xref: /linux/net/ipv4/route.c (revision b056e73267244a126545c52f6c36c62893bb643c)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		ROUTE - implementation of the IP router.
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
12  *		Linus Torvalds, <Linus.Torvalds@helsinki.fi>
13  *		Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
14  *
15  * Fixes:
16  *		Alan Cox	:	Verify area fixes.
17  *		Alan Cox	:	cli() protects routing changes
18  *		Rui Oliveira	:	ICMP routing table updates
19  *		(rco@di.uminho.pt)	Routing table insertion and update
20  *		Linus Torvalds	:	Rewrote bits to be sensible
21  *		Alan Cox	:	Added BSD route gw semantics
22  *		Alan Cox	:	Super /proc >4K
23  *		Alan Cox	:	MTU in route table
24  *		Alan Cox	:	MSS actually. Also added the window
25  *					clamper.
26  *		Sam Lantinga	:	Fixed route matching in rt_del()
27  *		Alan Cox	:	Routing cache support.
28  *		Alan Cox	:	Removed compatibility cruft.
29  *		Alan Cox	:	RTF_REJECT support.
30  *		Alan Cox	:	TCP irtt support.
31  *		Jonathan Naylor	:	Added Metric support.
32  *	Miquel van Smoorenburg	:	BSD API fixes.
33  *	Miquel van Smoorenburg	:	Metrics.
34  *		Alan Cox	:	Use __u32 properly
35  *		Alan Cox	:	Aligned routing errors more closely with BSD
36  *					our system is still very different.
37  *		Alan Cox	:	Faster /proc handling
38  *	Alexey Kuznetsov	:	Massive rework to support tree based routing,
39  *					routing caches and better behaviour.
40  *
41  *		Olaf Erb	:	irtt wasn't being copied right.
42  *		Bjorn Ekwall	:	Kerneld route support.
43  *		Alan Cox	:	Multicast fixed (I hope)
44  *		Pavel Krauz	:	Limited broadcast fixed
45  *		Mike McLagan	:	Routing by source
46  *	Alexey Kuznetsov	:	End of old history. Split to fib.c and
47  *					route.c and rewritten from scratch.
48  *		Andi Kleen	:	Load-limit warning messages.
49  *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
50  *	Vitaly E. Lavrov	:	Race condition in ip_route_input_slow.
51  *	Tobias Ringstrom	:	Uninitialized res.type in ip_route_output_slow.
52  *	Vladimir V. Ivanov	:	IP rule info (flowid) is really useful.
53  *		Marc Boucher	:	routing by fwmark
54  *	Robert Olsson		:	Added rt_cache statistics
55  *	Arnaldo C. Melo		:	Convert proc stuff to seq_file
56  *	Eric Dumazet		:	hashed spinlocks and rt_check_expire() fixes.
57  *	Ilia Sotnikov		:	Ignore TOS on PMTUD and Redirect
58  *	Ilia Sotnikov		:	Removed TOS from hash calculations
59  */
60 
61 #define pr_fmt(fmt) "IPv4: " fmt
62 
63 #include <linux/module.h>
64 #include <linux/bitops.h>
65 #include <linux/kernel.h>
66 #include <linux/mm.h>
67 #include <linux/memblock.h>
68 #include <linux/socket.h>
69 #include <linux/errno.h>
70 #include <linux/in.h>
71 #include <linux/inet.h>
72 #include <linux/netdevice.h>
73 #include <linux/proc_fs.h>
74 #include <linux/init.h>
75 #include <linux/skbuff.h>
76 #include <linux/inetdevice.h>
77 #include <linux/igmp.h>
78 #include <linux/pkt_sched.h>
79 #include <linux/mroute.h>
80 #include <linux/netfilter_ipv4.h>
81 #include <linux/random.h>
82 #include <linux/rcupdate.h>
83 #include <linux/slab.h>
84 #include <linux/jhash.h>
85 #include <net/dst.h>
86 #include <net/dst_metadata.h>
87 #include <net/flow.h>
88 #include <net/inet_dscp.h>
89 #include <net/net_namespace.h>
90 #include <net/ip.h>
91 #include <net/route.h>
92 #include <net/inetpeer.h>
93 #include <net/sock.h>
94 #include <net/ip_fib.h>
95 #include <net/nexthop.h>
96 #include <net/tcp.h>
97 #include <net/icmp.h>
98 #include <net/xfrm.h>
99 #include <net/lwtunnel.h>
100 #include <net/netevent.h>
101 #include <net/rtnetlink.h>
102 #ifdef CONFIG_SYSCTL
103 #include <linux/sysctl.h>
104 #endif
105 #include <net/secure_seq.h>
106 #include <net/ip_tunnels.h>
107 
108 #include "fib_lookup.h"
109 
110 #define RT_GC_TIMEOUT (300*HZ)
111 
112 #define DEFAULT_MIN_PMTU (512 + 20 + 20)
113 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
114 #define DEFAULT_MIN_ADVMSS 256
115 static int ip_rt_max_size;
116 static int ip_rt_redirect_number __read_mostly	= 9;
117 static int ip_rt_redirect_load __read_mostly	= HZ / 50;
118 static int ip_rt_redirect_silence __read_mostly	= ((HZ / 50) << (9 + 1));
119 static int ip_rt_error_cost __read_mostly	= HZ;
120 static int ip_rt_error_burst __read_mostly	= 5 * HZ;
121 
122 static int ip_rt_gc_timeout __read_mostly	= RT_GC_TIMEOUT;
123 
124 /*
125  *	Interface to generic destination cache.
