xref: /openbmc/linux/net/ipv6/route.c (revision c67ce71d)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *	Linux INET6 implementation
4  *	FIB front-end.
5  *
6  *	Authors:
7  *	Pedro Roque		<roque@di.fc.ul.pt>
8  */
9 
10 /*	Changes:
11  *
12  *	YOSHIFUJI Hideaki @USAGI
13  *		reworked default router selection.
14  *		- respect outgoing interface
15  *		- select from (probably) reachable routers (i.e.
16  *		routers in REACHABLE, STALE, DELAY or PROBE states).
17  *		- always select the same router if it is (probably)
18  *		reachable.  otherwise, round-robin the list.
19  *	Ville Nuorvala
20  *		Fixed routing subtrees.
21  */
22 
23 #define pr_fmt(fmt) "IPv6: " fmt
24 
25 #include <linux/capability.h>
26 #include <linux/errno.h>
27 #include <linux/export.h>
28 #include <linux/types.h>
29 #include <linux/times.h>
30 #include <linux/socket.h>
31 #include <linux/sockios.h>
32 #include <linux/net.h>
33 #include <linux/route.h>
34 #include <linux/netdevice.h>
35 #include <linux/in6.h>
36 #include <linux/mroute6.h>
37 #include <linux/init.h>
38 #include <linux/if_arp.h>
39 #include <linux/proc_fs.h>
40 #include <linux/seq_file.h>
41 #include <linux/nsproxy.h>
42 #include <linux/slab.h>
43 #include <linux/jhash.h>
44 #include <linux/siphash.h>
45 #include <net/net_namespace.h>
46 #include <net/snmp.h>
47 #include <net/ipv6.h>
48 #include <net/ip6_fib.h>
49 #include <net/ip6_route.h>
50 #include <net/ndisc.h>
51 #include <net/addrconf.h>
52 #include <net/tcp.h>
53 #include <linux/rtnetlink.h>
54 #include <net/dst.h>
55 #include <net/dst_metadata.h>
56 #include <net/xfrm.h>
57 #include <net/netevent.h>
58 #include <net/netlink.h>
59 #include <net/rtnh.h>
60 #include <net/lwtunnel.h>
61 #include <net/ip_tunnels.h>
62 #include <net/l3mdev.h>
63 #include <net/ip.h>
64 #include <linux/uaccess.h>
65 #include <linux/btf_ids.h>
66 
67 #ifdef CONFIG_SYSCTL
68 #include <linux/sysctl.h>
69 #endif
70 
71 static int ip6_rt_type_to_error(u8 fib6_type);
72 
73 #define CREATE_TRACE_POINTS
74 #include <trace/events/fib6.h>
75 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
76 #undef CREATE_TRACE_POINTS
77 
78 enum rt6_nud_state {
79 	RT6_NUD_FAIL_HARD = -3,
80 	RT6_NUD_FAIL_PROBE = -2,
81 	RT6_NUD_FAIL_DO_RR = -1,
82 	RT6_NUD_SUCCEED = 1
83 };
84 
85 INDIRECT_CALLABLE_SCOPE
86 struct dst_entry	*ip6_dst_check(struct dst_entry *dst, u32 cookie);
87 static unsigned int	 ip6_default_advmss(const struct dst_entry *dst);
88 INDIRECT_CALLABLE_SCOPE
89 unsigned int		ip6_mtu(const struct dst_entry *dst);
90 static void		ip6_negative_advice(struct sock *sk,
91 					    struct dst_entry *dst);
92 static void		ip6_dst_destroy(struct dst_entry *);
93 static void		ip6_dst_ifdown(struct dst_entry *,
94 				       struct net_device *dev);
95 static void		 ip6_dst_gc(struct dst_ops *ops);
96 
97 static int		ip6_pkt_discard(struct sk_buff *skb);
98 static int		ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
99 static int		ip6_pkt_prohibit(struct sk_buff *skb);
100 static int		ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb);
101 static void		ip6_link_failure(struct sk_buff *skb);
102 static void		ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
103 					   struct sk_buff *skb, u32 mtu,
104 					   bool confirm_neigh);
105 static void		rt6_do_redirect(struct dst_entry *dst, struct sock *sk,
106 					struct sk_buff *skb);
107 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
108 			   int strict);
109 static size_t rt6_nlmsg_size(struct fib6_info *f6i);
110 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
111 			 struct fib6_info *rt, struct dst_entry *dst,
112 			 struct in6_addr *dest, struct in6_addr *src,
113 			 int iif, int type, u32 portid, u32 seq,
114 			 unsigned int flags);
115 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
116 					   const struct in6_addr *daddr,
117 					   const struct in6_addr *saddr);
118 
119 #ifdef CONFIG_IPV6_ROUTE_INFO
120 static struct fib6_info *rt6_add_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 					   unsigned int pref);
125 static struct fib6_info *rt6_get_route_info(struct net *net,
126 					   const struct in6_addr *prefix, int prefixlen,
127 					   const struct in6_addr *gwaddr,
128 					   struct net_device *dev);
129 #endif
130 
131 struct uncached_list {
132 	spinlock_t		lock;
133 	struct list_head	head;
134 	struct list_head	quarantine;
135 };
136 
137 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
138 
139 void rt6_uncached_list_add(struct rt6_info *rt)
140 {
141 	struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
142 
143 	rt->dst.rt_uncached_list = ul;
144 
145 	spin_lock_bh(&ul->lock);
146 	list_add_tail(&rt->dst.rt_uncached, &ul->head);
147 	spin_unlock_bh(&ul->lock);
148 }
149 
150 void rt6_uncached_list_del(struct rt6_info *rt)
151 {
152 	if (!list_empty(&rt->dst.rt_uncached)) {
153 		struct uncached_list *ul = rt->dst.rt_uncached_list;
154 
155 		spin_lock_bh(&ul->lock);
156 		list_del_init(&rt->dst.rt_uncached);
157 		spin_unlock_bh(&ul->lock);
158 	}
159 }
160 
161 static void rt6_uncached_list_flush_dev(struct net_device *dev)
162 {
163 	int cpu;
164 
165 	for_each_possible_cpu(cpu) {
166 		struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
167 		struct rt6_info *rt, *safe;
168 
169 		if (list_empty(&ul->head))
170 			continue;
171 
172 		spin_lock_bh(&ul->lock);
173 		list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
174 			struct inet6_dev *rt_idev = rt->rt6i_idev;
175 			struct net_device *rt_dev = rt->dst.dev;
176 			bool handled = false;
177 
178 			if (rt_idev && rt_idev->dev == dev) {
179 				rt->rt6i_idev = in6_dev_get(blackhole_netdev);
180 				in6_dev_put(rt_idev);
181 				handled = true;
182 			}
183 
184 			if (rt_dev == dev) {
185 				rt->dst.dev = blackhole_netdev;
186 				netdev_ref_replace(rt_dev, blackhole_netdev,
187 						   &rt->dst.dev_tracker,
188 						   GFP_ATOMIC);
189 				handled = true;
190 			}
191 			if (handled)
192 				list_move(&rt->dst.rt_uncached,
193 					  &ul->quarantine);
194 		}
195 		spin_unlock_bh(&ul->lock);
196 	}
197 }
198 
199 static inline const void *choose_neigh_daddr(const struct in6_addr *p,
200 					     struct sk_buff *skb,
201 					     const void *daddr)
202 {
203 	if (!ipv6_addr_any(p))
204 		return (const void *) p;
205 	else if (skb)
206 		return &ipv6_hdr(skb)->daddr;
207 	return daddr;
208 }
209 
210 struct neighbour *ip6_neigh_lookup(const struct in6_addr *gw,
211 				   struct net_device *dev,
212 				   struct sk_buff *skb,
213 				   const void *daddr)
214 {
215 	struct neighbour *n;
216 
217 	daddr = choose_neigh_daddr(gw, skb, daddr);
218 	n = __ipv6_neigh_lookup(dev, daddr);
219 	if (n)
220 		return n;
221 
222 	n = neigh_create(&nd_tbl, daddr, dev);
223 	return IS_ERR(n) ? NULL : n;
224 }
225 
226 static struct neighbour *ip6_dst_neigh_lookup(const struct dst_entry *dst,
227 					      struct sk_buff *skb,
228 					      const void *daddr)
229 {
230 	const struct rt6_info *rt = container_of(dst, struct rt6_info, dst);
231 
232 	return ip6_neigh_lookup(rt6_nexthop(rt, &in6addr_any),
233 				dst->dev, skb, daddr);
234 }
235 
236 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
237 {
238 	struct net_device *dev = dst->dev;
239 	struct rt6_info *rt = (struct rt6_info *)dst;
240 
241 	daddr = choose_neigh_daddr(rt6_nexthop(rt, &in6addr_any), NULL, daddr);
242 	if (!daddr)
243 		return;
244 	if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
245 		return;
246 	if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
247 		return;
248 	__ipv6_confirm_neigh(dev, daddr);
249 }
250 
251 static struct dst_ops ip6_dst_ops_template = {
252 	.family			=	AF_INET6,
253 	.gc			=	ip6_dst_gc,
254 	.gc_thresh		=	1024,
255 	.check			=	ip6_dst_check,
256 	.default_advmss		=	ip6_default_advmss,
257 	.mtu			=	ip6_mtu,
258 	.cow_metrics		=	dst_cow_metrics_generic,
259 	.destroy		=	ip6_dst_destroy,
260 	.ifdown			=	ip6_dst_ifdown,
261 	.negative_advice	=	ip6_negative_advice,
262 	.link_failure		=	ip6_link_failure,
263 	.update_pmtu		=	ip6_rt_update_pmtu,
264 	.redirect		=	rt6_do_redirect,
265 	.local_out		=	__ip6_local_out,
266 	.neigh_lookup		=	ip6_dst_neigh_lookup,
267 	.confirm_neigh		=	ip6_confirm_neigh,
268 };
269 
270 static struct dst_ops ip6_dst_blackhole_ops = {
271 	.family			= AF_INET6,
272 	.default_advmss		= ip6_default_advmss,
273 	.neigh_lookup		= ip6_dst_neigh_lookup,
274 	.check			= ip6_dst_check,
275 	.destroy		= ip6_dst_destroy,
276 	.cow_metrics		= dst_cow_metrics_generic,
277 	.update_pmtu		= dst_blackhole_update_pmtu,
278 	.redirect		= dst_blackhole_redirect,
279 	.mtu			= dst_blackhole_mtu,
280 };
281 
282 static const u32 ip6_template_metrics[RTAX_MAX] = {
283 	[RTAX_HOPLIMIT - 1] = 0,
284 };
285 
286 static const struct fib6_info fib6_null_entry_template = {
287 	.fib6_flags	= (RTF_REJECT | RTF_NONEXTHOP),
288 	.fib6_protocol  = RTPROT_KERNEL,
289 	.fib6_metric	= ~(u32)0,
290 	.fib6_ref	= REFCOUNT_INIT(1),
291 	.fib6_type	= RTN_UNREACHABLE,
292 	.fib6_metrics	= (struct dst_metrics *)&dst_default_metrics,
293 };
294 
295 static const struct rt6_info ip6_null_entry_template = {
296 	.dst = {
297 		.__rcuref	= RCUREF_INIT(1),
298 		.__use		= 1,
299 		.obsolete	= DST_OBSOLETE_FORCE_CHK,
300 		.error		= -ENETUNREACH,
301 		.input		= ip6_pkt_discard,
302 		.output		= ip6_pkt_discard_out,
303 	},
304 	.rt6i_flags	= (RTF_REJECT | RTF_NONEXTHOP),
305 };
306 
307 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
308 
309 static const struct rt6_info ip6_prohibit_entry_template = {
310 	.dst = {
311 		.__rcuref	= RCUREF_INIT(1),
312 		.__use		= 1,
313 		.obsolete	= DST_OBSOLETE_FORCE_CHK,
314 		.error		= -EACCES,
315 		.input		= ip6_pkt_prohibit,
316 		.output		= ip6_pkt_prohibit_out,
317 	},
318 	.rt6i_flags	= (RTF_REJECT | RTF_NONEXTHOP),
319 };
320 
321 static const struct rt6_info ip6_blk_hole_entry_template = {
322 	.dst = {
323 		.__rcuref	= RCUREF_INIT(1),
324 		.__use		= 1,
325 		.obsolete	= DST_OBSOLETE_FORCE_CHK,
326 		.error		= -EINVAL,
327 		.input		= dst_discard,
328 		.output		= dst_discard_out,
329 	},
330 	.rt6i_flags	= (RTF_REJECT | RTF_NONEXTHOP),
331 };
332 
333 #endif
334 
335 static void rt6_info_init(struct rt6_info *rt)
336 {
337 	memset_after(rt, 0, dst);
338 }
339 
340 /* allocate dst with ip6_dst_ops */
341 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
342 			       int flags)
343 {
344 	struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
345 					1, DST_OBSOLETE_FORCE_CHK, flags);
346 
347 	if (rt) {
348 		rt6_info_init(rt);
349 		atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
350 	}
351 
352 	return rt;
353 }
354 EXPORT_SYMBOL(ip6_dst_alloc);
355 
356 static void ip6_dst_destroy(struct dst_entry *dst)
357 {
358 	struct rt6_info *rt = (struct rt6_info *)dst;
359 	struct fib6_info *from;
360 	struct inet6_dev *idev;
361 
362 	ip_dst_metrics_put(dst);
363 	rt6_uncached_list_del(rt);
364 
365 	idev = rt->rt6i_idev;
366 	if (idev) {
367 		rt->rt6i_idev = NULL;
368 		in6_dev_put(idev);
369 	}
370 
371 	from = xchg((__force struct fib6_info **)&rt->from, NULL);
372 	fib6_info_release(from);
373 }
374 
375 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
376 {
377 	struct rt6_info *rt = (struct rt6_info *)dst;
378 	struct inet6_dev *idev = rt->rt6i_idev;
379 
380 	if (idev && idev->dev != blackhole_netdev) {
381 		struct inet6_dev *blackhole_idev = in6_dev_get(blackhole_netdev);
382 
383 		if (blackhole_idev) {
384 			rt->rt6i_idev = blackhole_idev;
385 			in6_dev_put(idev);
386 		}
387 	}
388 }
389 
390 static bool __rt6_check_expired(const struct rt6_info *rt)
391 {
392 	if (rt->rt6i_flags & RTF_EXPIRES)
393 		return time_after(jiffies, rt->dst.expires);
394 	else
395 		return false;
396 }
397 
398 static bool rt6_check_expired(const struct rt6_info *rt)
399 {
400 	struct fib6_info *from;
401 
402 	from = rcu_dereference(rt->from);
403 
404 	if (rt->rt6i_flags & RTF_EXPIRES) {
405 		if (time_after(jiffies, rt->dst.expires))
406 			return true;
407 	} else if (from) {
408 		return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
409 			fib6_check_expired(from);
410 	}
411 	return false;
412 }
413 
414 void fib6_select_path(const struct net *net, struct fib6_result *res,
415 		      struct flowi6 *fl6, int oif, bool have_oif_match,
416 		      const struct sk_buff *skb, int strict)
417 {
418 	struct fib6_info *sibling, *next_sibling;
419 	struct fib6_info *match = res->f6i;
420 
421 	if (!match->nh && (!match->fib6_nsiblings || have_oif_match))
422 		goto out;
423 
424 	if (match->nh && have_oif_match && res->nh)
425 		return;
426 
427 	if (skb)
428 		IP6CB(skb)->flags |= IP6SKB_MULTIPATH;
429 
430 	/* We might have already computed the hash for ICMPv6 errors. In such
431 	 * case it will always be non-zero. Otherwise now is the time to do it.
432 	 */
433 	if (!fl6->mp_hash &&
434 	    (!match->nh || nexthop_is_multipath(match->nh)))
435 		fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
436 
437 	if (unlikely(match->nh)) {
438 		nexthop_path_fib6_result(res, fl6->mp_hash);
439 		return;
440 	}
441 
442 	if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
443 		goto out;
444 
445 	list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
446 				 fib6_siblings) {
447 		const struct fib6_nh *nh = sibling->fib6_nh;
448 		int nh_upper_bound;
449 
450 		nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
451 		if (fl6->mp_hash > nh_upper_bound)
452 			continue;
453 		if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
454 			break;
455 		match = sibling;
456 		break;
457 	}
458 
459 out:
460 	res->f6i = match;
461 	res->nh = match->fib6_nh;
462 }
463 
464 /*
465  *	Route lookup. rcu_read_lock() should be held.
466  */
467 
468 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
469 			       const struct in6_addr *saddr, int oif, int flags)
470 {
471 	const struct net_device *dev;
472 
473 	if (nh->fib_nh_flags & RTNH_F_DEAD)
474 		return false;
475 
476 	dev = nh->fib_nh_dev;
477 	if (oif) {
478 		if (dev->ifindex == oif)
479 			return true;
480 	} else {
481 		if (ipv6_chk_addr(net, saddr, dev,
482 				  flags & RT6_LOOKUP_F_IFACE))
483 			return true;
484 	}
485 
486 	return false;
487 }
488 
489 struct fib6_nh_dm_arg {
490 	struct net		*net;
491 	const struct in6_addr	*saddr;
492 	int			oif;
493 	int			flags;
494 	struct fib6_nh		*nh;
495 };
496 
497 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
498 {
499 	struct fib6_nh_dm_arg *arg = _arg;
500 
501 	arg->nh = nh;
502 	return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
503 				  arg->flags);
504 }
505 
506 /* returns fib6_nh from nexthop or NULL */
507 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
508 					struct fib6_result *res,
509 					const struct in6_addr *saddr,
510 					int oif, int flags)
511 {
512 	struct fib6_nh_dm_arg arg = {
513 		.net   = net,
514 		.saddr = saddr,
515 		.oif   = oif,
516 		.flags = flags,
517 	};
518 
519 	if (nexthop_is_blackhole(nh))
520 		return NULL;
521 
522 	if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
523 		return arg.nh;
524 
525 	return NULL;
526 }
527 
528 static void rt6_device_match(struct net *net, struct fib6_result *res,
529 			     const struct in6_addr *saddr, int oif, int flags)
530 {
531 	struct fib6_info *f6i = res->f6i;
532 	struct fib6_info *spf6i;
533 	struct fib6_nh *nh;
534 
535 	if (!oif && ipv6_addr_any(saddr)) {
536 		if (unlikely(f6i->nh)) {
537 			nh = nexthop_fib6_nh(f6i->nh);
538 			if (nexthop_is_blackhole(f6i->nh))
539 				goto out_blackhole;
540 		} else {
541 			nh = f6i->fib6_nh;
542 		}
543 		if (!(nh->fib_nh_flags & RTNH_F_DEAD))
544 			goto out;
545 	}
546 
547 	for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
548 		bool matched = false;
549 
550 		if (unlikely(spf6i->nh)) {
551 			nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
552 					      oif, flags);
553 			if (nh)
554 				matched = true;
555 		} else {
556 			nh = spf6i->fib6_nh;
557 			if (__rt6_device_match(net, nh, saddr, oif, flags))
558 				matched = true;
559 		}
560 		if (matched) {
561 			res->f6i = spf6i;
562 			goto out;
563 		}
564 	}
565 
566 	if (oif && flags & RT6_LOOKUP_F_IFACE) {
567 		res->f6i = net->ipv6.fib6_null_entry;
568 		nh = res->f6i->fib6_nh;
569 		goto out;
570 	}
571 
572 	if (unlikely(f6i->nh)) {
573 		nh = nexthop_fib6_nh(f6i->nh);
574 		if (nexthop_is_blackhole(f6i->nh))
575 			goto out_blackhole;
576 	} else {
577 		nh = f6i->fib6_nh;
578 	}
579 
580 	if (nh->fib_nh_flags & RTNH_F_DEAD) {
581 		res->f6i = net->ipv6.fib6_null_entry;
582 		nh = res->f6i->fib6_nh;
583 	}
584 out:
585 	res->nh = nh;
586 	res->fib6_type = res->f6i->fib6_type;
587 	res->fib6_flags = res->f6i->fib6_flags;
588 	return;
589 
590 out_blackhole:
591 	res->fib6_flags |= RTF_REJECT;
592 	res->fib6_type = RTN_BLACKHOLE;
593 	res->nh = nh;
594 }
595 
596 #ifdef CONFIG_IPV6_ROUTER_PREF
597 struct __rt6_probe_work {
598 	struct work_struct work;
599 	struct in6_addr target;
600 	struct net_device *dev;
601 	netdevice_tracker dev_tracker;
602 };
603 
604 static void rt6_probe_deferred(struct work_struct *w)
605 {
606 	struct in6_addr mcaddr;
607 	struct __rt6_probe_work *work =
608 		container_of(w, struct __rt6_probe_work, work);
609 
610 	addrconf_addr_solict_mult(&work->target, &mcaddr);
611 	ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
612 	netdev_put(work->dev, &work->dev_tracker);
613 	kfree(work);
614 }
615 
616 static void rt6_probe(struct fib6_nh *fib6_nh)
617 {
618 	struct __rt6_probe_work *work = NULL;
619 	const struct in6_addr *nh_gw;
620 	unsigned long last_probe;
621 	struct neighbour *neigh;
622 	struct net_device *dev;
623 	struct inet6_dev *idev;
624 
625 	/*
626 	 * Okay, this does not seem to be appropriate
627 	 * for now, however, we need to check if it
628 	 * is really so; aka Router Reachability Probing.
629 	 *
630 	 * Router Reachability Probe MUST be rate-limited
631 	 * to no more than one per minute.
632 	 */
633 	if (!fib6_nh->fib_nh_gw_family)
634 		return;
635 
636 	nh_gw = &fib6_nh->fib_nh_gw6;
637 	dev = fib6_nh->fib_nh_dev;
638 	rcu_read_lock();
639 	last_probe = READ_ONCE(fib6_nh->last_probe);
640 	idev = __in6_dev_get(dev);
641 	if (!idev)
642 		goto out;
643 	neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
644 	if (neigh) {
645 		if (READ_ONCE(neigh->nud_state) & NUD_VALID)
646 			goto out;
647 
648 		write_lock_bh(&neigh->lock);
649 		if (!(neigh->nud_state & NUD_VALID) &&
650 		    time_after(jiffies,
651 			       neigh->updated + idev->cnf.rtr_probe_interval)) {
652 			work = kmalloc(sizeof(*work), GFP_ATOMIC);
653 			if (work)
654 				__neigh_set_probe_once(neigh);
655 		}
656 		write_unlock_bh(&neigh->lock);
657 	} else if (time_after(jiffies, last_probe +
658 				       idev->cnf.rtr_probe_interval)) {
659 		work = kmalloc(sizeof(*work), GFP_ATOMIC);
660 	}
661 
662 	if (!work || cmpxchg(&fib6_nh->last_probe,
663 			     last_probe, jiffies) != last_probe) {
664 		kfree(work);
665 	} else {
666 		INIT_WORK(&work->work, rt6_probe_deferred);
667 		work->target = *nh_gw;
668 		netdev_hold(dev, &work->dev_tracker, GFP_ATOMIC);
669 		work->dev = dev;
670 		schedule_work(&work->work);
671 	}
672 
673 out:
674 	rcu_read_unlock();
675 }
676 #else
677 static inline void rt6_probe(struct fib6_nh *fib6_nh)
678 {
679 }
680 #endif
681 
682 /*
683  * Default Router Selection (RFC 2461 6.3.6)
684  */
685 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
686 {
687 	enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
688 	struct neighbour *neigh;
689 
690 	rcu_read_lock();
691 	neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
692 					  &fib6_nh->fib_nh_gw6);
693 	if (neigh) {
694 		u8 nud_state = READ_ONCE(neigh->nud_state);
695 
696 		if (nud_state & NUD_VALID)
697 			ret = RT6_NUD_SUCCEED;
698 #ifdef CONFIG_IPV6_ROUTER_PREF
699 		else if (!(nud_state & NUD_FAILED))
700 			ret = RT6_NUD_SUCCEED;
701 		else
702 			ret = RT6_NUD_FAIL_PROBE;
703 #endif
704 	} else {
705 		ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
706 		      RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
707 	}
708 	rcu_read_unlock();
709 
710 	return ret;
711 }
712 
713 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
714 			   int strict)
715 {
716 	int m = 0;
717 
718 	if (!oif || nh->fib_nh_dev->ifindex == oif)
719 		m = 2;
720 
721 	if (!m && (strict & RT6_LOOKUP_F_IFACE))
722 		return RT6_NUD_FAIL_HARD;
723 #ifdef CONFIG_IPV6_ROUTER_PREF
724 	m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
725 #endif
726 	if ((strict & RT6_LOOKUP_F_REACHABLE) &&
727 	    !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
728 		int n = rt6_check_neigh(nh);
729 		if (n < 0)
730 			return n;
731 	}
732 	return m;
733 }
734 
735 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
736 		       int oif, int strict, int *mpri, bool *do_rr)
737 {
738 	bool match_do_rr = false;
739 	bool rc = false;
740 	int m;
741 
742 	if (nh->fib_nh_flags & RTNH_F_DEAD)
743 		goto out;
744 
745 	if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
746 	    nh->fib_nh_flags & RTNH_F_LINKDOWN &&
747 	    !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
748 		goto out;
749 
750 	m = rt6_score_route(nh, fib6_flags, oif, strict);
751 	if (m == RT6_NUD_FAIL_DO_RR) {
752 		match_do_rr = true;
753 		m = 0; /* lowest valid score */
754 	} else if (m == RT6_NUD_FAIL_HARD) {
755 		goto out;
756 	}
757 
758 	if (strict & RT6_LOOKUP_F_REACHABLE)
759 		rt6_probe(nh);
760 
761 	/* note that m can be RT6_NUD_FAIL_PROBE at this point */
762 	if (m > *mpri) {
763 		*do_rr = match_do_rr;
764 		*mpri = m;
765 		rc = true;
766 	}
767 out:
768 	return rc;
769 }
770 
771 struct fib6_nh_frl_arg {
772 	u32		flags;
773 	int		oif;
774 	int		strict;
775 	int		*mpri;
776 	bool		*do_rr;
777 	struct fib6_nh	*nh;
778 };
779 
780 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
781 {
782 	struct fib6_nh_frl_arg *arg = _arg;
783 
784 	arg->nh = nh;
785 	return find_match(nh, arg->flags, arg->oif, arg->strict,
786 			  arg->mpri, arg->do_rr);
787 }
788 
789 static void __find_rr_leaf(struct fib6_info *f6i_start,
790 			   struct fib6_info *nomatch, u32 metric,
791 			   struct fib6_result *res, struct fib6_info **cont,
792 			   int oif, int strict, bool *do_rr, int *mpri)
793 {
794 	struct fib6_info *f6i;
795 
796 	for (f6i = f6i_start;
797 	     f6i && f6i != nomatch;
798 	     f6i = rcu_dereference(f6i->fib6_next)) {
799 		bool matched = false;
800 		struct fib6_nh *nh;
801 
802 		if (cont && f6i->fib6_metric != metric) {
803 			*cont = f6i;
804 			return;
805 		}
806 
807 		if (fib6_check_expired(f6i))
808 			continue;
809 
810 		if (unlikely(f6i->nh)) {
811 			struct fib6_nh_frl_arg arg = {
812 				.flags  = f6i->fib6_flags,
813 				.oif    = oif,
814 				.strict = strict,
815 				.mpri   = mpri,
816 				.do_rr  = do_rr
817 			};
818 
819 			if (nexthop_is_blackhole(f6i->nh)) {
820 				res->fib6_flags = RTF_REJECT;
821 				res->fib6_type = RTN_BLACKHOLE;
822 				res->f6i = f6i;
823 				res->nh = nexthop_fib6_nh(f6i->nh);
824 				return;
825 			}
826 			if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
827 						     &arg)) {
828 				matched = true;
829 				nh = arg.nh;
830 			}
831 		} else {
832 			nh = f6i->fib6_nh;
833 			if (find_match(nh, f6i->fib6_flags, oif, strict,
834 				       mpri, do_rr))
835 				matched = true;
836 		}
837 		if (matched) {
838 			res->f6i = f6i;
839 			res->nh = nh;
840 			res->fib6_flags = f6i->fib6_flags;
841 			res->fib6_type = f6i->fib6_type;
842 		}
843 	}
844 }
845 
846 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
847 			 struct fib6_info *rr_head, int oif, int strict,
848 			 bool *do_rr, struct fib6_result *res)
849 {
850 	u32 metric = rr_head->fib6_metric;
851 	struct fib6_info *cont = NULL;
852 	int mpri = -1;
853 
854 	__find_rr_leaf(rr_head, NULL, metric, res, &cont,
855 		       oif, strict, do_rr, &mpri);
856 
857 	__find_rr_leaf(leaf, rr_head, metric, res, &cont,
858 		       oif, strict, do_rr, &mpri);
859 
860 	if (res->f6i || !cont)
861 		return;
862 
863 	__find_rr_leaf(cont, NULL, metric, res, NULL,
864 		       oif, strict, do_rr, &mpri);
865 }
866 
867 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
868 		       struct fib6_result *res, int strict)
869 {
870 	struct fib6_info *leaf = rcu_dereference(fn->leaf);
871 	struct fib6_info *rt0;
872 	bool do_rr = false;
873 	int key_plen;
874 
875 	/* make sure this function or its helpers sets f6i */
876 	res->f6i = NULL;
877 
878 	if (!leaf || leaf == net->ipv6.fib6_null_entry)
879 		goto out;
880 
881 	rt0 = rcu_dereference(fn->rr_ptr);
882 	if (!rt0)
883 		rt0 = leaf;
884 
885 	/* Double check to make sure fn is not an intermediate node
886 	 * and fn->leaf does not points to its child's leaf
887 	 * (This might happen if all routes under fn are deleted from
888 	 * the tree and fib6_repair_tree() is called on the node.)
