xref: /openbmc/linux/net/ipv6/route.c (revision adb57164)
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 <net/net_namespace.h>
45 #include <net/snmp.h>
46 #include <net/ipv6.h>
47 #include <net/ip6_fib.h>
48 #include <net/ip6_route.h>
49 #include <net/ndisc.h>
50 #include <net/addrconf.h>
51 #include <net/tcp.h>
52 #include <linux/rtnetlink.h>
53 #include <net/dst.h>
54 #include <net/dst_metadata.h>
55 #include <net/xfrm.h>
56 #include <net/netevent.h>
57 #include <net/netlink.h>
58 #include <net/rtnh.h>
59 #include <net/lwtunnel.h>
60 #include <net/ip_tunnels.h>
61 #include <net/l3mdev.h>
62 #include <net/ip.h>
63 #include <linux/uaccess.h>
64 
65 #ifdef CONFIG_SYSCTL
66 #include <linux/sysctl.h>
67 #endif
68 
69 static int ip6_rt_type_to_error(u8 fib6_type);
70 
71 #define CREATE_TRACE_POINTS
72 #include <trace/events/fib6.h>
73 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
74 #undef CREATE_TRACE_POINTS
75 
76 enum rt6_nud_state {
77 	RT6_NUD_FAIL_HARD = -3,
78 	RT6_NUD_FAIL_PROBE = -2,
79 	RT6_NUD_FAIL_DO_RR = -1,
80 	RT6_NUD_SUCCEED = 1
81 };
82 
83 static struct dst_entry	*ip6_dst_check(struct dst_entry *dst, u32 cookie);
84 static unsigned int	 ip6_default_advmss(const struct dst_entry *dst);
85 static unsigned int	 ip6_mtu(const struct dst_entry *dst);
86 static struct dst_entry *ip6_negative_advice(struct dst_entry *);
87 static void		ip6_dst_destroy(struct dst_entry *);
88 static void		ip6_dst_ifdown(struct dst_entry *,
89 				       struct net_device *dev, int how);
90 static int		 ip6_dst_gc(struct dst_ops *ops);
91 
92 static int		ip6_pkt_discard(struct sk_buff *skb);
93 static int		ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
94 static int		ip6_pkt_prohibit(struct sk_buff *skb);
95 static int		ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb);
96 static void		ip6_link_failure(struct sk_buff *skb);
97 static void		ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
98 					   struct sk_buff *skb, u32 mtu,
99 					   bool confirm_neigh);
100 static void		rt6_do_redirect(struct dst_entry *dst, struct sock *sk,
101 					struct sk_buff *skb);
102 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
103 			   int strict);
104 static size_t rt6_nlmsg_size(struct fib6_info *f6i);
105 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
106 			 struct fib6_info *rt, struct dst_entry *dst,
107 			 struct in6_addr *dest, struct in6_addr *src,
108 			 int iif, int type, u32 portid, u32 seq,
109 			 unsigned int flags);
110 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
111 					   const struct in6_addr *daddr,
112 					   const struct in6_addr *saddr);
113 
114 #ifdef CONFIG_IPV6_ROUTE_INFO
115 static struct fib6_info *rt6_add_route_info(struct net *net,
116 					   const struct in6_addr *prefix, int prefixlen,
117 					   const struct in6_addr *gwaddr,
118 					   struct net_device *dev,
119 					   unsigned int pref);
120 static struct fib6_info *rt6_get_route_info(struct net *net,
121 					   const struct in6_addr *prefix, int prefixlen,
122 					   const struct in6_addr *gwaddr,
123 					   struct net_device *dev);
124 #endif
125 
126 struct uncached_list {
127 	spinlock_t		lock;
128 	struct list_head	head;
129 };
130 
131 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
132 
133 void rt6_uncached_list_add(struct rt6_info *rt)
134 {
135 	struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
136 
137 	rt->rt6i_uncached_list = ul;
138 
139 	spin_lock_bh(&ul->lock);
140 	list_add_tail(&rt->rt6i_uncached, &ul->head);
141 	spin_unlock_bh(&ul->lock);
142 }
143 
144 void rt6_uncached_list_del(struct rt6_info *rt)
145 {
146 	if (!list_empty(&rt->rt6i_uncached)) {
147 		struct uncached_list *ul = rt->rt6i_uncached_list;
148 		struct net *net = dev_net(rt->dst.dev);
149 
150 		spin_lock_bh(&ul->lock);
151 		list_del(&rt->rt6i_uncached);
152 		atomic_dec(&net->ipv6.rt6_stats->fib_rt_uncache);
153 		spin_unlock_bh(&ul->lock);
154 	}
155 }
156 
157 static void rt6_uncached_list_flush_dev(struct net *net, struct net_device *dev)
158 {
159 	struct net_device *loopback_dev = net->loopback_dev;
160 	int cpu;
161 
162 	if (dev == loopback_dev)
163 		return;
164 
165 	for_each_possible_cpu(cpu) {
166 		struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
167 		struct rt6_info *rt;
168 
169 		spin_lock_bh(&ul->lock);
170 		list_for_each_entry(rt, &ul->head, rt6i_uncached) {
171 			struct inet6_dev *rt_idev = rt->rt6i_idev;
172 			struct net_device *rt_dev = rt->dst.dev;
173 
174 			if (rt_idev->dev == dev) {
175 				rt->rt6i_idev = in6_dev_get(loopback_dev);
176 				in6_dev_put(rt_idev);
177 			}
178 
179 			if (rt_dev == dev) {
180 				rt->dst.dev = blackhole_netdev;
181 				dev_hold(rt->dst.dev);
182 				dev_put(rt_dev);
183 			}
184 		}
185 		spin_unlock_bh(&ul->lock);
186 	}
187 }
188 
189 static inline const void *choose_neigh_daddr(const struct in6_addr *p,
190 					     struct sk_buff *skb,
191 					     const void *daddr)
192 {
193 	if (!ipv6_addr_any(p))
194 		return (const void *) p;
195 	else if (skb)
196 		return &ipv6_hdr(skb)->daddr;
197 	return daddr;
198 }
199 
200 struct neighbour *ip6_neigh_lookup(const struct in6_addr *gw,
201 				   struct net_device *dev,
202 				   struct sk_buff *skb,
203 				   const void *daddr)
204 {
205 	struct neighbour *n;
206 
207 	daddr = choose_neigh_daddr(gw, skb, daddr);
208 	n = __ipv6_neigh_lookup(dev, daddr);
209 	if (n)
210 		return n;
211 
212 	n = neigh_create(&nd_tbl, daddr, dev);
213 	return IS_ERR(n) ? NULL : n;
214 }
215 
216 static struct neighbour *ip6_dst_neigh_lookup(const struct dst_entry *dst,
217 					      struct sk_buff *skb,
218 					      const void *daddr)
219 {
220 	const struct rt6_info *rt = container_of(dst, struct rt6_info, dst);
221 
222 	return ip6_neigh_lookup(rt6_nexthop(rt, &in6addr_any),
223 				dst->dev, skb, daddr);
224 }
225 
226 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
227 {
228 	struct net_device *dev = dst->dev;
229 	struct rt6_info *rt = (struct rt6_info *)dst;
230 
231 	daddr = choose_neigh_daddr(rt6_nexthop(rt, &in6addr_any), NULL, daddr);
232 	if (!daddr)
233 		return;
234 	if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
235 		return;
236 	if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
237 		return;
238 	__ipv6_confirm_neigh(dev, daddr);
239 }
240 
241 static struct dst_ops ip6_dst_ops_template = {
242 	.family			=	AF_INET6,
243 	.gc			=	ip6_dst_gc,
244 	.gc_thresh		=	1024,
245 	.check			=	ip6_dst_check,
246 	.default_advmss		=	ip6_default_advmss,
247 	.mtu			=	ip6_mtu,
248 	.cow_metrics		=	dst_cow_metrics_generic,
249 	.destroy		=	ip6_dst_destroy,
250 	.ifdown			=	ip6_dst_ifdown,
251 	.negative_advice	=	ip6_negative_advice,
252 	.link_failure		=	ip6_link_failure,
253 	.update_pmtu		=	ip6_rt_update_pmtu,
254 	.redirect		=	rt6_do_redirect,
255 	.local_out		=	__ip6_local_out,
256 	.neigh_lookup		=	ip6_dst_neigh_lookup,
257 	.confirm_neigh		=	ip6_confirm_neigh,
258 };
259 
260 static unsigned int ip6_blackhole_mtu(const struct dst_entry *dst)
261 {
262 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
263 
264 	return mtu ? : dst->dev->mtu;
265 }
266 
267 static void ip6_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
268 					 struct sk_buff *skb, u32 mtu,
269 					 bool confirm_neigh)
270 {
271 }
272 
273 static void ip6_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
274 				      struct sk_buff *skb)
275 {
276 }
277 
278 static struct dst_ops ip6_dst_blackhole_ops = {
279 	.family			=	AF_INET6,
280 	.destroy		=	ip6_dst_destroy,
281 	.check			=	ip6_dst_check,
282 	.mtu			=	ip6_blackhole_mtu,
283 	.default_advmss		=	ip6_default_advmss,
284 	.update_pmtu		=	ip6_rt_blackhole_update_pmtu,
285 	.redirect		=	ip6_rt_blackhole_redirect,
286 	.cow_metrics		=	dst_cow_metrics_generic,
287 	.neigh_lookup		=	ip6_dst_neigh_lookup,
288 };
289 
290 static const u32 ip6_template_metrics[RTAX_MAX] = {
291 	[RTAX_HOPLIMIT - 1] = 0,
292 };
293 
294 static const struct fib6_info fib6_null_entry_template = {
295 	.fib6_flags	= (RTF_REJECT | RTF_NONEXTHOP),
296 	.fib6_protocol  = RTPROT_KERNEL,
297 	.fib6_metric	= ~(u32)0,
298 	.fib6_ref	= REFCOUNT_INIT(1),
299 	.fib6_type	= RTN_UNREACHABLE,
300 	.fib6_metrics	= (struct dst_metrics *)&dst_default_metrics,
301 };
302 
303 static const struct rt6_info ip6_null_entry_template = {
304 	.dst = {
305 		.__refcnt	= ATOMIC_INIT(1),
306 		.__use		= 1,
307 		.obsolete	= DST_OBSOLETE_FORCE_CHK,
308 		.error		= -ENETUNREACH,
309 		.input		= ip6_pkt_discard,
310 		.output		= ip6_pkt_discard_out,
311 	},
312 	.rt6i_flags	= (RTF_REJECT | RTF_NONEXTHOP),
313 };
314 
315 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
316 
317 static const struct rt6_info ip6_prohibit_entry_template = {
318 	.dst = {
319 		.__refcnt	= ATOMIC_INIT(1),
320 		.__use		= 1,
321 		.obsolete	= DST_OBSOLETE_FORCE_CHK,
322 		.error		= -EACCES,
323 		.input		= ip6_pkt_prohibit,
324 		.output		= ip6_pkt_prohibit_out,
325 	},
326 	.rt6i_flags	= (RTF_REJECT | RTF_NONEXTHOP),
327 };
328 
329 static const struct rt6_info ip6_blk_hole_entry_template = {
330 	.dst = {
331 		.__refcnt	= ATOMIC_INIT(1),
332 		.__use		= 1,
333 		.obsolete	= DST_OBSOLETE_FORCE_CHK,
334 		.error		= -EINVAL,
335 		.input		= dst_discard,
336 		.output		= dst_discard_out,
337 	},
338 	.rt6i_flags	= (RTF_REJECT | RTF_NONEXTHOP),
339 };
340 
341 #endif
342 
343 static void rt6_info_init(struct rt6_info *rt)
344 {
345 	struct dst_entry *dst = &rt->dst;
346 
347 	memset(dst + 1, 0, sizeof(*rt) - sizeof(*dst));
348 	INIT_LIST_HEAD(&rt->rt6i_uncached);
349 }
350 
351 /* allocate dst with ip6_dst_ops */
352 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
353 			       int flags)
354 {
355 	struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
356 					1, DST_OBSOLETE_FORCE_CHK, flags);
357 
358 	if (rt) {
359 		rt6_info_init(rt);
360 		atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
361 	}
362 
363 	return rt;
364 }
365 EXPORT_SYMBOL(ip6_dst_alloc);
366 
367 static void ip6_dst_destroy(struct dst_entry *dst)
368 {
369 	struct rt6_info *rt = (struct rt6_info *)dst;
370 	struct fib6_info *from;
371 	struct inet6_dev *idev;
372 
373 	ip_dst_metrics_put(dst);
374 	rt6_uncached_list_del(rt);
375 
376 	idev = rt->rt6i_idev;
377 	if (idev) {
378 		rt->rt6i_idev = NULL;
379 		in6_dev_put(idev);
380 	}
381 
382 	from = xchg((__force struct fib6_info **)&rt->from, NULL);
383 	fib6_info_release(from);
384 }
385 
386 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
387 			   int how)
388 {
389 	struct rt6_info *rt = (struct rt6_info *)dst;
390 	struct inet6_dev *idev = rt->rt6i_idev;
391 	struct net_device *loopback_dev =
392 		dev_net(dev)->loopback_dev;
393 
394 	if (idev && idev->dev != loopback_dev) {
395 		struct inet6_dev *loopback_idev = in6_dev_get(loopback_dev);
396 		if (loopback_idev) {
397 			rt->rt6i_idev = loopback_idev;
398 			in6_dev_put(idev);
399 		}
400 	}
401 }
402 
403 static bool __rt6_check_expired(const struct rt6_info *rt)
404 {
405 	if (rt->rt6i_flags & RTF_EXPIRES)
406 		return time_after(jiffies, rt->dst.expires);
407 	else
408 		return false;
409 }
410 
411 static bool rt6_check_expired(const struct rt6_info *rt)
412 {
413 	struct fib6_info *from;
414 
415 	from = rcu_dereference(rt->from);
416 
417 	if (rt->rt6i_flags & RTF_EXPIRES) {
418 		if (time_after(jiffies, rt->dst.expires))
419 			return true;
420 	} else if (from) {
421 		return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
422 			fib6_check_expired(from);
423 	}
424 	return false;
425 }
426 
427 void fib6_select_path(const struct net *net, struct fib6_result *res,
428 		      struct flowi6 *fl6, int oif, bool have_oif_match,
429 		      const struct sk_buff *skb, int strict)
430 {
431 	struct fib6_info *sibling, *next_sibling;
432 	struct fib6_info *match = res->f6i;
433 
434 	if ((!match->fib6_nsiblings && !match->nh) || have_oif_match)
435 		goto out;
436 
437 	/* We might have already computed the hash for ICMPv6 errors. In such
438 	 * case it will always be non-zero. Otherwise now is the time to do it.
439 	 */
440 	if (!fl6->mp_hash &&
441 	    (!match->nh || nexthop_is_multipath(match->nh)))
442 		fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
443 
444 	if (unlikely(match->nh)) {
445 		nexthop_path_fib6_result(res, fl6->mp_hash);
446 		return;
447 	}
448 
449 	if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
450 		goto out;
451 
452 	list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
453 				 fib6_siblings) {
454 		const struct fib6_nh *nh = sibling->fib6_nh;
455 		int nh_upper_bound;
456 
457 		nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
458 		if (fl6->mp_hash > nh_upper_bound)
459 			continue;
460 		if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
461 			break;
462 		match = sibling;
463 		break;
464 	}
465 
466 out:
467 	res->f6i = match;
468 	res->nh = match->fib6_nh;
469 }
470 
471 /*
472  *	Route lookup. rcu_read_lock() should be held.
473  */
474 
475 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
476 			       const struct in6_addr *saddr, int oif, int flags)
477 {
478 	const struct net_device *dev;
479 
480 	if (nh->fib_nh_flags & RTNH_F_DEAD)
481 		return false;
482 
483 	dev = nh->fib_nh_dev;
484 	if (oif) {
485 		if (dev->ifindex == oif)
486 			return true;
487 	} else {
488 		if (ipv6_chk_addr(net, saddr, dev,
489 				  flags & RT6_LOOKUP_F_IFACE))
490 			return true;
491 	}
492 
493 	return false;
494 }
495 
496 struct fib6_nh_dm_arg {
497 	struct net		*net;
498 	const struct in6_addr	*saddr;
499 	int			oif;
500 	int			flags;
501 	struct fib6_nh		*nh;
502 };
503 
504 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
505 {
506 	struct fib6_nh_dm_arg *arg = _arg;
507 
508 	arg->nh = nh;
509 	return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
510 				  arg->flags);
511 }
512 
513 /* returns fib6_nh from nexthop or NULL */
514 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
515 					struct fib6_result *res,
516 					const struct in6_addr *saddr,
517 					int oif, int flags)
518 {
519 	struct fib6_nh_dm_arg arg = {
520 		.net   = net,
521 		.saddr = saddr,
522 		.oif   = oif,
523 		.flags = flags,
524 	};
525 
526 	if (nexthop_is_blackhole(nh))
527 		return NULL;
528 
529 	if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
530 		return arg.nh;
531 
532 	return NULL;
533 }
534 
535 static void rt6_device_match(struct net *net, struct fib6_result *res,
536 			     const struct in6_addr *saddr, int oif, int flags)
537 {
538 	struct fib6_info *f6i = res->f6i;
539 	struct fib6_info *spf6i;
540 	struct fib6_nh *nh;
541 
542 	if (!oif && ipv6_addr_any(saddr)) {
543 		if (unlikely(f6i->nh)) {
544 			nh = nexthop_fib6_nh(f6i->nh);
545 			if (nexthop_is_blackhole(f6i->nh))
546 				goto out_blackhole;
547 		} else {
548 			nh = f6i->fib6_nh;
549 		}
550 		if (!(nh->fib_nh_flags & RTNH_F_DEAD))
551 			goto out;
552 	}
553 
554 	for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
555 		bool matched = false;
556 
557 		if (unlikely(spf6i->nh)) {
558 			nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
559 					      oif, flags);
560 			if (nh)
561 				matched = true;
562 		} else {
563 			nh = spf6i->fib6_nh;
564 			if (__rt6_device_match(net, nh, saddr, oif, flags))
565 				matched = true;
566 		}
567 		if (matched) {
568 			res->f6i = spf6i;
569 			goto out;
570 		}
571 	}
572 
573 	if (oif && flags & RT6_LOOKUP_F_IFACE) {
574 		res->f6i = net->ipv6.fib6_null_entry;
575 		nh = res->f6i->fib6_nh;
576 		goto out;
577 	}
578 
579 	if (unlikely(f6i->nh)) {
580 		nh = nexthop_fib6_nh(f6i->nh);
581 		if (nexthop_is_blackhole(f6i->nh))
582 			goto out_blackhole;
583 	} else {
584 		nh = f6i->fib6_nh;
585 	}
586 
587 	if (nh->fib_nh_flags & RTNH_F_DEAD) {
588 		res->f6i = net->ipv6.fib6_null_entry;
589 		nh = res->f6i->fib6_nh;
590 	}
591 out:
592 	res->nh = nh;
593 	res->fib6_type = res->f6i->fib6_type;
594 	res->fib6_flags = res->f6i->fib6_flags;
595 	return;
596 
597 out_blackhole:
598 	res->fib6_flags |= RTF_REJECT;
599 	res->fib6_type = RTN_BLACKHOLE;
600 	res->nh = nh;
601 }
602 
603 #ifdef CONFIG_IPV6_ROUTER_PREF
604 struct __rt6_probe_work {
605 	struct work_struct work;
606 	struct in6_addr target;
607 	struct net_device *dev;
608 };
609 
610 static void rt6_probe_deferred(struct work_struct *w)
611 {
612 	struct in6_addr mcaddr;
613 	struct __rt6_probe_work *work =
614 		container_of(w, struct __rt6_probe_work, work);
615 
616 	addrconf_addr_solict_mult(&work->target, &mcaddr);
617 	ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
618 	dev_put(work->dev);
619 	kfree(work);
620 }
621 
622 static void rt6_probe(struct fib6_nh *fib6_nh)
623 {
624 	struct __rt6_probe_work *work = NULL;
625 	const struct in6_addr *nh_gw;
626 	unsigned long last_probe;
627 	struct neighbour *neigh;
628 	struct net_device *dev;
629 	struct inet6_dev *idev;
630 
631 	/*
632 	 * Okay, this does not seem to be appropriate
633 	 * for now, however, we need to check if it
634 	 * is really so; aka Router Reachability Probing.
635 	 *
636 	 * Router Reachability Probe MUST be rate-limited
637 	 * to no more than one per minute.
638 	 */
639 	if (!fib6_nh->fib_nh_gw_family)
640 		return;
641 
642 	nh_gw = &fib6_nh->fib_nh_gw6;
643 	dev = fib6_nh->fib_nh_dev;
644 	rcu_read_lock_bh();
645 	last_probe = READ_ONCE(fib6_nh->last_probe);
646 	idev = __in6_dev_get(dev);
647 	neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
648 	if (neigh) {
649 		if (neigh->nud_state & NUD_VALID)
650 			goto out;
651 
652 		write_lock(&neigh->lock);
653 		if (!(neigh->nud_state & NUD_VALID) &&
654 		    time_after(jiffies,
655 			       neigh->updated + idev->cnf.rtr_probe_interval)) {
656 			work = kmalloc(sizeof(*work), GFP_ATOMIC);
657 			if (work)
658 				__neigh_set_probe_once(neigh);
659 		}
660 		write_unlock(&neigh->lock);
661 	} else if (time_after(jiffies, last_probe +
662 				       idev->cnf.rtr_probe_interval)) {
663 		work = kmalloc(sizeof(*work), GFP_ATOMIC);
664 	}
665 
666 	if (!work || cmpxchg(&fib6_nh->last_probe,
667 			     last_probe, jiffies) != last_probe) {
668 		kfree(work);
669 	} else {
670 		INIT_WORK(&work->work, rt6_probe_deferred);
671 		work->target = *nh_gw;
672 		dev_hold(dev);
673 		work->dev = dev;
674 		schedule_work(&work->work);
675 	}
676 
677 out:
678 	rcu_read_unlock_bh();
679 }
680 #else
681 static inline void rt6_probe(struct fib6_nh *fib6_nh)
682 {
683 }
684 #endif
685 
686 /*
687  * Default Router Selection (RFC 2461 6.3.6)
688  */
689 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
690 {
691 	enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
692 	struct neighbour *neigh;
693 
694 	rcu_read_lock_bh();
695 	neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
696 					  &fib6_nh->fib_nh_gw6);
697 	if (neigh) {
698 		read_lock(&neigh->lock);
699 		if (neigh->nud_state & NUD_VALID)
700 			ret = RT6_NUD_SUCCEED;
701 #ifdef CONFIG_IPV6_ROUTER_PREF
702 		else if (!(neigh->nud_state & NUD_FAILED))
703 			ret = RT6_NUD_SUCCEED;
704 		else
705 			ret = RT6_NUD_FAIL_PROBE;
706 #endif
707 		read_unlock(&neigh->lock);
708 	} else {
709 		ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
710 		      RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
711 	}
712 	rcu_read_unlock_bh();
713 
714 	return ret;
715 }
716 
717 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
718 			   int strict)
719 {
720 	int m = 0;
721 
722 	if (!oif || nh->fib_nh_dev->ifindex == oif)
723 		m = 2;
724 
725 	if (!m && (strict & RT6_LOOKUP_F_IFACE))
726 		return RT6_NUD_FAIL_HARD;
727 #ifdef CONFIG_IPV6_ROUTER_PREF
728 	m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
729 #endif
730 	if ((strict & RT6_LOOKUP_F_REACHABLE) &&
731 	    !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
732 		int n = rt6_check_neigh(nh);
733 		if (n < 0)
734 			return n;
735 	}
736 	return m;
737 }
738 
739 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
740 		       int oif, int strict, int *mpri, bool *do_rr)
741 {
742 	bool match_do_rr = false;
743 	bool rc = false;
744 	int m;
745 
746 	if (nh->fib_nh_flags & RTNH_F_DEAD)
747 		goto out;
748 
749 	if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
750 	    nh->fib_nh_flags & RTNH_F_LINKDOWN &&
751 	    !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
752 		goto out;
753 
754 	m = rt6_score_route(nh, fib6_flags, oif, strict);
755 	if (m == RT6_NUD_FAIL_DO_RR) {
756 		match_do_rr = true;
757 		m = 0; /* lowest valid score */
758 	} else if (m == RT6_NUD_FAIL_HARD) {
759 		goto out;
760 	}
761 
762 	if (strict & RT6_LOOKUP_F_REACHABLE)
763 		rt6_probe(nh);
764 
765 	/* note that m can be RT6_NUD_FAIL_PROBE at this point */
766 	if (m > *mpri) {
767 		*do_rr = match_do_rr;
768 		*mpri = m;
769 		rc = true;
770 	}
771 out:
772 	return rc;
773 }
774 
775 struct fib6_nh_frl_arg {
776 	u32		flags;
777 	int		oif;
778 	int		strict;
779 	int		*mpri;
780 	bool		*do_rr;
781 	struct fib6_nh	*nh;
782 };
783 
784 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
785 {
786 	struct fib6_nh_frl_arg *arg = _arg;
787 
788 	arg->nh = nh;
789 	return find_match(nh, arg->flags, arg->oif, arg->strict,
790 			  arg->mpri, arg->do_rr);
791 }
792 
793 static void __find_rr_leaf(struct fib6_info *f6i_start,
794 			   struct fib6_info *nomatch, u32 metric,
795 			   struct fib6_result *res, struct fib6_info **cont,
796 			   int oif, int strict, bool *do_rr, int *mpri)
797 {
798 	struct fib6_info *f6i;
799 
800 	for (f6i = f6i_start;
801 	     f6i && f6i != nomatch;
802 	     f6i = rcu_dereference(f6i->fib6_next)) {
803 		bool matched = false;
804 		struct fib6_nh *nh;
805 
806 		if (cont && f6i->fib6_metric != metric) {
807 			*cont = f6i;
808 			return;
809 		}
810 
811 		if (fib6_check_expired(f6i))
812 			continue;
813 
814 		if (unlikely(f6i->nh)) {
815 			struct fib6_nh_frl_arg arg = {
816 				.flags  = f6i->fib6_flags,
817 				.oif    = oif,
818 				.strict = strict,
819 				.mpri   = mpri,
820 				.do_rr  = do_rr
821 			};
822 
823 			if (nexthop_is_blackhole(f6i->nh)) {
824 				res->fib6_flags = RTF_REJECT;
825 				res->fib6_type = RTN_BLACKHOLE;
826 				res->f6i = f6i;
827 				res->nh = nexthop_fib6_nh(f6i->nh);
828 				return;
829 			}
830 			if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
831 						     &arg)) {
832 				matched = true;
833 				nh = arg.nh;
834 			}
835 		} else {
836 			nh = f6i->fib6_nh;
837 			if (find_match(nh, f6i->fib6_flags, oif, strict,
838 				       mpri, do_rr))
839 				matched = true;
840 		}
841 		if (matched) {
842 			res->f6i = f6i;
843 			res->nh = nh;
844 			res->fib6_flags = f6i->fib6_flags;
845 			res->fib6_type = f6i->fib6_type;
846 		}
847 	}
848 }
849 
850 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
851 			 struct fib6_info *rr_head, int oif, int strict,
852 			 bool *do_rr, struct fib6_result *res)
853 {
854 	u32 metric = rr_head->fib6_metric;
855 	struct fib6_info *cont = NULL;
856 	int mpri = -1;
857 
858 	__find_rr_leaf(rr_head, NULL, metric, res, &cont,
859 		       oif, strict, do_rr, &mpri);
860 
861 	__find_rr_leaf(leaf, rr_head, metric, res, &cont,
862 		       oif, strict, do_rr, &mpri);
863 
864 	if (res->f6i || !cont)
865 		return;
866 
867 	__find_rr_leaf(cont, NULL, metric, res, NULL,
868 		       oif, strict, do_rr, &mpri);
869 }
870 
871 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
872 		       struct fib6_result *res, int strict)
873 {
874 	struct fib6_info *leaf = rcu_dereference(fn->leaf);
875 	struct fib6_info *rt0;
876 	bool do_rr = false;
877 	int key_plen;
878 
879 	/* make sure this function or its helpers sets f6i */
880 	res->f6i = NULL;
881 
882 	if (!leaf || leaf == net->ipv6.fib6_null_entry)
883 		goto out;
884 
885 	rt0 = rcu_dereference(fn->rr_ptr);
886 	if (!rt0)
887 		rt0 = leaf;
888 
889 	/* Double check to make sure fn is not an intermediate node
890 	 * and fn->leaf does not points to its child's leaf
891 	 * (This might happen if all routes under fn are deleted from
892 	 * the tree and fib6_repair_tree() is called on the node.)
