xref: /linux/net/ipv6/route.c (revision 1553a1c48281243359a9529a10ddb551f3b967ab)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *	Linux INET6 implementation
4  *	FIB front-end.
5  *
6  *	Authors:
7  *	Pedro Roque		<roque@di.fc.ul.pt>
8  */
9 
10 /*	Changes:
11  *
12  *	YOSHIFUJI Hideaki @USAGI
13  *		reworked default router selection.
14  *		- respect outgoing interface
15  *		- select from (probably) reachable routers (i.e.
16  *		routers in REACHABLE, STALE, DELAY or PROBE states).
17  *		- always select the same router if it is (probably)
18  *		reachable.  otherwise, round-robin the list.
19  *	Ville Nuorvala
20  *		Fixed routing subtrees.
21  */
22 
23 #define pr_fmt(fmt) "IPv6: " fmt
24 
25 #include <linux/capability.h>
26 #include <linux/errno.h>
27 #include <linux/export.h>
28 #include <linux/types.h>
29 #include <linux/times.h>
30 #include <linux/socket.h>
31 #include <linux/sockios.h>
32 #include <linux/net.h>
33 #include <linux/route.h>
34 #include <linux/netdevice.h>
35 #include <linux/in6.h>
36 #include <linux/mroute6.h>
37 #include <linux/init.h>
38 #include <linux/if_arp.h>
39 #include <linux/proc_fs.h>
40 #include <linux/seq_file.h>
41 #include <linux/nsproxy.h>
42 #include <linux/slab.h>
43 #include <linux/jhash.h>
44 #include <linux/siphash.h>
45 #include <net/net_namespace.h>
46 #include <net/snmp.h>
47 #include <net/ipv6.h>
48 #include <net/ip6_fib.h>
49 #include <net/ip6_route.h>
50 #include <net/ndisc.h>
51 #include <net/addrconf.h>
52 #include <net/tcp.h>
53 #include <linux/rtnetlink.h>
54 #include <net/dst.h>
55 #include <net/dst_metadata.h>
56 #include <net/xfrm.h>
57 #include <net/netevent.h>
58 #include <net/netlink.h>
59 #include <net/rtnh.h>
60 #include <net/lwtunnel.h>
61 #include <net/ip_tunnels.h>
62 #include <net/l3mdev.h>
63 #include <net/ip.h>
64 #include <linux/uaccess.h>
65 #include <linux/btf_ids.h>
66 
67 #ifdef CONFIG_SYSCTL
68 #include <linux/sysctl.h>
69 #endif
70 
71 static int ip6_rt_type_to_error(u8 fib6_type);
72 
73 #define CREATE_TRACE_POINTS
74 #include <trace/events/fib6.h>
75 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
76 #undef CREATE_TRACE_POINTS
77 
78 enum rt6_nud_state {
79 	RT6_NUD_FAIL_HARD = -3,
80 	RT6_NUD_FAIL_PROBE = -2,
81 	RT6_NUD_FAIL_DO_RR = -1,
82 	RT6_NUD_SUCCEED = 1
83 };
84 
85 INDIRECT_CALLABLE_SCOPE
86 struct dst_entry	*ip6_dst_check(struct dst_entry *dst, u32 cookie);
87 static unsigned int	 ip6_default_advmss(const struct dst_entry *dst);
88 INDIRECT_CALLABLE_SCOPE
89 unsigned int		ip6_mtu(const struct dst_entry *dst);
90 static struct dst_entry *ip6_negative_advice(struct dst_entry *);
91 static void		ip6_dst_destroy(struct dst_entry *);
92 static void		ip6_dst_ifdown(struct dst_entry *,
93 				       struct net_device *dev);
94 static void		 ip6_dst_gc(struct dst_ops *ops);
95 
96 static int		ip6_pkt_discard(struct sk_buff *skb);
97 static int		ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
98 static int		ip6_pkt_prohibit(struct sk_buff *skb);
99 static int		ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb);
100 static void		ip6_link_failure(struct sk_buff *skb);
101 static void		ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
102 					   struct sk_buff *skb, u32 mtu,
103 					   bool confirm_neigh);
104 static void		rt6_do_redirect(struct dst_entry *dst, struct sock *sk,
105 					struct sk_buff *skb);
106 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
107 			   int strict);
108 static size_t rt6_nlmsg_size(struct fib6_info *f6i);
109 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
110 			 struct fib6_info *rt, struct dst_entry *dst,
111 			 struct in6_addr *dest, struct in6_addr *src,
112 			 int iif, int type, u32 portid, u32 seq,
113 			 unsigned int flags);
114 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
115 					   const struct in6_addr *daddr,
116 					   const struct in6_addr *saddr);
117 
118 #ifdef CONFIG_IPV6_ROUTE_INFO
119 static struct fib6_info *rt6_add_route_info(struct net *net,
120 					   const struct in6_addr *prefix, int prefixlen,
121 					   const struct in6_addr *gwaddr,
122 					   struct net_device *dev,
123 					   unsigned int pref);
124 static struct fib6_info *rt6_get_route_info(struct net *net,
125 					   const struct in6_addr *prefix, int prefixlen,
126 					   const struct in6_addr *gwaddr,
127 					   struct net_device *dev);
128 #endif
129 
130 struct uncached_list {
131 	spinlock_t		lock;
132 	struct list_head	head;
133 	struct list_head	quarantine;
134 };
135 
136 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
137 
138 void rt6_uncached_list_add(struct rt6_info *rt)
139 {
140 	struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
141 
142 	rt->dst.rt_uncached_list = ul;
143 
144 	spin_lock_bh(&ul->lock);
145 	list_add_tail(&rt->dst.rt_uncached, &ul->head);
146 	spin_unlock_bh(&ul->lock);
147 }
148 
149 void rt6_uncached_list_del(struct rt6_info *rt)
150 {
151 	if (!list_empty(&rt->dst.rt_uncached)) {
152 		struct uncached_list *ul = rt->dst.rt_uncached_list;
153 
154 		spin_lock_bh(&ul->lock);
155 		list_del_init(&rt->dst.rt_uncached);
156 		spin_unlock_bh(&ul->lock);
157 	}
158 }
159 
160 static void rt6_uncached_list_flush_dev(struct net_device *dev)
161 {
162 	int cpu;
163 
164 	for_each_possible_cpu(cpu) {
165 		struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
166 		struct rt6_info *rt, *safe;
167 
168 		if (list_empty(&ul->head))
169 			continue;
170 
171 		spin_lock_bh(&ul->lock);
172 		list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
173 			struct inet6_dev *rt_idev = rt->rt6i_idev;
174 			struct net_device *rt_dev = rt->dst.dev;
175 			bool handled = false;
176 
177 			if (rt_idev->dev == dev) {
178 				rt->rt6i_idev = in6_dev_get(blackhole_netdev);
179 				in6_dev_put(rt_idev);
180 				handled = true;
181 			}
182 
183 			if (rt_dev == dev) {
184 				rt->dst.dev = blackhole_netdev;
185 				netdev_ref_replace(rt_dev, blackhole_netdev,
186 						   &rt->dst.dev_tracker,
187 						   GFP_ATOMIC);
188 				handled = true;
189 			}
190 			if (handled)
191 				list_move(&rt->dst.rt_uncached,
192 					  &ul->quarantine);
193 		}
194 		spin_unlock_bh(&ul->lock);
195 	}
196 }
197 
198 static inline const void *choose_neigh_daddr(const struct in6_addr *p,
199 					     struct sk_buff *skb,
200 					     const void *daddr)
201 {
202 	if (!ipv6_addr_any(p))
203 		return (const void *) p;
204 	else if (skb)
205 		return &ipv6_hdr(skb)->daddr;
206 	return daddr;
207 }
208 
209 struct neighbour *ip6_neigh_lookup(const struct in6_addr *gw,
210 				   struct net_device *dev,
211 				   struct sk_buff *skb,
212 				   const void *daddr)
213 {
214 	struct neighbour *n;
215 
216 	daddr = choose_neigh_daddr(gw, skb, daddr);
217 	n = __ipv6_neigh_lookup(dev, daddr);
218 	if (n)
219 		return n;
220 
221 	n = neigh_create(&nd_tbl, daddr, dev);
222 	return IS_ERR(n) ? NULL : n;
223 }
224 
225 static struct neighbour *ip6_dst_neigh_lookup(const struct dst_entry *dst,
226 					      struct sk_buff *skb,
227 					      const void *daddr)
228 {
229 	const struct rt6_info *rt = container_of(dst, struct rt6_info, dst);
230 
231 	return ip6_neigh_lookup(rt6_nexthop(rt, &in6addr_any),
232 				dst->dev, skb, daddr);
233 }
234 
235 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
236 {
237 	struct net_device *dev = dst->dev;
238 	struct rt6_info *rt = (struct rt6_info *)dst;
239 
240 	daddr = choose_neigh_daddr(rt6_nexthop(rt, &in6addr_any), NULL, daddr);
241 	if (!daddr)
242 		return;
243 	if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
244 		return;
245 	if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
246 		return;
247 	__ipv6_confirm_neigh(dev, daddr);
248 }
249 
250 static struct dst_ops ip6_dst_ops_template = {
251 	.family			=	AF_INET6,
252 	.gc			=	ip6_dst_gc,
253 	.gc_thresh		=	1024,
254 	.check			=	ip6_dst_check,
255 	.default_advmss		=	ip6_default_advmss,
256 	.mtu			=	ip6_mtu,
257 	.cow_metrics		=	dst_cow_metrics_generic,
258 	.destroy		=	ip6_dst_destroy,
259 	.ifdown			=	ip6_dst_ifdown,
260 	.negative_advice	=	ip6_negative_advice,
261 	.link_failure		=	ip6_link_failure,
262 	.update_pmtu		=	ip6_rt_update_pmtu,
263 	.redirect		=	rt6_do_redirect,
264 	.local_out		=	__ip6_local_out,
265 	.neigh_lookup		=	ip6_dst_neigh_lookup,
266 	.confirm_neigh		=	ip6_confirm_neigh,
267 };
268 
269 static struct dst_ops ip6_dst_blackhole_ops = {
270 	.family			= AF_INET6,
271 	.default_advmss		= ip6_default_advmss,
272 	.neigh_lookup		= ip6_dst_neigh_lookup,
273 	.check			= ip6_dst_check,
274 	.destroy		= ip6_dst_destroy,
275 	.cow_metrics		= dst_cow_metrics_generic,
276 	.update_pmtu		= dst_blackhole_update_pmtu,
277 	.redirect		= dst_blackhole_redirect,
278 	.mtu			= dst_blackhole_mtu,
279 };
280 
281 static const u32 ip6_template_metrics[RTAX_MAX] = {
282 	[RTAX_HOPLIMIT - 1] = 0,
283 };
284 
285 static const struct fib6_info fib6_null_entry_template = {
286 	.fib6_flags	= (RTF_REJECT | RTF_NONEXTHOP),
287 	.fib6_protocol  = RTPROT_KERNEL,
288 	.fib6_metric	= ~(u32)0,
289 	.fib6_ref	= REFCOUNT_INIT(1),
290 	.fib6_type	= RTN_UNREACHABLE,
291 	.fib6_metrics	= (struct dst_metrics *)&dst_default_metrics,
292 };
293 
294 static const struct rt6_info ip6_null_entry_template = {
295 	.dst = {
296 		.__rcuref	= RCUREF_INIT(1),
297 		.__use		= 1,
298 		.obsolete	= DST_OBSOLETE_FORCE_CHK,
299 		.error		= -ENETUNREACH,
300 		.input		= ip6_pkt_discard,
301 		.output		= ip6_pkt_discard_out,
302 	},
303 	.rt6i_flags	= (RTF_REJECT | RTF_NONEXTHOP),
304 };
305 
306 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
307 
308 static const struct rt6_info ip6_prohibit_entry_template = {
309 	.dst = {
310 		.__rcuref	= RCUREF_INIT(1),
311 		.__use		= 1,
312 		.obsolete	= DST_OBSOLETE_FORCE_CHK,
313 		.error		= -EACCES,
314 		.input		= ip6_pkt_prohibit,
315 		.output		= ip6_pkt_prohibit_out,
316 	},
317 	.rt6i_flags	= (RTF_REJECT | RTF_NONEXTHOP),
318 };
319 
320 static const struct rt6_info ip6_blk_hole_entry_template = {
321 	.dst = {
322 		.__rcuref	= RCUREF_INIT(1),
323 		.__use		= 1,
324 		.obsolete	= DST_OBSOLETE_FORCE_CHK,
325 		.error		= -EINVAL,
326 		.input		= dst_discard,
327 		.output		= dst_discard_out,
328 	},
329 	.rt6i_flags	= (RTF_REJECT | RTF_NONEXTHOP),
330 };
331 
332 #endif
333 
334 static void rt6_info_init(struct rt6_info *rt)
335 {
336 	memset_after(rt, 0, dst);
337 }
338 
339 /* allocate dst with ip6_dst_ops */
340 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
341 			       int flags)
342 {
343 	struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
344 					DST_OBSOLETE_FORCE_CHK, flags);
345 
346 	if (rt) {
347 		rt6_info_init(rt);
348 		atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
349 	}
350 
351 	return rt;
352 }
353 EXPORT_SYMBOL(ip6_dst_alloc);
354 
355 static void ip6_dst_destroy(struct dst_entry *dst)
356 {
357 	struct rt6_info *rt = (struct rt6_info *)dst;
358 	struct fib6_info *from;
359 	struct inet6_dev *idev;
360 
361 	ip_dst_metrics_put(dst);
362 	rt6_uncached_list_del(rt);
363 
364 	idev = rt->rt6i_idev;
365 	if (idev) {
366 		rt->rt6i_idev = NULL;
367 		in6_dev_put(idev);
368 	}
369 
370 	from = xchg((__force struct fib6_info **)&rt->from, NULL);
371 	fib6_info_release(from);
372 }
373 
374 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
375 {
376 	struct rt6_info *rt = (struct rt6_info *)dst;
377 	struct inet6_dev *idev = rt->rt6i_idev;
378 
379 	if (idev && idev->dev != blackhole_netdev) {
380 		struct inet6_dev *blackhole_idev = in6_dev_get(blackhole_netdev);
381 
382 		if (blackhole_idev) {
383 			rt->rt6i_idev = blackhole_idev;
384 			in6_dev_put(idev);
385 		}
386 	}
387 }
388 
389 static bool __rt6_check_expired(const struct rt6_info *rt)
390 {
391 	if (rt->rt6i_flags & RTF_EXPIRES)
392 		return time_after(jiffies, rt->dst.expires);
393 	else
394 		return false;
395 }
396 
397 static bool rt6_check_expired(const struct rt6_info *rt)
398 {
399 	struct fib6_info *from;
400 
401 	from = rcu_dereference(rt->from);
402 
403 	if (rt->rt6i_flags & RTF_EXPIRES) {
404 		if (time_after(jiffies, rt->dst.expires))
405 			return true;
406 	} else if (from) {
407 		return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
408 			fib6_check_expired(from);
409 	}
410 	return false;
411 }
412 
413 void fib6_select_path(const struct net *net, struct fib6_result *res,
414 		      struct flowi6 *fl6, int oif, bool have_oif_match,
415 		      const struct sk_buff *skb, int strict)
416 {
417 	struct fib6_info *sibling, *next_sibling;
418 	struct fib6_info *match = res->f6i;
419 
420 	if (!match->nh && (!match->fib6_nsiblings || have_oif_match))
421 		goto out;
422 
423 	if (match->nh && have_oif_match && res->nh)
424 		return;
425 
426 	if (skb)
427 		IP6CB(skb)->flags |= IP6SKB_MULTIPATH;
428 
429 	/* We might have already computed the hash for ICMPv6 errors. In such
430 	 * case it will always be non-zero. Otherwise now is the time to do it.
431 	 */
432 	if (!fl6->mp_hash &&
433 	    (!match->nh || nexthop_is_multipath(match->nh)))
434 		fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
435 
436 	if (unlikely(match->nh)) {
437 		nexthop_path_fib6_result(res, fl6->mp_hash);
438 		return;
439 	}
440 
441 	if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
442 		goto out;
443 
444 	list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
445 				 fib6_siblings) {
446 		const struct fib6_nh *nh = sibling->fib6_nh;
447 		int nh_upper_bound;
448 
449 		nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
450 		if (fl6->mp_hash > nh_upper_bound)
451 			continue;
452 		if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
453 			break;
454 		match = sibling;
455 		break;
456 	}
457 
458 out:
459 	res->f6i = match;
460 	res->nh = match->fib6_nh;
461 }
462 
463 /*
464  *	Route lookup. rcu_read_lock() should be held.
465  */
466 
467 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
468 			       const struct in6_addr *saddr, int oif, int flags)
469 {
470 	const struct net_device *dev;
471 
472 	if (nh->fib_nh_flags & RTNH_F_DEAD)
473 		return false;
474 
475 	dev = nh->fib_nh_dev;
476 	if (oif) {
477 		if (dev->ifindex == oif)
478 			return true;
479 	} else {
480 		if (ipv6_chk_addr(net, saddr, dev,
481 				  flags & RT6_LOOKUP_F_IFACE))
482 			return true;
483 	}
484 
485 	return false;
486 }
487 
488 struct fib6_nh_dm_arg {
489 	struct net		*net;
490 	const struct in6_addr	*saddr;
491 	int			oif;
492 	int			flags;
493 	struct fib6_nh		*nh;
494 };
495 
496 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
497 {
498 	struct fib6_nh_dm_arg *arg = _arg;
499 
500 	arg->nh = nh;
501 	return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
502 				  arg->flags);
503 }
504 
505 /* returns fib6_nh from nexthop or NULL */
506 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
507 					struct fib6_result *res,
508 					const struct in6_addr *saddr,
509 					int oif, int flags)
510 {
511 	struct fib6_nh_dm_arg arg = {
512 		.net   = net,
513 		.saddr = saddr,
514 		.oif   = oif,
515 		.flags = flags,
516 	};
517 
518 	if (nexthop_is_blackhole(nh))
519 		return NULL;
520 
521 	if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
522 		return arg.nh;
523 
524 	return NULL;
525 }
526 
527 static void rt6_device_match(struct net *net, struct fib6_result *res,
528 			     const struct in6_addr *saddr, int oif, int flags)
529 {
530 	struct fib6_info *f6i = res->f6i;
531 	struct fib6_info *spf6i;
532 	struct fib6_nh *nh;
533 
534 	if (!oif && ipv6_addr_any(saddr)) {
535 		if (unlikely(f6i->nh)) {
536 			nh = nexthop_fib6_nh(f6i->nh);
537 			if (nexthop_is_blackhole(f6i->nh))
538 				goto out_blackhole;
539 		} else {
540 			nh = f6i->fib6_nh;
541 		}
542 		if (!(nh->fib_nh_flags & RTNH_F_DEAD))
543 			goto out;
544 	}
545 
546 	for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
547 		bool matched = false;
548 
549 		if (unlikely(spf6i->nh)) {
550 			nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
551 					      oif, flags);
552 			if (nh)
553 				matched = true;
554 		} else {
555 			nh = spf6i->fib6_nh;
556 			if (__rt6_device_match(net, nh, saddr, oif, flags))
557 				matched = true;
558 		}
559 		if (matched) {
560 			res->f6i = spf6i;
561 			goto out;
562 		}
563 	}
564 
565 	if (oif && flags & RT6_LOOKUP_F_IFACE) {
566 		res->f6i = net->ipv6.fib6_null_entry;
567 		nh = res->f6i->fib6_nh;
568 		goto out;
569 	}
570 
571 	if (unlikely(f6i->nh)) {
572 		nh = nexthop_fib6_nh(f6i->nh);
573 		if (nexthop_is_blackhole(f6i->nh))
574 			goto out_blackhole;
575 	} else {
576 		nh = f6i->fib6_nh;
577 	}
578 
579 	if (nh->fib_nh_flags & RTNH_F_DEAD) {
580 		res->f6i = net->ipv6.fib6_null_entry;
581 		nh = res->f6i->fib6_nh;
582 	}
583 out:
584 	res->nh = nh;
585 	res->fib6_type = res->f6i->fib6_type;
586 	res->fib6_flags = res->f6i->fib6_flags;
587 	return;
588 
589 out_blackhole:
590 	res->fib6_flags |= RTF_REJECT;
591 	res->fib6_type = RTN_BLACKHOLE;
592 	res->nh = nh;
593 }
594 
595 #ifdef CONFIG_IPV6_ROUTER_PREF
596 struct __rt6_probe_work {
597 	struct work_struct work;
598 	struct in6_addr target;
599 	struct net_device *dev;
600 	netdevice_tracker dev_tracker;
601 };
602 
603 static void rt6_probe_deferred(struct work_struct *w)
604 {
605 	struct in6_addr mcaddr;
606 	struct __rt6_probe_work *work =
607 		container_of(w, struct __rt6_probe_work, work);
608 
609 	addrconf_addr_solict_mult(&work->target, &mcaddr);
610 	ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
611 	netdev_put(work->dev, &work->dev_tracker);
612 	kfree(work);
613 }
614 
615 static void rt6_probe(struct fib6_nh *fib6_nh)
616 {
617 	struct __rt6_probe_work *work = NULL;
618 	const struct in6_addr *nh_gw;
619 	unsigned long last_probe;
620 	struct neighbour *neigh;
621 	struct net_device *dev;
622 	struct inet6_dev *idev;
623 
624 	/*
625 	 * Okay, this does not seem to be appropriate
626 	 * for now, however, we need to check if it
627 	 * is really so; aka Router Reachability Probing.
628 	 *
629 	 * Router Reachability Probe MUST be rate-limited
630 	 * to no more than one per minute.
631 	 */
632 	if (!fib6_nh->fib_nh_gw_family)
633 		return;
634 
635 	nh_gw = &fib6_nh->fib_nh_gw6;
636 	dev = fib6_nh->fib_nh_dev;
637 	rcu_read_lock();
638 	last_probe = READ_ONCE(fib6_nh->last_probe);
639 	idev = __in6_dev_get(dev);
640 	neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
641 	if (neigh) {
642 		if (READ_ONCE(neigh->nud_state) & NUD_VALID)
643 			goto out;
644 
645 		write_lock_bh(&neigh->lock);
646 		if (!(neigh->nud_state & NUD_VALID) &&
647 		    time_after(jiffies,
648 			       neigh->updated +
649 			       READ_ONCE(idev->cnf.rtr_probe_interval))) {
650 			work = kmalloc(sizeof(*work), GFP_ATOMIC);
651 			if (work)
652 				__neigh_set_probe_once(neigh);
653 		}
654 		write_unlock_bh(&neigh->lock);
655 	} else if (time_after(jiffies, last_probe +
656 				       READ_ONCE(idev->cnf.rtr_probe_interval))) {
657 		work = kmalloc(sizeof(*work), GFP_ATOMIC);
658 	}
659 
660 	if (!work || cmpxchg(&fib6_nh->last_probe,
661 			     last_probe, jiffies) != last_probe) {
662 		kfree(work);
663 	} else {
664 		INIT_WORK(&work->work, rt6_probe_deferred);
665 		work->target = *nh_gw;
666 		netdev_hold(dev, &work->dev_tracker, GFP_ATOMIC);
667 		work->dev = dev;
668 		schedule_work(&work->work);
669 	}
670 
671 out:
672 	rcu_read_unlock();
673 }
674 #else
675 static inline void rt6_probe(struct fib6_nh *fib6_nh)
676 {
677 }
678 #endif
679 
680 /*
681  * Default Router Selection (RFC 2461 6.3.6)
682  */
683 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
684 {
685 	enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
686 	struct neighbour *neigh;
687 
688 	rcu_read_lock();
689 	neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
690 					  &fib6_nh->fib_nh_gw6);
691 	if (neigh) {
692 		u8 nud_state = READ_ONCE(neigh->nud_state);
693 
694 		if (nud_state & NUD_VALID)
695 			ret = RT6_NUD_SUCCEED;
696 #ifdef CONFIG_IPV6_ROUTER_PREF
697 		else if (!(nud_state & NUD_FAILED))
698 			ret = RT6_NUD_SUCCEED;
699 		else
700 			ret = RT6_NUD_FAIL_PROBE;
701 #endif
702 	} else {
703 		ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
704 		      RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
705 	}
706 	rcu_read_unlock();
707 
708 	return ret;
709 }
710 
711 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
712 			   int strict)
713 {
714 	int m = 0;
715 
716 	if (!oif || nh->fib_nh_dev->ifindex == oif)
717 		m = 2;
718 
719 	if (!m && (strict & RT6_LOOKUP_F_IFACE))
720 		return RT6_NUD_FAIL_HARD;
721 #ifdef CONFIG_IPV6_ROUTER_PREF
722 	m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
723 #endif
724 	if ((strict & RT6_LOOKUP_F_REACHABLE) &&
725 	    !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
726 		int n = rt6_check_neigh(nh);
727 		if (n < 0)
728 			return n;
729 	}
730 	return m;
731 }
732 
733 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
734 		       int oif, int strict, int *mpri, bool *do_rr)
735 {
736 	bool match_do_rr = false;
737 	bool rc = false;
738 	int m;
739 
740 	if (nh->fib_nh_flags & RTNH_F_DEAD)
741 		goto out;
742 
743 	if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
744 	    nh->fib_nh_flags & RTNH_F_LINKDOWN &&
745 	    !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
746 		goto out;
747 
748 	m = rt6_score_route(nh, fib6_flags, oif, strict);
749 	if (m == RT6_NUD_FAIL_DO_RR) {
750 		match_do_rr = true;
751 		m = 0; /* lowest valid score */
752 	} else if (m == RT6_NUD_FAIL_HARD) {
753 		goto out;
754 	}
755 
756 	if (strict & RT6_LOOKUP_F_REACHABLE)
757 		rt6_probe(nh);
758 
759 	/* note that m can be RT6_NUD_FAIL_PROBE at this point */
760 	if (m > *mpri) {
761 		*do_rr = match_do_rr;
762 		*mpri = m;
763 		rc = true;
764 	}
765 out:
766 	return rc;
767 }
768 
769 struct fib6_nh_frl_arg {
770 	u32		flags;
771 	int		oif;
772 	int		strict;
773 	int		*mpri;
774 	bool		*do_rr;
775 	struct fib6_nh	*nh;
776 };
777 
778 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
779 {
780 	struct fib6_nh_frl_arg *arg = _arg;
781 
782 	arg->nh = nh;
783 	return find_match(nh, arg->flags, arg->oif, arg->strict,
784 			  arg->mpri, arg->do_rr);
785 }
786 
787 static void __find_rr_leaf(struct fib6_info *f6i_start,
788 			   struct fib6_info *nomatch, u32 metric,
789 			   struct fib6_result *res, struct fib6_info **cont,
790 			   int oif, int strict, bool *do_rr, int *mpri)
791 {
792 	struct fib6_info *f6i;
793 
794 	for (f6i = f6i_start;
795 	     f6i && f6i != nomatch;
796 	     f6i = rcu_dereference(f6i->fib6_next)) {
797 		bool matched = false;
798 		struct fib6_nh *nh;
799 
800 		if (cont && f6i->fib6_metric != metric) {
801 			*cont = f6i;
802 			return;
803 		}
804 
805 		if (fib6_check_expired(f6i))
806 			continue;
807 
808 		if (unlikely(f6i->nh)) {
809 			struct fib6_nh_frl_arg arg = {
810 				.flags  = f6i->fib6_flags,
811 				.oif    = oif,
812 				.strict = strict,
813 				.mpri   = mpri,
814 				.do_rr  = do_rr
815 			};
816 
817 			if (nexthop_is_blackhole(f6i->nh)) {
818 				res->fib6_flags = RTF_REJECT;
819 				res->fib6_type = RTN_BLACKHOLE;
820 				res->f6i = f6i;
821 				res->nh = nexthop_fib6_nh(f6i->nh);
822 				return;
823 			}
824 			if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
825 						     &arg)) {
826 				matched = true;
827 				nh = arg.nh;
828 			}
829 		} else {
830 			nh = f6i->fib6_nh;
831 			if (find_match(nh, f6i->fib6_flags, oif, strict,
832 				       mpri, do_rr))
833 				matched = true;
834 		}
835 		if (matched) {
836 			res->f6i = f6i;
837 			res->nh = nh;
838 			res->fib6_flags = f6i->fib6_flags;
839 			res->fib6_type = f6i->fib6_type;
840 		}
841 	}
842 }
843 
844 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
845 			 struct fib6_info *rr_head, int oif, int strict,
846 			 bool *do_rr, struct fib6_result *res)
847 {
848 	u32 metric = rr_head->fib6_metric;
849 	struct fib6_info *cont = NULL;
850 	int mpri = -1;
851 
852 	__find_rr_leaf(rr_head, NULL, metric, res, &cont,
853 		       oif, strict, do_rr, &mpri);
854 
855 	__find_rr_leaf(leaf, rr_head, metric, res, &cont,
856 		       oif, strict, do_rr, &mpri);
857 
858 	if (res->f6i || !cont)
859 		return;
860 
861 	__find_rr_leaf(cont, NULL, metric, res, NULL,
862 		       oif, strict, do_rr, &mpri);
863 }
864 
865 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
866 		       struct fib6_result *res, int strict)
867 {
868 	struct fib6_info *leaf = rcu_dereference(fn->leaf);
869 	struct fib6_info *rt0;
870 	bool do_rr = false;
871 	int key_plen;
872 
873 	/* make sure this function or its helpers sets f6i */
874 	res->f6i = NULL;
875 
876 	if (!leaf || leaf == net->ipv6.fib6_null_entry)
877 		goto out;
878 
879 	rt0 = rcu_dereference(fn->rr_ptr);
880 	if (!rt0)
881 		rt0 = leaf;
882 
883 	/* Double check to make sure fn is not an intermediate node
884 	 * and fn->leaf does not points to its child's leaf
885 	 * (This might happen if all routes under fn are deleted from
886 	 * the tree and fib6_repair_tree() is called on the node.)