126  */
127 
128 INDIRECT_CALLABLE_SCOPE
129 struct dst_entry	*ipv4_dst_check(struct dst_entry *dst, u32 cookie);
130 static unsigned int	 ipv4_default_advmss(const struct dst_entry *dst);
131 INDIRECT_CALLABLE_SCOPE
132 unsigned int		ipv4_mtu(const struct dst_entry *dst);
133 static void		ipv4_negative_advice(struct sock *sk,
134 					     struct dst_entry *dst);
135 static void		 ipv4_link_failure(struct sk_buff *skb);
136 static void		 ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
137 					   struct sk_buff *skb, u32 mtu,
138 					   bool confirm_neigh);
139 static void		 ip_do_redirect(struct dst_entry *dst, struct sock *sk,
140 					struct sk_buff *skb);
141 static void		ipv4_dst_destroy(struct dst_entry *dst);
142 
143 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
144 {
145 	WARN_ON(1);
146 	return NULL;
147 }
148 
149 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
150 					   struct sk_buff *skb,
151 					   const void *daddr);
152 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
153 
154 static struct dst_ops ipv4_dst_ops = {
155 	.family =		AF_INET,
156 	.check =		ipv4_dst_check,
157 	.default_advmss =	ipv4_default_advmss,
158 	.mtu =			ipv4_mtu,
159 	.cow_metrics =		ipv4_cow_metrics,
160 	.destroy =		ipv4_dst_destroy,
161 	.negative_advice =	ipv4_negative_advice,
162 	.link_failure =		ipv4_link_failure,
163 	.update_pmtu =		ip_rt_update_pmtu,
164 	.redirect =		ip_do_redirect,
165 	.local_out =		__ip_local_out,
166 	.neigh_lookup =		ipv4_neigh_lookup,
167 	.confirm_neigh =	ipv4_confirm_neigh,
168 };
169 
170 #define ECN_OR_COST(class)	TC_PRIO_##class
171 
172 const __u8 ip_tos2prio[16] = {
173 	TC_PRIO_BESTEFFORT,
174 	ECN_OR_COST(BESTEFFORT),
175 	TC_PRIO_BESTEFFORT,
176 	ECN_OR_COST(BESTEFFORT),
177 	TC_PRIO_BULK,
178 	ECN_OR_COST(BULK),
179 	TC_PRIO_BULK,
180 	ECN_OR_COST(BULK),
181 	TC_PRIO_INTERACTIVE,
182 	ECN_OR_COST(INTERACTIVE),
183 	TC_PRIO_INTERACTIVE,
184 	ECN_OR_COST(INTERACTIVE),
185 	TC_PRIO_INTERACTIVE_BULK,
186 	ECN_OR_COST(INTERACTIVE_BULK),
187 	TC_PRIO_INTERACTIVE_BULK,
188 	ECN_OR_COST(INTERACTIVE_BULK)
189 };
190 EXPORT_SYMBOL(ip_tos2prio);
191 
192 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
193 #ifndef CONFIG_PREEMPT_RT
194 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
195 #else
196 #define RT_CACHE_STAT_INC(field) this_cpu_inc(rt_cache_stat.field)
197 #endif
198 
199 #ifdef CONFIG_PROC_FS
200 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
201 {
202 	if (*pos)
203 		return NULL;
204 	return SEQ_START_TOKEN;
205 }
206 
207 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
208 {
209 	++*pos;
210 	return NULL;
211 }
212 
213 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
214 {
215 }
216 
217 static int rt_cache_seq_show(struct seq_file *seq, void *v)
218 {
219 	if (v == SEQ_START_TOKEN)
220 		seq_printf(seq, "%-127s\n",
221 			   "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
222 			   "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
223 			   "HHUptod\tSpecDst");
224 	return 0;
225 }
226 
227 static const struct seq_operations rt_cache_seq_ops = {
228 	.start  = rt_cache_seq_start,
229 	.next   = rt_cache_seq_next,
230 	.stop   = rt_cache_seq_stop,
231 	.show   = rt_cache_seq_show,
232 };
233 
234 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
235 {
236 	int cpu;
237 
238 	if (*pos == 0)
239 		return SEQ_START_TOKEN;
240 
241 	for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
242 		if (!cpu_possible(cpu))
243 			continue;
244 		*pos = cpu+1;
245 		return &per_cpu(rt_cache_stat, cpu);
246 	}
247 	return NULL;
248 }
249 
250 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
251 {
252 	int cpu;
253 
254 	for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
255 		if (!cpu_possible(cpu))
256 			continue;
257 		*pos = cpu+1;
258 		return &per_cpu(rt_cache_stat, cpu);
259 	}
260 	(*pos)++;
261 	return NULL;
262 
263 }
264 
265 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
266 {
267 
268 }
269 
270 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
271 {
272 	struct rt_cache_stat *st = v;
273 
274 	if (v == SEQ_START_TOKEN) {
275 		seq_puts(seq, "entries  in_hit   in_slow_tot in_slow_mc in_no_route in_brd   in_martian_dst in_martian_src out_hit  out_slow_tot out_slow_mc gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
276 		return 0;
277 	}
278 
279 	seq_printf(seq, "%08x %08x %08x    %08x   %08x    %08x %08x       "
280 			"%08x       %08x %08x     %08x    %08x %08x   "
281 			"%08x     %08x        %08x        %08x\n",
282 		   dst_entries_get_slow(&ipv4_dst_ops),
283 		   0, /* st->in_hit */
284 		   st->in_slow_tot,
285 		   st->in_slow_mc,
286 		   st->in_no_route,
287 		   st->in_brd,
288 		   st->in_martian_dst,
289 		   st->in_martian_src,
290 
291 		   0, /* st->out_hit */
292 		   st->out_slow_tot,
293 		   st->out_slow_mc,
294 
295 		   0, /* st->gc_total */
296 		   0, /* st->gc_ignored */
297 		   0, /* st->gc_goal_miss */
298 		   0, /* st->gc_dst_overflow */
299 		   0, /* st->in_hlist_search */
300 		   0  /* st->out_hlist_search */
301 		);
302 	return 0;
303 }
304 
305 static const struct seq_operations rt_cpu_seq_ops = {
306 	.start  = rt_cpu_seq_start,
307 	.next   = rt_cpu_seq_next,
308 	.stop   = rt_cpu_seq_stop,
309 	.show   = rt_cpu_seq_show,
310 };
311 
312 #ifdef CONFIG_IP_ROUTE_CLASSID
313 static int rt_acct_proc_show(struct seq_file *m, void *v)
314 {
315 	struct ip_rt_acct *dst, *src;
316 	unsigned int i, j;
317 
318 	dst = kzalloc_objs(struct ip_rt_acct, 256);
319 	if (!dst)
320 		return -ENOMEM;
321 
322 	for_each_possible_cpu(i) {
323 		src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
324 		for (j = 0; j < 256; j++) {
325 			dst[j].o_bytes   += src[j].o_bytes;
326 			dst[j].o_packets += src[j].o_packets;
327 			dst[j].i_bytes   += src[j].i_bytes;
328 			dst[j].i_packets += src[j].i_packets;
329 		}
330 	}
331 
332 	seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
333 	kfree(dst);
334 	return 0;
335 }
336 #endif
337 
338 static int __net_init ip_rt_do_proc_init(struct net *net)
339 {
340 	struct proc_dir_entry *pde;
341 
342 	pde = proc_create_seq("rt_cache", 0444, net->proc_net,
343 			      &rt_cache_seq_ops);
344 	if (!pde)
345 		goto err1;
346 
347 	pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
348 			      &rt_cpu_seq_ops);
349 	if (!pde)
350 		goto err2;
351 
352 #ifdef CONFIG_IP_ROUTE_CLASSID
353 	pde = proc_create_single("rt_acct", 0, net->proc_net,
354 			rt_acct_proc_show);
355 	if (!pde)
356 		goto err3;
357 #endif
358 	return 0;
359 
360 #ifdef CONFIG_IP_ROUTE_CLASSID
361 err3:
362 	remove_proc_entry("rt_cache", net->proc_net_stat);
363 #endif
364 err2:
365 	remove_proc_entry("rt_cache", net->proc_net);
366 err1:
367 	return -ENOMEM;
368 }
369 
370 static void __net_exit ip_rt_do_proc_exit(struct net *net)
371 {
372 	remove_proc_entry("rt_cache", net->proc_net_stat);
373 	remove_proc_entry("rt_cache", net->proc_net);
374 #ifdef CONFIG_IP_ROUTE_CLASSID
375 	remove_proc_entry("rt_acct", net->proc_net);
376 #endif
377 }
378 
379 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
380 	.init = ip_rt_do_proc_init,
381 	.exit = ip_rt_do_proc_exit,
382 };
383 
384 static int __init ip_rt_proc_init(void)
385 {
386 	return register_pernet_subsys(&ip_rt_proc_ops);
387 }
388 
389 #else
390 static inline int ip_rt_proc_init(void)
391 {
392 	return 0;
393 }
394 #endif /* CONFIG_PROC_FS */
395 
396 static inline bool rt_is_expired(const struct rtable *rth)
397 {
398 	bool res;
399 
400 	rcu_read_lock();
401 	res = rth->rt_genid != rt_genid_ipv4(dev_net_rcu(rth->dst.dev));
402 	rcu_read_unlock();
403 
404 	return res;
405 }
406 
407 void rt_cache_flush(struct net *net)
408 {
409 	rt_genid_bump_ipv4(net);
410 }
411 
412 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
413 					   struct sk_buff *skb,
414 					   const void *daddr)
415 {
416 	const struct rtable *rt = container_of(dst, struct rtable, dst);
417 	struct net_device *dev;
418 	struct neighbour *n;
419 
420 	rcu_read_lock();
421 	dev = dst_dev_rcu(dst);
422 	if (likely(rt->rt_gw_family == AF_INET)) {
423 		n = ip_neigh_gw4(dev, rt->rt_gw4);
424 	} else if (rt->rt_gw_family == AF_INET6) {
425 		n = ip_neigh_gw6(dev, &rt->rt_gw6);
426         } else {
427 		__be32 pkey;
428 
429 		pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
430 		n = ip_neigh_gw4(dev, pkey);
431 	}
432 
433 	if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
434 		n = NULL;
435 
436 	rcu_read_unlock();
437 
438 	return n;
439 }
440 
441 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
442 {
443 	const struct rtable *rt = container_of(dst, struct rtable, dst);
444 	struct net_device *dev = dst_dev(dst);
445 	const __be32 *pkey = daddr;
446 
447 	if (rt->rt_gw_family == AF_INET) {
448 		pkey = (const __be32 *)&rt->rt_gw4;
449 	} else if (IS_ENABLED(CONFIG_IPV6) && rt->rt_gw_family == AF_INET6) {
450 		return __ipv6_confirm_neigh(dev, &rt->rt_gw6);
451 	} else if (!daddr ||
452 		 (rt->rt_flags &
453 		  (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
454 		return;
455 	}
456 	__ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
457 }
458 
459 /* Hash tables of size 2048..262144 depending on RAM size.
