					sibling->fib6_nsiblings--;
1242				rt->fib6_nsiblings = 0;
1243				list_del_init(&rt->fib6_siblings);
1244				rt6_multipath_rebalance(next_sibling);
1245				return err;
1246			}
1247		}
1248
1249		rcu_assign_pointer(rt->fib6_next, iter);
1250		fib6_info_hold(rt);
1251		rcu_assign_pointer(rt->fib6_node, fn);
1252		rcu_assign_pointer(*ins, rt);
1253		if (!info->skip_notify)
1254			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1255		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1256
1257		if (!(fn->fn_flags & RTN_RTINFO)) {
1258			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1259			fn->fn_flags |= RTN_RTINFO;
1260		}
1261
1262	} else {
1263		int nsiblings;
1264
1265		if (!found) {
1266			if (add)
1267				goto add;
1268			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1269			return -ENOENT;
1270		}
1271
1272		if (!info->skip_notify_kernel && ins == &fn->leaf) {
1273			err = call_fib6_entry_notifiers(info->nl_net,
1274							FIB_EVENT_ENTRY_REPLACE,
1275							rt, extack);
1276			if (err)
1277				return err;
1278		}
1279
1280		fib6_info_hold(rt);
1281		rcu_assign_pointer(rt->fib6_node, fn);
1282		rt->fib6_next = iter->fib6_next;
1283		rcu_assign_pointer(*ins, rt);
1284		if (!info->skip_notify)
1285			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1286		if (!(fn->fn_flags & RTN_RTINFO)) {
1287			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1288			fn->fn_flags |= RTN_RTINFO;
1289		}
1290		nsiblings = iter->fib6_nsiblings;
1291		iter->fib6_node = NULL;
1292		fib6_purge_rt(iter, fn, info->nl_net);
1293		if (rcu_access_pointer(fn->rr_ptr) == iter)
1294			fn->rr_ptr = NULL;
1295		fib6_info_release(iter);
1296
1297		if (nsiblings) {
1298			/* Replacing an ECMP route, remove all siblings */
1299			ins = &rt->fib6_next;
1300			iter = rcu_dereference_protected(*ins,
1301				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1302			while (iter) {
1303				if (iter->fib6_metric > rt->fib6_metric)
1304					break;
1305				if (rt6_qualify_for_ecmp(iter)) {
1306					*ins = iter->fib6_next;
1307					iter->fib6_node = NULL;
1308					fib6_purge_rt(iter, fn, info->nl_net);
1309					if (rcu_access_pointer(fn->rr_ptr) == iter)
1310						fn->rr_ptr = NULL;
1311					fib6_info_release(iter);
1312					nsiblings--;
1313					info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1314				} else {
1315					ins = &iter->fib6_next;
1316				}
1317				iter = rcu_dereference_protected(*ins,
1318					lockdep_is_held(&rt->fib6_table->tb6_lock));
1319			}
1320			WARN_ON(nsiblings != 0);
1321		}
1322	}
1323
1324	return 0;
1325}
1326
1327static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1328{
1329	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1330	    (rt->fib6_flags & RTF_EXPIRES))
1331		mod_timer(&net->ipv6.ip6_fib_timer,
1332			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1333}
1334
1335void fib6_force_start_gc(struct net *net)
1336{
1337	if (!timer_pending(&net->ipv6.ip6_fib_timer))
1338		mod_timer(&net->ipv6.ip6_fib_timer,
1339			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1340}
1341
1342static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1343					   int sernum)
1344{
1345	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1346				lockdep_is_held(&rt->fib6_table->tb6_lock));
1347
1348	/* paired with smp_rmb() in rt6_get_cookie_safe() */
1349	smp_wmb();
1350	while (fn) {
1351		WRITE_ONCE(fn->fn_sernum, sernum);
1352		fn = rcu_dereference_protected(fn->parent,
1353				lockdep_is_held(&rt->fib6_table->tb6_lock));
1354	}
1355}
1356
1357void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1358{
1359	__fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1360}
1361
1362/* allow ipv4 to update sernum via ipv6_stub */
1363void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1364{
1365	spin_lock_bh(&f6i->fib6_table->tb6_lock);
1366	fib6_update_sernum_upto_root(net, f6i);
1367	spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1368}
1369
1370/*
1371 *	Add routing information to the routing tree.
1372 *	<destination addr>/<source addr>
1373 *	with source addr info in sub-trees
1374 *	Need to own table->tb6_lock
1375 */
1376
1377int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1378	     struct nl_info *info, struct netlink_ext_ack *extack)
1379{
1380	struct fib6_table *table = rt->fib6_table;
1381	struct fib6_node *fn, *pn = NULL;
1382	int err = -ENOMEM;
1383	int allow_create = 1;
1384	int replace_required = 0;
1385
1386	if (info->nlh) {
1387		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1388			allow_create = 0;
1389		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1390			replace_required = 1;
1391	}
1392	if (!allow_create && !replace_required)
1393		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1394
1395	fn = fib6_add_1(info->nl_net, table, root,
1396			&rt->fib6_dst.addr, rt->fib6_dst.plen,
1397			offsetof(struct fib6_info, fib6_dst), allow_create,
1398			replace_required, extack);
1399	if (IS_ERR(fn)) {
1400		err = PTR_ERR(fn);
1401		fn = NULL;
1402		goto out;
1403	}
1404
1405	pn = fn;
1406
1407#ifdef CONFIG_IPV6_SUBTREES
1408	if (rt->fib6_src.plen) {
1409		struct fib6_node *sn;
1410
1411		if (!rcu_access_pointer(fn->subtree)) {
1412			struct fib6_node *sfn;
1413
1414			/*
1415			 * Create subtree.