887 	 */
888 	key_plen = rt0->fib6_dst.plen;
889 #ifdef CONFIG_IPV6_SUBTREES
890 	if (rt0->fib6_src.plen)
891 		key_plen = rt0->fib6_src.plen;
892 #endif
893 	if (fn->fn_bit != key_plen)
894 		goto out;
895 
896 	find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
897 	if (do_rr) {
898 		struct fib6_info *next = rcu_dereference(rt0->fib6_next);
899 
900 		/* no entries matched; do round-robin */
901 		if (!next || next->fib6_metric != rt0->fib6_metric)
902 			next = leaf;
903 
904 		if (next != rt0) {
905 			spin_lock_bh(&leaf->fib6_table->tb6_lock);
906 			/* make sure next is not being deleted from the tree */
907 			if (next->fib6_node)
908 				rcu_assign_pointer(fn->rr_ptr, next);
909 			spin_unlock_bh(&leaf->fib6_table->tb6_lock);
910 		}
911 	}
912 
913 out:
914 	if (!res->f6i) {
915 		res->f6i = net->ipv6.fib6_null_entry;
916 		res->nh = res->f6i->fib6_nh;
917 		res->fib6_flags = res->f6i->fib6_flags;
918 		res->fib6_type = res->f6i->fib6_type;
919 	}
920 }
921 
922 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
923 {
924 	return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
925 	       res->nh->fib_nh_gw_family;
926 }
927 
928 #ifdef CONFIG_IPV6_ROUTE_INFO
929 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
930 		  const struct in6_addr *gwaddr)
931 {
932 	struct net *net = dev_net(dev);
933 	struct route_info *rinfo = (struct route_info *) opt;
934 	struct in6_addr prefix_buf, *prefix;
935 	struct fib6_table *table;
936 	unsigned int pref;
937 	unsigned long lifetime;
938 	struct fib6_info *rt;
939 
940 	if (len < sizeof(struct route_info)) {
941 		return -EINVAL;
942 	}
943 
944 	/* Sanity check for prefix_len and length */
945 	if (rinfo->length > 3) {
946 		return -EINVAL;
947 	} else if (rinfo->prefix_len > 128) {
948 		return -EINVAL;
949 	} else if (rinfo->prefix_len > 64) {
950 		if (rinfo->length < 2) {
951 			return -EINVAL;
952 		}
953 	} else if (rinfo->prefix_len > 0) {
954 		if (rinfo->length < 1) {
955 			return -EINVAL;
956 		}
957 	}
958 
959 	pref = rinfo->route_pref;
960 	if (pref == ICMPV6_ROUTER_PREF_INVALID)
961 		return -EINVAL;
962 
963 	lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
964 
965 	if (rinfo->length == 3)
966 		prefix = (struct in6_addr *)rinfo->prefix;
967 	else {
968 		/* this function is safe */
969 		ipv6_addr_prefix(&prefix_buf,
970 				 (struct in6_addr *)rinfo->prefix,
971 				 rinfo->prefix_len);
972 		prefix = &prefix_buf;
973 	}
974 
975 	if (rinfo->prefix_len == 0)
976 		rt = rt6_get_dflt_router(net, gwaddr, dev);
977 	else
978 		rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
979 					gwaddr, dev);
980 
981 	if (rt && !lifetime) {
982 		ip6_del_rt(net, rt, false);
983 		rt = NULL;
984 	}
985 
986 	if (!rt && lifetime)
987 		rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
988 					dev, pref);
989 	else if (rt)
990 		rt->fib6_flags = RTF_ROUTEINFO |
991 				 (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
992 
993 	if (rt) {
994 		table = rt->fib6_table;
995 		spin_lock_bh(&table->tb6_lock);
996 
997 		if (!addrconf_finite_timeout(lifetime)) {
998 			fib6_clean_expires(rt);
999 			fib6_remove_gc_list(rt);
1000 		} else {
1001 			fib6_set_expires(rt, jiffies + HZ * lifetime);
1002 			fib6_add_gc_list(rt);
1003 		}
1004 
1005 		spin_unlock_bh(&table->tb6_lock);
1006 
1007 		fib6_info_release(rt);
1008 	}
1009 	return 0;
1010 }
1011 #endif
1012 
1013 /*
1014  *	Misc support functions
1015  */
1016 
1017 /* called with rcu_lock held */
1018 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1019 {
1020 	struct net_device *dev = res->nh->fib_nh_dev;
1021 
1022 	if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1023 		/* for copies of local routes, dst->dev needs to be the
1024 		 * device if it is a master device, the master device if
1025 		 * device is enslaved, and the loopback as the default
1026 		 */
1027 		if (netif_is_l3_slave(dev) &&
1028 		    !rt6_need_strict(&res->f6i->fib6_dst.addr))
1029 			dev = l3mdev_master_dev_rcu(dev);
1030 		else if (!netif_is_l3_master(dev))
1031 			dev = dev_net(dev)->loopback_dev;
1032 		/* last case is netif_is_l3_master(dev) is true in which
1033 		 * case we want dev returned to be dev
1034 		 */
1035 	}
1036 
1037 	return dev;
1038 }
1039 
1040 static const int fib6_prop[RTN_MAX + 1] = {
1041 	[RTN_UNSPEC]	= 0,
1042 	[RTN_UNICAST]	= 0,
1043 	[RTN_LOCAL]	= 0,
1044 	[RTN_BROADCAST]	= 0,
1045 	[RTN_ANYCAST]	= 0,
1046 	[RTN_MULTICAST]	= 0,
1047 	[RTN_BLACKHOLE]	= -EINVAL,
1048 	[RTN_UNREACHABLE] = -EHOSTUNREACH,
1049 	[RTN_PROHIBIT]	= -EACCES,
1050 	[RTN_THROW]	= -EAGAIN,
1051 	[RTN_NAT]	= -EINVAL,
1052 	[RTN_XRESOLVE]	= -EINVAL,
1053 };
1054 
1055 static int ip6_rt_type_to_error(u8 fib6_type)
1056 {
1057 	return fib6_prop[fib6_type];
1058 }
1059 
1060 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1061 {
1062 	unsigned short flags = 0;
1063 
1064 	if (rt->dst_nocount)
1065 		flags |= DST_NOCOUNT;
1066 	if (rt->dst_nopolicy)
1067 		flags |= DST_NOPOLICY;
1068 
1069 	return flags;
1070 }
1071 
1072 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1073 {
1074 	rt->dst.error = ip6_rt_type_to_error(fib6_type);
1075 
1076 	switch (fib6_type) {
1077 	case RTN_BLACKHOLE:
1078 		rt->dst.output = dst_discard_out;
1079 		rt->dst.input = dst_discard;
1080 		break;
1081 	case RTN_PROHIBIT:
1082 		rt->dst.output = ip6_pkt_prohibit_out;
1083 		rt->dst.input = ip6_pkt_prohibit;
1084 		break;
1085 	case RTN_THROW:
1086 	case RTN_UNREACHABLE:
1087 	default:
1088 		rt->dst.output = ip6_pkt_discard_out;
1089 		rt->dst.input = ip6_pkt_discard;
1090 		break;
1091 	}
1092 }
1093 
1094 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1095 {
1096 	struct fib6_info *f6i = res->f6i;
1097 
1098 	if (res->fib6_flags & RTF_REJECT) {
1099 		ip6_rt_init_dst_reject(rt, res->fib6_type);
1100 		return;
1101 	}
1102 
1103 	rt->dst.error = 0;
1104 	rt->dst.output = ip6_output;
1105 
1106 	if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1107 		rt->dst.input = ip6_input;
1108 	} else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1109 		rt->dst.input = ip6_mc_input;
1110 	} else {
1111 		rt->dst.input = ip6_forward;
1112 	}
1113 
1114 	if (res->nh->fib_nh_lws) {
1115 		rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1116 		lwtunnel_set_redirect(&rt->dst);
1117 	}
1118 
1119 	rt->dst.lastuse = jiffies;
1120 }
1121 
1122 /* Caller must already hold reference to @from */
1123 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1124 {
1125 	rt->rt6i_flags &= ~RTF_EXPIRES;
1126 	rcu_assign_pointer(rt->from, from);
1127 	ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1128 }
1129 
1130 /* Caller must already hold reference to f6i in result */
1131 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1132 {
1133 	const struct fib6_nh *nh = res->nh;
1134 	const struct net_device *dev = nh->fib_nh_dev;
1135 	struct fib6_info *f6i = res->f6i;
1136 
1137 	ip6_rt_init_dst(rt, res);
1138 
1139 	rt->rt6i_dst = f6i->fib6_dst;
1140 	rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1141 	rt->rt6i_flags = res->fib6_flags;
1142 	if (nh->fib_nh_gw_family) {
1143 		rt->rt6i_gateway = nh->fib_nh_gw6;
1144 		rt->rt6i_flags |= RTF_GATEWAY;
1145 	}
1146 	rt6_set_from(rt, f6i);
1147 #ifdef CONFIG_IPV6_SUBTREES
1148 	rt->rt6i_src = f6i->fib6_src;
1149 #endif
1150 }
1151 
1152 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1153 					struct in6_addr *saddr)
1154 {
1155 	struct fib6_node *pn, *sn;
1156 	while (1) {
1157 		if (fn->fn_flags & RTN_TL_ROOT)
1158 			return NULL;
1159 		pn = rcu_dereference(fn->parent);
1160 		sn = FIB6_SUBTREE(pn);
1161 		if (sn && sn != fn)
1162 			fn = fib6_node_lookup(sn, NULL, saddr);
1163 		else
1164 			fn = pn;
1165 		if (fn->fn_flags & RTN_RTINFO)
1166 			return fn;
1167 	}
1168 }
1169 
1170 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1171 {
1172 	struct rt6_info *rt = *prt;
1173 
1174 	if (dst_hold_safe(&rt->dst))
1175 		return true;
1176 	if (net) {
1177 		rt = net->ipv6.ip6_null_entry;
1178 		dst_hold(&rt->dst);
1179 	} else {
1180 		rt = NULL;
1181 	}
1182 	*prt = rt;
1183 	return false;
1184 }
1185 
1186 /* called with rcu_lock held */
1187 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1188 {
1189 	struct net_device *dev = res->nh->fib_nh_dev;
1190 	struct fib6_info *f6i = res->f6i;
1191 	unsigned short flags;
1192 	struct rt6_info *nrt;
1193 
1194 	if (!fib6_info_hold_safe(f6i))
1195 		goto fallback;
1196 
1197 	flags = fib6_info_dst_flags(f6i);
1198 	nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1199 	if (!nrt) {
1200 		fib6_info_release(f6i);
1201 		goto fallback;
1202 	}
1203 
1204 	ip6_rt_copy_init(nrt, res);
1205 	return nrt;
1206 
1207 fallback:
1208 	nrt = dev_net(dev)->ipv6.ip6_null_entry;
1209 	dst_hold(&nrt->dst);
1210 	return nrt;
1211 }
1212 
1213 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_lookup(struct net *net,
1214 					     struct fib6_table *table,
1215 					     struct flowi6 *fl6,
1216 					     const struct sk_buff *skb,
1217 					     int flags)
1218 {
1219 	struct fib6_result res = {};
1220 	struct fib6_node *fn;
1221 	struct rt6_info *rt;
1222 
1223 	rcu_read_lock();
1224 	fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1225 restart:
1226 	res.f6i = rcu_dereference(fn->leaf);
1227 	if (!res.f6i)
1228 		res.f6i = net->ipv6.fib6_null_entry;
1229 	else
1230 		rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1231 				 flags);
1232 
1233 	if (res.f6i == net->ipv6.fib6_null_entry) {
1234 		fn = fib6_backtrack(fn, &fl6->saddr);
1235 		if (fn)
1236 			goto restart;
1237 
1238 		rt = net->ipv6.ip6_null_entry;
1239 		dst_hold(&rt->dst);
1240 		goto out;
1241 	} else if (res.fib6_flags & RTF_REJECT) {
1242 		goto do_create;
1243 	}
1244 
1245 	fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1246 			 fl6->flowi6_oif != 0, skb, flags);
1247 
1248 	/* Search through exception table */
1249 	rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1250 	if (rt) {
1251 		if (ip6_hold_safe(net, &rt))
1252 			dst_use_noref(&rt->dst, jiffies);
1253 	} else {
1254 do_create:
1255 		rt = ip6_create_rt_rcu(&res);
1256 	}
1257 
1258 out:
1259 	trace_fib6_table_lookup(net, &res, table, fl6);
1260 
1261 	rcu_read_unlock();
1262 
1263 	return rt;
1264 }
1265 
1266 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1267 				   const struct sk_buff *skb, int flags)
1268 {
1269 	return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1270 }
1271 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1272 
1273 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1274 			    const struct in6_addr *saddr, int oif,
1275 			    const struct sk_buff *skb, int strict)
1276 {
1277 	struct flowi6 fl6 = {
1278 		.flowi6_oif = oif,
1279 		.daddr = *daddr,
1280 	};
1281 	struct dst_entry *dst;
1282 	int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1283 
1284 	if (saddr) {
1285 		memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1286 		flags |= RT6_LOOKUP_F_HAS_SADDR;
1287 	}
1288 
1289 	dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1290 	if (dst->error == 0)
1291 		return (struct rt6_info *) dst;
1292 
1293 	dst_release(dst);
1294 
1295 	return NULL;
1296 }
1297 EXPORT_SYMBOL(rt6_lookup);
1298 
1299 /* ip6_ins_rt is called with FREE table->tb6_lock.
1300  * It takes new route entry, the addition fails by any reason the
1301  * route is released.
1302  * Caller must hold dst before calling it.
1303  */
1304 
1305 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1306 			struct netlink_ext_ack *extack)
1307 {
1308 	int err;
1309 	struct fib6_table *table;
1310 
1311 	table = rt->fib6_table;
1312 	spin_lock_bh(&table->tb6_lock);
1313 	err = fib6_add(&table->tb6_root, rt, info, extack);
1314 	spin_unlock_bh(&table->tb6_lock);
1315 
1316 	return err;
1317 }
1318 
1319 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1320 {
1321 	struct nl_info info = {	.nl_net = net, };
1322 
1323 	return __ip6_ins_rt(rt, &info, NULL);
1324 }
1325 
1326 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1327 					   const struct in6_addr *daddr,
1328 					   const struct in6_addr *saddr)
1329 {
1330 	struct fib6_info *f6i = res->f6i;
1331 	struct net_device *dev;
1332 	struct rt6_info *rt;
1333 
1334 	/*
1335 	 *	Clone the route.
1336 	 */
1337 
1338 	if (!fib6_info_hold_safe(f6i))
1339 		return NULL;
1340 
1341 	dev = ip6_rt_get_dev_rcu(res);
1342 	rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1343 	if (!rt) {
1344 		fib6_info_release(f6i);
1345 		return NULL;
1346 	}
1347 
1348 	ip6_rt_copy_init(rt, res);
1349 	rt->rt6i_flags |= RTF_CACHE;
1350 	rt->rt6i_dst.addr = *daddr;
1351 	rt->rt6i_dst.plen = 128;
1352 
1353 	if (!rt6_is_gw_or_nonexthop(res)) {
1354 		if (f6i->fib6_dst.plen != 128 &&
1355 		    ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1356 			rt->rt6i_flags |= RTF_ANYCAST;
1357 #ifdef CONFIG_IPV6_SUBTREES
1358 		if (rt->rt6i_src.plen && saddr) {
1359 			rt->rt6i_src.addr = *saddr;
1360 			rt->rt6i_src.plen = 128;
1361 		}
1362 #endif
1363 	}
1364 
1365 	return rt;
1366 }
1367 
1368 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1369 {
1370 	struct fib6_info *f6i = res->f6i;
1371 	unsigned short flags = fib6_info_dst_flags(f6i);
1372 	struct net_device *dev;
1373 	struct rt6_info *pcpu_rt;
1374 
1375 	if (!fib6_info_hold_safe(f6i))
1376 		return NULL;
1377 
1378 	rcu_read_lock();
1379 	dev = ip6_rt_get_dev_rcu(res);
1380 	pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags | DST_NOCOUNT);
1381 	rcu_read_unlock();
1382 	if (!pcpu_rt) {
1383 		fib6_info_release(f6i);
1384 		return NULL;
1385 	}
1386 	ip6_rt_copy_init(pcpu_rt, res);
1387 	pcpu_rt->rt6i_flags |= RTF_PCPU;
1388 
1389 	if (f6i->nh)
1390 		pcpu_rt->sernum = rt_genid_ipv6(dev_net(dev));
1391 
1392 	return pcpu_rt;
1393 }
1394 
1395 static bool rt6_is_valid(const struct rt6_info *rt6)
1396 {
1397 	return rt6->sernum == rt_genid_ipv6(dev_net(rt6->dst.dev));
1398 }
1399 
1400 /* It should be called with rcu_read_lock() acquired */
1401 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1402 {
1403 	struct rt6_info *pcpu_rt;
1404 
1405 	pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1406 
1407 	if (pcpu_rt && pcpu_rt->sernum && !rt6_is_valid(pcpu_rt)) {
1408 		struct rt6_info *prev, **p;
1409 
1410 		p = this_cpu_ptr(res->nh->rt6i_pcpu);
1411 		prev = xchg(p, NULL);
1412 		if (prev) {
1413 			dst_dev_put(&prev->dst);
1414 			dst_release(&prev->dst);
1415 		}
1416 
1417 		pcpu_rt = NULL;
1418 	}
1419 
1420 	return pcpu_rt;
1421 }
1422 
1423 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1424 					    const struct fib6_result *res)
1425 {
1426 	struct rt6_info *pcpu_rt, *prev, **p;
1427 
1428 	pcpu_rt = ip6_rt_pcpu_alloc(res);
1429 	if (!pcpu_rt)
1430 		return NULL;
1431 
1432 	p = this_cpu_ptr(res->nh->rt6i_pcpu);
1433 	prev = cmpxchg(p, NULL, pcpu_rt);
1434 	BUG_ON(prev);
1435 
1436 	if (res->f6i->fib6_destroying) {
1437 		struct fib6_info *from;
1438 
1439 		from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1440 		fib6_info_release(from);
1441 	}
1442 
1443 	return pcpu_rt;
1444 }
1445 
1446 /* exception hash table implementation
1447  */
1448 static DEFINE_SPINLOCK(rt6_exception_lock);
1449 
1450 /* Remove rt6_ex from hash table and free the memory
1451  * Caller must hold rt6_exception_lock
1452  */
1453 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1454 				 struct rt6_exception *rt6_ex)
1455 {
1456 	struct fib6_info *from;
1457 	struct net *net;
1458 
1459 	if (!bucket || !rt6_ex)
1460 		return;
1461 
1462 	net = dev_net(rt6_ex->rt6i->dst.dev);
1463 	net->ipv6.rt6_stats->fib_rt_cache--;
1464 
1465 	/* purge completely the exception to allow releasing the held resources:
1466 	 * some [sk] cache may keep the dst around for unlimited time
1467 	 */
1468 	from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1469 	fib6_info_release(from);
1470 	dst_dev_put(&rt6_ex->rt6i->dst);
1471 
1472 	hlist_del_rcu(&rt6_ex->hlist);
1473 	dst_release(&rt6_ex->rt6i->dst);
1474 	kfree_rcu(rt6_ex, rcu);
1475 	WARN_ON_ONCE(!bucket->depth);
1476 	bucket->depth--;
1477 }
1478 
1479 /* Remove oldest rt6_ex in bucket and free the memory
1480  * Caller must hold rt6_exception_lock
1481  */
1482 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1483 {
1484 	struct rt6_exception *rt6_ex, *oldest = NULL;
1485 
1486 	if (!bucket)
1487 		return;
1488 
1489 	hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1490 		if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1491 			oldest = rt6_ex;
1492 	}
1493 	rt6_remove_exception(bucket, oldest);
1494 }
1495 
1496 static u32 rt6_exception_hash(const struct in6_addr *dst,
1497 			      const struct in6_addr *src)
1498 {
1499 	static siphash_aligned_key_t rt6_exception_key;
1500 	struct {
1501 		struct in6_addr dst;
1502 		struct in6_addr src;
1503 	} __aligned(SIPHASH_ALIGNMENT) combined = {
1504 		.dst = *dst,
1505 	};
1506 	u64 val;
1507 
1508 	net_get_random_once(&rt6_exception_key, sizeof(rt6_exception_key));
1509 
1510 #ifdef CONFIG_IPV6_SUBTREES
1511 	if (src)
1512 		combined.src = *src;
1513 #endif
1514 	val = siphash(&combined, sizeof(combined), &rt6_exception_key);
1515 
1516 	return hash_64(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1517 }
1518 
1519 /* Helper function to find the cached rt in the hash table
1520  * and update bucket pointer to point to the bucket for this
1521  * (daddr, saddr) pair
1522  * Caller must hold rt6_exception_lock
1523  */
1524 static struct rt6_exception *
1525 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1526 			      const struct in6_addr *daddr,
1527 			      const struct in6_addr *saddr)
1528 {
1529 	struct rt6_exception *rt6_ex;
1530 	u32 hval;
1531 
1532 	if (!(*bucket) || !daddr)
1533 		return NULL;
1534 
1535 	hval = rt6_exception_hash(daddr, saddr);
1536 	*bucket += hval;
1537 
1538 	hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1539 		struct rt6_info *rt6 = rt6_ex->rt6i;
1540 		bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1541 
1542 #ifdef CONFIG_IPV6_SUBTREES
1543 		if (matched && saddr)
1544 			matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1545 #endif
1546 		if (matched)
1547 			return rt6_ex;
1548 	}
1549 	return NULL;
1550 }
1551 
1552 /* Helper function to find the cached rt in the hash table
1553  * and update bucket pointer to point to the bucket for this
1554  * (daddr, saddr) pair
1555  * Caller must hold rcu_read_lock()
1556  */
1557 static struct rt6_exception *
1558 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1559 			 const struct in6_addr *daddr,
1560 			 const struct in6_addr *saddr)
1561 {
1562 	struct rt6_exception *rt6_ex;
1563 	u32 hval;
1564 
1565 	WARN_ON_ONCE(!rcu_read_lock_held());
1566 
1567 	if (!(*bucket) || !daddr)
1568 		return NULL;
1569 
1570 	hval = rt6_exception_hash(daddr, saddr);
1571 	*bucket += hval;
1572 
1573 	hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1574 		struct rt6_info *rt6 = rt6_ex->rt6i;
1575 		bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1576 
1577 #ifdef CONFIG_IPV6_SUBTREES
1578 		if (matched && saddr)
1579 			matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1580 #endif
1581 		if (matched)
1582 			return rt6_ex;
1583 	}
1584 	return NULL;
1585 }
1586 
1587 static unsigned int fib6_mtu(const struct fib6_result *res)
1588 {
1589 	const struct fib6_nh *nh = res->nh;
1590 	unsigned int mtu;
1591 
1592 	if (res->f6i->fib6_pmtu) {
1593 		mtu = res->f6i->fib6_pmtu;
1594 	} else {
1595 		struct net_device *dev = nh->fib_nh_dev;
1596 		struct inet6_dev *idev;
1597 
1598 		rcu_read_lock();
1599 		idev = __in6_dev_get(dev);
1600 		mtu = READ_ONCE(idev->cnf.mtu6);
1601 		rcu_read_unlock();
1602 	}
1603 
1604 	mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1605 
1606 	return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1607 }
1608 
1609 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1610 
1611 /* used when the flushed bit is not relevant, only access to the bucket
1612  * (ie., all bucket users except rt6_insert_exception);
1613  *
1614  * called under rcu lock; sometimes called with rt6_exception_lock held
1615  */
1616 static
1617 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1618 						       spinlock_t *lock)
1619 {
1620 	struct rt6_exception_bucket *bucket;
1621 
1622 	if (lock)
1623 		bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1624 						   lockdep_is_held(lock));
1625 	else
1626 		bucket = rcu_dereference(nh->rt6i_exception_bucket);
1627 
1628 	/* remove bucket flushed bit if set */
1629 	if (bucket) {
1630 		unsigned long p = (unsigned long)bucket;
1631 
1632 		p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1633 		bucket = (struct rt6_exception_bucket *)p;
1634 	}
1635 
1636 	return bucket;
1637 }
1638 
1639 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1640 {
1641 	unsigned long p = (unsigned long)bucket;
1642 
1643 	return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1644 }
1645 
1646 /* called with rt6_exception_lock held */
1647 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1648 					      spinlock_t *lock)
1649 {
1650 	struct rt6_exception_bucket *bucket;
1651 	unsigned long p;
1652 
1653 	bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1654 					   lockdep_is_held(lock));
1655 
1656 	p = (unsigned long)bucket;
1657 	p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1658 	bucket = (struct rt6_exception_bucket *)p;
1659 	rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1660 }
1661 
1662 static int rt6_insert_exception(struct rt6_info *nrt,
1663 				const struct fib6_result *res)
1664 {
1665 	struct net *net = dev_net(nrt->dst.dev);
1666 	struct rt6_exception_bucket *bucket;
1667 	struct fib6_info *f6i = res->f6i;
1668 	struct in6_addr *src_key = NULL;
1669 	struct rt6_exception *rt6_ex;
1670 	struct fib6_nh *nh = res->nh;
1671 	int max_depth;
1672 	int err = 0;
1673 
1674 	spin_lock_bh(&rt6_exception_lock);
1675 
1676 	bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1677 					  lockdep_is_held(&rt6_exception_lock));
1678 	if (!bucket) {
1679 		bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1680 				 GFP_ATOMIC);
1681 		if (!bucket) {
1682 			err = -ENOMEM;
1683 			goto out;
1684 		}
1685 		rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1686 	} else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1687 		err = -EINVAL;
1688 		goto out;
1689 	}
1690 
1691 #ifdef CONFIG_IPV6_SUBTREES
1692 	/* fib6_src.plen != 0 indicates f6i is in subtree
1693 	 * and exception table is indexed by a hash of
1694 	 * both fib6_dst and fib6_src.