889 	 */
890 	key_plen = rt0->fib6_dst.plen;
891 #ifdef CONFIG_IPV6_SUBTREES
892 	if (rt0->fib6_src.plen)
893 		key_plen = rt0->fib6_src.plen;
894 #endif
895 	if (fn->fn_bit != key_plen)
896 		goto out;
897 
898 	find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
899 	if (do_rr) {
900 		struct fib6_info *next = rcu_dereference(rt0->fib6_next);
901 
902 		/* no entries matched; do round-robin */
903 		if (!next || next->fib6_metric != rt0->fib6_metric)
904 			next = leaf;
905 
906 		if (next != rt0) {
907 			spin_lock_bh(&leaf->fib6_table->tb6_lock);
908 			/* make sure next is not being deleted from the tree */
909 			if (next->fib6_node)
910 				rcu_assign_pointer(fn->rr_ptr, next);
911 			spin_unlock_bh(&leaf->fib6_table->tb6_lock);
912 		}
913 	}
914 
915 out:
916 	if (!res->f6i) {
917 		res->f6i = net->ipv6.fib6_null_entry;
918 		res->nh = res->f6i->fib6_nh;
919 		res->fib6_flags = res->f6i->fib6_flags;
920 		res->fib6_type = res->f6i->fib6_type;
921 	}
922 }
923 
924 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
925 {
926 	return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
927 	       res->nh->fib_nh_gw_family;
928 }
929 
930 #ifdef CONFIG_IPV6_ROUTE_INFO
931 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
932 		  const struct in6_addr *gwaddr)
933 {
934 	struct net *net = dev_net(dev);
935 	struct route_info *rinfo = (struct route_info *) opt;
936 	struct in6_addr prefix_buf, *prefix;
937 	unsigned int pref;
938 	unsigned long lifetime;
939 	struct fib6_info *rt;
940 
941 	if (len < sizeof(struct route_info)) {
942 		return -EINVAL;
943 	}
944 
945 	/* Sanity check for prefix_len and length */
946 	if (rinfo->length > 3) {
947 		return -EINVAL;
948 	} else if (rinfo->prefix_len > 128) {
949 		return -EINVAL;
950 	} else if (rinfo->prefix_len > 64) {
951 		if (rinfo->length < 2) {
952 			return -EINVAL;
953 		}
954 	} else if (rinfo->prefix_len > 0) {
955 		if (rinfo->length < 1) {
956 			return -EINVAL;
957 		}
958 	}
959 
960 	pref = rinfo->route_pref;
961 	if (pref == ICMPV6_ROUTER_PREF_INVALID)
962 		return -EINVAL;
963 
964 	lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
965 
966 	if (rinfo->length == 3)
967 		prefix = (struct in6_addr *)rinfo->prefix;
968 	else {
969 		/* this function is safe */
970 		ipv6_addr_prefix(&prefix_buf,
971 				 (struct in6_addr *)rinfo->prefix,
972 				 rinfo->prefix_len);
973 		prefix = &prefix_buf;
974 	}
975 
976 	if (rinfo->prefix_len == 0)
977 		rt = rt6_get_dflt_router(net, gwaddr, dev);
978 	else
979 		rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
980 					gwaddr, dev);
981 
982 	if (rt && !lifetime) {
983 		ip6_del_rt(net, rt, false);
984 		rt = NULL;
985 	}
986 
987 	if (!rt && lifetime)
988 		rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
989 					dev, pref);
990 	else if (rt)
991 		rt->fib6_flags = RTF_ROUTEINFO |
992 				 (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
993 
994 	if (rt) {
995 		if (!addrconf_finite_timeout(lifetime))
996 			fib6_clean_expires(rt);
997 		else
998 			fib6_set_expires(rt, jiffies + HZ * lifetime);
999 
1000 		fib6_info_release(rt);
1001 	}
1002 	return 0;
1003 }
1004 #endif
1005 
1006 /*
1007  *	Misc support functions
1008  */
1009 
1010 /* called with rcu_lock held */
1011 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1012 {
1013 	struct net_device *dev = res->nh->fib_nh_dev;
1014 
1015 	if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1016 		/* for copies of local routes, dst->dev needs to be the
1017 		 * device if it is a master device, the master device if
1018 		 * device is enslaved, and the loopback as the default
1019 		 */
1020 		if (netif_is_l3_slave(dev) &&
1021 		    !rt6_need_strict(&res->f6i->fib6_dst.addr))
1022 			dev = l3mdev_master_dev_rcu(dev);
1023 		else if (!netif_is_l3_master(dev))
1024 			dev = dev_net(dev)->loopback_dev;
1025 		/* last case is netif_is_l3_master(dev) is true in which
1026 		 * case we want dev returned to be dev
1027 		 */
1028 	}
1029 
1030 	return dev;
1031 }
1032 
1033 static const int fib6_prop[RTN_MAX + 1] = {
1034 	[RTN_UNSPEC]	= 0,
1035 	[RTN_UNICAST]	= 0,
1036 	[RTN_LOCAL]	= 0,
1037 	[RTN_BROADCAST]	= 0,
1038 	[RTN_ANYCAST]	= 0,
1039 	[RTN_MULTICAST]	= 0,
1040 	[RTN_BLACKHOLE]	= -EINVAL,
1041 	[RTN_UNREACHABLE] = -EHOSTUNREACH,
1042 	[RTN_PROHIBIT]	= -EACCES,
1043 	[RTN_THROW]	= -EAGAIN,
1044 	[RTN_NAT]	= -EINVAL,
1045 	[RTN_XRESOLVE]	= -EINVAL,
1046 };
1047 
1048 static int ip6_rt_type_to_error(u8 fib6_type)
1049 {
1050 	return fib6_prop[fib6_type];
1051 }
1052 
1053 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1054 {
1055 	unsigned short flags = 0;
1056 
1057 	if (rt->dst_nocount)
1058 		flags |= DST_NOCOUNT;
1059 	if (rt->dst_nopolicy)
1060 		flags |= DST_NOPOLICY;
1061 
1062 	return flags;
1063 }
1064 
1065 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1066 {
1067 	rt->dst.error = ip6_rt_type_to_error(fib6_type);
1068 
1069 	switch (fib6_type) {
1070 	case RTN_BLACKHOLE:
1071 		rt->dst.output = dst_discard_out;
1072 		rt->dst.input = dst_discard;
1073 		break;
1074 	case RTN_PROHIBIT:
1075 		rt->dst.output = ip6_pkt_prohibit_out;
1076 		rt->dst.input = ip6_pkt_prohibit;
1077 		break;
1078 	case RTN_THROW:
1079 	case RTN_UNREACHABLE:
1080 	default:
1081 		rt->dst.output = ip6_pkt_discard_out;
1082 		rt->dst.input = ip6_pkt_discard;
1083 		break;
1084 	}
1085 }
1086 
1087 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1088 {
1089 	struct fib6_info *f6i = res->f6i;
1090 
1091 	if (res->fib6_flags & RTF_REJECT) {
1092 		ip6_rt_init_dst_reject(rt, res->fib6_type);
1093 		return;
1094 	}
1095 
1096 	rt->dst.error = 0;
1097 	rt->dst.output = ip6_output;
1098 
1099 	if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1100 		rt->dst.input = ip6_input;
1101 	} else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1102 		rt->dst.input = ip6_mc_input;
1103 	} else {
1104 		rt->dst.input = ip6_forward;
1105 	}
1106 
1107 	if (res->nh->fib_nh_lws) {
1108 		rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1109 		lwtunnel_set_redirect(&rt->dst);
1110 	}
1111 
1112 	rt->dst.lastuse = jiffies;
1113 }
1114 
1115 /* Caller must already hold reference to @from */
1116 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1117 {
1118 	rt->rt6i_flags &= ~RTF_EXPIRES;
1119 	rcu_assign_pointer(rt->from, from);
1120 	ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1121 }
1122 
1123 /* Caller must already hold reference to f6i in result */
1124 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1125 {
1126 	const struct fib6_nh *nh = res->nh;
1127 	const struct net_device *dev = nh->fib_nh_dev;
1128 	struct fib6_info *f6i = res->f6i;
1129 
1130 	ip6_rt_init_dst(rt, res);
1131 
1132 	rt->rt6i_dst = f6i->fib6_dst;
1133 	rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1134 	rt->rt6i_flags = res->fib6_flags;
1135 	if (nh->fib_nh_gw_family) {
1136 		rt->rt6i_gateway = nh->fib_nh_gw6;
1137 		rt->rt6i_flags |= RTF_GATEWAY;
1138 	}
1139 	rt6_set_from(rt, f6i);
1140 #ifdef CONFIG_IPV6_SUBTREES
1141 	rt->rt6i_src = f6i->fib6_src;
1142 #endif
1143 }
1144 
1145 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1146 					struct in6_addr *saddr)
1147 {
1148 	struct fib6_node *pn, *sn;
1149 	while (1) {
1150 		if (fn->fn_flags & RTN_TL_ROOT)
1151 			return NULL;
1152 		pn = rcu_dereference(fn->parent);
1153 		sn = FIB6_SUBTREE(pn);
1154 		if (sn && sn != fn)
1155 			fn = fib6_node_lookup(sn, NULL, saddr);
1156 		else
1157 			fn = pn;
1158 		if (fn->fn_flags & RTN_RTINFO)
1159 			return fn;
1160 	}
1161 }
1162 
1163 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1164 {
1165 	struct rt6_info *rt = *prt;
1166 
1167 	if (dst_hold_safe(&rt->dst))
1168 		return true;
1169 	if (net) {
1170 		rt = net->ipv6.ip6_null_entry;
1171 		dst_hold(&rt->dst);
1172 	} else {
1173 		rt = NULL;
1174 	}
1175 	*prt = rt;
1176 	return false;
1177 }
1178 
1179 /* called with rcu_lock held */
1180 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1181 {
1182 	struct net_device *dev = res->nh->fib_nh_dev;
1183 	struct fib6_info *f6i = res->f6i;
1184 	unsigned short flags;
1185 	struct rt6_info *nrt;
1186 
1187 	if (!fib6_info_hold_safe(f6i))
1188 		goto fallback;
1189 
1190 	flags = fib6_info_dst_flags(f6i);
1191 	nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1192 	if (!nrt) {
1193 		fib6_info_release(f6i);
1194 		goto fallback;
1195 	}
1196 
1197 	ip6_rt_copy_init(nrt, res);
1198 	return nrt;
1199 
1200 fallback:
1201 	nrt = dev_net(dev)->ipv6.ip6_null_entry;
1202 	dst_hold(&nrt->dst);
1203 	return nrt;
1204 }
1205 
1206 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_lookup(struct net *net,
1207 					     struct fib6_table *table,
1208 					     struct flowi6 *fl6,
1209 					     const struct sk_buff *skb,
1210 					     int flags)
1211 {
1212 	struct fib6_result res = {};
1213 	struct fib6_node *fn;
1214 	struct rt6_info *rt;
1215 
1216 	rcu_read_lock();
1217 	fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1218 restart:
1219 	res.f6i = rcu_dereference(fn->leaf);
1220 	if (!res.f6i)
1221 		res.f6i = net->ipv6.fib6_null_entry;
1222 	else
1223 		rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1224 				 flags);
1225 
1226 	if (res.f6i == net->ipv6.fib6_null_entry) {
1227 		fn = fib6_backtrack(fn, &fl6->saddr);
1228 		if (fn)
1229 			goto restart;
1230 
1231 		rt = net->ipv6.ip6_null_entry;
1232 		dst_hold(&rt->dst);
1233 		goto out;
1234 	} else if (res.fib6_flags & RTF_REJECT) {
1235 		goto do_create;
1236 	}
1237 
1238 	fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1239 			 fl6->flowi6_oif != 0, skb, flags);
1240 
1241 	/* Search through exception table */
1242 	rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1243 	if (rt) {
1244 		if (ip6_hold_safe(net, &rt))
1245 			dst_use_noref(&rt->dst, jiffies);
1246 	} else {
1247 do_create:
1248 		rt = ip6_create_rt_rcu(&res);
1249 	}
1250 
1251 out:
1252 	trace_fib6_table_lookup(net, &res, table, fl6);
1253 
1254 	rcu_read_unlock();
1255 
1256 	return rt;
1257 }
1258 
1259 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1260 				   const struct sk_buff *skb, int flags)
1261 {
1262 	return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1263 }
1264 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1265 
1266 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1267 			    const struct in6_addr *saddr, int oif,
1268 			    const struct sk_buff *skb, int strict)
1269 {
1270 	struct flowi6 fl6 = {
1271 		.flowi6_oif = oif,
1272 		.daddr = *daddr,
1273 	};
1274 	struct dst_entry *dst;
1275 	int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1276 
1277 	if (saddr) {
1278 		memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1279 		flags |= RT6_LOOKUP_F_HAS_SADDR;
1280 	}
1281 
1282 	dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1283 	if (dst->error == 0)
1284 		return (struct rt6_info *) dst;
1285 
1286 	dst_release(dst);
1287 
1288 	return NULL;
1289 }
1290 EXPORT_SYMBOL(rt6_lookup);
1291 
1292 /* ip6_ins_rt is called with FREE table->tb6_lock.
1293  * It takes new route entry, the addition fails by any reason the
1294  * route is released.
1295  * Caller must hold dst before calling it.
1296  */
1297 
1298 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1299 			struct netlink_ext_ack *extack)
1300 {
1301 	int err;
1302 	struct fib6_table *table;
1303 
1304 	table = rt->fib6_table;
1305 	spin_lock_bh(&table->tb6_lock);
1306 	err = fib6_add(&table->tb6_root, rt, info, extack);
1307 	spin_unlock_bh(&table->tb6_lock);
1308 
1309 	return err;
1310 }
1311 
1312 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1313 {
1314 	struct nl_info info = {	.nl_net = net, };
1315 
1316 	return __ip6_ins_rt(rt, &info, NULL);
1317 }
1318 
1319 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1320 					   const struct in6_addr *daddr,
1321 					   const struct in6_addr *saddr)
1322 {
1323 	struct fib6_info *f6i = res->f6i;
1324 	struct net_device *dev;
1325 	struct rt6_info *rt;
1326 
1327 	/*
1328 	 *	Clone the route.
1329 	 */
1330 
1331 	if (!fib6_info_hold_safe(f6i))
1332 		return NULL;
1333 
1334 	dev = ip6_rt_get_dev_rcu(res);
1335 	rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1336 	if (!rt) {
1337 		fib6_info_release(f6i);
1338 		return NULL;
1339 	}
1340 
1341 	ip6_rt_copy_init(rt, res);
1342 	rt->rt6i_flags |= RTF_CACHE;
1343 	rt->rt6i_dst.addr = *daddr;
1344 	rt->rt6i_dst.plen = 128;
1345 
1346 	if (!rt6_is_gw_or_nonexthop(res)) {
1347 		if (f6i->fib6_dst.plen != 128 &&
1348 		    ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1349 			rt->rt6i_flags |= RTF_ANYCAST;
1350 #ifdef CONFIG_IPV6_SUBTREES
1351 		if (rt->rt6i_src.plen && saddr) {
1352 			rt->rt6i_src.addr = *saddr;
1353 			rt->rt6i_src.plen = 128;
1354 		}
1355 #endif
1356 	}
1357 
1358 	return rt;
1359 }
1360 
1361 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1362 {
1363 	struct fib6_info *f6i = res->f6i;
1364 	unsigned short flags = fib6_info_dst_flags(f6i);
1365 	struct net_device *dev;
1366 	struct rt6_info *pcpu_rt;
1367 
1368 	if (!fib6_info_hold_safe(f6i))
1369 		return NULL;
1370 
1371 	rcu_read_lock();
1372 	dev = ip6_rt_get_dev_rcu(res);
1373 	pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags | DST_NOCOUNT);
1374 	rcu_read_unlock();
1375 	if (!pcpu_rt) {
1376 		fib6_info_release(f6i);
1377 		return NULL;
1378 	}
1379 	ip6_rt_copy_init(pcpu_rt, res);
1380 	pcpu_rt->rt6i_flags |= RTF_PCPU;
1381 
1382 	if (f6i->nh)
1383 		pcpu_rt->sernum = rt_genid_ipv6(dev_net(dev));
1384 
1385 	return pcpu_rt;
1386 }
1387 
1388 static bool rt6_is_valid(const struct rt6_info *rt6)
1389 {
1390 	return rt6->sernum == rt_genid_ipv6(dev_net(rt6->dst.dev));
1391 }
1392 
1393 /* It should be called with rcu_read_lock() acquired */
1394 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1395 {
1396 	struct rt6_info *pcpu_rt;
1397 
1398 	pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1399 
1400 	if (pcpu_rt && pcpu_rt->sernum && !rt6_is_valid(pcpu_rt)) {
1401 		struct rt6_info *prev, **p;
1402 
1403 		p = this_cpu_ptr(res->nh->rt6i_pcpu);
1404 		/* Paired with READ_ONCE() in __fib6_drop_pcpu_from() */
1405 		prev = xchg(p, NULL);
1406 		if (prev) {
1407 			dst_dev_put(&prev->dst);
1408 			dst_release(&prev->dst);
1409 		}
1410 
1411 		pcpu_rt = NULL;
1412 	}
1413 
1414 	return pcpu_rt;
1415 }
1416 
1417 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1418 					    const struct fib6_result *res)
1419 {
1420 	struct rt6_info *pcpu_rt, *prev, **p;
1421 
1422 	pcpu_rt = ip6_rt_pcpu_alloc(res);
1423 	if (!pcpu_rt)
1424 		return NULL;
1425 
1426 	p = this_cpu_ptr(res->nh->rt6i_pcpu);
1427 	prev = cmpxchg(p, NULL, pcpu_rt);
1428 	BUG_ON(prev);
1429 
1430 	if (res->f6i->fib6_destroying) {
1431 		struct fib6_info *from;
1432 
1433 		from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1434 		fib6_info_release(from);
1435 	}
1436 
1437 	return pcpu_rt;
1438 }
1439 
1440 /* exception hash table implementation
1441  */
1442 static DEFINE_SPINLOCK(rt6_exception_lock);
1443 
1444 /* Remove rt6_ex from hash table and free the memory
1445  * Caller must hold rt6_exception_lock
1446  */
1447 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1448 				 struct rt6_exception *rt6_ex)
1449 {
1450 	struct fib6_info *from;
1451 	struct net *net;
1452 
1453 	if (!bucket || !rt6_ex)
1454 		return;
1455 
1456 	net = dev_net(rt6_ex->rt6i->dst.dev);
1457 	net->ipv6.rt6_stats->fib_rt_cache--;
1458 
1459 	/* purge completely the exception to allow releasing the held resources:
1460 	 * some [sk] cache may keep the dst around for unlimited time
1461 	 */
1462 	from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1463 	fib6_info_release(from);
1464 	dst_dev_put(&rt6_ex->rt6i->dst);
1465 
1466 	hlist_del_rcu(&rt6_ex->hlist);
1467 	dst_release(&rt6_ex->rt6i->dst);
1468 	kfree_rcu(rt6_ex, rcu);
1469 	WARN_ON_ONCE(!bucket->depth);
1470 	bucket->depth--;
1471 }
1472 
1473 /* Remove oldest rt6_ex in bucket and free the memory
1474  * Caller must hold rt6_exception_lock
1475  */
1476 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1477 {
1478 	struct rt6_exception *rt6_ex, *oldest = NULL;
1479 
1480 	if (!bucket)
1481 		return;
1482 
1483 	hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1484 		if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1485 			oldest = rt6_ex;
1486 	}
1487 	rt6_remove_exception(bucket, oldest);
1488 }
1489 
1490 static u32 rt6_exception_hash(const struct in6_addr *dst,
1491 			      const struct in6_addr *src)
1492 {
1493 	static siphash_aligned_key_t rt6_exception_key;
1494 	struct {
1495 		struct in6_addr dst;
1496 		struct in6_addr src;
1497 	} __aligned(SIPHASH_ALIGNMENT) combined = {
1498 		.dst = *dst,
1499 	};
1500 	u64 val;
1501 
1502 	net_get_random_once(&rt6_exception_key, sizeof(rt6_exception_key));
1503 
1504 #ifdef CONFIG_IPV6_SUBTREES
1505 	if (src)
1506 		combined.src = *src;
1507 #endif
1508 	val = siphash(&combined, sizeof(combined), &rt6_exception_key);
1509 
1510 	return hash_64(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1511 }
1512 
1513 /* Helper function to find the cached rt in the hash table
1514  * and update bucket pointer to point to the bucket for this
1515  * (daddr, saddr) pair
1516  * Caller must hold rt6_exception_lock
1517  */
1518 static struct rt6_exception *
1519 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1520 			      const struct in6_addr *daddr,
1521 			      const struct in6_addr *saddr)
1522 {
1523 	struct rt6_exception *rt6_ex;
1524 	u32 hval;
1525 
1526 	if (!(*bucket) || !daddr)
1527 		return NULL;
1528 
1529 	hval = rt6_exception_hash(daddr, saddr);
1530 	*bucket += hval;
1531 
1532 	hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1533 		struct rt6_info *rt6 = rt6_ex->rt6i;
1534 		bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1535 
1536 #ifdef CONFIG_IPV6_SUBTREES
1537 		if (matched && saddr)
1538 			matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1539 #endif
1540 		if (matched)
1541 			return rt6_ex;
1542 	}
1543 	return NULL;
1544 }
1545 
1546 /* Helper function to find the cached rt in the hash table
1547  * and update bucket pointer to point to the bucket for this
1548  * (daddr, saddr) pair
1549  * Caller must hold rcu_read_lock()
1550  */
1551 static struct rt6_exception *
1552 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1553 			 const struct in6_addr *daddr,
1554 			 const struct in6_addr *saddr)
1555 {
1556 	struct rt6_exception *rt6_ex;
1557 	u32 hval;
1558 
1559 	WARN_ON_ONCE(!rcu_read_lock_held());
1560 
1561 	if (!(*bucket) || !daddr)
1562 		return NULL;
1563 
1564 	hval = rt6_exception_hash(daddr, saddr);
1565 	*bucket += hval;
1566 
1567 	hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1568 		struct rt6_info *rt6 = rt6_ex->rt6i;
1569 		bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1570 
1571 #ifdef CONFIG_IPV6_SUBTREES
1572 		if (matched && saddr)
1573 			matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1574 #endif
1575 		if (matched)
1576 			return rt6_ex;
1577 	}
1578 	return NULL;
1579 }
1580 
1581 static unsigned int fib6_mtu(const struct fib6_result *res)
1582 {
1583 	const struct fib6_nh *nh = res->nh;
1584 	unsigned int mtu;
1585 
1586 	if (res->f6i->fib6_pmtu) {
1587 		mtu = res->f6i->fib6_pmtu;
1588 	} else {
1589 		struct net_device *dev = nh->fib_nh_dev;
1590 		struct inet6_dev *idev;
1591 
1592 		rcu_read_lock();
1593 		idev = __in6_dev_get(dev);
1594 		mtu = idev->cnf.mtu6;
1595 		rcu_read_unlock();
1596 	}
1597 
1598 	mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1599 
1600 	return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1601 }
1602 
1603 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1604 
1605 /* used when the flushed bit is not relevant, only access to the bucket
1606  * (ie., all bucket users except rt6_insert_exception);
1607  *
1608  * called under rcu lock; sometimes called with rt6_exception_lock held
1609  */
1610 static
1611 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1612 						       spinlock_t *lock)
1613 {
1614 	struct rt6_exception_bucket *bucket;
1615 
1616 	if (lock)
1617 		bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1618 						   lockdep_is_held(lock));
1619 	else
1620 		bucket = rcu_dereference(nh->rt6i_exception_bucket);
1621 
1622 	/* remove bucket flushed bit if set */
1623 	if (bucket) {
1624 		unsigned long p = (unsigned long)bucket;
1625 
1626 		p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1627 		bucket = (struct rt6_exception_bucket *)p;
1628 	}
1629 
1630 	return bucket;
1631 }
1632 
1633 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1634 {
1635 	unsigned long p = (unsigned long)bucket;
1636 
1637 	return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1638 }
1639 
1640 /* called with rt6_exception_lock held */
1641 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1642 					      spinlock_t *lock)
1643 {
1644 	struct rt6_exception_bucket *bucket;
1645 	unsigned long p;
1646 
1647 	bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1648 					   lockdep_is_held(lock));
1649 
1650 	p = (unsigned long)bucket;
1651 	p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1652 	bucket = (struct rt6_exception_bucket *)p;
1653 	rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1654 }
1655 
1656 static int rt6_insert_exception(struct rt6_info *nrt,
1657 				const struct fib6_result *res)
1658 {
1659 	struct net *net = dev_net(nrt->dst.dev);
1660 	struct rt6_exception_bucket *bucket;
1661 	struct fib6_info *f6i = res->f6i;
1662 	struct in6_addr *src_key = NULL;
1663 	struct rt6_exception *rt6_ex;
1664 	struct fib6_nh *nh = res->nh;
1665 	int max_depth;
1666 	int err = 0;
1667 
1668 	spin_lock_bh(&rt6_exception_lock);
1669 
1670 	bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1671 					  lockdep_is_held(&rt6_exception_lock));
1672 	if (!bucket) {
1673 		bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1674 				 GFP_ATOMIC);
1675 		if (!bucket) {
1676 			err = -ENOMEM;
1677 			goto out;
1678 		}
1679 		rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1680 	} else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1681 		err = -EINVAL;
1682 		goto out;
1683 	}
1684 
1685 #ifdef CONFIG_IPV6_SUBTREES
1686 	/* fib6_src.plen != 0 indicates f6i is in subtree
1687 	 * and exception table is indexed by a hash of
1688 	 * both fib6_dst and fib6_src.