460  * Each bucket uses 8 bytes.
461  */
462 static u32 ip_idents_mask __read_mostly;
463 static atomic_t *ip_idents __read_mostly;
464 static u32 *ip_tstamps __read_mostly;
465 
466 /* In order to protect privacy, we add a perturbation to identifiers
467  * if one generator is seldom used. This makes hard for an attacker
468  * to infer how many packets were sent between two points in time.
469  */
470 static u32 ip_idents_reserve(u32 hash, int segs)
471 {
472 	u32 bucket, old, now = (u32)jiffies;
473 	atomic_t *p_id;
474 	u32 *p_tstamp;
475 	u32 delta = 0;
476 
477 	bucket = hash & ip_idents_mask;
478 	p_tstamp = ip_tstamps + bucket;
479 	p_id = ip_idents + bucket;
480 	old = READ_ONCE(*p_tstamp);
481 
482 	if (old != now && cmpxchg(p_tstamp, old, now) == old)
483 		delta = get_random_u32_below(now - old);
484 
485 	/* If UBSAN reports an error there, please make sure your compiler
486 	 * supports -fno-strict-overflow before reporting it that was a bug
487 	 * in UBSAN, and it has been fixed in GCC-8.
488 	 */
489 	return atomic_add_return(segs + delta, p_id) - segs;
490 }
491 
492 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
493 {
494 	u32 hash, id;
495 
496 	/* Note the following code is not safe, but this is okay. */
497 	if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
498 		get_random_bytes(&net->ipv4.ip_id_key,
499 				 sizeof(net->ipv4.ip_id_key));
500 
501 	hash = siphash_3u32((__force u32)iph->daddr,
502 			    (__force u32)iph->saddr,
503 			    iph->protocol,
504 			    &net->ipv4.ip_id_key);
505 	id = ip_idents_reserve(hash, segs);
506 	iph->id = htons(id);
507 }
508 EXPORT_SYMBOL(__ip_select_ident);
509 
510 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
511 			     const struct sock *sk, const struct iphdr *iph,
512 			     int oif, __u8 tos, u8 prot, u32 mark,
513 			     int flow_flags)
514 {
515 	__u8 scope = RT_SCOPE_UNIVERSE;
516 
517 	if (sk) {
518 		oif = sk->sk_bound_dev_if;
519 		mark = READ_ONCE(sk->sk_mark);
520 		tos = ip_sock_rt_tos(sk);
521 		scope = ip_sock_rt_scope(sk);
522 		prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW :
523 						    sk->sk_protocol;
524 	}
525 
526 	flowi4_init_output(fl4, oif, mark, tos & INET_DSCP_MASK, scope,
527 			   prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
528 			   sock_net_uid(net, sk));
529 }
530 
531 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
532 			       const struct sock *sk)
533 {
534 	const struct net *net = dev_net(skb->dev);
535 	const struct iphdr *iph = ip_hdr(skb);
536 	int oif = skb->dev->ifindex;
537 	u8 prot = iph->protocol;
538 	u32 mark = skb->mark;
539 	__u8 tos = iph->tos;
540 
541 	__build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
542 }
543 
544 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
545 {
546 	const struct inet_sock *inet = inet_sk(sk);
547 	const struct ip_options_rcu *inet_opt;
548 	__be32 daddr = inet->inet_daddr;
549 
550 	rcu_read_lock();
551 	inet_opt = rcu_dereference(inet->inet_opt);
552 	if (inet_opt && inet_opt->opt.srr)
553 		daddr = inet_opt->opt.faddr;
554 	flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark),
555 			   ip_sock_rt_tos(sk),
556 			   ip_sock_rt_scope(sk),
557 			   inet_test_bit(HDRINCL, sk) ?
558 				IPPROTO_RAW : sk->sk_protocol,
559 			   inet_sk_flowi_flags(sk),
560 			   daddr, inet->inet_saddr, 0, 0,
561 			   sk_uid(sk));
562 	rcu_read_unlock();
563 }
564 
565 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
566 				 const struct sk_buff *skb)
567 {
568 	if (skb)
569 		build_skb_flow_key(fl4, skb, sk);
570 	else
571 		build_sk_flow_key(fl4, sk);
572 }
573 
574 static DEFINE_SPINLOCK(fnhe_lock);
575 
576 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
577 {
578 	struct rtable *rt;
579 
580 	rt = rcu_dereference(fnhe->fnhe_rth_input);
581 	if (rt) {
582 		RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
583 		dst_dev_put(&rt->dst);
584 		dst_release(&rt->dst);
585 	}
586 	rt = rcu_dereference(fnhe->fnhe_rth_output);
587 	if (rt) {
588 		RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
589 		dst_dev_put(&rt->dst);
590 		dst_release(&rt->dst);
591 	}
592 }
593 
594 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
595 {
596 	struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
597 	struct fib_nh_exception *fnhe, *oldest = NULL;
598 
599 	for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
600 		fnhe = rcu_dereference_protected(*fnhe_p,
601 						 lockdep_is_held(&fnhe_lock));
602 		if (!fnhe)
603 			break;
604 		if (!oldest ||
605 		    time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
606 			oldest = fnhe;
607 			oldest_p = fnhe_p;
608 		}
609 	}
610 
611 	/* Clear oldest->fnhe_daddr to prevent this fnhe from being
612 	 * rebound with new dsts in rt_bind_exception().