1695 	 * Otherwise, the exception table is indexed by
1696 	 * a hash of only fib6_dst.
1697 	 */
1698 	if (f6i->fib6_src.plen)
1699 		src_key = &nrt->rt6i_src.addr;
1700 #endif
1701 	/* rt6_mtu_change() might lower mtu on f6i.
1702 	 * Only insert this exception route if its mtu
1703 	 * is less than f6i's mtu value.
1704 	 */
1705 	if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1706 		err = -EINVAL;
1707 		goto out;
1708 	}
1709 
1710 	rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1711 					       src_key);
1712 	if (rt6_ex)
1713 		rt6_remove_exception(bucket, rt6_ex);
1714 
1715 	rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1716 	if (!rt6_ex) {
1717 		err = -ENOMEM;
1718 		goto out;
1719 	}
1720 	rt6_ex->rt6i = nrt;
1721 	rt6_ex->stamp = jiffies;
1722 	hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1723 	bucket->depth++;
1724 	net->ipv6.rt6_stats->fib_rt_cache++;
1725 
1726 	/* Randomize max depth to avoid some side channels attacks. */
1727 	max_depth = FIB6_MAX_DEPTH + get_random_u32_below(FIB6_MAX_DEPTH);
1728 	while (bucket->depth > max_depth)
1729 		rt6_exception_remove_oldest(bucket);
1730 
1731 out:
1732 	spin_unlock_bh(&rt6_exception_lock);
1733 
1734 	/* Update fn->fn_sernum to invalidate all cached dst */
1735 	if (!err) {
1736 		spin_lock_bh(&f6i->fib6_table->tb6_lock);
1737 		fib6_update_sernum(net, f6i);
1738 		spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1739 		fib6_force_start_gc(net);
1740 	}
1741 
1742 	return err;
1743 }
1744 
1745 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1746 {
1747 	struct rt6_exception_bucket *bucket;
1748 	struct rt6_exception *rt6_ex;
1749 	struct hlist_node *tmp;
1750 	int i;
1751 
1752 	spin_lock_bh(&rt6_exception_lock);
1753 
1754 	bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1755 	if (!bucket)
1756 		goto out;
1757 
1758 	/* Prevent rt6_insert_exception() to recreate the bucket list */
1759 	if (!from)
1760 		fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1761 
1762 	for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1763 		hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1764 			if (!from ||
1765 			    rcu_access_pointer(rt6_ex->rt6i->from) == from)
1766 				rt6_remove_exception(bucket, rt6_ex);
1767 		}
1768 		WARN_ON_ONCE(!from && bucket->depth);
1769 		bucket++;
1770 	}
1771 out:
1772 	spin_unlock_bh(&rt6_exception_lock);
1773 }
1774 
1775 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1776 {
1777 	struct fib6_info *f6i = arg;
1778 
1779 	fib6_nh_flush_exceptions(nh, f6i);
1780 
1781 	return 0;
1782 }
1783 
1784 void rt6_flush_exceptions(struct fib6_info *f6i)
1785 {
1786 	if (f6i->nh)
1787 		nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1788 					 f6i);
1789 	else
1790 		fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1791 }
1792 
1793 /* Find cached rt in the hash table inside passed in rt
1794  * Caller has to hold rcu_read_lock()
1795  */
1796 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1797 					   const struct in6_addr *daddr,
1798 					   const struct in6_addr *saddr)
1799 {
1800 	const struct in6_addr *src_key = NULL;
1801 	struct rt6_exception_bucket *bucket;
1802 	struct rt6_exception *rt6_ex;
1803 	struct rt6_info *ret = NULL;
1804 
1805 #ifdef CONFIG_IPV6_SUBTREES
1806 	/* fib6i_src.plen != 0 indicates f6i is in subtree
1807 	 * and exception table is indexed by a hash of
1808 	 * both fib6_dst and fib6_src.
1809 	 * However, the src addr used to create the hash
1810 	 * might not be exactly the passed in saddr which
1811 	 * is a /128 addr from the flow.
1812 	 * So we need to use f6i->fib6_src to redo lookup
1813 	 * if the passed in saddr does not find anything.
1814 	 * (See the logic in ip6_rt_cache_alloc() on how
1815 	 * rt->rt6i_src is updated.)
1816 	 */
1817 	if (res->f6i->fib6_src.plen)
1818 		src_key = saddr;
1819 find_ex:
1820 #endif
1821 	bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1822 	rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1823 
1824 	if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1825 		ret = rt6_ex->rt6i;
1826 
1827 #ifdef CONFIG_IPV6_SUBTREES
1828 	/* Use fib6_src as src_key and redo lookup */
1829 	if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1830 		src_key = &res->f6i->fib6_src.addr;
1831 		goto find_ex;
1832 	}
1833 #endif
1834 
1835 	return ret;
1836 }
1837 
1838 /* Remove the passed in cached rt from the hash table that contains it */
1839 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1840 				    const struct rt6_info *rt)
1841 {
1842 	const struct in6_addr *src_key = NULL;
1843 	struct rt6_exception_bucket *bucket;
1844 	struct rt6_exception *rt6_ex;
1845 	int err;
1846 
1847 	if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1848 		return -ENOENT;
1849 
1850 	spin_lock_bh(&rt6_exception_lock);
1851 	bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1852 
1853 #ifdef CONFIG_IPV6_SUBTREES
1854 	/* rt6i_src.plen != 0 indicates 'from' is in subtree
1855 	 * and exception table is indexed by a hash of
1856 	 * both rt6i_dst and rt6i_src.
1857 	 * Otherwise, the exception table is indexed by
1858 	 * a hash of only rt6i_dst.
1859 	 */
1860 	if (plen)
1861 		src_key = &rt->rt6i_src.addr;
1862 #endif
1863 	rt6_ex = __rt6_find_exception_spinlock(&bucket,
1864 					       &rt->rt6i_dst.addr,
1865 					       src_key);
1866 	if (rt6_ex) {
1867 		rt6_remove_exception(bucket, rt6_ex);
1868 		err = 0;
1869 	} else {
1870 		err = -ENOENT;
1871 	}
1872 
1873 	spin_unlock_bh(&rt6_exception_lock);
1874 	return err;
1875 }
1876 
1877 struct fib6_nh_excptn_arg {
1878 	struct rt6_info	*rt;
1879 	int		plen;
1880 };
1881 
1882 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1883 {
1884 	struct fib6_nh_excptn_arg *arg = _arg;
1885 	int err;
1886 
1887 	err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1888 	if (err == 0)
1889 		return 1;
1890 
1891 	return 0;
1892 }
1893 
1894 static int rt6_remove_exception_rt(struct rt6_info *rt)
1895 {
1896 	struct fib6_info *from;
1897 
1898 	from = rcu_dereference(rt->from);
1899 	if (!from || !(rt->rt6i_flags & RTF_CACHE))
1900 		return -EINVAL;
1901 
1902 	if (from->nh) {
1903 		struct fib6_nh_excptn_arg arg = {
1904 			.rt = rt,
1905 			.plen = from->fib6_src.plen
1906 		};
1907 		int rc;
1908 
1909 		/* rc = 1 means an entry was found */
1910 		rc = nexthop_for_each_fib6_nh(from->nh,
1911 					      rt6_nh_remove_exception_rt,
1912 					      &arg);
1913 		return rc ? 0 : -ENOENT;
1914 	}
1915 
1916 	return fib6_nh_remove_exception(from->fib6_nh,
1917 					from->fib6_src.plen, rt);
1918 }
1919 
1920 /* Find rt6_ex which contains the passed in rt cache and
1921  * refresh its stamp
1922  */
1923 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1924 				     const struct rt6_info *rt)
1925 {
1926 	const struct in6_addr *src_key = NULL;
1927 	struct rt6_exception_bucket *bucket;
1928 	struct rt6_exception *rt6_ex;
1929 
1930 	bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1931 #ifdef CONFIG_IPV6_SUBTREES
1932 	/* rt6i_src.plen != 0 indicates 'from' is in subtree
1933 	 * and exception table is indexed by a hash of
1934 	 * both rt6i_dst and rt6i_src.
1935 	 * Otherwise, the exception table is indexed by
1936 	 * a hash of only rt6i_dst.
1937 	 */
1938 	if (plen)
1939 		src_key = &rt->rt6i_src.addr;
1940 #endif
1941 	rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1942 	if (rt6_ex)
1943 		rt6_ex->stamp = jiffies;
1944 }
1945 
1946 struct fib6_nh_match_arg {
1947 	const struct net_device *dev;
1948 	const struct in6_addr	*gw;
1949 	struct fib6_nh		*match;
1950 };
1951 
1952 /* determine if fib6_nh has given device and gateway */
1953 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1954 {
1955 	struct fib6_nh_match_arg *arg = _arg;
1956 
1957 	if (arg->dev != nh->fib_nh_dev ||
1958 	    (arg->gw && !nh->fib_nh_gw_family) ||
1959 	    (!arg->gw && nh->fib_nh_gw_family) ||
1960 	    (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1961 		return 0;
1962 
1963 	arg->match = nh;
1964 
1965 	/* found a match, break the loop */
1966 	return 1;
1967 }
1968 
1969 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1970 {
1971 	struct fib6_info *from;
1972 	struct fib6_nh *fib6_nh;
1973 
1974 	rcu_read_lock();
1975 
1976 	from = rcu_dereference(rt->from);
1977 	if (!from || !(rt->rt6i_flags & RTF_CACHE))
1978 		goto unlock;
1979 
1980 	if (from->nh) {
1981 		struct fib6_nh_match_arg arg = {
1982 			.dev = rt->dst.dev,
1983 			.gw = &rt->rt6i_gateway,
1984 		};
1985 
1986 		nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1987 
1988 		if (!arg.match)
1989 			goto unlock;
1990 		fib6_nh = arg.match;
1991 	} else {
1992 		fib6_nh = from->fib6_nh;
1993 	}
1994 	fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1995 unlock:
1996 	rcu_read_unlock();
1997 }
1998 
1999 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
2000 					 struct rt6_info *rt, int mtu)
2001 {
2002 	/* If the new MTU is lower than the route PMTU, this new MTU will be the
2003 	 * lowest MTU in the path: always allow updating the route PMTU to
2004 	 * reflect PMTU decreases.
2005 	 *
2006 	 * If the new MTU is higher, and the route PMTU is equal to the local
2007 	 * MTU, this means the old MTU is the lowest in the path, so allow
2008 	 * updating it: if other nodes now have lower MTUs, PMTU discovery will
2009 	 * handle this.
2010 	 */
2011 
2012 	if (dst_mtu(&rt->dst) >= mtu)
2013 		return true;
2014 
2015 	if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
2016 		return true;
2017 
2018 	return false;
2019 }
2020 
2021 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
2022 				       const struct fib6_nh *nh, int mtu)
2023 {
2024 	struct rt6_exception_bucket *bucket;
2025 	struct rt6_exception *rt6_ex;
2026 	int i;
2027 
2028 	bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2029 	if (!bucket)
2030 		return;
2031 
2032 	for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2033 		hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2034 			struct rt6_info *entry = rt6_ex->rt6i;
2035 
2036 			/* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2037 			 * route), the metrics of its rt->from have already
2038 			 * been updated.
2039 			 */
2040 			if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2041 			    rt6_mtu_change_route_allowed(idev, entry, mtu))
2042 				dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2043 		}
2044 		bucket++;
2045 	}
2046 }
2047 
2048 #define RTF_CACHE_GATEWAY	(RTF_GATEWAY | RTF_CACHE)
2049 
2050 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2051 					    const struct in6_addr *gateway)
2052 {
2053 	struct rt6_exception_bucket *bucket;
2054 	struct rt6_exception *rt6_ex;
2055 	struct hlist_node *tmp;
2056 	int i;
2057 
2058 	if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2059 		return;
2060 
2061 	spin_lock_bh(&rt6_exception_lock);
2062 	bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2063 	if (bucket) {
2064 		for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2065 			hlist_for_each_entry_safe(rt6_ex, tmp,
2066 						  &bucket->chain, hlist) {
2067 				struct rt6_info *entry = rt6_ex->rt6i;
2068 
2069 				if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2070 				    RTF_CACHE_GATEWAY &&
2071 				    ipv6_addr_equal(gateway,
2072 						    &entry->rt6i_gateway)) {
2073 					rt6_remove_exception(bucket, rt6_ex);
2074 				}
2075 			}
2076 			bucket++;
2077 		}
2078 	}
2079 
2080 	spin_unlock_bh(&rt6_exception_lock);
2081 }
2082 
2083 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2084 				      struct rt6_exception *rt6_ex,
2085 				      struct fib6_gc_args *gc_args,
2086 				      unsigned long now)
2087 {
2088 	struct rt6_info *rt = rt6_ex->rt6i;
2089 
2090 	/* we are pruning and obsoleting aged-out and non gateway exceptions
2091 	 * even if others have still references to them, so that on next
2092 	 * dst_check() such references can be dropped.
2093 	 * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2094 	 * expired, independently from their aging, as per RFC 8201 section 4
2095 	 */
2096 	if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2097 		if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2098 			pr_debug("aging clone %p\n", rt);
2099 			rt6_remove_exception(bucket, rt6_ex);
2100 			return;
2101 		}
2102 	} else if (time_after(jiffies, rt->dst.expires)) {
2103 		pr_debug("purging expired route %p\n", rt);
2104 		rt6_remove_exception(bucket, rt6_ex);
2105 		return;
2106 	}
2107 
2108 	if (rt->rt6i_flags & RTF_GATEWAY) {
2109 		struct neighbour *neigh;
2110 
2111 		neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2112 
2113 		if (!(neigh && (neigh->flags & NTF_ROUTER))) {
2114 			pr_debug("purging route %p via non-router but gateway\n",
2115 				 rt);
2116 			rt6_remove_exception(bucket, rt6_ex);
2117 			return;
2118 		}
2119 	}
2120 
2121 	gc_args->more++;
2122 }
2123 
2124 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2125 				   struct fib6_gc_args *gc_args,
2126 				   unsigned long now)
2127 {
2128 	struct rt6_exception_bucket *bucket;
2129 	struct rt6_exception *rt6_ex;
2130 	struct hlist_node *tmp;
2131 	int i;
2132 
2133 	if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2134 		return;
2135 
2136 	rcu_read_lock_bh();
2137 	spin_lock(&rt6_exception_lock);
2138 	bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2139 	if (bucket) {
2140 		for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2141 			hlist_for_each_entry_safe(rt6_ex, tmp,
2142 						  &bucket->chain, hlist) {
2143 				rt6_age_examine_exception(bucket, rt6_ex,
2144 							  gc_args, now);
2145 			}
2146 			bucket++;
2147 		}
2148 	}
2149 	spin_unlock(&rt6_exception_lock);
2150 	rcu_read_unlock_bh();
2151 }
2152 
2153 struct fib6_nh_age_excptn_arg {
2154 	struct fib6_gc_args	*gc_args;
2155 	unsigned long		now;
2156 };
2157 
2158 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2159 {
2160 	struct fib6_nh_age_excptn_arg *arg = _arg;
2161 
2162 	fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2163 	return 0;
2164 }
2165 
2166 void rt6_age_exceptions(struct fib6_info *f6i,
2167 			struct fib6_gc_args *gc_args,
2168 			unsigned long now)
2169 {
2170 	if (f6i->nh) {
2171 		struct fib6_nh_age_excptn_arg arg = {
2172 			.gc_args = gc_args,
2173 			.now = now
2174 		};
2175 
2176 		nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2177 					 &arg);
2178 	} else {
2179 		fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2180 	}
2181 }
2182 
2183 /* must be called with rcu lock held */
2184 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2185 		      struct flowi6 *fl6, struct fib6_result *res, int strict)
2186 {
2187 	struct fib6_node *fn, *saved_fn;
2188 
2189 	fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2190 	saved_fn = fn;
2191 
2192 redo_rt6_select:
2193 	rt6_select(net, fn, oif, res, strict);
2194 	if (res->f6i == net->ipv6.fib6_null_entry) {
2195 		fn = fib6_backtrack(fn, &fl6->saddr);
2196 		if (fn)
2197 			goto redo_rt6_select;
2198 		else if (strict & RT6_LOOKUP_F_REACHABLE) {
2199 			/* also consider unreachable route */
2200 			strict &= ~RT6_LOOKUP_F_REACHABLE;
2201 			fn = saved_fn;
2202 			goto redo_rt6_select;
2203 		}
2204 	}
2205 
2206 	trace_fib6_table_lookup(net, res, table, fl6);
2207 
2208 	return 0;
2209 }
2210 
2211 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2212 			       int oif, struct flowi6 *fl6,
2213 			       const struct sk_buff *skb, int flags)
2214 {
2215 	struct fib6_result res = {};
2216 	struct rt6_info *rt = NULL;
2217 	int strict = 0;
2218 
2219 	WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2220 		     !rcu_read_lock_held());
2221 
2222 	strict |= flags & RT6_LOOKUP_F_IFACE;
2223 	strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2224 	if (READ_ONCE(net->ipv6.devconf_all->forwarding) == 0)
2225 		strict |= RT6_LOOKUP_F_REACHABLE;
2226 
2227 	rcu_read_lock();
2228 
2229 	fib6_table_lookup(net, table, oif, fl6, &res, strict);
2230 	if (res.f6i == net->ipv6.fib6_null_entry)
2231 		goto out;
2232 
2233 	fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2234 
2235 	/*Search through exception table */
2236 	rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2237 	if (rt) {
2238 		goto out;
2239 	} else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2240 			    !res.nh->fib_nh_gw_family)) {
2241 		/* Create a RTF_CACHE clone which will not be
2242 		 * owned by the fib6 tree.  It is for the special case where
2243 		 * the daddr in the skb during the neighbor look-up is different
2244 		 * from the fl6->daddr used to look-up route here.
2245 		 */
2246 		rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2247 
2248 		if (rt) {
2249 			/* 1 refcnt is taken during ip6_rt_cache_alloc().
2250 			 * As rt6_uncached_list_add() does not consume refcnt,
2251 			 * this refcnt is always returned to the caller even
2252 			 * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2253 			 */
2254 			rt6_uncached_list_add(rt);
2255 			rcu_read_unlock();
2256 
2257 			return rt;
2258 		}
2259 	} else {
2260 		/* Get a percpu copy */
2261 		local_bh_disable();
2262 		rt = rt6_get_pcpu_route(&res);
2263 
2264 		if (!rt)
2265 			rt = rt6_make_pcpu_route(net, &res);
2266 
2267 		local_bh_enable();
2268 	}
2269 out:
2270 	if (!rt)
2271 		rt = net->ipv6.ip6_null_entry;
2272 	if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2273 		ip6_hold_safe(net, &rt);
2274 	rcu_read_unlock();
2275 
2276 	return rt;
2277 }
2278 EXPORT_SYMBOL_GPL(ip6_pol_route);
2279 
2280 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_input(struct net *net,
2281 					    struct fib6_table *table,
2282 					    struct flowi6 *fl6,
2283 					    const struct sk_buff *skb,
2284 					    int flags)
2285 {
2286 	return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2287 }
2288 
2289 struct dst_entry *ip6_route_input_lookup(struct net *net,
2290 					 struct net_device *dev,
2291 					 struct flowi6 *fl6,
2292 					 const struct sk_buff *skb,
2293 					 int flags)
2294 {
2295 	if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2296 		flags |= RT6_LOOKUP_F_IFACE;
2297 
2298 	return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2299 }
2300 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2301 
2302 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2303 				  struct flow_keys *keys,
2304 				  struct flow_keys *flkeys)
2305 {
2306 	const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2307 	const struct ipv6hdr *key_iph = outer_iph;
2308 	struct flow_keys *_flkeys = flkeys;
2309 	const struct ipv6hdr *inner_iph;
2310 	const struct icmp6hdr *icmph;
2311 	struct ipv6hdr _inner_iph;
2312 	struct icmp6hdr _icmph;
2313 
2314 	if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2315 		goto out;
2316 
2317 	icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2318 				   sizeof(_icmph), &_icmph);
2319 	if (!icmph)
2320 		goto out;
2321 
2322 	if (!icmpv6_is_err(icmph->icmp6_type))
2323 		goto out;
2324 
2325 	inner_iph = skb_header_pointer(skb,
2326 				       skb_transport_offset(skb) + sizeof(*icmph),
2327 				       sizeof(_inner_iph), &_inner_iph);
2328 	if (!inner_iph)
2329 		goto out;
2330 
2331 	key_iph = inner_iph;
2332 	_flkeys = NULL;
2333 out:
2334 	if (_flkeys) {
2335 		keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2336 		keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2337 		keys->tags.flow_label = _flkeys->tags.flow_label;
2338 		keys->basic.ip_proto = _flkeys->basic.ip_proto;
2339 	} else {
2340 		keys->addrs.v6addrs.src = key_iph->saddr;
2341 		keys->addrs.v6addrs.dst = key_iph->daddr;
2342 		keys->tags.flow_label = ip6_flowlabel(key_iph);
2343 		keys->basic.ip_proto = key_iph->nexthdr;
2344 	}
2345 }
2346 
2347 static u32 rt6_multipath_custom_hash_outer(const struct net *net,
2348 					   const struct sk_buff *skb,
2349 					   bool *p_has_inner)
2350 {
2351 	u32 hash_fields = ip6_multipath_hash_fields(net);
2352 	struct flow_keys keys, hash_keys;
2353 
2354 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2355 		return 0;
2356 
2357 	memset(&hash_keys, 0, sizeof(hash_keys));
2358 	skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
2359 
2360 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2361 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2362 		hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2363 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2364 		hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2365 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2366 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
2367 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2368 		hash_keys.tags.flow_label = keys.tags.flow_label;
2369 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2370 		hash_keys.ports.src = keys.ports.src;
2371 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2372 		hash_keys.ports.dst = keys.ports.dst;
2373 
2374 	*p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
2375 	return flow_hash_from_keys(&hash_keys);
2376 }
2377 
2378 static u32 rt6_multipath_custom_hash_inner(const struct net *net,
2379 					   const struct sk_buff *skb,
2380 					   bool has_inner)
2381 {
2382 	u32 hash_fields = ip6_multipath_hash_fields(net);
2383 	struct flow_keys keys, hash_keys;
2384 
2385 	/* We assume the packet carries an encapsulation, but if none was
2386 	 * encountered during dissection of the outer flow, then there is no
2387 	 * point in calling the flow dissector again.