1689 	 * Otherwise, the exception table is indexed by
1690 	 * a hash of only fib6_dst.
1691 	 */
1692 	if (f6i->fib6_src.plen)
1693 		src_key = &nrt->rt6i_src.addr;
1694 #endif
1695 	/* rt6_mtu_change() might lower mtu on f6i.
1696 	 * Only insert this exception route if its mtu
1697 	 * is less than f6i's mtu value.
1698 	 */
1699 	if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1700 		err = -EINVAL;
1701 		goto out;
1702 	}
1703 
1704 	rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1705 					       src_key);
1706 	if (rt6_ex)
1707 		rt6_remove_exception(bucket, rt6_ex);
1708 
1709 	rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1710 	if (!rt6_ex) {
1711 		err = -ENOMEM;
1712 		goto out;
1713 	}
1714 	rt6_ex->rt6i = nrt;
1715 	rt6_ex->stamp = jiffies;
1716 	hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1717 	bucket->depth++;
1718 	net->ipv6.rt6_stats->fib_rt_cache++;
1719 
1720 	/* Randomize max depth to avoid some side channels attacks. */
1721 	max_depth = FIB6_MAX_DEPTH + get_random_u32_below(FIB6_MAX_DEPTH);
1722 	while (bucket->depth > max_depth)
1723 		rt6_exception_remove_oldest(bucket);
1724 
1725 out:
1726 	spin_unlock_bh(&rt6_exception_lock);
1727 
1728 	/* Update fn->fn_sernum to invalidate all cached dst */
1729 	if (!err) {
1730 		spin_lock_bh(&f6i->fib6_table->tb6_lock);
1731 		fib6_update_sernum(net, f6i);
1732 		spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1733 		fib6_force_start_gc(net);
1734 	}
1735 
1736 	return err;
1737 }
1738 
1739 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1740 {
1741 	struct rt6_exception_bucket *bucket;
1742 	struct rt6_exception *rt6_ex;
1743 	struct hlist_node *tmp;
1744 	int i;
1745 
1746 	spin_lock_bh(&rt6_exception_lock);
1747 
1748 	bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1749 	if (!bucket)
1750 		goto out;
1751 
1752 	/* Prevent rt6_insert_exception() to recreate the bucket list */
1753 	if (!from)
1754 		fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1755 
1756 	for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1757 		hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1758 			if (!from ||
1759 			    rcu_access_pointer(rt6_ex->rt6i->from) == from)
1760 				rt6_remove_exception(bucket, rt6_ex);
1761 		}
1762 		WARN_ON_ONCE(!from && bucket->depth);
1763 		bucket++;
1764 	}
1765 out:
1766 	spin_unlock_bh(&rt6_exception_lock);
1767 }
1768 
1769 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1770 {
1771 	struct fib6_info *f6i = arg;
1772 
1773 	fib6_nh_flush_exceptions(nh, f6i);
1774 
1775 	return 0;
1776 }
1777 
1778 void rt6_flush_exceptions(struct fib6_info *f6i)
1779 {
1780 	if (f6i->nh)
1781 		nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1782 					 f6i);
1783 	else
1784 		fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1785 }
1786 
1787 /* Find cached rt in the hash table inside passed in rt
1788  * Caller has to hold rcu_read_lock()
1789  */
1790 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1791 					   const struct in6_addr *daddr,
1792 					   const struct in6_addr *saddr)
1793 {
1794 	const struct in6_addr *src_key = NULL;
1795 	struct rt6_exception_bucket *bucket;
1796 	struct rt6_exception *rt6_ex;
1797 	struct rt6_info *ret = NULL;
1798 
1799 #ifdef CONFIG_IPV6_SUBTREES
1800 	/* fib6i_src.plen != 0 indicates f6i is in subtree
1801 	 * and exception table is indexed by a hash of
1802 	 * both fib6_dst and fib6_src.
1803 	 * However, the src addr used to create the hash
1804 	 * might not be exactly the passed in saddr which
1805 	 * is a /128 addr from the flow.
1806 	 * So we need to use f6i->fib6_src to redo lookup
1807 	 * if the passed in saddr does not find anything.
1808 	 * (See the logic in ip6_rt_cache_alloc() on how
1809 	 * rt->rt6i_src is updated.)
1810 	 */
1811 	if (res->f6i->fib6_src.plen)
1812 		src_key = saddr;
1813 find_ex:
1814 #endif
1815 	bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1816 	rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1817 
1818 	if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1819 		ret = rt6_ex->rt6i;
1820 
1821 #ifdef CONFIG_IPV6_SUBTREES
1822 	/* Use fib6_src as src_key and redo lookup */
1823 	if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1824 		src_key = &res->f6i->fib6_src.addr;
1825 		goto find_ex;
1826 	}
1827 #endif
1828 
1829 	return ret;
1830 }
1831 
1832 /* Remove the passed in cached rt from the hash table that contains it */
1833 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1834 				    const struct rt6_info *rt)
1835 {
1836 	const struct in6_addr *src_key = NULL;
1837 	struct rt6_exception_bucket *bucket;
1838 	struct rt6_exception *rt6_ex;
1839 	int err;
1840 
1841 	if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1842 		return -ENOENT;
1843 
1844 	spin_lock_bh(&rt6_exception_lock);
1845 	bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1846 
1847 #ifdef CONFIG_IPV6_SUBTREES
1848 	/* rt6i_src.plen != 0 indicates 'from' is in subtree
1849 	 * and exception table is indexed by a hash of
1850 	 * both rt6i_dst and rt6i_src.
1851 	 * Otherwise, the exception table is indexed by
1852 	 * a hash of only rt6i_dst.
1853 	 */
1854 	if (plen)
1855 		src_key = &rt->rt6i_src.addr;
1856 #endif
1857 	rt6_ex = __rt6_find_exception_spinlock(&bucket,
1858 					       &rt->rt6i_dst.addr,
1859 					       src_key);
1860 	if (rt6_ex) {
1861 		rt6_remove_exception(bucket, rt6_ex);
1862 		err = 0;
1863 	} else {
1864 		err = -ENOENT;
1865 	}
1866 
1867 	spin_unlock_bh(&rt6_exception_lock);
1868 	return err;
1869 }
1870 
1871 struct fib6_nh_excptn_arg {
1872 	struct rt6_info	*rt;
1873 	int		plen;
1874 };
1875 
1876 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1877 {
1878 	struct fib6_nh_excptn_arg *arg = _arg;
1879 	int err;
1880 
1881 	err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1882 	if (err == 0)
1883 		return 1;
1884 
1885 	return 0;
1886 }
1887 
1888 static int rt6_remove_exception_rt(struct rt6_info *rt)
1889 {
1890 	struct fib6_info *from;
1891 
1892 	from = rcu_dereference(rt->from);
1893 	if (!from || !(rt->rt6i_flags & RTF_CACHE))
1894 		return -EINVAL;
1895 
1896 	if (from->nh) {
1897 		struct fib6_nh_excptn_arg arg = {
1898 			.rt = rt,
1899 			.plen = from->fib6_src.plen
1900 		};
1901 		int rc;
1902 
1903 		/* rc = 1 means an entry was found */
1904 		rc = nexthop_for_each_fib6_nh(from->nh,
1905 					      rt6_nh_remove_exception_rt,
1906 					      &arg);
1907 		return rc ? 0 : -ENOENT;
1908 	}
1909 
1910 	return fib6_nh_remove_exception(from->fib6_nh,
1911 					from->fib6_src.plen, rt);
1912 }
1913 
1914 /* Find rt6_ex which contains the passed in rt cache and
1915  * refresh its stamp
1916  */
1917 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1918 				     const struct rt6_info *rt)
1919 {
1920 	const struct in6_addr *src_key = NULL;
1921 	struct rt6_exception_bucket *bucket;
1922 	struct rt6_exception *rt6_ex;
1923 
1924 	bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1925 #ifdef CONFIG_IPV6_SUBTREES
1926 	/* rt6i_src.plen != 0 indicates 'from' is in subtree
1927 	 * and exception table is indexed by a hash of
1928 	 * both rt6i_dst and rt6i_src.
1929 	 * Otherwise, the exception table is indexed by
1930 	 * a hash of only rt6i_dst.
1931 	 */
1932 	if (plen)
1933 		src_key = &rt->rt6i_src.addr;
1934 #endif
1935 	rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1936 	if (rt6_ex)
1937 		rt6_ex->stamp = jiffies;
1938 }
1939 
1940 struct fib6_nh_match_arg {
1941 	const struct net_device *dev;
1942 	const struct in6_addr	*gw;
1943 	struct fib6_nh		*match;
1944 };
1945 
1946 /* determine if fib6_nh has given device and gateway */
1947 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1948 {
1949 	struct fib6_nh_match_arg *arg = _arg;
1950 
1951 	if (arg->dev != nh->fib_nh_dev ||
1952 	    (arg->gw && !nh->fib_nh_gw_family) ||
1953 	    (!arg->gw && nh->fib_nh_gw_family) ||
1954 	    (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1955 		return 0;
1956 
1957 	arg->match = nh;
1958 
1959 	/* found a match, break the loop */
1960 	return 1;
1961 }
1962 
1963 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1964 {
1965 	struct fib6_info *from;
1966 	struct fib6_nh *fib6_nh;
1967 
1968 	rcu_read_lock();
1969 
1970 	from = rcu_dereference(rt->from);
1971 	if (!from || !(rt->rt6i_flags & RTF_CACHE))
1972 		goto unlock;
1973 
1974 	if (from->nh) {
1975 		struct fib6_nh_match_arg arg = {
1976 			.dev = rt->dst.dev,
1977 			.gw = &rt->rt6i_gateway,
1978 		};
1979 
1980 		nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1981 
1982 		if (!arg.match)
1983 			goto unlock;
1984 		fib6_nh = arg.match;
1985 	} else {
1986 		fib6_nh = from->fib6_nh;
1987 	}
1988 	fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1989 unlock:
1990 	rcu_read_unlock();
1991 }
1992 
1993 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1994 					 struct rt6_info *rt, int mtu)
1995 {
1996 	/* If the new MTU is lower than the route PMTU, this new MTU will be the
1997 	 * lowest MTU in the path: always allow updating the route PMTU to
1998 	 * reflect PMTU decreases.
1999 	 *
2000 	 * If the new MTU is higher, and the route PMTU is equal to the local
2001 	 * MTU, this means the old MTU is the lowest in the path, so allow
2002 	 * updating it: if other nodes now have lower MTUs, PMTU discovery will
2003 	 * handle this.
2004 	 */
2005 
2006 	if (dst_mtu(&rt->dst) >= mtu)
2007 		return true;
2008 
2009 	if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
2010 		return true;
2011 
2012 	return false;
2013 }
2014 
2015 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
2016 				       const struct fib6_nh *nh, int mtu)
2017 {
2018 	struct rt6_exception_bucket *bucket;
2019 	struct rt6_exception *rt6_ex;
2020 	int i;
2021 
2022 	bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2023 	if (!bucket)
2024 		return;
2025 
2026 	for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2027 		hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2028 			struct rt6_info *entry = rt6_ex->rt6i;
2029 
2030 			/* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2031 			 * route), the metrics of its rt->from have already
2032 			 * been updated.
2033 			 */
2034 			if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2035 			    rt6_mtu_change_route_allowed(idev, entry, mtu))
2036 				dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2037 		}
2038 		bucket++;
2039 	}
2040 }
2041 
2042 #define RTF_CACHE_GATEWAY	(RTF_GATEWAY | RTF_CACHE)
2043 
2044 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2045 					    const struct in6_addr *gateway)
2046 {
2047 	struct rt6_exception_bucket *bucket;
2048 	struct rt6_exception *rt6_ex;
2049 	struct hlist_node *tmp;
2050 	int i;
2051 
2052 	if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2053 		return;
2054 
2055 	spin_lock_bh(&rt6_exception_lock);
2056 	bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2057 	if (bucket) {
2058 		for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2059 			hlist_for_each_entry_safe(rt6_ex, tmp,
2060 						  &bucket->chain, hlist) {
2061 				struct rt6_info *entry = rt6_ex->rt6i;
2062 
2063 				if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2064 				    RTF_CACHE_GATEWAY &&
2065 				    ipv6_addr_equal(gateway,
2066 						    &entry->rt6i_gateway)) {
2067 					rt6_remove_exception(bucket, rt6_ex);
2068 				}
2069 			}
2070 			bucket++;
2071 		}
2072 	}
2073 
2074 	spin_unlock_bh(&rt6_exception_lock);
2075 }
2076 
2077 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2078 				      struct rt6_exception *rt6_ex,
2079 				      struct fib6_gc_args *gc_args,
2080 				      unsigned long now)
2081 {
2082 	struct rt6_info *rt = rt6_ex->rt6i;
2083 
2084 	/* we are pruning and obsoleting aged-out and non gateway exceptions
2085 	 * even if others have still references to them, so that on next
2086 	 * dst_check() such references can be dropped.
2087 	 * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2088 	 * expired, independently from their aging, as per RFC 8201 section 4
2089 	 */
2090 	if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2091 		if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2092 			RT6_TRACE("aging clone %p\n", rt);
2093 			rt6_remove_exception(bucket, rt6_ex);
2094 			return;
2095 		}
2096 	} else if (time_after(jiffies, rt->dst.expires)) {
2097 		RT6_TRACE("purging expired route %p\n", rt);
2098 		rt6_remove_exception(bucket, rt6_ex);
2099 		return;
2100 	}
2101 
2102 	if (rt->rt6i_flags & RTF_GATEWAY) {
2103 		struct neighbour *neigh;
2104 
2105 		neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2106 
2107 		if (!(neigh && (neigh->flags & NTF_ROUTER))) {
2108 			RT6_TRACE("purging route %p via non-router but gateway\n",
2109 				  rt);
2110 			rt6_remove_exception(bucket, rt6_ex);
2111 			return;
2112 		}
2113 	}
2114 
2115 	gc_args->more++;
2116 }
2117 
2118 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2119 				   struct fib6_gc_args *gc_args,
2120 				   unsigned long now)
2121 {
2122 	struct rt6_exception_bucket *bucket;
2123 	struct rt6_exception *rt6_ex;
2124 	struct hlist_node *tmp;
2125 	int i;
2126 
2127 	if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2128 		return;
2129 
2130 	rcu_read_lock_bh();
2131 	spin_lock(&rt6_exception_lock);
2132 	bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2133 	if (bucket) {
2134 		for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2135 			hlist_for_each_entry_safe(rt6_ex, tmp,
2136 						  &bucket->chain, hlist) {
2137 				rt6_age_examine_exception(bucket, rt6_ex,
2138 							  gc_args, now);
2139 			}
2140 			bucket++;
2141 		}
2142 	}
2143 	spin_unlock(&rt6_exception_lock);
2144 	rcu_read_unlock_bh();
2145 }
2146 
2147 struct fib6_nh_age_excptn_arg {
2148 	struct fib6_gc_args	*gc_args;
2149 	unsigned long		now;
2150 };
2151 
2152 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2153 {
2154 	struct fib6_nh_age_excptn_arg *arg = _arg;
2155 
2156 	fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2157 	return 0;
2158 }
2159 
2160 void rt6_age_exceptions(struct fib6_info *f6i,
2161 			struct fib6_gc_args *gc_args,
2162 			unsigned long now)
2163 {
2164 	if (f6i->nh) {
2165 		struct fib6_nh_age_excptn_arg arg = {
2166 			.gc_args = gc_args,
2167 			.now = now
2168 		};
2169 
2170 		nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2171 					 &arg);
2172 	} else {
2173 		fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2174 	}
2175 }
2176 
2177 /* must be called with rcu lock held */
2178 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2179 		      struct flowi6 *fl6, struct fib6_result *res, int strict)
2180 {
2181 	struct fib6_node *fn, *saved_fn;
2182 
2183 	fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2184 	saved_fn = fn;
2185 
2186 redo_rt6_select:
2187 	rt6_select(net, fn, oif, res, strict);
2188 	if (res->f6i == net->ipv6.fib6_null_entry) {
2189 		fn = fib6_backtrack(fn, &fl6->saddr);
2190 		if (fn)
2191 			goto redo_rt6_select;
2192 		else if (strict & RT6_LOOKUP_F_REACHABLE) {
2193 			/* also consider unreachable route */
2194 			strict &= ~RT6_LOOKUP_F_REACHABLE;
2195 			fn = saved_fn;
2196 			goto redo_rt6_select;
2197 		}
2198 	}
2199 
2200 	trace_fib6_table_lookup(net, res, table, fl6);
2201 
2202 	return 0;
2203 }
2204 
2205 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2206 			       int oif, struct flowi6 *fl6,
2207 			       const struct sk_buff *skb, int flags)
2208 {
2209 	struct fib6_result res = {};
2210 	struct rt6_info *rt = NULL;
2211 	int strict = 0;
2212 
2213 	WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2214 		     !rcu_read_lock_held());
2215 
2216 	strict |= flags & RT6_LOOKUP_F_IFACE;
2217 	strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2218 	if (net->ipv6.devconf_all->forwarding == 0)
2219 		strict |= RT6_LOOKUP_F_REACHABLE;
2220 
2221 	rcu_read_lock();
2222 
2223 	fib6_table_lookup(net, table, oif, fl6, &res, strict);
2224 	if (res.f6i == net->ipv6.fib6_null_entry)
2225 		goto out;
2226 
2227 	fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2228 
2229 	/*Search through exception table */
2230 	rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2231 	if (rt) {
2232 		goto out;
2233 	} else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2234 			    !res.nh->fib_nh_gw_family)) {
2235 		/* Create a RTF_CACHE clone which will not be
2236 		 * owned by the fib6 tree.  It is for the special case where
2237 		 * the daddr in the skb during the neighbor look-up is different
2238 		 * from the fl6->daddr used to look-up route here.
2239 		 */
2240 		rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2241 
2242 		if (rt) {
2243 			/* 1 refcnt is taken during ip6_rt_cache_alloc().
2244 			 * As rt6_uncached_list_add() does not consume refcnt,
2245 			 * this refcnt is always returned to the caller even
2246 			 * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2247 			 */
2248 			rt6_uncached_list_add(rt);
2249 			rcu_read_unlock();
2250 
2251 			return rt;
2252 		}
2253 	} else {
2254 		/* Get a percpu copy */
2255 		local_bh_disable();
2256 		rt = rt6_get_pcpu_route(&res);
2257 
2258 		if (!rt)
2259 			rt = rt6_make_pcpu_route(net, &res);
2260 
2261 		local_bh_enable();
2262 	}
2263 out:
2264 	if (!rt)
2265 		rt = net->ipv6.ip6_null_entry;
2266 	if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2267 		ip6_hold_safe(net, &rt);
2268 	rcu_read_unlock();
2269 
2270 	return rt;
2271 }
2272 EXPORT_SYMBOL_GPL(ip6_pol_route);
2273 
2274 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_input(struct net *net,
2275 					    struct fib6_table *table,
2276 					    struct flowi6 *fl6,
2277 					    const struct sk_buff *skb,
2278 					    int flags)
2279 {
2280 	return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2281 }
2282 
2283 struct dst_entry *ip6_route_input_lookup(struct net *net,
2284 					 struct net_device *dev,
2285 					 struct flowi6 *fl6,
2286 					 const struct sk_buff *skb,
2287 					 int flags)
2288 {
2289 	if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2290 		flags |= RT6_LOOKUP_F_IFACE;
2291 
2292 	return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2293 }
2294 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2295 
2296 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2297 				  struct flow_keys *keys,
2298 				  struct flow_keys *flkeys)
2299 {
2300 	const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2301 	const struct ipv6hdr *key_iph = outer_iph;
2302 	struct flow_keys *_flkeys = flkeys;
2303 	const struct ipv6hdr *inner_iph;
2304 	const struct icmp6hdr *icmph;
2305 	struct ipv6hdr _inner_iph;
2306 	struct icmp6hdr _icmph;
2307 
2308 	if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2309 		goto out;
2310 
2311 	icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2312 				   sizeof(_icmph), &_icmph);
2313 	if (!icmph)
2314 		goto out;
2315 
2316 	if (!icmpv6_is_err(icmph->icmp6_type))
2317 		goto out;
2318 
2319 	inner_iph = skb_header_pointer(skb,
2320 				       skb_transport_offset(skb) + sizeof(*icmph),
2321 				       sizeof(_inner_iph), &_inner_iph);
2322 	if (!inner_iph)
2323 		goto out;
2324 
2325 	key_iph = inner_iph;
2326 	_flkeys = NULL;
2327 out:
2328 	if (_flkeys) {
2329 		keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2330 		keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2331 		keys->tags.flow_label = _flkeys->tags.flow_label;
2332 		keys->basic.ip_proto = _flkeys->basic.ip_proto;
2333 	} else {
2334 		keys->addrs.v6addrs.src = key_iph->saddr;
2335 		keys->addrs.v6addrs.dst = key_iph->daddr;
2336 		keys->tags.flow_label = ip6_flowlabel(key_iph);
2337 		keys->basic.ip_proto = key_iph->nexthdr;
2338 	}
2339 }
2340 
2341 static u32 rt6_multipath_custom_hash_outer(const struct net *net,
2342 					   const struct sk_buff *skb,
2343 					   bool *p_has_inner)
2344 {
2345 	u32 hash_fields = ip6_multipath_hash_fields(net);
2346 	struct flow_keys keys, hash_keys;
2347 
2348 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2349 		return 0;
2350 
2351 	memset(&hash_keys, 0, sizeof(hash_keys));
2352 	skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
2353 
2354 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2355 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2356 		hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2357 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2358 		hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2359 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2360 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
2361 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2362 		hash_keys.tags.flow_label = keys.tags.flow_label;
2363 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2364 		hash_keys.ports.src = keys.ports.src;
2365 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2366 		hash_keys.ports.dst = keys.ports.dst;
2367 
2368 	*p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
2369 	return flow_hash_from_keys(&hash_keys);
2370 }
2371 
2372 static u32 rt6_multipath_custom_hash_inner(const struct net *net,
2373 					   const struct sk_buff *skb,
2374 					   bool has_inner)
2375 {
2376 	u32 hash_fields = ip6_multipath_hash_fields(net);
2377 	struct flow_keys keys, hash_keys;
2378 
2379 	/* We assume the packet carries an encapsulation, but if none was
2380 	 * encountered during dissection of the outer flow, then there is no
2381 	 * point in calling the flow dissector again.