613 	 */
614 	oldest->fnhe_daddr = 0;
615 	fnhe_flush_routes(oldest);
616 	*oldest_p = oldest->fnhe_next;
617 	kfree_rcu(oldest, rcu);
618 }
619 
620 static u32 fnhe_hashfun(__be32 daddr)
621 {
622 	static siphash_aligned_key_t fnhe_hash_key;
623 	u64 hval;
624 
625 	net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
626 	hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
627 	return hash_64(hval, FNHE_HASH_SHIFT);
628 }
629 
630 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
631 {
632 	rt->rt_pmtu = fnhe->fnhe_pmtu;
633 	rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
634 	rt->dst.expires = fnhe->fnhe_expires;
635 
636 	if (fnhe->fnhe_gw) {
637 		rt->rt_flags |= RTCF_REDIRECTED;
638 		rt->rt_uses_gateway = 1;
639 		rt->rt_gw_family = AF_INET;
640 		rt->rt_gw4 = fnhe->fnhe_gw;
641 	}
642 }
643 
644 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
645 				  __be32 gw, u32 pmtu, bool lock,
646 				  unsigned long expires)
647 {
648 	struct fnhe_hash_bucket *hash;
649 	struct fib_nh_exception *fnhe;
650 	struct rtable *rt;
651 	u32 genid, hval;
652 	unsigned int i;
653 	int depth;
654 
655 	genid = fnhe_genid(dev_net(nhc->nhc_dev));
656 	hval = fnhe_hashfun(daddr);
657 
658 	spin_lock_bh(&fnhe_lock);
659 
660 	hash = rcu_dereference(nhc->nhc_exceptions);
661 	if (!hash) {
662 		hash = kzalloc_objs(*hash, FNHE_HASH_SIZE, GFP_ATOMIC);
663 		if (!hash)
664 			goto out_unlock;
665 		rcu_assign_pointer(nhc->nhc_exceptions, hash);
666 	}
667 
668 	hash += hval;
669 
670 	depth = 0;
671 	for (fnhe = rcu_dereference(hash->chain); fnhe;
672 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
673 		if (fnhe->fnhe_daddr == daddr)
674 			break;
675 		depth++;
676 	}
677 
678 	if (fnhe) {
679 		if (fnhe->fnhe_genid != genid)
680 			fnhe->fnhe_genid = genid;
681 		if (gw)
682 			fnhe->fnhe_gw = gw;
683 		if (pmtu) {
684 			fnhe->fnhe_pmtu = pmtu;
685 			fnhe->fnhe_mtu_locked = lock;
686 		}
687 		fnhe->fnhe_expires = max(1UL, expires);
688 		/* Update all cached dsts too */
689 		rt = rcu_dereference(fnhe->fnhe_rth_input);
690 		if (rt)
691 			fill_route_from_fnhe(rt, fnhe);
692 		rt = rcu_dereference(fnhe->fnhe_rth_output);
693 		if (rt)
694 			fill_route_from_fnhe(rt, fnhe);
695 	} else {
696 		/* Randomize max depth to avoid some side channels attacks. */
697 		int max_depth = FNHE_RECLAIM_DEPTH +
698 				get_random_u32_below(FNHE_RECLAIM_DEPTH);
699 
700 		while (depth > max_depth) {
701 			fnhe_remove_oldest(hash);
702 			depth--;
703 		}
704 
705 		fnhe = kzalloc_obj(*fnhe, GFP_ATOMIC);
706 		if (!fnhe)
707 			goto out_unlock;
708 
709 		fnhe->fnhe_next = hash->chain;
710 
711 		fnhe->fnhe_genid = genid;
712 		fnhe->fnhe_daddr = daddr;
713 		fnhe->fnhe_gw = gw;
714 		fnhe->fnhe_pmtu = pmtu;
715 		fnhe->fnhe_mtu_locked = lock;
716 		fnhe->fnhe_expires = max(1UL, expires);
717 
718 		rcu_assign_pointer(hash->chain, fnhe);
719 
720 		/* Exception created; mark the cached routes for the nexthop
721 		 * stale, so anyone caching it rechecks if this exception
722 		 * applies to them.
723 		 */
724 		rt = rcu_dereference(nhc->nhc_rth_input);
725 		if (rt)
726 			WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL);
727 
728 		for_each_possible_cpu(i) {
729 			struct rtable __rcu **prt;
730 
731 			prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
732 			rt = rcu_dereference(*prt);
733 			if (rt)
734 				WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL);
735 		}
736 	}
737 
738 	fnhe->fnhe_stamp = jiffies;
739 
740 out_unlock:
741 	spin_unlock_bh(&fnhe_lock);
742 }
743 
744 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
745 			     bool kill_route)
746 {
747 	__be32 new_gw = icmp_hdr(skb)->un.gateway;
748 	__be32 old_gw = ip_hdr(skb)->saddr;
749 	struct net_device *dev = skb->dev;
750 	struct in_device *in_dev;
751 	struct fib_result res;
752 	struct neighbour *n;
753 	struct net *net;
754 
755 	switch (icmp_hdr(skb)->code & 7) {
756 	case ICMP_REDIR_NET:
757 	case ICMP_REDIR_NETTOS:
758 	case ICMP_REDIR_HOST:
759 	case ICMP_REDIR_HOSTTOS:
760 		break;
761 
762 	default:
763 		return;
764 	}
765 
766 	if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
767 		return;
768 
769 	in_dev = __in_dev_get_rcu(dev);
770 	if (!in_dev)
771 		return;
772 
773 	net = dev_net(dev);
774 	if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
775 	    ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
776 	    ipv4_is_zeronet(new_gw))
777 		goto reject_redirect;
778 
779 	if (!IN_DEV_SHARED_MEDIA(in_dev)) {
780 		if (!inet_addr_onlink(in_dev, new_gw, old_gw))
781 			goto reject_redirect;
782 		if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
783 			goto reject_redirect;
784 	} else {
785 		if (inet_addr_type(net, new_gw) != RTN_UNICAST)
786 			goto reject_redirect;
787 	}
788 
789 	n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw);
790 	if (!n)
791 		n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
792 	if (!IS_ERR(n)) {
793 		if (!(READ_ONCE(n->nud_state) & NUD_VALID)) {
794 			neigh_event_send(n, NULL);
795 		} else {
796 			if (fib_lookup(net, fl4, &res, 0) == 0) {
797 				struct fib_nh_common *nhc;
798 
799 				fib_select_path(net, &res, fl4, skb);
800 				nhc = FIB_RES_NHC(res);
801 				update_or_create_fnhe(nhc, fl4->daddr, new_gw,
802 						0, false,
803 						jiffies + ip_rt_gc_timeout);
804 			}
805 			if (kill_route)
806 				WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL);
807 			call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
808 		}
809 		neigh_release(n);
810 	}
811 	return;
812 
813 reject_redirect:
814 #ifdef CONFIG_IP_ROUTE_VERBOSE
815 	if (IN_DEV_LOG_MARTIANS(in_dev)) {
816 		const struct iphdr *iph = (const struct iphdr *) skb->data;
817 		__be32 daddr = iph->daddr;
818 		__be32 saddr = iph->saddr;
819 
820 		net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
821 				     "  Advised path = %pI4 -> %pI4\n",
822 				     &old_gw, dev->name, &new_gw,
823 				     &saddr, &daddr);
824 	}
825 #endif
826 	;
827 }
828 
829 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
830 {
831 	struct rtable *rt;
832 	struct flowi4 fl4;
833 	const struct iphdr *iph = (const struct iphdr *) skb->data;
834 	struct net *net = dev_net(skb->dev);
835 	int oif = skb->dev->ifindex;
836 	u8 prot = iph->protocol;
837 	u32 mark = skb->mark;
838 	__u8 tos = iph->tos;
839 
840 	rt = dst_rtable(dst);
841 
842 	__build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
843 	__ip_do_redirect(rt, skb, &fl4, true);
844 }
845 
846 static void ipv4_negative_advice(struct sock *sk,
847 				 struct dst_entry *dst)
848 {
849 	struct rtable *rt = dst_rtable(dst);
850 
851 	if ((READ_ONCE(dst->obsolete) > 0) ||
852 	    (rt->rt_flags & RTCF_REDIRECTED) ||
853 	    READ_ONCE(rt->dst.expires))
854 		sk_dst_reset(sk);
855 }
856 
857 /*
858  * Algorithm:
859  *	1. The first ip_rt_redirect_number redirects are sent
860  *	   with exponential backoff, then we stop sending them at all,
861  *	   assuming that the host ignores our redirects.