893 	 */
894 	key_plen = rt0->fib6_dst.plen;
895 #ifdef CONFIG_IPV6_SUBTREES
896 	if (rt0->fib6_src.plen)
897 		key_plen = rt0->fib6_src.plen;
898 #endif
899 	if (fn->fn_bit != key_plen)
900 		goto out;
901 
902 	find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
903 	if (do_rr) {
904 		struct fib6_info *next = rcu_dereference(rt0->fib6_next);
905 
906 		/* no entries matched; do round-robin */
907 		if (!next || next->fib6_metric != rt0->fib6_metric)
908 			next = leaf;
909 
910 		if (next != rt0) {
911 			spin_lock_bh(&leaf->fib6_table->tb6_lock);
912 			/* make sure next is not being deleted from the tree */
913 			if (next->fib6_node)
914 				rcu_assign_pointer(fn->rr_ptr, next);
915 			spin_unlock_bh(&leaf->fib6_table->tb6_lock);
916 		}
917 	}
918 
919 out:
920 	if (!res->f6i) {
921 		res->f6i = net->ipv6.fib6_null_entry;
922 		res->nh = res->f6i->fib6_nh;
923 		res->fib6_flags = res->f6i->fib6_flags;
924 		res->fib6_type = res->f6i->fib6_type;
925 	}
926 }
927 
928 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
929 {
930 	return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
931 	       res->nh->fib_nh_gw_family;
932 }
933 
934 #ifdef CONFIG_IPV6_ROUTE_INFO
935 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
936 		  const struct in6_addr *gwaddr)
937 {
938 	struct net *net = dev_net(dev);
939 	struct route_info *rinfo = (struct route_info *) opt;
940 	struct in6_addr prefix_buf, *prefix;
941 	unsigned int pref;
942 	unsigned long lifetime;
943 	struct fib6_info *rt;
944 
945 	if (len < sizeof(struct route_info)) {
946 		return -EINVAL;
947 	}
948 
949 	/* Sanity check for prefix_len and length */
950 	if (rinfo->length > 3) {
951 		return -EINVAL;
952 	} else if (rinfo->prefix_len > 128) {
953 		return -EINVAL;
954 	} else if (rinfo->prefix_len > 64) {
955 		if (rinfo->length < 2) {
956 			return -EINVAL;
957 		}
958 	} else if (rinfo->prefix_len > 0) {
959 		if (rinfo->length < 1) {
960 			return -EINVAL;
961 		}
962 	}
963 
964 	pref = rinfo->route_pref;
965 	if (pref == ICMPV6_ROUTER_PREF_INVALID)
966 		return -EINVAL;
967 
968 	lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
969 
970 	if (rinfo->length == 3)
971 		prefix = (struct in6_addr *)rinfo->prefix;
972 	else {
973 		/* this function is safe */
974 		ipv6_addr_prefix(&prefix_buf,
975 				 (struct in6_addr *)rinfo->prefix,
976 				 rinfo->prefix_len);
977 		prefix = &prefix_buf;
978 	}
979 
980 	if (rinfo->prefix_len == 0)
981 		rt = rt6_get_dflt_router(net, gwaddr, dev);
982 	else
983 		rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
984 					gwaddr, dev);
985 
986 	if (rt && !lifetime) {
987 		ip6_del_rt(net, rt);
988 		rt = NULL;
989 	}
990 
991 	if (!rt && lifetime)
992 		rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
993 					dev, pref);
994 	else if (rt)
995 		rt->fib6_flags = RTF_ROUTEINFO |
996 				 (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
997 
998 	if (rt) {
999 		if (!addrconf_finite_timeout(lifetime))
1000 			fib6_clean_expires(rt);
1001 		else
1002 			fib6_set_expires(rt, jiffies + HZ * lifetime);
1003 
1004 		fib6_info_release(rt);
1005 	}
1006 	return 0;
1007 }
1008 #endif
1009 
1010 /*
1011  *	Misc support functions
1012  */
1013 
1014 /* called with rcu_lock held */
1015 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1016 {
1017 	struct net_device *dev = res->nh->fib_nh_dev;
1018 
1019 	if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1020 		/* for copies of local routes, dst->dev needs to be the
1021 		 * device if it is a master device, the master device if
1022 		 * device is enslaved, and the loopback as the default
1023 		 */
1024 		if (netif_is_l3_slave(dev) &&
1025 		    !rt6_need_strict(&res->f6i->fib6_dst.addr))
1026 			dev = l3mdev_master_dev_rcu(dev);
1027 		else if (!netif_is_l3_master(dev))
1028 			dev = dev_net(dev)->loopback_dev;
1029 		/* last case is netif_is_l3_master(dev) is true in which
1030 		 * case we want dev returned to be dev
1031 		 */
1032 	}
1033 
1034 	return dev;
1035 }
1036 
1037 static const int fib6_prop[RTN_MAX + 1] = {
1038 	[RTN_UNSPEC]	= 0,
1039 	[RTN_UNICAST]	= 0,
1040 	[RTN_LOCAL]	= 0,
1041 	[RTN_BROADCAST]	= 0,
1042 	[RTN_ANYCAST]	= 0,
1043 	[RTN_MULTICAST]	= 0,
1044 	[RTN_BLACKHOLE]	= -EINVAL,
1045 	[RTN_UNREACHABLE] = -EHOSTUNREACH,
1046 	[RTN_PROHIBIT]	= -EACCES,
1047 	[RTN_THROW]	= -EAGAIN,
1048 	[RTN_NAT]	= -EINVAL,
1049 	[RTN_XRESOLVE]	= -EINVAL,
1050 };
1051 
1052 static int ip6_rt_type_to_error(u8 fib6_type)
1053 {
1054 	return fib6_prop[fib6_type];
1055 }
1056 
1057 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1058 {
1059 	unsigned short flags = 0;
1060 
1061 	if (rt->dst_nocount)
1062 		flags |= DST_NOCOUNT;
1063 	if (rt->dst_nopolicy)
1064 		flags |= DST_NOPOLICY;
1065 
1066 	return flags;
1067 }
1068 
1069 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1070 {
1071 	rt->dst.error = ip6_rt_type_to_error(fib6_type);
1072 
1073 	switch (fib6_type) {
1074 	case RTN_BLACKHOLE:
1075 		rt->dst.output = dst_discard_out;
1076 		rt->dst.input = dst_discard;
1077 		break;
1078 	case RTN_PROHIBIT:
1079 		rt->dst.output = ip6_pkt_prohibit_out;
1080 		rt->dst.input = ip6_pkt_prohibit;
1081 		break;
1082 	case RTN_THROW:
1083 	case RTN_UNREACHABLE:
1084 	default:
1085 		rt->dst.output = ip6_pkt_discard_out;
1086 		rt->dst.input = ip6_pkt_discard;
1087 		break;
1088 	}
1089 }
1090 
1091 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1092 {
1093 	struct fib6_info *f6i = res->f6i;
1094 
1095 	if (res->fib6_flags & RTF_REJECT) {
1096 		ip6_rt_init_dst_reject(rt, res->fib6_type);
1097 		return;
1098 	}
1099 
1100 	rt->dst.error = 0;
1101 	rt->dst.output = ip6_output;
1102 
1103 	if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1104 		rt->dst.input = ip6_input;
1105 	} else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1106 		rt->dst.input = ip6_mc_input;
1107 	} else {
1108 		rt->dst.input = ip6_forward;
1109 	}
1110 
1111 	if (res->nh->fib_nh_lws) {
1112 		rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1113 		lwtunnel_set_redirect(&rt->dst);
1114 	}
1115 
1116 	rt->dst.lastuse = jiffies;
1117 }
1118 
1119 /* Caller must already hold reference to @from */
1120 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1121 {
1122 	rt->rt6i_flags &= ~RTF_EXPIRES;
1123 	rcu_assign_pointer(rt->from, from);
1124 	ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1125 }
1126 
1127 /* Caller must already hold reference to f6i in result */
1128 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1129 {
1130 	const struct fib6_nh *nh = res->nh;
1131 	const struct net_device *dev = nh->fib_nh_dev;
1132 	struct fib6_info *f6i = res->f6i;
1133 
1134 	ip6_rt_init_dst(rt, res);
1135 
1136 	rt->rt6i_dst = f6i->fib6_dst;
1137 	rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1138 	rt->rt6i_flags = res->fib6_flags;
1139 	if (nh->fib_nh_gw_family) {
1140 		rt->rt6i_gateway = nh->fib_nh_gw6;
1141 		rt->rt6i_flags |= RTF_GATEWAY;
1142 	}
1143 	rt6_set_from(rt, f6i);
1144 #ifdef CONFIG_IPV6_SUBTREES
1145 	rt->rt6i_src = f6i->fib6_src;
1146 #endif
1147 }
1148 
1149 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1150 					struct in6_addr *saddr)
1151 {
1152 	struct fib6_node *pn, *sn;
1153 	while (1) {
1154 		if (fn->fn_flags & RTN_TL_ROOT)
1155 			return NULL;
1156 		pn = rcu_dereference(fn->parent);
1157 		sn = FIB6_SUBTREE(pn);
1158 		if (sn && sn != fn)
1159 			fn = fib6_node_lookup(sn, NULL, saddr);
1160 		else
1161 			fn = pn;
1162 		if (fn->fn_flags & RTN_RTINFO)
1163 			return fn;
1164 	}
1165 }
1166 
1167 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1168 {
1169 	struct rt6_info *rt = *prt;
1170 
1171 	if (dst_hold_safe(&rt->dst))
1172 		return true;
1173 	if (net) {
1174 		rt = net->ipv6.ip6_null_entry;
1175 		dst_hold(&rt->dst);
1176 	} else {
1177 		rt = NULL;
1178 	}
1179 	*prt = rt;
1180 	return false;
1181 }
1182 
1183 /* called with rcu_lock held */
1184 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1185 {
1186 	struct net_device *dev = res->nh->fib_nh_dev;
1187 	struct fib6_info *f6i = res->f6i;
1188 	unsigned short flags;
1189 	struct rt6_info *nrt;
1190 
1191 	if (!fib6_info_hold_safe(f6i))
1192 		goto fallback;
1193 
1194 	flags = fib6_info_dst_flags(f6i);
1195 	nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1196 	if (!nrt) {
1197 		fib6_info_release(f6i);
1198 		goto fallback;
1199 	}
1200 
1201 	ip6_rt_copy_init(nrt, res);
1202 	return nrt;
1203 
1204 fallback:
1205 	nrt = dev_net(dev)->ipv6.ip6_null_entry;
1206 	dst_hold(&nrt->dst);
1207 	return nrt;
1208 }
1209 
1210 static struct rt6_info *ip6_pol_route_lookup(struct net *net,
1211 					     struct fib6_table *table,
1212 					     struct flowi6 *fl6,
1213 					     const struct sk_buff *skb,
1214 					     int flags)
1215 {
1216 	struct fib6_result res = {};
1217 	struct fib6_node *fn;
1218 	struct rt6_info *rt;
1219 
1220 	if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
1221 		flags &= ~RT6_LOOKUP_F_IFACE;
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);
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 	return pcpu_rt;
1389 }
1390 
1391 /* It should be called with rcu_read_lock() acquired */
1392 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1393 {
1394 	struct rt6_info *pcpu_rt;
1395 
1396 	pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1397 
1398 	return pcpu_rt;
1399 }
1400 
1401 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1402 					    const struct fib6_result *res)
1403 {
1404 	struct rt6_info *pcpu_rt, *prev, **p;
1405 
1406 	pcpu_rt = ip6_rt_pcpu_alloc(res);
1407 	if (!pcpu_rt)
1408 		return NULL;
1409 
1410 	p = this_cpu_ptr(res->nh->rt6i_pcpu);
1411 	prev = cmpxchg(p, NULL, pcpu_rt);
1412 	BUG_ON(prev);
1413 
1414 	if (res->f6i->fib6_destroying) {
1415 		struct fib6_info *from;
1416 
1417 		from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1418 		fib6_info_release(from);
1419 	}
1420 
1421 	return pcpu_rt;
1422 }
1423 
1424 /* exception hash table implementation
1425  */
1426 static DEFINE_SPINLOCK(rt6_exception_lock);
1427 
1428 /* Remove rt6_ex from hash table and free the memory
1429  * Caller must hold rt6_exception_lock
1430  */
1431 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1432 				 struct rt6_exception *rt6_ex)
1433 {
1434 	struct fib6_info *from;
1435 	struct net *net;
1436 
1437 	if (!bucket || !rt6_ex)
1438 		return;
1439 
1440 	net = dev_net(rt6_ex->rt6i->dst.dev);
1441 	net->ipv6.rt6_stats->fib_rt_cache--;
1442 
1443 	/* purge completely the exception to allow releasing the held resources:
1444 	 * some [sk] cache may keep the dst around for unlimited time
1445 	 */
1446 	from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1447 	fib6_info_release(from);
1448 	dst_dev_put(&rt6_ex->rt6i->dst);
1449 
1450 	hlist_del_rcu(&rt6_ex->hlist);
1451 	dst_release(&rt6_ex->rt6i->dst);
1452 	kfree_rcu(rt6_ex, rcu);
1453 	WARN_ON_ONCE(!bucket->depth);
1454 	bucket->depth--;
1455 }
1456 
1457 /* Remove oldest rt6_ex in bucket and free the memory
1458  * Caller must hold rt6_exception_lock
1459  */
1460 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1461 {
1462 	struct rt6_exception *rt6_ex, *oldest = NULL;
1463 
1464 	if (!bucket)
1465 		return;
1466 
1467 	hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1468 		if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1469 			oldest = rt6_ex;
1470 	}
1471 	rt6_remove_exception(bucket, oldest);
1472 }
1473 
1474 static u32 rt6_exception_hash(const struct in6_addr *dst,
1475 			      const struct in6_addr *src)
1476 {
1477 	static u32 seed __read_mostly;
1478 	u32 val;
1479 
1480 	net_get_random_once(&seed, sizeof(seed));
1481 	val = jhash2((const u32 *)dst, sizeof(*dst)/sizeof(u32), seed);
1482 
1483 #ifdef CONFIG_IPV6_SUBTREES
1484 	if (src)
1485 		val = jhash2((const u32 *)src, sizeof(*src)/sizeof(u32), val);
1486 #endif
1487 	return hash_32(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1488 }
1489 
1490 /* Helper function to find the cached rt in the hash table
1491  * and update bucket pointer to point to the bucket for this
1492  * (daddr, saddr) pair
1493  * Caller must hold rt6_exception_lock
1494  */
1495 static struct rt6_exception *
1496 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1497 			      const struct in6_addr *daddr,
1498 			      const struct in6_addr *saddr)
1499 {
1500 	struct rt6_exception *rt6_ex;
1501 	u32 hval;
1502 
1503 	if (!(*bucket) || !daddr)
1504 		return NULL;
1505 
1506 	hval = rt6_exception_hash(daddr, saddr);
1507 	*bucket += hval;
1508 
1509 	hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1510 		struct rt6_info *rt6 = rt6_ex->rt6i;
1511 		bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1512 
1513 #ifdef CONFIG_IPV6_SUBTREES
1514 		if (matched && saddr)
1515 			matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1516 #endif
1517 		if (matched)
1518 			return rt6_ex;
1519 	}
1520 	return NULL;
1521 }
1522 
1523 /* Helper function to find the cached rt in the hash table
1524  * and update bucket pointer to point to the bucket for this
1525  * (daddr, saddr) pair
1526  * Caller must hold rcu_read_lock()
1527  */
1528 static struct rt6_exception *
1529 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1530 			 const struct in6_addr *daddr,
1531 			 const struct in6_addr *saddr)
1532 {
1533 	struct rt6_exception *rt6_ex;
1534 	u32 hval;
1535 
1536 	WARN_ON_ONCE(!rcu_read_lock_held());
1537 
1538 	if (!(*bucket) || !daddr)
1539 		return NULL;
1540 
1541 	hval = rt6_exception_hash(daddr, saddr);
1542 	*bucket += hval;
1543 
1544 	hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1545 		struct rt6_info *rt6 = rt6_ex->rt6i;
1546 		bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1547 
1548 #ifdef CONFIG_IPV6_SUBTREES
1549 		if (matched && saddr)
1550 			matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1551 #endif
1552 		if (matched)
1553 			return rt6_ex;
1554 	}
1555 	return NULL;
1556 }
1557 
1558 static unsigned int fib6_mtu(const struct fib6_result *res)
1559 {
1560 	const struct fib6_nh *nh = res->nh;
1561 	unsigned int mtu;
1562 
1563 	if (res->f6i->fib6_pmtu) {
1564 		mtu = res->f6i->fib6_pmtu;
1565 	} else {
1566 		struct net_device *dev = nh->fib_nh_dev;
1567 		struct inet6_dev *idev;
1568 
1569 		rcu_read_lock();
1570 		idev = __in6_dev_get(dev);
1571 		mtu = idev->cnf.mtu6;
1572 		rcu_read_unlock();
1573 	}
1574 
1575 	mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1576 
1577 	return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1578 }
1579 
1580 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1581 
1582 /* used when the flushed bit is not relevant, only access to the bucket
1583  * (ie., all bucket users except rt6_insert_exception);
1584  *
1585  * called under rcu lock; sometimes called with rt6_exception_lock held
1586  */
1587 static
1588 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1589 						       spinlock_t *lock)
1590 {
1591 	struct rt6_exception_bucket *bucket;
1592 
1593 	if (lock)
1594 		bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1595 						   lockdep_is_held(lock));
1596 	else
1597 		bucket = rcu_dereference(nh->rt6i_exception_bucket);
1598 
1599 	/* remove bucket flushed bit if set */
1600 	if (bucket) {
1601 		unsigned long p = (unsigned long)bucket;
1602 
1603 		p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1604 		bucket = (struct rt6_exception_bucket *)p;
1605 	}
1606 
1607 	return bucket;
1608 }
1609 
1610 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1611 {
1612 	unsigned long p = (unsigned long)bucket;
1613 
1614 	return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1615 }
1616 
1617 /* called with rt6_exception_lock held */
1618 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1619 					      spinlock_t *lock)
1620 {
1621 	struct rt6_exception_bucket *bucket;
1622 	unsigned long p;
1623 
1624 	bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1625 					   lockdep_is_held(lock));
1626 
1627 	p = (unsigned long)bucket;
1628 	p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1629 	bucket = (struct rt6_exception_bucket *)p;
1630 	rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1631 }
1632 
1633 static int rt6_insert_exception(struct rt6_info *nrt,
1634 				const struct fib6_result *res)
1635 {
1636 	struct net *net = dev_net(nrt->dst.dev);
1637 	struct rt6_exception_bucket *bucket;
1638 	struct fib6_info *f6i = res->f6i;
1639 	struct in6_addr *src_key = NULL;
1640 	struct rt6_exception *rt6_ex;
1641 	struct fib6_nh *nh = res->nh;
1642 	int err = 0;
1643 
1644 	spin_lock_bh(&rt6_exception_lock);
1645 
1646 	bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1647 					  lockdep_is_held(&rt6_exception_lock));
1648 	if (!bucket) {
1649 		bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1650 				 GFP_ATOMIC);
1651 		if (!bucket) {
1652 			err = -ENOMEM;
1653 			goto out;
1654 		}
1655 		rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1656 	} else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1657 		err = -EINVAL;
1658 		goto out;
1659 	}
1660 
1661 #ifdef CONFIG_IPV6_SUBTREES
1662 	/* fib6_src.plen != 0 indicates f6i is in subtree
1663 	 * and exception table is indexed by a hash of
1664 	 * both fib6_dst and fib6_src.
1665 	 * Otherwise, the exception table is indexed by
1666 	 * a hash of only fib6_dst.
1667 	 */
1668 	if (f6i->fib6_src.plen)
1669 		src_key = &nrt->rt6i_src.addr;
1670 #endif
1671 	/* rt6_mtu_change() might lower mtu on f6i.
1672 	 * Only insert this exception route if its mtu
1673 	 * is less than f6i's mtu value.
1674 	 */
1675 	if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1676 		err = -EINVAL;
1677 		goto out;
1678 	}
1679 
1680 	rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1681 					       src_key);
1682 	if (rt6_ex)
1683 		rt6_remove_exception(bucket, rt6_ex);
1684 
1685 	rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1686 	if (!rt6_ex) {
1687 		err = -ENOMEM;
1688 		goto out;
1689 	}
1690 	rt6_ex->rt6i = nrt;
1691 	rt6_ex->stamp = jiffies;
1692 	hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1693 	bucket->depth++;
1694 	net->ipv6.rt6_stats->fib_rt_cache++;
1695 
1696 	if (bucket->depth > FIB6_MAX_DEPTH)
1697 		rt6_exception_remove_oldest(bucket);
1698 
1699 out:
1700 	spin_unlock_bh(&rt6_exception_lock);
1701 
1702 	/* Update fn->fn_sernum to invalidate all cached dst */
1703 	if (!err) {
1704 		spin_lock_bh(&f6i->fib6_table->tb6_lock);
1705 		fib6_update_sernum(net, f6i);
1706 		spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1707 		fib6_force_start_gc(net);
1708 	}
1709 
1710 	return err;
1711 }
1712 
1713 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1714 {
1715 	struct rt6_exception_bucket *bucket;
1716 	struct rt6_exception *rt6_ex;
1717 	struct hlist_node *tmp;
1718 	int i;
1719 
1720 	spin_lock_bh(&rt6_exception_lock);
1721 
1722 	bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1723 	if (!bucket)
1724 		goto out;
1725 
1726 	/* Prevent rt6_insert_exception() to recreate the bucket list */
1727 	if (!from)
1728 		fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1729 
1730 	for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1731 		hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1732 			if (!from ||
1733 			    rcu_access_pointer(rt6_ex->rt6i->from) == from)
1734 				rt6_remove_exception(bucket, rt6_ex);
1735 		}
1736 		WARN_ON_ONCE(!from && bucket->depth);
1737 		bucket++;
1738 	}
1739 out:
1740 	spin_unlock_bh(&rt6_exception_lock);
1741 }
1742 
1743 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1744 {
1745 	struct fib6_info *f6i = arg;
1746 
1747 	fib6_nh_flush_exceptions(nh, f6i);
1748 
1749 	return 0;
1750 }
1751 
1752 void rt6_flush_exceptions(struct fib6_info *f6i)
1753 {
1754 	if (f6i->nh)
1755 		nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1756 					 f6i);
1757 	else
1758 		fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1759 }
1760 
1761 /* Find cached rt in the hash table inside passed in rt
1762  * Caller has to hold rcu_read_lock()
1763  */
1764 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1765 					   const struct in6_addr *daddr,
1766 					   const struct in6_addr *saddr)
1767 {
1768 	const struct in6_addr *src_key = NULL;
1769 	struct rt6_exception_bucket *bucket;
1770 	struct rt6_exception *rt6_ex;
1771 	struct rt6_info *ret = NULL;
1772 
1773 #ifdef CONFIG_IPV6_SUBTREES
1774 	/* fib6i_src.plen != 0 indicates f6i is in subtree
1775 	 * and exception table is indexed by a hash of
1776 	 * both fib6_dst and fib6_src.