2382 	 */
2383 	if (!has_inner)
2384 		return 0;
2385 
2386 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
2387 		return 0;
2388 
2389 	memset(&hash_keys, 0, sizeof(hash_keys));
2390 	skb_flow_dissect_flow_keys(skb, &keys, 0);
2391 
2392 	if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
2393 		return 0;
2394 
2395 	if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2396 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2397 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2398 			hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2399 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2400 			hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2401 	} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2402 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2403 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2404 			hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2405 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2406 			hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2407 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2408 			hash_keys.tags.flow_label = keys.tags.flow_label;
2409 	}
2410 
2411 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2412 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
2413 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2414 		hash_keys.ports.src = keys.ports.src;
2415 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2416 		hash_keys.ports.dst = keys.ports.dst;
2417 
2418 	return flow_hash_from_keys(&hash_keys);
2419 }
2420 
2421 static u32 rt6_multipath_custom_hash_skb(const struct net *net,
2422 					 const struct sk_buff *skb)
2423 {
2424 	u32 mhash, mhash_inner;
2425 	bool has_inner = true;
2426 
2427 	mhash = rt6_multipath_custom_hash_outer(net, skb, &has_inner);
2428 	mhash_inner = rt6_multipath_custom_hash_inner(net, skb, has_inner);
2429 
2430 	return jhash_2words(mhash, mhash_inner, 0);
2431 }
2432 
2433 static u32 rt6_multipath_custom_hash_fl6(const struct net *net,
2434 					 const struct flowi6 *fl6)
2435 {
2436 	u32 hash_fields = ip6_multipath_hash_fields(net);
2437 	struct flow_keys hash_keys;
2438 
2439 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2440 		return 0;
2441 
2442 	memset(&hash_keys, 0, sizeof(hash_keys));
2443 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2444 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2445 		hash_keys.addrs.v6addrs.src = fl6->saddr;
2446 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2447 		hash_keys.addrs.v6addrs.dst = fl6->daddr;
2448 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2449 		hash_keys.basic.ip_proto = fl6->flowi6_proto;
2450 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2451 		hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2452 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2453 		hash_keys.ports.src = fl6->fl6_sport;
2454 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2455 		hash_keys.ports.dst = fl6->fl6_dport;
2456 
2457 	return flow_hash_from_keys(&hash_keys);
2458 }
2459 
2460 /* if skb is set it will be used and fl6 can be NULL */
2461 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2462 		       const struct sk_buff *skb, struct flow_keys *flkeys)
2463 {
2464 	struct flow_keys hash_keys;
2465 	u32 mhash = 0;
2466 
2467 	switch (ip6_multipath_hash_policy(net)) {
2468 	case 0:
2469 		memset(&hash_keys, 0, sizeof(hash_keys));
2470 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2471 		if (skb) {
2472 			ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2473 		} else {
2474 			hash_keys.addrs.v6addrs.src = fl6->saddr;
2475 			hash_keys.addrs.v6addrs.dst = fl6->daddr;
2476 			hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2477 			hash_keys.basic.ip_proto = fl6->flowi6_proto;
2478 		}
2479 		mhash = flow_hash_from_keys(&hash_keys);
2480 		break;
2481 	case 1:
2482 		if (skb) {
2483 			unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2484 			struct flow_keys keys;
2485 
2486 			/* short-circuit if we already have L4 hash present */
2487 			if (skb->l4_hash)
2488 				return skb_get_hash_raw(skb) >> 1;
2489 
2490 			memset(&hash_keys, 0, sizeof(hash_keys));
2491 
2492 			if (!flkeys) {
2493 				skb_flow_dissect_flow_keys(skb, &keys, flag);
2494 				flkeys = &keys;
2495 			}
2496 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2497 			hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2498 			hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2499 			hash_keys.ports.src = flkeys->ports.src;
2500 			hash_keys.ports.dst = flkeys->ports.dst;
2501 			hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2502 		} else {
2503 			memset(&hash_keys, 0, sizeof(hash_keys));
2504 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2505 			hash_keys.addrs.v6addrs.src = fl6->saddr;
2506 			hash_keys.addrs.v6addrs.dst = fl6->daddr;
2507 			hash_keys.ports.src = fl6->fl6_sport;
2508 			hash_keys.ports.dst = fl6->fl6_dport;
2509 			hash_keys.basic.ip_proto = fl6->flowi6_proto;
2510 		}
2511 		mhash = flow_hash_from_keys(&hash_keys);
2512 		break;
2513 	case 2:
2514 		memset(&hash_keys, 0, sizeof(hash_keys));
2515 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2516 		if (skb) {
2517 			struct flow_keys keys;
2518 
2519 			if (!flkeys) {
2520 				skb_flow_dissect_flow_keys(skb, &keys, 0);
2521 				flkeys = &keys;
2522 			}
2523 
2524 			/* Inner can be v4 or v6 */
2525 			if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2526 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2527 				hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2528 				hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2529 			} else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2530 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2531 				hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2532 				hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2533 				hash_keys.tags.flow_label = flkeys->tags.flow_label;
2534 				hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2535 			} else {
2536 				/* Same as case 0 */
2537 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2538 				ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2539 			}
2540 		} else {
2541 			/* Same as case 0 */
2542 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2543 			hash_keys.addrs.v6addrs.src = fl6->saddr;
2544 			hash_keys.addrs.v6addrs.dst = fl6->daddr;
2545 			hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2546 			hash_keys.basic.ip_proto = fl6->flowi6_proto;
2547 		}
2548 		mhash = flow_hash_from_keys(&hash_keys);
2549 		break;
2550 	case 3:
2551 		if (skb)
2552 			mhash = rt6_multipath_custom_hash_skb(net, skb);
2553 		else
2554 			mhash = rt6_multipath_custom_hash_fl6(net, fl6);
2555 		break;
2556 	}
2557 
2558 	return mhash >> 1;
2559 }
2560 
2561 /* Called with rcu held */
2562 void ip6_route_input(struct sk_buff *skb)
2563 {
2564 	const struct ipv6hdr *iph = ipv6_hdr(skb);
2565 	struct net *net = dev_net(skb->dev);
2566 	int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2567 	struct ip_tunnel_info *tun_info;
2568 	struct flowi6 fl6 = {
2569 		.flowi6_iif = skb->dev->ifindex,
2570 		.daddr = iph->daddr,
2571 		.saddr = iph->saddr,
2572 		.flowlabel = ip6_flowinfo(iph),
2573 		.flowi6_mark = skb->mark,
2574 		.flowi6_proto = iph->nexthdr,
2575 	};
2576 	struct flow_keys *flkeys = NULL, _flkeys;
2577 
2578 	tun_info = skb_tunnel_info(skb);
2579 	if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2580 		fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2581 
2582 	if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2583 		flkeys = &_flkeys;
2584 
2585 	if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2586 		fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2587 	skb_dst_drop(skb);
2588 	skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2589 						      &fl6, skb, flags));
2590 }
2591 
2592 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_output(struct net *net,
2593 					     struct fib6_table *table,
2594 					     struct flowi6 *fl6,
2595 					     const struct sk_buff *skb,
2596 					     int flags)
2597 {
2598 	return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2599 }
2600 
2601 static struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2602 						      const struct sock *sk,
2603 						      struct flowi6 *fl6,
2604 						      int flags)
2605 {
2606 	bool any_src;
2607 
2608 	if (ipv6_addr_type(&fl6->daddr) &
2609 	    (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2610 		struct dst_entry *dst;
2611 
2612 		/* This function does not take refcnt on the dst */
2613 		dst = l3mdev_link_scope_lookup(net, fl6);
2614 		if (dst)
2615 			return dst;
2616 	}
2617 
2618 	fl6->flowi6_iif = LOOPBACK_IFINDEX;
2619 
2620 	flags |= RT6_LOOKUP_F_DST_NOREF;
2621 	any_src = ipv6_addr_any(&fl6->saddr);
2622 	if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2623 	    (fl6->flowi6_oif && any_src))
2624 		flags |= RT6_LOOKUP_F_IFACE;
2625 
2626 	if (!any_src)
2627 		flags |= RT6_LOOKUP_F_HAS_SADDR;
2628 	else if (sk)
2629 		flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2630 
2631 	return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2632 }
2633 
2634 struct dst_entry *ip6_route_output_flags(struct net *net,
2635 					 const struct sock *sk,
2636 					 struct flowi6 *fl6,
2637 					 int flags)
2638 {
2639 	struct dst_entry *dst;
2640 	struct rt6_info *rt6;
2641 
2642 	rcu_read_lock();
2643 	dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2644 	rt6 = (struct rt6_info *)dst;
2645 	/* For dst cached in uncached_list, refcnt is already taken. */
2646 	if (list_empty(&rt6->dst.rt_uncached) && !dst_hold_safe(dst)) {
2647 		dst = &net->ipv6.ip6_null_entry->dst;
2648 		dst_hold(dst);
2649 	}
2650 	rcu_read_unlock();
2651 
2652 	return dst;
2653 }
2654 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2655 
2656 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2657 {
2658 	struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2659 	struct net_device *loopback_dev = net->loopback_dev;
2660 	struct dst_entry *new = NULL;
2661 
2662 	rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2663 		       DST_OBSOLETE_DEAD, 0);
2664 	if (rt) {
2665 		rt6_info_init(rt);
2666 		atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2667 
2668 		new = &rt->dst;
2669 		new->__use = 1;
2670 		new->input = dst_discard;
2671 		new->output = dst_discard_out;
2672 
2673 		dst_copy_metrics(new, &ort->dst);
2674 
2675 		rt->rt6i_idev = in6_dev_get(loopback_dev);
2676 		rt->rt6i_gateway = ort->rt6i_gateway;
2677 		rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2678 
2679 		memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2680 #ifdef CONFIG_IPV6_SUBTREES
2681 		memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2682 #endif
2683 	}
2684 
2685 	dst_release(dst_orig);
2686 	return new ? new : ERR_PTR(-ENOMEM);
2687 }
2688 
2689 /*
2690  *	Destination cache support functions
2691  */
2692 
2693 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2694 {
2695 	u32 rt_cookie = 0;
2696 
2697 	if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2698 		return false;
2699 
2700 	if (fib6_check_expired(f6i))
2701 		return false;
2702 
2703 	return true;
2704 }
2705 
2706 static struct dst_entry *rt6_check(struct rt6_info *rt,
2707 				   struct fib6_info *from,
2708 				   u32 cookie)
2709 {
2710 	u32 rt_cookie = 0;
2711 
2712 	if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2713 	    rt_cookie != cookie)
2714 		return NULL;
2715 
2716 	if (rt6_check_expired(rt))
2717 		return NULL;
2718 
2719 	return &rt->dst;
2720 }
2721 
2722 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2723 					    struct fib6_info *from,
2724 					    u32 cookie)
2725 {
2726 	if (!__rt6_check_expired(rt) &&
2727 	    rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2728 	    fib6_check(from, cookie))
2729 		return &rt->dst;
2730 	else
2731 		return NULL;
2732 }
2733 
2734 INDIRECT_CALLABLE_SCOPE struct dst_entry *ip6_dst_check(struct dst_entry *dst,
2735 							u32 cookie)
2736 {
2737 	struct dst_entry *dst_ret;
2738 	struct fib6_info *from;
2739 	struct rt6_info *rt;
2740 
2741 	rt = container_of(dst, struct rt6_info, dst);
2742 
2743 	if (rt->sernum)
2744 		return rt6_is_valid(rt) ? dst : NULL;
2745 
2746 	rcu_read_lock();
2747 
2748 	/* All IPV6 dsts are created with ->obsolete set to the value
2749 	 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2750 	 * into this function always.
2751 	 */
2752 
2753 	from = rcu_dereference(rt->from);
2754 
2755 	if (from && (rt->rt6i_flags & RTF_PCPU ||
2756 	    unlikely(!list_empty(&rt->dst.rt_uncached))))
2757 		dst_ret = rt6_dst_from_check(rt, from, cookie);
2758 	else
2759 		dst_ret = rt6_check(rt, from, cookie);
2760 
2761 	rcu_read_unlock();
2762 
2763 	return dst_ret;
2764 }
2765 EXPORT_INDIRECT_CALLABLE(ip6_dst_check);
2766 
2767 static void ip6_negative_advice(struct sock *sk,
2768 				struct dst_entry *dst)
2769 {
2770 	struct rt6_info *rt = (struct rt6_info *) dst;
2771 
2772 	if (rt->rt6i_flags & RTF_CACHE) {
2773 		rcu_read_lock();
2774 		if (rt6_check_expired(rt)) {
2775 			/* counteract the dst_release() in sk_dst_reset() */
2776 			dst_hold(dst);
2777 			sk_dst_reset(sk);
2778 
2779 			rt6_remove_exception_rt(rt);
2780 		}
2781 		rcu_read_unlock();
2782 		return;
2783 	}
2784 	sk_dst_reset(sk);
2785 }
2786 
2787 static void ip6_link_failure(struct sk_buff *skb)
2788 {
2789 	struct rt6_info *rt;
2790 
2791 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2792 
2793 	rt = (struct rt6_info *) skb_dst(skb);
2794 	if (rt) {
2795 		rcu_read_lock();
2796 		if (rt->rt6i_flags & RTF_CACHE) {
2797 			rt6_remove_exception_rt(rt);
2798 		} else {
2799 			struct fib6_info *from;
2800 			struct fib6_node *fn;
2801 
2802 			from = rcu_dereference(rt->from);
2803 			if (from) {
2804 				fn = rcu_dereference(from->fib6_node);
2805 				if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2806 					WRITE_ONCE(fn->fn_sernum, -1);
2807 			}
2808 		}
2809 		rcu_read_unlock();
2810 	}
2811 }
2812 
2813 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2814 {
2815 	if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2816 		struct fib6_info *from;
2817 
2818 		rcu_read_lock();
2819 		from = rcu_dereference(rt0->from);
2820 		if (from)
2821 			rt0->dst.expires = from->expires;
2822 		rcu_read_unlock();
2823 	}
2824 
2825 	dst_set_expires(&rt0->dst, timeout);
2826 	rt0->rt6i_flags |= RTF_EXPIRES;
2827 }
2828 
2829 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2830 {
2831 	struct net *net = dev_net(rt->dst.dev);
2832 
2833 	dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2834 	rt->rt6i_flags |= RTF_MODIFIED;
2835 	rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2836 }
2837 
2838 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2839 {
2840 	return !(rt->rt6i_flags & RTF_CACHE) &&
2841 		(rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2842 }
2843 
2844 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2845 				 const struct ipv6hdr *iph, u32 mtu,
2846 				 bool confirm_neigh)
2847 {
2848 	const struct in6_addr *daddr, *saddr;
2849 	struct rt6_info *rt6 = (struct rt6_info *)dst;
2850 
2851 	/* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2852 	 * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2853 	 * [see also comment in rt6_mtu_change_route()]
2854 	 */
2855 
2856 	if (iph) {
2857 		daddr = &iph->daddr;
2858 		saddr = &iph->saddr;
2859 	} else if (sk) {
2860 		daddr = &sk->sk_v6_daddr;
2861 		saddr = &inet6_sk(sk)->saddr;
2862 	} else {
2863 		daddr = NULL;
2864 		saddr = NULL;
2865 	}
2866 
2867 	if (confirm_neigh)
2868 		dst_confirm_neigh(dst, daddr);
2869 
2870 	if (mtu < IPV6_MIN_MTU)
2871 		return;
2872 	if (mtu >= dst_mtu(dst))
2873 		return;
2874 
2875 	if (!rt6_cache_allowed_for_pmtu(rt6)) {
2876 		rt6_do_update_pmtu(rt6, mtu);
2877 		/* update rt6_ex->stamp for cache */
2878 		if (rt6->rt6i_flags & RTF_CACHE)
2879 			rt6_update_exception_stamp_rt(rt6);
2880 	} else if (daddr) {
2881 		struct fib6_result res = {};
2882 		struct rt6_info *nrt6;
2883 
2884 		rcu_read_lock();
2885 		res.f6i = rcu_dereference(rt6->from);
2886 		if (!res.f6i)
2887 			goto out_unlock;
2888 
2889 		res.fib6_flags = res.f6i->fib6_flags;
2890 		res.fib6_type = res.f6i->fib6_type;
2891 
2892 		if (res.f6i->nh) {
2893 			struct fib6_nh_match_arg arg = {
2894 				.dev = dst->dev,
2895 				.gw = &rt6->rt6i_gateway,
2896 			};
2897 
2898 			nexthop_for_each_fib6_nh(res.f6i->nh,
2899 						 fib6_nh_find_match, &arg);
2900 
2901 			/* fib6_info uses a nexthop that does not have fib6_nh
2902 			 * using the dst->dev + gw. Should be impossible.
2903 			 */
2904 			if (!arg.match)
2905 				goto out_unlock;
2906 
2907 			res.nh = arg.match;
2908 		} else {
2909 			res.nh = res.f6i->fib6_nh;
2910 		}
2911 
2912 		nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2913 		if (nrt6) {
2914 			rt6_do_update_pmtu(nrt6, mtu);
2915 			if (rt6_insert_exception(nrt6, &res))
2916 				dst_release_immediate(&nrt6->dst);
2917 		}
2918 out_unlock:
2919 		rcu_read_unlock();
2920 	}
2921 }
2922 
2923 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2924 			       struct sk_buff *skb, u32 mtu,
2925 			       bool confirm_neigh)
2926 {
2927 	__ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2928 			     confirm_neigh);
2929 }
2930 
2931 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2932 		     int oif, u32 mark, kuid_t uid)
2933 {
2934 	const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2935 	struct dst_entry *dst;
2936 	struct flowi6 fl6 = {
2937 		.flowi6_oif = oif,
2938 		.flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2939 		.daddr = iph->daddr,
2940 		.saddr = iph->saddr,
2941 		.flowlabel = ip6_flowinfo(iph),
2942 		.flowi6_uid = uid,
2943 	};
2944 
2945 	dst = ip6_route_output(net, NULL, &fl6);
2946 	if (!dst->error)
2947 		__ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2948 	dst_release(dst);
2949 }
2950 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2951 
2952 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2953 {
2954 	int oif = sk->sk_bound_dev_if;
2955 	struct dst_entry *dst;
2956 
2957 	if (!oif && skb->dev)
2958 		oif = l3mdev_master_ifindex(skb->dev);
2959 
2960 	ip6_update_pmtu(skb, sock_net(sk), mtu, oif, READ_ONCE(sk->sk_mark),
2961 			sk->sk_uid);
2962 
2963 	dst = __sk_dst_get(sk);
2964 	if (!dst || !dst->obsolete ||
2965 	    dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2966 		return;
2967 
2968 	bh_lock_sock(sk);
2969 	if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2970 		ip6_datagram_dst_update(sk, false);
2971 	bh_unlock_sock(sk);
2972 }
2973 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2974 
2975 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2976 			   const struct flowi6 *fl6)
2977 {
2978 #ifdef CONFIG_IPV6_SUBTREES
2979 	struct ipv6_pinfo *np = inet6_sk(sk);
2980 #endif
2981 
2982 	ip6_dst_store(sk, dst,
2983 		      ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2984 		      &sk->sk_v6_daddr : NULL,
2985 #ifdef CONFIG_IPV6_SUBTREES
2986 		      ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2987 		      &np->saddr :
2988 #endif
2989 		      NULL);
2990 }
2991 
2992 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2993 				  struct flowi6 *fl6,
2994 				  const struct in6_addr *gw,
2995 				  struct rt6_info **ret)
2996 {
2997 	const struct fib6_nh *nh = res->nh;
2998 
2999 	if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
3000 	    fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
3001 		return false;
3002 
3003 	/* rt_cache's gateway might be different from its 'parent'
3004 	 * in the case of an ip redirect.
3005 	 * So we keep searching in the exception table if the gateway
3006 	 * is different.
3007 	 */
3008 	if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
3009 		struct rt6_info *rt_cache;
3010 
3011 		rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
3012 		if (rt_cache &&
3013 		    ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
3014 			*ret = rt_cache;
3015 			return true;
3016 		}
3017 		return false;
3018 	}
3019 	return true;
3020 }
3021 
3022 struct fib6_nh_rd_arg {
3023 	struct fib6_result	*res;
3024 	struct flowi6		*fl6;
3025 	const struct in6_addr	*gw;
3026 	struct rt6_info		**ret;
3027 };
3028 
3029 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
3030 {
3031 	struct fib6_nh_rd_arg *arg = _arg;
3032 
3033 	arg->res->nh = nh;
3034 	return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
3035 }
3036 
3037 /* Handle redirects */
3038 struct ip6rd_flowi {
3039 	struct flowi6 fl6;
3040 	struct in6_addr gateway;
3041 };
3042 
3043 INDIRECT_CALLABLE_SCOPE struct rt6_info *__ip6_route_redirect(struct net *net,
3044 					     struct fib6_table *table,
3045 					     struct flowi6 *fl6,
3046 					     const struct sk_buff *skb,
3047 					     int flags)
3048 {
3049 	struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
3050 	struct rt6_info *ret = NULL;
3051 	struct fib6_result res = {};
3052 	struct fib6_nh_rd_arg arg = {
3053 		.res = &res,
3054 		.fl6 = fl6,
3055 		.gw  = &rdfl->gateway,
3056 		.ret = &ret
3057 	};
3058 	struct fib6_info *rt;
3059 	struct fib6_node *fn;
3060 
3061 	/* Get the "current" route for this destination and
3062 	 * check if the redirect has come from appropriate router.
3063 	 *
3064 	 * RFC 4861 specifies that redirects should only be
3065 	 * accepted if they come from the nexthop to the target.
3066 	 * Due to the way the routes are chosen, this notion
3067 	 * is a bit fuzzy and one might need to check all possible
3068 	 * routes.
3069 	 */
3070 
3071 	rcu_read_lock();
3072 	fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
3073 restart:
3074 	for_each_fib6_node_rt_rcu(fn) {
3075 		res.f6i = rt;
3076 		if (fib6_check_expired(rt))
3077 			continue;
3078 		if (rt->fib6_flags & RTF_REJECT)
3079 			break;
3080 		if (unlikely(rt->nh)) {
3081 			if (nexthop_is_blackhole(rt->nh))
3082 				continue;
3083 			/* on match, res->nh is filled in and potentially ret */
3084 			if (nexthop_for_each_fib6_nh(rt->nh,
3085 						     fib6_nh_redirect_match,
3086 						     &arg))
3087 				goto out;
3088 		} else {
3089 			res.nh = rt->fib6_nh;
3090 			if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
3091 						  &ret))
3092 				goto out;
3093 		}
3094 	}
3095 
3096 	if (!rt)
3097 		rt = net->ipv6.fib6_null_entry;
3098 	else if (rt->fib6_flags & RTF_REJECT) {
3099 		ret = net->ipv6.ip6_null_entry;
3100 		goto out;
3101 	}
3102 
3103 	if (rt == net->ipv6.fib6_null_entry) {
3104 		fn = fib6_backtrack(fn, &fl6->saddr);
3105 		if (fn)
3106 			goto restart;
3107 	}
3108 
3109 	res.f6i = rt;
3110 	res.nh = rt->fib6_nh;
3111 out:
3112 	if (ret) {
3113 		ip6_hold_safe(net, &ret);
3114 	} else {
3115 		res.fib6_flags = res.f6i->fib6_flags;
3116 		res.fib6_type = res.f6i->fib6_type;
3117 		ret = ip6_create_rt_rcu(&res);
3118 	}
3119 
3120 	rcu_read_unlock();
3121 
3122 	trace_fib6_table_lookup(net, &res, table, fl6);
3123 	return ret;
3124 };
3125 
3126 static struct dst_entry *ip6_route_redirect(struct net *net,
3127 					    const struct flowi6 *fl6,
3128 					    const struct sk_buff *skb,
3129 					    const struct in6_addr *gateway)
3130 {
3131 	int flags = RT6_LOOKUP_F_HAS_SADDR;
3132 	struct ip6rd_flowi rdfl;
3133 
3134 	rdfl.fl6 = *fl6;
3135 	rdfl.gateway = *gateway;
3136 
3137 	return fib6_rule_lookup(net, &rdfl.fl6, skb,
3138 				flags, __ip6_route_redirect);
3139 }
3140 
3141 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
3142 		  kuid_t uid)
3143 {
3144 	const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
3145 	struct dst_entry *dst;
3146 	struct flowi6 fl6 = {
3147 		.flowi6_iif = LOOPBACK_IFINDEX,
3148 		.flowi6_oif = oif,
3149 		.flowi6_mark = mark,
3150 		.daddr = iph->daddr,
3151 		.saddr = iph->saddr,
3152 		.flowlabel = ip6_flowinfo(iph),
3153 		.flowi6_uid = uid,
3154 	};
3155 
3156 	dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3157 	rt6_do_redirect(dst, NULL, skb);
3158 	dst_release(dst);
3159 }
3160 EXPORT_SYMBOL_GPL(ip6_redirect);
3161 
3162 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3163 {
3164 	const struct ipv6hdr *iph = ipv6_hdr(skb);
3165 	const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3166 	struct dst_entry *dst;
3167 	struct flowi6 fl6 = {
3168 		.flowi6_iif = LOOPBACK_IFINDEX,
3169 		.flowi6_oif = oif,
3170 		.daddr = msg->dest,
3171 		.saddr = iph->daddr,
3172 		.flowi6_uid = sock_net_uid(net, NULL),
3173 	};
3174 
3175 	dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3176 	rt6_do_redirect(dst, NULL, skb);
3177 	dst_release(dst);
3178 }
3179 
3180 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3181 {
3182 	ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if,
3183 		     READ_ONCE(sk->sk_mark), sk->sk_uid);
3184 }
3185 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3186 
3187 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3188 {
3189 	struct net_device *dev = dst->dev;
3190 	unsigned int mtu = dst_mtu(dst);
3191 	struct net *net = dev_net(dev);
3192 
3193 	mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3194 
3195 	if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3196 		mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3197 
3198 	/*
3199 	 * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3200 	 * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3201 	 * IPV6_MAXPLEN is also valid and means: "any MSS,
3202 	 * rely only on pmtu discovery"
3203 	 */
3204 	if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3205 		mtu = IPV6_MAXPLEN;
3206 	return mtu;
3207 }
3208 
3209 INDIRECT_CALLABLE_SCOPE unsigned int ip6_mtu(const struct dst_entry *dst)
3210 {
3211 	return ip6_dst_mtu_maybe_forward(dst, false);
3212 }
3213 EXPORT_INDIRECT_CALLABLE(ip6_mtu);
3214 
3215 /* MTU selection:
3216  * 1. mtu on route is locked - use it
3217  * 2. mtu from nexthop exception
3218  * 3. mtu from egress device
3219  *
3220  * based on ip6_dst_mtu_forward and exception logic of
3221  * rt6_find_cached_rt; called with rcu_read_lock
3222  */
3223 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3224 		      const struct in6_addr *daddr,
3225 		      const struct in6_addr *saddr)
3226 {
3227 	const struct fib6_nh *nh = res->nh;
3228 	struct fib6_info *f6i = res->f6i;
3229 	struct inet6_dev *idev;
3230 	struct rt6_info *rt;
3231 	u32 mtu = 0;
3232 
3233 	if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3234 		mtu = f6i->fib6_pmtu;
3235 		if (mtu)
3236 			goto out;
3237 	}
3238 
3239 	rt = rt6_find_cached_rt(res, daddr, saddr);
3240 	if (unlikely(rt)) {
3241 		mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3242 	} else {
3243 		struct net_device *dev = nh->fib_nh_dev;
3244 
3245 		mtu = IPV6_MIN_MTU;
3246 		idev = __in6_dev_get(dev);
3247 		if (idev && idev->cnf.mtu6 > mtu)
3248 			mtu = idev->cnf.mtu6;
3249 	}
3250 
3251 	mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3252 out:
3253 	return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3254 }
3255 
3256 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3257 				  struct flowi6 *fl6)
3258 {
3259 	struct dst_entry *dst;
3260 	struct rt6_info *rt;
3261 	struct inet6_dev *idev = in6_dev_get(dev);
3262 	struct net *net = dev_net(dev);
3263 
3264 	if (unlikely(!idev))
3265 		return ERR_PTR(-ENODEV);
3266 
3267 	rt = ip6_dst_alloc(net, dev, 0);
3268 	if (unlikely(!rt)) {
3269 		in6_dev_put(idev);
3270 		dst = ERR_PTR(-ENOMEM);
3271 		goto out;
3272 	}
3273 
3274 	rt->dst.input = ip6_input;
3275 	rt->dst.output  = ip6_output;
3276 	rt->rt6i_gateway  = fl6->daddr;
3277 	rt->rt6i_dst.addr = fl6->daddr;
3278 	rt->rt6i_dst.plen = 128;
3279 	rt->rt6i_idev     = idev;
3280 	dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3281 
3282 	/* Add this dst into uncached_list so that rt6_disable_ip() can
3283 	 * do proper release of the net_device
3284 	 */
3285 	rt6_uncached_list_add(rt);
3286 
3287 	dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3288 
3289 out:
3290 	return dst;
3291 }
3292 
3293 static void ip6_dst_gc(struct dst_ops *ops)
3294 {
3295 	struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3296 	int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3297 	int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3298 	int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3299 	unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3300 	unsigned int val;
3301 	int entries;
3302 
3303 	if (time_after(rt_last_gc + rt_min_interval, jiffies))
3304 		goto out;
3305 
3306 	fib6_run_gc(atomic_inc_return(&net->ipv6.ip6_rt_gc_expire), net, true);
3307 	entries = dst_entries_get_slow(ops);
3308 	if (entries < ops->gc_thresh)
3309 		atomic_set(&net->ipv6.ip6_rt_gc_expire, rt_gc_timeout >> 1);
3310 out:
3311 	val = atomic_read(&net->ipv6.ip6_rt_gc_expire);
3312 	atomic_set(&net->ipv6.ip6_rt_gc_expire, val - (val >> rt_elasticity));
3313 }
3314 
3315 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3316 			       const struct in6_addr *gw_addr, u32 tbid,
3317 			       int flags, struct fib6_result *res)
3318 {
3319 	struct flowi6 fl6 = {
3320 		.flowi6_oif = cfg->fc_ifindex,
3321 		.daddr = *gw_addr,
3322 		.saddr = cfg->fc_prefsrc,
3323 	};
3324 	struct fib6_table *table;
3325 	int err;
3326 
3327 	table = fib6_get_table(net, tbid);
3328 	if (!table)
3329 		return -EINVAL;
3330 
3331 	if (!ipv6_addr_any(&cfg->fc_prefsrc))
3332 		flags |= RT6_LOOKUP_F_HAS_SADDR;
3333 
3334 	flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3335 
3336 	err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3337 	if (!err && res->f6i != net->ipv6.fib6_null_entry)
3338 		fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3339 				 cfg->fc_ifindex != 0, NULL, flags);
3340 
3341 	return err;
3342 }
3343 
3344 static int ip6_route_check_nh_onlink(struct net *net,
3345 				     struct fib6_config *cfg,
3346 				     const struct net_device *dev,
3347 				     struct netlink_ext_ack *extack)
3348 {
3349 	u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3350 	const struct in6_addr *gw_addr = &cfg->fc_gateway;
3351 	struct fib6_result res = {};
3352 	int err;
3353 
3354 	err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3355 	if (!err && !(res.fib6_flags & RTF_REJECT) &&
3356 	    /* ignore match if it is the default route */
3357 	    !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3358 	    (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3359 		NL_SET_ERR_MSG(extack,
3360 			       "Nexthop has invalid gateway or device mismatch");
3361 		err = -EINVAL;
3362 	}
3363 
3364 	return err;
3365 }
3366 
3367 static int ip6_route_check_nh(struct net *net,
3368 			      struct fib6_config *cfg,
3369 			      struct net_device **_dev,
3370 			      netdevice_tracker *dev_tracker,
3371 			      struct inet6_dev **idev)
3372 {
3373 	const struct in6_addr *gw_addr = &cfg->fc_gateway;
3374 	struct net_device *dev = _dev ? *_dev : NULL;
3375 	int flags = RT6_LOOKUP_F_IFACE;
3376 	struct fib6_result res = {};
3377 	int err = -EHOSTUNREACH;
3378 
3379 	if (cfg->fc_table) {
3380 		err = ip6_nh_lookup_table(net, cfg, gw_addr,
3381 					  cfg->fc_table, flags, &res);
3382 		/* gw_addr can not require a gateway or resolve to a reject
3383 		 * route. If a device is given, it must match the result.