862  *	2. If we did not see packets requiring redirects
863  *	   during ip_rt_redirect_silence, we assume that the host
864  *	   forgot redirected route and start to send redirects again.
865  *
866  * This algorithm is much cheaper and more intelligent than dumb load limiting
867  * in icmp.c.
868  *
869  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
870  * and "frag. need" (breaks PMTU discovery) in icmp.c.
871  */
872 
873 void ip_rt_send_redirect(struct sk_buff *skb)
874 {
875 	struct rtable *rt = skb_rtable(skb);
876 	struct in_device *in_dev;
877 	struct inet_peer *peer;
878 	struct net *net;
879 	int log_martians;
880 	int vif;
881 
882 	rcu_read_lock();
883 	in_dev = __in_dev_get_rcu(rt->dst.dev);
884 	if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
885 		rcu_read_unlock();
886 		return;
887 	}
888 	log_martians = IN_DEV_LOG_MARTIANS(in_dev);
889 	vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
890 
891 	net = dev_net(rt->dst.dev);
892 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif);
893 	if (!peer) {
894 		rcu_read_unlock();
895 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
896 			  rt_nexthop(rt, ip_hdr(skb)->daddr));
897 		return;
898 	}
899 
900 	/* No redirected packets during ip_rt_redirect_silence;
901 	 * reset the algorithm.
902 	 */
903 	if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
904 		peer->rate_tokens = 0;
905 		peer->n_redirects = 0;
906 	}
907 
908 	/* Too many ignored redirects; do not send anything
909 	 * set dst.rate_last to the last seen redirected packet.
910 	 */
911 	if (peer->n_redirects >= ip_rt_redirect_number) {
912 		peer->rate_last = jiffies;
913 		goto out_unlock;
914 	}
915 
916 	/* Check for load limit; set rate_last to the latest sent
917 	 * redirect.
918 	 */
919 	if (peer->n_redirects == 0 ||
920 	    time_after(jiffies,
921 		       (peer->rate_last +
922 			(ip_rt_redirect_load << peer->n_redirects)))) {
923 		__be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
924 
925 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
926 		peer->rate_last = jiffies;
927 		++peer->n_redirects;
928 		if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians &&
929 		    peer->n_redirects == ip_rt_redirect_number)
930 			net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
931 					     &ip_hdr(skb)->saddr, inet_iif(skb),
932 					     &ip_hdr(skb)->daddr, &gw);
933 	}
934 out_unlock:
935 	rcu_read_unlock();
936 }
937 
938 static int ip_error(struct sk_buff *skb)
939 {
940 	struct rtable *rt = skb_rtable(skb);
941 	struct net_device *dev = skb->dev;
942 	struct in_device *in_dev;
943 	struct inet_peer *peer;
944 	unsigned long now;
945 	struct net *net;
946 	SKB_DR(reason);
947 	bool send;
948 	int code;
949 
950 	if (netif_is_l3_master(skb->dev)) {
951 		dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
952 		if (!dev)
953 			goto out;
954 	}
955 
956 	in_dev = __in_dev_get_rcu(dev);
957 
958 	/* IP on this device is disabled. */
959 	if (!in_dev)
960 		goto out;
961 
962 	net = dev_net(rt->dst.dev);
963 	if (!IN_DEV_FORWARD(in_dev)) {
964 		switch (rt->dst.error) {
965 		case EHOSTUNREACH:
966 			SKB_DR_SET(reason, IP_INADDRERRORS);
967 			__IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
968 			break;
969 
970 		case ENETUNREACH:
971 			SKB_DR_SET(reason, IP_INNOROUTES);
972 			__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
973 			break;
974 		}
975 		goto out;
976 	}
977 
978 	switch (rt->dst.error) {
979 	case EINVAL:
980 	default:
981 		goto out;
982 	case EHOSTUNREACH:
983 		code = ICMP_HOST_UNREACH;
984 		break;
985 	case ENETUNREACH:
986 		code = ICMP_NET_UNREACH;
987 		SKB_DR_SET(reason, IP_INNOROUTES);
988 		__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
989 		break;
990 	case EACCES:
991 		code = ICMP_PKT_FILTERED;
992 		break;
993 	}
994 
995 	rcu_read_lock();
996 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
997 			       l3mdev_master_ifindex_rcu(skb->dev));
998 	send = true;
999 	if (peer) {
1000 		now = jiffies;
1001 		peer->rate_tokens += now - peer->rate_last;
1002 		if (peer->rate_tokens > ip_rt_error_burst)
1003 			peer->rate_tokens = ip_rt_error_burst;
1004 		peer->rate_last = now;
1005 		if (peer->rate_tokens >= ip_rt_error_cost)
1006 			peer->rate_tokens -= ip_rt_error_cost;
1007 		else
1008 			send = false;
1009 	}
1010 	rcu_read_unlock();
1011 
1012 	if (send)
1013 		icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1014 
1015 out:	kfree_skb_reason(skb, reason);
1016 	return 0;
1017 }
1018 
1019 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1020 {
1021 	struct dst_entry *dst = &rt->dst;
1022 	struct fib_result res;
1023 	bool lock = false;
1024 	struct net *net;
1025 	u32 old_mtu;
1026 
1027 	if (ip_mtu_locked(dst))
1028 		return;
1029 
1030 	old_mtu = ipv4_mtu(dst);
1031 	if (old_mtu < mtu)
1032 		return;
1033 
1034 	rcu_read_lock();
1035 	net = dst_dev_net_rcu(dst);
1036 	if (mtu < net->ipv4.ip_rt_min_pmtu) {
1037 		lock = true;
1038 		mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1039 	}
1040 
1041 	if (rt->rt_pmtu == mtu && !lock &&
1042 	    time_before(jiffies, READ_ONCE(dst->expires) -
1043 				 net->ipv4.ip_rt_mtu_expires / 2))
1044 		goto out;
1045 
1046 	if (fib_lookup(net, fl4, &res, 0) == 0) {
1047 		struct fib_nh_common *nhc;
1048 
1049 		fib_select_path(net, &res, fl4, NULL);
1050 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1051 		if (fib_info_num_path(res.fi) > 1) {
1052 			int nhsel;
1053 
1054 			for (nhsel = 0; nhsel < fib_info_num_path(res.fi); nhsel++) {
1055 				nhc = fib_info_nhc(res.fi, nhsel);
1056 				update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1057 						      jiffies + net->ipv4.ip_rt_mtu_expires);
1058 			}
1059 			goto out;
1060 		}
1061 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1062 		nhc = FIB_RES_NHC(res);
1063 		update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1064 				      jiffies + net->ipv4.ip_rt_mtu_expires);
1065 	}
1066 out:
1067 	rcu_read_unlock();
1068 }
1069 
1070 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1071 			      struct sk_buff *skb, u32 mtu,