1777 	 * However, the src addr used to create the hash
1778 	 * might not be exactly the passed in saddr which
1779 	 * is a /128 addr from the flow.
1780 	 * So we need to use f6i->fib6_src to redo lookup
1781 	 * if the passed in saddr does not find anything.
1782 	 * (See the logic in ip6_rt_cache_alloc() on how
1783 	 * rt->rt6i_src is updated.)
1784 	 */
1785 	if (res->f6i->fib6_src.plen)
1786 		src_key = saddr;
1787 find_ex:
1788 #endif
1789 	bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1790 	rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1791 
1792 	if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1793 		ret = rt6_ex->rt6i;
1794 
1795 #ifdef CONFIG_IPV6_SUBTREES
1796 	/* Use fib6_src as src_key and redo lookup */
1797 	if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1798 		src_key = &res->f6i->fib6_src.addr;
1799 		goto find_ex;
1800 	}
1801 #endif
1802 
1803 	return ret;
1804 }
1805 
1806 /* Remove the passed in cached rt from the hash table that contains it */
1807 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1808 				    const struct rt6_info *rt)
1809 {
1810 	const struct in6_addr *src_key = NULL;
1811 	struct rt6_exception_bucket *bucket;
1812 	struct rt6_exception *rt6_ex;
1813 	int err;
1814 
1815 	if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1816 		return -ENOENT;
1817 
1818 	spin_lock_bh(&rt6_exception_lock);
1819 	bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1820 
1821 #ifdef CONFIG_IPV6_SUBTREES
1822 	/* rt6i_src.plen != 0 indicates 'from' is in subtree
1823 	 * and exception table is indexed by a hash of
1824 	 * both rt6i_dst and rt6i_src.
1825 	 * Otherwise, the exception table is indexed by
1826 	 * a hash of only rt6i_dst.
1827 	 */
1828 	if (plen)
1829 		src_key = &rt->rt6i_src.addr;
1830 #endif
1831 	rt6_ex = __rt6_find_exception_spinlock(&bucket,
1832 					       &rt->rt6i_dst.addr,
1833 					       src_key);
1834 	if (rt6_ex) {
1835 		rt6_remove_exception(bucket, rt6_ex);
1836 		err = 0;
1837 	} else {
1838 		err = -ENOENT;
1839 	}
1840 
1841 	spin_unlock_bh(&rt6_exception_lock);
1842 	return err;
1843 }
1844 
1845 struct fib6_nh_excptn_arg {
1846 	struct rt6_info	*rt;
1847 	int		plen;
1848 };
1849 
1850 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1851 {
1852 	struct fib6_nh_excptn_arg *arg = _arg;
1853 	int err;
1854 
1855 	err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1856 	if (err == 0)
1857 		return 1;
1858 
1859 	return 0;
1860 }
1861 
1862 static int rt6_remove_exception_rt(struct rt6_info *rt)
1863 {
1864 	struct fib6_info *from;
1865 
1866 	from = rcu_dereference(rt->from);
1867 	if (!from || !(rt->rt6i_flags & RTF_CACHE))
1868 		return -EINVAL;
1869 
1870 	if (from->nh) {
1871 		struct fib6_nh_excptn_arg arg = {
1872 			.rt = rt,
1873 			.plen = from->fib6_src.plen
1874 		};
1875 		int rc;
1876 
1877 		/* rc = 1 means an entry was found */
1878 		rc = nexthop_for_each_fib6_nh(from->nh,
1879 					      rt6_nh_remove_exception_rt,
1880 					      &arg);
1881 		return rc ? 0 : -ENOENT;
1882 	}
1883 
1884 	return fib6_nh_remove_exception(from->fib6_nh,
1885 					from->fib6_src.plen, rt);
1886 }
1887 
1888 /* Find rt6_ex which contains the passed in rt cache and
1889  * refresh its stamp
1890  */
1891 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1892 				     const struct rt6_info *rt)
1893 {
1894 	const struct in6_addr *src_key = NULL;
1895 	struct rt6_exception_bucket *bucket;
1896 	struct rt6_exception *rt6_ex;
1897 
1898 	bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1899 #ifdef CONFIG_IPV6_SUBTREES
1900 	/* rt6i_src.plen != 0 indicates 'from' is in subtree
1901 	 * and exception table is indexed by a hash of
1902 	 * both rt6i_dst and rt6i_src.
1903 	 * Otherwise, the exception table is indexed by
1904 	 * a hash of only rt6i_dst.
1905 	 */
1906 	if (plen)
1907 		src_key = &rt->rt6i_src.addr;
1908 #endif
1909 	rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1910 	if (rt6_ex)
1911 		rt6_ex->stamp = jiffies;
1912 }
1913 
1914 struct fib6_nh_match_arg {
1915 	const struct net_device *dev;
1916 	const struct in6_addr	*gw;
1917 	struct fib6_nh		*match;
1918 };
1919 
1920 /* determine if fib6_nh has given device and gateway */
1921 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1922 {
1923 	struct fib6_nh_match_arg *arg = _arg;
1924 
1925 	if (arg->dev != nh->fib_nh_dev ||
1926 	    (arg->gw && !nh->fib_nh_gw_family) ||
1927 	    (!arg->gw && nh->fib_nh_gw_family) ||
1928 	    (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1929 		return 0;
1930 
1931 	arg->match = nh;
1932 
1933 	/* found a match, break the loop */
1934 	return 1;
1935 }
1936 
1937 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1938 {
1939 	struct fib6_info *from;
1940 	struct fib6_nh *fib6_nh;
1941 
1942 	rcu_read_lock();
1943 
1944 	from = rcu_dereference(rt->from);
1945 	if (!from || !(rt->rt6i_flags & RTF_CACHE))
1946 		goto unlock;
1947 
1948 	if (from->nh) {
1949 		struct fib6_nh_match_arg arg = {
1950 			.dev = rt->dst.dev,
1951 			.gw = &rt->rt6i_gateway,
1952 		};
1953 
1954 		nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1955 
1956 		if (!arg.match)
1957 			goto unlock;
1958 		fib6_nh = arg.match;
1959 	} else {
1960 		fib6_nh = from->fib6_nh;
1961 	}
1962 	fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1963 unlock:
1964 	rcu_read_unlock();
1965 }
1966 
1967 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1968 					 struct rt6_info *rt, int mtu)
1969 {
1970 	/* If the new MTU is lower than the route PMTU, this new MTU will be the
1971 	 * lowest MTU in the path: always allow updating the route PMTU to
1972 	 * reflect PMTU decreases.
1973 	 *
1974 	 * If the new MTU is higher, and the route PMTU is equal to the local
1975 	 * MTU, this means the old MTU is the lowest in the path, so allow
1976 	 * updating it: if other nodes now have lower MTUs, PMTU discovery will
1977 	 * handle this.
1978 	 */
1979 
1980 	if (dst_mtu(&rt->dst) >= mtu)
1981 		return true;
1982 
1983 	if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
1984 		return true;
1985 
1986 	return false;
1987 }
1988 
1989 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
1990 				       const struct fib6_nh *nh, int mtu)
1991 {
1992 	struct rt6_exception_bucket *bucket;
1993 	struct rt6_exception *rt6_ex;
1994 	int i;
1995 
1996 	bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1997 	if (!bucket)
1998 		return;
1999 
2000 	for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2001 		hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2002 			struct rt6_info *entry = rt6_ex->rt6i;
2003 
2004 			/* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2005 			 * route), the metrics of its rt->from have already
2006 			 * been updated.
2007 			 */
2008 			if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2009 			    rt6_mtu_change_route_allowed(idev, entry, mtu))
2010 				dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2011 		}
2012 		bucket++;
2013 	}
2014 }
2015 
2016 #define RTF_CACHE_GATEWAY	(RTF_GATEWAY | RTF_CACHE)
2017 
2018 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2019 					    const struct in6_addr *gateway)
2020 {
2021 	struct rt6_exception_bucket *bucket;
2022 	struct rt6_exception *rt6_ex;
2023 	struct hlist_node *tmp;
2024 	int i;
2025 
2026 	if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2027 		return;
2028 
2029 	spin_lock_bh(&rt6_exception_lock);
2030 	bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2031 	if (bucket) {
2032 		for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2033 			hlist_for_each_entry_safe(rt6_ex, tmp,
2034 						  &bucket->chain, hlist) {
2035 				struct rt6_info *entry = rt6_ex->rt6i;
2036 
2037 				if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2038 				    RTF_CACHE_GATEWAY &&
2039 				    ipv6_addr_equal(gateway,
2040 						    &entry->rt6i_gateway)) {
2041 					rt6_remove_exception(bucket, rt6_ex);
2042 				}
2043 			}
2044 			bucket++;
2045 		}
2046 	}
2047 
2048 	spin_unlock_bh(&rt6_exception_lock);
2049 }
2050 
2051 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2052 				      struct rt6_exception *rt6_ex,
2053 				      struct fib6_gc_args *gc_args,
2054 				      unsigned long now)
2055 {
2056 	struct rt6_info *rt = rt6_ex->rt6i;
2057 
2058 	/* we are pruning and obsoleting aged-out and non gateway exceptions
2059 	 * even if others have still references to them, so that on next
2060 	 * dst_check() such references can be dropped.
2061 	 * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2062 	 * expired, independently from their aging, as per RFC 8201 section 4
2063 	 */
2064 	if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2065 		if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2066 			RT6_TRACE("aging clone %p\n", rt);
2067 			rt6_remove_exception(bucket, rt6_ex);
2068 			return;
2069 		}
2070 	} else if (time_after(jiffies, rt->dst.expires)) {
2071 		RT6_TRACE("purging expired route %p\n", rt);
2072 		rt6_remove_exception(bucket, rt6_ex);
2073 		return;
2074 	}
2075 
2076 	if (rt->rt6i_flags & RTF_GATEWAY) {
2077 		struct neighbour *neigh;
2078 		__u8 neigh_flags = 0;
2079 
2080 		neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2081 		if (neigh)
2082 			neigh_flags = neigh->flags;
2083 
2084 		if (!(neigh_flags & NTF_ROUTER)) {
2085 			RT6_TRACE("purging route %p via non-router but gateway\n",
2086 				  rt);
2087 			rt6_remove_exception(bucket, rt6_ex);
2088 			return;
2089 		}
2090 	}
2091 
2092 	gc_args->more++;
2093 }
2094 
2095 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2096 				   struct fib6_gc_args *gc_args,
2097 				   unsigned long now)
2098 {
2099 	struct rt6_exception_bucket *bucket;
2100 	struct rt6_exception *rt6_ex;
2101 	struct hlist_node *tmp;
2102 	int i;
2103 
2104 	if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2105 		return;
2106 
2107 	rcu_read_lock_bh();
2108 	spin_lock(&rt6_exception_lock);
2109 	bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2110 	if (bucket) {
2111 		for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2112 			hlist_for_each_entry_safe(rt6_ex, tmp,
2113 						  &bucket->chain, hlist) {
2114 				rt6_age_examine_exception(bucket, rt6_ex,
2115 							  gc_args, now);
2116 			}
2117 			bucket++;
2118 		}
2119 	}
2120 	spin_unlock(&rt6_exception_lock);
2121 	rcu_read_unlock_bh();
2122 }
2123 
2124 struct fib6_nh_age_excptn_arg {
2125 	struct fib6_gc_args	*gc_args;
2126 	unsigned long		now;
2127 };
2128 
2129 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2130 {
2131 	struct fib6_nh_age_excptn_arg *arg = _arg;
2132 
2133 	fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2134 	return 0;
2135 }
2136 
2137 void rt6_age_exceptions(struct fib6_info *f6i,
2138 			struct fib6_gc_args *gc_args,
2139 			unsigned long now)
2140 {
2141 	if (f6i->nh) {
2142 		struct fib6_nh_age_excptn_arg arg = {
2143 			.gc_args = gc_args,
2144 			.now = now
2145 		};
2146 
2147 		nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2148 					 &arg);
2149 	} else {
2150 		fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2151 	}
2152 }
2153 
2154 /* must be called with rcu lock held */
2155 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2156 		      struct flowi6 *fl6, struct fib6_result *res, int strict)
2157 {
2158 	struct fib6_node *fn, *saved_fn;
2159 
2160 	fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2161 	saved_fn = fn;
2162 
2163 	if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2164 		oif = 0;
2165 
2166 redo_rt6_select:
2167 	rt6_select(net, fn, oif, res, strict);
2168 	if (res->f6i == net->ipv6.fib6_null_entry) {
2169 		fn = fib6_backtrack(fn, &fl6->saddr);
2170 		if (fn)
2171 			goto redo_rt6_select;
2172 		else if (strict & RT6_LOOKUP_F_REACHABLE) {
2173 			/* also consider unreachable route */
2174 			strict &= ~RT6_LOOKUP_F_REACHABLE;
2175 			fn = saved_fn;
2176 			goto redo_rt6_select;
2177 		}
2178 	}
2179 
2180 	trace_fib6_table_lookup(net, res, table, fl6);
2181 
2182 	return 0;
2183 }
2184 
2185 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2186 			       int oif, struct flowi6 *fl6,
2187 			       const struct sk_buff *skb, int flags)
2188 {
2189 	struct fib6_result res = {};
2190 	struct rt6_info *rt = NULL;
2191 	int strict = 0;
2192 
2193 	WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2194 		     !rcu_read_lock_held());
2195 
2196 	strict |= flags & RT6_LOOKUP_F_IFACE;
2197 	strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2198 	if (net->ipv6.devconf_all->forwarding == 0)
2199 		strict |= RT6_LOOKUP_F_REACHABLE;
2200 
2201 	rcu_read_lock();
2202 
2203 	fib6_table_lookup(net, table, oif, fl6, &res, strict);
2204 	if (res.f6i == net->ipv6.fib6_null_entry)
2205 		goto out;
2206 
2207 	fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2208 
2209 	/*Search through exception table */
2210 	rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2211 	if (rt) {
2212 		goto out;
2213 	} else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2214 			    !res.nh->fib_nh_gw_family)) {
2215 		/* Create a RTF_CACHE clone which will not be
2216 		 * owned by the fib6 tree.  It is for the special case where
2217 		 * the daddr in the skb during the neighbor look-up is different
2218 		 * from the fl6->daddr used to look-up route here.
2219 		 */
2220 		rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2221 
2222 		if (rt) {
2223 			/* 1 refcnt is taken during ip6_rt_cache_alloc().
2224 			 * As rt6_uncached_list_add() does not consume refcnt,
2225 			 * this refcnt is always returned to the caller even
2226 			 * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2227 			 */
2228 			rt6_uncached_list_add(rt);
2229 			atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2230 			rcu_read_unlock();
2231 
2232 			return rt;
2233 		}
2234 	} else {
2235 		/* Get a percpu copy */
2236 		local_bh_disable();
2237 		rt = rt6_get_pcpu_route(&res);
2238 
2239 		if (!rt)
2240 			rt = rt6_make_pcpu_route(net, &res);
2241 
2242 		local_bh_enable();
2243 	}
2244 out:
2245 	if (!rt)
2246 		rt = net->ipv6.ip6_null_entry;
2247 	if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2248 		ip6_hold_safe(net, &rt);
2249 	rcu_read_unlock();
2250 
2251 	return rt;
2252 }
2253 EXPORT_SYMBOL_GPL(ip6_pol_route);
2254 
2255 static struct rt6_info *ip6_pol_route_input(struct net *net,
2256 					    struct fib6_table *table,
2257 					    struct flowi6 *fl6,
2258 					    const struct sk_buff *skb,
2259 					    int flags)
2260 {
2261 	return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2262 }
2263 
2264 struct dst_entry *ip6_route_input_lookup(struct net *net,
2265 					 struct net_device *dev,
2266 					 struct flowi6 *fl6,
2267 					 const struct sk_buff *skb,
2268 					 int flags)
2269 {
2270 	if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2271 		flags |= RT6_LOOKUP_F_IFACE;
2272 
2273 	return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2274 }
2275 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2276 
2277 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2278 				  struct flow_keys *keys,
2279 				  struct flow_keys *flkeys)
2280 {
2281 	const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2282 	const struct ipv6hdr *key_iph = outer_iph;
2283 	struct flow_keys *_flkeys = flkeys;
2284 	const struct ipv6hdr *inner_iph;
2285 	const struct icmp6hdr *icmph;
2286 	struct ipv6hdr _inner_iph;
2287 	struct icmp6hdr _icmph;
2288 
2289 	if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2290 		goto out;
2291 
2292 	icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2293 				   sizeof(_icmph), &_icmph);
2294 	if (!icmph)
2295 		goto out;
2296 
2297 	if (!icmpv6_is_err(icmph->icmp6_type))
2298 		goto out;
2299 
2300 	inner_iph = skb_header_pointer(skb,
2301 				       skb_transport_offset(skb) + sizeof(*icmph),
2302 				       sizeof(_inner_iph), &_inner_iph);
2303 	if (!inner_iph)
2304 		goto out;
2305 
2306 	key_iph = inner_iph;
2307 	_flkeys = NULL;
2308 out:
2309 	if (_flkeys) {
2310 		keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2311 		keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2312 		keys->tags.flow_label = _flkeys->tags.flow_label;
2313 		keys->basic.ip_proto = _flkeys->basic.ip_proto;
2314 	} else {
2315 		keys->addrs.v6addrs.src = key_iph->saddr;
2316 		keys->addrs.v6addrs.dst = key_iph->daddr;
2317 		keys->tags.flow_label = ip6_flowlabel(key_iph);
2318 		keys->basic.ip_proto = key_iph->nexthdr;
2319 	}
2320 }
2321 
2322 /* if skb is set it will be used and fl6 can be NULL */
2323 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2324 		       const struct sk_buff *skb, struct flow_keys *flkeys)
2325 {
2326 	struct flow_keys hash_keys;
2327 	u32 mhash;
2328 
2329 	switch (ip6_multipath_hash_policy(net)) {
2330 	case 0:
2331 		memset(&hash_keys, 0, sizeof(hash_keys));
2332 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2333 		if (skb) {
2334 			ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2335 		} else {
2336 			hash_keys.addrs.v6addrs.src = fl6->saddr;
2337 			hash_keys.addrs.v6addrs.dst = fl6->daddr;
2338 			hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2339 			hash_keys.basic.ip_proto = fl6->flowi6_proto;
2340 		}
2341 		break;
2342 	case 1:
2343 		if (skb) {
2344 			unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2345 			struct flow_keys keys;
2346 
2347 			/* short-circuit if we already have L4 hash present */
2348 			if (skb->l4_hash)
2349 				return skb_get_hash_raw(skb) >> 1;
2350 
2351 			memset(&hash_keys, 0, sizeof(hash_keys));
2352 
2353                         if (!flkeys) {
2354 				skb_flow_dissect_flow_keys(skb, &keys, flag);
2355 				flkeys = &keys;
2356 			}
2357 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2358 			hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2359 			hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2360 			hash_keys.ports.src = flkeys->ports.src;
2361 			hash_keys.ports.dst = flkeys->ports.dst;
2362 			hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2363 		} else {
2364 			memset(&hash_keys, 0, sizeof(hash_keys));
2365 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2366 			hash_keys.addrs.v6addrs.src = fl6->saddr;
2367 			hash_keys.addrs.v6addrs.dst = fl6->daddr;
2368 			hash_keys.ports.src = fl6->fl6_sport;
2369 			hash_keys.ports.dst = fl6->fl6_dport;
2370 			hash_keys.basic.ip_proto = fl6->flowi6_proto;
2371 		}
2372 		break;
2373 	case 2:
2374 		memset(&hash_keys, 0, sizeof(hash_keys));
2375 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2376 		if (skb) {
2377 			struct flow_keys keys;
2378 
2379 			if (!flkeys) {
2380 				skb_flow_dissect_flow_keys(skb, &keys, 0);
2381 				flkeys = &keys;
2382 			}
2383 
2384 			/* Inner can be v4 or v6 */
2385 			if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2386 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2387 				hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2388 				hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2389 			} else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2390 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2391 				hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2392 				hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2393 				hash_keys.tags.flow_label = flkeys->tags.flow_label;
2394 				hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2395 			} else {
2396 				/* Same as case 0 */
2397 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2398 				ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2399 			}
2400 		} else {
2401 			/* Same as case 0 */
2402 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2403 			hash_keys.addrs.v6addrs.src = fl6->saddr;
2404 			hash_keys.addrs.v6addrs.dst = fl6->daddr;
2405 			hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2406 			hash_keys.basic.ip_proto = fl6->flowi6_proto;
2407 		}
2408 		break;
2409 	}
2410 	mhash = flow_hash_from_keys(&hash_keys);
2411 
2412 	return mhash >> 1;
2413 }
2414 
2415 /* Called with rcu held */
2416 void ip6_route_input(struct sk_buff *skb)
2417 {
2418 	const struct ipv6hdr *iph = ipv6_hdr(skb);
2419 	struct net *net = dev_net(skb->dev);
2420 	int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2421 	struct ip_tunnel_info *tun_info;
2422 	struct flowi6 fl6 = {
2423 		.flowi6_iif = skb->dev->ifindex,
2424 		.daddr = iph->daddr,
2425 		.saddr = iph->saddr,
2426 		.flowlabel = ip6_flowinfo(iph),
2427 		.flowi6_mark = skb->mark,
2428 		.flowi6_proto = iph->nexthdr,
2429 	};
2430 	struct flow_keys *flkeys = NULL, _flkeys;
2431 
2432 	tun_info = skb_tunnel_info(skb);
2433 	if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2434 		fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2435 
2436 	if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2437 		flkeys = &_flkeys;
2438 
2439 	if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2440 		fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2441 	skb_dst_drop(skb);
2442 	skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2443 						      &fl6, skb, flags));
2444 }
2445 
2446 static struct rt6_info *ip6_pol_route_output(struct net *net,
2447 					     struct fib6_table *table,
2448 					     struct flowi6 *fl6,
2449 					     const struct sk_buff *skb,
2450 					     int flags)
2451 {
2452 	return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2453 }
2454 
2455 struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2456 					       const struct sock *sk,
2457 					       struct flowi6 *fl6, int flags)
2458 {
2459 	bool any_src;
2460 
2461 	if (ipv6_addr_type(&fl6->daddr) &
2462 	    (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2463 		struct dst_entry *dst;
2464 
2465 		/* This function does not take refcnt on the dst */
2466 		dst = l3mdev_link_scope_lookup(net, fl6);
2467 		if (dst)
2468 			return dst;
2469 	}
2470 
2471 	fl6->flowi6_iif = LOOPBACK_IFINDEX;
2472 
2473 	flags |= RT6_LOOKUP_F_DST_NOREF;
2474 	any_src = ipv6_addr_any(&fl6->saddr);
2475 	if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2476 	    (fl6->flowi6_oif && any_src))
2477 		flags |= RT6_LOOKUP_F_IFACE;
2478 
2479 	if (!any_src)
2480 		flags |= RT6_LOOKUP_F_HAS_SADDR;
2481 	else if (sk)
2482 		flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2483 
2484 	return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2485 }
2486 EXPORT_SYMBOL_GPL(ip6_route_output_flags_noref);
2487 
2488 struct dst_entry *ip6_route_output_flags(struct net *net,
2489 					 const struct sock *sk,
2490 					 struct flowi6 *fl6,
2491 					 int flags)
2492 {
2493         struct dst_entry *dst;
2494         struct rt6_info *rt6;
2495 
2496         rcu_read_lock();
2497         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2498         rt6 = (struct rt6_info *)dst;
2499         /* For dst cached in uncached_list, refcnt is already taken. */
2500         if (list_empty(&rt6->rt6i_uncached) && !dst_hold_safe(dst)) {
2501                 dst = &net->ipv6.ip6_null_entry->dst;
2502                 dst_hold(dst);
2503         }
2504         rcu_read_unlock();
2505 
2506         return dst;
2507 }
2508 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2509 
2510 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2511 {
2512 	struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2513 	struct net_device *loopback_dev = net->loopback_dev;
2514 	struct dst_entry *new = NULL;
2515 
2516 	rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2517 		       DST_OBSOLETE_DEAD, 0);
2518 	if (rt) {
2519 		rt6_info_init(rt);
2520 		atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2521 
2522 		new = &rt->dst;
2523 		new->__use = 1;
2524 		new->input = dst_discard;
2525 		new->output = dst_discard_out;
2526 
2527 		dst_copy_metrics(new, &ort->dst);
2528 
2529 		rt->rt6i_idev = in6_dev_get(loopback_dev);
2530 		rt->rt6i_gateway = ort->rt6i_gateway;
2531 		rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2532 
2533 		memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2534 #ifdef CONFIG_IPV6_SUBTREES
2535 		memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2536 #endif
2537 	}
2538 
2539 	dst_release(dst_orig);
2540 	return new ? new : ERR_PTR(-ENOMEM);
2541 }
2542 
2543 /*
2544  *	Destination cache support functions
2545  */
2546 
2547 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2548 {
2549 	u32 rt_cookie = 0;
2550 
2551 	if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2552 		return false;
2553 
2554 	if (fib6_check_expired(f6i))
2555 		return false;
2556 
2557 	return true;
2558 }
2559 
2560 static struct dst_entry *rt6_check(struct rt6_info *rt,
2561 				   struct fib6_info *from,
2562 				   u32 cookie)
2563 {
2564 	u32 rt_cookie = 0;
2565 
2566 	if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2567 	    rt_cookie != cookie)
2568 		return NULL;
2569 
2570 	if (rt6_check_expired(rt))
2571 		return NULL;
2572 
2573 	return &rt->dst;
2574 }
2575 
2576 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2577 					    struct fib6_info *from,
2578 					    u32 cookie)
2579 {
2580 	if (!__rt6_check_expired(rt) &&
2581 	    rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2582 	    fib6_check(from, cookie))
2583 		return &rt->dst;
2584 	else
2585 		return NULL;
2586 }
2587 
2588 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie)
2589 {
2590 	struct dst_entry *dst_ret;
2591 	struct fib6_info *from;
2592 	struct rt6_info *rt;
2593 
2594 	rt = container_of(dst, struct rt6_info, dst);
2595 
2596 	rcu_read_lock();
2597 
2598 	/* All IPV6 dsts are created with ->obsolete set to the value
2599 	 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2600 	 * into this function always.