1416			 *
1417			 *		fn[main tree]
1418			 *		|
1419			 *		sfn[subtree root]
1420			 *		   \
1421			 *		    sn[new leaf node]
1422			 */
1423
1424			/* Create subtree root node */
1425			sfn = node_alloc(info->nl_net);
1426			if (!sfn)
1427				goto failure;
1428
1429			fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1430			rcu_assign_pointer(sfn->leaf,
1431					   info->nl_net->ipv6.fib6_null_entry);
1432			sfn->fn_flags = RTN_ROOT;
1433
1434			/* Now add the first leaf node to new subtree */
1435
1436			sn = fib6_add_1(info->nl_net, table, sfn,
1437					&rt->fib6_src.addr, rt->fib6_src.plen,
1438					offsetof(struct fib6_info, fib6_src),
1439					allow_create, replace_required, extack);
1440
1441			if (IS_ERR(sn)) {
1442				/* If it is failed, discard just allocated
1443				   root, and then (in failure) stale node
1444				   in main tree.
1445				 */
1446				node_free_immediate(info->nl_net, sfn);
1447				err = PTR_ERR(sn);
1448				goto failure;
1449			}
1450
1451			/* Now link new subtree to main tree */
1452			rcu_assign_pointer(sfn->parent, fn);
1453			rcu_assign_pointer(fn->subtree, sfn);
1454		} else {
1455			sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1456					&rt->fib6_src.addr, rt->fib6_src.plen,
1457					offsetof(struct fib6_info, fib6_src),
1458					allow_create, replace_required, extack);
1459
1460			if (IS_ERR(sn)) {
1461				err = PTR_ERR(sn);
1462				goto failure;
1463			}
1464		}
1465
1466		if (!rcu_access_pointer(fn->leaf)) {
1467			if (fn->fn_flags & RTN_TL_ROOT) {
1468				/* put back null_entry for root node */
1469				rcu_assign_pointer(fn->leaf,
1470					    info->nl_net->ipv6.fib6_null_entry);
1471			} else {
1472				fib6_info_hold(rt);
1473				rcu_assign_pointer(fn->leaf, rt);
1474			}
1475		}
1476		fn = sn;
1477	}
1478#endif
1479
1480	err = fib6_add_rt2node(fn, rt, info, extack);
1481	if (!err) {
1482		if (rt->nh)
1483			list_add(&rt->nh_list, &rt->nh->f6i_list);
1484		__fib6_update_sernum_upto_root(rt, fib6_new_sernum(info->nl_net));
1485		fib6_start_gc(info->nl_net, rt);
1486	}
1487
1488out:
1489	if (err) {
1490#ifdef CONFIG_IPV6_SUBTREES
1491		/*
1492		 * If fib6_add_1 has cleared the old leaf pointer in the
1493		 * super-tree leaf node we have to find a new one for it.
1494		 */
1495		if (pn != fn) {
1496			struct fib6_info *pn_leaf =
1497				rcu_dereference_protected(pn->leaf,
1498				    lockdep_is_held(&table->tb6_lock));
1499			if (pn_leaf == rt) {
1500				pn_leaf = NULL;
1501				RCU_INIT_POINTER(pn->leaf, NULL);
1502				fib6_info_release(rt);
1503			}
1504			if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1505				pn_leaf = fib6_find_prefix(info->nl_net, table,
1506							   pn);
1507				if (!pn_leaf)
1508					pn_leaf =
1509					    info->nl_net->ipv6.fib6_null_entry;
1510				fib6_info_hold(pn_leaf);
1511				rcu_assign_pointer(pn->leaf, pn_leaf);
1512			}
1513		}
1514#endif
1515		goto failure;
1516	} else if (fib6_requires_src(rt)) {
1517		fib6_routes_require_src_inc(info->nl_net);
1518	}
1519	return err;
1520
1521failure:
1522	/* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1523	 * 1. fn is an intermediate node and we failed to add the new
1524	 * route to it in both subtree creation failure and fib6_add_rt2node()
1525	 * failure case.
1526	 * 2. fn is the root node in the table and we fail to add the first
1527	 * default route to it.
1528	 */
1529	if (fn &&
1530	    (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1531	     (fn->fn_flags & RTN_TL_ROOT &&
1532	      !rcu_access_pointer(fn->leaf))))
1533		fib6_repair_tree(info->nl_net, table, fn);
1534	return err;
1535}
1536
1537/*
1538 *	Routing tree lookup
1539 *
1540 */
1541
1542struct lookup_args {
1543	int			offset;		/* key offset on fib6_info */
1544	const struct in6_addr	*addr;		/* search key			*/
1545};
1546
1547static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1548					    struct lookup_args *args)
1549{
1550	struct fib6_node *fn;
1551	__be32 dir;
1552
1553	if (unlikely(args->offset == 0))
1554		return NULL;
1555
1556	/*
1557	 *	Descend on a tree
1558	 */
1559
1560	fn = root;
1561
1562	for (;;) {
1563		struct fib6_node *next;
1564
1565		dir = addr_bit_set(args->addr, fn->fn_bit);
1566
1567		next = dir ? rcu_dereference(fn->right) :
1568			     rcu_dereference(fn->left);
1569
1570		if (next) {
1571			fn = next;
1572			continue;
1573		}
1574		break;
1575	}
1576
1577	while (fn) {
1578		struct fib6_node *subtree = FIB6_SUBTREE(fn);
1579
1580		if (subtree || fn->fn_flags & RTN_RTINFO) {
1581			struct fib6_info *leaf = rcu_dereference(fn->leaf);
1582			struct rt6key *key;
1583
1584			if (!leaf)
1585				goto backtrack;
1586
1587			key = (struct rt6key *) ((u8 *)leaf + args->offset);
1588
1589			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1590#ifdef CONFIG_IPV6_SUBTREES
1591				if (subtree) {
1592					struct fib6_node *sfn;
1593					sfn = fib6_node_lookup_1(subtree,
1594								 args + 1);
1595					if (!sfn)
1596						goto backtrack;
1597					fn = sfn;
1598				}
1599#endif
1600				if (fn->fn_flags & RTN_RTINFO)
1601					return fn;
1602			}
1603		}
1604backtrack:
1605		if (fn->fn_flags & RTN_ROOT)
1606			break;
1607
1608		fn = rcu_dereference(fn->parent);
1609	}
1610
1611	return NULL;
1612}
1613
1614/* called with rcu_read_lock() held
1615 */
1616struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1617				   const struct in6_addr *daddr,
1618				   const struct in6_addr *saddr)
1619{
1620	struct fib6_node *fn;
1621	struct lookup_args args[] = {
1622		{
1623			.offset = offsetof(struct fib6_info, fib6_dst),
1624			.addr = daddr,
1625		},
1626#ifdef CONFIG_IPV6_SUBTREES
1627		{
1628			.offset = offsetof(struct fib6_info, fib6_src),
1629			.addr = saddr,
1630		},
1631#endif
1632		{
1633			.offset = 0,	/* sentinel */
1634		}
1635	};
1636
1637	fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1638	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1639		fn = root;
1640
1641	return fn;
1642}
1643
1644/*
1645 *	Get node with specified destination prefix (and source prefix,
1646 *	if subtrees are used)
1647 *	exact_match == true means we try to find fn with exact match of
1648 *	the passed in prefix addr
1649 *	exact_match == false means we try to find fn with longest prefix
1650 *	match of the passed in prefix addr. This is useful for finding fn
1651 *	for cached route as it will be stored in the exception table under
1652 *	the node with longest prefix length.