3384 		 */
3385 		if (err || res.fib6_flags & RTF_REJECT ||
3386 		    res.nh->fib_nh_gw_family ||
3387 		    (dev && dev != res.nh->fib_nh_dev))
3388 			err = -EHOSTUNREACH;
3389 	}
3390 
3391 	if (err < 0) {
3392 		struct flowi6 fl6 = {
3393 			.flowi6_oif = cfg->fc_ifindex,
3394 			.daddr = *gw_addr,
3395 		};
3396 
3397 		err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3398 		if (err || res.fib6_flags & RTF_REJECT ||
3399 		    res.nh->fib_nh_gw_family)
3400 			err = -EHOSTUNREACH;
3401 
3402 		if (err)
3403 			return err;
3404 
3405 		fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3406 				 cfg->fc_ifindex != 0, NULL, flags);
3407 	}
3408 
3409 	err = 0;
3410 	if (dev) {
3411 		if (dev != res.nh->fib_nh_dev)
3412 			err = -EHOSTUNREACH;
3413 	} else {
3414 		*_dev = dev = res.nh->fib_nh_dev;
3415 		netdev_hold(dev, dev_tracker, GFP_ATOMIC);
3416 		*idev = in6_dev_get(dev);
3417 	}
3418 
3419 	return err;
3420 }
3421 
3422 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3423 			   struct net_device **_dev,
3424 			   netdevice_tracker *dev_tracker,
3425 			   struct inet6_dev **idev,
3426 			   struct netlink_ext_ack *extack)
3427 {
3428 	const struct in6_addr *gw_addr = &cfg->fc_gateway;
3429 	int gwa_type = ipv6_addr_type(gw_addr);
3430 	bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3431 	const struct net_device *dev = *_dev;
3432 	bool need_addr_check = !dev;
3433 	int err = -EINVAL;
3434 
3435 	/* if gw_addr is local we will fail to detect this in case
3436 	 * address is still TENTATIVE (DAD in progress). rt6_lookup()
3437 	 * will return already-added prefix route via interface that
3438 	 * prefix route was assigned to, which might be non-loopback.
3439 	 */
3440 	if (dev &&
3441 	    ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3442 		NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3443 		goto out;
3444 	}
3445 
3446 	if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3447 		/* IPv6 strictly inhibits using not link-local
3448 		 * addresses as nexthop address.
3449 		 * Otherwise, router will not able to send redirects.
3450 		 * It is very good, but in some (rare!) circumstances
3451 		 * (SIT, PtP, NBMA NOARP links) it is handy to allow
3452 		 * some exceptions. --ANK
3453 		 * We allow IPv4-mapped nexthops to support RFC4798-type
3454 		 * addressing
3455 		 */
3456 		if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3457 			NL_SET_ERR_MSG(extack, "Invalid gateway address");
3458 			goto out;
3459 		}
3460 
3461 		rcu_read_lock();
3462 
3463 		if (cfg->fc_flags & RTNH_F_ONLINK)
3464 			err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3465 		else
3466 			err = ip6_route_check_nh(net, cfg, _dev, dev_tracker,
3467 						 idev);
3468 
3469 		rcu_read_unlock();
3470 
3471 		if (err)
3472 			goto out;
3473 	}
3474 
3475 	/* reload in case device was changed */
3476 	dev = *_dev;
3477 
3478 	err = -EINVAL;
3479 	if (!dev) {
3480 		NL_SET_ERR_MSG(extack, "Egress device not specified");
3481 		goto out;
3482 	} else if (dev->flags & IFF_LOOPBACK) {
3483 		NL_SET_ERR_MSG(extack,
3484 			       "Egress device can not be loopback device for this route");
3485 		goto out;
3486 	}
3487 
3488 	/* if we did not check gw_addr above, do so now that the
3489 	 * egress device has been resolved.
3490 	 */
3491 	if (need_addr_check &&
3492 	    ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3493 		NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3494 		goto out;
3495 	}
3496 
3497 	err = 0;
3498 out:
3499 	return err;
3500 }
3501 
3502 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3503 {
3504 	if ((flags & RTF_REJECT) ||
3505 	    (dev && (dev->flags & IFF_LOOPBACK) &&
3506 	     !(addr_type & IPV6_ADDR_LOOPBACK) &&
3507 	     !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3508 		return true;
3509 
3510 	return false;
3511 }
3512 
3513 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3514 		 struct fib6_config *cfg, gfp_t gfp_flags,
3515 		 struct netlink_ext_ack *extack)
3516 {
3517 	netdevice_tracker *dev_tracker = &fib6_nh->fib_nh_dev_tracker;
3518 	struct net_device *dev = NULL;
3519 	struct inet6_dev *idev = NULL;
3520 	int addr_type;
3521 	int err;
3522 
3523 	fib6_nh->fib_nh_family = AF_INET6;
3524 #ifdef CONFIG_IPV6_ROUTER_PREF
3525 	fib6_nh->last_probe = jiffies;
3526 #endif
3527 	if (cfg->fc_is_fdb) {
3528 		fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3529 		fib6_nh->fib_nh_gw_family = AF_INET6;
3530 		return 0;
3531 	}
3532 
3533 	err = -ENODEV;
3534 	if (cfg->fc_ifindex) {
3535 		dev = netdev_get_by_index(net, cfg->fc_ifindex,
3536 					  dev_tracker, gfp_flags);
3537 		if (!dev)
3538 			goto out;
3539 		idev = in6_dev_get(dev);
3540 		if (!idev)
3541 			goto out;
3542 	}
3543 
3544 	if (cfg->fc_flags & RTNH_F_ONLINK) {
3545 		if (!dev) {
3546 			NL_SET_ERR_MSG(extack,
3547 				       "Nexthop device required for onlink");
3548 			goto out;
3549 		}
3550 
3551 		if (!(dev->flags & IFF_UP)) {
3552 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3553 			err = -ENETDOWN;
3554 			goto out;
3555 		}
3556 
3557 		fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3558 	}
3559 
3560 	fib6_nh->fib_nh_weight = 1;
3561 
3562 	/* We cannot add true routes via loopback here,
3563 	 * they would result in kernel looping; promote them to reject routes
3564 	 */
3565 	addr_type = ipv6_addr_type(&cfg->fc_dst);
3566 	if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3567 		/* hold loopback dev/idev if we haven't done so. */
3568 		if (dev != net->loopback_dev) {
3569 			if (dev) {
3570 				netdev_put(dev, dev_tracker);
3571 				in6_dev_put(idev);
3572 			}
3573 			dev = net->loopback_dev;
3574 			netdev_hold(dev, dev_tracker, gfp_flags);
3575 			idev = in6_dev_get(dev);
3576 			if (!idev) {
3577 				err = -ENODEV;
3578 				goto out;
3579 			}
3580 		}
3581 		goto pcpu_alloc;
3582 	}
3583 
3584 	if (cfg->fc_flags & RTF_GATEWAY) {
3585 		err = ip6_validate_gw(net, cfg, &dev, dev_tracker,
3586 				      &idev, extack);
3587 		if (err)
3588 			goto out;
3589 
3590 		fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3591 		fib6_nh->fib_nh_gw_family = AF_INET6;
3592 	}
3593 
3594 	err = -ENODEV;
3595 	if (!dev)
3596 		goto out;
3597 
3598 	if (!idev || idev->cnf.disable_ipv6) {
3599 		NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3600 		err = -EACCES;
3601 		goto out;
3602 	}
3603 
3604 	if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3605 		NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3606 		err = -ENETDOWN;
3607 		goto out;
3608 	}
3609 
3610 	if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3611 	    !netif_carrier_ok(dev))
3612 		fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3613 
3614 	err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3615 				 cfg->fc_encap_type, cfg, gfp_flags, extack);
3616 	if (err)
3617 		goto out;
3618 
3619 pcpu_alloc:
3620 	fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3621 	if (!fib6_nh->rt6i_pcpu) {
3622 		err = -ENOMEM;
3623 		goto out;
3624 	}
3625 
3626 	fib6_nh->fib_nh_dev = dev;
3627 	fib6_nh->fib_nh_oif = dev->ifindex;
3628 	err = 0;
3629 out:
3630 	if (idev)
3631 		in6_dev_put(idev);
3632 
3633 	if (err) {
3634 		lwtstate_put(fib6_nh->fib_nh_lws);
3635 		fib6_nh->fib_nh_lws = NULL;
3636 		netdev_put(dev, dev_tracker);
3637 	}
3638 
3639 	return err;
3640 }
3641 
3642 void fib6_nh_release(struct fib6_nh *fib6_nh)
3643 {
3644 	struct rt6_exception_bucket *bucket;
3645 
3646 	rcu_read_lock();
3647 
3648 	fib6_nh_flush_exceptions(fib6_nh, NULL);
3649 	bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3650 	if (bucket) {
3651 		rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3652 		kfree(bucket);
3653 	}
3654 
3655 	rcu_read_unlock();
3656 
3657 	fib6_nh_release_dsts(fib6_nh);
3658 	free_percpu(fib6_nh->rt6i_pcpu);
3659 
3660 	fib_nh_common_release(&fib6_nh->nh_common);
3661 }
3662 
3663 void fib6_nh_release_dsts(struct fib6_nh *fib6_nh)
3664 {
3665 	int cpu;
3666 
3667 	if (!fib6_nh->rt6i_pcpu)
3668 		return;
3669 
3670 	for_each_possible_cpu(cpu) {
3671 		struct rt6_info *pcpu_rt, **ppcpu_rt;
3672 
3673 		ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3674 		pcpu_rt = xchg(ppcpu_rt, NULL);
3675 		if (pcpu_rt) {
3676 			dst_dev_put(&pcpu_rt->dst);
3677 			dst_release(&pcpu_rt->dst);
3678 		}
3679 	}
3680 }
3681 
3682 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3683 					      gfp_t gfp_flags,
3684 					      struct netlink_ext_ack *extack)
3685 {
3686 	struct net *net = cfg->fc_nlinfo.nl_net;
3687 	struct fib6_info *rt = NULL;
3688 	struct nexthop *nh = NULL;
3689 	struct fib6_table *table;
3690 	struct fib6_nh *fib6_nh;
3691 	int err = -EINVAL;
3692 	int addr_type;
3693 
3694 	/* RTF_PCPU is an internal flag; can not be set by userspace */
3695 	if (cfg->fc_flags & RTF_PCPU) {
3696 		NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3697 		goto out;
3698 	}
3699 
3700 	/* RTF_CACHE is an internal flag; can not be set by userspace */
3701 	if (cfg->fc_flags & RTF_CACHE) {
3702 		NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3703 		goto out;
3704 	}
3705 
3706 	if (cfg->fc_type > RTN_MAX) {
3707 		NL_SET_ERR_MSG(extack, "Invalid route type");
3708 		goto out;
3709 	}
3710 
3711 	if (cfg->fc_dst_len > 128) {
3712 		NL_SET_ERR_MSG(extack, "Invalid prefix length");
3713 		goto out;
3714 	}
3715 	if (cfg->fc_src_len > 128) {
3716 		NL_SET_ERR_MSG(extack, "Invalid source address length");
3717 		goto out;
3718 	}
3719 #ifndef CONFIG_IPV6_SUBTREES
3720 	if (cfg->fc_src_len) {
3721 		NL_SET_ERR_MSG(extack,
3722 			       "Specifying source address requires IPV6_SUBTREES to be enabled");
3723 		goto out;
3724 	}
3725 #endif
3726 	if (cfg->fc_nh_id) {
3727 		nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3728 		if (!nh) {
3729 			NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3730 			goto out;
3731 		}
3732 		err = fib6_check_nexthop(nh, cfg, extack);
3733 		if (err)
3734 			goto out;
3735 	}
3736 
3737 	err = -ENOBUFS;
3738 	if (cfg->fc_nlinfo.nlh &&
3739 	    !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3740 		table = fib6_get_table(net, cfg->fc_table);
3741 		if (!table) {
3742 			pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3743 			table = fib6_new_table(net, cfg->fc_table);
3744 		}
3745 	} else {
3746 		table = fib6_new_table(net, cfg->fc_table);
3747 	}
3748 
3749 	if (!table)
3750 		goto out;
3751 
3752 	err = -ENOMEM;
3753 	rt = fib6_info_alloc(gfp_flags, !nh);
3754 	if (!rt)
3755 		goto out;
3756 
3757 	rt->fib6_metrics = ip_fib_metrics_init(cfg->fc_mx, cfg->fc_mx_len,
3758 					       extack);
3759 	if (IS_ERR(rt->fib6_metrics)) {
3760 		err = PTR_ERR(rt->fib6_metrics);
3761 		/* Do not leave garbage there. */
3762 		rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3763 		goto out_free;
3764 	}
3765 
3766 	if (cfg->fc_flags & RTF_ADDRCONF)
3767 		rt->dst_nocount = true;
3768 
3769 	if (cfg->fc_flags & RTF_EXPIRES)
3770 		fib6_set_expires(rt, jiffies +
3771 				clock_t_to_jiffies(cfg->fc_expires));
3772 	else
3773 		fib6_clean_expires(rt);
3774 
3775 	if (cfg->fc_protocol == RTPROT_UNSPEC)
3776 		cfg->fc_protocol = RTPROT_BOOT;
3777 	rt->fib6_protocol = cfg->fc_protocol;
3778 
3779 	rt->fib6_table = table;
3780 	rt->fib6_metric = cfg->fc_metric;
3781 	rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3782 	rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3783 
3784 	ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3785 	rt->fib6_dst.plen = cfg->fc_dst_len;
3786 
3787 #ifdef CONFIG_IPV6_SUBTREES
3788 	ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3789 	rt->fib6_src.plen = cfg->fc_src_len;
3790 #endif
3791 	if (nh) {
3792 		if (rt->fib6_src.plen) {
3793 			NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3794 			goto out_free;
3795 		}
3796 		if (!nexthop_get(nh)) {
3797 			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3798 			goto out_free;
3799 		}
3800 		rt->nh = nh;
3801 		fib6_nh = nexthop_fib6_nh(rt->nh);
3802 	} else {
3803 		err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3804 		if (err)
3805 			goto out;
3806 
3807 		fib6_nh = rt->fib6_nh;
3808 
3809 		/* We cannot add true routes via loopback here, they would
3810 		 * result in kernel looping; promote them to reject routes
3811 		 */
3812 		addr_type = ipv6_addr_type(&cfg->fc_dst);
3813 		if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3814 				   addr_type))
3815 			rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3816 	}
3817 
3818 	if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3819 		struct net_device *dev = fib6_nh->fib_nh_dev;
3820 
3821 		if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3822 			NL_SET_ERR_MSG(extack, "Invalid source address");
3823 			err = -EINVAL;
3824 			goto out;
3825 		}
3826 		rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3827 		rt->fib6_prefsrc.plen = 128;
3828 	} else
3829 		rt->fib6_prefsrc.plen = 0;
3830 
3831 	return rt;
3832 out:
3833 	fib6_info_release(rt);
3834 	return ERR_PTR(err);
3835 out_free:
3836 	ip_fib_metrics_put(rt->fib6_metrics);
3837 	kfree(rt);
3838 	return ERR_PTR(err);
3839 }
3840 
3841 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3842 		  struct netlink_ext_ack *extack)
3843 {
3844 	struct fib6_info *rt;
3845 	int err;
3846 
3847 	rt = ip6_route_info_create(cfg, gfp_flags, extack);
3848 	if (IS_ERR(rt))
3849 		return PTR_ERR(rt);
3850 
3851 	err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3852 	fib6_info_release(rt);
3853 
3854 	return err;
3855 }
3856 
3857 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3858 {
3859 	struct net *net = info->nl_net;
3860 	struct fib6_table *table;
3861 	int err;
3862 
3863 	if (rt == net->ipv6.fib6_null_entry) {
3864 		err = -ENOENT;
3865 		goto out;
3866 	}
3867 
3868 	table = rt->fib6_table;
3869 	spin_lock_bh(&table->tb6_lock);
3870 	err = fib6_del(rt, info);
3871 	spin_unlock_bh(&table->tb6_lock);
3872 
3873 out:
3874 	fib6_info_release(rt);
3875 	return err;
3876 }
3877 
3878 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3879 {
3880 	struct nl_info info = {
3881 		.nl_net = net,
3882 		.skip_notify = skip_notify
3883 	};
3884 
3885 	return __ip6_del_rt(rt, &info);
3886 }
3887 
3888 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3889 {
3890 	struct nl_info *info = &cfg->fc_nlinfo;
3891 	struct net *net = info->nl_net;
3892 	struct sk_buff *skb = NULL;
3893 	struct fib6_table *table;
3894 	int err = -ENOENT;
3895 
3896 	if (rt == net->ipv6.fib6_null_entry)
3897 		goto out_put;
3898 	table = rt->fib6_table;
3899 	spin_lock_bh(&table->tb6_lock);
3900 
3901 	if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3902 		struct fib6_info *sibling, *next_sibling;
3903 		struct fib6_node *fn;
3904 
3905 		/* prefer to send a single notification with all hops */
3906 		skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3907 		if (skb) {
3908 			u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3909 
3910 			if (rt6_fill_node(net, skb, rt, NULL,
3911 					  NULL, NULL, 0, RTM_DELROUTE,
3912 					  info->portid, seq, 0) < 0) {
3913 				kfree_skb(skb);
3914 				skb = NULL;
3915 			} else
3916 				info->skip_notify = 1;
3917 		}
3918 
3919 		/* 'rt' points to the first sibling route. If it is not the
3920 		 * leaf, then we do not need to send a notification. Otherwise,
3921 		 * we need to check if the last sibling has a next route or not
3922 		 * and emit a replace or delete notification, respectively.
3923 		 */
3924 		info->skip_notify_kernel = 1;
3925 		fn = rcu_dereference_protected(rt->fib6_node,
3926 					    lockdep_is_held(&table->tb6_lock));
3927 		if (rcu_access_pointer(fn->leaf) == rt) {
3928 			struct fib6_info *last_sibling, *replace_rt;
3929 
3930 			last_sibling = list_last_entry(&rt->fib6_siblings,
3931 						       struct fib6_info,
3932 						       fib6_siblings);
3933 			replace_rt = rcu_dereference_protected(
3934 					    last_sibling->fib6_next,
3935 					    lockdep_is_held(&table->tb6_lock));
3936 			if (replace_rt)
3937 				call_fib6_entry_notifiers_replace(net,
3938 								  replace_rt);
3939 			else
3940 				call_fib6_multipath_entry_notifiers(net,
3941 						       FIB_EVENT_ENTRY_DEL,
3942 						       rt, rt->fib6_nsiblings,
3943 						       NULL);
3944 		}
3945 		list_for_each_entry_safe(sibling, next_sibling,
3946 					 &rt->fib6_siblings,
3947 					 fib6_siblings) {
3948 			err = fib6_del(sibling, info);
3949 			if (err)
3950 				goto out_unlock;
3951 		}
3952 	}
3953 
3954 	err = fib6_del(rt, info);
3955 out_unlock:
3956 	spin_unlock_bh(&table->tb6_lock);
3957 out_put:
3958 	fib6_info_release(rt);
3959 
3960 	if (skb) {
3961 		rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3962 			    info->nlh, gfp_any());
3963 	}
3964 	return err;
3965 }
3966 
3967 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3968 {
3969 	int rc = -ESRCH;
3970 
3971 	if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3972 		goto out;
3973 
3974 	if (cfg->fc_flags & RTF_GATEWAY &&
3975 	    !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3976 		goto out;
3977 
3978 	rc = rt6_remove_exception_rt(rt);
3979 out:
3980 	return rc;
3981 }
3982 
3983 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3984 			     struct fib6_nh *nh)
3985 {
3986 	struct fib6_result res = {
3987 		.f6i = rt,
3988 		.nh = nh,
3989 	};
3990 	struct rt6_info *rt_cache;
3991 
3992 	rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3993 	if (rt_cache)
3994 		return __ip6_del_cached_rt(rt_cache, cfg);
3995 
3996 	return 0;
3997 }
3998 
3999 struct fib6_nh_del_cached_rt_arg {
4000 	struct fib6_config *cfg;
4001 	struct fib6_info *f6i;
4002 };
4003 
4004 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
4005 {
4006 	struct fib6_nh_del_cached_rt_arg *arg = _arg;
4007 	int rc;
4008 
4009 	rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
4010 	return rc != -ESRCH ? rc : 0;
4011 }
4012 
4013 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
4014 {
4015 	struct fib6_nh_del_cached_rt_arg arg = {
4016 		.cfg = cfg,
4017 		.f6i = f6i
4018 	};
4019 
4020 	return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
4021 }
4022 
4023 static int ip6_route_del(struct fib6_config *cfg,
4024 			 struct netlink_ext_ack *extack)
4025 {
4026 	struct fib6_table *table;
4027 	struct fib6_info *rt;
4028 	struct fib6_node *fn;
4029 	int err = -ESRCH;
4030 
4031 	table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
4032 	if (!table) {
4033 		NL_SET_ERR_MSG(extack, "FIB table does not exist");
4034 		return err;
4035 	}
4036 
4037 	rcu_read_lock();
4038 
4039 	fn = fib6_locate(&table->tb6_root,
4040 			 &cfg->fc_dst, cfg->fc_dst_len,
4041 			 &cfg->fc_src, cfg->fc_src_len,
4042 			 !(cfg->fc_flags & RTF_CACHE));
4043 
4044 	if (fn) {
4045 		for_each_fib6_node_rt_rcu(fn) {
4046 			struct fib6_nh *nh;
4047 
4048 			if (rt->nh && cfg->fc_nh_id &&
4049 			    rt->nh->id != cfg->fc_nh_id)
4050 				continue;
4051 
4052 			if (cfg->fc_flags & RTF_CACHE) {
4053 				int rc = 0;
4054 
4055 				if (rt->nh) {
4056 					rc = ip6_del_cached_rt_nh(cfg, rt);
4057 				} else if (cfg->fc_nh_id) {
4058 					continue;
4059 				} else {
4060 					nh = rt->fib6_nh;
4061 					rc = ip6_del_cached_rt(cfg, rt, nh);
4062 				}
4063 				if (rc != -ESRCH) {
4064 					rcu_read_unlock();
4065 					return rc;
4066 				}
4067 				continue;
4068 			}
4069 
4070 			if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
4071 				continue;
4072 			if (cfg->fc_protocol &&
4073 			    cfg->fc_protocol != rt->fib6_protocol)
4074 				continue;
4075 
4076 			if (rt->nh) {
4077 				if (!fib6_info_hold_safe(rt))
4078 					continue;
4079 				rcu_read_unlock();
4080 
4081 				return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4082 			}
4083 			if (cfg->fc_nh_id)
4084 				continue;
4085 
4086 			nh = rt->fib6_nh;
4087 			if (cfg->fc_ifindex &&
4088 			    (!nh->fib_nh_dev ||
4089 			     nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
4090 				continue;
4091 			if (cfg->fc_flags & RTF_GATEWAY &&
4092 			    !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
4093 				continue;
4094 			if (!fib6_info_hold_safe(rt))
4095 				continue;
4096 			rcu_read_unlock();
4097 
4098 			/* if gateway was specified only delete the one hop */
4099 			if (cfg->fc_flags & RTF_GATEWAY)
4100 				return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4101 
4102 			return __ip6_del_rt_siblings(rt, cfg);
4103 		}
4104 	}
4105 	rcu_read_unlock();
4106 
4107 	return err;
4108 }
4109 
4110 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4111 {
4112 	struct netevent_redirect netevent;
4113 	struct rt6_info *rt, *nrt = NULL;
4114 	struct fib6_result res = {};
4115 	struct ndisc_options ndopts;
4116 	struct inet6_dev *in6_dev;
4117 	struct neighbour *neigh;
4118 	struct rd_msg *msg;
4119 	int optlen, on_link;
4120 	u8 *lladdr;
4121 
4122 	optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4123 	optlen -= sizeof(*msg);
4124 
4125 	if (optlen < 0) {
4126 		net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4127 		return;
4128 	}
4129 
4130 	msg = (struct rd_msg *)icmp6_hdr(skb);
4131 
4132 	if (ipv6_addr_is_multicast(&msg->dest)) {
4133 		net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4134 		return;
4135 	}
4136 
4137 	on_link = 0;
4138 	if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4139 		on_link = 1;
4140 	} else if (ipv6_addr_type(&msg->target) !=
4141 		   (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4142 		net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4143 		return;
4144 	}
4145 
4146 	in6_dev = __in6_dev_get(skb->dev);
4147 	if (!in6_dev)
4148 		return;
4149 	if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
4150 		return;
4151 
4152 	/* RFC2461 8.1:
4153 	 *	The IP source address of the Redirect MUST be the same as the current
4154 	 *	first-hop router for the specified ICMP Destination Address.
4155 	 */
4156 
4157 	if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4158 		net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4159 		return;
4160 	}
4161 
4162 	lladdr = NULL;
4163 	if (ndopts.nd_opts_tgt_lladdr) {
4164 		lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4165 					     skb->dev);
4166 		if (!lladdr) {
4167 			net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4168 			return;
4169 		}
4170 	}
4171 
4172 	rt = (struct rt6_info *) dst;
4173 	if (rt->rt6i_flags & RTF_REJECT) {
4174 		net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4175 		return;
4176 	}
4177 
4178 	/* Redirect received -> path was valid.
4179 	 * Look, redirects are sent only in response to data packets,
4180 	 * so that this nexthop apparently is reachable. --ANK
4181 	 */
4182 	dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4183 
4184 	neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4185 	if (!neigh)
4186 		return;
4187 
4188 	/*
4189 	 *	We have finally decided to accept it.