2601 	 */
2602 
2603 	from = rcu_dereference(rt->from);
2604 
2605 	if (from && (rt->rt6i_flags & RTF_PCPU ||
2606 	    unlikely(!list_empty(&rt->rt6i_uncached))))
2607 		dst_ret = rt6_dst_from_check(rt, from, cookie);
2608 	else
2609 		dst_ret = rt6_check(rt, from, cookie);
2610 
2611 	rcu_read_unlock();
2612 
2613 	return dst_ret;
2614 }
2615 
2616 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2617 {
2618 	struct rt6_info *rt = (struct rt6_info *) dst;
2619 
2620 	if (rt) {
2621 		if (rt->rt6i_flags & RTF_CACHE) {
2622 			rcu_read_lock();
2623 			if (rt6_check_expired(rt)) {
2624 				rt6_remove_exception_rt(rt);
2625 				dst = NULL;
2626 			}
2627 			rcu_read_unlock();
2628 		} else {
2629 			dst_release(dst);
2630 			dst = NULL;
2631 		}
2632 	}
2633 	return dst;
2634 }
2635 
2636 static void ip6_link_failure(struct sk_buff *skb)
2637 {
2638 	struct rt6_info *rt;
2639 
2640 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2641 
2642 	rt = (struct rt6_info *) skb_dst(skb);
2643 	if (rt) {
2644 		rcu_read_lock();
2645 		if (rt->rt6i_flags & RTF_CACHE) {
2646 			rt6_remove_exception_rt(rt);
2647 		} else {
2648 			struct fib6_info *from;
2649 			struct fib6_node *fn;
2650 
2651 			from = rcu_dereference(rt->from);
2652 			if (from) {
2653 				fn = rcu_dereference(from->fib6_node);
2654 				if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2655 					fn->fn_sernum = -1;
2656 			}
2657 		}
2658 		rcu_read_unlock();
2659 	}
2660 }
2661 
2662 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2663 {
2664 	if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2665 		struct fib6_info *from;
2666 
2667 		rcu_read_lock();
2668 		from = rcu_dereference(rt0->from);
2669 		if (from)
2670 			rt0->dst.expires = from->expires;
2671 		rcu_read_unlock();
2672 	}
2673 
2674 	dst_set_expires(&rt0->dst, timeout);
2675 	rt0->rt6i_flags |= RTF_EXPIRES;
2676 }
2677 
2678 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2679 {
2680 	struct net *net = dev_net(rt->dst.dev);
2681 
2682 	dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2683 	rt->rt6i_flags |= RTF_MODIFIED;
2684 	rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2685 }
2686 
2687 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2688 {
2689 	return !(rt->rt6i_flags & RTF_CACHE) &&
2690 		(rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2691 }
2692 
2693 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2694 				 const struct ipv6hdr *iph, u32 mtu,
2695 				 bool confirm_neigh)
2696 {
2697 	const struct in6_addr *daddr, *saddr;
2698 	struct rt6_info *rt6 = (struct rt6_info *)dst;
2699 
2700 	if (dst_metric_locked(dst, RTAX_MTU))
2701 		return;
2702 
2703 	if (iph) {
2704 		daddr = &iph->daddr;
2705 		saddr = &iph->saddr;
2706 	} else if (sk) {
2707 		daddr = &sk->sk_v6_daddr;
2708 		saddr = &inet6_sk(sk)->saddr;
2709 	} else {
2710 		daddr = NULL;
2711 		saddr = NULL;
2712 	}
2713 
2714 	if (confirm_neigh)
2715 		dst_confirm_neigh(dst, daddr);
2716 
2717 	mtu = max_t(u32, mtu, IPV6_MIN_MTU);
2718 	if (mtu >= dst_mtu(dst))
2719 		return;
2720 
2721 	if (!rt6_cache_allowed_for_pmtu(rt6)) {
2722 		rt6_do_update_pmtu(rt6, mtu);
2723 		/* update rt6_ex->stamp for cache */
2724 		if (rt6->rt6i_flags & RTF_CACHE)
2725 			rt6_update_exception_stamp_rt(rt6);
2726 	} else if (daddr) {
2727 		struct fib6_result res = {};
2728 		struct rt6_info *nrt6;
2729 
2730 		rcu_read_lock();
2731 		res.f6i = rcu_dereference(rt6->from);
2732 		if (!res.f6i)
2733 			goto out_unlock;
2734 
2735 		res.fib6_flags = res.f6i->fib6_flags;
2736 		res.fib6_type = res.f6i->fib6_type;
2737 
2738 		if (res.f6i->nh) {
2739 			struct fib6_nh_match_arg arg = {
2740 				.dev = dst->dev,
2741 				.gw = &rt6->rt6i_gateway,
2742 			};
2743 
2744 			nexthop_for_each_fib6_nh(res.f6i->nh,
2745 						 fib6_nh_find_match, &arg);
2746 
2747 			/* fib6_info uses a nexthop that does not have fib6_nh
2748 			 * using the dst->dev + gw. Should be impossible.
2749 			 */
2750 			if (!arg.match)
2751 				goto out_unlock;
2752 
2753 			res.nh = arg.match;
2754 		} else {
2755 			res.nh = res.f6i->fib6_nh;
2756 		}
2757 
2758 		nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2759 		if (nrt6) {
2760 			rt6_do_update_pmtu(nrt6, mtu);
2761 			if (rt6_insert_exception(nrt6, &res))
2762 				dst_release_immediate(&nrt6->dst);
2763 		}
2764 out_unlock:
2765 		rcu_read_unlock();
2766 	}
2767 }
2768 
2769 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2770 			       struct sk_buff *skb, u32 mtu,
2771 			       bool confirm_neigh)
2772 {
2773 	__ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2774 			     confirm_neigh);
2775 }
2776 
2777 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2778 		     int oif, u32 mark, kuid_t uid)
2779 {
2780 	const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2781 	struct dst_entry *dst;
2782 	struct flowi6 fl6 = {
2783 		.flowi6_oif = oif,
2784 		.flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2785 		.daddr = iph->daddr,
2786 		.saddr = iph->saddr,
2787 		.flowlabel = ip6_flowinfo(iph),
2788 		.flowi6_uid = uid,
2789 	};
2790 
2791 	dst = ip6_route_output(net, NULL, &fl6);
2792 	if (!dst->error)
2793 		__ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2794 	dst_release(dst);
2795 }
2796 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2797 
2798 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2799 {
2800 	int oif = sk->sk_bound_dev_if;
2801 	struct dst_entry *dst;
2802 
2803 	if (!oif && skb->dev)
2804 		oif = l3mdev_master_ifindex(skb->dev);
2805 
2806 	ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, sk->sk_uid);
2807 
2808 	dst = __sk_dst_get(sk);
2809 	if (!dst || !dst->obsolete ||
2810 	    dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2811 		return;
2812 
2813 	bh_lock_sock(sk);
2814 	if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2815 		ip6_datagram_dst_update(sk, false);
2816 	bh_unlock_sock(sk);
2817 }
2818 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2819 
2820 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2821 			   const struct flowi6 *fl6)
2822 {
2823 #ifdef CONFIG_IPV6_SUBTREES
2824 	struct ipv6_pinfo *np = inet6_sk(sk);
2825 #endif
2826 
2827 	ip6_dst_store(sk, dst,
2828 		      ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2829 		      &sk->sk_v6_daddr : NULL,
2830 #ifdef CONFIG_IPV6_SUBTREES
2831 		      ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2832 		      &np->saddr :
2833 #endif
2834 		      NULL);
2835 }
2836 
2837 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2838 				  struct flowi6 *fl6,
2839 				  const struct in6_addr *gw,
2840 				  struct rt6_info **ret)
2841 {
2842 	const struct fib6_nh *nh = res->nh;
2843 
2844 	if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2845 	    fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2846 		return false;
2847 
2848 	/* rt_cache's gateway might be different from its 'parent'
2849 	 * in the case of an ip redirect.
2850 	 * So we keep searching in the exception table if the gateway
2851 	 * is different.
2852 	 */
2853 	if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
2854 		struct rt6_info *rt_cache;
2855 
2856 		rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
2857 		if (rt_cache &&
2858 		    ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
2859 			*ret = rt_cache;
2860 			return true;
2861 		}
2862 		return false;
2863 	}
2864 	return true;
2865 }
2866 
2867 struct fib6_nh_rd_arg {
2868 	struct fib6_result	*res;
2869 	struct flowi6		*fl6;
2870 	const struct in6_addr	*gw;
2871 	struct rt6_info		**ret;
2872 };
2873 
2874 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
2875 {
2876 	struct fib6_nh_rd_arg *arg = _arg;
2877 
2878 	arg->res->nh = nh;
2879 	return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
2880 }
2881 
2882 /* Handle redirects */
2883 struct ip6rd_flowi {
2884 	struct flowi6 fl6;
2885 	struct in6_addr gateway;
2886 };
2887 
2888 static struct rt6_info *__ip6_route_redirect(struct net *net,
2889 					     struct fib6_table *table,
2890 					     struct flowi6 *fl6,
2891 					     const struct sk_buff *skb,
2892 					     int flags)
2893 {
2894 	struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
2895 	struct rt6_info *ret = NULL;
2896 	struct fib6_result res = {};
2897 	struct fib6_nh_rd_arg arg = {
2898 		.res = &res,
2899 		.fl6 = fl6,
2900 		.gw  = &rdfl->gateway,
2901 		.ret = &ret
2902 	};
2903 	struct fib6_info *rt;
2904 	struct fib6_node *fn;
2905 
2906 	/* l3mdev_update_flow overrides oif if the device is enslaved; in
2907 	 * this case we must match on the real ingress device, so reset it
2908 	 */
2909 	if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2910 		fl6->flowi6_oif = skb->dev->ifindex;
2911 
2912 	/* Get the "current" route for this destination and
2913 	 * check if the redirect has come from appropriate router.
2914 	 *
2915 	 * RFC 4861 specifies that redirects should only be
2916 	 * accepted if they come from the nexthop to the target.
2917 	 * Due to the way the routes are chosen, this notion
2918 	 * is a bit fuzzy and one might need to check all possible
2919 	 * routes.
2920 	 */
2921 
2922 	rcu_read_lock();
2923 	fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2924 restart:
2925 	for_each_fib6_node_rt_rcu(fn) {
2926 		res.f6i = rt;
2927 		if (fib6_check_expired(rt))
2928 			continue;
2929 		if (rt->fib6_flags & RTF_REJECT)
2930 			break;
2931 		if (unlikely(rt->nh)) {
2932 			if (nexthop_is_blackhole(rt->nh))
2933 				continue;
2934 			/* on match, res->nh is filled in and potentially ret */
2935 			if (nexthop_for_each_fib6_nh(rt->nh,
2936 						     fib6_nh_redirect_match,
2937 						     &arg))
2938 				goto out;
2939 		} else {
2940 			res.nh = rt->fib6_nh;
2941 			if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
2942 						  &ret))
2943 				goto out;
2944 		}
2945 	}
2946 
2947 	if (!rt)
2948 		rt = net->ipv6.fib6_null_entry;
2949 	else if (rt->fib6_flags & RTF_REJECT) {
2950 		ret = net->ipv6.ip6_null_entry;
2951 		goto out;
2952 	}
2953 
2954 	if (rt == net->ipv6.fib6_null_entry) {
2955 		fn = fib6_backtrack(fn, &fl6->saddr);
2956 		if (fn)
2957 			goto restart;
2958 	}
2959 
2960 	res.f6i = rt;
2961 	res.nh = rt->fib6_nh;
2962 out:
2963 	if (ret) {
2964 		ip6_hold_safe(net, &ret);
2965 	} else {
2966 		res.fib6_flags = res.f6i->fib6_flags;
2967 		res.fib6_type = res.f6i->fib6_type;
2968 		ret = ip6_create_rt_rcu(&res);
2969 	}
2970 
2971 	rcu_read_unlock();
2972 
2973 	trace_fib6_table_lookup(net, &res, table, fl6);
2974 	return ret;
2975 };
2976 
2977 static struct dst_entry *ip6_route_redirect(struct net *net,
2978 					    const struct flowi6 *fl6,
2979 					    const struct sk_buff *skb,
2980 					    const struct in6_addr *gateway)
2981 {
2982 	int flags = RT6_LOOKUP_F_HAS_SADDR;
2983 	struct ip6rd_flowi rdfl;
2984 
2985 	rdfl.fl6 = *fl6;
2986 	rdfl.gateway = *gateway;
2987 
2988 	return fib6_rule_lookup(net, &rdfl.fl6, skb,
2989 				flags, __ip6_route_redirect);
2990 }
2991 
2992 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
2993 		  kuid_t uid)
2994 {
2995 	const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2996 	struct dst_entry *dst;
2997 	struct flowi6 fl6 = {
2998 		.flowi6_iif = LOOPBACK_IFINDEX,
2999 		.flowi6_oif = oif,
3000 		.flowi6_mark = mark,
3001 		.daddr = iph->daddr,
3002 		.saddr = iph->saddr,
3003 		.flowlabel = ip6_flowinfo(iph),
3004 		.flowi6_uid = uid,
3005 	};
3006 
3007 	dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3008 	rt6_do_redirect(dst, NULL, skb);
3009 	dst_release(dst);
3010 }
3011 EXPORT_SYMBOL_GPL(ip6_redirect);
3012 
3013 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3014 {
3015 	const struct ipv6hdr *iph = ipv6_hdr(skb);
3016 	const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3017 	struct dst_entry *dst;
3018 	struct flowi6 fl6 = {
3019 		.flowi6_iif = LOOPBACK_IFINDEX,
3020 		.flowi6_oif = oif,
3021 		.daddr = msg->dest,
3022 		.saddr = iph->daddr,
3023 		.flowi6_uid = sock_net_uid(net, NULL),
3024 	};
3025 
3026 	dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3027 	rt6_do_redirect(dst, NULL, skb);
3028 	dst_release(dst);
3029 }
3030 
3031 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3032 {
3033 	ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if, sk->sk_mark,
3034 		     sk->sk_uid);
3035 }
3036 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3037 
3038 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3039 {
3040 	struct net_device *dev = dst->dev;
3041 	unsigned int mtu = dst_mtu(dst);
3042 	struct net *net = dev_net(dev);
3043 
3044 	mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3045 
3046 	if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3047 		mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3048 
3049 	/*
3050 	 * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3051 	 * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3052 	 * IPV6_MAXPLEN is also valid and means: "any MSS,
3053 	 * rely only on pmtu discovery"
3054 	 */
3055 	if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3056 		mtu = IPV6_MAXPLEN;
3057 	return mtu;
3058 }
3059 
3060 static unsigned int ip6_mtu(const struct dst_entry *dst)
3061 {
3062 	struct inet6_dev *idev;
3063 	unsigned int mtu;
3064 
3065 	mtu = dst_metric_raw(dst, RTAX_MTU);
3066 	if (mtu)
3067 		goto out;
3068 
3069 	mtu = IPV6_MIN_MTU;
3070 
3071 	rcu_read_lock();
3072 	idev = __in6_dev_get(dst->dev);
3073 	if (idev)
3074 		mtu = idev->cnf.mtu6;
3075 	rcu_read_unlock();
3076 
3077 out:
3078 	mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3079 
3080 	return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
3081 }
3082 
3083 /* MTU selection:
3084  * 1. mtu on route is locked - use it
3085  * 2. mtu from nexthop exception
3086  * 3. mtu from egress device
3087  *
3088  * based on ip6_dst_mtu_forward and exception logic of
3089  * rt6_find_cached_rt; called with rcu_read_lock
3090  */
3091 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3092 		      const struct in6_addr *daddr,
3093 		      const struct in6_addr *saddr)
3094 {
3095 	const struct fib6_nh *nh = res->nh;
3096 	struct fib6_info *f6i = res->f6i;
3097 	struct inet6_dev *idev;
3098 	struct rt6_info *rt;
3099 	u32 mtu = 0;
3100 
3101 	if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3102 		mtu = f6i->fib6_pmtu;
3103 		if (mtu)
3104 			goto out;
3105 	}
3106 
3107 	rt = rt6_find_cached_rt(res, daddr, saddr);
3108 	if (unlikely(rt)) {
3109 		mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3110 	} else {
3111 		struct net_device *dev = nh->fib_nh_dev;
3112 
3113 		mtu = IPV6_MIN_MTU;
3114 		idev = __in6_dev_get(dev);
3115 		if (idev && idev->cnf.mtu6 > mtu)
3116 			mtu = idev->cnf.mtu6;
3117 	}
3118 
3119 	mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3120 out:
3121 	return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3122 }
3123 
3124 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3125 				  struct flowi6 *fl6)
3126 {
3127 	struct dst_entry *dst;
3128 	struct rt6_info *rt;
3129 	struct inet6_dev *idev = in6_dev_get(dev);
3130 	struct net *net = dev_net(dev);
3131 
3132 	if (unlikely(!idev))
3133 		return ERR_PTR(-ENODEV);
3134 
3135 	rt = ip6_dst_alloc(net, dev, 0);
3136 	if (unlikely(!rt)) {
3137 		in6_dev_put(idev);
3138 		dst = ERR_PTR(-ENOMEM);
3139 		goto out;
3140 	}
3141 
3142 	rt->dst.input = ip6_input;
3143 	rt->dst.output  = ip6_output;
3144 	rt->rt6i_gateway  = fl6->daddr;
3145 	rt->rt6i_dst.addr = fl6->daddr;
3146 	rt->rt6i_dst.plen = 128;
3147 	rt->rt6i_idev     = idev;
3148 	dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3149 
3150 	/* Add this dst into uncached_list so that rt6_disable_ip() can
3151 	 * do proper release of the net_device
3152 	 */
3153 	rt6_uncached_list_add(rt);
3154 	atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
3155 
3156 	dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3157 
3158 out:
3159 	return dst;
3160 }
3161 
3162 static int ip6_dst_gc(struct dst_ops *ops)
3163 {
3164 	struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3165 	int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3166 	int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;
3167 	int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3168 	int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3169 	unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3170 	int entries;
3171 
3172 	entries = dst_entries_get_fast(ops);
3173 	if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
3174 	    entries <= rt_max_size)
3175 		goto out;
3176 
3177 	net->ipv6.ip6_rt_gc_expire++;
3178 	fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net, true);
3179 	entries = dst_entries_get_slow(ops);
3180 	if (entries < ops->gc_thresh)
3181 		net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1;
3182 out:
3183 	net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity;
3184 	return entries > rt_max_size;
3185 }
3186 
3187 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3188 			       const struct in6_addr *gw_addr, u32 tbid,
3189 			       int flags, struct fib6_result *res)
3190 {
3191 	struct flowi6 fl6 = {
3192 		.flowi6_oif = cfg->fc_ifindex,
3193 		.daddr = *gw_addr,
3194 		.saddr = cfg->fc_prefsrc,
3195 	};
3196 	struct fib6_table *table;
3197 	int err;
3198 
3199 	table = fib6_get_table(net, tbid);
3200 	if (!table)
3201 		return -EINVAL;
3202 
3203 	if (!ipv6_addr_any(&cfg->fc_prefsrc))
3204 		flags |= RT6_LOOKUP_F_HAS_SADDR;
3205 
3206 	flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3207 
3208 	err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3209 	if (!err && res->f6i != net->ipv6.fib6_null_entry)
3210 		fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3211 				 cfg->fc_ifindex != 0, NULL, flags);
3212 
3213 	return err;
3214 }
3215 
3216 static int ip6_route_check_nh_onlink(struct net *net,
3217 				     struct fib6_config *cfg,
3218 				     const struct net_device *dev,
3219 				     struct netlink_ext_ack *extack)
3220 {
3221 	u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3222 	const struct in6_addr *gw_addr = &cfg->fc_gateway;
3223 	struct fib6_result res = {};
3224 	int err;
3225 
3226 	err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3227 	if (!err && !(res.fib6_flags & RTF_REJECT) &&
3228 	    /* ignore match if it is the default route */
3229 	    !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3230 	    (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3231 		NL_SET_ERR_MSG(extack,
3232 			       "Nexthop has invalid gateway or device mismatch");
3233 		err = -EINVAL;
3234 	}
3235 
3236 	return err;
3237 }
3238 
3239 static int ip6_route_check_nh(struct net *net,
3240 			      struct fib6_config *cfg,
3241 			      struct net_device **_dev,
3242 			      struct inet6_dev **idev)
3243 {
3244 	const struct in6_addr *gw_addr = &cfg->fc_gateway;
3245 	struct net_device *dev = _dev ? *_dev : NULL;
3246 	int flags = RT6_LOOKUP_F_IFACE;
3247 	struct fib6_result res = {};
3248 	int err = -EHOSTUNREACH;
3249 
3250 	if (cfg->fc_table) {
3251 		err = ip6_nh_lookup_table(net, cfg, gw_addr,
3252 					  cfg->fc_table, flags, &res);
3253 		/* gw_addr can not require a gateway or resolve to a reject
3254 		 * route. If a device is given, it must match the result.
3255 		 */
3256 		if (err || res.fib6_flags & RTF_REJECT ||
3257 		    res.nh->fib_nh_gw_family ||
3258 		    (dev && dev != res.nh->fib_nh_dev))
3259 			err = -EHOSTUNREACH;
3260 	}
3261 
3262 	if (err < 0) {
3263 		struct flowi6 fl6 = {
3264 			.flowi6_oif = cfg->fc_ifindex,
3265 			.daddr = *gw_addr,
3266 		};
3267 
3268 		err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3269 		if (err || res.fib6_flags & RTF_REJECT ||
3270 		    res.nh->fib_nh_gw_family)
3271 			err = -EHOSTUNREACH;
3272 
3273 		if (err)
3274 			return err;
3275 
3276 		fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3277 				 cfg->fc_ifindex != 0, NULL, flags);
3278 	}
3279 
3280 	err = 0;
3281 	if (dev) {
3282 		if (dev != res.nh->fib_nh_dev)
3283 			err = -EHOSTUNREACH;
3284 	} else {
3285 		*_dev = dev = res.nh->fib_nh_dev;
3286 		dev_hold(dev);
3287 		*idev = in6_dev_get(dev);
3288 	}
3289 
3290 	return err;
3291 }
3292 
3293 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3294 			   struct net_device **_dev, struct inet6_dev **idev,
3295 			   struct netlink_ext_ack *extack)
3296 {
3297 	const struct in6_addr *gw_addr = &cfg->fc_gateway;
3298 	int gwa_type = ipv6_addr_type(gw_addr);
3299 	bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3300 	const struct net_device *dev = *_dev;
3301 	bool need_addr_check = !dev;
3302 	int err = -EINVAL;
3303 
3304 	/* if gw_addr is local we will fail to detect this in case
3305 	 * address is still TENTATIVE (DAD in progress). rt6_lookup()
3306 	 * will return already-added prefix route via interface that
3307 	 * prefix route was assigned to, which might be non-loopback.
3308 	 */
3309 	if (dev &&
3310 	    ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3311 		NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3312 		goto out;
3313 	}
3314 
3315 	if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3316 		/* IPv6 strictly inhibits using not link-local
3317 		 * addresses as nexthop address.
3318 		 * Otherwise, router will not able to send redirects.