1653 */
1654
1655
1656static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1657				       const struct in6_addr *addr,
1658				       int plen, int offset,
1659				       bool exact_match)
1660{
1661	struct fib6_node *fn, *prev = NULL;
1662
1663	for (fn = root; fn ; ) {
1664		struct fib6_info *leaf = rcu_dereference(fn->leaf);
1665		struct rt6key *key;
1666
1667		/* This node is being deleted */
1668		if (!leaf) {
1669			if (plen <= fn->fn_bit)
1670				goto out;
1671			else
1672				goto next;
1673		}
1674
1675		key = (struct rt6key *)((u8 *)leaf + offset);
1676
1677		/*
1678		 *	Prefix match
1679		 */
1680		if (plen < fn->fn_bit ||
1681		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1682			goto out;
1683
1684		if (plen == fn->fn_bit)
1685			return fn;
1686
1687		if (fn->fn_flags & RTN_RTINFO)
1688			prev = fn;
1689
1690next:
1691		/*
1692		 *	We have more bits to go
1693		 */
1694		if (addr_bit_set(addr, fn->fn_bit))
1695			fn = rcu_dereference(fn->right);
1696		else
1697			fn = rcu_dereference(fn->left);
1698	}
1699out:
1700	if (exact_match)
1701		return NULL;
1702	else
1703		return prev;
1704}
1705
1706struct fib6_node *fib6_locate(struct fib6_node *root,
1707			      const struct in6_addr *daddr, int dst_len,
1708			      const struct in6_addr *saddr, int src_len,
1709			      bool exact_match)
1710{
1711	struct fib6_node *fn;
1712
1713	fn = fib6_locate_1(root, daddr, dst_len,
1714			   offsetof(struct fib6_info, fib6_dst),
1715			   exact_match);
1716
1717#ifdef CONFIG_IPV6_SUBTREES
1718	if (src_len) {
1719		WARN_ON(saddr == NULL);
1720		if (fn) {
1721			struct fib6_node *subtree = FIB6_SUBTREE(fn);
1722
1723			if (subtree) {
1724				fn = fib6_locate_1(subtree, saddr, src_len,
1725					   offsetof(struct fib6_info, fib6_src),
1726					   exact_match);
1727			}
1728		}
1729	}
1730#endif
1731
1732	if (fn && fn->fn_flags & RTN_RTINFO)
1733		return fn;
1734
1735	return NULL;
1736}
1737
1738
1739/*
1740 *	Deletion
1741 *
1742 */
1743
1744static struct fib6_info *fib6_find_prefix(struct net *net,
1745					 struct fib6_table *table,
1746					 struct fib6_node *fn)
1747{
1748	struct fib6_node *child_left, *child_right;
1749
1750	if (fn->fn_flags & RTN_ROOT)
1751		return net->ipv6.fib6_null_entry;
1752
1753	while (fn) {
1754		child_left = rcu_dereference_protected(fn->left,
1755				    lockdep_is_held(&table->tb6_lock));
1756		child_right = rcu_dereference_protected(fn->right,
1757				    lockdep_is_held(&table->tb6_lock));
1758		if (child_left)
1759			return rcu_dereference_protected(child_left->leaf,
1760					lockdep_is_held(&table->tb6_lock));
1761		if (child_right)
1762			return rcu_dereference_protected(child_right->leaf,
1763					lockdep_is_held(&table->tb6_lock));
1764
1765		fn = FIB6_SUBTREE(fn);
1766	}
1767	return NULL;
1768}
1769
1770/*
1771 *	Called to trim the tree of intermediate nodes when possible. "fn"
1772 *	is the node we want to try and remove.