4190 	 */
4191 
4192 	ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4193 		     NEIGH_UPDATE_F_WEAK_OVERRIDE|
4194 		     NEIGH_UPDATE_F_OVERRIDE|
4195 		     (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4196 				     NEIGH_UPDATE_F_ISROUTER)),
4197 		     NDISC_REDIRECT, &ndopts);
4198 
4199 	rcu_read_lock();
4200 	res.f6i = rcu_dereference(rt->from);
4201 	if (!res.f6i)
4202 		goto out;
4203 
4204 	if (res.f6i->nh) {
4205 		struct fib6_nh_match_arg arg = {
4206 			.dev = dst->dev,
4207 			.gw = &rt->rt6i_gateway,
4208 		};
4209 
4210 		nexthop_for_each_fib6_nh(res.f6i->nh,
4211 					 fib6_nh_find_match, &arg);
4212 
4213 		/* fib6_info uses a nexthop that does not have fib6_nh
4214 		 * using the dst->dev. Should be impossible
4215 		 */
4216 		if (!arg.match)
4217 			goto out;
4218 		res.nh = arg.match;
4219 	} else {
4220 		res.nh = res.f6i->fib6_nh;
4221 	}
4222 
4223 	res.fib6_flags = res.f6i->fib6_flags;
4224 	res.fib6_type = res.f6i->fib6_type;
4225 	nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4226 	if (!nrt)
4227 		goto out;
4228 
4229 	nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4230 	if (on_link)
4231 		nrt->rt6i_flags &= ~RTF_GATEWAY;
4232 
4233 	nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4234 
4235 	/* rt6_insert_exception() will take care of duplicated exceptions */
4236 	if (rt6_insert_exception(nrt, &res)) {
4237 		dst_release_immediate(&nrt->dst);
4238 		goto out;
4239 	}
4240 
4241 	netevent.old = &rt->dst;
4242 	netevent.new = &nrt->dst;
4243 	netevent.daddr = &msg->dest;
4244 	netevent.neigh = neigh;
4245 	call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4246 
4247 out:
4248 	rcu_read_unlock();
4249 	neigh_release(neigh);
4250 }
4251 
4252 #ifdef CONFIG_IPV6_ROUTE_INFO
4253 static struct fib6_info *rt6_get_route_info(struct net *net,
4254 					   const struct in6_addr *prefix, int prefixlen,
4255 					   const struct in6_addr *gwaddr,
4256 					   struct net_device *dev)
4257 {
4258 	u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4259 	int ifindex = dev->ifindex;
4260 	struct fib6_node *fn;
4261 	struct fib6_info *rt = NULL;
4262 	struct fib6_table *table;
4263 
4264 	table = fib6_get_table(net, tb_id);
4265 	if (!table)
4266 		return NULL;
4267 
4268 	rcu_read_lock();
4269 	fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4270 	if (!fn)
4271 		goto out;
4272 
4273 	for_each_fib6_node_rt_rcu(fn) {
4274 		/* these routes do not use nexthops */
4275 		if (rt->nh)
4276 			continue;
4277 		if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4278 			continue;
4279 		if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4280 		    !rt->fib6_nh->fib_nh_gw_family)
4281 			continue;
4282 		if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4283 			continue;
4284 		if (!fib6_info_hold_safe(rt))
4285 			continue;
4286 		break;
4287 	}
4288 out:
4289 	rcu_read_unlock();
4290 	return rt;
4291 }
4292 
4293 static struct fib6_info *rt6_add_route_info(struct net *net,
4294 					   const struct in6_addr *prefix, int prefixlen,
4295 					   const struct in6_addr *gwaddr,
4296 					   struct net_device *dev,
4297 					   unsigned int pref)
4298 {
4299 	struct fib6_config cfg = {
4300 		.fc_metric	= IP6_RT_PRIO_USER,
4301 		.fc_ifindex	= dev->ifindex,
4302 		.fc_dst_len	= prefixlen,
4303 		.fc_flags	= RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4304 				  RTF_UP | RTF_PREF(pref),
4305 		.fc_protocol = RTPROT_RA,
4306 		.fc_type = RTN_UNICAST,
4307 		.fc_nlinfo.portid = 0,
4308 		.fc_nlinfo.nlh = NULL,
4309 		.fc_nlinfo.nl_net = net,
4310 	};
4311 
4312 	cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4313 	cfg.fc_dst = *prefix;
4314 	cfg.fc_gateway = *gwaddr;
4315 
4316 	/* We should treat it as a default route if prefix length is 0. */
4317 	if (!prefixlen)
4318 		cfg.fc_flags |= RTF_DEFAULT;
4319 
4320 	ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4321 
4322 	return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4323 }
4324 #endif
4325 
4326 struct fib6_info *rt6_get_dflt_router(struct net *net,
4327 				     const struct in6_addr *addr,
4328 				     struct net_device *dev)
4329 {
4330 	u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4331 	struct fib6_info *rt;
4332 	struct fib6_table *table;
4333 
4334 	table = fib6_get_table(net, tb_id);
4335 	if (!table)
4336 		return NULL;
4337 
4338 	rcu_read_lock();
4339 	for_each_fib6_node_rt_rcu(&table->tb6_root) {
4340 		struct fib6_nh *nh;
4341 
4342 		/* RA routes do not use nexthops */
4343 		if (rt->nh)
4344 			continue;
4345 
4346 		nh = rt->fib6_nh;
4347 		if (dev == nh->fib_nh_dev &&
4348 		    ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4349 		    ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4350 			break;
4351 	}
4352 	if (rt && !fib6_info_hold_safe(rt))
4353 		rt = NULL;
4354 	rcu_read_unlock();
4355 	return rt;
4356 }
4357 
4358 struct fib6_info *rt6_add_dflt_router(struct net *net,
4359 				     const struct in6_addr *gwaddr,
4360 				     struct net_device *dev,
4361 				     unsigned int pref,
4362 				     u32 defrtr_usr_metric)
4363 {
4364 	struct fib6_config cfg = {
4365 		.fc_table	= l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4366 		.fc_metric	= defrtr_usr_metric,
4367 		.fc_ifindex	= dev->ifindex,
4368 		.fc_flags	= RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4369 				  RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4370 		.fc_protocol = RTPROT_RA,
4371 		.fc_type = RTN_UNICAST,
4372 		.fc_nlinfo.portid = 0,
4373 		.fc_nlinfo.nlh = NULL,
4374 		.fc_nlinfo.nl_net = net,
4375 	};
4376 
4377 	cfg.fc_gateway = *gwaddr;
4378 
4379 	if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4380 		struct fib6_table *table;
4381 
4382 		table = fib6_get_table(dev_net(dev), cfg.fc_table);
4383 		if (table)
4384 			table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4385 	}
4386 
4387 	return rt6_get_dflt_router(net, gwaddr, dev);
4388 }
4389 
4390 static void __rt6_purge_dflt_routers(struct net *net,
4391 				     struct fib6_table *table)
4392 {
4393 	struct fib6_info *rt;
4394 
4395 restart:
4396 	rcu_read_lock();
4397 	for_each_fib6_node_rt_rcu(&table->tb6_root) {
4398 		struct net_device *dev = fib6_info_nh_dev(rt);
4399 		struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4400 
4401 		if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4402 		    (!idev || idev->cnf.accept_ra != 2) &&
4403 		    fib6_info_hold_safe(rt)) {
4404 			rcu_read_unlock();
4405 			ip6_del_rt(net, rt, false);
4406 			goto restart;
4407 		}
4408 	}
4409 	rcu_read_unlock();
4410 
4411 	table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4412 }
4413 
4414 void rt6_purge_dflt_routers(struct net *net)
4415 {
4416 	struct fib6_table *table;
4417 	struct hlist_head *head;
4418 	unsigned int h;
4419 
4420 	rcu_read_lock();
4421 
4422 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4423 		head = &net->ipv6.fib_table_hash[h];
4424 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4425 			if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4426 				__rt6_purge_dflt_routers(net, table);
4427 		}
4428 	}
4429 
4430 	rcu_read_unlock();
4431 }
4432 
4433 static void rtmsg_to_fib6_config(struct net *net,
4434 				 struct in6_rtmsg *rtmsg,
4435 				 struct fib6_config *cfg)
4436 {
4437 	*cfg = (struct fib6_config){
4438 		.fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4439 			 : RT6_TABLE_MAIN,
4440 		.fc_ifindex = rtmsg->rtmsg_ifindex,
4441 		.fc_metric = rtmsg->rtmsg_metric,
4442 		.fc_expires = rtmsg->rtmsg_info,
4443 		.fc_dst_len = rtmsg->rtmsg_dst_len,
4444 		.fc_src_len = rtmsg->rtmsg_src_len,
4445 		.fc_flags = rtmsg->rtmsg_flags,
4446 		.fc_type = rtmsg->rtmsg_type,
4447 
4448 		.fc_nlinfo.nl_net = net,
4449 
4450 		.fc_dst = rtmsg->rtmsg_dst,
4451 		.fc_src = rtmsg->rtmsg_src,
4452 		.fc_gateway = rtmsg->rtmsg_gateway,
4453 	};
4454 }
4455 
4456 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4457 {
4458 	struct fib6_config cfg;
4459 	int err;
4460 
4461 	if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4462 		return -EINVAL;
4463 	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4464 		return -EPERM;
4465 
4466 	rtmsg_to_fib6_config(net, rtmsg, &cfg);
4467 
4468 	rtnl_lock();
4469 	switch (cmd) {
4470 	case SIOCADDRT:
4471 		/* Only do the default setting of fc_metric in route adding */
4472 		if (cfg.fc_metric == 0)
4473 			cfg.fc_metric = IP6_RT_PRIO_USER;
4474 		err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4475 		break;
4476 	case SIOCDELRT:
4477 		err = ip6_route_del(&cfg, NULL);
4478 		break;
4479 	}
4480 	rtnl_unlock();
4481 	return err;
4482 }
4483 
4484 /*
4485  *	Drop the packet on the floor
4486  */
4487 
4488 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4489 {
4490 	struct dst_entry *dst = skb_dst(skb);
4491 	struct net *net = dev_net(dst->dev);
4492 	struct inet6_dev *idev;
4493 	SKB_DR(reason);
4494 	int type;
4495 
4496 	if (netif_is_l3_master(skb->dev) ||
4497 	    dst->dev == net->loopback_dev)
4498 		idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4499 	else
4500 		idev = ip6_dst_idev(dst);
4501 
4502 	switch (ipstats_mib_noroutes) {
4503 	case IPSTATS_MIB_INNOROUTES:
4504 		type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4505 		if (type == IPV6_ADDR_ANY) {
4506 			SKB_DR_SET(reason, IP_INADDRERRORS);
4507 			IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4508 			break;
4509 		}
4510 		SKB_DR_SET(reason, IP_INNOROUTES);
4511 		fallthrough;
4512 	case IPSTATS_MIB_OUTNOROUTES:
4513 		SKB_DR_OR(reason, IP_OUTNOROUTES);
4514 		IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4515 		break;
4516 	}
4517 
4518 	/* Start over by dropping the dst for l3mdev case */
4519 	if (netif_is_l3_master(skb->dev))
4520 		skb_dst_drop(skb);
4521 
4522 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4523 	kfree_skb_reason(skb, reason);
4524 	return 0;
4525 }
4526 
4527 static int ip6_pkt_discard(struct sk_buff *skb)
4528 {
4529 	return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4530 }
4531 
4532 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4533 {
4534 	skb->dev = skb_dst(skb)->dev;
4535 	return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4536 }
4537 
4538 static int ip6_pkt_prohibit(struct sk_buff *skb)
4539 {
4540 	return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4541 }
4542 
4543 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4544 {
4545 	skb->dev = skb_dst(skb)->dev;
4546 	return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4547 }
4548 
4549 /*
4550  *	Allocate a dst for local (unicast / anycast) address.
4551  */
4552 
4553 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4554 				     struct inet6_dev *idev,
4555 				     const struct in6_addr *addr,
4556 				     bool anycast, gfp_t gfp_flags,
4557 				     struct netlink_ext_ack *extack)
4558 {
4559 	struct fib6_config cfg = {
4560 		.fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4561 		.fc_ifindex = idev->dev->ifindex,
4562 		.fc_flags = RTF_UP | RTF_NONEXTHOP,
4563 		.fc_dst = *addr,
4564 		.fc_dst_len = 128,
4565 		.fc_protocol = RTPROT_KERNEL,
4566 		.fc_nlinfo.nl_net = net,
4567 		.fc_ignore_dev_down = true,
4568 	};
4569 	struct fib6_info *f6i;
4570 
4571 	if (anycast) {
4572 		cfg.fc_type = RTN_ANYCAST;
4573 		cfg.fc_flags |= RTF_ANYCAST;
4574 	} else {
4575 		cfg.fc_type = RTN_LOCAL;
4576 		cfg.fc_flags |= RTF_LOCAL;
4577 	}
4578 
4579 	f6i = ip6_route_info_create(&cfg, gfp_flags, extack);
4580 	if (!IS_ERR(f6i)) {
4581 		f6i->dst_nocount = true;
4582 
4583 		if (!anycast &&
4584 		    (net->ipv6.devconf_all->disable_policy ||
4585 		     idev->cnf.disable_policy))
4586 			f6i->dst_nopolicy = true;
4587 	}
4588 
4589 	return f6i;
4590 }
4591 
4592 /* remove deleted ip from prefsrc entries */
4593 struct arg_dev_net_ip {
4594 	struct net *net;
4595 	struct in6_addr *addr;
4596 };
4597 
4598 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4599 {
4600 	struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4601 	struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4602 
4603 	if (!rt->nh &&
4604 	    rt != net->ipv6.fib6_null_entry &&
4605 	    ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr) &&
4606 	    !ipv6_chk_addr(net, addr, rt->fib6_nh->fib_nh_dev, 0)) {
4607 		spin_lock_bh(&rt6_exception_lock);
4608 		/* remove prefsrc entry */
4609 		rt->fib6_prefsrc.plen = 0;
4610 		spin_unlock_bh(&rt6_exception_lock);
4611 	}
4612 	return 0;
4613 }
4614 
4615 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4616 {
4617 	struct net *net = dev_net(ifp->idev->dev);
4618 	struct arg_dev_net_ip adni = {
4619 		.net = net,
4620 		.addr = &ifp->addr,
4621 	};
4622 	fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4623 }
4624 
4625 #define RTF_RA_ROUTER		(RTF_ADDRCONF | RTF_DEFAULT)
4626 
4627 /* Remove routers and update dst entries when gateway turn into host. */
4628 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4629 {
4630 	struct in6_addr *gateway = (struct in6_addr *)arg;
4631 	struct fib6_nh *nh;
4632 
4633 	/* RA routes do not use nexthops */
4634 	if (rt->nh)
4635 		return 0;
4636 
4637 	nh = rt->fib6_nh;
4638 	if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4639 	    nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4640 		return -1;
4641 
4642 	/* Further clean up cached routes in exception table.
4643 	 * This is needed because cached route may have a different
4644 	 * gateway than its 'parent' in the case of an ip redirect.
4645 	 */
4646 	fib6_nh_exceptions_clean_tohost(nh, gateway);
4647 
4648 	return 0;
4649 }
4650 
4651 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4652 {
4653 	fib6_clean_all(net, fib6_clean_tohost, gateway);
4654 }
4655 
4656 struct arg_netdev_event {
4657 	const struct net_device *dev;
4658 	union {
4659 		unsigned char nh_flags;
4660 		unsigned long event;
4661 	};
4662 };
4663 
4664 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4665 {
4666 	struct fib6_info *iter;
4667 	struct fib6_node *fn;
4668 
4669 	fn = rcu_dereference_protected(rt->fib6_node,
4670 			lockdep_is_held(&rt->fib6_table->tb6_lock));
4671 	iter = rcu_dereference_protected(fn->leaf,
4672 			lockdep_is_held(&rt->fib6_table->tb6_lock));
4673 	while (iter) {
4674 		if (iter->fib6_metric == rt->fib6_metric &&
4675 		    rt6_qualify_for_ecmp(iter))
4676 			return iter;
4677 		iter = rcu_dereference_protected(iter->fib6_next,
4678 				lockdep_is_held(&rt->fib6_table->tb6_lock));
4679 	}
4680 
4681 	return NULL;
4682 }
4683 
4684 /* only called for fib entries with builtin fib6_nh */
4685 static bool rt6_is_dead(const struct fib6_info *rt)
4686 {
4687 	if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4688 	    (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4689 	     ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4690 		return true;
4691 
4692 	return false;
4693 }
4694 
4695 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4696 {
4697 	struct fib6_info *iter;
4698 	int total = 0;
4699 
4700 	if (!rt6_is_dead(rt))
4701 		total += rt->fib6_nh->fib_nh_weight;
4702 
4703 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4704 		if (!rt6_is_dead(iter))
4705 			total += iter->fib6_nh->fib_nh_weight;
4706 	}
4707 
4708 	return total;
4709 }
4710 
4711 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4712 {
4713 	int upper_bound = -1;
4714 
4715 	if (!rt6_is_dead(rt)) {
4716 		*weight += rt->fib6_nh->fib_nh_weight;
4717 		upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4718 						    total) - 1;
4719 	}
4720 	atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4721 }
4722 
4723 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4724 {
4725 	struct fib6_info *iter;
4726 	int weight = 0;
4727 
4728 	rt6_upper_bound_set(rt, &weight, total);
4729 
4730 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4731 		rt6_upper_bound_set(iter, &weight, total);
4732 }
4733 
4734 void rt6_multipath_rebalance(struct fib6_info *rt)
4735 {
4736 	struct fib6_info *first;
4737 	int total;
4738 
4739 	/* In case the entire multipath route was marked for flushing,
4740 	 * then there is no need to rebalance upon the removal of every
4741 	 * sibling route.
4742 	 */
4743 	if (!rt->fib6_nsiblings || rt->should_flush)
4744 		return;
4745 
4746 	/* During lookup routes are evaluated in order, so we need to
4747 	 * make sure upper bounds are assigned from the first sibling
4748 	 * onwards.
4749 	 */
4750 	first = rt6_multipath_first_sibling(rt);
4751 	if (WARN_ON_ONCE(!first))
4752 		return;
4753 
4754 	total = rt6_multipath_total_weight(first);
4755 	rt6_multipath_upper_bound_set(first, total);
4756 }
4757 
4758 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4759 {
4760 	const struct arg_netdev_event *arg = p_arg;
4761 	struct net *net = dev_net(arg->dev);
4762 
4763 	if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4764 	    rt->fib6_nh->fib_nh_dev == arg->dev) {
4765 		rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4766 		fib6_update_sernum_upto_root(net, rt);
4767 		rt6_multipath_rebalance(rt);
4768 	}
4769 
4770 	return 0;
4771 }
4772 
4773 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4774 {
4775 	struct arg_netdev_event arg = {
4776 		.dev = dev,
4777 		{
4778 			.nh_flags = nh_flags,
4779 		},
4780 	};
4781 
4782 	if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4783 		arg.nh_flags |= RTNH_F_LINKDOWN;
4784 
4785 	fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4786 }
4787 
4788 /* only called for fib entries with inline fib6_nh */
4789 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4790 				   const struct net_device *dev)
4791 {
4792 	struct fib6_info *iter;
4793 
4794 	if (rt->fib6_nh->fib_nh_dev == dev)
4795 		return true;
4796 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4797 		if (iter->fib6_nh->fib_nh_dev == dev)
4798 			return true;
4799 
4800 	return false;
4801 }
4802 
4803 static void rt6_multipath_flush(struct fib6_info *rt)
4804 {
4805 	struct fib6_info *iter;
4806 
4807 	rt->should_flush = 1;
4808 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4809 		iter->should_flush = 1;
4810 }
4811 
4812 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4813 					     const struct net_device *down_dev)
4814 {
4815 	struct fib6_info *iter;
4816 	unsigned int dead = 0;
4817 
4818 	if (rt->fib6_nh->fib_nh_dev == down_dev ||
4819 	    rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4820 		dead++;
4821 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4822 		if (iter->fib6_nh->fib_nh_dev == down_dev ||
4823 		    iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4824 			dead++;
4825 
4826 	return dead;
4827 }
4828 
4829 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4830 				       const struct net_device *dev,
4831 				       unsigned char nh_flags)
4832 {
4833 	struct fib6_info *iter;
4834 
4835 	if (rt->fib6_nh->fib_nh_dev == dev)
4836 		rt->fib6_nh->fib_nh_flags |= nh_flags;
4837 	list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4838 		if (iter->fib6_nh->fib_nh_dev == dev)
4839 			iter->fib6_nh->fib_nh_flags |= nh_flags;
4840 }
4841 
4842 /* called with write lock held for table with rt */
4843 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4844 {
4845 	const struct arg_netdev_event *arg = p_arg;
4846 	const struct net_device *dev = arg->dev;
4847 	struct net *net = dev_net(dev);
4848 
4849 	if (rt == net->ipv6.fib6_null_entry || rt->nh)
4850 		return 0;
4851 
4852 	switch (arg->event) {
4853 	case NETDEV_UNREGISTER:
4854 		return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4855 	case NETDEV_DOWN:
4856 		if (rt->should_flush)
4857 			return -1;
4858 		if (!rt->fib6_nsiblings)
4859 			return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4860 		if (rt6_multipath_uses_dev(rt, dev)) {
4861 			unsigned int count;
4862 
4863 			count = rt6_multipath_dead_count(rt, dev);
4864 			if (rt->fib6_nsiblings + 1 == count) {
4865 				rt6_multipath_flush(rt);
4866 				return -1;
4867 			}
4868 			rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4869 						   RTNH_F_LINKDOWN);
4870 			fib6_update_sernum(net, rt);
4871 			rt6_multipath_rebalance(rt);
4872 		}
4873 		return -2;
4874 	case NETDEV_CHANGE:
4875 		if (rt->fib6_nh->fib_nh_dev != dev ||
4876 		    rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4877 			break;
4878 		rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4879 		rt6_multipath_rebalance(rt);
4880 		break;
4881 	}
4882 
4883 	return 0;
4884 }
4885 
4886 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4887 {
4888 	struct arg_netdev_event arg = {
4889 		.dev = dev,
4890 		{
4891 			.event = event,
4892 		},
4893 	};
4894 	struct net *net = dev_net(dev);
4895 
4896 	if (net->ipv6.sysctl.skip_notify_on_dev_down)
4897 		fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4898 	else
4899 		fib6_clean_all(net, fib6_ifdown, &arg);
4900 }
4901 
4902 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4903 {
4904 	rt6_sync_down_dev(dev, event);
4905 	rt6_uncached_list_flush_dev(dev);
4906 	neigh_ifdown(&nd_tbl, dev);
4907 }
4908 
4909 struct rt6_mtu_change_arg {
4910 	struct net_device *dev;
4911 	unsigned int mtu;
4912 	struct fib6_info *f6i;
4913 };
4914 
4915 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4916 {
4917 	struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4918 	struct fib6_info *f6i = arg->f6i;
4919 
4920 	/* For administrative MTU increase, there is no way to discover
4921 	 * IPv6 PMTU increase, so PMTU increase should be updated here.
4922 	 * Since RFC 1981 doesn't include administrative MTU increase
4923 	 * update PMTU increase is a MUST. (i.e. jumbo frame)
4924 	 */
4925 	if (nh->fib_nh_dev == arg->dev) {
4926 		struct inet6_dev *idev = __in6_dev_get(arg->dev);
4927 		u32 mtu = f6i->fib6_pmtu;
4928 
4929 		if (mtu >= arg->mtu ||
4930 		    (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4931 			fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4932 
4933 		spin_lock_bh(&rt6_exception_lock);
4934 		rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4935 		spin_unlock_bh(&rt6_exception_lock);
4936 	}
4937 
4938 	return 0;
4939 }
4940 
4941 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4942 {
4943 	struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4944 	struct inet6_dev *idev;
4945 
4946 	/* In IPv6 pmtu discovery is not optional,
4947 	   so that RTAX_MTU lock cannot disable it.
4948 	   We still use this lock to block changes
4949 	   caused by addrconf/ndisc.