3319 		 * It is very good, but in some (rare!) circumstances
3320 		 * (SIT, PtP, NBMA NOARP links) it is handy to allow
3321 		 * some exceptions. --ANK
3322 		 * We allow IPv4-mapped nexthops to support RFC4798-type
3323 		 * addressing
3324 		 */
3325 		if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3326 			NL_SET_ERR_MSG(extack, "Invalid gateway address");
3327 			goto out;
3328 		}
3329 
3330 		rcu_read_lock();
3331 
3332 		if (cfg->fc_flags & RTNH_F_ONLINK)
3333 			err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3334 		else
3335 			err = ip6_route_check_nh(net, cfg, _dev, idev);
3336 
3337 		rcu_read_unlock();
3338 
3339 		if (err)
3340 			goto out;
3341 	}
3342 
3343 	/* reload in case device was changed */
3344 	dev = *_dev;
3345 
3346 	err = -EINVAL;
3347 	if (!dev) {
3348 		NL_SET_ERR_MSG(extack, "Egress device not specified");
3349 		goto out;
3350 	} else if (dev->flags & IFF_LOOPBACK) {
3351 		NL_SET_ERR_MSG(extack,
3352 			       "Egress device can not be loopback device for this route");
3353 		goto out;
3354 	}
3355 
3356 	/* if we did not check gw_addr above, do so now that the
3357 	 * egress device has been resolved.
3358 	 */
3359 	if (need_addr_check &&
3360 	    ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3361 		NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3362 		goto out;
3363 	}
3364 
3365 	err = 0;
3366 out:
3367 	return err;
3368 }
3369 
3370 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3371 {
3372 	if ((flags & RTF_REJECT) ||
3373 	    (dev && (dev->flags & IFF_LOOPBACK) &&
3374 	     !(addr_type & IPV6_ADDR_LOOPBACK) &&
3375 	     !(flags & RTF_LOCAL)))
3376 		return true;
3377 
3378 	return false;
3379 }
3380 
3381 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3382 		 struct fib6_config *cfg, gfp_t gfp_flags,
3383 		 struct netlink_ext_ack *extack)
3384 {
3385 	struct net_device *dev = NULL;
3386 	struct inet6_dev *idev = NULL;
3387 	int addr_type;
3388 	int err;
3389 
3390 	fib6_nh->fib_nh_family = AF_INET6;
3391 #ifdef CONFIG_IPV6_ROUTER_PREF
3392 	fib6_nh->last_probe = jiffies;
3393 #endif
3394 
3395 	err = -ENODEV;
3396 	if (cfg->fc_ifindex) {
3397 		dev = dev_get_by_index(net, cfg->fc_ifindex);
3398 		if (!dev)
3399 			goto out;
3400 		idev = in6_dev_get(dev);
3401 		if (!idev)
3402 			goto out;
3403 	}
3404 
3405 	if (cfg->fc_flags & RTNH_F_ONLINK) {
3406 		if (!dev) {
3407 			NL_SET_ERR_MSG(extack,
3408 				       "Nexthop device required for onlink");
3409 			goto out;
3410 		}
3411 
3412 		if (!(dev->flags & IFF_UP)) {
3413 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3414 			err = -ENETDOWN;
3415 			goto out;
3416 		}
3417 
3418 		fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3419 	}
3420 
3421 	fib6_nh->fib_nh_weight = 1;
3422 
3423 	/* We cannot add true routes via loopback here,
3424 	 * they would result in kernel looping; promote them to reject routes
3425 	 */
3426 	addr_type = ipv6_addr_type(&cfg->fc_dst);
3427 	if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3428 		/* hold loopback dev/idev if we haven't done so. */
3429 		if (dev != net->loopback_dev) {
3430 			if (dev) {
3431 				dev_put(dev);
3432 				in6_dev_put(idev);
3433 			}
3434 			dev = net->loopback_dev;
3435 			dev_hold(dev);
3436 			idev = in6_dev_get(dev);
3437 			if (!idev) {
3438 				err = -ENODEV;
3439 				goto out;
3440 			}
3441 		}
3442 		goto pcpu_alloc;
3443 	}
3444 
3445 	if (cfg->fc_flags & RTF_GATEWAY) {
3446 		err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3447 		if (err)
3448 			goto out;
3449 
3450 		fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3451 		fib6_nh->fib_nh_gw_family = AF_INET6;
3452 	}
3453 
3454 	err = -ENODEV;
3455 	if (!dev)
3456 		goto out;
3457 
3458 	if (idev->cnf.disable_ipv6) {
3459 		NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3460 		err = -EACCES;
3461 		goto out;
3462 	}
3463 
3464 	if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3465 		NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3466 		err = -ENETDOWN;
3467 		goto out;
3468 	}
3469 
3470 	if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3471 	    !netif_carrier_ok(dev))
3472 		fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3473 
3474 	err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3475 				 cfg->fc_encap_type, cfg, gfp_flags, extack);
3476 	if (err)
3477 		goto out;
3478 
3479 pcpu_alloc:
3480 	fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3481 	if (!fib6_nh->rt6i_pcpu) {
3482 		err = -ENOMEM;
3483 		goto out;
3484 	}
3485 
3486 	fib6_nh->fib_nh_dev = dev;
3487 	fib6_nh->fib_nh_oif = dev->ifindex;
3488 	err = 0;
3489 out:
3490 	if (idev)
3491 		in6_dev_put(idev);
3492 
3493 	if (err) {
3494 		lwtstate_put(fib6_nh->fib_nh_lws);
3495 		fib6_nh->fib_nh_lws = NULL;
3496 		if (dev)
3497 			dev_put(dev);
3498 	}
3499 
3500 	return err;
3501 }
3502 
3503 void fib6_nh_release(struct fib6_nh *fib6_nh)
3504 {
3505 	struct rt6_exception_bucket *bucket;
3506 
3507 	rcu_read_lock();
3508 
3509 	fib6_nh_flush_exceptions(fib6_nh, NULL);
3510 	bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3511 	if (bucket) {
3512 		rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3513 		kfree(bucket);
3514 	}
3515 
3516 	rcu_read_unlock();
3517 
3518 	if (fib6_nh->rt6i_pcpu) {
3519 		int cpu;
3520 
3521 		for_each_possible_cpu(cpu) {
3522 			struct rt6_info **ppcpu_rt;
3523 			struct rt6_info *pcpu_rt;
3524 
3525 			ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3526 			pcpu_rt = *ppcpu_rt;
3527 			if (pcpu_rt) {
3528 				dst_dev_put(&pcpu_rt->dst);
3529 				dst_release(&pcpu_rt->dst);
3530 				*ppcpu_rt = NULL;
3531 			}
3532 		}
3533 
3534 		free_percpu(fib6_nh->rt6i_pcpu);
3535 	}
3536 
3537 	fib_nh_common_release(&fib6_nh->nh_common);
3538 }
3539 
3540 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3541 					      gfp_t gfp_flags,
3542 					      struct netlink_ext_ack *extack)
3543 {
3544 	struct net *net = cfg->fc_nlinfo.nl_net;
3545 	struct fib6_info *rt = NULL;
3546 	struct nexthop *nh = NULL;
3547 	struct fib6_table *table;
3548 	struct fib6_nh *fib6_nh;
3549 	int err = -EINVAL;
3550 	int addr_type;
3551 
3552 	/* RTF_PCPU is an internal flag; can not be set by userspace */
3553 	if (cfg->fc_flags & RTF_PCPU) {
3554 		NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3555 		goto out;
3556 	}
3557 
3558 	/* RTF_CACHE is an internal flag; can not be set by userspace */
3559 	if (cfg->fc_flags & RTF_CACHE) {
3560 		NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3561 		goto out;
3562 	}
3563 
3564 	if (cfg->fc_type > RTN_MAX) {
3565 		NL_SET_ERR_MSG(extack, "Invalid route type");
3566 		goto out;
3567 	}
3568 
3569 	if (cfg->fc_dst_len > 128) {
3570 		NL_SET_ERR_MSG(extack, "Invalid prefix length");
3571 		goto out;
3572 	}
3573 	if (cfg->fc_src_len > 128) {
3574 		NL_SET_ERR_MSG(extack, "Invalid source address length");
3575 		goto out;
3576 	}
3577 #ifndef CONFIG_IPV6_SUBTREES
3578 	if (cfg->fc_src_len) {
3579 		NL_SET_ERR_MSG(extack,
3580 			       "Specifying source address requires IPV6_SUBTREES to be enabled");
3581 		goto out;
3582 	}
3583 #endif
3584 	if (cfg->fc_nh_id) {
3585 		nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3586 		if (!nh) {
3587 			NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3588 			goto out;
3589 		}
3590 		err = fib6_check_nexthop(nh, cfg, extack);
3591 		if (err)
3592 			goto out;
3593 	}
3594 
3595 	err = -ENOBUFS;
3596 	if (cfg->fc_nlinfo.nlh &&
3597 	    !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3598 		table = fib6_get_table(net, cfg->fc_table);
3599 		if (!table) {
3600 			pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3601 			table = fib6_new_table(net, cfg->fc_table);
3602 		}
3603 	} else {
3604 		table = fib6_new_table(net, cfg->fc_table);
3605 	}
3606 
3607 	if (!table)
3608 		goto out;
3609 
3610 	err = -ENOMEM;
3611 	rt = fib6_info_alloc(gfp_flags, !nh);
3612 	if (!rt)
3613 		goto out;
3614 
3615 	rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3616 					       extack);
3617 	if (IS_ERR(rt->fib6_metrics)) {
3618 		err = PTR_ERR(rt->fib6_metrics);
3619 		/* Do not leave garbage there. */
3620 		rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3621 		goto out;
3622 	}
3623 
3624 	if (cfg->fc_flags & RTF_ADDRCONF)
3625 		rt->dst_nocount = true;
3626 
3627 	if (cfg->fc_flags & RTF_EXPIRES)
3628 		fib6_set_expires(rt, jiffies +
3629 				clock_t_to_jiffies(cfg->fc_expires));
3630 	else
3631 		fib6_clean_expires(rt);
3632 
3633 	if (cfg->fc_protocol == RTPROT_UNSPEC)
3634 		cfg->fc_protocol = RTPROT_BOOT;
3635 	rt->fib6_protocol = cfg->fc_protocol;
3636 
3637 	rt->fib6_table = table;
3638 	rt->fib6_metric = cfg->fc_metric;
3639 	rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3640 	rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3641 
3642 	ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3643 	rt->fib6_dst.plen = cfg->fc_dst_len;
3644 
3645 #ifdef CONFIG_IPV6_SUBTREES
3646 	ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3647 	rt->fib6_src.plen = cfg->fc_src_len;
3648 #endif
3649 	if (nh) {
3650 		if (!nexthop_get(nh)) {
3651 			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3652 			goto out;
3653 		}
3654 		if (rt->fib6_src.plen) {
3655 			NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3656 			goto out;
3657 		}
3658 		rt->nh = nh;
3659 		fib6_nh = nexthop_fib6_nh(rt->nh);
3660 	} else {
3661 		err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3662 		if (err)
3663 			goto out;
3664 
3665 		fib6_nh = rt->fib6_nh;
3666 
3667 		/* We cannot add true routes via loopback here, they would
3668 		 * result in kernel looping; promote them to reject routes
3669 		 */
3670 		addr_type = ipv6_addr_type(&cfg->fc_dst);
3671 		if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3672 				   addr_type))
3673 			rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3674 	}
3675 
3676 	if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3677 		struct net_device *dev = fib6_nh->fib_nh_dev;
3678 
3679 		if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3680 			NL_SET_ERR_MSG(extack, "Invalid source address");
3681 			err = -EINVAL;
3682 			goto out;
3683 		}
3684 		rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3685 		rt->fib6_prefsrc.plen = 128;
3686 	} else
3687 		rt->fib6_prefsrc.plen = 0;
3688 
3689 	return rt;
3690 out:
3691 	fib6_info_release(rt);
3692 	return ERR_PTR(err);
3693 }
3694 
3695 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3696 		  struct netlink_ext_ack *extack)
3697 {
3698 	struct fib6_info *rt;
3699 	int err;
3700 
3701 	rt = ip6_route_info_create(cfg, gfp_flags, extack);
3702 	if (IS_ERR(rt))
3703 		return PTR_ERR(rt);
3704 
3705 	err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3706 	fib6_info_release(rt);
3707 
3708 	return err;
3709 }
3710 
3711 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3712 {
3713 	struct net *net = info->nl_net;
3714 	struct fib6_table *table;
3715 	int err;
3716 
3717 	if (rt == net->ipv6.fib6_null_entry) {
3718 		err = -ENOENT;
3719 		goto out;
3720 	}
3721 
3722 	table = rt->fib6_table;
3723 	spin_lock_bh(&table->tb6_lock);
3724 	err = fib6_del(rt, info);
3725 	spin_unlock_bh(&table->tb6_lock);
3726 
3727 out:
3728 	fib6_info_release(rt);
3729 	return err;
3730 }
3731 
3732 int ip6_del_rt(struct net *net, struct fib6_info *rt)
3733 {
3734 	struct nl_info info = { .nl_net = net };
3735 
3736 	return __ip6_del_rt(rt, &info);
3737 }
3738 
3739 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3740 {
3741 	struct nl_info *info = &cfg->fc_nlinfo;
3742 	struct net *net = info->nl_net;
3743 	struct sk_buff *skb = NULL;
3744 	struct fib6_table *table;
3745 	int err = -ENOENT;
3746 
3747 	if (rt == net->ipv6.fib6_null_entry)
3748 		goto out_put;
3749 	table = rt->fib6_table;
3750 	spin_lock_bh(&table->tb6_lock);
3751 
3752 	if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3753 		struct fib6_info *sibling, *next_sibling;
3754 		struct fib6_node *fn;
3755 
3756 		/* prefer to send a single notification with all hops */
3757 		skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3758 		if (skb) {
3759 			u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3760 
3761 			if (rt6_fill_node(net, skb, rt, NULL,
3762 					  NULL, NULL, 0, RTM_DELROUTE,
3763 					  info->portid, seq, 0) < 0) {
3764 				kfree_skb(skb);
3765 				skb = NULL;
3766 			} else
3767 				info->skip_notify = 1;
3768 		}
3769 
3770 		/* 'rt' points to the first sibling route. If it is not the
3771 		 * leaf, then we do not need to send a notification. Otherwise,
3772 		 * we need to check if the last sibling has a next route or not
3773 		 * and emit a replace or delete notification, respectively.
3774 		 */
3775 		info->skip_notify_kernel = 1;
3776 		fn = rcu_dereference_protected(rt->fib6_node,
3777 					    lockdep_is_held(&table->tb6_lock));
3778 		if (rcu_access_pointer(fn->leaf) == rt) {
3779 			struct fib6_info *last_sibling, *replace_rt;
3780 
3781 			last_sibling = list_last_entry(&rt->fib6_siblings,
3782 						       struct fib6_info,
3783 						       fib6_siblings);
3784 			replace_rt = rcu_dereference_protected(
3785 					    last_sibling->fib6_next,
3786 					    lockdep_is_held(&table->tb6_lock));
3787 			if (replace_rt)
3788 				call_fib6_entry_notifiers_replace(net,
3789 								  replace_rt);
3790 			else
3791 				call_fib6_multipath_entry_notifiers(net,
3792 						       FIB_EVENT_ENTRY_DEL,
3793 						       rt, rt->fib6_nsiblings,
3794 						       NULL);
3795 		}
3796 		list_for_each_entry_safe(sibling, next_sibling,
3797 					 &rt->fib6_siblings,
3798 					 fib6_siblings) {
3799 			err = fib6_del(sibling, info);
3800 			if (err)
3801 				goto out_unlock;
3802 		}
3803 	}
3804 
3805 	err = fib6_del(rt, info);
3806 out_unlock:
3807 	spin_unlock_bh(&table->tb6_lock);
3808 out_put:
3809 	fib6_info_release(rt);
3810 
3811 	if (skb) {
3812 		rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3813 			    info->nlh, gfp_any());
3814 	}
3815 	return err;
3816 }
3817 
3818 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3819 {
3820 	int rc = -ESRCH;
3821 
3822 	if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3823 		goto out;
3824 
3825 	if (cfg->fc_flags & RTF_GATEWAY &&
3826 	    !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3827 		goto out;
3828 
3829 	rc = rt6_remove_exception_rt(rt);
3830 out:
3831 	return rc;
3832 }
3833 
3834 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3835 			     struct fib6_nh *nh)
3836 {
3837 	struct fib6_result res = {
3838 		.f6i = rt,
3839 		.nh = nh,
3840 	};
3841 	struct rt6_info *rt_cache;
3842 
3843 	rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3844 	if (rt_cache)
3845 		return __ip6_del_cached_rt(rt_cache, cfg);
3846 
3847 	return 0;
3848 }
3849 
3850 struct fib6_nh_del_cached_rt_arg {
3851 	struct fib6_config *cfg;
3852 	struct fib6_info *f6i;
3853 };
3854 
3855 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
3856 {
3857 	struct fib6_nh_del_cached_rt_arg *arg = _arg;
3858 	int rc;
3859 
3860 	rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
3861 	return rc != -ESRCH ? rc : 0;
3862 }
3863 
3864 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
3865 {
3866 	struct fib6_nh_del_cached_rt_arg arg = {
3867 		.cfg = cfg,
3868 		.f6i = f6i
3869 	};
3870 
3871 	return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
3872 }
3873 
3874 static int ip6_route_del(struct fib6_config *cfg,
3875 			 struct netlink_ext_ack *extack)
3876 {
3877 	struct fib6_table *table;
3878 	struct fib6_info *rt;
3879 	struct fib6_node *fn;
3880 	int err = -ESRCH;
3881 
3882 	table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
3883 	if (!table) {
3884 		NL_SET_ERR_MSG(extack, "FIB table does not exist");
3885 		return err;
3886 	}
3887 
3888 	rcu_read_lock();
3889 
3890 	fn = fib6_locate(&table->tb6_root,
3891 			 &cfg->fc_dst, cfg->fc_dst_len,
3892 			 &cfg->fc_src, cfg->fc_src_len,
3893 			 !(cfg->fc_flags & RTF_CACHE));
3894 
3895 	if (fn) {
3896 		for_each_fib6_node_rt_rcu(fn) {
3897 			struct fib6_nh *nh;
3898 
3899 			if (rt->nh && cfg->fc_nh_id &&
3900 			    rt->nh->id != cfg->fc_nh_id)
3901 				continue;
3902 
3903 			if (cfg->fc_flags & RTF_CACHE) {
3904 				int rc = 0;
3905 
3906 				if (rt->nh) {
3907 					rc = ip6_del_cached_rt_nh(cfg, rt);
3908 				} else if (cfg->fc_nh_id) {
3909 					continue;
3910 				} else {
3911 					nh = rt->fib6_nh;
3912 					rc = ip6_del_cached_rt(cfg, rt, nh);
3913 				}
3914 				if (rc != -ESRCH) {
3915 					rcu_read_unlock();
3916 					return rc;
3917 				}
3918 				continue;
3919 			}
3920 
3921 			if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
3922 				continue;
3923 			if (cfg->fc_protocol &&
3924 			    cfg->fc_protocol != rt->fib6_protocol)
3925 				continue;
3926 
3927 			if (rt->nh) {
3928 				if (!fib6_info_hold_safe(rt))
3929 					continue;
3930 				rcu_read_unlock();
3931 
3932 				return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3933 			}
3934 			if (cfg->fc_nh_id)
3935 				continue;
3936 
3937 			nh = rt->fib6_nh;
3938 			if (cfg->fc_ifindex &&
3939 			    (!nh->fib_nh_dev ||
3940 			     nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
3941 				continue;
3942 			if (cfg->fc_flags & RTF_GATEWAY &&
3943 			    !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
3944 				continue;
3945 			if (!fib6_info_hold_safe(rt))
3946 				continue;
3947 			rcu_read_unlock();
3948 
3949 			/* if gateway was specified only delete the one hop */
3950 			if (cfg->fc_flags & RTF_GATEWAY)
3951 				return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3952 
3953 			return __ip6_del_rt_siblings(rt, cfg);
3954 		}
3955 	}
3956 	rcu_read_unlock();
3957 
3958 	return err;
3959 }
3960 
3961 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
3962 {
3963 	struct netevent_redirect netevent;
3964 	struct rt6_info *rt, *nrt = NULL;
3965 	struct fib6_result res = {};
3966 	struct ndisc_options ndopts;
3967 	struct inet6_dev *in6_dev;
3968 	struct neighbour *neigh;
3969 	struct rd_msg *msg;
3970 	int optlen, on_link;
3971 	u8 *lladdr;
3972 
3973 	optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
3974 	optlen -= sizeof(*msg);
3975 
3976 	if (optlen < 0) {
3977 		net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
3978 		return;
3979 	}
3980 
3981 	msg = (struct rd_msg *)icmp6_hdr(skb);
3982 
3983 	if (ipv6_addr_is_multicast(&msg->dest)) {
3984 		net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
3985 		return;
3986 	}
3987 
3988 	on_link = 0;
3989 	if (ipv6_addr_equal(&msg->dest, &msg->target)) {
3990 		on_link = 1;
3991 	} else if (ipv6_addr_type(&msg->target) !=
3992 		   (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
3993 		net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
3994 		return;
3995 	}
3996 
3997 	in6_dev = __in6_dev_get(skb->dev);
3998 	if (!in6_dev)
3999 		return;
4000 	if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
4001 		return;
4002 
4003 	/* RFC2461 8.1:
4004 	 *	The IP source address of the Redirect MUST be the same as the current
4005 	 *	first-hop router for the specified ICMP Destination Address.
4006 	 */
4007 
4008 	if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4009 		net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4010 		return;
4011 	}
4012 
4013 	lladdr = NULL;
4014 	if (ndopts.nd_opts_tgt_lladdr) {
4015 		lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4016 					     skb->dev);
4017 		if (!lladdr) {
4018 			net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4019 			return;
4020 		}
4021 	}
4022 
4023 	rt = (struct rt6_info *) dst;
4024 	if (rt->rt6i_flags & RTF_REJECT) {
4025 		net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4026 		return;
4027 	}
4028 
4029 	/* Redirect received -> path was valid.
4030 	 * Look, redirects are sent only in response to data packets,
4031 	 * so that this nexthop apparently is reachable. --ANK
4032 	 */
4033 	dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4034 
4035 	neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4036 	if (!neigh)
4037 		return;
4038 
4039 	/*
4040 	 *	We have finally decided to accept it.