1773 *	Need to own table->tb6_lock
1774 */
1775
1776static struct fib6_node *fib6_repair_tree(struct net *net,
1777					  struct fib6_table *table,
1778					  struct fib6_node *fn)
1779{
1780	int children;
1781	int nstate;
1782	struct fib6_node *child;
1783	struct fib6_walker *w;
1784	int iter = 0;
1785
1786	/* Set fn->leaf to null_entry for root node. */
1787	if (fn->fn_flags & RTN_TL_ROOT) {
1788		rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1789		return fn;
1790	}
1791
1792	for (;;) {
1793		struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1794					    lockdep_is_held(&table->tb6_lock));
1795		struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1796					    lockdep_is_held(&table->tb6_lock));
1797		struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1798					    lockdep_is_held(&table->tb6_lock));
1799		struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1800					    lockdep_is_held(&table->tb6_lock));
1801		struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1802					    lockdep_is_held(&table->tb6_lock));
1803		struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1804					    lockdep_is_held(&table->tb6_lock));
1805		struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1806					    lockdep_is_held(&table->tb6_lock));
1807		struct fib6_info *new_fn_leaf;
1808
1809		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1810		iter++;
1811
1812		WARN_ON(fn->fn_flags & RTN_RTINFO);
1813		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1814		WARN_ON(fn_leaf);
1815
1816		children = 0;
1817		child = NULL;
1818		if (fn_r) {
1819			child = fn_r;
1820			children |= 1;
1821		}
1822		if (fn_l) {
1823			child = fn_l;
1824			children |= 2;
1825		}
1826
1827		if (children == 3 || FIB6_SUBTREE(fn)
1828#ifdef CONFIG_IPV6_SUBTREES
1829		    /* Subtree root (i.e. fn) may have one child */
1830		    || (children && fn->fn_flags & RTN_ROOT)
1831#endif
1832		    ) {
1833			new_fn_leaf = fib6_find_prefix(net, table, fn);
1834#if RT6_DEBUG >= 2
1835			if (!new_fn_leaf) {
1836				WARN_ON(!new_fn_leaf);
1837				new_fn_leaf = net->ipv6.fib6_null_entry;
1838			}
1839#endif
1840			fib6_info_hold(new_fn_leaf);
1841			rcu_assign_pointer(fn->leaf, new_fn_leaf);
1842			return pn;
1843		}
1844
1845#ifdef CONFIG_IPV6_SUBTREES
1846		if (FIB6_SUBTREE(pn) == fn) {
1847			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1848			RCU_INIT_POINTER(pn->subtree, NULL);
1849			nstate = FWS_L;
1850		} else {
1851			WARN_ON(fn->fn_flags & RTN_ROOT);
1852#endif
1853			if (pn_r == fn)
1854				rcu_assign_pointer(pn->right, child);
1855			else if (pn_l == fn)
1856				rcu_assign_pointer(pn->left, child);
1857#if RT6_DEBUG >= 2
1858			else
1859				WARN_ON(1);
1860#endif
1861			if (child)
1862				rcu_assign_pointer(child->parent, pn);
1863			nstate = FWS_R;
1864#ifdef CONFIG_IPV6_SUBTREES
1865		}
1866#endif
1867
1868		read_lock(&net->ipv6.fib6_walker_lock);
1869		FOR_WALKERS(net, w) {
1870			if (!child) {
1871				if (w->node == fn) {
1872					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1873					w->node = pn;
1874					w->state = nstate;
1875				}
1876			} else {
1877				if (w->node == fn) {
1878					w->node = child;
1879					if (children&2) {
1880						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1881						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1882					} else {
1883						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1884						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1885					}
1886				}
1887			}
1888		}
1889		read_unlock(&net->ipv6.fib6_walker_lock);
1890
1891		node_free(net, fn);
1892		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1893			return pn;
1894
1895		RCU_INIT_POINTER(pn->leaf, NULL);
1896		fib6_info_release(pn_leaf);
1897		fn = pn;
1898	}
1899}
1900
1901static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1902			   struct fib6_info __rcu **rtp, struct nl_info *info)
1903{
1904	struct fib6_info *leaf, *replace_rt = NULL;
1905	struct fib6_walker *w;
1906	struct fib6_info *rt = rcu_dereference_protected(*rtp,
1907				    lockdep_is_held(&table->tb6_lock));
1908	struct net *net = info->nl_net;
1909	bool notify_del = false;
1910
1911	RT6_TRACE("fib6_del_route\n");
1912
1913	/* If the deleted route is the first in the node and it is not part of
1914	 * a multipath route, then we need to replace it with the next route
1915	 * in the node, if exists.
1916	 */
1917	leaf = rcu_dereference_protected(fn->leaf,
1918					 lockdep_is_held(&table->tb6_lock));
1919	if (leaf == rt && !rt->fib6_nsiblings) {
1920		if (rcu_access_pointer(rt->fib6_next))
1921			replace_rt = rcu_dereference_protected(rt->fib6_next,
1922					    lockdep_is_held(&table->tb6_lock));
1923		else
1924			notify_del = true;
1925	}
1926
1927	/* Unlink it */
1928	*rtp = rt->fib6_next;
1929	rt->fib6_node = NULL;
1930	net->ipv6.rt6_stats->fib_rt_entries--;
1931	net->ipv6.rt6_stats->fib_discarded_routes++;
1932
1933	/* Reset round-robin state, if necessary */
1934	if (rcu_access_pointer(fn->rr_ptr) == rt)
1935		fn->rr_ptr = NULL;
1936
1937	/* Remove this entry from other siblings */
1938	if (rt->fib6_nsiblings) {
1939		struct fib6_info *sibling, *next_sibling;
1940
1941		/* The route is deleted from a multipath route. If this
1942		 * multipath route is the first route in the node, then we need
1943		 * to emit a delete notification. Otherwise, we need to skip
1944		 * the notification.