2388 	 */
2389 	if (!has_inner)
2390 		return 0;
2391 
2392 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
2393 		return 0;
2394 
2395 	memset(&hash_keys, 0, sizeof(hash_keys));
2396 	skb_flow_dissect_flow_keys(skb, &keys, 0);
2397 
2398 	if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
2399 		return 0;
2400 
2401 	if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2402 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2403 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2404 			hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2405 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2406 			hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2407 	} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2408 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2409 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2410 			hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2411 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2412 			hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2413 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2414 			hash_keys.tags.flow_label = keys.tags.flow_label;
2415 	}
2416 
2417 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2418 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
2419 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2420 		hash_keys.ports.src = keys.ports.src;
2421 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2422 		hash_keys.ports.dst = keys.ports.dst;
2423 
2424 	return flow_hash_from_keys(&hash_keys);
2425 }
2426 
2427 static u32 rt6_multipath_custom_hash_skb(const struct net *net,
2428 					 const struct sk_buff *skb)
2429 {
2430 	u32 mhash, mhash_inner;
2431 	bool has_inner = true;
2432 
2433 	mhash = rt6_multipath_custom_hash_outer(net, skb, &has_inner);
2434 	mhash_inner = rt6_multipath_custom_hash_inner(net, skb, has_inner);
2435 
2436 	return jhash_2words(mhash, mhash_inner, 0);
2437 }
2438 
2439 static u32 rt6_multipath_custom_hash_fl6(const struct net *net,
2440 					 const struct flowi6 *fl6)
2441 {
2442 	u32 hash_fields = ip6_multipath_hash_fields(net);
2443 	struct flow_keys hash_keys;
2444 
2445 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2446 		return 0;
2447 
2448 	memset(&hash_keys, 0, sizeof(hash_keys));
2449 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2450 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2451 		hash_keys.addrs.v6addrs.src = fl6->saddr;
2452 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2453 		hash_keys.addrs.v6addrs.dst = fl6->daddr;
2454 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2455 		hash_keys.basic.ip_proto = fl6->flowi6_proto;
2456 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2457 		hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2458 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2459 		hash_keys.ports.src = fl6->fl6_sport;
2460 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2461 		hash_keys.ports.dst = fl6->fl6_dport;
2462 
2463 	return flow_hash_from_keys(&hash_keys);
2464 }
2465 
2466 /* if skb is set it will be used and fl6 can be NULL */
2467 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2468 		       const struct sk_buff *skb, struct flow_keys *flkeys)
2469 {
2470 	struct flow_keys hash_keys;
2471 	u32 mhash = 0;
2472 
2473 	switch (ip6_multipath_hash_policy(net)) {
2474 	case 0:
2475 		memset(&hash_keys, 0, sizeof(hash_keys));
2476 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2477 		if (skb) {
2478 			ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2479 		} else {
2480 			hash_keys.addrs.v6addrs.src = fl6->saddr;
2481 			hash_keys.addrs.v6addrs.dst = fl6->daddr;
2482 			hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2483 			hash_keys.basic.ip_proto = fl6->flowi6_proto;
2484 		}
2485 		mhash = flow_hash_from_keys(&hash_keys);
2486 		break;
2487 	case 1:
2488 		if (skb) {
2489 			unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2490 			struct flow_keys keys;
2491 
2492 			/* short-circuit if we already have L4 hash present */
2493 			if (skb->l4_hash)
2494 				return skb_get_hash_raw(skb) >> 1;
2495 
2496 			memset(&hash_keys, 0, sizeof(hash_keys));
2497 
2498 			if (!flkeys) {
2499 				skb_flow_dissect_flow_keys(skb, &keys, flag);
2500 				flkeys = &keys;
2501 			}
2502 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2503 			hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2504 			hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2505 			hash_keys.ports.src = flkeys->ports.src;
2506 			hash_keys.ports.dst = flkeys->ports.dst;
2507 			hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2508 		} else {
2509 			memset(&hash_keys, 0, sizeof(hash_keys));
2510 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2511 			hash_keys.addrs.v6addrs.src = fl6->saddr;
2512 			hash_keys.addrs.v6addrs.dst = fl6->daddr;
2513 			hash_keys.ports.src = fl6->fl6_sport;
2514 			hash_keys.ports.dst = fl6->fl6_dport;
2515 			hash_keys.basic.ip_proto = fl6->flowi6_proto;
2516 		}
2517 		mhash = flow_hash_from_keys(&hash_keys);
2518 		break;
2519 	case 2:
2520 		memset(&hash_keys, 0, sizeof(hash_keys));
2521 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2522 		if (skb) {
2523 			struct flow_keys keys;
2524 
2525 			if (!flkeys) {
2526 				skb_flow_dissect_flow_keys(skb, &keys, 0);
2527 				flkeys = &keys;
2528 			}
2529 
2530 			/* Inner can be v4 or v6 */
2531 			if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2532 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2533 				hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2534 				hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2535 			} else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2536 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2537 				hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2538 				hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2539 				hash_keys.tags.flow_label = flkeys->tags.flow_label;
2540 				hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2541 			} else {
2542 				/* Same as case 0 */
2543 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2544 				ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2545 			}
2546 		} else {
2547 			/* Same as case 0 */
2548 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2549 			hash_keys.addrs.v6addrs.src = fl6->saddr;
2550 			hash_keys.addrs.v6addrs.dst = fl6->daddr;
2551 			hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2552 			hash_keys.basic.ip_proto = fl6->flowi6_proto;
2553 		}
2554 		mhash = flow_hash_from_keys(&hash_keys);
2555 		break;
2556 	case 3:
2557 		if (skb)
2558 			mhash = rt6_multipath_custom_hash_skb(net, skb);
2559 		else
2560 			mhash = rt6_multipath_custom_hash_fl6(net, fl6);
2561 		break;
2562 	}
2563 
2564 	return mhash >> 1;
2565 }
2566 
2567 /* Called with rcu held */
2568 void ip6_route_input(struct sk_buff *skb)
2569 {
2570 	const struct ipv6hdr *iph = ipv6_hdr(skb);
2571 	struct net *net = dev_net(skb->dev);
2572 	int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2573 	struct ip_tunnel_info *tun_info;
2574 	struct flowi6 fl6 = {
2575 		.flowi6_iif = skb->dev->ifindex,
2576 		.daddr = iph->daddr,
2577 		.saddr = iph->saddr,
2578 		.flowlabel = ip6_flowinfo(iph),
2579 		.flowi6_mark = skb->mark,
2580 		.flowi6_proto = iph->nexthdr,
2581 	};
2582 	struct flow_keys *flkeys = NULL, _flkeys;
2583 
2584 	tun_info = skb_tunnel_info(skb);
2585 	if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2586 		fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2587 
2588 	if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2589 		flkeys = &_flkeys;
2590 
2591 	if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2592 		fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2593 	skb_dst_drop(skb);
2594 	skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2595 						      &fl6, skb, flags));
2596 }
2597 
2598 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_output(struct net *net,
2599 					     struct fib6_table *table,
2600 					     struct flowi6 *fl6,
2601 					     const struct sk_buff *skb,
2602 					     int flags)
2603 {
2604 	return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2605 }
2606 
2607 static struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2608 						      const struct sock *sk,
2609 						      struct flowi6 *fl6,
2610 						      int flags)
2611 {
2612 	bool any_src;
2613 
2614 	if (ipv6_addr_type(&fl6->daddr) &
2615 	    (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2616 		struct dst_entry *dst;
2617 
2618 		/* This function does not take refcnt on the dst */
2619 		dst = l3mdev_link_scope_lookup(net, fl6);
2620 		if (dst)
2621 			return dst;
2622 	}
2623 
2624 	fl6->flowi6_iif = LOOPBACK_IFINDEX;
2625 
2626 	flags |= RT6_LOOKUP_F_DST_NOREF;
2627 	any_src = ipv6_addr_any(&fl6->saddr);
2628 	if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2629 	    (fl6->flowi6_oif && any_src))
2630 		flags |= RT6_LOOKUP_F_IFACE;
2631 
2632 	if (!any_src)
2633 		flags |= RT6_LOOKUP_F_HAS_SADDR;
2634 	else if (sk)
2635 		flags |= rt6_srcprefs2flags(READ_ONCE(inet6_sk(sk)->srcprefs));
2636 
2637 	return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2638 }
2639 
2640 struct dst_entry *ip6_route_output_flags(struct net *net,
2641 					 const struct sock *sk,
2642 					 struct flowi6 *fl6,
2643 					 int flags)
2644 {
2645 	struct dst_entry *dst;
2646 	struct rt6_info *rt6;
2647 
2648 	rcu_read_lock();
2649 	dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2650 	rt6 = (struct rt6_info *)dst;
2651 	/* For dst cached in uncached_list, refcnt is already taken. */
2652 	if (list_empty(&rt6->dst.rt_uncached) && !dst_hold_safe(dst)) {
2653 		dst = &net->ipv6.ip6_null_entry->dst;
2654 		dst_hold(dst);
2655 	}
2656 	rcu_read_unlock();
2657 
2658 	return dst;
2659 }
2660 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2661 
2662 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2663 {
2664 	struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2665 	struct net_device *loopback_dev = net->loopback_dev;
2666 	struct dst_entry *new = NULL;
2667 
2668 	rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev,
2669 		       DST_OBSOLETE_DEAD, 0);
2670 	if (rt) {
2671 		rt6_info_init(rt);
2672 		atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2673 
2674 		new = &rt->dst;
2675 		new->__use = 1;
2676 		new->input = dst_discard;
2677 		new->output = dst_discard_out;
2678 
2679 		dst_copy_metrics(new, &ort->dst);
2680 
2681 		rt->rt6i_idev = in6_dev_get(loopback_dev);
2682 		rt->rt6i_gateway = ort->rt6i_gateway;
2683 		rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2684 
2685 		memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2686 #ifdef CONFIG_IPV6_SUBTREES
2687 		memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2688 #endif
2689 	}
2690 
2691 	dst_release(dst_orig);
2692 	return new ? new : ERR_PTR(-ENOMEM);
2693 }
2694 
2695 /*
2696  *	Destination cache support functions
2697  */
2698 
2699 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2700 {
2701 	u32 rt_cookie = 0;
2702 
2703 	if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2704 		return false;
2705 
2706 	if (fib6_check_expired(f6i))
2707 		return false;
2708 
2709 	return true;
2710 }
2711 
2712 static struct dst_entry *rt6_check(struct rt6_info *rt,
2713 				   struct fib6_info *from,
2714 				   u32 cookie)
2715 {
2716 	u32 rt_cookie = 0;
2717 
2718 	if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2719 	    rt_cookie != cookie)
2720 		return NULL;
2721 
2722 	if (rt6_check_expired(rt))
2723 		return NULL;
2724 
2725 	return &rt->dst;
2726 }
2727 
2728 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2729 					    struct fib6_info *from,
2730 					    u32 cookie)
2731 {
2732 	if (!__rt6_check_expired(rt) &&
2733 	    rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2734 	    fib6_check(from, cookie))
2735 		return &rt->dst;
2736 	else
2737 		return NULL;
2738 }
2739 
2740 INDIRECT_CALLABLE_SCOPE struct dst_entry *ip6_dst_check(struct dst_entry *dst,
2741 							u32 cookie)
2742 {
2743 	struct dst_entry *dst_ret;
2744 	struct fib6_info *from;
2745 	struct rt6_info *rt;
2746 
2747 	rt = container_of(dst, struct rt6_info, dst);
2748 
2749 	if (rt->sernum)
2750 		return rt6_is_valid(rt) ? dst : NULL;
2751 
2752 	rcu_read_lock();
2753 
2754 	/* All IPV6 dsts are created with ->obsolete set to the value
2755 	 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2756 	 * into this function always.
2757 	 */
2758 
2759 	from = rcu_dereference(rt->from);
2760 
2761 	if (from && (rt->rt6i_flags & RTF_PCPU ||
2762 	    unlikely(!list_empty(&rt->dst.rt_uncached))))
2763 		dst_ret = rt6_dst_from_check(rt, from, cookie);
2764 	else
2765 		dst_ret = rt6_check(rt, from, cookie);
2766 
2767 	rcu_read_unlock();
2768 
2769 	return dst_ret;
2770 }
2771 EXPORT_INDIRECT_CALLABLE(ip6_dst_check);
2772 
2773 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2774 {
2775 	struct rt6_info *rt = (struct rt6_info *) dst;
2776 
2777 	if (rt) {
2778 		if (rt->rt6i_flags & RTF_CACHE) {
2779 			rcu_read_lock();
2780 			if (rt6_check_expired(rt)) {
2781 				rt6_remove_exception_rt(rt);
2782 				dst = NULL;
2783 			}
2784 			rcu_read_unlock();
2785 		} else {
2786 			dst_release(dst);
2787 			dst = NULL;
2788 		}
2789 	}
2790 	return dst;
2791 }
2792 
2793 static void ip6_link_failure(struct sk_buff *skb)
2794 {
2795 	struct rt6_info *rt;
2796 
2797 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2798 
2799 	rt = (struct rt6_info *) skb_dst(skb);
2800 	if (rt) {
2801 		rcu_read_lock();
2802 		if (rt->rt6i_flags & RTF_CACHE) {
2803 			rt6_remove_exception_rt(rt);
2804 		} else {
2805 			struct fib6_info *from;
2806 			struct fib6_node *fn;
2807 
2808 			from = rcu_dereference(rt->from);
2809 			if (from) {
2810 				fn = rcu_dereference(from->fib6_node);
2811 				if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2812 					WRITE_ONCE(fn->fn_sernum, -1);
2813 			}
2814 		}
2815 		rcu_read_unlock();
2816 	}
2817 }
2818 
2819 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2820 {
2821 	if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2822 		struct fib6_info *from;
2823 
2824 		rcu_read_lock();
2825 		from = rcu_dereference(rt0->from);
2826 		if (from)
2827 			rt0->dst.expires = from->expires;
2828 		rcu_read_unlock();
2829 	}
2830 
2831 	dst_set_expires(&rt0->dst, timeout);
2832 	rt0->rt6i_flags |= RTF_EXPIRES;
2833 }
2834 
2835 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2836 {
2837 	struct net *net = dev_net(rt->dst.dev);
2838 
2839 	dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2840 	rt->rt6i_flags |= RTF_MODIFIED;
2841 	rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2842 }
2843 
2844 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2845 {
2846 	return !(rt->rt6i_flags & RTF_CACHE) &&
2847 		(rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2848 }
2849 
2850 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2851 				 const struct ipv6hdr *iph, u32 mtu,
2852 				 bool confirm_neigh)
2853 {
2854 	const struct in6_addr *daddr, *saddr;
2855 	struct rt6_info *rt6 = (struct rt6_info *)dst;
2856 
2857 	/* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2858 	 * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2859 	 * [see also comment in rt6_mtu_change_route()]
2860 	 */
2861 
2862 	if (iph) {
2863 		daddr = &iph->daddr;
2864 		saddr = &iph->saddr;
2865 	} else if (sk) {
2866 		daddr = &sk->sk_v6_daddr;
2867 		saddr = &inet6_sk(sk)->saddr;
2868 	} else {
2869 		daddr = NULL;
2870 		saddr = NULL;
2871 	}
2872 
2873 	if (confirm_neigh)
2874 		dst_confirm_neigh(dst, daddr);
2875 
2876 	if (mtu < IPV6_MIN_MTU)
2877 		return;
2878 	if (mtu >= dst_mtu(dst))
2879 		return;
2880 
2881 	if (!rt6_cache_allowed_for_pmtu(rt6)) {
2882 		rt6_do_update_pmtu(rt6, mtu);
2883 		/* update rt6_ex->stamp for cache */
2884 		if (rt6->rt6i_flags & RTF_CACHE)
2885 			rt6_update_exception_stamp_rt(rt6);
2886 	} else if (daddr) {
2887 		struct fib6_result res = {};
2888 		struct rt6_info *nrt6;
2889 
2890 		rcu_read_lock();
2891 		res.f6i = rcu_dereference(rt6->from);
2892 		if (!res.f6i)
2893 			goto out_unlock;
2894 
2895 		res.fib6_flags = res.f6i->fib6_flags;
2896 		res.fib6_type = res.f6i->fib6_type;
2897 
2898 		if (res.f6i->nh) {
2899 			struct fib6_nh_match_arg arg = {
2900 				.dev = dst->dev,
2901 				.gw = &rt6->rt6i_gateway,
2902 			};
2903 
2904 			nexthop_for_each_fib6_nh(res.f6i->nh,
2905 						 fib6_nh_find_match, &arg);
2906 
2907 			/* fib6_info uses a nexthop that does not have fib6_nh
2908 			 * using the dst->dev + gw. Should be impossible.
2909 			 */
2910 			if (!arg.match)
2911 				goto out_unlock;
2912 
2913 			res.nh = arg.match;
2914 		} else {
2915 			res.nh = res.f6i->fib6_nh;
2916 		}
2917 
2918 		nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2919 		if (nrt6) {
2920 			rt6_do_update_pmtu(nrt6, mtu);
2921 			if (rt6_insert_exception(nrt6, &res))
2922 				dst_release_immediate(&nrt6->dst);
2923 		}
2924 out_unlock:
2925 		rcu_read_unlock();
2926 	}
2927 }
2928 
2929 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2930 			       struct sk_buff *skb, u32 mtu,
2931 			       bool confirm_neigh)
2932 {
2933 	__ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2934 			     confirm_neigh);
2935 }
2936 
2937 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2938 		     int oif, u32 mark, kuid_t uid)
2939 {
2940 	const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2941 	struct dst_entry *dst;
2942 	struct flowi6 fl6 = {
2943 		.flowi6_oif = oif,
2944 		.flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2945 		.daddr = iph->daddr,
2946 		.saddr = iph->saddr,
2947 		.flowlabel = ip6_flowinfo(iph),
2948 		.flowi6_uid = uid,
2949 	};
2950 
2951 	dst = ip6_route_output(net, NULL, &fl6);
2952 	if (!dst->error)
2953 		__ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2954 	dst_release(dst);
2955 }
2956 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2957 
2958 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2959 {
2960 	int oif = sk->sk_bound_dev_if;
2961 	struct dst_entry *dst;
2962 
2963 	if (!oif && skb->dev)
2964 		oif = l3mdev_master_ifindex(skb->dev);
2965 
2966 	ip6_update_pmtu(skb, sock_net(sk), mtu, oif, READ_ONCE(sk->sk_mark),
2967 			sk->sk_uid);
2968 
2969 	dst = __sk_dst_get(sk);
2970 	if (!dst || !dst->obsolete ||
2971 	    dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2972 		return;
2973 
2974 	bh_lock_sock(sk);
2975 	if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2976 		ip6_datagram_dst_update(sk, false);
2977 	bh_unlock_sock(sk);
2978 }
2979 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2980 
2981 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2982 			   const struct flowi6 *fl6)
2983 {
2984 #ifdef CONFIG_IPV6_SUBTREES
2985 	struct ipv6_pinfo *np = inet6_sk(sk);
2986 #endif
2987 
2988 	ip6_dst_store(sk, dst,
2989 		      ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2990 		      &sk->sk_v6_daddr : NULL,
2991 #ifdef CONFIG_IPV6_SUBTREES
2992 		      ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2993 		      &np->saddr :
2994 #endif
2995 		      NULL);
2996 }
2997 
2998 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2999 				  struct flowi6 *fl6,
3000 				  const struct in6_addr *gw,
3001 				  struct rt6_info **ret)
3002 {
3003 	const struct fib6_nh *nh = res->nh;
3004 
3005 	if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
3006 	    fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
3007 		return false;
3008 
3009 	/* rt_cache's gateway might be different from its 'parent'
3010 	 * in the case of an ip redirect.
3011 	 * So we keep searching in the exception table if the gateway
3012 	 * is different.
3013 	 */
3014 	if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
3015 		struct rt6_info *rt_cache;
3016 
3017 		rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
3018 		if (rt_cache &&
3019 		    ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
3020 			*ret = rt_cache;
3021 			return true;
3022 		}
3023 		return false;
3024 	}
3025 	return true;
3026 }
3027 
3028 struct fib6_nh_rd_arg {
3029 	struct fib6_result	*res;
3030 	struct flowi6		*fl6;
3031 	const struct in6_addr	*gw;
3032 	struct rt6_info		**ret;
3033 };
3034 
3035 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
3036 {
3037 	struct fib6_nh_rd_arg *arg = _arg;
3038 
3039 	arg->res->nh = nh;
3040 	return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
3041 }
3042 
3043 /* Handle redirects */
3044 struct ip6rd_flowi {
3045 	struct flowi6 fl6;
3046 	struct in6_addr gateway;
3047 };
3048 
3049 INDIRECT_CALLABLE_SCOPE struct rt6_info *__ip6_route_redirect(struct net *net,
3050 					     struct fib6_table *table,
3051 					     struct flowi6 *fl6,
3052 					     const struct sk_buff *skb,
3053 					     int flags)
3054 {
3055 	struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
3056 	struct rt6_info *ret = NULL;
3057 	struct fib6_result res = {};
3058 	struct fib6_nh_rd_arg arg = {
3059 		.res = &res,
3060 		.fl6 = fl6,
3061 		.gw  = &rdfl->gateway,
3062 		.ret = &ret
3063 	};
3064 	struct fib6_info *rt;
3065 	struct fib6_node *fn;
3066 
3067 	/* Get the "current" route for this destination and
3068 	 * check if the redirect has come from appropriate router.
3069 	 *
3070 	 * RFC 4861 specifies that redirects should only be
3071 	 * accepted if they come from the nexthop to the target.
3072 	 * Due to the way the routes are chosen, this notion
3073 	 * is a bit fuzzy and one might need to check all possible
3074 	 * routes.
3075 	 */
3076 
3077 	rcu_read_lock();
3078 	fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
3079 restart:
3080 	for_each_fib6_node_rt_rcu(fn) {
3081 		res.f6i = rt;
3082 		if (fib6_check_expired(rt))
3083 			continue;
3084 		if (rt->fib6_flags & RTF_REJECT)
3085 			break;
3086 		if (unlikely(rt->nh)) {
3087 			if (nexthop_is_blackhole(rt->nh))
3088 				continue;
3089 			/* on match, res->nh is filled in and potentially ret */
3090 			if (nexthop_for_each_fib6_nh(rt->nh,
3091 						     fib6_nh_redirect_match,
3092 						     &arg))
3093 				goto out;
3094 		} else {
3095 			res.nh = rt->fib6_nh;
3096 			if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
3097 						  &ret))
3098 				goto out;
3099 		}
3100 	}
3101 
3102 	if (!rt)
3103 		rt = net->ipv6.fib6_null_entry;
3104 	else if (rt->fib6_flags & RTF_REJECT) {
3105 		ret = net->ipv6.ip6_null_entry;
3106 		goto out;
3107 	}
3108 
3109 	if (rt == net->ipv6.fib6_null_entry) {
3110 		fn = fib6_backtrack(fn, &fl6->saddr);
3111 		if (fn)
3112 			goto restart;
3113 	}
3114 
3115 	res.f6i = rt;
3116 	res.nh = rt->fib6_nh;
3117 out:
3118 	if (ret) {
3119 		ip6_hold_safe(net, &ret);
3120 	} else {
3121 		res.fib6_flags = res.f6i->fib6_flags;
3122 		res.fib6_type = res.f6i->fib6_type;
3123 		ret = ip6_create_rt_rcu(&res);
3124 	}
3125 
3126 	rcu_read_unlock();
3127 
3128 	trace_fib6_table_lookup(net, &res, table, fl6);
3129 	return ret;
3130 };
3131 
3132 static struct dst_entry *ip6_route_redirect(struct net *net,
3133 					    const struct flowi6 *fl6,
3134 					    const struct sk_buff *skb,
3135 					    const struct in6_addr *gateway)
3136 {
3137 	int flags = RT6_LOOKUP_F_HAS_SADDR;
3138 	struct ip6rd_flowi rdfl;
3139 
3140 	rdfl.fl6 = *fl6;
3141 	rdfl.gateway = *gateway;
3142 
3143 	return fib6_rule_lookup(net, &rdfl.fl6, skb,
3144 				flags, __ip6_route_redirect);
3145 }
3146 
3147 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
3148 		  kuid_t uid)
3149 {
3150 	const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
3151 	struct dst_entry *dst;
3152 	struct flowi6 fl6 = {
3153 		.flowi6_iif = LOOPBACK_IFINDEX,
3154 		.flowi6_oif = oif,
3155 		.flowi6_mark = mark,
3156 		.daddr = iph->daddr,
3157 		.saddr = iph->saddr,
3158 		.flowlabel = ip6_flowinfo(iph),
3159 		.flowi6_uid = uid,
3160 	};
3161 
3162 	dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3163 	rt6_do_redirect(dst, NULL, skb);
3164 	dst_release(dst);
3165 }
3166 EXPORT_SYMBOL_GPL(ip6_redirect);
3167 
3168 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3169 {
3170 	const struct ipv6hdr *iph = ipv6_hdr(skb);
3171 	const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3172 	struct dst_entry *dst;
3173 	struct flowi6 fl6 = {
3174 		.flowi6_iif = LOOPBACK_IFINDEX,
3175 		.flowi6_oif = oif,
3176 		.daddr = msg->dest,
3177 		.saddr = iph->daddr,
3178 		.flowi6_uid = sock_net_uid(net, NULL),
3179 	};
3180 
3181 	dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3182 	rt6_do_redirect(dst, NULL, skb);
3183 	dst_release(dst);
3184 }
3185 
3186 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3187 {
3188 	ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if,
3189 		     READ_ONCE(sk->sk_mark), sk->sk_uid);
3190 }
3191 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3192 
3193 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3194 {
3195 	struct net_device *dev = dst->dev;
3196 	unsigned int mtu = dst_mtu(dst);
3197 	struct net *net = dev_net(dev);
3198 
3199 	mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3200 
3201 	if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3202 		mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3203 
3204 	/*
3205 	 * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3206 	 * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3207 	 * IPV6_MAXPLEN is also valid and means: "any MSS,
3208 	 * rely only on pmtu discovery"
3209 	 */
3210 	if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3211 		mtu = IPV6_MAXPLEN;
3212 	return mtu;
3213 }
3214 
3215 INDIRECT_CALLABLE_SCOPE unsigned int ip6_mtu(const struct dst_entry *dst)
3216 {
3217 	return ip6_dst_mtu_maybe_forward(dst, false);
3218 }
3219 EXPORT_INDIRECT_CALLABLE(ip6_mtu);
3220 
3221 /* MTU selection:
3222  * 1. mtu on route is locked - use it
3223  * 2. mtu from nexthop exception
3224  * 3. mtu from egress device
3225  *
3226  * based on ip6_dst_mtu_forward and exception logic of
3227  * rt6_find_cached_rt; called with rcu_read_lock
3228  */
3229 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3230 		      const struct in6_addr *daddr,
3231 		      const struct in6_addr *saddr)
3232 {
3233 	const struct fib6_nh *nh = res->nh;
3234 	struct fib6_info *f6i = res->f6i;
3235 	struct inet6_dev *idev;
3236 	struct rt6_info *rt;
3237 	u32 mtu = 0;
3238 
3239 	if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3240 		mtu = f6i->fib6_pmtu;
3241 		if (mtu)
3242 			goto out;
3243 	}
3244 
3245 	rt = rt6_find_cached_rt(res, daddr, saddr);
3246 	if (unlikely(rt)) {
3247 		mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3248 	} else {
3249 		struct net_device *dev = nh->fib_nh_dev;
3250 
3251 		mtu = IPV6_MIN_MTU;
3252 		idev = __in6_dev_get(dev);
3253 		if (idev)
3254 			mtu = max_t(u32, mtu, READ_ONCE(idev->cnf.mtu6));
3255 	}
3256 
3257 	mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3258 out:
3259 	return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3260 }
3261 
3262 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3263 				  struct flowi6 *fl6)
3264 {
3265 	struct dst_entry *dst;
3266 	struct rt6_info *rt;
3267 	struct inet6_dev *idev = in6_dev_get(dev);
3268 	struct net *net = dev_net(dev);
3269 
3270 	if (unlikely(!idev))
3271 		return ERR_PTR(-ENODEV);
3272 
3273 	rt = ip6_dst_alloc(net, dev, 0);
3274 	if (unlikely(!rt)) {
3275 		in6_dev_put(idev);
3276 		dst = ERR_PTR(-ENOMEM);
3277 		goto out;
3278 	}
3279 
3280 	rt->dst.input = ip6_input;
3281 	rt->dst.output  = ip6_output;
3282 	rt->rt6i_gateway  = fl6->daddr;
3283 	rt->rt6i_dst.addr = fl6->daddr;
3284 	rt->rt6i_dst.plen = 128;
3285 	rt->rt6i_idev     = idev;
3286 	dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3287 
3288 	/* Add this dst into uncached_list so that rt6_disable_ip() can
3289 	 * do proper release of the net_device
3290 	 */
3291 	rt6_uncached_list_add(rt);
3292 
3293 	dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3294 
3295 out:
3296 	return dst;
3297 }
3298 
3299 static void ip6_dst_gc(struct dst_ops *ops)
3300 {
3301 	struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3302 	int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3303 	int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3304 	int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3305 	unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3306 	unsigned int val;
3307 	int entries;
3308 
3309 	if (time_after(rt_last_gc + rt_min_interval, jiffies))
3310 		goto out;
3311 
3312 	fib6_run_gc(atomic_inc_return(&net->ipv6.ip6_rt_gc_expire), net, true);
3313 	entries = dst_entries_get_slow(ops);
3314 	if (entries < ops->gc_thresh)
3315 		atomic_set(&net->ipv6.ip6_rt_gc_expire, rt_gc_timeout >> 1);
3316 out:
3317 	val = atomic_read(&net->ipv6.ip6_rt_gc_expire);
3318 	atomic_set(&net->ipv6.ip6_rt_gc_expire, val - (val >> rt_elasticity));
3319 }
3320 
3321 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3322 			       const struct in6_addr *gw_addr, u32 tbid,
3323 			       int flags, struct fib6_result *res)
3324 {
3325 	struct flowi6 fl6 = {
3326 		.flowi6_oif = cfg->fc_ifindex,
3327 		.daddr = *gw_addr,
3328 		.saddr = cfg->fc_prefsrc,
3329 	};
3330 	struct fib6_table *table;
3331 	int err;
3332 
3333 	table = fib6_get_table(net, tbid);
3334 	if (!table)
3335 		return -EINVAL;
3336 
3337 	if (!ipv6_addr_any(&cfg->fc_prefsrc))
3338 		flags |= RT6_LOOKUP_F_HAS_SADDR;
3339 
3340 	flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3341 
3342 	err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3343 	if (!err && res->f6i != net->ipv6.fib6_null_entry)
3344 		fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3345 				 cfg->fc_ifindex != 0, NULL, flags);
3346 
3347 	return err;
3348 }
3349 
3350 static int ip6_route_check_nh_onlink(struct net *net,
3351 				     struct fib6_config *cfg,
3352 				     const struct net_device *dev,
3353 				     struct netlink_ext_ack *extack)
3354 {
3355 	u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3356 	const struct in6_addr *gw_addr = &cfg->fc_gateway;
3357 	struct fib6_result res = {};
3358 	int err;
3359 
3360 	err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3361 	if (!err && !(res.fib6_flags & RTF_REJECT) &&
3362 	    /* ignore match if it is the default route */
3363 	    !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3364 	    (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3365 		NL_SET_ERR_MSG(extack,
3366 			       "Nexthop has invalid gateway or device mismatch");
3367 		err = -EINVAL;
3368 	}
3369 
3370 	return err;
3371 }
3372 
3373 static int ip6_route_check_nh(struct net *net,
3374 			      struct fib6_config *cfg,
3375 			      struct net_device **_dev,
3376 			      netdevice_tracker *dev_tracker,
3377 			      struct inet6_dev **idev)
3378 {
3379 	const struct in6_addr *gw_addr = &cfg->fc_gateway;
3380 	struct net_device *dev = _dev ? *_dev : NULL;
3381 	int flags = RT6_LOOKUP_F_IFACE;
3382 	struct fib6_result res = {};
3383 	int err = -EHOSTUNREACH;
3384 
3385 	if (cfg->fc_table) {
3386 		err = ip6_nh_lookup_table(net, cfg, gw_addr,
3387 					  cfg->fc_table, flags, &res);
3388 		/* gw_addr can not require a gateway or resolve to a reject
3389 		 * route. If a device is given, it must match the result.