4041 	 */
4042 
4043 	ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4044 		     NEIGH_UPDATE_F_WEAK_OVERRIDE|
4045 		     NEIGH_UPDATE_F_OVERRIDE|
4046 		     (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4047 				     NEIGH_UPDATE_F_ISROUTER)),
4048 		     NDISC_REDIRECT, &ndopts);
4049 
4050 	rcu_read_lock();
4051 	res.f6i = rcu_dereference(rt->from);
4052 	if (!res.f6i)
4053 		goto out;
4054 
4055 	if (res.f6i->nh) {
4056 		struct fib6_nh_match_arg arg = {
4057 			.dev = dst->dev,
4058 			.gw = &rt->rt6i_gateway,
4059 		};
4060 
4061 		nexthop_for_each_fib6_nh(res.f6i->nh,
4062 					 fib6_nh_find_match, &arg);
4063 
4064 		/* fib6_info uses a nexthop that does not have fib6_nh
4065 		 * using the dst->dev. Should be impossible
4066 		 */
4067 		if (!arg.match)
4068 			goto out;
4069 		res.nh = arg.match;
4070 	} else {
4071 		res.nh = res.f6i->fib6_nh;
4072 	}
4073 
4074 	res.fib6_flags = res.f6i->fib6_flags;
4075 	res.fib6_type = res.f6i->fib6_type;
4076 	nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4077 	if (!nrt)
4078 		goto out;
4079 
4080 	nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4081 	if (on_link)
4082 		nrt->rt6i_flags &= ~RTF_GATEWAY;
4083 
4084 	nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4085 
4086 	/* rt6_insert_exception() will take care of duplicated exceptions */
4087 	if (rt6_insert_exception(nrt, &res)) {
4088 		dst_release_immediate(&nrt->dst);
4089 		goto out;
4090 	}
4091 
4092 	netevent.old = &rt->dst;
4093 	netevent.new = &nrt->dst;
4094 	netevent.daddr = &msg->dest;
4095 	netevent.neigh = neigh;
4096 	call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4097 
4098 out:
4099 	rcu_read_unlock();
4100 	neigh_release(neigh);
4101 }
4102 
4103 #ifdef CONFIG_IPV6_ROUTE_INFO
4104 static struct fib6_info *rt6_get_route_info(struct net *net,
4105 					   const struct in6_addr *prefix, int prefixlen,
4106 					   const struct in6_addr *gwaddr,
4107 					   struct net_device *dev)
4108 {
4109 	u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4110 	int ifindex = dev->ifindex;
4111 	struct fib6_node *fn;
4112 	struct fib6_info *rt = NULL;
4113 	struct fib6_table *table;
4114 
4115 	table = fib6_get_table(net, tb_id);
4116 	if (!table)
4117 		return NULL;
4118 
4119 	rcu_read_lock();
4120 	fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4121 	if (!fn)
4122 		goto out;
4123 
4124 	for_each_fib6_node_rt_rcu(fn) {
4125 		/* these routes do not use nexthops */
4126 		if (rt->nh)
4127 			continue;
4128 		if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4129 			continue;
4130 		if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4131 		    !rt->fib6_nh->fib_nh_gw_family)
4132 			continue;
4133 		if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4134 			continue;
4135 		if (!fib6_info_hold_safe(rt))
4136 			continue;
4137 		break;
4138 	}
4139 out:
4140 	rcu_read_unlock();
4141 	return rt;
4142 }
4143 
4144 static struct fib6_info *rt6_add_route_info(struct net *net,
4145 					   const struct in6_addr *prefix, int prefixlen,
4146 					   const struct in6_addr *gwaddr,
4147 					   struct net_device *dev,
4148 					   unsigned int pref)
4149 {
4150 	struct fib6_config cfg = {
4151 		.fc_metric	= IP6_RT_PRIO_USER,
4152 		.fc_ifindex	= dev->ifindex,
4153 		.fc_dst_len	= prefixlen,
4154 		.fc_flags	= RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4155 				  RTF_UP | RTF_PREF(pref),
4156 		.fc_protocol = RTPROT_RA,
4157 		.fc_type = RTN_UNICAST,
4158 		.fc_nlinfo.portid = 0,
4159 		.fc_nlinfo.nlh = NULL,
4160 		.fc_nlinfo.nl_net = net,
4161 	};
4162 
4163 	cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO,
4164 	cfg.fc_dst = *prefix;
4165 	cfg.fc_gateway = *gwaddr;
4166 
4167 	/* We should treat it as a default route if prefix length is 0. */
4168 	if (!prefixlen)
4169 		cfg.fc_flags |= RTF_DEFAULT;
4170 
4171 	ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4172 
4173 	return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4174 }
4175 #endif
4176 
4177 struct fib6_info *rt6_get_dflt_router(struct net *net,
4178 				     const struct in6_addr *addr,
4179 				     struct net_device *dev)
4180 {
4181 	u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4182 	struct fib6_info *rt;
4183 	struct fib6_table *table;
4184 
4185 	table = fib6_get_table(net, tb_id);
4186 	if (!table)
4187 		return NULL;
4188 
4189 	rcu_read_lock();
4190 	for_each_fib6_node_rt_rcu(&table->tb6_root) {
4191 		struct fib6_nh *nh;
4192 
4193 		/* RA routes do not use nexthops */
4194 		if (rt->nh)
4195 			continue;
4196 
4197 		nh = rt->fib6_nh;
4198 		if (dev == nh->fib_nh_dev &&
4199 		    ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4200 		    ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4201 			break;
4202 	}
4203 	if (rt && !fib6_info_hold_safe(rt))
4204 		rt = NULL;
4205 	rcu_read_unlock();
4206 	return rt;
4207 }
4208 
4209 struct fib6_info *rt6_add_dflt_router(struct net *net,
4210 				     const struct in6_addr *gwaddr,
4211 				     struct net_device *dev,
4212 				     unsigned int pref)
4213 {
4214 	struct fib6_config cfg = {
4215 		.fc_table	= l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4216 		.fc_metric	= IP6_RT_PRIO_USER,
4217 		.fc_ifindex	= dev->ifindex,
4218 		.fc_flags	= RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4219 				  RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4220 		.fc_protocol = RTPROT_RA,
4221 		.fc_type = RTN_UNICAST,
4222 		.fc_nlinfo.portid = 0,
4223 		.fc_nlinfo.nlh = NULL,
4224 		.fc_nlinfo.nl_net = net,
4225 	};
4226 
4227 	cfg.fc_gateway = *gwaddr;
4228 
4229 	if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4230 		struct fib6_table *table;
4231 
4232 		table = fib6_get_table(dev_net(dev), cfg.fc_table);
4233 		if (table)
4234 			table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4235 	}
4236 
4237 	return rt6_get_dflt_router(net, gwaddr, dev);
4238 }
4239 
4240 static void __rt6_purge_dflt_routers(struct net *net,
4241 				     struct fib6_table *table)
4242 {
4243 	struct fib6_info *rt;
4244 
4245 restart:
4246 	rcu_read_lock();
4247 	for_each_fib6_node_rt_rcu(&table->tb6_root) {
4248 		struct net_device *dev = fib6_info_nh_dev(rt);
4249 		struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4250 
4251 		if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4252 		    (!idev || idev->cnf.accept_ra != 2) &&
4253 		    fib6_info_hold_safe(rt)) {
4254 			rcu_read_unlock();
4255 			ip6_del_rt(net, rt);
4256 			goto restart;
4257 		}
4258 	}
4259 	rcu_read_unlock();
4260 
4261 	table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4262 }
4263 
4264 void rt6_purge_dflt_routers(struct net *net)
4265 {
4266 	struct fib6_table *table;
4267 	struct hlist_head *head;
4268 	unsigned int h;
4269 
4270 	rcu_read_lock();
4271 
4272 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4273 		head = &net->ipv6.fib_table_hash[h];
4274 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4275 			if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4276 				__rt6_purge_dflt_routers(net, table);
4277 		}
4278 	}
4279 
4280 	rcu_read_unlock();
4281 }
4282 
4283 static void rtmsg_to_fib6_config(struct net *net,
4284 				 struct in6_rtmsg *rtmsg,
4285 				 struct fib6_config *cfg)
4286 {
4287 	*cfg = (struct fib6_config){
4288 		.fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4289 			 : RT6_TABLE_MAIN,
4290 		.fc_ifindex = rtmsg->rtmsg_ifindex,
4291 		.fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4292 		.fc_expires = rtmsg->rtmsg_info,
4293 		.fc_dst_len = rtmsg->rtmsg_dst_len,
4294 		.fc_src_len = rtmsg->rtmsg_src_len,
4295 		.fc_flags = rtmsg->rtmsg_flags,
4296 		.fc_type = rtmsg->rtmsg_type,
4297 
4298 		.fc_nlinfo.nl_net = net,
4299 
4300 		.fc_dst = rtmsg->rtmsg_dst,
4301 		.fc_src = rtmsg->rtmsg_src,
4302 		.fc_gateway = rtmsg->rtmsg_gateway,
4303 	};
4304 }
4305 
4306 int ipv6_route_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4307 {
4308 	struct fib6_config cfg;
4309 	struct in6_rtmsg rtmsg;
4310 	int err;
4311 
4312 	switch (cmd) {
4313 	case SIOCADDRT:		/* Add a route */
4314 	case SIOCDELRT:		/* Delete a route */
4315 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4316 			return -EPERM;
4317 		err = copy_from_user(&rtmsg, arg,
4318 				     sizeof(struct in6_rtmsg));
4319 		if (err)
4320 			return -EFAULT;
4321 
4322 		rtmsg_to_fib6_config(net, &rtmsg, &cfg);
4323 
4324 		rtnl_lock();
4325 		switch (cmd) {
4326 		case SIOCADDRT:
4327 			err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4328 			break;
4329 		case SIOCDELRT:
4330 			err = ip6_route_del(&cfg, NULL);
4331 			break;
4332 		default:
4333 			err = -EINVAL;
4334 		}
4335 		rtnl_unlock();
4336 
4337 		return err;
4338 	}
4339 
4340 	return -EINVAL;
4341 }
4342 
4343 /*
4344  *	Drop the packet on the floor
4345  */
4346 
4347 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4348 {
4349 	struct dst_entry *dst = skb_dst(skb);
4350 	struct net *net = dev_net(dst->dev);
4351 	struct inet6_dev *idev;
4352 	int type;
4353 
4354 	if (netif_is_l3_master(skb->dev) &&
4355 	    dst->dev == net->loopback_dev)
4356 		idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4357 	else
4358 		idev = ip6_dst_idev(dst);
4359 
4360 	switch (ipstats_mib_noroutes) {
4361 	case IPSTATS_MIB_INNOROUTES:
4362 		type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4363 		if (type == IPV6_ADDR_ANY) {
4364 			IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4365 			break;
4366 		}
4367 		fallthrough;
4368 	case IPSTATS_MIB_OUTNOROUTES:
4369 		IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4370 		break;
4371 	}
4372 
4373 	/* Start over by dropping the dst for l3mdev case */
4374 	if (netif_is_l3_master(skb->dev))
4375 		skb_dst_drop(skb);
4376 
4377 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4378 	kfree_skb(skb);
4379 	return 0;
4380 }
4381 
4382 static int ip6_pkt_discard(struct sk_buff *skb)
4383 {
4384 	return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4385 }
4386 
4387 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4388 {
4389 	skb->dev = skb_dst(skb)->dev;
4390 	return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4391 }
4392 
4393 static int ip6_pkt_prohibit(struct sk_buff *skb)
4394 {
4395 	return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4396 }
4397 
4398 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4399 {
4400 	skb->dev = skb_dst(skb)->dev;
4401 	return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4402 }
4403 
4404 /*
4405  *	Allocate a dst for local (unicast / anycast) address.
4406  */
4407 
4408 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4409 				     struct inet6_dev *idev,
4410 				     const struct in6_addr *addr,
4411 				     bool anycast, gfp_t gfp_flags)
4412 {
4413 	struct fib6_config cfg = {
4414 		.fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4415 		.fc_ifindex = idev->dev->ifindex,
4416 		.fc_flags = RTF_UP | RTF_NONEXTHOP,
4417 		.fc_dst = *addr,
4418 		.fc_dst_len = 128,
4419 		.fc_protocol = RTPROT_KERNEL,
4420 		.fc_nlinfo.nl_net = net,
4421 		.fc_ignore_dev_down = true,
4422 	};
4423 	struct fib6_info *f6i;
4424 
4425 	if (anycast) {
4426 		cfg.fc_type = RTN_ANYCAST;
4427 		cfg.fc_flags |= RTF_ANYCAST;
4428 	} else {
4429 		cfg.fc_type = RTN_LOCAL;
4430 		cfg.fc_flags |= RTF_LOCAL;
4431 	}
4432 
4433 	f6i = ip6_route_info_create(&cfg, gfp_flags, NULL);
4434 	if (!IS_ERR(f6i))
4435 		f6i->dst_nocount = true;
4436 	return f6i;
4437 }
4438 
4439 /* remove deleted ip from prefsrc entries */
4440 struct arg_dev_net_ip {
4441 	struct net_device *dev;
4442 	struct net *net;
4443 	struct in6_addr *addr;
4444 };
4445 
4446 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4447 {
4448 	struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4449 	struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4450 	struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4451 
4452 	if (!rt->nh &&
4453 	    ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4454 	    rt != net->ipv6.fib6_null_entry &&
4455 	    ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4456 		spin_lock_bh(&rt6_exception_lock);
4457 		/* remove prefsrc entry */
4458 		rt->fib6_prefsrc.plen = 0;
4459 		spin_unlock_bh(&rt6_exception_lock);
4460 	}
4461 	return 0;
4462 }
4463 
4464 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4465 {
4466 	struct net *net = dev_net(ifp->idev->dev);
4467 	struct arg_dev_net_ip adni = {
4468 		.dev = ifp->idev->dev,
4469 		.net = net,
4470 		.addr = &ifp->addr,
4471 	};
4472 	fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4473 }
4474 
4475 #define RTF_RA_ROUTER		(RTF_ADDRCONF | RTF_DEFAULT)
4476 
4477 /* Remove routers and update dst entries when gateway turn into host. */
4478 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4479 {
4480 	struct in6_addr *gateway = (struct in6_addr *)arg;
4481 	struct fib6_nh *nh;
4482 
4483 	/* RA routes do not use nexthops */
4484 	if (rt->nh)
4485 		return 0;
4486 
4487 	nh = rt->fib6_nh;
4488 	if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4489 	    nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4490 		return -1;
4491 
4492 	/* Further clean up cached routes in exception table.
4493 	 * This is needed because cached route may have a different
4494 	 * gateway than its 'parent' in the case of an ip redirect.
4495 	 */
4496 	fib6_nh_exceptions_clean_tohost(nh, gateway);
4497 
4498 	return 0;
4499 }
4500 
4501 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4502 {
4503 	fib6_clean_all(net, fib6_clean_tohost, gateway);
4504 }
4505 
4506 struct arg_netdev_event {
4507 	const struct net_device *dev;
4508 	union {
4509 		unsigned char nh_flags;
4510 		unsigned long event;
4511 	};
4512 };
4513 
4514 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4515 {
4516 	struct fib6_info *iter;
4517 	struct fib6_node *fn;
4518 
4519 	fn = rcu_dereference_protected(rt->fib6_node,
4520 			lockdep_is_held(&rt->fib6_table->tb6_lock));
4521 	iter = rcu_dereference_protected(fn->leaf,
4522 			lockdep_is_held(&rt->fib6_table->tb6_lock));
4523 	while (iter) {
4524 		if (iter->fib6_metric == rt->fib6_metric &&
4525 		    rt6_qualify_for_ecmp(iter))
4526 			return iter;
4527 		iter = rcu_dereference_protected(iter->fib6_next,
4528 				lockdep_is_held(&rt->fib6_table->tb6_lock));
4529 	}
4530 
4531 	return NULL;
4532 }
4533 
4534 /* only called for fib entries with builtin fib6_nh */
4535 static bool rt6_is_dead(const struct fib6_info *rt)
4536 {
4537 	if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4538 	    (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4539 	     ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4540 		return true;
4541 
4542 	return false;
4543 }
4544 
4545 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4546 {
4547 	struct fib6_info *iter;
4548 	int total = 0;
4549 
4550 	if (!rt6_is_dead(rt))
4551 		total += rt->fib6_nh->fib_nh_weight;
4552 
4553 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4554 		if (!rt6_is_dead(iter))
4555 			total += iter->fib6_nh->fib_nh_weight;
4556 	}
4557 
4558 	return total;
4559 }
4560 
4561 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4562 {
4563 	int upper_bound = -1;
4564 
4565 	if (!rt6_is_dead(rt)) {
4566 		*weight += rt->fib6_nh->fib_nh_weight;
4567 		upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4568 						    total) - 1;
4569 	}
4570 	atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4571 }
4572 
4573 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4574 {
4575 	struct fib6_info *iter;
4576 	int weight = 0;
4577 
4578 	rt6_upper_bound_set(rt, &weight, total);
4579 
4580 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4581 		rt6_upper_bound_set(iter, &weight, total);
4582 }
4583 
4584 void rt6_multipath_rebalance(struct fib6_info *rt)
4585 {
4586 	struct fib6_info *first;
4587 	int total;
4588 
4589 	/* In case the entire multipath route was marked for flushing,
4590 	 * then there is no need to rebalance upon the removal of every
4591 	 * sibling route.
4592 	 */
4593 	if (!rt->fib6_nsiblings || rt->should_flush)
4594 		return;
4595 
4596 	/* During lookup routes are evaluated in order, so we need to
4597 	 * make sure upper bounds are assigned from the first sibling
4598 	 * onwards.
4599 	 */
4600 	first = rt6_multipath_first_sibling(rt);
4601 	if (WARN_ON_ONCE(!first))
4602 		return;
4603 
4604 	total = rt6_multipath_total_weight(first);
4605 	rt6_multipath_upper_bound_set(first, total);
4606 }
4607 
4608 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4609 {
4610 	const struct arg_netdev_event *arg = p_arg;
4611 	struct net *net = dev_net(arg->dev);
4612 
4613 	if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4614 	    rt->fib6_nh->fib_nh_dev == arg->dev) {
4615 		rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4616 		fib6_update_sernum_upto_root(net, rt);
4617 		rt6_multipath_rebalance(rt);
4618 	}
4619 
4620 	return 0;
4621 }
4622 
4623 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4624 {
4625 	struct arg_netdev_event arg = {
4626 		.dev = dev,
4627 		{
4628 			.nh_flags = nh_flags,
4629 		},
4630 	};
4631 
4632 	if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4633 		arg.nh_flags |= RTNH_F_LINKDOWN;
4634 
4635 	fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4636 }
4637 
4638 /* only called for fib entries with inline fib6_nh */
4639 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4640 				   const struct net_device *dev)
4641 {
4642 	struct fib6_info *iter;
4643 
4644 	if (rt->fib6_nh->fib_nh_dev == dev)
4645 		return true;
4646 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4647 		if (iter->fib6_nh->fib_nh_dev == dev)
4648 			return true;
4649 
4650 	return false;
4651 }
4652 
4653 static void rt6_multipath_flush(struct fib6_info *rt)
4654 {
4655 	struct fib6_info *iter;
4656 
4657 	rt->should_flush = 1;
4658 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4659 		iter->should_flush = 1;
4660 }
4661 
4662 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4663 					     const struct net_device *down_dev)
4664 {
4665 	struct fib6_info *iter;
4666 	unsigned int dead = 0;
4667 
4668 	if (rt->fib6_nh->fib_nh_dev == down_dev ||
4669 	    rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4670 		dead++;
4671 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4672 		if (iter->fib6_nh->fib_nh_dev == down_dev ||
4673 		    iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4674 			dead++;
4675 
4676 	return dead;
4677 }
4678 
4679 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4680 				       const struct net_device *dev,
4681 				       unsigned char nh_flags)
4682 {
4683 	struct fib6_info *iter;
4684 
4685 	if (rt->fib6_nh->fib_nh_dev == dev)
4686 		rt->fib6_nh->fib_nh_flags |= nh_flags;
4687 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4688 		if (iter->fib6_nh->fib_nh_dev == dev)
4689 			iter->fib6_nh->fib_nh_flags |= nh_flags;
4690 }
4691 
4692 /* called with write lock held for table with rt */
4693 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4694 {
4695 	const struct arg_netdev_event *arg = p_arg;
4696 	const struct net_device *dev = arg->dev;
4697 	struct net *net = dev_net(dev);
4698 
4699 	if (rt == net->ipv6.fib6_null_entry || rt->nh)
4700 		return 0;
4701 
4702 	switch (arg->event) {
4703 	case NETDEV_UNREGISTER:
4704 		return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4705 	case NETDEV_DOWN:
4706 		if (rt->should_flush)
4707 			return -1;
4708 		if (!rt->fib6_nsiblings)
4709 			return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4710 		if (rt6_multipath_uses_dev(rt, dev)) {
4711 			unsigned int count;
4712 
4713 			count = rt6_multipath_dead_count(rt, dev);
4714 			if (rt->fib6_nsiblings + 1 == count) {
4715 				rt6_multipath_flush(rt);
4716 				return -1;
4717 			}
4718 			rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4719 						   RTNH_F_LINKDOWN);
4720 			fib6_update_sernum(net, rt);
4721 			rt6_multipath_rebalance(rt);
4722 		}
4723 		return -2;
4724 	case NETDEV_CHANGE:
4725 		if (rt->fib6_nh->fib_nh_dev != dev ||
4726 		    rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4727 			break;
4728 		rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4729 		rt6_multipath_rebalance(rt);
4730 		break;
4731 	}
4732 
4733 	return 0;
4734 }
4735 
4736 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4737 {
4738 	struct arg_netdev_event arg = {
4739 		.dev = dev,
4740 		{
4741 			.event = event,
4742 		},
4743 	};
4744 	struct net *net = dev_net(dev);
4745 
4746 	if (net->ipv6.sysctl.skip_notify_on_dev_down)
4747 		fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4748 	else
4749 		fib6_clean_all(net, fib6_ifdown, &arg);
4750 }
4751 
4752 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4753 {
4754 	rt6_sync_down_dev(dev, event);
4755 	rt6_uncached_list_flush_dev(dev_net(dev), dev);
4756 	neigh_ifdown(&nd_tbl, dev);
4757 }
4758 
4759 struct rt6_mtu_change_arg {
4760 	struct net_device *dev;
4761 	unsigned int mtu;
4762 	struct fib6_info *f6i;
4763 };
4764 
4765 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4766 {
4767 	struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4768 	struct fib6_info *f6i = arg->f6i;
4769 
4770 	/* For administrative MTU increase, there is no way to discover
4771 	 * IPv6 PMTU increase, so PMTU increase should be updated here.
4772 	 * Since RFC 1981 doesn't include administrative MTU increase
4773 	 * update PMTU increase is a MUST. (i.e. jumbo frame)
4774 	 */
4775 	if (nh->fib_nh_dev == arg->dev) {
4776 		struct inet6_dev *idev = __in6_dev_get(arg->dev);
4777 		u32 mtu = f6i->fib6_pmtu;
4778 
4779 		if (mtu >= arg->mtu ||
4780 		    (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4781 			fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4782 
4783 		spin_lock_bh(&rt6_exception_lock);
4784 		rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4785 		spin_unlock_bh(&rt6_exception_lock);
4786 	}
4787 
4788 	return 0;
4789 }
4790 
4791 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4792 {
4793 	struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4794 	struct inet6_dev *idev;
4795 
4796 	/* In IPv6 pmtu discovery is not optional,
4797 	   so that RTAX_MTU lock cannot disable it.
4798 	   We still use this lock to block changes
4799 	   caused by addrconf/ndisc.