4950 	*/
4951 
4952 	idev = __in6_dev_get(arg->dev);
4953 	if (!idev)
4954 		return 0;
4955 
4956 	if (fib6_metric_locked(f6i, RTAX_MTU))
4957 		return 0;
4958 
4959 	arg->f6i = f6i;
4960 	if (f6i->nh) {
4961 		/* fib6_nh_mtu_change only returns 0, so this is safe */
4962 		return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4963 						arg);
4964 	}
4965 
4966 	return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4967 }
4968 
4969 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4970 {
4971 	struct rt6_mtu_change_arg arg = {
4972 		.dev = dev,
4973 		.mtu = mtu,
4974 	};
4975 
4976 	fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4977 }
4978 
4979 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4980 	[RTA_UNSPEC]		= { .strict_start_type = RTA_DPORT + 1 },
4981 	[RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4982 	[RTA_PREFSRC]		= { .len = sizeof(struct in6_addr) },
4983 	[RTA_OIF]               = { .type = NLA_U32 },
4984 	[RTA_IIF]		= { .type = NLA_U32 },
4985 	[RTA_PRIORITY]          = { .type = NLA_U32 },
4986 	[RTA_METRICS]           = { .type = NLA_NESTED },
4987 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
4988 	[RTA_PREF]              = { .type = NLA_U8 },
4989 	[RTA_ENCAP_TYPE]	= { .type = NLA_U16 },
4990 	[RTA_ENCAP]		= { .type = NLA_NESTED },
4991 	[RTA_EXPIRES]		= { .type = NLA_U32 },
4992 	[RTA_UID]		= { .type = NLA_U32 },
4993 	[RTA_MARK]		= { .type = NLA_U32 },
4994 	[RTA_TABLE]		= { .type = NLA_U32 },
4995 	[RTA_IP_PROTO]		= { .type = NLA_U8 },
4996 	[RTA_SPORT]		= { .type = NLA_U16 },
4997 	[RTA_DPORT]		= { .type = NLA_U16 },
4998 	[RTA_NH_ID]		= { .type = NLA_U32 },
4999 };
5000 
5001 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
5002 			      struct fib6_config *cfg,
5003 			      struct netlink_ext_ack *extack)
5004 {
5005 	struct rtmsg *rtm;
5006 	struct nlattr *tb[RTA_MAX+1];
5007 	unsigned int pref;
5008 	int err;
5009 
5010 	err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5011 				     rtm_ipv6_policy, extack);
5012 	if (err < 0)
5013 		goto errout;
5014 
5015 	err = -EINVAL;
5016 	rtm = nlmsg_data(nlh);
5017 
5018 	if (rtm->rtm_tos) {
5019 		NL_SET_ERR_MSG(extack,
5020 			       "Invalid dsfield (tos): option not available for IPv6");
5021 		goto errout;
5022 	}
5023 
5024 	*cfg = (struct fib6_config){
5025 		.fc_table = rtm->rtm_table,
5026 		.fc_dst_len = rtm->rtm_dst_len,
5027 		.fc_src_len = rtm->rtm_src_len,
5028 		.fc_flags = RTF_UP,
5029 		.fc_protocol = rtm->rtm_protocol,
5030 		.fc_type = rtm->rtm_type,
5031 
5032 		.fc_nlinfo.portid = NETLINK_CB(skb).portid,
5033 		.fc_nlinfo.nlh = nlh,
5034 		.fc_nlinfo.nl_net = sock_net(skb->sk),
5035 	};
5036 
5037 	if (rtm->rtm_type == RTN_UNREACHABLE ||
5038 	    rtm->rtm_type == RTN_BLACKHOLE ||
5039 	    rtm->rtm_type == RTN_PROHIBIT ||
5040 	    rtm->rtm_type == RTN_THROW)
5041 		cfg->fc_flags |= RTF_REJECT;
5042 
5043 	if (rtm->rtm_type == RTN_LOCAL)
5044 		cfg->fc_flags |= RTF_LOCAL;
5045 
5046 	if (rtm->rtm_flags & RTM_F_CLONED)
5047 		cfg->fc_flags |= RTF_CACHE;
5048 
5049 	cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
5050 
5051 	if (tb[RTA_NH_ID]) {
5052 		if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
5053 		    tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
5054 			NL_SET_ERR_MSG(extack,
5055 				       "Nexthop specification and nexthop id are mutually exclusive");
5056 			goto errout;
5057 		}
5058 		cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
5059 	}
5060 
5061 	if (tb[RTA_GATEWAY]) {
5062 		cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
5063 		cfg->fc_flags |= RTF_GATEWAY;
5064 	}
5065 	if (tb[RTA_VIA]) {
5066 		NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
5067 		goto errout;
5068 	}
5069 
5070 	if (tb[RTA_DST]) {
5071 		int plen = (rtm->rtm_dst_len + 7) >> 3;
5072 
5073 		if (nla_len(tb[RTA_DST]) < plen)
5074 			goto errout;
5075 
5076 		nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
5077 	}
5078 
5079 	if (tb[RTA_SRC]) {
5080 		int plen = (rtm->rtm_src_len + 7) >> 3;
5081 
5082 		if (nla_len(tb[RTA_SRC]) < plen)
5083 			goto errout;
5084 
5085 		nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
5086 	}
5087 
5088 	if (tb[RTA_PREFSRC])
5089 		cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
5090 
5091 	if (tb[RTA_OIF])
5092 		cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
5093 
5094 	if (tb[RTA_PRIORITY])
5095 		cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
5096 
5097 	if (tb[RTA_METRICS]) {
5098 		cfg->fc_mx = nla_data(tb[RTA_METRICS]);
5099 		cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
5100 	}
5101 
5102 	if (tb[RTA_TABLE])
5103 		cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
5104 
5105 	if (tb[RTA_MULTIPATH]) {
5106 		cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
5107 		cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
5108 
5109 		err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
5110 						     cfg->fc_mp_len, extack);
5111 		if (err < 0)
5112 			goto errout;
5113 	}
5114 
5115 	if (tb[RTA_PREF]) {
5116 		pref = nla_get_u8(tb[RTA_PREF]);
5117 		if (pref != ICMPV6_ROUTER_PREF_LOW &&
5118 		    pref != ICMPV6_ROUTER_PREF_HIGH)
5119 			pref = ICMPV6_ROUTER_PREF_MEDIUM;
5120 		cfg->fc_flags |= RTF_PREF(pref);
5121 	}
5122 
5123 	if (tb[RTA_ENCAP])
5124 		cfg->fc_encap = tb[RTA_ENCAP];
5125 
5126 	if (tb[RTA_ENCAP_TYPE]) {
5127 		cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5128 
5129 		err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5130 		if (err < 0)
5131 			goto errout;
5132 	}
5133 
5134 	if (tb[RTA_EXPIRES]) {
5135 		unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5136 
5137 		if (addrconf_finite_timeout(timeout)) {
5138 			cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5139 			cfg->fc_flags |= RTF_EXPIRES;
5140 		}
5141 	}
5142 
5143 	err = 0;
5144 errout:
5145 	return err;
5146 }
5147 
5148 struct rt6_nh {
5149 	struct fib6_info *fib6_info;
5150 	struct fib6_config r_cfg;
5151 	struct list_head next;
5152 };
5153 
5154 static int ip6_route_info_append(struct net *net,
5155 				 struct list_head *rt6_nh_list,
5156 				 struct fib6_info *rt,
5157 				 struct fib6_config *r_cfg)
5158 {
5159 	struct rt6_nh *nh;
5160 	int err = -EEXIST;
5161 
5162 	list_for_each_entry(nh, rt6_nh_list, next) {
5163 		/* check if fib6_info already exists */
5164 		if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5165 			return err;
5166 	}
5167 
5168 	nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5169 	if (!nh)
5170 		return -ENOMEM;
5171 	nh->fib6_info = rt;
5172 	memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5173 	list_add_tail(&nh->next, rt6_nh_list);
5174 
5175 	return 0;
5176 }
5177 
5178 static void ip6_route_mpath_notify(struct fib6_info *rt,
5179 				   struct fib6_info *rt_last,
5180 				   struct nl_info *info,
5181 				   __u16 nlflags)
5182 {
5183 	/* if this is an APPEND route, then rt points to the first route
5184 	 * inserted and rt_last points to last route inserted. Userspace
5185 	 * wants a consistent dump of the route which starts at the first
5186 	 * nexthop. Since sibling routes are always added at the end of
5187 	 * the list, find the first sibling of the last route appended
5188 	 */
5189 	if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5190 		rt = list_first_entry(&rt_last->fib6_siblings,
5191 				      struct fib6_info,
5192 				      fib6_siblings);
5193 	}
5194 
5195 	if (rt)
5196 		inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5197 }
5198 
5199 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5200 {
5201 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5202 	bool should_notify = false;
5203 	struct fib6_info *leaf;
5204 	struct fib6_node *fn;
5205 
5206 	rcu_read_lock();
5207 	fn = rcu_dereference(rt->fib6_node);
5208 	if (!fn)
5209 		goto out;
5210 
5211 	leaf = rcu_dereference(fn->leaf);
5212 	if (!leaf)
5213 		goto out;
5214 
5215 	if (rt == leaf ||
5216 	    (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5217 	     rt6_qualify_for_ecmp(leaf)))
5218 		should_notify = true;
5219 out:
5220 	rcu_read_unlock();
5221 
5222 	return should_notify;
5223 }
5224 
5225 static int fib6_gw_from_attr(struct in6_addr *gw, struct nlattr *nla,
5226 			     struct netlink_ext_ack *extack)
5227 {
5228 	if (nla_len(nla) < sizeof(*gw)) {
5229 		NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_GATEWAY");
5230 		return -EINVAL;
5231 	}
5232 
5233 	*gw = nla_get_in6_addr(nla);
5234 
5235 	return 0;
5236 }
5237 
5238 static int ip6_route_multipath_add(struct fib6_config *cfg,
5239 				   struct netlink_ext_ack *extack)
5240 {
5241 	struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5242 	struct nl_info *info = &cfg->fc_nlinfo;
5243 	struct fib6_config r_cfg;
5244 	struct rtnexthop *rtnh;
5245 	struct fib6_info *rt;
5246 	struct rt6_nh *err_nh;
5247 	struct rt6_nh *nh, *nh_safe;
5248 	__u16 nlflags;
5249 	int remaining;
5250 	int attrlen;
5251 	int err = 1;
5252 	int nhn = 0;
5253 	int replace = (cfg->fc_nlinfo.nlh &&
5254 		       (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5255 	LIST_HEAD(rt6_nh_list);
5256 
5257 	nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5258 	if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5259 		nlflags |= NLM_F_APPEND;
5260 
5261 	remaining = cfg->fc_mp_len;
5262 	rtnh = (struct rtnexthop *)cfg->fc_mp;
5263 
5264 	/* Parse a Multipath Entry and build a list (rt6_nh_list) of
5265 	 * fib6_info structs per nexthop
5266 	 */
5267 	while (rtnh_ok(rtnh, remaining)) {
5268 		memcpy(&r_cfg, cfg, sizeof(*cfg));
5269 		if (rtnh->rtnh_ifindex)
5270 			r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5271 
5272 		attrlen = rtnh_attrlen(rtnh);
5273 		if (attrlen > 0) {
5274 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5275 
5276 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5277 			if (nla) {
5278 				err = fib6_gw_from_attr(&r_cfg.fc_gateway, nla,
5279 							extack);
5280 				if (err)
5281 					goto cleanup;
5282 
5283 				r_cfg.fc_flags |= RTF_GATEWAY;
5284 			}
5285 			r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5286 
5287 			/* RTA_ENCAP_TYPE length checked in
5288 			 * lwtunnel_valid_encap_type_attr
5289 			 */
5290 			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5291 			if (nla)
5292 				r_cfg.fc_encap_type = nla_get_u16(nla);
5293 		}
5294 
5295 		r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5296 		rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5297 		if (IS_ERR(rt)) {
5298 			err = PTR_ERR(rt);
5299 			rt = NULL;
5300 			goto cleanup;
5301 		}
5302 		if (!rt6_qualify_for_ecmp(rt)) {
5303 			err = -EINVAL;
5304 			NL_SET_ERR_MSG(extack,
5305 				       "Device only routes can not be added for IPv6 using the multipath API.");
5306 			fib6_info_release(rt);
5307 			goto cleanup;
5308 		}
5309 
5310 		rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5311 
5312 		err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5313 					    rt, &r_cfg);
5314 		if (err) {
5315 			fib6_info_release(rt);
5316 			goto cleanup;
5317 		}
5318 
5319 		rtnh = rtnh_next(rtnh, &remaining);
5320 	}
5321 
5322 	if (list_empty(&rt6_nh_list)) {
5323 		NL_SET_ERR_MSG(extack,
5324 			       "Invalid nexthop configuration - no valid nexthops");
5325 		return -EINVAL;
5326 	}
5327 
5328 	/* for add and replace send one notification with all nexthops.
5329 	 * Skip the notification in fib6_add_rt2node and send one with
5330 	 * the full route when done
5331 	 */
5332 	info->skip_notify = 1;
5333 
5334 	/* For add and replace, send one notification with all nexthops. For
5335 	 * append, send one notification with all appended nexthops.
5336 	 */
5337 	info->skip_notify_kernel = 1;
5338 
5339 	err_nh = NULL;
5340 	list_for_each_entry(nh, &rt6_nh_list, next) {
5341 		err = __ip6_ins_rt(nh->fib6_info, info, extack);
5342 
5343 		if (err) {
5344 			if (replace && nhn)
5345 				NL_SET_ERR_MSG_MOD(extack,
5346 						   "multipath route replace failed (check consistency of installed routes)");
5347 			err_nh = nh;
5348 			goto add_errout;
5349 		}
5350 		/* save reference to last route successfully inserted */
5351 		rt_last = nh->fib6_info;
5352 
5353 		/* save reference to first route for notification */
5354 		if (!rt_notif)
5355 			rt_notif = nh->fib6_info;
5356 
5357 		/* Because each route is added like a single route we remove
5358 		 * these flags after the first nexthop: if there is a collision,
5359 		 * we have already failed to add the first nexthop:
5360 		 * fib6_add_rt2node() has rejected it; when replacing, old
5361 		 * nexthops have been replaced by first new, the rest should
5362 		 * be added to it.
5363 		 */
5364 		if (cfg->fc_nlinfo.nlh) {
5365 			cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5366 							     NLM_F_REPLACE);
5367 			cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5368 		}
5369 		nhn++;
5370 	}
5371 
5372 	/* An in-kernel notification should only be sent in case the new
5373 	 * multipath route is added as the first route in the node, or if
5374 	 * it was appended to it. We pass 'rt_notif' since it is the first
5375 	 * sibling and might allow us to skip some checks in the replace case.
5376 	 */
5377 	if (ip6_route_mpath_should_notify(rt_notif)) {
5378 		enum fib_event_type fib_event;
5379 
5380 		if (rt_notif->fib6_nsiblings != nhn - 1)
5381 			fib_event = FIB_EVENT_ENTRY_APPEND;
5382 		else
5383 			fib_event = FIB_EVENT_ENTRY_REPLACE;
5384 
5385 		err = call_fib6_multipath_entry_notifiers(info->nl_net,
5386 							  fib_event, rt_notif,
5387 							  nhn - 1, extack);
5388 		if (err) {
5389 			/* Delete all the siblings that were just added */
5390 			err_nh = NULL;
5391 			goto add_errout;
5392 		}
5393 	}
5394 
5395 	/* success ... tell user about new route */
5396 	ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5397 	goto cleanup;
5398 
5399 add_errout:
5400 	/* send notification for routes that were added so that
5401 	 * the delete notifications sent by ip6_route_del are
5402 	 * coherent
5403 	 */
5404 	if (rt_notif)
5405 		ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5406 
5407 	/* Delete routes that were already added */
5408 	list_for_each_entry(nh, &rt6_nh_list, next) {
5409 		if (err_nh == nh)
5410 			break;
5411 		ip6_route_del(&nh->r_cfg, extack);
5412 	}
5413 
5414 cleanup:
5415 	list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5416 		fib6_info_release(nh->fib6_info);
5417 		list_del(&nh->next);
5418 		kfree(nh);
5419 	}
5420 
5421 	return err;
5422 }
5423 
5424 static int ip6_route_multipath_del(struct fib6_config *cfg,
5425 				   struct netlink_ext_ack *extack)
5426 {
5427 	struct fib6_config r_cfg;
5428 	struct rtnexthop *rtnh;
5429 	int last_err = 0;
5430 	int remaining;
5431 	int attrlen;
5432 	int err;
5433 
5434 	remaining = cfg->fc_mp_len;
5435 	rtnh = (struct rtnexthop *)cfg->fc_mp;
5436 
5437 	/* Parse a Multipath Entry */
5438 	while (rtnh_ok(rtnh, remaining)) {
5439 		memcpy(&r_cfg, cfg, sizeof(*cfg));
5440 		if (rtnh->rtnh_ifindex)
5441 			r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5442 
5443 		attrlen = rtnh_attrlen(rtnh);
5444 		if (attrlen > 0) {
5445 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5446 
5447 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5448 			if (nla) {
5449 				err = fib6_gw_from_attr(&r_cfg.fc_gateway, nla,
5450 							extack);
5451 				if (err) {
5452 					last_err = err;
5453 					goto next_rtnh;
5454 				}
5455 
5456 				r_cfg.fc_flags |= RTF_GATEWAY;
5457 			}
5458 		}
5459 		err = ip6_route_del(&r_cfg, extack);
5460 		if (err)
5461 			last_err = err;
5462 
5463 next_rtnh:
5464 		rtnh = rtnh_next(rtnh, &remaining);
5465 	}
5466 
5467 	return last_err;
5468 }
5469 
5470 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5471 			      struct netlink_ext_ack *extack)
5472 {
5473 	struct fib6_config cfg;
5474 	int err;
5475 
5476 	err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5477 	if (err < 0)
5478 		return err;
5479 
5480 	if (cfg.fc_nh_id &&
5481 	    !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5482 		NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5483 		return -EINVAL;
5484 	}
5485 
5486 	if (cfg.fc_mp)
5487 		return ip6_route_multipath_del(&cfg, extack);
5488 	else {
5489 		cfg.fc_delete_all_nh = 1;
5490 		return ip6_route_del(&cfg, extack);
5491 	}
5492 }
5493 
5494 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5495 			      struct netlink_ext_ack *extack)
5496 {
5497 	struct fib6_config cfg;
5498 	int err;
5499 
5500 	err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5501 	if (err < 0)
5502 		return err;
5503 
5504 	if (cfg.fc_metric == 0)
5505 		cfg.fc_metric = IP6_RT_PRIO_USER;
5506 
5507 	if (cfg.fc_mp)
5508 		return ip6_route_multipath_add(&cfg, extack);
5509 	else
5510 		return ip6_route_add(&cfg, GFP_KERNEL, extack);
5511 }
5512 
5513 /* add the overhead of this fib6_nh to nexthop_len */
5514 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5515 {
5516 	int *nexthop_len = arg;
5517 
5518 	*nexthop_len += nla_total_size(0)	 /* RTA_MULTIPATH */
5519 		     + NLA_ALIGN(sizeof(struct rtnexthop))
5520 		     + nla_total_size(16); /* RTA_GATEWAY */
5521 
5522 	if (nh->fib_nh_lws) {
5523 		/* RTA_ENCAP_TYPE */
5524 		*nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5525 		/* RTA_ENCAP */
5526 		*nexthop_len += nla_total_size(2);
5527 	}
5528 
5529 	return 0;
5530 }
5531 
5532 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5533 {
5534 	int nexthop_len;
5535 
5536 	if (f6i->nh) {
5537 		nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5538 		nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5539 					 &nexthop_len);
5540 	} else {
5541 		struct fib6_info *sibling, *next_sibling;
5542 		struct fib6_nh *nh = f6i->fib6_nh;
5543 
5544 		nexthop_len = 0;
5545 		if (f6i->fib6_nsiblings) {
5546 			rt6_nh_nlmsg_size(nh, &nexthop_len);
5547 
5548 			list_for_each_entry_safe(sibling, next_sibling,
5549 						 &f6i->fib6_siblings, fib6_siblings) {
5550 				rt6_nh_nlmsg_size(sibling->fib6_nh, &nexthop_len);
5551 			}
5552 		}
5553 		nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5554 	}
5555 
5556 	return NLMSG_ALIGN(sizeof(struct rtmsg))
5557 	       + nla_total_size(16) /* RTA_SRC */
5558 	       + nla_total_size(16) /* RTA_DST */
5559 	       + nla_total_size(16) /* RTA_GATEWAY */
5560 	       + nla_total_size(16) /* RTA_PREFSRC */
5561 	       + nla_total_size(4) /* RTA_TABLE */
5562 	       + nla_total_size(4) /* RTA_IIF */
5563 	       + nla_total_size(4) /* RTA_OIF */
5564 	       + nla_total_size(4) /* RTA_PRIORITY */
5565 	       + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5566 	       + nla_total_size(sizeof(struct rta_cacheinfo))
5567 	       + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5568 	       + nla_total_size(1) /* RTA_PREF */
5569 	       + nexthop_len;
5570 }
5571 
5572 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5573 				 unsigned char *flags)
5574 {
5575 	if (nexthop_is_multipath(nh)) {
5576 		struct nlattr *mp;
5577 
5578 		mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5579 		if (!mp)
5580 			goto nla_put_failure;
5581 
5582 		if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5583 			goto nla_put_failure;
5584 
5585 		nla_nest_end(skb, mp);
5586 	} else {
5587 		struct fib6_nh *fib6_nh;
5588 
5589 		fib6_nh = nexthop_fib6_nh(nh);
5590 		if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5591 				     flags, false) < 0)
5592 			goto nla_put_failure;
5593 	}
5594 
5595 	return 0;
5596 
5597 nla_put_failure:
5598 	return -EMSGSIZE;
5599 }
5600 
5601 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5602 			 struct fib6_info *rt, struct dst_entry *dst,
5603 			 struct in6_addr *dest, struct in6_addr *src,
5604 			 int iif, int type, u32 portid, u32 seq,
5605 			 unsigned int flags)
5606 {
5607 	struct rt6_info *rt6 = (struct rt6_info *)dst;
5608 	struct rt6key *rt6_dst, *rt6_src;
5609 	u32 *pmetrics, table, rt6_flags;
5610 	unsigned char nh_flags = 0;
5611 	struct nlmsghdr *nlh;
5612 	struct rtmsg *rtm;
5613 	long expires = 0;
5614 
5615 	nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5616 	if (!nlh)
5617 		return -EMSGSIZE;
5618 
5619 	if (rt6) {
5620 		rt6_dst = &rt6->rt6i_dst;
5621 		rt6_src = &rt6->rt6i_src;
5622 		rt6_flags = rt6->rt6i_flags;
5623 	} else {
5624 		rt6_dst = &rt->fib6_dst;
5625 		rt6_src = &rt->fib6_src;
5626 		rt6_flags = rt->fib6_flags;
5627 	}
5628 
5629 	rtm = nlmsg_data(nlh);
5630 	rtm->rtm_family = AF_INET6;
5631 	rtm->rtm_dst_len = rt6_dst->plen;
5632 	rtm->rtm_src_len = rt6_src->plen;
5633 	rtm->rtm_tos = 0;
5634 	if (rt->fib6_table)
5635 		table = rt->fib6_table->tb6_id;
5636 	else
5637 		table = RT6_TABLE_UNSPEC;
5638 	rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5639 	if (nla_put_u32(skb, RTA_TABLE, table))
5640 		goto nla_put_failure;
5641 
5642 	rtm->rtm_type = rt->fib6_type;
5643 	rtm->rtm_flags = 0;
5644 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5645 	rtm->rtm_protocol = rt->fib6_protocol;
5646 
5647 	if (rt6_flags & RTF_CACHE)
5648 		rtm->rtm_flags |= RTM_F_CLONED;
5649 
5650 	if (dest) {
5651 		if (nla_put_in6_addr(skb, RTA_DST, dest))
5652 			goto nla_put_failure;
5653 		rtm->rtm_dst_len = 128;
5654 	} else if (rtm->rtm_dst_len)
5655 		if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5656 			goto nla_put_failure;
5657 #ifdef CONFIG_IPV6_SUBTREES
5658 	if (src) {
5659 		if (nla_put_in6_addr(skb, RTA_SRC, src))
5660 			goto nla_put_failure;
5661 		rtm->rtm_src_len = 128;
5662 	} else if (rtm->rtm_src_len &&
5663 		   nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5664 		goto nla_put_failure;
5665 #endif
5666 	if (iif) {
5667 #ifdef CONFIG_IPV6_MROUTE
5668 		if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5669 			int err = ip6mr_get_route(net, skb, rtm, portid);
5670 
5671 			if (err == 0)
5672 				return 0;
5673 			if (err < 0)
5674 				goto nla_put_failure;
5675 		} else
5676 #endif
5677 			if (nla_put_u32(skb, RTA_IIF, iif))
5678 				goto nla_put_failure;
5679 	} else if (dest) {
5680 		struct in6_addr saddr_buf;
5681 		if (ip6_route_get_saddr(net, rt, dest, 0, 0, &saddr_buf) == 0 &&
5682 		    nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5683 			goto nla_put_failure;
5684 	}
5685 
5686 	if (rt->fib6_prefsrc.plen) {
5687 		struct in6_addr saddr_buf;
5688 		saddr_buf = rt->fib6_prefsrc.addr;
5689 		if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5690 			goto nla_put_failure;
5691 	}
5692 
5693 	pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5694 	if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5695 		goto nla_put_failure;
5696 
5697 	if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5698 		goto nla_put_failure;
5699 
5700 	/* For multipath routes, walk the siblings list and add
5701 	 * each as a nexthop within RTA_MULTIPATH.
5702 	 */
5703 	if (rt6) {
5704 		if (rt6_flags & RTF_GATEWAY &&
5705 		    nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5706 			goto nla_put_failure;
5707 
5708 		if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5709 			goto nla_put_failure;
5710 
5711 		if (dst->lwtstate &&
5712 		    lwtunnel_fill_encap(skb, dst->lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
5713 			goto nla_put_failure;
5714 	} else if (rt->fib6_nsiblings) {
5715 		struct fib6_info *sibling, *next_sibling;
5716 		struct nlattr *mp;
5717 
5718 		mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5719 		if (!mp)
5720 			goto nla_put_failure;
5721 
5722 		if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5723 				    rt->fib6_nh->fib_nh_weight, AF_INET6,
5724 				    0) < 0)
5725 			goto nla_put_failure;
5726 
5727 		list_for_each_entry_safe(sibling, next_sibling,
5728 					 &rt->fib6_siblings, fib6_siblings) {
5729 			if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5730 					    sibling->fib6_nh->fib_nh_weight,
5731 					    AF_INET6, 0) < 0)
5732 				goto nla_put_failure;
5733 		}
5734 
5735 		nla_nest_end(skb, mp);
5736 	} else if (rt->nh) {
5737 		if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5738 			goto nla_put_failure;
5739 
5740 		if (nexthop_is_blackhole(rt->nh))
5741 			rtm->rtm_type = RTN_BLACKHOLE;
5742 
5743 		if (READ_ONCE(net->ipv4.sysctl_nexthop_compat_mode) &&
5744 		    rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5745 			goto nla_put_failure;
5746 
5747 		rtm->rtm_flags |= nh_flags;
5748 	} else {
5749 		if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5750 				     &nh_flags, false) < 0)
5751 			goto nla_put_failure;
5752 
5753 		rtm->rtm_flags |= nh_flags;
5754 	}
5755 
5756 	if (rt6_flags & RTF_EXPIRES) {
5757 		expires = dst ? dst->expires : rt->expires;
5758 		expires -= jiffies;
5759 	}
5760 
5761 	if (!dst) {
5762 		if (READ_ONCE(rt->offload))
5763 			rtm->rtm_flags |= RTM_F_OFFLOAD;
5764 		if (READ_ONCE(rt->trap))
5765 			rtm->rtm_flags |= RTM_F_TRAP;
5766 		if (READ_ONCE(rt->offload_failed))
5767 			rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
5768 	}
5769 
5770 	if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5771 		goto nla_put_failure;
5772 
5773 	if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5774 		goto nla_put_failure;
5775 
5776 
5777 	nlmsg_end(skb, nlh);
5778 	return 0;
5779 
5780 nla_put_failure:
5781 	nlmsg_cancel(skb, nlh);
5782 	return -EMSGSIZE;
5783 }
5784 
5785 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5786 {
5787 	const struct net_device *dev = arg;
5788 
5789 	if (nh->fib_nh_dev == dev)
5790 		return 1;
5791 
5792 	return 0;
5793 }
5794 
5795 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5796 			       const struct net_device *dev)
5797 {
5798 	if (f6i->nh) {
5799 		struct net_device *_dev = (struct net_device *)dev;
5800 
5801 		return !!nexthop_for_each_fib6_nh(f6i->nh,
5802 						  fib6_info_nh_uses_dev,
5803 						  _dev);
5804 	}
5805 
5806 	if (f6i->fib6_nh->fib_nh_dev == dev)
5807 		return true;
5808 
5809 	if (f6i->fib6_nsiblings) {
5810 		struct fib6_info *sibling, *next_sibling;
5811 
5812 		list_for_each_entry_safe(sibling, next_sibling,
5813 					 &f6i->fib6_siblings, fib6_siblings) {
5814 			if (sibling->fib6_nh->fib_nh_dev == dev)
5815 				return true;
5816 		}
5817 	}
5818 
5819 	return false;
5820 }
5821 
5822 struct fib6_nh_exception_dump_walker {
5823 	struct rt6_rtnl_dump_arg *dump;
5824 	struct fib6_info *rt;
5825 	unsigned int flags;
5826 	unsigned int skip;
5827 	unsigned int count;
5828 };
5829 
5830 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5831 {
5832 	struct fib6_nh_exception_dump_walker *w = arg;
5833 	struct rt6_rtnl_dump_arg *dump = w->dump;
5834 	struct rt6_exception_bucket *bucket;
5835 	struct rt6_exception *rt6_ex;
5836 	int i, err;
5837 
5838 	bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5839 	if (!bucket)
5840 		return 0;
5841 
5842 	for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5843 		hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5844 			if (w->skip) {
5845 				w->skip--;
5846 				continue;
5847 			}
5848 
5849 			/* Expiration of entries doesn't bump sernum, insertion
5850 			 * does. Removal is triggered by insertion, so we can
5851 			 * rely on the fact that if entries change between two
5852 			 * partial dumps, this node is scanned again completely,
5853 			 * see rt6_insert_exception() and fib6_dump_table().
5854 			 *
5855 			 * Count expired entries we go through as handled
5856 			 * entries that we'll skip next time, in case of partial
5857 			 * node dump. Otherwise, if entries expire meanwhile,
5858 			 * we'll skip the wrong amount.