1945		 */
1946		if (rt->fib6_metric == leaf->fib6_metric &&
1947		    rt6_qualify_for_ecmp(leaf))
1948			notify_del = true;
1949		list_for_each_entry_safe(sibling, next_sibling,
1950					 &rt->fib6_siblings, fib6_siblings)
1951			sibling->fib6_nsiblings--;
1952		rt->fib6_nsiblings = 0;
1953		list_del_init(&rt->fib6_siblings);
1954		rt6_multipath_rebalance(next_sibling);
1955	}
1956
1957	/* Adjust walkers */
1958	read_lock(&net->ipv6.fib6_walker_lock);
1959	FOR_WALKERS(net, w) {
1960		if (w->state == FWS_C && w->leaf == rt) {
1961			RT6_TRACE("walker %p adjusted by delroute\n", w);
1962			w->leaf = rcu_dereference_protected(rt->fib6_next,
1963					    lockdep_is_held(&table->tb6_lock));
1964			if (!w->leaf)
1965				w->state = FWS_U;
1966		}
1967	}
1968	read_unlock(&net->ipv6.fib6_walker_lock);
1969
1970	/* If it was last route, call fib6_repair_tree() to:
1971	 * 1. For root node, put back null_entry as how the table was created.
1972	 * 2. For other nodes, expunge its radix tree node.
1973	 */
1974	if (!rcu_access_pointer(fn->leaf)) {
1975		if (!(fn->fn_flags & RTN_TL_ROOT)) {
1976			fn->fn_flags &= ~RTN_RTINFO;
1977			net->ipv6.rt6_stats->fib_route_nodes--;
1978		}
1979		fn = fib6_repair_tree(net, table, fn);
1980	}
1981
1982	fib6_purge_rt(rt, fn, net);
1983
1984	if (!info->skip_notify_kernel) {
1985		if (notify_del)
1986			call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1987						  rt, NULL);
1988		else if (replace_rt)
1989			call_fib6_entry_notifiers_replace(net, replace_rt);
1990	}
1991	if (!info->skip_notify)
1992		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1993
1994	fib6_info_release(rt);
1995}
1996
1997/* Need to own table->tb6_lock */
1998int fib6_del(struct fib6_info *rt, struct nl_info *info)
1999{
2000	struct net *net = info->nl_net;
2001	struct fib6_info __rcu **rtp;
2002	struct fib6_info __rcu **rtp_next;
2003	struct fib6_table *table;
2004	struct fib6_node *fn;
2005
2006	if (rt == net->ipv6.fib6_null_entry)
2007		return -ENOENT;
2008
2009	table = rt->fib6_table;
2010	fn = rcu_dereference_protected(rt->fib6_node,
2011				       lockdep_is_held(&table->tb6_lock));
2012	if (!fn)
2013		return -ENOENT;
2014
2015	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
2016
2017	/*
2018	 *	Walk the leaf entries looking for ourself
2019	 */
2020
2021	for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
2022		struct fib6_info *cur = rcu_dereference_protected(*rtp,
2023					lockdep_is_held(&table->tb6_lock));
2024		if (rt == cur) {
2025			if (fib6_requires_src(cur))
2026				fib6_routes_require_src_dec(info->nl_net);
2027			fib6_del_route(table, fn, rtp, info);
2028			return 0;
2029		}
2030		rtp_next = &cur->fib6_next;
2031	}
2032	return -ENOENT;
2033}
2034
2035/*
2036 *	Tree traversal function.
2037 *
2038 *	Certainly, it is not interrupt safe.
2039 *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
2040 *	It means, that we can modify tree during walking
2041 *	and use this function for garbage collection, clone pruning,
2042 *	cleaning tree when a device goes down etc. etc.
2043 *
2044 *	It guarantees that every node will be traversed,
2045 *	and that it will be traversed only once.
2046 *
2047 *	Callback function w->func may return:
2048 *	0 -> continue walking.
2049 *	positive value -> walking is suspended (used by tree dumps,
2050 *	and probably by gc, if it will be split to several slices)
2051 *	negative value -> terminate walking.
2052 *
2053 *	The function itself returns:
2054 *	0   -> walk is complete.
2055 *	>0  -> walk is incomplete (i.e. suspended)
2056 *	<0  -> walk is terminated by an error.
2057 *
2058 *	This function is called with tb6_lock held.
2059 */
2060
2061static int fib6_walk_continue(struct fib6_walker *w)
2062{
2063	struct fib6_node *fn, *pn, *left, *right;
2064
2065	/* w->root should always be table->tb6_root */
2066	WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
2067
2068	for (;;) {
2069		fn = w->node;
2070		if (!fn)
2071			return 0;
2072
2073		switch (w->state) {
2074#ifdef CONFIG_IPV6_SUBTREES
2075		case FWS_S:
2076			if (FIB6_SUBTREE(fn)) {
2077				w->node = FIB6_SUBTREE(fn);
2078				continue;
2079			}
2080			w->state = FWS_L;
2081			fallthrough;
2082#endif
2083		case FWS_L:
2084			left = rcu_dereference_protected(fn->left, 1);
2085			if (left) {
2086				w->node = left;
2087				w->state = FWS_INIT;
2088				continue;
2089			}
2090			w->state = FWS_R;
2091			fallthrough;
2092		case FWS_R:
2093			right = rcu_dereference_protected(fn->right, 1);
2094			if (right) {
2095				w->node = right;
2096				w->state = FWS_INIT;
2097				continue;
2098			}
2099			w->state = FWS_C;
2100			w->leaf = rcu_dereference_protected(fn->leaf, 1);
2101			fallthrough;
2102		case FWS_C:
2103			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
2104				int err;
2105
2106				if (w->skip) {
2107					w->skip--;
2108					goto skip;
2109				}
2110
2111				err = w->func(w);
2112				if (err)