4800 	*/
4801 
4802 	idev = __in6_dev_get(arg->dev);
4803 	if (!idev)
4804 		return 0;
4805 
4806 	if (fib6_metric_locked(f6i, RTAX_MTU))
4807 		return 0;
4808 
4809 	arg->f6i = f6i;
4810 	if (f6i->nh) {
4811 		/* fib6_nh_mtu_change only returns 0, so this is safe */
4812 		return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4813 						arg);
4814 	}
4815 
4816 	return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4817 }
4818 
4819 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4820 {
4821 	struct rt6_mtu_change_arg arg = {
4822 		.dev = dev,
4823 		.mtu = mtu,
4824 	};
4825 
4826 	fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4827 }
4828 
4829 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4830 	[RTA_UNSPEC]		= { .strict_start_type = RTA_DPORT + 1 },
4831 	[RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4832 	[RTA_PREFSRC]		= { .len = sizeof(struct in6_addr) },
4833 	[RTA_OIF]               = { .type = NLA_U32 },
4834 	[RTA_IIF]		= { .type = NLA_U32 },
4835 	[RTA_PRIORITY]          = { .type = NLA_U32 },
4836 	[RTA_METRICS]           = { .type = NLA_NESTED },
4837 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
4838 	[RTA_PREF]              = { .type = NLA_U8 },
4839 	[RTA_ENCAP_TYPE]	= { .type = NLA_U16 },
4840 	[RTA_ENCAP]		= { .type = NLA_NESTED },
4841 	[RTA_EXPIRES]		= { .type = NLA_U32 },
4842 	[RTA_UID]		= { .type = NLA_U32 },
4843 	[RTA_MARK]		= { .type = NLA_U32 },
4844 	[RTA_TABLE]		= { .type = NLA_U32 },
4845 	[RTA_IP_PROTO]		= { .type = NLA_U8 },
4846 	[RTA_SPORT]		= { .type = NLA_U16 },
4847 	[RTA_DPORT]		= { .type = NLA_U16 },
4848 	[RTA_NH_ID]		= { .type = NLA_U32 },
4849 };
4850 
4851 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4852 			      struct fib6_config *cfg,
4853 			      struct netlink_ext_ack *extack)
4854 {
4855 	struct rtmsg *rtm;
4856 	struct nlattr *tb[RTA_MAX+1];
4857 	unsigned int pref;
4858 	int err;
4859 
4860 	err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
4861 				     rtm_ipv6_policy, extack);
4862 	if (err < 0)
4863 		goto errout;
4864 
4865 	err = -EINVAL;
4866 	rtm = nlmsg_data(nlh);
4867 
4868 	*cfg = (struct fib6_config){
4869 		.fc_table = rtm->rtm_table,
4870 		.fc_dst_len = rtm->rtm_dst_len,
4871 		.fc_src_len = rtm->rtm_src_len,
4872 		.fc_flags = RTF_UP,
4873 		.fc_protocol = rtm->rtm_protocol,
4874 		.fc_type = rtm->rtm_type,
4875 
4876 		.fc_nlinfo.portid = NETLINK_CB(skb).portid,
4877 		.fc_nlinfo.nlh = nlh,
4878 		.fc_nlinfo.nl_net = sock_net(skb->sk),
4879 	};
4880 
4881 	if (rtm->rtm_type == RTN_UNREACHABLE ||
4882 	    rtm->rtm_type == RTN_BLACKHOLE ||
4883 	    rtm->rtm_type == RTN_PROHIBIT ||
4884 	    rtm->rtm_type == RTN_THROW)
4885 		cfg->fc_flags |= RTF_REJECT;
4886 
4887 	if (rtm->rtm_type == RTN_LOCAL)
4888 		cfg->fc_flags |= RTF_LOCAL;
4889 
4890 	if (rtm->rtm_flags & RTM_F_CLONED)
4891 		cfg->fc_flags |= RTF_CACHE;
4892 
4893 	cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
4894 
4895 	if (tb[RTA_NH_ID]) {
4896 		if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
4897 		    tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
4898 			NL_SET_ERR_MSG(extack,
4899 				       "Nexthop specification and nexthop id are mutually exclusive");
4900 			goto errout;
4901 		}
4902 		cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
4903 	}
4904 
4905 	if (tb[RTA_GATEWAY]) {
4906 		cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
4907 		cfg->fc_flags |= RTF_GATEWAY;
4908 	}
4909 	if (tb[RTA_VIA]) {
4910 		NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
4911 		goto errout;
4912 	}
4913 
4914 	if (tb[RTA_DST]) {
4915 		int plen = (rtm->rtm_dst_len + 7) >> 3;
4916 
4917 		if (nla_len(tb[RTA_DST]) < plen)
4918 			goto errout;
4919 
4920 		nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
4921 	}
4922 
4923 	if (tb[RTA_SRC]) {
4924 		int plen = (rtm->rtm_src_len + 7) >> 3;
4925 
4926 		if (nla_len(tb[RTA_SRC]) < plen)
4927 			goto errout;
4928 
4929 		nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
4930 	}
4931 
4932 	if (tb[RTA_PREFSRC])
4933 		cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
4934 
4935 	if (tb[RTA_OIF])
4936 		cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
4937 
4938 	if (tb[RTA_PRIORITY])
4939 		cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
4940 
4941 	if (tb[RTA_METRICS]) {
4942 		cfg->fc_mx = nla_data(tb[RTA_METRICS]);
4943 		cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
4944 	}
4945 
4946 	if (tb[RTA_TABLE])
4947 		cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
4948 
4949 	if (tb[RTA_MULTIPATH]) {
4950 		cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
4951 		cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
4952 
4953 		err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
4954 						     cfg->fc_mp_len, extack);
4955 		if (err < 0)
4956 			goto errout;
4957 	}
4958 
4959 	if (tb[RTA_PREF]) {
4960 		pref = nla_get_u8(tb[RTA_PREF]);
4961 		if (pref != ICMPV6_ROUTER_PREF_LOW &&
4962 		    pref != ICMPV6_ROUTER_PREF_HIGH)
4963 			pref = ICMPV6_ROUTER_PREF_MEDIUM;
4964 		cfg->fc_flags |= RTF_PREF(pref);
4965 	}
4966 
4967 	if (tb[RTA_ENCAP])
4968 		cfg->fc_encap = tb[RTA_ENCAP];
4969 
4970 	if (tb[RTA_ENCAP_TYPE]) {
4971 		cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
4972 
4973 		err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
4974 		if (err < 0)
4975 			goto errout;
4976 	}
4977 
4978 	if (tb[RTA_EXPIRES]) {
4979 		unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
4980 
4981 		if (addrconf_finite_timeout(timeout)) {
4982 			cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
4983 			cfg->fc_flags |= RTF_EXPIRES;
4984 		}
4985 	}
4986 
4987 	err = 0;
4988 errout:
4989 	return err;
4990 }
4991 
4992 struct rt6_nh {
4993 	struct fib6_info *fib6_info;
4994 	struct fib6_config r_cfg;
4995 	struct list_head next;
4996 };
4997 
4998 static int ip6_route_info_append(struct net *net,
4999 				 struct list_head *rt6_nh_list,
5000 				 struct fib6_info *rt,
5001 				 struct fib6_config *r_cfg)
5002 {
5003 	struct rt6_nh *nh;
5004 	int err = -EEXIST;
5005 
5006 	list_for_each_entry(nh, rt6_nh_list, next) {
5007 		/* check if fib6_info already exists */
5008 		if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5009 			return err;
5010 	}
5011 
5012 	nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5013 	if (!nh)
5014 		return -ENOMEM;
5015 	nh->fib6_info = rt;
5016 	memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5017 	list_add_tail(&nh->next, rt6_nh_list);
5018 
5019 	return 0;
5020 }
5021 
5022 static void ip6_route_mpath_notify(struct fib6_info *rt,
5023 				   struct fib6_info *rt_last,
5024 				   struct nl_info *info,
5025 				   __u16 nlflags)
5026 {
5027 	/* if this is an APPEND route, then rt points to the first route
5028 	 * inserted and rt_last points to last route inserted. Userspace
5029 	 * wants a consistent dump of the route which starts at the first
5030 	 * nexthop. Since sibling routes are always added at the end of
5031 	 * the list, find the first sibling of the last route appended
5032 	 */
5033 	if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5034 		rt = list_first_entry(&rt_last->fib6_siblings,
5035 				      struct fib6_info,
5036 				      fib6_siblings);
5037 	}
5038 
5039 	if (rt)
5040 		inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5041 }
5042 
5043 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5044 {
5045 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5046 	bool should_notify = false;
5047 	struct fib6_info *leaf;
5048 	struct fib6_node *fn;
5049 
5050 	rcu_read_lock();
5051 	fn = rcu_dereference(rt->fib6_node);
5052 	if (!fn)
5053 		goto out;
5054 
5055 	leaf = rcu_dereference(fn->leaf);
5056 	if (!leaf)
5057 		goto out;
5058 
5059 	if (rt == leaf ||
5060 	    (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5061 	     rt6_qualify_for_ecmp(leaf)))
5062 		should_notify = true;
5063 out:
5064 	rcu_read_unlock();
5065 
5066 	return should_notify;
5067 }
5068 
5069 static int ip6_route_multipath_add(struct fib6_config *cfg,
5070 				   struct netlink_ext_ack *extack)
5071 {
5072 	struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5073 	struct nl_info *info = &cfg->fc_nlinfo;
5074 	struct fib6_config r_cfg;
5075 	struct rtnexthop *rtnh;
5076 	struct fib6_info *rt;
5077 	struct rt6_nh *err_nh;
5078 	struct rt6_nh *nh, *nh_safe;
5079 	__u16 nlflags;
5080 	int remaining;
5081 	int attrlen;
5082 	int err = 1;
5083 	int nhn = 0;
5084 	int replace = (cfg->fc_nlinfo.nlh &&
5085 		       (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5086 	LIST_HEAD(rt6_nh_list);
5087 
5088 	nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5089 	if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5090 		nlflags |= NLM_F_APPEND;
5091 
5092 	remaining = cfg->fc_mp_len;
5093 	rtnh = (struct rtnexthop *)cfg->fc_mp;
5094 
5095 	/* Parse a Multipath Entry and build a list (rt6_nh_list) of
5096 	 * fib6_info structs per nexthop
5097 	 */
5098 	while (rtnh_ok(rtnh, remaining)) {
5099 		memcpy(&r_cfg, cfg, sizeof(*cfg));
5100 		if (rtnh->rtnh_ifindex)
5101 			r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5102 
5103 		attrlen = rtnh_attrlen(rtnh);
5104 		if (attrlen > 0) {
5105 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5106 
5107 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5108 			if (nla) {
5109 				r_cfg.fc_gateway = nla_get_in6_addr(nla);
5110 				r_cfg.fc_flags |= RTF_GATEWAY;
5111 			}
5112 			r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5113 			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5114 			if (nla)
5115 				r_cfg.fc_encap_type = nla_get_u16(nla);
5116 		}
5117 
5118 		r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5119 		rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5120 		if (IS_ERR(rt)) {
5121 			err = PTR_ERR(rt);
5122 			rt = NULL;
5123 			goto cleanup;
5124 		}
5125 		if (!rt6_qualify_for_ecmp(rt)) {
5126 			err = -EINVAL;
5127 			NL_SET_ERR_MSG(extack,
5128 				       "Device only routes can not be added for IPv6 using the multipath API.");
5129 			fib6_info_release(rt);
5130 			goto cleanup;
5131 		}
5132 
5133 		rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5134 
5135 		err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5136 					    rt, &r_cfg);
5137 		if (err) {
5138 			fib6_info_release(rt);
5139 			goto cleanup;
5140 		}
5141 
5142 		rtnh = rtnh_next(rtnh, &remaining);
5143 	}
5144 
5145 	if (list_empty(&rt6_nh_list)) {
5146 		NL_SET_ERR_MSG(extack,
5147 			       "Invalid nexthop configuration - no valid nexthops");
5148 		return -EINVAL;
5149 	}
5150 
5151 	/* for add and replace send one notification with all nexthops.
5152 	 * Skip the notification in fib6_add_rt2node and send one with
5153 	 * the full route when done
5154 	 */
5155 	info->skip_notify = 1;
5156 
5157 	/* For add and replace, send one notification with all nexthops. For
5158 	 * append, send one notification with all appended nexthops.
5159 	 */
5160 	info->skip_notify_kernel = 1;
5161 
5162 	err_nh = NULL;
5163 	list_for_each_entry(nh, &rt6_nh_list, next) {
5164 		err = __ip6_ins_rt(nh->fib6_info, info, extack);
5165 		fib6_info_release(nh->fib6_info);
5166 
5167 		if (!err) {
5168 			/* save reference to last route successfully inserted */
5169 			rt_last = nh->fib6_info;
5170 
5171 			/* save reference to first route for notification */
5172 			if (!rt_notif)
5173 				rt_notif = nh->fib6_info;
5174 		}
5175 
5176 		/* nh->fib6_info is used or freed at this point, reset to NULL*/
5177 		nh->fib6_info = NULL;
5178 		if (err) {
5179 			if (replace && nhn)
5180 				NL_SET_ERR_MSG_MOD(extack,
5181 						   "multipath route replace failed (check consistency of installed routes)");
5182 			err_nh = nh;
5183 			goto add_errout;
5184 		}
5185 
5186 		/* Because each route is added like a single route we remove
5187 		 * these flags after the first nexthop: if there is a collision,
5188 		 * we have already failed to add the first nexthop:
5189 		 * fib6_add_rt2node() has rejected it; when replacing, old
5190 		 * nexthops have been replaced by first new, the rest should
5191 		 * be added to it.
5192 		 */
5193 		cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5194 						     NLM_F_REPLACE);
5195 		cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5196 		nhn++;
5197 	}
5198 
5199 	/* An in-kernel notification should only be sent in case the new
5200 	 * multipath route is added as the first route in the node, or if
5201 	 * it was appended to it. We pass 'rt_notif' since it is the first
5202 	 * sibling and might allow us to skip some checks in the replace case.
5203 	 */
5204 	if (ip6_route_mpath_should_notify(rt_notif)) {
5205 		enum fib_event_type fib_event;
5206 
5207 		if (rt_notif->fib6_nsiblings != nhn - 1)
5208 			fib_event = FIB_EVENT_ENTRY_APPEND;
5209 		else
5210 			fib_event = FIB_EVENT_ENTRY_REPLACE;
5211 
5212 		err = call_fib6_multipath_entry_notifiers(info->nl_net,
5213 							  fib_event, rt_notif,
5214 							  nhn - 1, extack);
5215 		if (err) {
5216 			/* Delete all the siblings that were just added */
5217 			err_nh = NULL;
5218 			goto add_errout;
5219 		}
5220 	}
5221 
5222 	/* success ... tell user about new route */
5223 	ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5224 	goto cleanup;
5225 
5226 add_errout:
5227 	/* send notification for routes that were added so that
5228 	 * the delete notifications sent by ip6_route_del are
5229 	 * coherent
5230 	 */
5231 	if (rt_notif)
5232 		ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5233 
5234 	/* Delete routes that were already added */
5235 	list_for_each_entry(nh, &rt6_nh_list, next) {
5236 		if (err_nh == nh)
5237 			break;
5238 		ip6_route_del(&nh->r_cfg, extack);
5239 	}
5240 
5241 cleanup:
5242 	list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5243 		if (nh->fib6_info)
5244 			fib6_info_release(nh->fib6_info);
5245 		list_del(&nh->next);
5246 		kfree(nh);
5247 	}
5248 
5249 	return err;
5250 }
5251 
5252 static int ip6_route_multipath_del(struct fib6_config *cfg,
5253 				   struct netlink_ext_ack *extack)
5254 {
5255 	struct fib6_config r_cfg;
5256 	struct rtnexthop *rtnh;
5257 	int remaining;
5258 	int attrlen;
5259 	int err = 1, last_err = 0;
5260 
5261 	remaining = cfg->fc_mp_len;
5262 	rtnh = (struct rtnexthop *)cfg->fc_mp;
5263 
5264 	/* Parse a Multipath Entry */
5265 	while (rtnh_ok(rtnh, remaining)) {
5266 		memcpy(&r_cfg, cfg, sizeof(*cfg));
5267 		if (rtnh->rtnh_ifindex)
5268 			r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5269 
5270 		attrlen = rtnh_attrlen(rtnh);
5271 		if (attrlen > 0) {
5272 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5273 
5274 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5275 			if (nla) {
5276 				nla_memcpy(&r_cfg.fc_gateway, nla, 16);
5277 				r_cfg.fc_flags |= RTF_GATEWAY;
5278 			}
5279 		}
5280 		err = ip6_route_del(&r_cfg, extack);
5281 		if (err)
5282 			last_err = err;
5283 
5284 		rtnh = rtnh_next(rtnh, &remaining);
5285 	}
5286 
5287 	return last_err;
5288 }
5289 
5290 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5291 			      struct netlink_ext_ack *extack)
5292 {
5293 	struct fib6_config cfg;
5294 	int err;
5295 
5296 	err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5297 	if (err < 0)
5298 		return err;
5299 
5300 	if (cfg.fc_nh_id &&
5301 	    !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5302 		NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5303 		return -EINVAL;
5304 	}
5305 
5306 	if (cfg.fc_mp)
5307 		return ip6_route_multipath_del(&cfg, extack);
5308 	else {
5309 		cfg.fc_delete_all_nh = 1;
5310 		return ip6_route_del(&cfg, extack);
5311 	}
5312 }
5313 
5314 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5315 			      struct netlink_ext_ack *extack)
5316 {
5317 	struct fib6_config cfg;
5318 	int err;
5319 
5320 	err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5321 	if (err < 0)
5322 		return err;
5323 
5324 	if (cfg.fc_metric == 0)
5325 		cfg.fc_metric = IP6_RT_PRIO_USER;
5326 
5327 	if (cfg.fc_mp)
5328 		return ip6_route_multipath_add(&cfg, extack);
5329 	else
5330 		return ip6_route_add(&cfg, GFP_KERNEL, extack);
5331 }
5332 
5333 /* add the overhead of this fib6_nh to nexthop_len */
5334 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5335 {
5336 	int *nexthop_len = arg;
5337 
5338 	*nexthop_len += nla_total_size(0)	 /* RTA_MULTIPATH */
5339 		     + NLA_ALIGN(sizeof(struct rtnexthop))
5340 		     + nla_total_size(16); /* RTA_GATEWAY */
5341 
5342 	if (nh->fib_nh_lws) {
5343 		/* RTA_ENCAP_TYPE */
5344 		*nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5345 		/* RTA_ENCAP */
5346 		*nexthop_len += nla_total_size(2);
5347 	}
5348 
5349 	return 0;
5350 }
5351 
5352 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5353 {
5354 	int nexthop_len;
5355 
5356 	if (f6i->nh) {
5357 		nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5358 		nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5359 					 &nexthop_len);
5360 	} else {
5361 		struct fib6_nh *nh = f6i->fib6_nh;
5362 
5363 		nexthop_len = 0;
5364 		if (f6i->fib6_nsiblings) {
5365 			nexthop_len = nla_total_size(0)	 /* RTA_MULTIPATH */
5366 				    + NLA_ALIGN(sizeof(struct rtnexthop))
5367 				    + nla_total_size(16) /* RTA_GATEWAY */
5368 				    + lwtunnel_get_encap_size(nh->fib_nh_lws);
5369 
5370 			nexthop_len *= f6i->fib6_nsiblings;
5371 		}
5372 		nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5373 	}
5374 
5375 	return NLMSG_ALIGN(sizeof(struct rtmsg))
5376 	       + nla_total_size(16) /* RTA_SRC */
5377 	       + nla_total_size(16) /* RTA_DST */
5378 	       + nla_total_size(16) /* RTA_GATEWAY */
5379 	       + nla_total_size(16) /* RTA_PREFSRC */
5380 	       + nla_total_size(4) /* RTA_TABLE */
5381 	       + nla_total_size(4) /* RTA_IIF */
5382 	       + nla_total_size(4) /* RTA_OIF */
5383 	       + nla_total_size(4) /* RTA_PRIORITY */
5384 	       + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5385 	       + nla_total_size(sizeof(struct rta_cacheinfo))
5386 	       + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5387 	       + nla_total_size(1) /* RTA_PREF */
5388 	       + nexthop_len;
5389 }
5390 
5391 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5392 				 unsigned char *flags)
5393 {
5394 	if (nexthop_is_multipath(nh)) {
5395 		struct nlattr *mp;
5396 
5397 		mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5398 		if (!mp)
5399 			goto nla_put_failure;
5400 
5401 		if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5402 			goto nla_put_failure;
5403 
5404 		nla_nest_end(skb, mp);
5405 	} else {
5406 		struct fib6_nh *fib6_nh;
5407 
5408 		fib6_nh = nexthop_fib6_nh(nh);
5409 		if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5410 				     flags, false) < 0)
5411 			goto nla_put_failure;
5412 	}
5413 
5414 	return 0;
5415 
5416 nla_put_failure:
5417 	return -EMSGSIZE;
5418 }
5419 
5420 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5421 			 struct fib6_info *rt, struct dst_entry *dst,
5422 			 struct in6_addr *dest, struct in6_addr *src,
5423 			 int iif, int type, u32 portid, u32 seq,
5424 			 unsigned int flags)
5425 {
5426 	struct rt6_info *rt6 = (struct rt6_info *)dst;
5427 	struct rt6key *rt6_dst, *rt6_src;
5428 	u32 *pmetrics, table, rt6_flags;
5429 	unsigned char nh_flags = 0;
5430 	struct nlmsghdr *nlh;
5431 	struct rtmsg *rtm;
5432 	long expires = 0;
5433 
5434 	nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5435 	if (!nlh)
5436 		return -EMSGSIZE;
5437 
5438 	if (rt6) {
5439 		rt6_dst = &rt6->rt6i_dst;
5440 		rt6_src = &rt6->rt6i_src;
5441 		rt6_flags = rt6->rt6i_flags;
5442 	} else {
5443 		rt6_dst = &rt->fib6_dst;
5444 		rt6_src = &rt->fib6_src;
5445 		rt6_flags = rt->fib6_flags;
5446 	}
5447 
5448 	rtm = nlmsg_data(nlh);
5449 	rtm->rtm_family = AF_INET6;
5450 	rtm->rtm_dst_len = rt6_dst->plen;
5451 	rtm->rtm_src_len = rt6_src->plen;
5452 	rtm->rtm_tos = 0;
5453 	if (rt->fib6_table)
5454 		table = rt->fib6_table->tb6_id;
5455 	else
5456 		table = RT6_TABLE_UNSPEC;
5457 	rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5458 	if (nla_put_u32(skb, RTA_TABLE, table))
5459 		goto nla_put_failure;
5460 
5461 	rtm->rtm_type = rt->fib6_type;
5462 	rtm->rtm_flags = 0;
5463 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5464 	rtm->rtm_protocol = rt->fib6_protocol;
5465 
5466 	if (rt6_flags & RTF_CACHE)
5467 		rtm->rtm_flags |= RTM_F_CLONED;
5468 
5469 	if (dest) {
5470 		if (nla_put_in6_addr(skb, RTA_DST, dest))
5471 			goto nla_put_failure;
5472 		rtm->rtm_dst_len = 128;
5473 	} else if (rtm->rtm_dst_len)
5474 		if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5475 			goto nla_put_failure;
5476 #ifdef CONFIG_IPV6_SUBTREES
5477 	if (src) {
5478 		if (nla_put_in6_addr(skb, RTA_SRC, src))
5479 			goto nla_put_failure;
5480 		rtm->rtm_src_len = 128;
5481 	} else if (rtm->rtm_src_len &&
5482 		   nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5483 		goto nla_put_failure;
5484 #endif
5485 	if (iif) {
5486 #ifdef CONFIG_IPV6_MROUTE
5487 		if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5488 			int err = ip6mr_get_route(net, skb, rtm, portid);
5489 
5490 			if (err == 0)
5491 				return 0;
5492 			if (err < 0)
5493 				goto nla_put_failure;
5494 		} else
5495 #endif
5496 			if (nla_put_u32(skb, RTA_IIF, iif))
5497 				goto nla_put_failure;
5498 	} else if (dest) {
5499 		struct in6_addr saddr_buf;
5500 		if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5501 		    nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5502 			goto nla_put_failure;
5503 	}
5504 
5505 	if (rt->fib6_prefsrc.plen) {
5506 		struct in6_addr saddr_buf;
5507 		saddr_buf = rt->fib6_prefsrc.addr;
5508 		if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5509 			goto nla_put_failure;
5510 	}
5511 
5512 	pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5513 	if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5514 		goto nla_put_failure;
5515 
5516 	if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5517 		goto nla_put_failure;
5518 
5519 	/* For multipath routes, walk the siblings list and add
5520 	 * each as a nexthop within RTA_MULTIPATH.
5521 	 */
5522 	if (rt6) {
5523 		if (rt6_flags & RTF_GATEWAY &&
5524 		    nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5525 			goto nla_put_failure;
5526 
5527 		if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5528 			goto nla_put_failure;
5529 	} else if (rt->fib6_nsiblings) {
5530 		struct fib6_info *sibling, *next_sibling;
5531 		struct nlattr *mp;
5532 
5533 		mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5534 		if (!mp)
5535 			goto nla_put_failure;
5536 
5537 		if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5538 				    rt->fib6_nh->fib_nh_weight, AF_INET6) < 0)
5539 			goto nla_put_failure;
5540 
5541 		list_for_each_entry_safe(sibling, next_sibling,
5542 					 &rt->fib6_siblings, fib6_siblings) {
5543 			if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5544 					    sibling->fib6_nh->fib_nh_weight,
5545 					    AF_INET6) < 0)
5546 				goto nla_put_failure;
5547 		}
5548 
5549 		nla_nest_end(skb, mp);
5550 	} else if (rt->nh) {
5551 		if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5552 			goto nla_put_failure;
5553 
5554 		if (nexthop_is_blackhole(rt->nh))
5555 			rtm->rtm_type = RTN_BLACKHOLE;
5556 
5557 		if (rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5558 			goto nla_put_failure;
5559 
5560 		rtm->rtm_flags |= nh_flags;
5561 	} else {
5562 		if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5563 				     &nh_flags, false) < 0)
5564 			goto nla_put_failure;
5565 
5566 		rtm->rtm_flags |= nh_flags;
5567 	}
5568 
5569 	if (rt6_flags & RTF_EXPIRES) {
5570 		expires = dst ? dst->expires : rt->expires;
5571 		expires -= jiffies;
5572 	}
5573 
5574 	if (!dst) {
5575 		if (rt->offload)
5576 			rtm->rtm_flags |= RTM_F_OFFLOAD;
5577 		if (rt->trap)
5578 			rtm->rtm_flags |= RTM_F_TRAP;
5579 	}
5580 
5581 	if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5582 		goto nla_put_failure;
5583 
5584 	if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5585 		goto nla_put_failure;
5586 
5587 
5588 	nlmsg_end(skb, nlh);
5589 	return 0;
5590 
5591 nla_put_failure:
5592 	nlmsg_cancel(skb, nlh);
5593 	return -EMSGSIZE;
5594 }
5595 
5596 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5597 {
5598 	const struct net_device *dev = arg;
5599 
5600 	if (nh->fib_nh_dev == dev)
5601 		return 1;
5602 
5603 	return 0;
5604 }
5605 
5606 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5607 			       const struct net_device *dev)
5608 {
5609 	if (f6i->nh) {
5610 		struct net_device *_dev = (struct net_device *)dev;
5611 
5612 		return !!nexthop_for_each_fib6_nh(f6i->nh,
5613 						  fib6_info_nh_uses_dev,
5614 						  _dev);
5615 	}
5616 
5617 	if (f6i->fib6_nh->fib_nh_dev == dev)
5618 		return true;
5619 
5620 	if (f6i->fib6_nsiblings) {
5621 		struct fib6_info *sibling, *next_sibling;
5622 
5623 		list_for_each_entry_safe(sibling, next_sibling,
5624 					 &f6i->fib6_siblings, fib6_siblings) {
5625 			if (sibling->fib6_nh->fib_nh_dev == dev)
5626 				return true;
5627 		}
5628 	}
5629 
5630 	return false;
5631 }
5632 
5633 struct fib6_nh_exception_dump_walker {
5634 	struct rt6_rtnl_dump_arg *dump;
5635 	struct fib6_info *rt;
5636 	unsigned int flags;
5637 	unsigned int skip;
5638 	unsigned int count;
5639 };
5640 
5641 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5642 {
5643 	struct fib6_nh_exception_dump_walker *w = arg;
5644 	struct rt6_rtnl_dump_arg *dump = w->dump;
5645 	struct rt6_exception_bucket *bucket;
5646 	struct rt6_exception *rt6_ex;
5647 	int i, err;
5648 
5649 	bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5650 	if (!bucket)
5651 		return 0;
5652 
5653 	for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5654 		hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5655 			if (w->skip) {
5656 				w->skip--;
5657 				continue;
5658 			}
5659 
5660 			/* Expiration of entries doesn't bump sernum, insertion
5661 			 * does. Removal is triggered by insertion, so we can
5662 			 * rely on the fact that if entries change between two
5663 			 * partial dumps, this node is scanned again completely,
5664 			 * see rt6_insert_exception() and fib6_dump_table().
5665 			 *
5666 			 * Count expired entries we go through as handled
5667 			 * entries that we'll skip next time, in case of partial
5668 			 * node dump. Otherwise, if entries expire meanwhile,
5669 			 * we'll skip the wrong amount.