5859 			 */
5860 			if (rt6_check_expired(rt6_ex->rt6i)) {
5861 				w->count++;
5862 				continue;
5863 			}
5864 
5865 			err = rt6_fill_node(dump->net, dump->skb, w->rt,
5866 					    &rt6_ex->rt6i->dst, NULL, NULL, 0,
5867 					    RTM_NEWROUTE,
5868 					    NETLINK_CB(dump->cb->skb).portid,
5869 					    dump->cb->nlh->nlmsg_seq, w->flags);
5870 			if (err)
5871 				return err;
5872 
5873 			w->count++;
5874 		}
5875 		bucket++;
5876 	}
5877 
5878 	return 0;
5879 }
5880 
5881 /* Return -1 if done with node, number of handled routes on partial dump */
5882 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5883 {
5884 	struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5885 	struct fib_dump_filter *filter = &arg->filter;
5886 	unsigned int flags = NLM_F_MULTI;
5887 	struct net *net = arg->net;
5888 	int count = 0;
5889 
5890 	if (rt == net->ipv6.fib6_null_entry)
5891 		return -1;
5892 
5893 	if ((filter->flags & RTM_F_PREFIX) &&
5894 	    !(rt->fib6_flags & RTF_PREFIX_RT)) {
5895 		/* success since this is not a prefix route */
5896 		return -1;
5897 	}
5898 	if (filter->filter_set &&
5899 	    ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5900 	     (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5901 	     (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5902 		return -1;
5903 	}
5904 
5905 	if (filter->filter_set ||
5906 	    !filter->dump_routes || !filter->dump_exceptions) {
5907 		flags |= NLM_F_DUMP_FILTERED;
5908 	}
5909 
5910 	if (filter->dump_routes) {
5911 		if (skip) {
5912 			skip--;
5913 		} else {
5914 			if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5915 					  0, RTM_NEWROUTE,
5916 					  NETLINK_CB(arg->cb->skb).portid,
5917 					  arg->cb->nlh->nlmsg_seq, flags)) {
5918 				return 0;
5919 			}
5920 			count++;
5921 		}
5922 	}
5923 
5924 	if (filter->dump_exceptions) {
5925 		struct fib6_nh_exception_dump_walker w = { .dump = arg,
5926 							   .rt = rt,
5927 							   .flags = flags,
5928 							   .skip = skip,
5929 							   .count = 0 };
5930 		int err;
5931 
5932 		rcu_read_lock();
5933 		if (rt->nh) {
5934 			err = nexthop_for_each_fib6_nh(rt->nh,
5935 						       rt6_nh_dump_exceptions,
5936 						       &w);
5937 		} else {
5938 			err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5939 		}
5940 		rcu_read_unlock();
5941 
5942 		if (err)
5943 			return count + w.count;
5944 	}
5945 
5946 	return -1;
5947 }
5948 
5949 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5950 					const struct nlmsghdr *nlh,
5951 					struct nlattr **tb,
5952 					struct netlink_ext_ack *extack)
5953 {
5954 	struct rtmsg *rtm;
5955 	int i, err;
5956 
5957 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5958 		NL_SET_ERR_MSG_MOD(extack,
5959 				   "Invalid header for get route request");
5960 		return -EINVAL;
5961 	}
5962 
5963 	if (!netlink_strict_get_check(skb))
5964 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5965 					      rtm_ipv6_policy, extack);
5966 
5967 	rtm = nlmsg_data(nlh);
5968 	if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5969 	    (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5970 	    rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5971 	    rtm->rtm_type) {
5972 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5973 		return -EINVAL;
5974 	}
5975 	if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5976 		NL_SET_ERR_MSG_MOD(extack,
5977 				   "Invalid flags for get route request");
5978 		return -EINVAL;
5979 	}
5980 
5981 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5982 					    rtm_ipv6_policy, extack);
5983 	if (err)
5984 		return err;
5985 
5986 	if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5987 	    (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5988 		NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5989 		return -EINVAL;
5990 	}
5991 
5992 	for (i = 0; i <= RTA_MAX; i++) {
5993 		if (!tb[i])
5994 			continue;
5995 
5996 		switch (i) {
5997 		case RTA_SRC:
5998 		case RTA_DST:
5999 		case RTA_IIF:
6000 		case RTA_OIF:
6001 		case RTA_MARK:
6002 		case RTA_UID:
6003 		case RTA_SPORT:
6004 		case RTA_DPORT:
6005 		case RTA_IP_PROTO:
6006 			break;
6007 		default:
6008 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
6009 			return -EINVAL;
6010 		}
6011 	}
6012 
6013 	return 0;
6014 }
6015 
6016 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
6017 			      struct netlink_ext_ack *extack)
6018 {
6019 	struct net *net = sock_net(in_skb->sk);
6020 	struct nlattr *tb[RTA_MAX+1];
6021 	int err, iif = 0, oif = 0;
6022 	struct fib6_info *from;
6023 	struct dst_entry *dst;
6024 	struct rt6_info *rt;
6025 	struct sk_buff *skb;
6026 	struct rtmsg *rtm;
6027 	struct flowi6 fl6 = {};
6028 	bool fibmatch;
6029 
6030 	err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
6031 	if (err < 0)
6032 		goto errout;
6033 
6034 	err = -EINVAL;
6035 	rtm = nlmsg_data(nlh);
6036 	fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
6037 	fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
6038 
6039 	if (tb[RTA_SRC]) {
6040 		if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
6041 			goto errout;
6042 
6043 		fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
6044 	}
6045 
6046 	if (tb[RTA_DST]) {
6047 		if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
6048 			goto errout;
6049 
6050 		fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
6051 	}
6052 
6053 	if (tb[RTA_IIF])
6054 		iif = nla_get_u32(tb[RTA_IIF]);
6055 
6056 	if (tb[RTA_OIF])
6057 		oif = nla_get_u32(tb[RTA_OIF]);
6058 
6059 	if (tb[RTA_MARK])
6060 		fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
6061 
6062 	if (tb[RTA_UID])
6063 		fl6.flowi6_uid = make_kuid(current_user_ns(),
6064 					   nla_get_u32(tb[RTA_UID]));
6065 	else
6066 		fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
6067 
6068 	if (tb[RTA_SPORT])
6069 		fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
6070 
6071 	if (tb[RTA_DPORT])
6072 		fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
6073 
6074 	if (tb[RTA_IP_PROTO]) {
6075 		err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
6076 						  &fl6.flowi6_proto, AF_INET6,
6077 						  extack);
6078 		if (err)
6079 			goto errout;
6080 	}
6081 
6082 	if (iif) {
6083 		struct net_device *dev;
6084 		int flags = 0;
6085 
6086 		rcu_read_lock();
6087 
6088 		dev = dev_get_by_index_rcu(net, iif);
6089 		if (!dev) {
6090 			rcu_read_unlock();
6091 			err = -ENODEV;
6092 			goto errout;
6093 		}
6094 
6095 		fl6.flowi6_iif = iif;
6096 
6097 		if (!ipv6_addr_any(&fl6.saddr))
6098 			flags |= RT6_LOOKUP_F_HAS_SADDR;
6099 
6100 		dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
6101 
6102 		rcu_read_unlock();
6103 	} else {
6104 		fl6.flowi6_oif = oif;
6105 
6106 		dst = ip6_route_output(net, NULL, &fl6);
6107 	}
6108 
6109 
6110 	rt = container_of(dst, struct rt6_info, dst);
6111 	if (rt->dst.error) {
6112 		err = rt->dst.error;
6113 		ip6_rt_put(rt);
6114 		goto errout;
6115 	}
6116 
6117 	if (rt == net->ipv6.ip6_null_entry) {
6118 		err = rt->dst.error;
6119 		ip6_rt_put(rt);
6120 		goto errout;
6121 	}
6122 
6123 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
6124 	if (!skb) {
6125 		ip6_rt_put(rt);
6126 		err = -ENOBUFS;
6127 		goto errout;
6128 	}
6129 
6130 	skb_dst_set(skb, &rt->dst);
6131 
6132 	rcu_read_lock();
6133 	from = rcu_dereference(rt->from);
6134 	if (from) {
6135 		if (fibmatch)
6136 			err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
6137 					    iif, RTM_NEWROUTE,
6138 					    NETLINK_CB(in_skb).portid,
6139 					    nlh->nlmsg_seq, 0);
6140 		else
6141 			err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
6142 					    &fl6.saddr, iif, RTM_NEWROUTE,
6143 					    NETLINK_CB(in_skb).portid,
6144 					    nlh->nlmsg_seq, 0);
6145 	} else {
6146 		err = -ENETUNREACH;
6147 	}
6148 	rcu_read_unlock();
6149 
6150 	if (err < 0) {
6151 		kfree_skb(skb);
6152 		goto errout;
6153 	}
6154 
6155 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
6156 errout:
6157 	return err;
6158 }
6159 
6160 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6161 		     unsigned int nlm_flags)
6162 {
6163 	struct sk_buff *skb;
6164 	struct net *net = info->nl_net;
6165 	u32 seq;
6166 	int err;
6167 
6168 	err = -ENOBUFS;
6169 	seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6170 
6171 	skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6172 	if (!skb)
6173 		goto errout;
6174 
6175 	err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6176 			    event, info->portid, seq, nlm_flags);
6177 	if (err < 0) {
6178 		/* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6179 		WARN_ON(err == -EMSGSIZE);
6180 		kfree_skb(skb);
6181 		goto errout;
6182 	}
6183 	rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6184 		    info->nlh, gfp_any());
6185 	return;
6186 errout:
6187 	if (err < 0)
6188 		rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6189 }
6190 
6191 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6192 		    struct nl_info *info)
6193 {
6194 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6195 	struct sk_buff *skb;
6196 	int err = -ENOBUFS;
6197 
6198 	skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6199 	if (!skb)
6200 		goto errout;
6201 
6202 	err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6203 			    RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6204 	if (err < 0) {
6205 		/* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6206 		WARN_ON(err == -EMSGSIZE);
6207 		kfree_skb(skb);
6208 		goto errout;
6209 	}
6210 	rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6211 		    info->nlh, gfp_any());
6212 	return;
6213 errout:
6214 	if (err < 0)
6215 		rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6216 }
6217 
6218 void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
6219 			    bool offload, bool trap, bool offload_failed)
6220 {
6221 	struct sk_buff *skb;
6222 	int err;
6223 
6224 	if (READ_ONCE(f6i->offload) == offload &&
6225 	    READ_ONCE(f6i->trap) == trap &&
6226 	    READ_ONCE(f6i->offload_failed) == offload_failed)
6227 		return;
6228 
6229 	WRITE_ONCE(f6i->offload, offload);
6230 	WRITE_ONCE(f6i->trap, trap);
6231 
6232 	/* 2 means send notifications only if offload_failed was changed. */
6233 	if (net->ipv6.sysctl.fib_notify_on_flag_change == 2 &&
6234 	    READ_ONCE(f6i->offload_failed) == offload_failed)
6235 		return;
6236 
6237 	WRITE_ONCE(f6i->offload_failed, offload_failed);
6238 
6239 	if (!rcu_access_pointer(f6i->fib6_node))
6240 		/* The route was removed from the tree, do not send
6241 		 * notification.
6242 		 */
6243 		return;
6244 
6245 	if (!net->ipv6.sysctl.fib_notify_on_flag_change)
6246 		return;
6247 
6248 	skb = nlmsg_new(rt6_nlmsg_size(f6i), GFP_KERNEL);
6249 	if (!skb) {
6250 		err = -ENOBUFS;
6251 		goto errout;
6252 	}
6253 
6254 	err = rt6_fill_node(net, skb, f6i, NULL, NULL, NULL, 0, RTM_NEWROUTE, 0,
6255 			    0, 0);
6256 	if (err < 0) {
6257 		/* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6258 		WARN_ON(err == -EMSGSIZE);
6259 		kfree_skb(skb);
6260 		goto errout;
6261 	}
6262 
6263 	rtnl_notify(skb, net, 0, RTNLGRP_IPV6_ROUTE, NULL, GFP_KERNEL);
6264 	return;
6265 
6266 errout:
6267 	rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6268 }
6269 EXPORT_SYMBOL(fib6_info_hw_flags_set);
6270 
6271 static int ip6_route_dev_notify(struct notifier_block *this,
6272 				unsigned long event, void *ptr)
6273 {
6274 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6275 	struct net *net = dev_net(dev);
6276 
6277 	if (!(dev->flags & IFF_LOOPBACK))
6278 		return NOTIFY_OK;
6279 
6280 	if (event == NETDEV_REGISTER) {
6281 		net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6282 		net->ipv6.ip6_null_entry->dst.dev = dev;
6283 		net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6284 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6285 		net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6286 		net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6287 		net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6288 		net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6289 #endif
6290 	 } else if (event == NETDEV_UNREGISTER &&
6291 		    dev->reg_state != NETREG_UNREGISTERED) {
6292 		/* NETDEV_UNREGISTER could be fired for multiple times by
6293 		 * netdev_wait_allrefs(). Make sure we only call this once.
6294 		 */
6295 		in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6296 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6297 		in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6298 		in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6299 #endif
6300 	}
6301 
6302 	return NOTIFY_OK;
6303 }
6304 
6305 /*
6306  *	/proc
6307  */
6308 
6309 #ifdef CONFIG_PROC_FS
6310 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6311 {
6312 	struct net *net = (struct net *)seq->private;
6313 	seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6314 		   net->ipv6.rt6_stats->fib_nodes,
6315 		   net->ipv6.rt6_stats->fib_route_nodes,
6316 		   atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6317 		   net->ipv6.rt6_stats->fib_rt_entries,
6318 		   net->ipv6.rt6_stats->fib_rt_cache,
6319 		   dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6320 		   net->ipv6.rt6_stats->fib_discarded_routes);
6321 
6322 	return 0;
6323 }
6324 #endif	/* CONFIG_PROC_FS */
6325 
6326 #ifdef CONFIG_SYSCTL
6327 
6328 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6329 			      void *buffer, size_t *lenp, loff_t *ppos)
6330 {
6331 	struct net *net;
6332 	int delay;
6333 	int ret;
6334 	if (!write)
6335 		return -EINVAL;
6336 
6337 	ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6338 	if (ret)
6339 		return ret;
6340 
6341 	net = (struct net *)ctl->extra1;
6342 	delay = net->ipv6.sysctl.flush_delay;
6343 	fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6344 	return 0;
6345 }
6346 
6347 static struct ctl_table ipv6_route_table_template[] = {
6348 	{
6349 		.procname	=	"max_size",
6350 		.data		=	&init_net.ipv6.sysctl.ip6_rt_max_size,
6351 		.maxlen		=	sizeof(int),
6352 		.mode		=	0644,
6353 		.proc_handler	=	proc_dointvec,
6354 	},
6355 	{
6356 		.procname	=	"gc_thresh",
6357 		.data		=	&ip6_dst_ops_template.gc_thresh,
6358 		.maxlen		=	sizeof(int),
6359 		.mode		=	0644,
6360 		.proc_handler	=	proc_dointvec,
6361 	},
6362 	{
6363 		.procname	=	"flush",
6364 		.data		=	&init_net.ipv6.sysctl.flush_delay,
6365 		.maxlen		=	sizeof(int),
6366 		.mode		=	0200,
6367 		.proc_handler	=	ipv6_sysctl_rtcache_flush
6368 	},
6369 	{
6370 		.procname	=	"gc_min_interval",
6371 		.data		=	&init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6372 		.maxlen		=	sizeof(int),
6373 		.mode		=	0644,
6374 		.proc_handler	=	proc_dointvec_jiffies,
6375 	},
6376 	{
6377 		.procname	=	"gc_timeout",
6378 		.data		=	&init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6379 		.maxlen		=	sizeof(int),
6380 		.mode		=	0644,
6381 		.proc_handler	=	proc_dointvec_jiffies,
6382 	},
6383 	{
6384 		.procname	=	"gc_interval",
6385 		.data		=	&init_net.ipv6.sysctl.ip6_rt_gc_interval,
6386 		.maxlen		=	sizeof(int),
6387 		.mode		=	0644,
6388 		.proc_handler	=	proc_dointvec_jiffies,
6389 	},
6390 	{
6391 		.procname	=	"gc_elasticity",
6392 		.data		=	&init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6393 		.maxlen		=	sizeof(int),
6394 		.mode		=	0644,
6395 		.proc_handler	=	proc_dointvec,
6396 	},
6397 	{
6398 		.procname	=	"mtu_expires",
6399 		.data		=	&init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6400 		.maxlen		=	sizeof(int),
6401 		.mode		=	0644,
6402 		.proc_handler	=	proc_dointvec_jiffies,
6403 	},
6404 	{
6405 		.procname	=	"min_adv_mss",
6406 		.data		=	&init_net.ipv6.sysctl.ip6_rt_min_advmss,
6407 		.maxlen		=	sizeof(int),
6408 		.mode		=	0644,
6409 		.proc_handler	=	proc_dointvec,
6410 	},
6411 	{
6412 		.procname	=	"gc_min_interval_ms",
6413 		.data		=	&init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6414 		.maxlen		=	sizeof(int),
6415 		.mode		=	0644,
6416 		.proc_handler	=	proc_dointvec_ms_jiffies,
6417 	},
6418 	{
6419 		.procname	=	"skip_notify_on_dev_down",
6420 		.data		=	&init_net.ipv6.sysctl.skip_notify_on_dev_down,
6421 		.maxlen		=	sizeof(u8),
6422 		.mode		=	0644,
6423 		.proc_handler	=	proc_dou8vec_minmax,
6424 		.extra1		=	SYSCTL_ZERO,
6425 		.extra2		=	SYSCTL_ONE,
6426 	},
6427 	{ }
6428 };
6429 
6430 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6431 {
6432 	struct ctl_table *table;
6433 
6434 	table = kmemdup(ipv6_route_table_template,
6435 			sizeof(ipv6_route_table_template),
6436 			GFP_KERNEL);
6437 
6438 	if (table) {
6439 		table[0].data = &net->ipv6.sysctl.ip6_rt_max_size;
6440 		table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6441 		table[2].data = &net->ipv6.sysctl.flush_delay;
6442 		table[2].extra1 = net;
6443 		table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6444 		table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6445 		table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6446 		table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6447 		table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6448 		table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6449 		table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6450 		table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6451 
6452 		/* Don't export sysctls to unprivileged users */
6453 		if (net->user_ns != &init_user_ns)
6454 			table[1].procname = NULL;
6455 	}
6456 
6457 	return table;
6458 }
6459 
6460 size_t ipv6_route_sysctl_table_size(struct net *net)
6461 {
6462 	/* Don't export sysctls to unprivileged users */
6463 	if (net->user_ns != &init_user_ns)
6464 		return 1;
6465 
6466 	return ARRAY_SIZE(ipv6_route_table_template);
6467 }
6468 #endif
6469 
6470 static int __net_init ip6_route_net_init(struct net *net)
6471 {
6472 	int ret = -ENOMEM;
6473 
6474 	memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6475 	       sizeof(net->ipv6.ip6_dst_ops));
6476 
6477 	if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6478 		goto out_ip6_dst_ops;
6479 
6480 	net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6481 	if (!net->ipv6.fib6_null_entry)
6482 		goto out_ip6_dst_entries;
6483 	memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6484 	       sizeof(*net->ipv6.fib6_null_entry));
6485 
6486 	net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6487 					   sizeof(*net->ipv6.ip6_null_entry),
6488 					   GFP_KERNEL);
6489 	if (!net->ipv6.ip6_null_entry)
6490 		goto out_fib6_null_entry;
6491 	net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6492 	dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6493 			 ip6_template_metrics, true);
6494 	INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->dst.rt_uncached);
6495 
6496 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6497 	net->ipv6.fib6_has_custom_rules = false;
6498 	net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6499 					       sizeof(*net->ipv6.ip6_prohibit_entry),
6500 					       GFP_KERNEL);
6501 	if (!net->ipv6.ip6_prohibit_entry)
6502 		goto out_ip6_null_entry;
6503 	net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6504 	dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6505 			 ip6_template_metrics, true);
6506 	INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->dst.rt_uncached);
6507 
6508 	net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6509 					       sizeof(*net->ipv6.ip6_blk_hole_entry),
6510 					       GFP_KERNEL);
6511 	if (!net->ipv6.ip6_blk_hole_entry)
6512 		goto out_ip6_prohibit_entry;
6513 	net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6514 	dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6515 			 ip6_template_metrics, true);
6516 	INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->dst.rt_uncached);
6517 #ifdef CONFIG_IPV6_SUBTREES
6518 	net->ipv6.fib6_routes_require_src = 0;
6519 #endif
6520 #endif
6521 
6522 	net->ipv6.sysctl.flush_delay = 0;
6523 	net->ipv6.sysctl.ip6_rt_max_size = INT_MAX;
6524 	net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6525 	net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6526 	net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6527 	net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6528 	net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6529 	net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6530 	net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6531 
6532 	atomic_set(&net->ipv6.ip6_rt_gc_expire, 30*HZ);
6533 
6534 	ret = 0;
6535 out:
6536 	return ret;
6537 
6538 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6539 out_ip6_prohibit_entry:
6540 	kfree(net->ipv6.ip6_prohibit_entry);
6541 out_ip6_null_entry:
6542 	kfree(net->ipv6.ip6_null_entry);
6543 #endif
6544 out_fib6_null_entry:
6545 	kfree(net->ipv6.fib6_null_entry);
6546 out_ip6_dst_entries:
6547 	dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6548 out_ip6_dst_ops:
6549 	goto out;
6550 }
6551 
6552 static void __net_exit ip6_route_net_exit(struct net *net)
6553 {
6554 	kfree(net->ipv6.fib6_null_entry);
6555 	kfree(net->ipv6.ip6_null_entry);
6556 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6557 	kfree(net->ipv6.ip6_prohibit_entry);
6558 	kfree(net->ipv6.ip6_blk_hole_entry);
6559 #endif
6560 	dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6561 }
6562 
6563 static int __net_init ip6_route_net_init_late(struct net *net)
6564 {
6565 #ifdef CONFIG_PROC_FS
6566 	if (!proc_create_net("ipv6_route", 0, net->proc_net,
6567 			     &ipv6_route_seq_ops,
6568 			     sizeof(struct ipv6_route_iter)))
6569 		return -ENOMEM;
6570 
6571 	if (!proc_create_net_single("rt6_stats", 0444, net->proc_net,
6572 				    rt6_stats_seq_show, NULL)) {
6573 		remove_proc_entry("ipv6_route", net->proc_net);
6574 		return -ENOMEM;
6575 	}
6576 #endif
6577 	return 0;
6578 }
6579 
6580 static void __net_exit ip6_route_net_exit_late(struct net *net)
6581 {
6582 #ifdef CONFIG_PROC_FS
6583 	remove_proc_entry("ipv6_route", net->proc_net);
6584 	remove_proc_entry("rt6_stats", net->proc_net);
6585 #endif
6586 }
6587 
6588 static struct pernet_operations ip6_route_net_ops = {
6589 	.init = ip6_route_net_init,
6590 	.exit = ip6_route_net_exit,
6591 };
6592 
6593 static int __net_init ipv6_inetpeer_init(struct net *net)
6594 {
6595 	struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6596 
6597 	if (!bp)
6598 		return -ENOMEM;
6599 	inet_peer_base_init(bp);
6600 	net->ipv6.peers = bp;
6601 	return 0;
6602 }
6603 
6604 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6605 {
6606 	struct inet_peer_base *bp = net->ipv6.peers;
6607 
6608 	net->ipv6.peers = NULL;
6609 	inetpeer_invalidate_tree(bp);
6610 	kfree(bp);
6611 }
6612 
6613 static struct pernet_operations ipv6_inetpeer_ops = {
6614 	.init	=	ipv6_inetpeer_init,
6615 	.exit	=	ipv6_inetpeer_exit,
6616 };
6617 
6618 static struct pernet_operations ip6_route_net_late_ops = {
6619 	.init = ip6_route_net_init_late,
6620 	.exit = ip6_route_net_exit_late,
6621 };
6622 
6623 static struct notifier_block ip6_route_dev_notifier = {
6624 	.notifier_call = ip6_route_dev_notify,
6625 	.priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6626 };
6627 
6628 void __init ip6_route_init_special_entries(void)
6629 {
6630 	/* Registering of the loopback is done before this portion of code,
6631 	 * the loopback reference in rt6_info will not be taken, do it
6632 	 * manually for init_net */
6633 	init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6634 	init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6635 	init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6636   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6637 	init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6638 	init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6639 	init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6640 	init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6641   #endif
6642 }
6643 
6644 #if IS_BUILTIN(CONFIG_IPV6)
6645 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6646 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6647 
6648 BTF_ID_LIST(btf_fib6_info_id)
6649 BTF_ID(struct, fib6_info)
6650 
6651 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6652 	.seq_ops		= &ipv6_route_seq_ops,
6653 	.init_seq_private	= bpf_iter_init_seq_net,
6654 	.fini_seq_private	= bpf_iter_fini_seq_net,
6655 	.seq_priv_size		= sizeof(struct ipv6_route_iter),
6656 };
6657 
6658 static struct bpf_iter_reg ipv6_route_reg_info = {
6659 	.target			= "ipv6_route",
6660 	.ctx_arg_info_size	= 1,
6661 	.ctx_arg_info		= {
6662 		{ offsetof(struct bpf_iter__ipv6_route, rt),
6663 		  PTR_TO_BTF_ID_OR_NULL },
6664 	},
6665 	.seq_info		= &ipv6_route_seq_info,
6666 };
6667 
6668 static int __init bpf_iter_register(void)
6669 {
6670 	ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6671 	return bpf_iter_reg_target(&ipv6_route_reg_info);
6672 }
6673 
6674 static void bpf_iter_unregister(void)
6675 {
6676 	bpf_iter_unreg_target(&ipv6_route_reg_info);
6677 }
6678 #endif
6679 #endif
6680 
6681 int __init ip6_route_init(void)
6682 {
6683 	int ret;
6684 	int cpu;
6685 
6686 	ret = -ENOMEM;
6687 	ip6_dst_ops_template.kmem_cachep =
6688 		kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6689 				  SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
6690 	if (!ip6_dst_ops_template.kmem_cachep)
6691 		goto out;
6692 
6693 	ret = dst_entries_init(&ip6_dst_blackhole_ops);
6694 	if (ret)
6695 		goto out_kmem_cache;
6696 
6697 	ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6698 	if (ret)
6699 		goto out_dst_entries;
6700 
6701 	ret = register_pernet_subsys(&ip6_route_net_ops);
6702 	if (ret)
6703 		goto out_register_inetpeer;
6704 
6705 	ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6706 
6707 	ret = fib6_init();
6708 	if (ret)
6709 		goto out_register_subsys;
6710 
6711 	ret = xfrm6_init();
6712 	if (ret)
6713 		goto out_fib6_init;
6714 
6715 	ret = fib6_rules_init();
6716 	if (ret)
6717 		goto xfrm6_init;
6718 
6719 	ret = register_pernet_subsys(&ip6_route_net_late_ops);
6720 	if (ret)
6721 		goto fib6_rules_init;
6722 
6723 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6724 				   inet6_rtm_newroute, NULL, 0);
6725 	if (ret < 0)
6726 		goto out_register_late_subsys;
6727 
6728 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6729 				   inet6_rtm_delroute, NULL, 0);
6730 	if (ret < 0)
6731 		goto out_register_late_subsys;
6732 
6733 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6734 				   inet6_rtm_getroute, NULL,
6735 				   RTNL_FLAG_DOIT_UNLOCKED);
6736 	if (ret < 0)
6737 		goto out_register_late_subsys;
6738 
6739 	ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6740 	if (ret)
6741 		goto out_register_late_subsys;
6742 
6743 #if IS_BUILTIN(CONFIG_IPV6)
6744 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6745 	ret = bpf_iter_register();
6746 	if (ret)
6747 		goto out_register_late_subsys;
6748 #endif
6749 #endif
6750 
6751 	for_each_possible_cpu(cpu) {
6752 		struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6753 
6754 		INIT_LIST_HEAD(&ul->head);
6755 		INIT_LIST_HEAD(&ul->quarantine);
6756 		spin_lock_init(&ul->lock);
6757 	}
6758 
6759 out:
6760 	return ret;
6761 
6762 out_register_late_subsys:
6763 	rtnl_unregister_all(PF_INET6);
6764 	unregister_pernet_subsys(&ip6_route_net_late_ops);
6765 fib6_rules_init:
6766 	fib6_rules_cleanup();
6767 xfrm6_init:
6768 	xfrm6_fini();
6769 out_fib6_init:
6770 	fib6_gc_cleanup();
6771 out_register_subsys:
6772 	unregister_pernet_subsys(&ip6_route_net_ops);
6773 out_register_inetpeer:
6774 	unregister_pernet_subsys(&ipv6_inetpeer_ops);
6775 out_dst_entries:
6776 	dst_entries_destroy(&ip6_dst_blackhole_ops);
6777 out_kmem_cache:
6778 	kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6779 	goto out;
6780 }
6781 
6782 void ip6_route_cleanup(void)
6783 {
6784 #if IS_BUILTIN(CONFIG_IPV6)
6785 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6786 	bpf_iter_unregister();
6787 #endif
6788 #endif
6789 	unregister_netdevice_notifier(&ip6_route_dev_notifier);
6790 	unregister_pernet_subsys(&ip6_route_net_late_ops);
6791 	fib6_rules_cleanup();
6792 	xfrm6_fini();
6793 	fib6_gc_cleanup();
6794 	unregister_pernet_subsys(&ipv6_inetpeer_ops);
6795 	unregister_pernet_subsys(&ip6_route_net_ops);
6796 	dst_entries_destroy(&ip6_dst_blackhole_ops);
6797 	kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6798 }
6799