3390 		 */
3391 		if (err || res.fib6_flags & RTF_REJECT ||
3392 		    res.nh->fib_nh_gw_family ||
3393 		    (dev && dev != res.nh->fib_nh_dev))
3394 			err = -EHOSTUNREACH;
3395 	}
3396 
3397 	if (err < 0) {
3398 		struct flowi6 fl6 = {
3399 			.flowi6_oif = cfg->fc_ifindex,
3400 			.daddr = *gw_addr,
3401 		};
3402 
3403 		err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3404 		if (err || res.fib6_flags & RTF_REJECT ||
3405 		    res.nh->fib_nh_gw_family)
3406 			err = -EHOSTUNREACH;
3407 
3408 		if (err)
3409 			return err;
3410 
3411 		fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3412 				 cfg->fc_ifindex != 0, NULL, flags);
3413 	}
3414 
3415 	err = 0;
3416 	if (dev) {
3417 		if (dev != res.nh->fib_nh_dev)
3418 			err = -EHOSTUNREACH;
3419 	} else {
3420 		*_dev = dev = res.nh->fib_nh_dev;
3421 		netdev_hold(dev, dev_tracker, GFP_ATOMIC);
3422 		*idev = in6_dev_get(dev);
3423 	}
3424 
3425 	return err;
3426 }
3427 
3428 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3429 			   struct net_device **_dev,
3430 			   netdevice_tracker *dev_tracker,
3431 			   struct inet6_dev **idev,
3432 			   struct netlink_ext_ack *extack)
3433 {
3434 	const struct in6_addr *gw_addr = &cfg->fc_gateway;
3435 	int gwa_type = ipv6_addr_type(gw_addr);
3436 	bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3437 	const struct net_device *dev = *_dev;
3438 	bool need_addr_check = !dev;
3439 	int err = -EINVAL;
3440 
3441 	/* if gw_addr is local we will fail to detect this in case
3442 	 * address is still TENTATIVE (DAD in progress). rt6_lookup()
3443 	 * will return already-added prefix route via interface that
3444 	 * prefix route was assigned to, which might be non-loopback.
3445 	 */
3446 	if (dev &&
3447 	    ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3448 		NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3449 		goto out;
3450 	}
3451 
3452 	if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3453 		/* IPv6 strictly inhibits using not link-local
3454 		 * addresses as nexthop address.
3455 		 * Otherwise, router will not able to send redirects.
3456 		 * It is very good, but in some (rare!) circumstances
3457 		 * (SIT, PtP, NBMA NOARP links) it is handy to allow
3458 		 * some exceptions. --ANK
3459 		 * We allow IPv4-mapped nexthops to support RFC4798-type
3460 		 * addressing
3461 		 */
3462 		if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3463 			NL_SET_ERR_MSG(extack, "Invalid gateway address");
3464 			goto out;
3465 		}
3466 
3467 		rcu_read_lock();
3468 
3469 		if (cfg->fc_flags & RTNH_F_ONLINK)
3470 			err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3471 		else
3472 			err = ip6_route_check_nh(net, cfg, _dev, dev_tracker,
3473 						 idev);
3474 
3475 		rcu_read_unlock();
3476 
3477 		if (err)
3478 			goto out;
3479 	}
3480 
3481 	/* reload in case device was changed */
3482 	dev = *_dev;
3483 
3484 	err = -EINVAL;
3485 	if (!dev) {
3486 		NL_SET_ERR_MSG(extack, "Egress device not specified");
3487 		goto out;
3488 	} else if (dev->flags & IFF_LOOPBACK) {
3489 		NL_SET_ERR_MSG(extack,
3490 			       "Egress device can not be loopback device for this route");
3491 		goto out;
3492 	}
3493 
3494 	/* if we did not check gw_addr above, do so now that the
3495 	 * egress device has been resolved.
3496 	 */
3497 	if (need_addr_check &&
3498 	    ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3499 		NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3500 		goto out;
3501 	}
3502 
3503 	err = 0;
3504 out:
3505 	return err;
3506 }
3507 
3508 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3509 {
3510 	if ((flags & RTF_REJECT) ||
3511 	    (dev && (dev->flags & IFF_LOOPBACK) &&
3512 	     !(addr_type & IPV6_ADDR_LOOPBACK) &&
3513 	     !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3514 		return true;
3515 
3516 	return false;
3517 }
3518 
3519 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3520 		 struct fib6_config *cfg, gfp_t gfp_flags,
3521 		 struct netlink_ext_ack *extack)
3522 {
3523 	netdevice_tracker *dev_tracker = &fib6_nh->fib_nh_dev_tracker;
3524 	struct net_device *dev = NULL;
3525 	struct inet6_dev *idev = NULL;
3526 	int addr_type;
3527 	int err;
3528 
3529 	fib6_nh->fib_nh_family = AF_INET6;
3530 #ifdef CONFIG_IPV6_ROUTER_PREF
3531 	fib6_nh->last_probe = jiffies;
3532 #endif
3533 	if (cfg->fc_is_fdb) {
3534 		fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3535 		fib6_nh->fib_nh_gw_family = AF_INET6;
3536 		return 0;
3537 	}
3538 
3539 	err = -ENODEV;
3540 	if (cfg->fc_ifindex) {
3541 		dev = netdev_get_by_index(net, cfg->fc_ifindex,
3542 					  dev_tracker, gfp_flags);
3543 		if (!dev)
3544 			goto out;
3545 		idev = in6_dev_get(dev);
3546 		if (!idev)
3547 			goto out;
3548 	}
3549 
3550 	if (cfg->fc_flags & RTNH_F_ONLINK) {
3551 		if (!dev) {
3552 			NL_SET_ERR_MSG(extack,
3553 				       "Nexthop device required for onlink");
3554 			goto out;
3555 		}
3556 
3557 		if (!(dev->flags & IFF_UP)) {
3558 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3559 			err = -ENETDOWN;
3560 			goto out;
3561 		}
3562 
3563 		fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3564 	}
3565 
3566 	fib6_nh->fib_nh_weight = 1;
3567 
3568 	/* We cannot add true routes via loopback here,
3569 	 * they would result in kernel looping; promote them to reject routes
3570 	 */
3571 	addr_type = ipv6_addr_type(&cfg->fc_dst);
3572 	if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3573 		/* hold loopback dev/idev if we haven't done so. */
3574 		if (dev != net->loopback_dev) {
3575 			if (dev) {
3576 				netdev_put(dev, dev_tracker);
3577 				in6_dev_put(idev);
3578 			}
3579 			dev = net->loopback_dev;
3580 			netdev_hold(dev, dev_tracker, gfp_flags);
3581 			idev = in6_dev_get(dev);
3582 			if (!idev) {
3583 				err = -ENODEV;
3584 				goto out;
3585 			}
3586 		}
3587 		goto pcpu_alloc;
3588 	}
3589 
3590 	if (cfg->fc_flags & RTF_GATEWAY) {
3591 		err = ip6_validate_gw(net, cfg, &dev, dev_tracker,
3592 				      &idev, extack);
3593 		if (err)
3594 			goto out;
3595 
3596 		fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3597 		fib6_nh->fib_nh_gw_family = AF_INET6;
3598 	}
3599 
3600 	err = -ENODEV;
3601 	if (!dev)
3602 		goto out;
3603 
3604 	if (idev->cnf.disable_ipv6) {
3605 		NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3606 		err = -EACCES;
3607 		goto out;
3608 	}
3609 
3610 	if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3611 		NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3612 		err = -ENETDOWN;
3613 		goto out;
3614 	}
3615 
3616 	if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3617 	    !netif_carrier_ok(dev))
3618 		fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3619 
3620 	err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3621 				 cfg->fc_encap_type, cfg, gfp_flags, extack);
3622 	if (err)
3623 		goto out;
3624 
3625 pcpu_alloc:
3626 	fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3627 	if (!fib6_nh->rt6i_pcpu) {
3628 		err = -ENOMEM;
3629 		goto out;
3630 	}
3631 
3632 	fib6_nh->fib_nh_dev = dev;
3633 	fib6_nh->fib_nh_oif = dev->ifindex;
3634 	err = 0;
3635 out:
3636 	if (idev)
3637 		in6_dev_put(idev);
3638 
3639 	if (err) {
3640 		lwtstate_put(fib6_nh->fib_nh_lws);
3641 		fib6_nh->fib_nh_lws = NULL;
3642 		netdev_put(dev, dev_tracker);
3643 	}
3644 
3645 	return err;
3646 }
3647 
3648 void fib6_nh_release(struct fib6_nh *fib6_nh)
3649 {
3650 	struct rt6_exception_bucket *bucket;
3651 
3652 	rcu_read_lock();
3653 
3654 	fib6_nh_flush_exceptions(fib6_nh, NULL);
3655 	bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3656 	if (bucket) {
3657 		rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3658 		kfree(bucket);
3659 	}
3660 
3661 	rcu_read_unlock();
3662 
3663 	fib6_nh_release_dsts(fib6_nh);
3664 	free_percpu(fib6_nh->rt6i_pcpu);
3665 
3666 	fib_nh_common_release(&fib6_nh->nh_common);
3667 }
3668 
3669 void fib6_nh_release_dsts(struct fib6_nh *fib6_nh)
3670 {
3671 	int cpu;
3672 
3673 	if (!fib6_nh->rt6i_pcpu)
3674 		return;
3675 
3676 	for_each_possible_cpu(cpu) {
3677 		struct rt6_info *pcpu_rt, **ppcpu_rt;
3678 
3679 		ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3680 		pcpu_rt = xchg(ppcpu_rt, NULL);
3681 		if (pcpu_rt) {
3682 			dst_dev_put(&pcpu_rt->dst);
3683 			dst_release(&pcpu_rt->dst);
3684 		}
3685 	}
3686 }
3687 
3688 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3689 					      gfp_t gfp_flags,
3690 					      struct netlink_ext_ack *extack)
3691 {
3692 	struct net *net = cfg->fc_nlinfo.nl_net;
3693 	struct fib6_info *rt = NULL;
3694 	struct nexthop *nh = NULL;
3695 	struct fib6_table *table;
3696 	struct fib6_nh *fib6_nh;
3697 	int err = -EINVAL;
3698 	int addr_type;
3699 
3700 	/* RTF_PCPU is an internal flag; can not be set by userspace */
3701 	if (cfg->fc_flags & RTF_PCPU) {
3702 		NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3703 		goto out;
3704 	}
3705 
3706 	/* RTF_CACHE is an internal flag; can not be set by userspace */
3707 	if (cfg->fc_flags & RTF_CACHE) {
3708 		NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3709 		goto out;
3710 	}
3711 
3712 	if (cfg->fc_type > RTN_MAX) {
3713 		NL_SET_ERR_MSG(extack, "Invalid route type");
3714 		goto out;
3715 	}
3716 
3717 	if (cfg->fc_dst_len > 128) {
3718 		NL_SET_ERR_MSG(extack, "Invalid prefix length");
3719 		goto out;
3720 	}
3721 	if (cfg->fc_src_len > 128) {
3722 		NL_SET_ERR_MSG(extack, "Invalid source address length");
3723 		goto out;
3724 	}
3725 #ifndef CONFIG_IPV6_SUBTREES
3726 	if (cfg->fc_src_len) {
3727 		NL_SET_ERR_MSG(extack,
3728 			       "Specifying source address requires IPV6_SUBTREES to be enabled");
3729 		goto out;
3730 	}
3731 #endif
3732 	if (cfg->fc_nh_id) {
3733 		nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3734 		if (!nh) {
3735 			NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3736 			goto out;
3737 		}
3738 		err = fib6_check_nexthop(nh, cfg, extack);
3739 		if (err)
3740 			goto out;
3741 	}
3742 
3743 	err = -ENOBUFS;
3744 	if (cfg->fc_nlinfo.nlh &&
3745 	    !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3746 		table = fib6_get_table(net, cfg->fc_table);
3747 		if (!table) {
3748 			pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3749 			table = fib6_new_table(net, cfg->fc_table);
3750 		}
3751 	} else {
3752 		table = fib6_new_table(net, cfg->fc_table);
3753 	}
3754 
3755 	if (!table)
3756 		goto out;
3757 
3758 	err = -ENOMEM;
3759 	rt = fib6_info_alloc(gfp_flags, !nh);
3760 	if (!rt)
3761 		goto out;
3762 
3763 	rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3764 					       extack);
3765 	if (IS_ERR(rt->fib6_metrics)) {
3766 		err = PTR_ERR(rt->fib6_metrics);
3767 		/* Do not leave garbage there. */
3768 		rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3769 		goto out_free;
3770 	}
3771 
3772 	if (cfg->fc_flags & RTF_ADDRCONF)
3773 		rt->dst_nocount = true;
3774 
3775 	if (cfg->fc_flags & RTF_EXPIRES)
3776 		fib6_set_expires(rt, jiffies +
3777 				clock_t_to_jiffies(cfg->fc_expires));
3778 
3779 	if (cfg->fc_protocol == RTPROT_UNSPEC)
3780 		cfg->fc_protocol = RTPROT_BOOT;
3781 	rt->fib6_protocol = cfg->fc_protocol;
3782 
3783 	rt->fib6_table = table;
3784 	rt->fib6_metric = cfg->fc_metric;
3785 	rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3786 	rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3787 
3788 	ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3789 	rt->fib6_dst.plen = cfg->fc_dst_len;
3790 
3791 #ifdef CONFIG_IPV6_SUBTREES
3792 	ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3793 	rt->fib6_src.plen = cfg->fc_src_len;
3794 #endif
3795 	if (nh) {
3796 		if (rt->fib6_src.plen) {
3797 			NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3798 			goto out_free;
3799 		}
3800 		if (!nexthop_get(nh)) {
3801 			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3802 			goto out_free;
3803 		}
3804 		rt->nh = nh;
3805 		fib6_nh = nexthop_fib6_nh(rt->nh);
3806 	} else {
3807 		err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3808 		if (err)
3809 			goto out;
3810 
3811 		fib6_nh = rt->fib6_nh;
3812 
3813 		/* We cannot add true routes via loopback here, they would
3814 		 * result in kernel looping; promote them to reject routes
3815 		 */
3816 		addr_type = ipv6_addr_type(&cfg->fc_dst);
3817 		if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3818 				   addr_type))
3819 			rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3820 	}
3821 
3822 	if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3823 		struct net_device *dev = fib6_nh->fib_nh_dev;
3824 
3825 		if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3826 			NL_SET_ERR_MSG(extack, "Invalid source address");
3827 			err = -EINVAL;
3828 			goto out;
3829 		}
3830 		rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3831 		rt->fib6_prefsrc.plen = 128;
3832 	} else
3833 		rt->fib6_prefsrc.plen = 0;
3834 
3835 	return rt;
3836 out:
3837 	fib6_info_release(rt);
3838 	return ERR_PTR(err);
3839 out_free:
3840 	ip_fib_metrics_put(rt->fib6_metrics);
3841 	kfree(rt);
3842 	return ERR_PTR(err);
3843 }
3844 
3845 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3846 		  struct netlink_ext_ack *extack)
3847 {
3848 	struct fib6_info *rt;
3849 	int err;
3850 
3851 	rt = ip6_route_info_create(cfg, gfp_flags, extack);
3852 	if (IS_ERR(rt))
3853 		return PTR_ERR(rt);
3854 
3855 	err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3856 	fib6_info_release(rt);
3857 
3858 	return err;
3859 }
3860 
3861 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3862 {
3863 	struct net *net = info->nl_net;
3864 	struct fib6_table *table;
3865 	int err;
3866 
3867 	if (rt == net->ipv6.fib6_null_entry) {
3868 		err = -ENOENT;
3869 		goto out;
3870 	}
3871 
3872 	table = rt->fib6_table;
3873 	spin_lock_bh(&table->tb6_lock);
3874 	err = fib6_del(rt, info);
3875 	spin_unlock_bh(&table->tb6_lock);
3876 
3877 out:
3878 	fib6_info_release(rt);
3879 	return err;
3880 }
3881 
3882 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3883 {
3884 	struct nl_info info = {
3885 		.nl_net = net,
3886 		.skip_notify = skip_notify
3887 	};
3888 
3889 	return __ip6_del_rt(rt, &info);
3890 }
3891 
3892 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3893 {
3894 	struct nl_info *info = &cfg->fc_nlinfo;
3895 	struct net *net = info->nl_net;
3896 	struct sk_buff *skb = NULL;
3897 	struct fib6_table *table;
3898 	int err = -ENOENT;
3899 
3900 	if (rt == net->ipv6.fib6_null_entry)
3901 		goto out_put;
3902 	table = rt->fib6_table;
3903 	spin_lock_bh(&table->tb6_lock);
3904 
3905 	if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3906 		struct fib6_info *sibling, *next_sibling;
3907 		struct fib6_node *fn;
3908 
3909 		/* prefer to send a single notification with all hops */
3910 		skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3911 		if (skb) {
3912 			u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3913 
3914 			if (rt6_fill_node(net, skb, rt, NULL,
3915 					  NULL, NULL, 0, RTM_DELROUTE,
3916 					  info->portid, seq, 0) < 0) {
3917 				kfree_skb(skb);
3918 				skb = NULL;
3919 			} else
3920 				info->skip_notify = 1;
3921 		}
3922 
3923 		/* 'rt' points to the first sibling route. If it is not the
3924 		 * leaf, then we do not need to send a notification. Otherwise,
3925 		 * we need to check if the last sibling has a next route or not
3926 		 * and emit a replace or delete notification, respectively.
3927 		 */
3928 		info->skip_notify_kernel = 1;
3929 		fn = rcu_dereference_protected(rt->fib6_node,
3930 					    lockdep_is_held(&table->tb6_lock));
3931 		if (rcu_access_pointer(fn->leaf) == rt) {
3932 			struct fib6_info *last_sibling, *replace_rt;
3933 
3934 			last_sibling = list_last_entry(&rt->fib6_siblings,
3935 						       struct fib6_info,
3936 						       fib6_siblings);
3937 			replace_rt = rcu_dereference_protected(
3938 					    last_sibling->fib6_next,
3939 					    lockdep_is_held(&table->tb6_lock));
3940 			if (replace_rt)
3941 				call_fib6_entry_notifiers_replace(net,
3942 								  replace_rt);
3943 			else
3944 				call_fib6_multipath_entry_notifiers(net,
3945 						       FIB_EVENT_ENTRY_DEL,
3946 						       rt, rt->fib6_nsiblings,
3947 						       NULL);
3948 		}
3949 		list_for_each_entry_safe(sibling, next_sibling,
3950 					 &rt->fib6_siblings,
3951 					 fib6_siblings) {
3952 			err = fib6_del(sibling, info);
3953 			if (err)
3954 				goto out_unlock;
3955 		}
3956 	}
3957 
3958 	err = fib6_del(rt, info);
3959 out_unlock:
3960 	spin_unlock_bh(&table->tb6_lock);
3961 out_put:
3962 	fib6_info_release(rt);
3963 
3964 	if (skb) {
3965 		rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3966 			    info->nlh, gfp_any());
3967 	}
3968 	return err;
3969 }
3970 
3971 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3972 {
3973 	int rc = -ESRCH;
3974 
3975 	if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3976 		goto out;
3977 
3978 	if (cfg->fc_flags & RTF_GATEWAY &&
3979 	    !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3980 		goto out;
3981 
3982 	rc = rt6_remove_exception_rt(rt);
3983 out:
3984 	return rc;
3985 }
3986 
3987 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3988 			     struct fib6_nh *nh)
3989 {
3990 	struct fib6_result res = {
3991 		.f6i = rt,
3992 		.nh = nh,
3993 	};
3994 	struct rt6_info *rt_cache;
3995 
3996 	rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3997 	if (rt_cache)
3998 		return __ip6_del_cached_rt(rt_cache, cfg);
3999 
4000 	return 0;
4001 }
4002 
4003 struct fib6_nh_del_cached_rt_arg {
4004 	struct fib6_config *cfg;
4005 	struct fib6_info *f6i;
4006 };
4007 
4008 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
4009 {
4010 	struct fib6_nh_del_cached_rt_arg *arg = _arg;
4011 	int rc;
4012 
4013 	rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
4014 	return rc != -ESRCH ? rc : 0;
4015 }
4016 
4017 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
4018 {
4019 	struct fib6_nh_del_cached_rt_arg arg = {
4020 		.cfg = cfg,
4021 		.f6i = f6i
4022 	};
4023 
4024 	return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
4025 }
4026 
4027 static int ip6_route_del(struct fib6_config *cfg,
4028 			 struct netlink_ext_ack *extack)
4029 {
4030 	struct fib6_table *table;
4031 	struct fib6_info *rt;
4032 	struct fib6_node *fn;
4033 	int err = -ESRCH;
4034 
4035 	table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
4036 	if (!table) {
4037 		NL_SET_ERR_MSG(extack, "FIB table does not exist");
4038 		return err;
4039 	}
4040 
4041 	rcu_read_lock();
4042 
4043 	fn = fib6_locate(&table->tb6_root,
4044 			 &cfg->fc_dst, cfg->fc_dst_len,
4045 			 &cfg->fc_src, cfg->fc_src_len,
4046 			 !(cfg->fc_flags & RTF_CACHE));
4047 
4048 	if (fn) {
4049 		for_each_fib6_node_rt_rcu(fn) {
4050 			struct fib6_nh *nh;
4051 
4052 			if (rt->nh && cfg->fc_nh_id &&
4053 			    rt->nh->id != cfg->fc_nh_id)
4054 				continue;
4055 
4056 			if (cfg->fc_flags & RTF_CACHE) {
4057 				int rc = 0;
4058 
4059 				if (rt->nh) {
4060 					rc = ip6_del_cached_rt_nh(cfg, rt);
4061 				} else if (cfg->fc_nh_id) {
4062 					continue;
4063 				} else {
4064 					nh = rt->fib6_nh;
4065 					rc = ip6_del_cached_rt(cfg, rt, nh);
4066 				}
4067 				if (rc != -ESRCH) {
4068 					rcu_read_unlock();
4069 					return rc;
4070 				}
4071 				continue;
4072 			}
4073 
4074 			if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
4075 				continue;
4076 			if (cfg->fc_protocol &&
4077 			    cfg->fc_protocol != rt->fib6_protocol)
4078 				continue;
4079 
4080 			if (rt->nh) {
4081 				if (!fib6_info_hold_safe(rt))
4082 					continue;
4083 				rcu_read_unlock();
4084 
4085 				return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4086 			}
4087 			if (cfg->fc_nh_id)
4088 				continue;
4089 
4090 			nh = rt->fib6_nh;
4091 			if (cfg->fc_ifindex &&
4092 			    (!nh->fib_nh_dev ||
4093 			     nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
4094 				continue;
4095 			if (cfg->fc_flags & RTF_GATEWAY &&
4096 			    !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
4097 				continue;
4098 			if (!fib6_info_hold_safe(rt))
4099 				continue;
4100 			rcu_read_unlock();
4101 
4102 			/* if gateway was specified only delete the one hop */
4103 			if (cfg->fc_flags & RTF_GATEWAY)
4104 				return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4105 
4106 			return __ip6_del_rt_siblings(rt, cfg);
4107 		}
4108 	}
4109 	rcu_read_unlock();
4110 
4111 	return err;
4112 }
4113 
4114 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4115 {
4116 	struct netevent_redirect netevent;
4117 	struct rt6_info *rt, *nrt = NULL;
4118 	struct fib6_result res = {};
4119 	struct ndisc_options ndopts;
4120 	struct inet6_dev *in6_dev;
4121 	struct neighbour *neigh;
4122 	struct rd_msg *msg;
4123 	int optlen, on_link;
4124 	u8 *lladdr;
4125 
4126 	optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4127 	optlen -= sizeof(*msg);
4128 
4129 	if (optlen < 0) {
4130 		net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4131 		return;
4132 	}
4133 
4134 	msg = (struct rd_msg *)icmp6_hdr(skb);
4135 
4136 	if (ipv6_addr_is_multicast(&msg->dest)) {
4137 		net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4138 		return;
4139 	}
4140 
4141 	on_link = 0;
4142 	if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4143 		on_link = 1;
4144 	} else if (ipv6_addr_type(&msg->target) !=
4145 		   (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4146 		net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4147 		return;
4148 	}
4149 
4150 	in6_dev = __in6_dev_get(skb->dev);
4151 	if (!in6_dev)
4152 		return;
4153 	if (READ_ONCE(in6_dev->cnf.forwarding) ||
4154 	    !READ_ONCE(in6_dev->cnf.accept_redirects))
4155 		return;
4156 
4157 	/* RFC2461 8.1:
4158 	 *	The IP source address of the Redirect MUST be the same as the current
4159 	 *	first-hop router for the specified ICMP Destination Address.
4160 	 */
4161 
4162 	if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4163 		net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4164 		return;
4165 	}
4166 
4167 	lladdr = NULL;
4168 	if (ndopts.nd_opts_tgt_lladdr) {
4169 		lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4170 					     skb->dev);
4171 		if (!lladdr) {
4172 			net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4173 			return;
4174 		}
4175 	}
4176 
4177 	rt = (struct rt6_info *) dst;
4178 	if (rt->rt6i_flags & RTF_REJECT) {
4179 		net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4180 		return;
4181 	}
4182 
4183 	/* Redirect received -> path was valid.
4184 	 * Look, redirects are sent only in response to data packets,
4185 	 * so that this nexthop apparently is reachable. --ANK
4186 	 */
4187 	dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4188 
4189 	neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4190 	if (!neigh)
4191 		return;
4192 
4193 	/*
4194 	 *	We have finally decided to accept it.