5670 			 */
5671 			if (rt6_check_expired(rt6_ex->rt6i)) {
5672 				w->count++;
5673 				continue;
5674 			}
5675 
5676 			err = rt6_fill_node(dump->net, dump->skb, w->rt,
5677 					    &rt6_ex->rt6i->dst, NULL, NULL, 0,
5678 					    RTM_NEWROUTE,
5679 					    NETLINK_CB(dump->cb->skb).portid,
5680 					    dump->cb->nlh->nlmsg_seq, w->flags);
5681 			if (err)
5682 				return err;
5683 
5684 			w->count++;
5685 		}
5686 		bucket++;
5687 	}
5688 
5689 	return 0;
5690 }
5691 
5692 /* Return -1 if done with node, number of handled routes on partial dump */
5693 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5694 {
5695 	struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5696 	struct fib_dump_filter *filter = &arg->filter;
5697 	unsigned int flags = NLM_F_MULTI;
5698 	struct net *net = arg->net;
5699 	int count = 0;
5700 
5701 	if (rt == net->ipv6.fib6_null_entry)
5702 		return -1;
5703 
5704 	if ((filter->flags & RTM_F_PREFIX) &&
5705 	    !(rt->fib6_flags & RTF_PREFIX_RT)) {
5706 		/* success since this is not a prefix route */
5707 		return -1;
5708 	}
5709 	if (filter->filter_set &&
5710 	    ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5711 	     (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5712 	     (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5713 		return -1;
5714 	}
5715 
5716 	if (filter->filter_set ||
5717 	    !filter->dump_routes || !filter->dump_exceptions) {
5718 		flags |= NLM_F_DUMP_FILTERED;
5719 	}
5720 
5721 	if (filter->dump_routes) {
5722 		if (skip) {
5723 			skip--;
5724 		} else {
5725 			if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5726 					  0, RTM_NEWROUTE,
5727 					  NETLINK_CB(arg->cb->skb).portid,
5728 					  arg->cb->nlh->nlmsg_seq, flags)) {
5729 				return 0;
5730 			}
5731 			count++;
5732 		}
5733 	}
5734 
5735 	if (filter->dump_exceptions) {
5736 		struct fib6_nh_exception_dump_walker w = { .dump = arg,
5737 							   .rt = rt,
5738 							   .flags = flags,
5739 							   .skip = skip,
5740 							   .count = 0 };
5741 		int err;
5742 
5743 		rcu_read_lock();
5744 		if (rt->nh) {
5745 			err = nexthop_for_each_fib6_nh(rt->nh,
5746 						       rt6_nh_dump_exceptions,
5747 						       &w);
5748 		} else {
5749 			err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5750 		}
5751 		rcu_read_unlock();
5752 
5753 		if (err)
5754 			return count += w.count;
5755 	}
5756 
5757 	return -1;
5758 }
5759 
5760 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5761 					const struct nlmsghdr *nlh,
5762 					struct nlattr **tb,
5763 					struct netlink_ext_ack *extack)
5764 {
5765 	struct rtmsg *rtm;
5766 	int i, err;
5767 
5768 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5769 		NL_SET_ERR_MSG_MOD(extack,
5770 				   "Invalid header for get route request");
5771 		return -EINVAL;
5772 	}
5773 
5774 	if (!netlink_strict_get_check(skb))
5775 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5776 					      rtm_ipv6_policy, extack);
5777 
5778 	rtm = nlmsg_data(nlh);
5779 	if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5780 	    (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5781 	    rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5782 	    rtm->rtm_type) {
5783 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5784 		return -EINVAL;
5785 	}
5786 	if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5787 		NL_SET_ERR_MSG_MOD(extack,
5788 				   "Invalid flags for get route request");
5789 		return -EINVAL;
5790 	}
5791 
5792 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5793 					    rtm_ipv6_policy, extack);
5794 	if (err)
5795 		return err;
5796 
5797 	if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5798 	    (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5799 		NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5800 		return -EINVAL;
5801 	}
5802 
5803 	for (i = 0; i <= RTA_MAX; i++) {
5804 		if (!tb[i])
5805 			continue;
5806 
5807 		switch (i) {
5808 		case RTA_SRC:
5809 		case RTA_DST:
5810 		case RTA_IIF:
5811 		case RTA_OIF:
5812 		case RTA_MARK:
5813 		case RTA_UID:
5814 		case RTA_SPORT:
5815 		case RTA_DPORT:
5816 		case RTA_IP_PROTO:
5817 			break;
5818 		default:
5819 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
5820 			return -EINVAL;
5821 		}
5822 	}
5823 
5824 	return 0;
5825 }
5826 
5827 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
5828 			      struct netlink_ext_ack *extack)
5829 {
5830 	struct net *net = sock_net(in_skb->sk);
5831 	struct nlattr *tb[RTA_MAX+1];
5832 	int err, iif = 0, oif = 0;
5833 	struct fib6_info *from;
5834 	struct dst_entry *dst;
5835 	struct rt6_info *rt;
5836 	struct sk_buff *skb;
5837 	struct rtmsg *rtm;
5838 	struct flowi6 fl6 = {};
5839 	bool fibmatch;
5840 
5841 	err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
5842 	if (err < 0)
5843 		goto errout;
5844 
5845 	err = -EINVAL;
5846 	rtm = nlmsg_data(nlh);
5847 	fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
5848 	fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
5849 
5850 	if (tb[RTA_SRC]) {
5851 		if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
5852 			goto errout;
5853 
5854 		fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
5855 	}
5856 
5857 	if (tb[RTA_DST]) {
5858 		if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
5859 			goto errout;
5860 
5861 		fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
5862 	}
5863 
5864 	if (tb[RTA_IIF])
5865 		iif = nla_get_u32(tb[RTA_IIF]);
5866 
5867 	if (tb[RTA_OIF])
5868 		oif = nla_get_u32(tb[RTA_OIF]);
5869 
5870 	if (tb[RTA_MARK])
5871 		fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
5872 
5873 	if (tb[RTA_UID])
5874 		fl6.flowi6_uid = make_kuid(current_user_ns(),
5875 					   nla_get_u32(tb[RTA_UID]));
5876 	else
5877 		fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
5878 
5879 	if (tb[RTA_SPORT])
5880 		fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
5881 
5882 	if (tb[RTA_DPORT])
5883 		fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
5884 
5885 	if (tb[RTA_IP_PROTO]) {
5886 		err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
5887 						  &fl6.flowi6_proto, AF_INET6,
5888 						  extack);
5889 		if (err)
5890 			goto errout;
5891 	}
5892 
5893 	if (iif) {
5894 		struct net_device *dev;
5895 		int flags = 0;
5896 
5897 		rcu_read_lock();
5898 
5899 		dev = dev_get_by_index_rcu(net, iif);
5900 		if (!dev) {
5901 			rcu_read_unlock();
5902 			err = -ENODEV;
5903 			goto errout;
5904 		}
5905 
5906 		fl6.flowi6_iif = iif;
5907 
5908 		if (!ipv6_addr_any(&fl6.saddr))
5909 			flags |= RT6_LOOKUP_F_HAS_SADDR;
5910 
5911 		dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
5912 
5913 		rcu_read_unlock();
5914 	} else {
5915 		fl6.flowi6_oif = oif;
5916 
5917 		dst = ip6_route_output(net, NULL, &fl6);
5918 	}
5919 
5920 
5921 	rt = container_of(dst, struct rt6_info, dst);
5922 	if (rt->dst.error) {
5923 		err = rt->dst.error;
5924 		ip6_rt_put(rt);
5925 		goto errout;
5926 	}
5927 
5928 	if (rt == net->ipv6.ip6_null_entry) {
5929 		err = rt->dst.error;
5930 		ip6_rt_put(rt);
5931 		goto errout;
5932 	}
5933 
5934 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
5935 	if (!skb) {
5936 		ip6_rt_put(rt);
5937 		err = -ENOBUFS;
5938 		goto errout;
5939 	}
5940 
5941 	skb_dst_set(skb, &rt->dst);
5942 
5943 	rcu_read_lock();
5944 	from = rcu_dereference(rt->from);
5945 	if (from) {
5946 		if (fibmatch)
5947 			err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
5948 					    iif, RTM_NEWROUTE,
5949 					    NETLINK_CB(in_skb).portid,
5950 					    nlh->nlmsg_seq, 0);
5951 		else
5952 			err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
5953 					    &fl6.saddr, iif, RTM_NEWROUTE,
5954 					    NETLINK_CB(in_skb).portid,
5955 					    nlh->nlmsg_seq, 0);
5956 	} else {
5957 		err = -ENETUNREACH;
5958 	}
5959 	rcu_read_unlock();
5960 
5961 	if (err < 0) {
5962 		kfree_skb(skb);
5963 		goto errout;
5964 	}
5965 
5966 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
5967 errout:
5968 	return err;
5969 }
5970 
5971 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
5972 		     unsigned int nlm_flags)
5973 {
5974 	struct sk_buff *skb;
5975 	struct net *net = info->nl_net;
5976 	u32 seq;
5977 	int err;
5978 
5979 	err = -ENOBUFS;
5980 	seq = info->nlh ? info->nlh->nlmsg_seq : 0;
5981 
5982 	skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
5983 	if (!skb)
5984 		goto errout;
5985 
5986 	err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
5987 			    event, info->portid, seq, nlm_flags);
5988 	if (err < 0) {
5989 		/* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
5990 		WARN_ON(err == -EMSGSIZE);
5991 		kfree_skb(skb);
5992 		goto errout;
5993 	}
5994 	rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
5995 		    info->nlh, gfp_any());
5996 	return;
5997 errout:
5998 	if (err < 0)
5999 		rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6000 }
6001 
6002 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6003 		    struct nl_info *info)
6004 {
6005 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6006 	struct sk_buff *skb;
6007 	int err = -ENOBUFS;
6008 
6009 	/* call_fib6_entry_notifiers will be removed when in-kernel notifier
6010 	 * is implemented and supported for nexthop objects
6011 	 */
6012 	call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, rt, NULL);
6013 
6014 	skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6015 	if (!skb)
6016 		goto errout;
6017 
6018 	err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6019 			    RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6020 	if (err < 0) {
6021 		/* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6022 		WARN_ON(err == -EMSGSIZE);
6023 		kfree_skb(skb);
6024 		goto errout;
6025 	}
6026 	rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6027 		    info->nlh, gfp_any());
6028 	return;
6029 errout:
6030 	if (err < 0)
6031 		rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6032 }
6033 
6034 static int ip6_route_dev_notify(struct notifier_block *this,
6035 				unsigned long event, void *ptr)
6036 {
6037 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6038 	struct net *net = dev_net(dev);
6039 
6040 	if (!(dev->flags & IFF_LOOPBACK))
6041 		return NOTIFY_OK;
6042 
6043 	if (event == NETDEV_REGISTER) {
6044 		net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6045 		net->ipv6.ip6_null_entry->dst.dev = dev;
6046 		net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6047 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6048 		net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6049 		net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6050 		net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6051 		net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6052 #endif
6053 	 } else if (event == NETDEV_UNREGISTER &&
6054 		    dev->reg_state != NETREG_UNREGISTERED) {
6055 		/* NETDEV_UNREGISTER could be fired for multiple times by
6056 		 * netdev_wait_allrefs(). Make sure we only call this once.
6057 		 */
6058 		in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6059 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6060 		in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6061 		in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6062 #endif
6063 	}
6064 
6065 	return NOTIFY_OK;
6066 }
6067 
6068 /*
6069  *	/proc
6070  */
6071 
6072 #ifdef CONFIG_PROC_FS
6073 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6074 {
6075 	struct net *net = (struct net *)seq->private;
6076 	seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6077 		   net->ipv6.rt6_stats->fib_nodes,
6078 		   net->ipv6.rt6_stats->fib_route_nodes,
6079 		   atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6080 		   net->ipv6.rt6_stats->fib_rt_entries,
6081 		   net->ipv6.rt6_stats->fib_rt_cache,
6082 		   dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6083 		   net->ipv6.rt6_stats->fib_discarded_routes);
6084 
6085 	return 0;
6086 }
6087 #endif	/* CONFIG_PROC_FS */
6088 
6089 #ifdef CONFIG_SYSCTL
6090 
6091 static
6092 int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6093 			      void __user *buffer, size_t *lenp, loff_t *ppos)
6094 {
6095 	struct net *net;
6096 	int delay;
6097 	int ret;
6098 	if (!write)
6099 		return -EINVAL;
6100 
6101 	net = (struct net *)ctl->extra1;
6102 	delay = net->ipv6.sysctl.flush_delay;
6103 	ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6104 	if (ret)
6105 		return ret;
6106 
6107 	fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6108 	return 0;
6109 }
6110 
6111 static struct ctl_table ipv6_route_table_template[] = {
6112 	{
6113 		.procname	=	"flush",
6114 		.data		=	&init_net.ipv6.sysctl.flush_delay,
6115 		.maxlen		=	sizeof(int),
6116 		.mode		=	0200,
6117 		.proc_handler	=	ipv6_sysctl_rtcache_flush
6118 	},
6119 	{
6120 		.procname	=	"gc_thresh",
6121 		.data		=	&ip6_dst_ops_template.gc_thresh,
6122 		.maxlen		=	sizeof(int),
6123 		.mode		=	0644,
6124 		.proc_handler	=	proc_dointvec,
6125 	},
6126 	{
6127 		.procname	=	"max_size",
6128 		.data		=	&init_net.ipv6.sysctl.ip6_rt_max_size,
6129 		.maxlen		=	sizeof(int),
6130 		.mode		=	0644,
6131 		.proc_handler	=	proc_dointvec,
6132 	},
6133 	{
6134 		.procname	=	"gc_min_interval",
6135 		.data		=	&init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6136 		.maxlen		=	sizeof(int),
6137 		.mode		=	0644,
6138 		.proc_handler	=	proc_dointvec_jiffies,
6139 	},
6140 	{
6141 		.procname	=	"gc_timeout",
6142 		.data		=	&init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6143 		.maxlen		=	sizeof(int),
6144 		.mode		=	0644,
6145 		.proc_handler	=	proc_dointvec_jiffies,
6146 	},
6147 	{
6148 		.procname	=	"gc_interval",
6149 		.data		=	&init_net.ipv6.sysctl.ip6_rt_gc_interval,
6150 		.maxlen		=	sizeof(int),
6151 		.mode		=	0644,
6152 		.proc_handler	=	proc_dointvec_jiffies,
6153 	},
6154 	{
6155 		.procname	=	"gc_elasticity",
6156 		.data		=	&init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6157 		.maxlen		=	sizeof(int),
6158 		.mode		=	0644,
6159 		.proc_handler	=	proc_dointvec,
6160 	},
6161 	{
6162 		.procname	=	"mtu_expires",
6163 		.data		=	&init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6164 		.maxlen		=	sizeof(int),
6165 		.mode		=	0644,
6166 		.proc_handler	=	proc_dointvec_jiffies,
6167 	},
6168 	{
6169 		.procname	=	"min_adv_mss",
6170 		.data		=	&init_net.ipv6.sysctl.ip6_rt_min_advmss,
6171 		.maxlen		=	sizeof(int),
6172 		.mode		=	0644,
6173 		.proc_handler	=	proc_dointvec,
6174 	},
6175 	{
6176 		.procname	=	"gc_min_interval_ms",
6177 		.data		=	&init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6178 		.maxlen		=	sizeof(int),
6179 		.mode		=	0644,
6180 		.proc_handler	=	proc_dointvec_ms_jiffies,
6181 	},
6182 	{
6183 		.procname	=	"skip_notify_on_dev_down",
6184 		.data		=	&init_net.ipv6.sysctl.skip_notify_on_dev_down,
6185 		.maxlen		=	sizeof(int),
6186 		.mode		=	0644,
6187 		.proc_handler	=	proc_dointvec_minmax,
6188 		.extra1		=	SYSCTL_ZERO,
6189 		.extra2		=	SYSCTL_ONE,
6190 	},
6191 	{ }
6192 };
6193 
6194 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6195 {
6196 	struct ctl_table *table;
6197 
6198 	table = kmemdup(ipv6_route_table_template,
6199 			sizeof(ipv6_route_table_template),
6200 			GFP_KERNEL);
6201 
6202 	if (table) {
6203 		table[0].data = &net->ipv6.sysctl.flush_delay;
6204 		table[0].extra1 = net;
6205 		table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6206 		table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
6207 		table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6208 		table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6209 		table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6210 		table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6211 		table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6212 		table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6213 		table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6214 		table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6215 
6216 		/* Don't export sysctls to unprivileged users */
6217 		if (net->user_ns != &init_user_ns)
6218 			table[0].procname = NULL;
6219 	}
6220 
6221 	return table;
6222 }
6223 #endif
6224 
6225 static int __net_init ip6_route_net_init(struct net *net)
6226 {
6227 	int ret = -ENOMEM;
6228 
6229 	memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6230 	       sizeof(net->ipv6.ip6_dst_ops));
6231 
6232 	if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6233 		goto out_ip6_dst_ops;
6234 
6235 	net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6236 	if (!net->ipv6.fib6_null_entry)
6237 		goto out_ip6_dst_entries;
6238 	memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6239 	       sizeof(*net->ipv6.fib6_null_entry));
6240 
6241 	net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6242 					   sizeof(*net->ipv6.ip6_null_entry),
6243 					   GFP_KERNEL);
6244 	if (!net->ipv6.ip6_null_entry)
6245 		goto out_fib6_null_entry;
6246 	net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6247 	dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6248 			 ip6_template_metrics, true);
6249 	INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->rt6i_uncached);
6250 
6251 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6252 	net->ipv6.fib6_has_custom_rules = false;
6253 	net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6254 					       sizeof(*net->ipv6.ip6_prohibit_entry),
6255 					       GFP_KERNEL);
6256 	if (!net->ipv6.ip6_prohibit_entry)
6257 		goto out_ip6_null_entry;
6258 	net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6259 	dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6260 			 ip6_template_metrics, true);
6261 	INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->rt6i_uncached);
6262 
6263 	net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6264 					       sizeof(*net->ipv6.ip6_blk_hole_entry),
6265 					       GFP_KERNEL);
6266 	if (!net->ipv6.ip6_blk_hole_entry)
6267 		goto out_ip6_prohibit_entry;
6268 	net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6269 	dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6270 			 ip6_template_metrics, true);
6271 	INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->rt6i_uncached);
6272 #ifdef CONFIG_IPV6_SUBTREES
6273 	net->ipv6.fib6_routes_require_src = 0;
6274 #endif
6275 #endif
6276 
6277 	net->ipv6.sysctl.flush_delay = 0;
6278 	net->ipv6.sysctl.ip6_rt_max_size = 4096;
6279 	net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6280 	net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6281 	net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6282 	net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6283 	net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6284 	net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6285 	net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6286 
6287 	net->ipv6.ip6_rt_gc_expire = 30*HZ;
6288 
6289 	ret = 0;
6290 out:
6291 	return ret;
6292 
6293 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6294 out_ip6_prohibit_entry:
6295 	kfree(net->ipv6.ip6_prohibit_entry);
6296 out_ip6_null_entry:
6297 	kfree(net->ipv6.ip6_null_entry);
6298 #endif
6299 out_fib6_null_entry:
6300 	kfree(net->ipv6.fib6_null_entry);
6301 out_ip6_dst_entries:
6302 	dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6303 out_ip6_dst_ops:
6304 	goto out;
6305 }
6306 
6307 static void __net_exit ip6_route_net_exit(struct net *net)
6308 {
6309 	kfree(net->ipv6.fib6_null_entry);
6310 	kfree(net->ipv6.ip6_null_entry);
6311 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6312 	kfree(net->ipv6.ip6_prohibit_entry);
6313 	kfree(net->ipv6.ip6_blk_hole_entry);
6314 #endif
6315 	dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6316 }
6317 
6318 static int __net_init ip6_route_net_init_late(struct net *net)
6319 {
6320 #ifdef CONFIG_PROC_FS
6321 	proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
6322 			sizeof(struct ipv6_route_iter));
6323 	proc_create_net_single("rt6_stats", 0444, net->proc_net,
6324 			rt6_stats_seq_show, NULL);
6325 #endif
6326 	return 0;
6327 }
6328 
6329 static void __net_exit ip6_route_net_exit_late(struct net *net)
6330 {
6331 #ifdef CONFIG_PROC_FS
6332 	remove_proc_entry("ipv6_route", net->proc_net);
6333 	remove_proc_entry("rt6_stats", net->proc_net);
6334 #endif
6335 }
6336 
6337 static struct pernet_operations ip6_route_net_ops = {
6338 	.init = ip6_route_net_init,
6339 	.exit = ip6_route_net_exit,
6340 };
6341 
6342 static int __net_init ipv6_inetpeer_init(struct net *net)
6343 {
6344 	struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6345 
6346 	if (!bp)
6347 		return -ENOMEM;
6348 	inet_peer_base_init(bp);
6349 	net->ipv6.peers = bp;
6350 	return 0;
6351 }
6352 
6353 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6354 {
6355 	struct inet_peer_base *bp = net->ipv6.peers;
6356 
6357 	net->ipv6.peers = NULL;
6358 	inetpeer_invalidate_tree(bp);
6359 	kfree(bp);
6360 }
6361 
6362 static struct pernet_operations ipv6_inetpeer_ops = {
6363 	.init	=	ipv6_inetpeer_init,
6364 	.exit	=	ipv6_inetpeer_exit,
6365 };
6366 
6367 static struct pernet_operations ip6_route_net_late_ops = {
6368 	.init = ip6_route_net_init_late,
6369 	.exit = ip6_route_net_exit_late,
6370 };
6371 
6372 static struct notifier_block ip6_route_dev_notifier = {
6373 	.notifier_call = ip6_route_dev_notify,
6374 	.priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6375 };
6376 
6377 void __init ip6_route_init_special_entries(void)
6378 {
6379 	/* Registering of the loopback is done before this portion of code,
6380 	 * the loopback reference in rt6_info will not be taken, do it
6381 	 * manually for init_net */
6382 	init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6383 	init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6384 	init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6385   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6386 	init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6387 	init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6388 	init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6389 	init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6390   #endif
6391 }
6392 
6393 int __init ip6_route_init(void)
6394 {
6395 	int ret;
6396 	int cpu;
6397 
6398 	ret = -ENOMEM;
6399 	ip6_dst_ops_template.kmem_cachep =
6400 		kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6401 				  SLAB_HWCACHE_ALIGN, NULL);
6402 	if (!ip6_dst_ops_template.kmem_cachep)
6403 		goto out;
6404 
6405 	ret = dst_entries_init(&ip6_dst_blackhole_ops);
6406 	if (ret)
6407 		goto out_kmem_cache;
6408 
6409 	ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6410 	if (ret)
6411 		goto out_dst_entries;
6412 
6413 	ret = register_pernet_subsys(&ip6_route_net_ops);
6414 	if (ret)
6415 		goto out_register_inetpeer;
6416 
6417 	ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6418 
6419 	ret = fib6_init();
6420 	if (ret)
6421 		goto out_register_subsys;
6422 
6423 	ret = xfrm6_init();
6424 	if (ret)
6425 		goto out_fib6_init;
6426 
6427 	ret = fib6_rules_init();
6428 	if (ret)
6429 		goto xfrm6_init;
6430 
6431 	ret = register_pernet_subsys(&ip6_route_net_late_ops);
6432 	if (ret)
6433 		goto fib6_rules_init;
6434 
6435 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6436 				   inet6_rtm_newroute, NULL, 0);
6437 	if (ret < 0)
6438 		goto out_register_late_subsys;
6439 
6440 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6441 				   inet6_rtm_delroute, NULL, 0);
6442 	if (ret < 0)
6443 		goto out_register_late_subsys;
6444 
6445 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6446 				   inet6_rtm_getroute, NULL,
6447 				   RTNL_FLAG_DOIT_UNLOCKED);
6448 	if (ret < 0)
6449 		goto out_register_late_subsys;
6450 
6451 	ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6452 	if (ret)
6453 		goto out_register_late_subsys;
6454 
6455 	for_each_possible_cpu(cpu) {
6456 		struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6457 
6458 		INIT_LIST_HEAD(&ul->head);
6459 		spin_lock_init(&ul->lock);
6460 	}
6461 
6462 out:
6463 	return ret;
6464 
6465 out_register_late_subsys:
6466 	rtnl_unregister_all(PF_INET6);
6467 	unregister_pernet_subsys(&ip6_route_net_late_ops);
6468 fib6_rules_init:
6469 	fib6_rules_cleanup();
6470 xfrm6_init:
6471 	xfrm6_fini();
6472 out_fib6_init:
6473 	fib6_gc_cleanup();
6474 out_register_subsys:
6475 	unregister_pernet_subsys(&ip6_route_net_ops);
6476 out_register_inetpeer:
6477 	unregister_pernet_subsys(&ipv6_inetpeer_ops);
6478 out_dst_entries:
6479 	dst_entries_destroy(&ip6_dst_blackhole_ops);
6480 out_kmem_cache:
6481 	kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6482 	goto out;
6483 }
6484 
6485 void ip6_route_cleanup(void)
6486 {
6487 	unregister_netdevice_notifier(&ip6_route_dev_notifier);
6488 	unregister_pernet_subsys(&ip6_route_net_late_ops);
6489 	fib6_rules_cleanup();
6490 	xfrm6_fini();
6491 	fib6_gc_cleanup();
6492 	unregister_pernet_subsys(&ipv6_inetpeer_ops);
6493 	unregister_pernet_subsys(&ip6_route_net_ops);
6494 	dst_entries_destroy(&ip6_dst_blackhole_ops);
6495 	kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6496 }
6497