2113					return err;
2114
2115				w->count++;
2116				continue;
2117			}
2118skip:
2119			w->state = FWS_U;
2120			fallthrough;
2121		case FWS_U:
2122			if (fn == w->root)
2123				return 0;
2124			pn = rcu_dereference_protected(fn->parent, 1);
2125			left = rcu_dereference_protected(pn->left, 1);
2126			right = rcu_dereference_protected(pn->right, 1);
2127			w->node = pn;
2128#ifdef CONFIG_IPV6_SUBTREES
2129			if (FIB6_SUBTREE(pn) == fn) {
2130				WARN_ON(!(fn->fn_flags & RTN_ROOT));
2131				w->state = FWS_L;
2132				continue;
2133			}
2134#endif
2135			if (left == fn) {
2136				w->state = FWS_R;
2137				continue;
2138			}
2139			if (right == fn) {
2140				w->state = FWS_C;
2141				w->leaf = rcu_dereference_protected(w->node->leaf, 1);
2142				continue;
2143			}
2144#if RT6_DEBUG >= 2
2145			WARN_ON(1);
2146#endif
2147		}
2148	}
2149}
2150
2151static int fib6_walk(struct net *net, struct fib6_walker *w)
2152{
2153	int res;
2154
2155	w->state = FWS_INIT;
2156	w->node = w->root;
2157
2158	fib6_walker_link(net, w);
2159	res = fib6_walk_continue(w);
2160	if (res <= 0)
2161		fib6_walker_unlink(net, w);
2162	return res;
2163}
2164
2165static int fib6_clean_node(struct fib6_walker *w)
2166{
2167	int res;
2168	struct fib6_info *rt;
2169	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
2170	struct nl_info info = {
2171		.nl_net = c->net,
2172		.skip_notify = c->skip_notify,
2173	};
2174
2175	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
2176	    READ_ONCE(w->node->fn_sernum) != c->sernum)
2177		WRITE_ONCE(w->node->fn_sernum, c->sernum);
2178
2179	if (!c->func) {
2180		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
2181		w->leaf = NULL;
2182		return 0;
2183	}
2184
2185	for_each_fib6_walker_rt(w) {
2186		res = c->func(rt, c->arg);
2187		if (res == -1) {
2188			w->leaf = rt;
2189			res = fib6_del(rt, &info);
2190			if (res) {
2191#if RT6_DEBUG >= 2
2192				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2193					 __func__, rt,
2194					 rcu_access_pointer(rt->fib6_node),
2195					 res);
2196#endif
2197				continue;
2198			}
2199			return 0;
2200		} else if (res == -2) {
2201			if (WARN_ON(!rt->fib6_nsiblings))
2202				continue;
2203			rt = list_last_entry(&rt->fib6_siblings,
2204					     struct fib6_info, fib6_siblings);
2205			continue;
2206		}
2207		WARN_ON(res != 0);
2208	}
2209	w->leaf = rt;
2210	return 0;
2211}
2212
2213/*
2214 *	Convenient frontend to tree walker.
2215 *
2216 *	func is called on each route.
2217 *		It may return -2 -> skip multipath route.
2218 *			      -1 -> delete this route.
2219 *		              0  -> continue walking
2220 */
2221
2222static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2223			    int (*func)(struct fib6_info *, void *arg),
2224			    int sernum, void *arg, bool skip_notify)
2225{
2226	struct fib6_cleaner c;
2227
2228	c.w.root = root;
2229	c.w.func = fib6_clean_node;
2230	c.w.count = 0;
2231	c.w.skip = 0;
2232	c.w.skip_in_node = 0;
2233	c.func = func;
2234	c.sernum = sernum;
2235	c.arg = arg;
2236	c.net = net;
2237	c.skip_notify = skip_notify;
2238
2239	fib6_walk(net, &c.w);
2240}
2241
2242static void __fib6_clean_all(struct net *net,
2243			     int (*func)(struct fib6_info *, void *),
2244			     int sernum, void *arg, bool skip_notify)
2245{
2246	struct fib6_table *table;
2247	struct hlist_head *head;
2248	unsigned int h;
2249
2250	rcu_read_lock();
2251	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2252		head = &net->ipv6.fib_table_hash[h];
2253		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2254			spin_lock_bh(&table->tb6_lock);
2255			fib6_clean_tree(net, &table->tb6_root,
2256					func, sernum, arg, skip_notify);
2257			spin_unlock_bh(&table->tb6_lock);
2258		}
2259	}
2260	rcu_read_unlock();
2261}
2262
2263void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2264		    void *arg)
2265{
2266	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2267}
2268
2269void fib6_clean_all_skip_notify(struct net *net,
2270				int (*func)(struct fib6_info *, void *),
2271				void *arg)
2272{
2273	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2274}
2275
2276static void fib6_flush_trees(struct net *net)
2277{
2278	int new_sernum = fib6_new_sernum(net);
2279
2280	__fib6_clean_all(net, NULL, new_sernum, NULL, false);
2281}
2282
2283/*
2284 *	Garbage collection
2285 */
2286
2287static int fib6_age(struct fib6_info *rt, void *arg)
2288{
2289	struct fib6_gc_args *gc_args = arg;
2290	unsigned long now = jiffies;
2291
2292	/*
2293	 *	check addrconf expiration here.
2294	 *	Routes are expired even if they are in use.
2295	 */
2296
2297	if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2298		if (time_after(now, rt->expires)) {
2299			RT6_TRACE("expiring %p\n", rt);
2300			return -1;
2301		}
2302		gc_args->more++;
2303	}
2304
2305	/*	Also age clones in the exception table.
2306	 *	Note, that clones are aged out
2307	 *	only if they are not in use now.