4195 	 */
4196 
4197 	ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4198 		     NEIGH_UPDATE_F_WEAK_OVERRIDE|
4199 		     NEIGH_UPDATE_F_OVERRIDE|
4200 		     (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4201 				     NEIGH_UPDATE_F_ISROUTER)),
4202 		     NDISC_REDIRECT, &ndopts);
4203 
4204 	rcu_read_lock();
4205 	res.f6i = rcu_dereference(rt->from);
4206 	if (!res.f6i)
4207 		goto out;
4208 
4209 	if (res.f6i->nh) {
4210 		struct fib6_nh_match_arg arg = {
4211 			.dev = dst->dev,
4212 			.gw = &rt->rt6i_gateway,
4213 		};
4214 
4215 		nexthop_for_each_fib6_nh(res.f6i->nh,
4216 					 fib6_nh_find_match, &arg);
4217 
4218 		/* fib6_info uses a nexthop that does not have fib6_nh
4219 		 * using the dst->dev. Should be impossible
4220 		 */
4221 		if (!arg.match)
4222 			goto out;
4223 		res.nh = arg.match;
4224 	} else {
4225 		res.nh = res.f6i->fib6_nh;
4226 	}
4227 
4228 	res.fib6_flags = res.f6i->fib6_flags;
4229 	res.fib6_type = res.f6i->fib6_type;
4230 	nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4231 	if (!nrt)
4232 		goto out;
4233 
4234 	nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4235 	if (on_link)
4236 		nrt->rt6i_flags &= ~RTF_GATEWAY;
4237 
4238 	nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4239 
4240 	/* rt6_insert_exception() will take care of duplicated exceptions */
4241 	if (rt6_insert_exception(nrt, &res)) {
4242 		dst_release_immediate(&nrt->dst);
4243 		goto out;
4244 	}
4245 
4246 	netevent.old = &rt->dst;
4247 	netevent.new = &nrt->dst;
4248 	netevent.daddr = &msg->dest;
4249 	netevent.neigh = neigh;
4250 	call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4251 
4252 out:
4253 	rcu_read_unlock();
4254 	neigh_release(neigh);
4255 }
4256 
4257 #ifdef CONFIG_IPV6_ROUTE_INFO
4258 static struct fib6_info *rt6_get_route_info(struct net *net,
4259 					   const struct in6_addr *prefix, int prefixlen,
4260 					   const struct in6_addr *gwaddr,
4261 					   struct net_device *dev)
4262 {
4263 	u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4264 	int ifindex = dev->ifindex;
4265 	struct fib6_node *fn;
4266 	struct fib6_info *rt = NULL;
4267 	struct fib6_table *table;
4268 
4269 	table = fib6_get_table(net, tb_id);
4270 	if (!table)
4271 		return NULL;
4272 
4273 	rcu_read_lock();
4274 	fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4275 	if (!fn)
4276 		goto out;
4277 
4278 	for_each_fib6_node_rt_rcu(fn) {
4279 		/* these routes do not use nexthops */
4280 		if (rt->nh)
4281 			continue;
4282 		if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4283 			continue;
4284 		if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4285 		    !rt->fib6_nh->fib_nh_gw_family)
4286 			continue;
4287 		if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4288 			continue;
4289 		if (!fib6_info_hold_safe(rt))
4290 			continue;
4291 		break;
4292 	}
4293 out:
4294 	rcu_read_unlock();
4295 	return rt;
4296 }
4297 
4298 static struct fib6_info *rt6_add_route_info(struct net *net,
4299 					   const struct in6_addr *prefix, int prefixlen,
4300 					   const struct in6_addr *gwaddr,
4301 					   struct net_device *dev,
4302 					   unsigned int pref)
4303 {
4304 	struct fib6_config cfg = {
4305 		.fc_metric	= IP6_RT_PRIO_USER,
4306 		.fc_ifindex	= dev->ifindex,
4307 		.fc_dst_len	= prefixlen,
4308 		.fc_flags	= RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4309 				  RTF_UP | RTF_PREF(pref),
4310 		.fc_protocol = RTPROT_RA,
4311 		.fc_type = RTN_UNICAST,
4312 		.fc_nlinfo.portid = 0,
4313 		.fc_nlinfo.nlh = NULL,
4314 		.fc_nlinfo.nl_net = net,
4315 	};
4316 
4317 	cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4318 	cfg.fc_dst = *prefix;
4319 	cfg.fc_gateway = *gwaddr;
4320 
4321 	/* We should treat it as a default route if prefix length is 0. */
4322 	if (!prefixlen)
4323 		cfg.fc_flags |= RTF_DEFAULT;
4324 
4325 	ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4326 
4327 	return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4328 }
4329 #endif
4330 
4331 struct fib6_info *rt6_get_dflt_router(struct net *net,
4332 				     const struct in6_addr *addr,
4333 				     struct net_device *dev)
4334 {
4335 	u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4336 	struct fib6_info *rt;
4337 	struct fib6_table *table;
4338 
4339 	table = fib6_get_table(net, tb_id);
4340 	if (!table)
4341 		return NULL;
4342 
4343 	rcu_read_lock();
4344 	for_each_fib6_node_rt_rcu(&table->tb6_root) {
4345 		struct fib6_nh *nh;
4346 
4347 		/* RA routes do not use nexthops */
4348 		if (rt->nh)
4349 			continue;
4350 
4351 		nh = rt->fib6_nh;
4352 		if (dev == nh->fib_nh_dev &&
4353 		    ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4354 		    ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4355 			break;
4356 	}
4357 	if (rt && !fib6_info_hold_safe(rt))
4358 		rt = NULL;
4359 	rcu_read_unlock();
4360 	return rt;
4361 }
4362 
4363 struct fib6_info *rt6_add_dflt_router(struct net *net,
4364 				     const struct in6_addr *gwaddr,
4365 				     struct net_device *dev,
4366 				     unsigned int pref,
4367 				     u32 defrtr_usr_metric,
4368 				     int lifetime)
4369 {
4370 	struct fib6_config cfg = {
4371 		.fc_table	= l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4372 		.fc_metric	= defrtr_usr_metric,
4373 		.fc_ifindex	= dev->ifindex,
4374 		.fc_flags	= RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4375 				  RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4376 		.fc_protocol = RTPROT_RA,
4377 		.fc_type = RTN_UNICAST,
4378 		.fc_nlinfo.portid = 0,
4379 		.fc_nlinfo.nlh = NULL,
4380 		.fc_nlinfo.nl_net = net,
4381 		.fc_expires = jiffies_to_clock_t(lifetime * HZ),
4382 	};
4383 
4384 	cfg.fc_gateway = *gwaddr;
4385 
4386 	if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4387 		struct fib6_table *table;
4388 
4389 		table = fib6_get_table(dev_net(dev), cfg.fc_table);
4390 		if (table)
4391 			table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4392 	}
4393 
4394 	return rt6_get_dflt_router(net, gwaddr, dev);
4395 }
4396 
4397 static void __rt6_purge_dflt_routers(struct net *net,
4398 				     struct fib6_table *table)
4399 {
4400 	struct fib6_info *rt;
4401 
4402 restart:
4403 	rcu_read_lock();
4404 	for_each_fib6_node_rt_rcu(&table->tb6_root) {
4405 		struct net_device *dev = fib6_info_nh_dev(rt);
4406 		struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4407 
4408 		if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4409 		    (!idev || idev->cnf.accept_ra != 2) &&
4410 		    fib6_info_hold_safe(rt)) {
4411 			rcu_read_unlock();
4412 			ip6_del_rt(net, rt, false);
4413 			goto restart;
4414 		}
4415 	}
4416 	rcu_read_unlock();
4417 
4418 	table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4419 }
4420 
4421 void rt6_purge_dflt_routers(struct net *net)
4422 {
4423 	struct fib6_table *table;
4424 	struct hlist_head *head;
4425 	unsigned int h;
4426 
4427 	rcu_read_lock();
4428 
4429 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4430 		head = &net->ipv6.fib_table_hash[h];
4431 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4432 			if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4433 				__rt6_purge_dflt_routers(net, table);
4434 		}
4435 	}
4436 
4437 	rcu_read_unlock();
4438 }
4439 
4440 static void rtmsg_to_fib6_config(struct net *net,
4441 				 struct in6_rtmsg *rtmsg,
4442 				 struct fib6_config *cfg)
4443 {
4444 	*cfg = (struct fib6_config){
4445 		.fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4446 			 : RT6_TABLE_MAIN,
4447 		.fc_ifindex = rtmsg->rtmsg_ifindex,
4448 		.fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4449 		.fc_expires = rtmsg->rtmsg_info,
4450 		.fc_dst_len = rtmsg->rtmsg_dst_len,
4451 		.fc_src_len = rtmsg->rtmsg_src_len,
4452 		.fc_flags = rtmsg->rtmsg_flags,
4453 		.fc_type = rtmsg->rtmsg_type,
4454 
4455 		.fc_nlinfo.nl_net = net,
4456 
4457 		.fc_dst = rtmsg->rtmsg_dst,
4458 		.fc_src = rtmsg->rtmsg_src,
4459 		.fc_gateway = rtmsg->rtmsg_gateway,
4460 	};
4461 }
4462 
4463 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4464 {
4465 	struct fib6_config cfg;
4466 	int err;
4467 
4468 	if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4469 		return -EINVAL;
4470 	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4471 		return -EPERM;
4472 
4473 	rtmsg_to_fib6_config(net, rtmsg, &cfg);
4474 
4475 	rtnl_lock();
4476 	switch (cmd) {
4477 	case SIOCADDRT:
4478 		err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4479 		break;
4480 	case SIOCDELRT:
4481 		err = ip6_route_del(&cfg, NULL);
4482 		break;
4483 	}
4484 	rtnl_unlock();
4485 	return err;
4486 }
4487 
4488 /*
4489  *	Drop the packet on the floor
4490  */
4491 
4492 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4493 {
4494 	struct dst_entry *dst = skb_dst(skb);
4495 	struct net *net = dev_net(dst->dev);
4496 	struct inet6_dev *idev;
4497 	SKB_DR(reason);
4498 	int type;
4499 
4500 	if (netif_is_l3_master(skb->dev) ||
4501 	    dst->dev == net->loopback_dev)
4502 		idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4503 	else
4504 		idev = ip6_dst_idev(dst);
4505 
4506 	switch (ipstats_mib_noroutes) {
4507 	case IPSTATS_MIB_INNOROUTES:
4508 		type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4509 		if (type == IPV6_ADDR_ANY) {
4510 			SKB_DR_SET(reason, IP_INADDRERRORS);
4511 			IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4512 			break;
4513 		}
4514 		SKB_DR_SET(reason, IP_INNOROUTES);
4515 		fallthrough;
4516 	case IPSTATS_MIB_OUTNOROUTES:
4517 		SKB_DR_OR(reason, IP_OUTNOROUTES);
4518 		IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4519 		break;
4520 	}
4521 
4522 	/* Start over by dropping the dst for l3mdev case */
4523 	if (netif_is_l3_master(skb->dev))
4524 		skb_dst_drop(skb);
4525 
4526 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4527 	kfree_skb_reason(skb, reason);
4528 	return 0;
4529 }
4530 
4531 static int ip6_pkt_discard(struct sk_buff *skb)
4532 {
4533 	return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4534 }
4535 
4536 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4537 {
4538 	skb->dev = skb_dst(skb)->dev;
4539 	return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4540 }
4541 
4542 static int ip6_pkt_prohibit(struct sk_buff *skb)
4543 {
4544 	return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4545 }
4546 
4547 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4548 {
4549 	skb->dev = skb_dst(skb)->dev;
4550 	return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4551 }
4552 
4553 /*
4554  *	Allocate a dst for local (unicast / anycast) address.
4555  */
4556 
4557 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4558 				     struct inet6_dev *idev,
4559 				     const struct in6_addr *addr,
4560 				     bool anycast, gfp_t gfp_flags,
4561 				     struct netlink_ext_ack *extack)
4562 {
4563 	struct fib6_config cfg = {
4564 		.fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4565 		.fc_ifindex = idev->dev->ifindex,
4566 		.fc_flags = RTF_UP | RTF_NONEXTHOP,
4567 		.fc_dst = *addr,
4568 		.fc_dst_len = 128,
4569 		.fc_protocol = RTPROT_KERNEL,
4570 		.fc_nlinfo.nl_net = net,
4571 		.fc_ignore_dev_down = true,
4572 	};
4573 	struct fib6_info *f6i;
4574 
4575 	if (anycast) {
4576 		cfg.fc_type = RTN_ANYCAST;
4577 		cfg.fc_flags |= RTF_ANYCAST;
4578 	} else {
4579 		cfg.fc_type = RTN_LOCAL;
4580 		cfg.fc_flags |= RTF_LOCAL;
4581 	}
4582 
4583 	f6i = ip6_route_info_create(&cfg, gfp_flags, extack);
4584 	if (!IS_ERR(f6i)) {
4585 		f6i->dst_nocount = true;
4586 
4587 		if (!anycast &&
4588 		    (READ_ONCE(net->ipv6.devconf_all->disable_policy) ||
4589 		     READ_ONCE(idev->cnf.disable_policy)))
4590 			f6i->dst_nopolicy = true;
4591 	}
4592 
4593 	return f6i;
4594 }
4595 
4596 /* remove deleted ip from prefsrc entries */
4597 struct arg_dev_net_ip {
4598 	struct net *net;
4599 	struct in6_addr *addr;
4600 };
4601 
4602 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4603 {
4604 	struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4605 	struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4606 
4607 	if (!rt->nh &&
4608 	    rt != net->ipv6.fib6_null_entry &&
4609 	    ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr) &&
4610 	    !ipv6_chk_addr(net, addr, rt->fib6_nh->fib_nh_dev, 0)) {
4611 		spin_lock_bh(&rt6_exception_lock);
4612 		/* remove prefsrc entry */
4613 		rt->fib6_prefsrc.plen = 0;
4614 		spin_unlock_bh(&rt6_exception_lock);
4615 	}
4616 	return 0;
4617 }
4618 
4619 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4620 {
4621 	struct net *net = dev_net(ifp->idev->dev);
4622 	struct arg_dev_net_ip adni = {
4623 		.net = net,
4624 		.addr = &ifp->addr,
4625 	};
4626 	fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4627 }
4628 
4629 #define RTF_RA_ROUTER		(RTF_ADDRCONF | RTF_DEFAULT)
4630 
4631 /* Remove routers and update dst entries when gateway turn into host. */
4632 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4633 {
4634 	struct in6_addr *gateway = (struct in6_addr *)arg;
4635 	struct fib6_nh *nh;
4636 
4637 	/* RA routes do not use nexthops */
4638 	if (rt->nh)
4639 		return 0;
4640 
4641 	nh = rt->fib6_nh;
4642 	if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4643 	    nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4644 		return -1;
4645 
4646 	/* Further clean up cached routes in exception table.
4647 	 * This is needed because cached route may have a different
4648 	 * gateway than its 'parent' in the case of an ip redirect.
4649 	 */
4650 	fib6_nh_exceptions_clean_tohost(nh, gateway);
4651 
4652 	return 0;
4653 }
4654 
4655 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4656 {
4657 	fib6_clean_all(net, fib6_clean_tohost, gateway);
4658 }
4659 
4660 struct arg_netdev_event {
4661 	const struct net_device *dev;
4662 	union {
4663 		unsigned char nh_flags;
4664 		unsigned long event;
4665 	};
4666 };
4667 
4668 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4669 {
4670 	struct fib6_info *iter;
4671 	struct fib6_node *fn;
4672 
4673 	fn = rcu_dereference_protected(rt->fib6_node,
4674 			lockdep_is_held(&rt->fib6_table->tb6_lock));
4675 	iter = rcu_dereference_protected(fn->leaf,
4676 			lockdep_is_held(&rt->fib6_table->tb6_lock));
4677 	while (iter) {
4678 		if (iter->fib6_metric == rt->fib6_metric &&
4679 		    rt6_qualify_for_ecmp(iter))
4680 			return iter;
4681 		iter = rcu_dereference_protected(iter->fib6_next,
4682 				lockdep_is_held(&rt->fib6_table->tb6_lock));
4683 	}
4684 
4685 	return NULL;
4686 }
4687 
4688 /* only called for fib entries with builtin fib6_nh */
4689 static bool rt6_is_dead(const struct fib6_info *rt)
4690 {
4691 	if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4692 	    (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4693 	     ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4694 		return true;
4695 
4696 	return false;
4697 }
4698 
4699 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4700 {
4701 	struct fib6_info *iter;
4702 	int total = 0;
4703 
4704 	if (!rt6_is_dead(rt))
4705 		total += rt->fib6_nh->fib_nh_weight;
4706 
4707 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4708 		if (!rt6_is_dead(iter))
4709 			total += iter->fib6_nh->fib_nh_weight;
4710 	}
4711 
4712 	return total;
4713 }
4714 
4715 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4716 {
4717 	int upper_bound = -1;
4718 
4719 	if (!rt6_is_dead(rt)) {
4720 		*weight += rt->fib6_nh->fib_nh_weight;
4721 		upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4722 						    total) - 1;
4723 	}
4724 	atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4725 }
4726 
4727 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4728 {
4729 	struct fib6_info *iter;
4730 	int weight = 0;
4731 
4732 	rt6_upper_bound_set(rt, &weight, total);
4733 
4734 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4735 		rt6_upper_bound_set(iter, &weight, total);
4736 }
4737 
4738 void rt6_multipath_rebalance(struct fib6_info *rt)
4739 {
4740 	struct fib6_info *first;
4741 	int total;
4742 
4743 	/* In case the entire multipath route was marked for flushing,
4744 	 * then there is no need to rebalance upon the removal of every
4745 	 * sibling route.
4746 	 */
4747 	if (!rt->fib6_nsiblings || rt->should_flush)
4748 		return;
4749 
4750 	/* During lookup routes are evaluated in order, so we need to
4751 	 * make sure upper bounds are assigned from the first sibling
4752 	 * onwards.
4753 	 */
4754 	first = rt6_multipath_first_sibling(rt);
4755 	if (WARN_ON_ONCE(!first))
4756 		return;
4757 
4758 	total = rt6_multipath_total_weight(first);
4759 	rt6_multipath_upper_bound_set(first, total);
4760 }
4761 
4762 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4763 {
4764 	const struct arg_netdev_event *arg = p_arg;
4765 	struct net *net = dev_net(arg->dev);
4766 
4767 	if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4768 	    rt->fib6_nh->fib_nh_dev == arg->dev) {
4769 		rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4770 		fib6_update_sernum_upto_root(net, rt);
4771 		rt6_multipath_rebalance(rt);
4772 	}
4773 
4774 	return 0;
4775 }
4776 
4777 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4778 {
4779 	struct arg_netdev_event arg = {
4780 		.dev = dev,
4781 		{
4782 			.nh_flags = nh_flags,
4783 		},
4784 	};
4785 
4786 	if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4787 		arg.nh_flags |= RTNH_F_LINKDOWN;
4788 
4789 	fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4790 }
4791 
4792 /* only called for fib entries with inline fib6_nh */
4793 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4794 				   const struct net_device *dev)
4795 {
4796 	struct fib6_info *iter;
4797 
4798 	if (rt->fib6_nh->fib_nh_dev == dev)
4799 		return true;
4800 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4801 		if (iter->fib6_nh->fib_nh_dev == dev)
4802 			return true;
4803 
4804 	return false;
4805 }
4806 
4807 static void rt6_multipath_flush(struct fib6_info *rt)
4808 {
4809 	struct fib6_info *iter;
4810 
4811 	rt->should_flush = 1;
4812 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4813 		iter->should_flush = 1;
4814 }
4815 
4816 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4817 					     const struct net_device *down_dev)
4818 {
4819 	struct fib6_info *iter;
4820 	unsigned int dead = 0;
4821 
4822 	if (rt->fib6_nh->fib_nh_dev == down_dev ||
4823 	    rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4824 		dead++;
4825 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4826 		if (iter->fib6_nh->fib_nh_dev == down_dev ||
4827 		    iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4828 			dead++;
4829 
4830 	return dead;
4831 }
4832 
4833 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4834 				       const struct net_device *dev,
4835 				       unsigned char nh_flags)
4836 {
4837 	struct fib6_info *iter;
4838 
4839 	if (rt->fib6_nh->fib_nh_dev == dev)
4840 		rt->fib6_nh->fib_nh_flags |= nh_flags;
4841 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4842 		if (iter->fib6_nh->fib_nh_dev == dev)
4843 			iter->fib6_nh->fib_nh_flags |= nh_flags;
4844 }
4845 
4846 /* called with write lock held for table with rt */
4847 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4848 {
4849 	const struct arg_netdev_event *arg = p_arg;
4850 	const struct net_device *dev = arg->dev;
4851 	struct net *net = dev_net(dev);
4852 
4853 	if (rt == net->ipv6.fib6_null_entry || rt->nh)
4854 		return 0;
4855 
4856 	switch (arg->event) {
4857 	case NETDEV_UNREGISTER:
4858 		return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4859 	case NETDEV_DOWN:
4860 		if (rt->should_flush)
4861 			return -1;
4862 		if (!rt->fib6_nsiblings)
4863 			return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4864 		if (rt6_multipath_uses_dev(rt, dev)) {
4865 			unsigned int count;
4866 
4867 			count = rt6_multipath_dead_count(rt, dev);
4868 			if (rt->fib6_nsiblings + 1 == count) {
4869 				rt6_multipath_flush(rt);
4870 				return -1;
4871 			}
4872 			rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4873 						   RTNH_F_LINKDOWN);
4874 			fib6_update_sernum(net, rt);
4875 			rt6_multipath_rebalance(rt);
4876 		}
4877 		return -2;
4878 	case NETDEV_CHANGE:
4879 		if (rt->fib6_nh->fib_nh_dev != dev ||
4880 		    rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4881 			break;
4882 		rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4883 		rt6_multipath_rebalance(rt);
4884 		break;
4885 	}
4886 
4887 	return 0;
4888 }
4889 
4890 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4891 {
4892 	struct arg_netdev_event arg = {
4893 		.dev = dev,
4894 		{
4895 			.event = event,
4896 		},
4897 	};
4898 	struct net *net = dev_net(dev);
4899 
4900 	if (net->ipv6.sysctl.skip_notify_on_dev_down)
4901 		fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4902 	else
4903 		fib6_clean_all(net, fib6_ifdown, &arg);
4904 }
4905 
4906 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4907 {
4908 	rt6_sync_down_dev(dev, event);
4909 	rt6_uncached_list_flush_dev(dev);
4910 	neigh_ifdown(&nd_tbl, dev);
4911 }
4912 
4913 struct rt6_mtu_change_arg {
4914 	struct net_device *dev;
4915 	unsigned int mtu;
4916 	struct fib6_info *f6i;
4917 };
4918 
4919 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4920 {
4921 	struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4922 	struct fib6_info *f6i = arg->f6i;
4923 
4924 	/* For administrative MTU increase, there is no way to discover
4925 	 * IPv6 PMTU increase, so PMTU increase should be updated here.
4926 	 * Since RFC 1981 doesn't include administrative MTU increase
4927 	 * update PMTU increase is a MUST. (i.e. jumbo frame)
4928 	 */
4929 	if (nh->fib_nh_dev == arg->dev) {
4930 		struct inet6_dev *idev = __in6_dev_get(arg->dev);
4931 		u32 mtu = f6i->fib6_pmtu;
4932 
4933 		if (mtu >= arg->mtu ||
4934 		    (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4935 			fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4936 
4937 		spin_lock_bh(&rt6_exception_lock);
4938 		rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4939 		spin_unlock_bh(&rt6_exception_lock);
4940 	}
4941 
4942 	return 0;
4943 }
4944 
4945 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4946 {
4947 	struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4948 	struct inet6_dev *idev;
4949 
4950 	/* In IPv6 pmtu discovery is not optional,
4951 	   so that RTAX_MTU lock cannot disable it.
4952 	   We still use this lock to block changes
4953 	   caused by addrconf/ndisc.