1072 			      bool confirm_neigh)
1073 {
1074 	struct rtable *rt = dst_rtable(dst);
1075 	struct flowi4 fl4;
1076 
1077 	ip_rt_build_flow_key(&fl4, sk, skb);
1078 
1079 	/* Don't make lookup fail for bridged encapsulations */
1080 	if (skb && netif_is_any_bridge_port(skb->dev))
1081 		fl4.flowi4_oif = 0;
1082 
1083 	__ip_rt_update_pmtu(rt, &fl4, mtu);
1084 }
1085 
1086 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1087 		      int oif, u8 protocol)
1088 {
1089 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1090 	struct flowi4 fl4;
1091 	struct rtable *rt;
1092 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
1093 
1094 	__build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1095 			 0);
1096 	rt = __ip_route_output_key(net, &fl4);
1097 	if (!IS_ERR(rt)) {
1098 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1099 		ip_rt_put(rt);
1100 	}
1101 }
1102 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1103 
1104 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1105 {
1106 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1107 	struct flowi4 fl4;
1108 	struct rtable *rt;
1109 
1110 	__build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1111 
1112 	if (!fl4.flowi4_mark)
1113 		fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1114 
1115 	rt = __ip_route_output_key(sock_net(sk), &fl4);
1116 	if (!IS_ERR(rt)) {
1117 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1118 		ip_rt_put(rt);
1119 	}
1120 }
1121 
1122 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1123 {
1124 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1125 	struct flowi4 fl4;
1126 	struct rtable *rt;
1127 	struct dst_entry *odst = NULL;
1128 	bool new = false;
1129 	struct net *net = sock_net(sk);
1130 
1131 	bh_lock_sock(sk);
1132 
1133 	if (!ip_sk_accept_pmtu(sk))
1134 		goto out;
1135 
1136 	odst = sk_dst_get(sk);
1137 
1138 	if (sock_owned_by_user(sk) || !odst) {
1139 		__ipv4_sk_update_pmtu(skb, sk, mtu);
1140 		goto out;
1141 	}
1142 
1143 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1144 
1145 	rt = dst_rtable(odst);
1146 	if (READ_ONCE(odst->obsolete) && !odst->ops->check(odst, 0)) {
1147 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1148 		if (IS_ERR(rt))
1149 			goto out;
1150 
1151 		new = true;
1152 	}
1153 
1154 	__ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu);
1155 
1156 	if (!dst_check(&rt->dst, 0)) {
1157 		if (new)
1158 			dst_release(&rt->dst);
1159 
1160 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1161 		if (IS_ERR(rt))
1162 			goto out;
1163 
1164 		new = true;
1165 	}
1166 
1167 	if (new)
1168 		sk_dst_set(sk, &rt->dst);
1169 
1170 out:
1171 	bh_unlock_sock(sk);
1172 	dst_release(odst);
1173 }
1174 
1175 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1176 		   int oif, u8 protocol)
1177 {
1178 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1179 	struct flowi4 fl4;
1180 	struct rtable *rt;
1181 
1182 	__build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1183 	rt = __ip_route_output_key(net, &fl4);
1184 	if (!IS_ERR(rt)) {
1185 		__ip_do_redirect(rt, skb, &fl4, false);
1186 		ip_rt_put(rt);
1187 	}
1188 }
1189 EXPORT_SYMBOL_GPL(ipv4_redirect);
1190 
1191 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1192 {
1193 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1194 	struct flowi4 fl4;
1195 	struct rtable *rt;
1196 	struct net *net = sock_net(sk);
1197 
1198 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1199 	rt = __ip_route_output_key(net, &fl4);
1200 	if (!IS_ERR(rt)) {
1201 		__ip_do_redirect(rt, skb, &fl4, false);
1202 		ip_rt_put(rt);
1203 	}
1204 }
1205 
1206 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1207 							 u32 cookie)
1208 {
1209 	struct rtable *rt = dst_rtable(dst);
1210 
1211 	/* All IPV4 dsts are created with ->obsolete set to the value
1212 	 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1213 	 * into this function always.
1214 	 *
1215 	 * When a PMTU/redirect information update invalidates a route,
1216 	 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1217 	 * DST_OBSOLETE_DEAD.
1218 	 */
1219 	if (READ_ONCE(dst->obsolete) != DST_OBSOLETE_FORCE_CHK ||
1220 	    rt_is_expired(rt))
1221 		return NULL;
1222 	return dst;
1223 }
1224 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1225 
1226 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1227 {
1228 	struct inet_skb_parm parm;
1229 	struct net_device *dev;
1230 	int res;
1231 
1232 	/* Recompile ip options since IPCB may not be valid anymore.
1233 	 * Also check we have a reasonable ipv4 header.
1234 	 */
1235 	if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1236 	    ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1237 		return;
1238 
1239 	memset(&parm, 0, sizeof(parm));
1240 	if (ip_hdr(skb)->ihl > 5) {
1241 		if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1242 			return;
1243 		parm.opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1244 
1245 		rcu_read_lock();
1246 		dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1247 		res = __ip_options_compile(dev_net(dev), &parm.opt, skb, NULL);
1248 		rcu_read_unlock();
1249 
1250 		if (res)
1251 			return;
1252 	}
1253 	__icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &parm);
1254 }
1255 
1256 static void ipv4_link_failure(struct sk_buff *skb)
1257 {
1258 	struct rtable *rt;
1259 
1260 	ipv4_send_dest_unreach(skb);
1261 
1262 	rt = skb_rtable(skb);
1263 	if (rt)
1264 		dst_set_expires(&rt->dst, 0);
1265 }
1266 
1267 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1268 {
1269 	pr_debug("%s: %pI4 -> %pI4, %s\n",
1270 		 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1271 		 skb->dev ? skb->dev->name : "?");
1272 	kfree_skb(skb);
1273 	WARN_ON_ONCE(1);
1274 	return 0;
1275 }
1276 
1277 /*
1278  * We do not cache source address of outgoing interface,
1279  * because it is used only by IP RR, TS and SRR options,
1280  * so that it out of fast path.
1281  *
1282  * BTW remember: "addr" is allowed to be not aligned
1283  * in IP options!