2308	 */
2309	rt6_age_exceptions(rt, gc_args, now);
2310
2311	return 0;
2312}
2313
2314void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2315{
2316	struct fib6_gc_args gc_args;
2317	unsigned long now;
2318
2319	if (force) {
2320		spin_lock_bh(&net->ipv6.fib6_gc_lock);
2321	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2322		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2323		return;
2324	}
2325	gc_args.timeout = expires ? (int)expires :
2326			  net->ipv6.sysctl.ip6_rt_gc_interval;
2327	gc_args.more = 0;
2328
2329	fib6_clean_all(net, fib6_age, &gc_args);
2330	now = jiffies;
2331	net->ipv6.ip6_rt_last_gc = now;
2332
2333	if (gc_args.more)
2334		mod_timer(&net->ipv6.ip6_fib_timer,
2335			  round_jiffies(now
2336					+ net->ipv6.sysctl.ip6_rt_gc_interval));
2337	else
2338		del_timer(&net->ipv6.ip6_fib_timer);
2339	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2340}
2341
2342static void fib6_gc_timer_cb(struct timer_list *t)
2343{
2344	struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2345
2346	fib6_run_gc(0, arg, true);
2347}
2348
2349static int __net_init fib6_net_init(struct net *net)
2350{
2351	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2352	int err;
2353
2354	err = fib6_notifier_init(net);
2355	if (err)
2356		return err;
2357
2358	spin_lock_init(&net->ipv6.fib6_gc_lock);
2359	rwlock_init(&net->ipv6.fib6_walker_lock);
2360	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2361	timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2362
2363	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2364	if (!net->ipv6.rt6_stats)
2365		goto out_timer;
2366
2367	/* Avoid false sharing : Use at least a full cache line */
2368	size = max_t(size_t, size, L1_CACHE_BYTES);
2369
2370	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2371	if (!net->ipv6.fib_table_hash)
2372		goto out_rt6_stats;
2373
2374	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2375					  GFP_KERNEL);
2376	if (!net->ipv6.fib6_main_tbl)
2377		goto out_fib_table_hash;
2378
2379	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2380	rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2381			   net->ipv6.fib6_null_entry);
2382	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2383		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2384	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2385
2386#ifdef CONFIG_IPV6_MULTIPLE_TABLES
2387	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2388					   GFP_KERNEL);
2389	if (!net->ipv6.fib6_local_tbl)
2390		goto out_fib6_main_tbl;
2391	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2392	rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2393			   net->ipv6.fib6_null_entry);
2394	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2395		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2396	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2397#endif
2398	fib6_tables_init(net);
2399
2400	return 0;
2401
2402#ifdef CONFIG_IPV6_MULTIPLE_TABLES
2403out_fib6_main_tbl:
2404	kfree(net->ipv6.fib6_main_tbl);
2405#endif
2406out_fib_table_hash:
2407	kfree(net->ipv6.fib_table_hash);
2408out_rt6_stats:
2409	kfree(net->ipv6.rt6_stats);
2410out_timer:
2411	fib6_notifier_exit(net);
2412	return -ENOMEM;
2413}
2414
2415static void fib6_net_exit(struct net *net)
2416{
2417	unsigned int i;
2418
2419	del_timer_sync(&net->ipv6.ip6_fib_timer);
2420
2421	for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2422		struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2423		struct hlist_node *tmp;
2424		struct fib6_table *tb;
2425
2426		hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2427			hlist_del(&tb->tb6_hlist);
2428			fib6_free_table(tb);
2429		}
2430	}
2431
2432	kfree(net->ipv6.fib_table_hash);
2433	kfree(net->ipv6.rt6_stats);
2434	fib6_notifier_exit(net);
2435}
2436
2437static struct pernet_operations fib6_net_ops = {
2438	.init = fib6_net_init,
2439	.exit = fib6_net_exit,
2440};
2441
2442int __init fib6_init(void)
2443{
2444	int ret = -ENOMEM;
2445
2446	fib6_node_kmem = kmem_cache_create("fib6_nodes",
2447					   sizeof(struct fib6_node),
2448					   0, SLAB_HWCACHE_ALIGN,
2449					   NULL);
2450	if (!fib6_node_kmem)
2451		goto out;
2452
2453	ret = register_pernet_subsys(&fib6_net_ops);
2454	if (ret)
2455		goto out_kmem_cache_create;
2456
2457	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2458				   inet6_dump_fib, 0);
2459	if (ret)
2460		goto out_unregister_subsys;
2461
2462	__fib6_flush_trees = fib6_flush_trees;
2463out:
2464	return ret;
2465
2466out_unregister_subsys:
2467	unregister_pernet_subsys(&fib6_net_ops);
2468out_kmem_cache_create:
2469	kmem_cache_destroy(fib6_node_kmem);
2470	goto out;
2471}
2472
2473void fib6_gc_cleanup(void)
2474{
2475	unregister_pernet_subsys(&fib6_net_ops);
2476	kmem_cache_destroy(fib6_node_kmem);
2477}
2478
2479#ifdef CONFIG_PROC_FS
2480static int ipv6_route_native_seq_show(struct seq_file *seq, void *v)
2481{
2482	struct fib6_info *rt = v;
2483	struct ipv6_route_iter *iter = seq->private;
2484	struct fib6_nh *fib6_nh = rt->fib6_nh;
2485	unsigned int flags = rt->fib6_flags;
2486	const struct net_device *dev;
2487
2488	if (rt->nh)
2489		fib6_nh = nexthop_fib6_nh_bh(rt->nh);
2490
2491	seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2492
2493#ifdef CONFIG_IPV6_SUBTREES
2494	seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2495#else
2496	seq_puts(seq, "00000000000000000000000000000000 00 ");
2497#endif
2498	if (fib6_nh->fib_nh_gw_family) {
2499		flags |= RTF_GATEWAY;
2500		seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6);
2501	} else {