4954 	*/
4955 
4956 	idev = __in6_dev_get(arg->dev);
4957 	if (!idev)
4958 		return 0;
4959 
4960 	if (fib6_metric_locked(f6i, RTAX_MTU))
4961 		return 0;
4962 
4963 	arg->f6i = f6i;
4964 	if (f6i->nh) {
4965 		/* fib6_nh_mtu_change only returns 0, so this is safe */
4966 		return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4967 						arg);
4968 	}
4969 
4970 	return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4971 }
4972 
4973 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4974 {
4975 	struct rt6_mtu_change_arg arg = {
4976 		.dev = dev,
4977 		.mtu = mtu,
4978 	};
4979 
4980 	fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4981 }
4982 
4983 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4984 	[RTA_UNSPEC]		= { .strict_start_type = RTA_DPORT + 1 },
4985 	[RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4986 	[RTA_PREFSRC]		= { .len = sizeof(struct in6_addr) },
4987 	[RTA_OIF]               = { .type = NLA_U32 },
4988 	[RTA_IIF]		= { .type = NLA_U32 },
4989 	[RTA_PRIORITY]          = { .type = NLA_U32 },
4990 	[RTA_METRICS]           = { .type = NLA_NESTED },
4991 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
4992 	[RTA_PREF]              = { .type = NLA_U8 },
4993 	[RTA_ENCAP_TYPE]	= { .type = NLA_U16 },
4994 	[RTA_ENCAP]		= { .type = NLA_NESTED },
4995 	[RTA_EXPIRES]		= { .type = NLA_U32 },
4996 	[RTA_UID]		= { .type = NLA_U32 },
4997 	[RTA_MARK]		= { .type = NLA_U32 },
4998 	[RTA_TABLE]		= { .type = NLA_U32 },
4999 	[RTA_IP_PROTO]		= { .type = NLA_U8 },
5000 	[RTA_SPORT]		= { .type = NLA_U16 },
5001 	[RTA_DPORT]		= { .type = NLA_U16 },
5002 	[RTA_NH_ID]		= { .type = NLA_U32 },
5003 };
5004 
5005 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
5006 			      struct fib6_config *cfg,
5007 			      struct netlink_ext_ack *extack)
5008 {
5009 	struct rtmsg *rtm;
5010 	struct nlattr *tb[RTA_MAX+1];
5011 	unsigned int pref;
5012 	int err;
5013 
5014 	err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5015 				     rtm_ipv6_policy, extack);
5016 	if (err < 0)
5017 		goto errout;
5018 
5019 	err = -EINVAL;
5020 	rtm = nlmsg_data(nlh);
5021 
5022 	if (rtm->rtm_tos) {
5023 		NL_SET_ERR_MSG(extack,
5024 			       "Invalid dsfield (tos): option not available for IPv6");
5025 		goto errout;
5026 	}
5027 
5028 	*cfg = (struct fib6_config){
5029 		.fc_table = rtm->rtm_table,
5030 		.fc_dst_len = rtm->rtm_dst_len,
5031 		.fc_src_len = rtm->rtm_src_len,
5032 		.fc_flags = RTF_UP,
5033 		.fc_protocol = rtm->rtm_protocol,
5034 		.fc_type = rtm->rtm_type,
5035 
5036 		.fc_nlinfo.portid = NETLINK_CB(skb).portid,
5037 		.fc_nlinfo.nlh = nlh,
5038 		.fc_nlinfo.nl_net = sock_net(skb->sk),
5039 	};
5040 
5041 	if (rtm->rtm_type == RTN_UNREACHABLE ||
5042 	    rtm->rtm_type == RTN_BLACKHOLE ||
5043 	    rtm->rtm_type == RTN_PROHIBIT ||
5044 	    rtm->rtm_type == RTN_THROW)
5045 		cfg->fc_flags |= RTF_REJECT;
5046 
5047 	if (rtm->rtm_type == RTN_LOCAL)
5048 		cfg->fc_flags |= RTF_LOCAL;
5049 
5050 	if (rtm->rtm_flags & RTM_F_CLONED)
5051 		cfg->fc_flags |= RTF_CACHE;
5052 
5053 	cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
5054 
5055 	if (tb[RTA_NH_ID]) {
5056 		if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
5057 		    tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
5058 			NL_SET_ERR_MSG(extack,
5059 				       "Nexthop specification and nexthop id are mutually exclusive");
5060 			goto errout;
5061 		}
5062 		cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
5063 	}
5064 
5065 	if (tb[RTA_GATEWAY]) {
5066 		cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
5067 		cfg->fc_flags |= RTF_GATEWAY;
5068 	}
5069 	if (tb[RTA_VIA]) {
5070 		NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
5071 		goto errout;
5072 	}
5073 
5074 	if (tb[RTA_DST]) {
5075 		int plen = (rtm->rtm_dst_len + 7) >> 3;
5076 
5077 		if (nla_len(tb[RTA_DST]) < plen)
5078 			goto errout;
5079 
5080 		nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
5081 	}
5082 
5083 	if (tb[RTA_SRC]) {
5084 		int plen = (rtm->rtm_src_len + 7) >> 3;
5085 
5086 		if (nla_len(tb[RTA_SRC]) < plen)
5087 			goto errout;
5088 
5089 		nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
5090 	}
5091 
5092 	if (tb[RTA_PREFSRC])
5093 		cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
5094 
5095 	if (tb[RTA_OIF])
5096 		cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
5097 
5098 	if (tb[RTA_PRIORITY])
5099 		cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
5100 
5101 	if (tb[RTA_METRICS]) {
5102 		cfg->fc_mx = nla_data(tb[RTA_METRICS]);
5103 		cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
5104 	}
5105 
5106 	if (tb[RTA_TABLE])
5107 		cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
5108 
5109 	if (tb[RTA_MULTIPATH]) {
5110 		cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
5111 		cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
5112 
5113 		err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
5114 						     cfg->fc_mp_len, extack);
5115 		if (err < 0)
5116 			goto errout;
5117 	}
5118 
5119 	if (tb[RTA_PREF]) {
5120 		pref = nla_get_u8(tb[RTA_PREF]);
5121 		if (pref != ICMPV6_ROUTER_PREF_LOW &&
5122 		    pref != ICMPV6_ROUTER_PREF_HIGH)
5123 			pref = ICMPV6_ROUTER_PREF_MEDIUM;
5124 		cfg->fc_flags |= RTF_PREF(pref);
5125 	}
5126 
5127 	if (tb[RTA_ENCAP])
5128 		cfg->fc_encap = tb[RTA_ENCAP];
5129 
5130 	if (tb[RTA_ENCAP_TYPE]) {
5131 		cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5132 
5133 		err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5134 		if (err < 0)
5135 			goto errout;
5136 	}
5137 
5138 	if (tb[RTA_EXPIRES]) {
5139 		unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5140 
5141 		if (addrconf_finite_timeout(timeout)) {
5142 			cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5143 			cfg->fc_flags |= RTF_EXPIRES;
5144 		}
5145 	}
5146 
5147 	err = 0;
5148 errout:
5149 	return err;
5150 }
5151 
5152 struct rt6_nh {
5153 	struct fib6_info *fib6_info;
5154 	struct fib6_config r_cfg;
5155 	struct list_head next;
5156 };
5157 
5158 static int ip6_route_info_append(struct net *net,
5159 				 struct list_head *rt6_nh_list,
5160 				 struct fib6_info *rt,
5161 				 struct fib6_config *r_cfg)
5162 {
5163 	struct rt6_nh *nh;
5164 	int err = -EEXIST;
5165 
5166 	list_for_each_entry(nh, rt6_nh_list, next) {
5167 		/* check if fib6_info already exists */
5168 		if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5169 			return err;
5170 	}
5171 
5172 	nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5173 	if (!nh)
5174 		return -ENOMEM;
5175 	nh->fib6_info = rt;
5176 	memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5177 	list_add_tail(&nh->next, rt6_nh_list);
5178 
5179 	return 0;
5180 }
5181 
5182 static void ip6_route_mpath_notify(struct fib6_info *rt,
5183 				   struct fib6_info *rt_last,
5184 				   struct nl_info *info,
5185 				   __u16 nlflags)
5186 {
5187 	/* if this is an APPEND route, then rt points to the first route
5188 	 * inserted and rt_last points to last route inserted. Userspace
5189 	 * wants a consistent dump of the route which starts at the first
5190 	 * nexthop. Since sibling routes are always added at the end of
5191 	 * the list, find the first sibling of the last route appended
5192 	 */
5193 	if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5194 		rt = list_first_entry(&rt_last->fib6_siblings,
5195 				      struct fib6_info,
5196 				      fib6_siblings);
5197 	}
5198 
5199 	if (rt)
5200 		inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5201 }
5202 
5203 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5204 {
5205 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5206 	bool should_notify = false;
5207 	struct fib6_info *leaf;
5208 	struct fib6_node *fn;
5209 
5210 	rcu_read_lock();
5211 	fn = rcu_dereference(rt->fib6_node);
5212 	if (!fn)
5213 		goto out;
5214 
5215 	leaf = rcu_dereference(fn->leaf);
5216 	if (!leaf)
5217 		goto out;
5218 
5219 	if (rt == leaf ||
5220 	    (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5221 	     rt6_qualify_for_ecmp(leaf)))
5222 		should_notify = true;
5223 out:
5224 	rcu_read_unlock();
5225 
5226 	return should_notify;
5227 }
5228 
5229 static int fib6_gw_from_attr(struct in6_addr *gw, struct nlattr *nla,
5230 			     struct netlink_ext_ack *extack)
5231 {
5232 	if (nla_len(nla) < sizeof(*gw)) {
5233 		NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_GATEWAY");
5234 		return -EINVAL;
5235 	}
5236 
5237 	*gw = nla_get_in6_addr(nla);
5238 
5239 	return 0;
5240 }
5241 
5242 static int ip6_route_multipath_add(struct fib6_config *cfg,
5243 				   struct netlink_ext_ack *extack)
5244 {
5245 	struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5246 	struct nl_info *info = &cfg->fc_nlinfo;
5247 	struct fib6_config r_cfg;
5248 	struct rtnexthop *rtnh;
5249 	struct fib6_info *rt;
5250 	struct rt6_nh *err_nh;
5251 	struct rt6_nh *nh, *nh_safe;
5252 	__u16 nlflags;
5253 	int remaining;
5254 	int attrlen;
5255 	int err = 1;
5256 	int nhn = 0;
5257 	int replace = (cfg->fc_nlinfo.nlh &&
5258 		       (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5259 	LIST_HEAD(rt6_nh_list);
5260 
5261 	nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5262 	if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5263 		nlflags |= NLM_F_APPEND;
5264 
5265 	remaining = cfg->fc_mp_len;
5266 	rtnh = (struct rtnexthop *)cfg->fc_mp;
5267 
5268 	/* Parse a Multipath Entry and build a list (rt6_nh_list) of
5269 	 * fib6_info structs per nexthop
5270 	 */
5271 	while (rtnh_ok(rtnh, remaining)) {
5272 		memcpy(&r_cfg, cfg, sizeof(*cfg));
5273 		if (rtnh->rtnh_ifindex)
5274 			r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5275 
5276 		attrlen = rtnh_attrlen(rtnh);
5277 		if (attrlen > 0) {
5278 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5279 
5280 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5281 			if (nla) {
5282 				err = fib6_gw_from_attr(&r_cfg.fc_gateway, nla,
5283 							extack);
5284 				if (err)
5285 					goto cleanup;
5286 
5287 				r_cfg.fc_flags |= RTF_GATEWAY;
5288 			}
5289 			r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5290 
5291 			/* RTA_ENCAP_TYPE length checked in
5292 			 * lwtunnel_valid_encap_type_attr
5293 			 */
5294 			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5295 			if (nla)
5296 				r_cfg.fc_encap_type = nla_get_u16(nla);
5297 		}
5298 
5299 		r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5300 		rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5301 		if (IS_ERR(rt)) {
5302 			err = PTR_ERR(rt);
5303 			rt = NULL;
5304 			goto cleanup;
5305 		}
5306 		if (!rt6_qualify_for_ecmp(rt)) {
5307 			err = -EINVAL;
5308 			NL_SET_ERR_MSG(extack,
5309 				       "Device only routes can not be added for IPv6 using the multipath API.");
5310 			fib6_info_release(rt);
5311 			goto cleanup;
5312 		}
5313 
5314 		rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5315 
5316 		err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5317 					    rt, &r_cfg);
5318 		if (err) {
5319 			fib6_info_release(rt);
5320 			goto cleanup;
5321 		}
5322 
5323 		rtnh = rtnh_next(rtnh, &remaining);
5324 	}
5325 
5326 	if (list_empty(&rt6_nh_list)) {
5327 		NL_SET_ERR_MSG(extack,
5328 			       "Invalid nexthop configuration - no valid nexthops");
5329 		return -EINVAL;
5330 	}
5331 
5332 	/* for add and replace send one notification with all nexthops.
5333 	 * Skip the notification in fib6_add_rt2node and send one with
5334 	 * the full route when done
5335 	 */
5336 	info->skip_notify = 1;
5337 
5338 	/* For add and replace, send one notification with all nexthops. For
5339 	 * append, send one notification with all appended nexthops.
5340 	 */
5341 	info->skip_notify_kernel = 1;
5342 
5343 	err_nh = NULL;
5344 	list_for_each_entry(nh, &rt6_nh_list, next) {
5345 		err = __ip6_ins_rt(nh->fib6_info, info, extack);
5346 
5347 		if (err) {
5348 			if (replace && nhn)
5349 				NL_SET_ERR_MSG_MOD(extack,
5350 						   "multipath route replace failed (check consistency of installed routes)");
5351 			err_nh = nh;
5352 			goto add_errout;
5353 		}
5354 		/* save reference to last route successfully inserted */
5355 		rt_last = nh->fib6_info;
5356 
5357 		/* save reference to first route for notification */
5358 		if (!rt_notif)
5359 			rt_notif = nh->fib6_info;
5360 
5361 		/* Because each route is added like a single route we remove
5362 		 * these flags after the first nexthop: if there is a collision,
5363 		 * we have already failed to add the first nexthop:
5364 		 * fib6_add_rt2node() has rejected it; when replacing, old
5365 		 * nexthops have been replaced by first new, the rest should
5366 		 * be added to it.
5367 		 */
5368 		if (cfg->fc_nlinfo.nlh) {
5369 			cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5370 							     NLM_F_REPLACE);
5371 			cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5372 		}
5373 		nhn++;
5374 	}
5375 
5376 	/* An in-kernel notification should only be sent in case the new
5377 	 * multipath route is added as the first route in the node, or if
5378 	 * it was appended to it. We pass 'rt_notif' since it is the first
5379 	 * sibling and might allow us to skip some checks in the replace case.
5380 	 */
5381 	if (ip6_route_mpath_should_notify(rt_notif)) {
5382 		enum fib_event_type fib_event;
5383 
5384 		if (rt_notif->fib6_nsiblings != nhn - 1)
5385 			fib_event = FIB_EVENT_ENTRY_APPEND;
5386 		else
5387 			fib_event = FIB_EVENT_ENTRY_REPLACE;
5388 
5389 		err = call_fib6_multipath_entry_notifiers(info->nl_net,
5390 							  fib_event, rt_notif,
5391 							  nhn - 1, extack);
5392 		if (err) {
5393 			/* Delete all the siblings that were just added */
5394 			err_nh = NULL;
5395 			goto add_errout;
5396 		}
5397 	}
5398 
5399 	/* success ... tell user about new route */
5400 	ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5401 	goto cleanup;
5402 
5403 add_errout:
5404 	/* send notification for routes that were added so that
5405 	 * the delete notifications sent by ip6_route_del are
5406 	 * coherent
5407 	 */
5408 	if (rt_notif)
5409 		ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5410 
5411 	/* Delete routes that were already added */
5412 	list_for_each_entry(nh, &rt6_nh_list, next) {
5413 		if (err_nh == nh)
5414 			break;
5415 		ip6_route_del(&nh->r_cfg, extack);
5416 	}
5417 
5418 cleanup:
5419 	list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5420 		fib6_info_release(nh->fib6_info);
5421 		list_del(&nh->next);
5422 		kfree(nh);
5423 	}
5424 
5425 	return err;
5426 }
5427 
5428 static int ip6_route_multipath_del(struct fib6_config *cfg,
5429 				   struct netlink_ext_ack *extack)
5430 {
5431 	struct fib6_config r_cfg;
5432 	struct rtnexthop *rtnh;
5433 	int last_err = 0;
5434 	int remaining;
5435 	int attrlen;
5436 	int err;
5437 
5438 	remaining = cfg->fc_mp_len;
5439 	rtnh = (struct rtnexthop *)cfg->fc_mp;
5440 
5441 	/* Parse a Multipath Entry */
5442 	while (rtnh_ok(rtnh, remaining)) {
5443 		memcpy(&r_cfg, cfg, sizeof(*cfg));
5444 		if (rtnh->rtnh_ifindex)
5445 			r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5446 
5447 		attrlen = rtnh_attrlen(rtnh);
5448 		if (attrlen > 0) {
5449 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5450 
5451 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5452 			if (nla) {
5453 				err = fib6_gw_from_attr(&r_cfg.fc_gateway, nla,
5454 							extack);
5455 				if (err) {
5456 					last_err = err;
5457 					goto next_rtnh;
5458 				}
5459 
5460 				r_cfg.fc_flags |= RTF_GATEWAY;
5461 			}
5462 		}
5463 		err = ip6_route_del(&r_cfg, extack);
5464 		if (err)
5465 			last_err = err;
5466 
5467 next_rtnh:
5468 		rtnh = rtnh_next(rtnh, &remaining);
5469 	}
5470 
5471 	return last_err;
5472 }
5473 
5474 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5475 			      struct netlink_ext_ack *extack)
5476 {
5477 	struct fib6_config cfg;
5478 	int err;
5479 
5480 	err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5481 	if (err < 0)
5482 		return err;
5483 
5484 	if (cfg.fc_nh_id &&
5485 	    !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5486 		NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5487 		return -EINVAL;
5488 	}
5489 
5490 	if (cfg.fc_mp)
5491 		return ip6_route_multipath_del(&cfg, extack);
5492 	else {
5493 		cfg.fc_delete_all_nh = 1;
5494 		return ip6_route_del(&cfg, extack);
5495 	}
5496 }
5497 
5498 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5499 			      struct netlink_ext_ack *extack)
5500 {
5501 	struct fib6_config cfg;
5502 	int err;
5503 
5504 	err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5505 	if (err < 0)
5506 		return err;
5507 
5508 	if (cfg.fc_metric == 0)
5509 		cfg.fc_metric = IP6_RT_PRIO_USER;
5510 
5511 	if (cfg.fc_mp)
5512 		return ip6_route_multipath_add(&cfg, extack);
5513 	else
5514 		return ip6_route_add(&cfg, GFP_KERNEL, extack);
5515 }
5516 
5517 /* add the overhead of this fib6_nh to nexthop_len */
5518 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5519 {
5520 	int *nexthop_len = arg;
5521 
5522 	*nexthop_len += nla_total_size(0)	 /* RTA_MULTIPATH */
5523 		     + NLA_ALIGN(sizeof(struct rtnexthop))
5524 		     + nla_total_size(16); /* RTA_GATEWAY */
5525 
5526 	if (nh->fib_nh_lws) {
5527 		/* RTA_ENCAP_TYPE */
5528 		*nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5529 		/* RTA_ENCAP */
5530 		*nexthop_len += nla_total_size(2);
5531 	}
5532 
5533 	return 0;
5534 }
5535 
5536 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5537 {
5538 	int nexthop_len;
5539 
5540 	if (f6i->nh) {
5541 		nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5542 		nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5543 					 &nexthop_len);
5544 	} else {
5545 		struct fib6_info *sibling, *next_sibling;
5546 		struct fib6_nh *nh = f6i->fib6_nh;
5547 
5548 		nexthop_len = 0;
5549 		if (f6i->fib6_nsiblings) {
5550 			rt6_nh_nlmsg_size(nh, &nexthop_len);
5551 
5552 			list_for_each_entry_safe(sibling, next_sibling,
5553 						 &f6i->fib6_siblings, fib6_siblings) {
5554 				rt6_nh_nlmsg_size(sibling->fib6_nh, &nexthop_len);
5555 			}
5556 		}
5557 		nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5558 	}
5559 
5560 	return NLMSG_ALIGN(sizeof(struct rtmsg))
5561 	       + nla_total_size(16) /* RTA_SRC */
5562 	       + nla_total_size(16) /* RTA_DST */
5563 	       + nla_total_size(16) /* RTA_GATEWAY */
5564 	       + nla_total_size(16) /* RTA_PREFSRC */
5565 	       + nla_total_size(4) /* RTA_TABLE */
5566 	       + nla_total_size(4) /* RTA_IIF */
5567 	       + nla_total_size(4) /* RTA_OIF */
5568 	       + nla_total_size(4) /* RTA_PRIORITY */
5569 	       + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5570 	       + nla_total_size(sizeof(struct rta_cacheinfo))
5571 	       + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5572 	       + nla_total_size(1) /* RTA_PREF */
5573 	       + nexthop_len;
5574 }
5575 
5576 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5577 				 unsigned char *flags)
5578 {
5579 	if (nexthop_is_multipath(nh)) {
5580 		struct nlattr *mp;
5581 
5582 		mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5583 		if (!mp)
5584 			goto nla_put_failure;
5585 
5586 		if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5587 			goto nla_put_failure;
5588 
5589 		nla_nest_end(skb, mp);
5590 	} else {
5591 		struct fib6_nh *fib6_nh;
5592 
5593 		fib6_nh = nexthop_fib6_nh(nh);
5594 		if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5595 				     flags, false) < 0)
5596 			goto nla_put_failure;
5597 	}
5598 
5599 	return 0;
5600 
5601 nla_put_failure:
5602 	return -EMSGSIZE;
5603 }
5604 
5605 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5606 			 struct fib6_info *rt, struct dst_entry *dst,
5607 			 struct in6_addr *dest, struct in6_addr *src,
5608 			 int iif, int type, u32 portid, u32 seq,
5609 			 unsigned int flags)
5610 {
5611 	struct rt6_info *rt6 = (struct rt6_info *)dst;
5612 	struct rt6key *rt6_dst, *rt6_src;
5613 	u32 *pmetrics, table, rt6_flags;
5614 	unsigned char nh_flags = 0;
5615 	struct nlmsghdr *nlh;
5616 	struct rtmsg *rtm;
5617 	long expires = 0;
5618 
5619 	nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5620 	if (!nlh)
5621 		return -EMSGSIZE;
5622 
5623 	if (rt6) {
5624 		rt6_dst = &rt6->rt6i_dst;
5625 		rt6_src = &rt6->rt6i_src;
5626 		rt6_flags = rt6->rt6i_flags;
5627 	} else {
5628 		rt6_dst = &rt->fib6_dst;
5629 		rt6_src = &rt->fib6_src;
5630 		rt6_flags = rt->fib6_flags;
5631 	}
5632 
5633 	rtm = nlmsg_data(nlh);
5634 	rtm->rtm_family = AF_INET6;
5635 	rtm->rtm_dst_len = rt6_dst->plen;
5636 	rtm->rtm_src_len = rt6_src->plen;
5637 	rtm->rtm_tos = 0;
5638 	if (rt->fib6_table)
5639 		table = rt->fib6_table->tb6_id;
5640 	else
5641 		table = RT6_TABLE_UNSPEC;
5642 	rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5643 	if (nla_put_u32(skb, RTA_TABLE, table))
5644 		goto nla_put_failure;
5645 
5646 	rtm->rtm_type = rt->fib6_type;
5647 	rtm->rtm_flags = 0;
5648 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5649 	rtm->rtm_protocol = rt->fib6_protocol;
5650 
5651 	if (rt6_flags & RTF_CACHE)
5652 		rtm->rtm_flags |= RTM_F_CLONED;
5653 
5654 	if (dest) {
5655 		if (nla_put_in6_addr(skb, RTA_DST, dest))
5656 			goto nla_put_failure;
5657 		rtm->rtm_dst_len = 128;
5658 	} else if (rtm->rtm_dst_len)
5659 		if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5660 			goto nla_put_failure;
5661 #ifdef CONFIG_IPV6_SUBTREES
5662 	if (src) {
5663 		if (nla_put_in6_addr(skb, RTA_SRC, src))
5664 			goto nla_put_failure;
5665 		rtm->rtm_src_len = 128;
5666 	} else if (rtm->rtm_src_len &&
5667 		   nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5668 		goto nla_put_failure;
5669 #endif
5670 	if (iif) {
5671 #ifdef CONFIG_IPV6_MROUTE
5672 		if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5673 			int err = ip6mr_get_route(net, skb, rtm, portid);
5674 
5675 			if (err == 0)
5676 				return 0;
5677 			if (err < 0)
5678 				goto nla_put_failure;
5679 		} else
5680 #endif
5681 			if (nla_put_u32(skb, RTA_IIF, iif))
5682 				goto nla_put_failure;
5683 	} else if (dest) {
5684 		struct in6_addr saddr_buf;
5685 		if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5686 		    nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5687 			goto nla_put_failure;
5688 	}
5689 
5690 	if (rt->fib6_prefsrc.plen) {
5691 		struct in6_addr saddr_buf;
5692 		saddr_buf = rt->fib6_prefsrc.addr;
5693 		if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5694 			goto nla_put_failure;
5695 	}
5696 
5697 	pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5698 	if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5699 		goto nla_put_failure;
5700 
5701 	if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5702 		goto nla_put_failure;
5703 
5704 	/* For multipath routes, walk the siblings list and add
5705 	 * each as a nexthop within RTA_MULTIPATH.
5706 	 */
5707 	if (rt6) {
5708 		if (rt6_flags & RTF_GATEWAY &&
5709 		    nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5710 			goto nla_put_failure;
5711 
5712 		if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5713 			goto nla_put_failure;
5714 
5715 		if (dst->lwtstate &&
5716 		    lwtunnel_fill_encap(skb, dst->lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
5717 			goto nla_put_failure;
5718 	} else if (rt->fib6_nsiblings) {
5719 		struct fib6_info *sibling, *next_sibling;
5720 		struct nlattr *mp;
5721 
5722 		mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5723 		if (!mp)
5724 			goto nla_put_failure;
5725 
5726 		if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5727 				    rt->fib6_nh->fib_nh_weight, AF_INET6,
5728 				    0) < 0)
5729 			goto nla_put_failure;
5730 
5731 		list_for_each_entry_safe(sibling, next_sibling,
5732 					 &rt->fib6_siblings, fib6_siblings) {
5733 			if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5734 					    sibling->fib6_nh->fib_nh_weight,
5735 					    AF_INET6, 0) < 0)
5736 				goto nla_put_failure;
5737 		}
5738 
5739 		nla_nest_end(skb, mp);
5740 	} else if (rt->nh) {
5741 		if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5742 			goto nla_put_failure;
5743 
5744 		if (nexthop_is_blackhole(rt->nh))
5745 			rtm->rtm_type = RTN_BLACKHOLE;
5746 
5747 		if (READ_ONCE(net->ipv4.sysctl_nexthop_compat_mode) &&
5748 		    rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5749 			goto nla_put_failure;
5750 
5751 		rtm->rtm_flags |= nh_flags;
5752 	} else {
5753 		if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5754 				     &nh_flags, false) < 0)
5755 			goto nla_put_failure;
5756 
5757 		rtm->rtm_flags |= nh_flags;
5758 	}
5759 
5760 	if (rt6_flags & RTF_EXPIRES) {
5761 		expires = dst ? dst->expires : rt->expires;
5762 		expires -= jiffies;
5763 	}
5764 
5765 	if (!dst) {
5766 		if (READ_ONCE(rt->offload))
5767 			rtm->rtm_flags |= RTM_F_OFFLOAD;
5768 		if (READ_ONCE(rt->trap))
5769 			rtm->rtm_flags |= RTM_F_TRAP;
5770 		if (READ_ONCE(rt->offload_failed))
5771 			rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
5772 	}
5773 
5774 	if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5775 		goto nla_put_failure;
5776 
5777 	if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5778 		goto nla_put_failure;
5779 
5780 
5781 	nlmsg_end(skb, nlh);
5782 	return 0;
5783 
5784 nla_put_failure:
5785 	nlmsg_cancel(skb, nlh);
5786 	return -EMSGSIZE;
5787 }
5788 
5789 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5790 {
5791 	const struct net_device *dev = arg;
5792 
5793 	if (nh->fib_nh_dev == dev)
5794 		return 1;
5795 
5796 	return 0;
5797 }
5798 
5799 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5800 			       const struct net_device *dev)
5801 {
5802 	if (f6i->nh) {
5803 		struct net_device *_dev = (struct net_device *)dev;
5804 
5805 		return !!nexthop_for_each_fib6_nh(f6i->nh,
5806 						  fib6_info_nh_uses_dev,
5807 						  _dev);
5808 	}
5809 
5810 	if (f6i->fib6_nh->fib_nh_dev == dev)
5811 		return true;
5812 
5813 	if (f6i->fib6_nsiblings) {
5814 		struct fib6_info *sibling, *next_sibling;
5815 
5816 		list_for_each_entry_safe(sibling, next_sibling,
5817 					 &f6i->fib6_siblings, fib6_siblings) {
5818 			if (sibling->fib6_nh->fib_nh_dev == dev)
5819 				return true;
5820 		}
5821 	}
5822 
5823 	return false;
5824 }
5825 
5826 struct fib6_nh_exception_dump_walker {
5827 	struct rt6_rtnl_dump_arg *dump;
5828 	struct fib6_info *rt;
5829 	unsigned int flags;
5830 	unsigned int skip;
5831 	unsigned int count;
5832 };
5833 
5834 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5835 {
5836 	struct fib6_nh_exception_dump_walker *w = arg;
5837 	struct rt6_rtnl_dump_arg *dump = w->dump;
5838 	struct rt6_exception_bucket *bucket;
5839 	struct rt6_exception *rt6_ex;
5840 	int i, err;
5841 
5842 	bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5843 	if (!bucket)
5844 		return 0;
5845 
5846 	for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5847 		hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5848 			if (w->skip) {
5849 				w->skip--;
5850 				continue;
5851 			}
5852 
5853 			/* Expiration of entries doesn't bump sernum, insertion
5854 			 * does. Removal is triggered by insertion, so we can
5855 			 * rely on the fact that if entries change between two
5856 			 * partial dumps, this node is scanned again completely,
5857 			 * see rt6_insert_exception() and fib6_dump_table().
5858 			 *
5859 			 * Count expired entries we go through as handled
5860 			 * entries that we'll skip next time, in case of partial
5861 			 * node dump. Otherwise, if entries expire meanwhile,
5862 			 * we'll skip the wrong amount.