1284  */
1285 
1286 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1287 {
1288 	__be32 src;
1289 
1290 	if (rt_is_output_route(rt))
1291 		src = ip_hdr(skb)->saddr;
1292 	else {
1293 		struct fib_result res;
1294 		struct iphdr *iph = ip_hdr(skb);
1295 		struct flowi4 fl4 = {
1296 			.daddr = iph->daddr,
1297 			.saddr = iph->saddr,
1298 			.flowi4_dscp = ip4h_dscp(iph),
1299 			.flowi4_oif = rt->dst.dev->ifindex,
1300 			.flowi4_iif = skb->dev->ifindex,
1301 			.flowi4_mark = skb->mark,
1302 		};
1303 
1304 		rcu_read_lock();
1305 		if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1306 			src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1307 		else
1308 			src = inet_select_addr(rt->dst.dev,
1309 					       rt_nexthop(rt, iph->daddr),
1310 					       RT_SCOPE_UNIVERSE);
1311 		rcu_read_unlock();
1312 	}
1313 	memcpy(addr, &src, 4);
1314 }
1315 
1316 #ifdef CONFIG_IP_ROUTE_CLASSID
1317 static void set_class_tag(struct rtable *rt, u32 tag)
1318 {
1319 	if (!(rt->dst.tclassid & 0xFFFF))
1320 		rt->dst.tclassid |= tag & 0xFFFF;
1321 	if (!(rt->dst.tclassid & 0xFFFF0000))
1322 		rt->dst.tclassid |= tag & 0xFFFF0000;
1323 }
1324 #endif
1325 
1326 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1327 {
1328 	unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1329 	unsigned int advmss;
1330 	struct net *net;
1331 
1332 	rcu_read_lock();
1333 	net = dst_dev_net_rcu(dst);
1334 	advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1335 				   net->ipv4.ip_rt_min_advmss);
1336 	rcu_read_unlock();
1337 
1338 	return min(advmss, IPV4_MAX_PMTU - header_size);
1339 }
1340 
1341 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1342 {
1343 	return ip_dst_mtu_maybe_forward(dst, false);
1344 }
1345 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1346 
1347 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1348 {
1349 	struct fnhe_hash_bucket *hash;
1350 	struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1351 	u32 hval = fnhe_hashfun(daddr);
1352 
1353 	spin_lock_bh(&fnhe_lock);
1354 
1355 	hash = rcu_dereference_protected(nhc->nhc_exceptions,
1356 					 lockdep_is_held(&fnhe_lock));
1357 	hash += hval;
1358 
1359 	fnhe_p = &hash->chain;
1360 	fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1361 	while (fnhe) {
1362 		if (fnhe->fnhe_daddr == daddr) {
1363 			rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1364 				fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1365 			/* set fnhe_daddr to 0 to ensure it won't bind with
1366 			 * new dsts in rt_bind_exception().
1367 			 */
1368 			fnhe->fnhe_daddr = 0;
1369 			fnhe_flush_routes(fnhe);
1370 			kfree_rcu(fnhe, rcu);
1371 			break;
1372 		}
1373 		fnhe_p = &fnhe->fnhe_next;
1374 		fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1375 						 lockdep_is_held(&fnhe_lock));
1376 	}
1377 
1378 	spin_unlock_bh(&fnhe_lock);
1379 }
1380 
1381 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1382 					       __be32 daddr)
1383 {
1384 	struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1385 	struct fib_nh_exception *fnhe;
1386 	u32 hval;
1387 
1388 	if (!hash)
1389 		return NULL;
1390 
1391 	hval = fnhe_hashfun(daddr);
1392 
1393 	for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1394 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
1395 		if (fnhe->fnhe_daddr == daddr) {
1396 			if (fnhe->fnhe_expires &&
1397 			    time_after(jiffies, fnhe->fnhe_expires)) {
1398 				ip_del_fnhe(nhc, daddr);
1399 				break;
1400 			}
1401 			return fnhe;
1402 		}
1403 	}
1404 	return NULL;
1405 }
1406 
1407 /* MTU selection:
1408  * 1. mtu on route is locked - use it
1409  * 2. mtu from nexthop exception
1410  * 3. mtu from egress device
1411  */
1412 
1413 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1414 {
1415 	struct fib_nh_common *nhc = res->nhc;
1416 	struct net_device *dev = nhc->nhc_dev;
1417 	struct fib_info *fi = res->fi;
1418 	u32 mtu = 0;
1419 
1420 	if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1421 	    fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1422 		mtu = fi->fib_mtu;
1423 
1424 	if (likely(!mtu)) {
1425 		struct fib_nh_exception *fnhe;
1426 
1427 		fnhe = find_exception(nhc, daddr);
1428 		if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1429 			mtu = fnhe->fnhe_pmtu;
1430 	}
1431 
1432 	if (likely(!mtu))
1433 		mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1434 
1435 	return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1436 }
1437 
1438 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1439 			      __be32 daddr, const bool do_cache)
1440 {
1441 	bool ret = false;
1442 
1443 	spin_lock_bh(&fnhe_lock);
1444 
1445 	if (daddr == fnhe->fnhe_daddr) {
1446 		struct rtable __rcu **porig;
1447 		struct rtable *orig;
1448 		int genid = fnhe_genid(dev_net(rt->dst.dev));
1449 
1450 		if (rt_is_input_route(rt))
1451 			porig = &fnhe->fnhe_rth_input;
1452 		else
1453 			porig = &fnhe->fnhe_rth_output;
1454 		orig = rcu_dereference(*porig);
1455 
1456 		if (fnhe->fnhe_genid != genid) {
1457 			fnhe->fnhe_genid = genid;
1458 			fnhe->fnhe_gw = 0;
1459 			fnhe->fnhe_pmtu = 0;
1460 			fnhe->fnhe_expires = 0;
1461 			fnhe->fnhe_mtu_locked = false;
1462 			fnhe_flush_routes(fnhe);
1463 			orig = NULL;
1464 		}
1465 		fill_route_from_fnhe(rt, fnhe);
1466 		if (!rt->rt_gw4) {
1467 			rt->rt_gw4 = daddr;
1468 			rt->rt_gw_family = AF_INET;
1469 		}
1470 
1471 		if (do_cache) {
1472 			dst_hold(&rt->dst);
1473 			rcu_assign_pointer(*porig, rt);
1474 			if (orig) {
1475 				dst_dev_put(&orig->dst);
1476 				dst_release(&orig->dst);
1477 			}
1478 			ret = true;
1479 		}
1480 
1481 		fnhe->fnhe_stamp = jiffies;
1482 	}
1483 	spin_unlock_bh(&fnhe_lock);
1484 
1485 	return ret;
1486 }
1487 
1488 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1489 {
1490 	struct rtable *orig, *prev, **p;
1491 	bool ret = true;
1492 
1493 	if (rt_is_input_route(rt)) {
1494 		p = (struct rtable **)&nhc->nhc_rth_input;
1495 	} else {
1496 		p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1497 	}
1498 	orig = *p;
1499 
1500 	/* hold dst before doing cmpxchg() to avoid race condition
1501 	 * on this dst
1502 	 */
1503 	dst_hold(&rt->dst);
1504 	prev = cmpxchg(p, orig, rt);