2502		seq_puts(seq, "00000000000000000000000000000000");
2503	}
2504
2505	dev = fib6_nh->fib_nh_dev;
2506	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2507		   rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2508		   flags, dev ? dev->name : "");
2509	iter->w.leaf = NULL;
2510	return 0;
2511}
2512
2513static int ipv6_route_yield(struct fib6_walker *w)
2514{
2515	struct ipv6_route_iter *iter = w->args;
2516
2517	if (!iter->skip)
2518		return 1;
2519
2520	do {
2521		iter->w.leaf = rcu_dereference_protected(
2522				iter->w.leaf->fib6_next,
2523				lockdep_is_held(&iter->tbl->tb6_lock));
2524		iter->skip--;
2525		if (!iter->skip && iter->w.leaf)
2526			return 1;
2527	} while (iter->w.leaf);
2528
2529	return 0;
2530}
2531
2532static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2533				      struct net *net)
2534{
2535	memset(&iter->w, 0, sizeof(iter->w));
2536	iter->w.func = ipv6_route_yield;
2537	iter->w.root = &iter->tbl->tb6_root;
2538	iter->w.state = FWS_INIT;
2539	iter->w.node = iter->w.root;
2540	iter->w.args = iter;
2541	iter->sernum = READ_ONCE(iter->w.root->fn_sernum);
2542	INIT_LIST_HEAD(&iter->w.lh);
2543	fib6_walker_link(net, &iter->w);
2544}
2545
2546static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2547						    struct net *net)
2548{
2549	unsigned int h;
2550	struct hlist_node *node;
2551
2552	if (tbl) {
2553		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2554		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2555	} else {
2556		h = 0;
2557		node = NULL;
2558	}
2559
2560	while (!node && h < FIB6_TABLE_HASHSZ) {
2561		node = rcu_dereference_bh(
2562			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2563	}
2564	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2565}
2566
2567static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2568{
2569	int sernum = READ_ONCE(iter->w.root->fn_sernum);
2570
2571	if (iter->sernum != sernum) {
2572		iter->sernum = sernum;
2573		iter->w.state = FWS_INIT;
2574		iter->w.node = iter->w.root;
2575		WARN_ON(iter->w.skip);
2576		iter->w.skip = iter->w.count;
2577	}
2578}
2579
2580static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2581{
2582	int r;
2583	struct fib6_info *n;
2584	struct net *net = seq_file_net(seq);
2585	struct ipv6_route_iter *iter = seq->private;
2586
2587	++(*pos);
2588	if (!v)
2589		goto iter_table;
2590
2591	n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2592	if (n)
2593		return n;
2594
2595iter_table:
2596	ipv6_route_check_sernum(iter);
2597	spin_lock_bh(&iter->tbl->tb6_lock);
2598	r = fib6_walk_continue(&iter->w);
2599	spin_unlock_bh(&iter->tbl->tb6_lock);
2600	if (r > 0) {
2601		return iter->w.leaf;
2602	} else if (r < 0) {
2603		fib6_walker_unlink(net, &iter->w);
2604		return NULL;
2605	}
2606	fib6_walker_unlink(net, &iter->w);
2607
2608	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2609	if (!iter->tbl)
2610		return NULL;
2611
2612	ipv6_route_seq_setup_walk(iter, net);
2613	goto iter_table;
2614}
2615
2616static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2617	__acquires(RCU_BH)
2618{
2619	struct net *net = seq_file_net(seq);
2620	struct ipv6_route_iter *iter = seq->private;
2621
2622	rcu_read_lock_bh();
2623	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2624	iter->skip = *pos;
2625
2626	if (iter->tbl) {
2627		loff_t p = 0;
2628
2629		ipv6_route_seq_setup_walk(iter, net);
2630		return ipv6_route_seq_next(seq, NULL, &p);
2631	} else {
2632		return NULL;
2633	}
2634}
2635
2636static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2637{
2638	struct fib6_walker *w = &iter->w;
2639	return w->node && !(w->state == FWS_U && w->node == w->root);
2640}
2641
2642static void ipv6_route_native_seq_stop(struct seq_file *seq, void *v)
2643	__releases(RCU_BH)
2644{
2645	struct net *net = seq_file_net(seq);
2646	struct ipv6_route_iter *iter = seq->private;
2647
2648	if (ipv6_route_iter_active(iter))
2649		fib6_walker_unlink(net, &iter->w);
2650
2651	rcu_read_unlock_bh();
2652}
2653
2654#if IS_BUILTIN(CONFIG_IPV6) && defined(CONFIG_BPF_SYSCALL)
2655static int ipv6_route_prog_seq_show(struct bpf_prog *prog,
2656				    struct bpf_iter_meta *meta,
2657				    void *v)
2658{
2659	struct bpf_iter__ipv6_route ctx;
2660
2661	ctx.meta = meta;
2662	ctx.rt = v;
2663	return bpf_iter_run_prog(prog, &ctx);
2664}
2665
2666static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2667{
2668	struct ipv6_route_iter *iter = seq->private;
2669	struct bpf_iter_meta meta;
2670	struct bpf_prog *prog;
2671	int ret;
2672
2673	meta.seq = seq;
2674	prog = bpf_iter_get_info(&meta, false);
2675	if (!prog)
2676		return ipv6_route_native_seq_show(seq, v);
2677
2678	ret = ipv6_route_prog_seq_show(prog, &meta, v);
2679	iter->w.leaf = NULL;
2680
2681	return ret;
2682}
2683
2684static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2685{
2686	struct bpf_iter_meta meta;
2687	struct bpf_prog *prog;
2688
2689	if (!v) {
2690		meta.seq = seq;
2691		prog = bpf_iter_get_info(&meta, true);
2692		if (prog)
2693			(void)ipv6_route_prog_seq_show(prog, &meta, v);
2694	}
2695
2696	ipv6_route_native_seq_stop(seq, v);
2697}
2698#else
2699static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2700{
2701	return ipv6_route_native_seq_show(seq, v);
2702}
2703
2704static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2705{
2706	ipv6_route_native_seq_stop(seq, v);
2707}
2708#endif
2709
2710const struct seq_operations ipv6_route_seq_ops = {
2711	.start	= ipv6_route_seq_start,
2712	.next	= ipv6_route_seq_next,
2713	.stop	= ipv6_route_seq_stop,
2714	.show	= ipv6_route_seq_show
2715};
2716#endif /* CONFIG_PROC_FS */
2717