5863 			 */
5864 			if (rt6_check_expired(rt6_ex->rt6i)) {
5865 				w->count++;
5866 				continue;
5867 			}
5868 
5869 			err = rt6_fill_node(dump->net, dump->skb, w->rt,
5870 					    &rt6_ex->rt6i->dst, NULL, NULL, 0,
5871 					    RTM_NEWROUTE,
5872 					    NETLINK_CB(dump->cb->skb).portid,
5873 					    dump->cb->nlh->nlmsg_seq, w->flags);
5874 			if (err)
5875 				return err;
5876 
5877 			w->count++;
5878 		}
5879 		bucket++;
5880 	}
5881 
5882 	return 0;
5883 }
5884 
5885 /* Return -1 if done with node, number of handled routes on partial dump */
5886 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5887 {
5888 	struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5889 	struct fib_dump_filter *filter = &arg->filter;
5890 	unsigned int flags = NLM_F_MULTI;
5891 	struct net *net = arg->net;
5892 	int count = 0;
5893 
5894 	if (rt == net->ipv6.fib6_null_entry)
5895 		return -1;
5896 
5897 	if ((filter->flags & RTM_F_PREFIX) &&
5898 	    !(rt->fib6_flags & RTF_PREFIX_RT)) {
5899 		/* success since this is not a prefix route */
5900 		return -1;
5901 	}
5902 	if (filter->filter_set &&
5903 	    ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5904 	     (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5905 	     (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5906 		return -1;
5907 	}
5908 
5909 	if (filter->filter_set ||
5910 	    !filter->dump_routes || !filter->dump_exceptions) {
5911 		flags |= NLM_F_DUMP_FILTERED;
5912 	}
5913 
5914 	if (filter->dump_routes) {
5915 		if (skip) {
5916 			skip--;
5917 		} else {
5918 			if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5919 					  0, RTM_NEWROUTE,
5920 					  NETLINK_CB(arg->cb->skb).portid,
5921 					  arg->cb->nlh->nlmsg_seq, flags)) {
5922 				return 0;
5923 			}
5924 			count++;
5925 		}
5926 	}
5927 
5928 	if (filter->dump_exceptions) {
5929 		struct fib6_nh_exception_dump_walker w = { .dump = arg,
5930 							   .rt = rt,
5931 							   .flags = flags,
5932 							   .skip = skip,
5933 							   .count = 0 };
5934 		int err;
5935 
5936 		rcu_read_lock();
5937 		if (rt->nh) {
5938 			err = nexthop_for_each_fib6_nh(rt->nh,
5939 						       rt6_nh_dump_exceptions,
5940 						       &w);
5941 		} else {
5942 			err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5943 		}
5944 		rcu_read_unlock();
5945 
5946 		if (err)
5947 			return count + w.count;
5948 	}
5949 
5950 	return -1;
5951 }
5952 
5953 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5954 					const struct nlmsghdr *nlh,
5955 					struct nlattr **tb,
5956 					struct netlink_ext_ack *extack)
5957 {
5958 	struct rtmsg *rtm;
5959 	int i, err;
5960 
5961 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5962 		NL_SET_ERR_MSG_MOD(extack,
5963 				   "Invalid header for get route request");
5964 		return -EINVAL;
5965 	}
5966 
5967 	if (!netlink_strict_get_check(skb))
5968 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5969 					      rtm_ipv6_policy, extack);
5970 
5971 	rtm = nlmsg_data(nlh);
5972 	if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5973 	    (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5974 	    rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5975 	    rtm->rtm_type) {
5976 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5977 		return -EINVAL;
5978 	}
5979 	if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5980 		NL_SET_ERR_MSG_MOD(extack,
5981 				   "Invalid flags for get route request");
5982 		return -EINVAL;
5983 	}
5984 
5985 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5986 					    rtm_ipv6_policy, extack);
5987 	if (err)
5988 		return err;
5989 
5990 	if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5991 	    (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5992 		NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5993 		return -EINVAL;
5994 	}
5995 
5996 	for (i = 0; i <= RTA_MAX; i++) {
5997 		if (!tb[i])
5998 			continue;
5999 
6000 		switch (i) {
6001 		case RTA_SRC:
6002 		case RTA_DST:
6003 		case RTA_IIF:
6004 		case RTA_OIF:
6005 		case RTA_MARK:
6006 		case RTA_UID:
6007 		case RTA_SPORT:
6008 		case RTA_DPORT:
6009 		case RTA_IP_PROTO:
6010 			break;
6011 		default:
6012 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
6013 			return -EINVAL;
6014 		}
6015 	}
6016 
6017 	return 0;
6018 }
6019 
6020 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
6021 			      struct netlink_ext_ack *extack)
6022 {
6023 	struct net *net = sock_net(in_skb->sk);
6024 	struct nlattr *tb[RTA_MAX+1];
6025 	int err, iif = 0, oif = 0;
6026 	struct fib6_info *from;
6027 	struct dst_entry *dst;
6028 	struct rt6_info *rt;
6029 	struct sk_buff *skb;
6030 	struct rtmsg *rtm;
6031 	struct flowi6 fl6 = {};
6032 	bool fibmatch;
6033 
6034 	err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
6035 	if (err < 0)
6036 		goto errout;
6037 
6038 	err = -EINVAL;
6039 	rtm = nlmsg_data(nlh);
6040 	fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
6041 	fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
6042 
6043 	if (tb[RTA_SRC]) {
6044 		if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
6045 			goto errout;
6046 
6047 		fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
6048 	}
6049 
6050 	if (tb[RTA_DST]) {
6051 		if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
6052 			goto errout;
6053 
6054 		fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
6055 	}
6056 
6057 	if (tb[RTA_IIF])
6058 		iif = nla_get_u32(tb[RTA_IIF]);
6059 
6060 	if (tb[RTA_OIF])
6061 		oif = nla_get_u32(tb[RTA_OIF]);
6062 
6063 	if (tb[RTA_MARK])
6064 		fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
6065 
6066 	if (tb[RTA_UID])
6067 		fl6.flowi6_uid = make_kuid(current_user_ns(),
6068 					   nla_get_u32(tb[RTA_UID]));
6069 	else
6070 		fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
6071 
6072 	if (tb[RTA_SPORT])
6073 		fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
6074 
6075 	if (tb[RTA_DPORT])
6076 		fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
6077 
6078 	if (tb[RTA_IP_PROTO]) {
6079 		err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
6080 						  &fl6.flowi6_proto, AF_INET6,
6081 						  extack);
6082 		if (err)
6083 			goto errout;
6084 	}
6085 
6086 	if (iif) {
6087 		struct net_device *dev;
6088 		int flags = 0;
6089 
6090 		rcu_read_lock();
6091 
6092 		dev = dev_get_by_index_rcu(net, iif);
6093 		if (!dev) {
6094 			rcu_read_unlock();
6095 			err = -ENODEV;
6096 			goto errout;
6097 		}
6098 
6099 		fl6.flowi6_iif = iif;
6100 
6101 		if (!ipv6_addr_any(&fl6.saddr))
6102 			flags |= RT6_LOOKUP_F_HAS_SADDR;
6103 
6104 		dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
6105 
6106 		rcu_read_unlock();
6107 	} else {
6108 		fl6.flowi6_oif = oif;
6109 
6110 		dst = ip6_route_output(net, NULL, &fl6);
6111 	}
6112 
6113 
6114 	rt = container_of(dst, struct rt6_info, dst);
6115 	if (rt->dst.error) {
6116 		err = rt->dst.error;
6117 		ip6_rt_put(rt);
6118 		goto errout;
6119 	}
6120 
6121 	if (rt == net->ipv6.ip6_null_entry) {
6122 		err = rt->dst.error;
6123 		ip6_rt_put(rt);
6124 		goto errout;
6125 	}
6126 
6127 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
6128 	if (!skb) {
6129 		ip6_rt_put(rt);
6130 		err = -ENOBUFS;
6131 		goto errout;
6132 	}
6133 
6134 	skb_dst_set(skb, &rt->dst);
6135 
6136 	rcu_read_lock();
6137 	from = rcu_dereference(rt->from);
6138 	if (from) {
6139 		if (fibmatch)
6140 			err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
6141 					    iif, RTM_NEWROUTE,
6142 					    NETLINK_CB(in_skb).portid,
6143 					    nlh->nlmsg_seq, 0);
6144 		else
6145 			err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
6146 					    &fl6.saddr, iif, RTM_NEWROUTE,
6147 					    NETLINK_CB(in_skb).portid,
6148 					    nlh->nlmsg_seq, 0);
6149 	} else {
6150 		err = -ENETUNREACH;
6151 	}
6152 	rcu_read_unlock();
6153 
6154 	if (err < 0) {
6155 		kfree_skb(skb);
6156 		goto errout;
6157 	}
6158 
6159 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
6160 errout:
6161 	return err;
6162 }
6163 
6164 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6165 		     unsigned int nlm_flags)
6166 {
6167 	struct sk_buff *skb;
6168 	struct net *net = info->nl_net;
6169 	u32 seq;
6170 	int err;
6171 
6172 	err = -ENOBUFS;
6173 	seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6174 
6175 	skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6176 	if (!skb)
6177 		goto errout;
6178 
6179 	err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6180 			    event, info->portid, seq, nlm_flags);
6181 	if (err < 0) {
6182 		/* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6183 		WARN_ON(err == -EMSGSIZE);
6184 		kfree_skb(skb);
6185 		goto errout;
6186 	}
6187 	rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6188 		    info->nlh, gfp_any());
6189 	return;
6190 errout:
6191 	if (err < 0)
6192 		rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6193 }
6194 
6195 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6196 		    struct nl_info *info)
6197 {
6198 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6199 	struct sk_buff *skb;
6200 	int err = -ENOBUFS;
6201 
6202 	skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6203 	if (!skb)
6204 		goto errout;
6205 
6206 	err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6207 			    RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6208 	if (err < 0) {
6209 		/* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6210 		WARN_ON(err == -EMSGSIZE);
6211 		kfree_skb(skb);
6212 		goto errout;
6213 	}
6214 	rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6215 		    info->nlh, gfp_any());
6216 	return;
6217 errout:
6218 	if (err < 0)
6219 		rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6220 }
6221 
6222 void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
6223 			    bool offload, bool trap, bool offload_failed)
6224 {
6225 	struct sk_buff *skb;
6226 	int err;
6227 
6228 	if (READ_ONCE(f6i->offload) == offload &&
6229 	    READ_ONCE(f6i->trap) == trap &&
6230 	    READ_ONCE(f6i->offload_failed) == offload_failed)
6231 		return;
6232 
6233 	WRITE_ONCE(f6i->offload, offload);
6234 	WRITE_ONCE(f6i->trap, trap);
6235 
6236 	/* 2 means send notifications only if offload_failed was changed. */
6237 	if (net->ipv6.sysctl.fib_notify_on_flag_change == 2 &&
6238 	    READ_ONCE(f6i->offload_failed) == offload_failed)
6239 		return;
6240 
6241 	WRITE_ONCE(f6i->offload_failed, offload_failed);
6242 
6243 	if (!rcu_access_pointer(f6i->fib6_node))
6244 		/* The route was removed from the tree, do not send
6245 		 * notification.
6246 		 */
6247 		return;
6248 
6249 	if (!net->ipv6.sysctl.fib_notify_on_flag_change)
6250 		return;
6251 
6252 	skb = nlmsg_new(rt6_nlmsg_size(f6i), GFP_KERNEL);
6253 	if (!skb) {
6254 		err = -ENOBUFS;
6255 		goto errout;
6256 	}
6257 
6258 	err = rt6_fill_node(net, skb, f6i, NULL, NULL, NULL, 0, RTM_NEWROUTE, 0,
6259 			    0, 0);
6260 	if (err < 0) {
6261 		/* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6262 		WARN_ON(err == -EMSGSIZE);
6263 		kfree_skb(skb);
6264 		goto errout;
6265 	}
6266 
6267 	rtnl_notify(skb, net, 0, RTNLGRP_IPV6_ROUTE, NULL, GFP_KERNEL);
6268 	return;
6269 
6270 errout:
6271 	rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6272 }
6273 EXPORT_SYMBOL(fib6_info_hw_flags_set);
6274 
6275 static int ip6_route_dev_notify(struct notifier_block *this,
6276 				unsigned long event, void *ptr)
6277 {
6278 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6279 	struct net *net = dev_net(dev);
6280 
6281 	if (!(dev->flags & IFF_LOOPBACK))
6282 		return NOTIFY_OK;
6283 
6284 	if (event == NETDEV_REGISTER) {
6285 		net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6286 		net->ipv6.ip6_null_entry->dst.dev = dev;
6287 		net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6288 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6289 		net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6290 		net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6291 		net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6292 		net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6293 #endif
6294 	 } else if (event == NETDEV_UNREGISTER &&
6295 		    dev->reg_state != NETREG_UNREGISTERED) {
6296 		/* NETDEV_UNREGISTER could be fired for multiple times by
6297 		 * netdev_wait_allrefs(). Make sure we only call this once.
6298 		 */
6299 		in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6300 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6301 		in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6302 		in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6303 #endif
6304 	}
6305 
6306 	return NOTIFY_OK;
6307 }
6308 
6309 /*
6310  *	/proc
6311  */
6312 
6313 #ifdef CONFIG_PROC_FS
6314 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6315 {
6316 	struct net *net = (struct net *)seq->private;
6317 	seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6318 		   net->ipv6.rt6_stats->fib_nodes,
6319 		   net->ipv6.rt6_stats->fib_route_nodes,
6320 		   atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6321 		   net->ipv6.rt6_stats->fib_rt_entries,
6322 		   net->ipv6.rt6_stats->fib_rt_cache,
6323 		   dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6324 		   net->ipv6.rt6_stats->fib_discarded_routes);
6325 
6326 	return 0;
6327 }
6328 #endif	/* CONFIG_PROC_FS */
6329 
6330 #ifdef CONFIG_SYSCTL
6331 
6332 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6333 			      void *buffer, size_t *lenp, loff_t *ppos)
6334 {
6335 	struct net *net;
6336 	int delay;
6337 	int ret;
6338 	if (!write)
6339 		return -EINVAL;
6340 
6341 	net = (struct net *)ctl->extra1;
6342 	delay = net->ipv6.sysctl.flush_delay;
6343 	ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6344 	if (ret)
6345 		return ret;
6346 
6347 	fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6348 	return 0;
6349 }
6350 
6351 static struct ctl_table ipv6_route_table_template[] = {
6352 	{
6353 		.procname	=	"max_size",
6354 		.data		=	&init_net.ipv6.sysctl.ip6_rt_max_size,
6355 		.maxlen		=	sizeof(int),
6356 		.mode		=	0644,
6357 		.proc_handler	=	proc_dointvec,
6358 	},
6359 	{
6360 		.procname	=	"gc_thresh",
6361 		.data		=	&ip6_dst_ops_template.gc_thresh,
6362 		.maxlen		=	sizeof(int),
6363 		.mode		=	0644,
6364 		.proc_handler	=	proc_dointvec,
6365 	},
6366 	{
6367 		.procname	=	"flush",
6368 		.data		=	&init_net.ipv6.sysctl.flush_delay,
6369 		.maxlen		=	sizeof(int),
6370 		.mode		=	0200,
6371 		.proc_handler	=	ipv6_sysctl_rtcache_flush
6372 	},
6373 	{
6374 		.procname	=	"gc_min_interval",
6375 		.data		=	&init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6376 		.maxlen		=	sizeof(int),
6377 		.mode		=	0644,
6378 		.proc_handler	=	proc_dointvec_jiffies,
6379 	},
6380 	{
6381 		.procname	=	"gc_timeout",
6382 		.data		=	&init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6383 		.maxlen		=	sizeof(int),
6384 		.mode		=	0644,
6385 		.proc_handler	=	proc_dointvec_jiffies,
6386 	},
6387 	{
6388 		.procname	=	"gc_interval",
6389 		.data		=	&init_net.ipv6.sysctl.ip6_rt_gc_interval,
6390 		.maxlen		=	sizeof(int),
6391 		.mode		=	0644,
6392 		.proc_handler	=	proc_dointvec_jiffies,
6393 	},
6394 	{
6395 		.procname	=	"gc_elasticity",
6396 		.data		=	&init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6397 		.maxlen		=	sizeof(int),
6398 		.mode		=	0644,
6399 		.proc_handler	=	proc_dointvec,
6400 	},
6401 	{
6402 		.procname	=	"mtu_expires",
6403 		.data		=	&init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6404 		.maxlen		=	sizeof(int),
6405 		.mode		=	0644,
6406 		.proc_handler	=	proc_dointvec_jiffies,
6407 	},
6408 	{
6409 		.procname	=	"min_adv_mss",
6410 		.data		=	&init_net.ipv6.sysctl.ip6_rt_min_advmss,
6411 		.maxlen		=	sizeof(int),
6412 		.mode		=	0644,
6413 		.proc_handler	=	proc_dointvec,
6414 	},
6415 	{
6416 		.procname	=	"gc_min_interval_ms",
6417 		.data		=	&init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6418 		.maxlen		=	sizeof(int),
6419 		.mode		=	0644,
6420 		.proc_handler	=	proc_dointvec_ms_jiffies,
6421 	},
6422 	{
6423 		.procname	=	"skip_notify_on_dev_down",
6424 		.data		=	&init_net.ipv6.sysctl.skip_notify_on_dev_down,
6425 		.maxlen		=	sizeof(u8),
6426 		.mode		=	0644,
6427 		.proc_handler	=	proc_dou8vec_minmax,
6428 		.extra1		=	SYSCTL_ZERO,
6429 		.extra2		=	SYSCTL_ONE,
6430 	},
6431 	{ }
6432 };
6433 
6434 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6435 {
6436 	struct ctl_table *table;
6437 
6438 	table = kmemdup(ipv6_route_table_template,
6439 			sizeof(ipv6_route_table_template),
6440 			GFP_KERNEL);
6441 
6442 	if (table) {
6443 		table[0].data = &net->ipv6.sysctl.ip6_rt_max_size;
6444 		table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6445 		table[2].data = &net->ipv6.sysctl.flush_delay;
6446 		table[2].extra1 = net;
6447 		table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6448 		table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6449 		table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6450 		table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6451 		table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6452 		table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6453 		table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6454 		table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6455 
6456 		/* Don't export sysctls to unprivileged users */
6457 		if (net->user_ns != &init_user_ns)
6458 			table[1].procname = NULL;
6459 	}
6460 
6461 	return table;
6462 }
6463 
6464 size_t ipv6_route_sysctl_table_size(struct net *net)
6465 {
6466 	/* Don't export sysctls to unprivileged users */
6467 	if (net->user_ns != &init_user_ns)
6468 		return 1;
6469 
6470 	return ARRAY_SIZE(ipv6_route_table_template);
6471 }
6472 #endif
6473 
6474 static int __net_init ip6_route_net_init(struct net *net)
6475 {
6476 	int ret = -ENOMEM;
6477 
6478 	memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6479 	       sizeof(net->ipv6.ip6_dst_ops));
6480 
6481 	if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6482 		goto out_ip6_dst_ops;
6483 
6484 	net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6485 	if (!net->ipv6.fib6_null_entry)
6486 		goto out_ip6_dst_entries;
6487 	memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6488 	       sizeof(*net->ipv6.fib6_null_entry));
6489 
6490 	net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6491 					   sizeof(*net->ipv6.ip6_null_entry),
6492 					   GFP_KERNEL);
6493 	if (!net->ipv6.ip6_null_entry)
6494 		goto out_fib6_null_entry;
6495 	net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6496 	dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6497 			 ip6_template_metrics, true);
6498 	INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->dst.rt_uncached);
6499 
6500 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6501 	net->ipv6.fib6_has_custom_rules = false;
6502 	net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6503 					       sizeof(*net->ipv6.ip6_prohibit_entry),
6504 					       GFP_KERNEL);
6505 	if (!net->ipv6.ip6_prohibit_entry)
6506 		goto out_ip6_null_entry;
6507 	net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6508 	dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6509 			 ip6_template_metrics, true);
6510 	INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->dst.rt_uncached);
6511 
6512 	net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6513 					       sizeof(*net->ipv6.ip6_blk_hole_entry),
6514 					       GFP_KERNEL);
6515 	if (!net->ipv6.ip6_blk_hole_entry)
6516 		goto out_ip6_prohibit_entry;
6517 	net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6518 	dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6519 			 ip6_template_metrics, true);
6520 	INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->dst.rt_uncached);
6521 #ifdef CONFIG_IPV6_SUBTREES
6522 	net->ipv6.fib6_routes_require_src = 0;
6523 #endif
6524 #endif
6525 
6526 	net->ipv6.sysctl.flush_delay = 0;
6527 	net->ipv6.sysctl.ip6_rt_max_size = INT_MAX;
6528 	net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6529 	net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6530 	net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6531 	net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6532 	net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6533 	net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6534 	net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6535 
6536 	atomic_set(&net->ipv6.ip6_rt_gc_expire, 30*HZ);
6537 
6538 	ret = 0;
6539 out:
6540 	return ret;
6541 
6542 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6543 out_ip6_prohibit_entry:
6544 	kfree(net->ipv6.ip6_prohibit_entry);
6545 out_ip6_null_entry:
6546 	kfree(net->ipv6.ip6_null_entry);
6547 #endif
6548 out_fib6_null_entry:
6549 	kfree(net->ipv6.fib6_null_entry);
6550 out_ip6_dst_entries:
6551 	dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6552 out_ip6_dst_ops:
6553 	goto out;
6554 }
6555 
6556 static void __net_exit ip6_route_net_exit(struct net *net)
6557 {
6558 	kfree(net->ipv6.fib6_null_entry);
6559 	kfree(net->ipv6.ip6_null_entry);
6560 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6561 	kfree(net->ipv6.ip6_prohibit_entry);
6562 	kfree(net->ipv6.ip6_blk_hole_entry);
6563 #endif
6564 	dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6565 }
6566 
6567 static int __net_init ip6_route_net_init_late(struct net *net)
6568 {
6569 #ifdef CONFIG_PROC_FS
6570 	if (!proc_create_net("ipv6_route", 0, net->proc_net,
6571 			     &ipv6_route_seq_ops,
6572 			     sizeof(struct ipv6_route_iter)))
6573 		return -ENOMEM;
6574 
6575 	if (!proc_create_net_single("rt6_stats", 0444, net->proc_net,
6576 				    rt6_stats_seq_show, NULL)) {
6577 		remove_proc_entry("ipv6_route", net->proc_net);
6578 		return -ENOMEM;
6579 	}
6580 #endif
6581 	return 0;
6582 }
6583 
6584 static void __net_exit ip6_route_net_exit_late(struct net *net)
6585 {
6586 #ifdef CONFIG_PROC_FS
6587 	remove_proc_entry("ipv6_route", net->proc_net);
6588 	remove_proc_entry("rt6_stats", net->proc_net);
6589 #endif
6590 }
6591 
6592 static struct pernet_operations ip6_route_net_ops = {
6593 	.init = ip6_route_net_init,
6594 	.exit = ip6_route_net_exit,
6595 };
6596 
6597 static int __net_init ipv6_inetpeer_init(struct net *net)
6598 {
6599 	struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6600 
6601 	if (!bp)
6602 		return -ENOMEM;
6603 	inet_peer_base_init(bp);
6604 	net->ipv6.peers = bp;
6605 	return 0;
6606 }
6607 
6608 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6609 {
6610 	struct inet_peer_base *bp = net->ipv6.peers;
6611 
6612 	net->ipv6.peers = NULL;
6613 	inetpeer_invalidate_tree(bp);
6614 	kfree(bp);
6615 }
6616 
6617 static struct pernet_operations ipv6_inetpeer_ops = {
6618 	.init	=	ipv6_inetpeer_init,
6619 	.exit	=	ipv6_inetpeer_exit,
6620 };
6621 
6622 static struct pernet_operations ip6_route_net_late_ops = {
6623 	.init = ip6_route_net_init_late,
6624 	.exit = ip6_route_net_exit_late,
6625 };
6626 
6627 static struct notifier_block ip6_route_dev_notifier = {
6628 	.notifier_call = ip6_route_dev_notify,
6629 	.priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6630 };
6631 
6632 void __init ip6_route_init_special_entries(void)
6633 {
6634 	/* Registering of the loopback is done before this portion of code,
6635 	 * the loopback reference in rt6_info will not be taken, do it
6636 	 * manually for init_net */
6637 	init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6638 	init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6639 	init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6640   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6641 	init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6642 	init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6643 	init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6644 	init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6645   #endif
6646 }
6647 
6648 #if IS_BUILTIN(CONFIG_IPV6)
6649 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6650 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6651 
6652 BTF_ID_LIST(btf_fib6_info_id)
6653 BTF_ID(struct, fib6_info)
6654 
6655 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6656 	.seq_ops		= &ipv6_route_seq_ops,
6657 	.init_seq_private	= bpf_iter_init_seq_net,
6658 	.fini_seq_private	= bpf_iter_fini_seq_net,
6659 	.seq_priv_size		= sizeof(struct ipv6_route_iter),
6660 };
6661 
6662 static struct bpf_iter_reg ipv6_route_reg_info = {
6663 	.target			= "ipv6_route",
6664 	.ctx_arg_info_size	= 1,
6665 	.ctx_arg_info		= {
6666 		{ offsetof(struct bpf_iter__ipv6_route, rt),
6667 		  PTR_TO_BTF_ID_OR_NULL },
6668 	},
6669 	.seq_info		= &ipv6_route_seq_info,
6670 };
6671 
6672 static int __init bpf_iter_register(void)
6673 {
6674 	ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6675 	return bpf_iter_reg_target(&ipv6_route_reg_info);
6676 }
6677 
6678 static void bpf_iter_unregister(void)
6679 {
6680 	bpf_iter_unreg_target(&ipv6_route_reg_info);
6681 }
6682 #endif
6683 #endif
6684 
6685 int __init ip6_route_init(void)
6686 {
6687 	int ret;
6688 	int cpu;
6689 
6690 	ret = -ENOMEM;
6691 	ip6_dst_ops_template.kmem_cachep =
6692 		kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6693 				  SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
6694 	if (!ip6_dst_ops_template.kmem_cachep)
6695 		goto out;
6696 
6697 	ret = dst_entries_init(&ip6_dst_blackhole_ops);
6698 	if (ret)
6699 		goto out_kmem_cache;
6700 
6701 	ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6702 	if (ret)
6703 		goto out_dst_entries;
6704 
6705 	ret = register_pernet_subsys(&ip6_route_net_ops);
6706 	if (ret)
6707 		goto out_register_inetpeer;
6708 
6709 	ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6710 
6711 	ret = fib6_init();
6712 	if (ret)
6713 		goto out_register_subsys;
6714 
6715 	ret = xfrm6_init();
6716 	if (ret)
6717 		goto out_fib6_init;
6718 
6719 	ret = fib6_rules_init();
6720 	if (ret)
6721 		goto xfrm6_init;
6722 
6723 	ret = register_pernet_subsys(&ip6_route_net_late_ops);
6724 	if (ret)
6725 		goto fib6_rules_init;
6726 
6727 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6728 				   inet6_rtm_newroute, NULL, 0);
6729 	if (ret < 0)
6730 		goto out_register_late_subsys;
6731 
6732 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6733 				   inet6_rtm_delroute, NULL, 0);
6734 	if (ret < 0)
6735 		goto out_register_late_subsys;
6736 
6737 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6738 				   inet6_rtm_getroute, NULL,
6739 				   RTNL_FLAG_DOIT_UNLOCKED);
6740 	if (ret < 0)
6741 		goto out_register_late_subsys;
6742 
6743 	ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6744 	if (ret)
6745 		goto out_register_late_subsys;
6746 
6747 #if IS_BUILTIN(CONFIG_IPV6)
6748 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6749 	ret = bpf_iter_register();
6750 	if (ret)
6751 		goto out_register_late_subsys;
6752 #endif
6753 #endif
6754 
6755 	for_each_possible_cpu(cpu) {
6756 		struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6757 
6758 		INIT_LIST_HEAD(&ul->head);
6759 		INIT_LIST_HEAD(&ul->quarantine);
6760 		spin_lock_init(&ul->lock);
6761 	}
6762 
6763 out:
6764 	return ret;
6765 
6766 out_register_late_subsys:
6767 	rtnl_unregister_all(PF_INET6);
6768 	unregister_pernet_subsys(&ip6_route_net_late_ops);
6769 fib6_rules_init:
6770 	fib6_rules_cleanup();
6771 xfrm6_init:
6772 	xfrm6_fini();
6773 out_fib6_init:
6774 	fib6_gc_cleanup();
6775 out_register_subsys:
6776 	unregister_pernet_subsys(&ip6_route_net_ops);
6777 out_register_inetpeer:
6778 	unregister_pernet_subsys(&ipv6_inetpeer_ops);
6779 out_dst_entries:
6780 	dst_entries_destroy(&ip6_dst_blackhole_ops);
6781 out_kmem_cache:
6782 	kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6783 	goto out;
6784 }
6785 
6786 void ip6_route_cleanup(void)
6787 {
6788 #if IS_BUILTIN(CONFIG_IPV6)
6789 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6790 	bpf_iter_unregister();
6791 #endif
6792 #endif
6793 	unregister_netdevice_notifier(&ip6_route_dev_notifier);
6794 	unregister_pernet_subsys(&ip6_route_net_late_ops);
6795 	fib6_rules_cleanup();
6796 	xfrm6_fini();
6797 	fib6_gc_cleanup();
6798 	unregister_pernet_subsys(&ipv6_inetpeer_ops);
6799 	unregister_pernet_subsys(&ip6_route_net_ops);
6800 	dst_entries_destroy(&ip6_dst_blackhole_ops);
6801 	kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6802 }
6803