1505 	if (prev == orig) {
1506 		if (orig) {
1507 			rt_add_uncached_list(orig);
1508 			dst_release(&orig->dst);
1509 		}
1510 	} else {
1511 		dst_release(&rt->dst);
1512 		ret = false;
1513 	}
1514 
1515 	return ret;
1516 }
1517 
1518 struct uncached_list {
1519 	spinlock_t		lock;
1520 	struct list_head	head;
1521 };
1522 
1523 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1524 
1525 void rt_add_uncached_list(struct rtable *rt)
1526 {
1527 	struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1528 
1529 	rt->dst.rt_uncached_list = ul;
1530 
1531 	spin_lock_bh(&ul->lock);
1532 	list_add_tail(&rt->dst.rt_uncached, &ul->head);
1533 	spin_unlock_bh(&ul->lock);
1534 }
1535 
1536 void rt_del_uncached_list(struct rtable *rt)
1537 {
1538 	struct uncached_list *ul = rt->dst.rt_uncached_list;
1539 
1540 	if (ul) {
1541 		spin_lock_bh(&ul->lock);
1542 		list_del_init(&rt->dst.rt_uncached);
1543 		spin_unlock_bh(&ul->lock);
1544 	}
1545 }
1546 
1547 static void ipv4_dst_destroy(struct dst_entry *dst)
1548 {
1549 	ip_dst_metrics_put(dst);
1550 	rt_del_uncached_list(dst_rtable(dst));
1551 }
1552 
1553 void rt_flush_dev(struct net_device *dev)
1554 {
1555 	struct rtable *rt, *safe;
1556 	int cpu;
1557 
1558 	for_each_possible_cpu(cpu) {
1559 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1560 
1561 		if (list_empty(&ul->head))
1562 			continue;
1563 
1564 		spin_lock_bh(&ul->lock);
1565 		list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1566 			if (rt->dst.dev != dev)
1567 				continue;
1568 			rt->dst.dev = blackhole_netdev;
1569 			netdev_ref_replace(dev, blackhole_netdev,
1570 					   &rt->dst.dev_tracker, GFP_ATOMIC);
1571 			list_del_init(&rt->dst.rt_uncached);
1572 		}
1573 		spin_unlock_bh(&ul->lock);
1574 	}
1575 }
1576 
1577 static bool rt_cache_valid(const struct rtable *rt)
1578 {
1579 	return	rt &&
1580 		READ_ONCE(rt->dst.obsolete) == DST_OBSOLETE_FORCE_CHK &&
1581 		!rt_is_expired(rt);
1582 }
1583 
1584 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1585 			   const struct fib_result *res,
1586 			   struct fib_nh_exception *fnhe,
1587 			   struct fib_info *fi, u16 type, u32 itag,
1588 			   const bool do_cache)
1589 {
1590 	bool cached = false;
1591 
1592 	if (fi) {
1593 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1594 
1595 		if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1596 			rt->rt_uses_gateway = 1;
1597 			rt->rt_gw_family = nhc->nhc_gw_family;
1598 			/* only INET and INET6 are supported */
1599 			if (likely(nhc->nhc_gw_family == AF_INET))
1600 				rt->rt_gw4 = nhc->nhc_gw.ipv4;
1601 			else
1602 				rt->rt_gw6 = nhc->nhc_gw.ipv6;
1603 		}
1604 
1605 		ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1606 
1607 #ifdef CONFIG_IP_ROUTE_CLASSID
1608 		if (nhc->nhc_family == AF_INET) {
1609 			struct fib_nh *nh;
1610 
1611 			nh = container_of(nhc, struct fib_nh, nh_common);
1612 			rt->dst.tclassid = nh->nh_tclassid;
1613 		}
1614 #endif
1615 		rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1616 		if (unlikely(fnhe))
1617 			cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1618 		else if (do_cache)
1619 			cached = rt_cache_route(nhc, rt);
1620 		if (unlikely(!cached)) {
1621 			/* Routes we intend to cache in nexthop exception or
1622 			 * FIB nexthop have the DST_NOCACHE bit clear.
1623 			 * However, if we are unsuccessful at storing this
1624 			 * route into the cache we really need to set it.
1625 			 */
1626 			if (!rt->rt_gw4) {
1627 				rt->rt_gw_family = AF_INET;
1628 				rt->rt_gw4 = daddr;
1629 			}
1630 			rt_add_uncached_list(rt);
1631 		}
1632 	} else
1633 		rt_add_uncached_list(rt);
1634 
1635 #ifdef CONFIG_IP_ROUTE_CLASSID
1636 #ifdef CONFIG_IP_MULTIPLE_TABLES
1637 	set_class_tag(rt, res->tclassid);
1638 #endif
1639 	set_class_tag(rt, itag);
1640 #endif
1641 }
1642 
1643 struct rtable *rt_dst_alloc(struct net_device *dev,
1644 			    unsigned int flags, u16 type,
1645 			    bool noxfrm)
1646 {
1647 	struct rtable *rt;
1648 
1649 	rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1650 		       (noxfrm ? DST_NOXFRM : 0));
1651 
1652 	if (rt) {
1653 		rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1654 		rt->rt_flags = flags;
1655 		rt->rt_type = type;
1656 		rt->rt_is_input = 0;
1657 		rt->rt_iif = 0;
1658 		rt->rt_pmtu = 0;
1659 		rt->rt_mtu_locked = 0;
1660 		rt->rt_uses_gateway = 0;
1661 		rt->rt_gw_family = 0;
1662 		rt->rt_gw4 = 0;
1663 
1664 		rt->dst.output = ip_output;
1665 		if (flags & RTCF_LOCAL)
1666 			rt->dst.input = ip_local_deliver;
1667 	}
1668 
1669 	return rt;
1670 }
1671 EXPORT_SYMBOL(rt_dst_alloc);
1672 
1673 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1674 {
1675 	struct rtable *new_rt;
1676 
1677 	new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1678 			   rt->dst.flags);
1679 
1680 	if (new_rt) {
1681 		new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1682 		new_rt->rt_flags = rt->rt_flags;
1683 		new_rt->rt_type = rt->rt_type;
1684 		new_rt->rt_is_input = rt->rt_is_input;
1685 		new_rt->rt_iif = rt->rt_iif;
1686 		new_rt->rt_pmtu = rt->rt_pmtu;
1687 		new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1688 		new_rt->rt_gw_family = rt->rt_gw_family;
1689 		if (rt->rt_gw_family == AF_INET)
1690 			new_rt->rt_gw4 = rt->rt_gw4;
1691 		else if (rt->rt_gw_family == AF_INET6)
1692 			new_rt->rt_gw6 = rt->rt_gw6;
1693 
1694 		new_rt->dst.input = READ_ONCE(rt->dst.input);
1695 		new_rt->dst.output = READ_ONCE(rt->dst.output);
1696 		new_rt->dst.error = rt->dst.error;
1697 		new_rt->dst.lastuse = jiffies;
1698 		new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1699 	}
1700 	return new_rt;
1701 }
1702 
1703 /* called in rcu_read_lock() section */
1704 enum skb_drop_reason
1705 ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1706 		      dscp_t dscp, struct net_device *dev,
1707 		      struct in_device *in_dev, u32 *itag)
1708 {
1709 	enum skb_drop_reason reason;
1710 
1711 	/* Primary sanity checks. */
1712 	if (!in_dev)
1713 		return SKB_DROP_REASON_NOT_SPECIFIED;
1714 
1715 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1716 		return SKB_DROP_REASON_IP_INVALID_SOURCE;
1717 
1718 	if (skb->protocol != htons(ETH_P_IP))
1719 		return SKB_DROP_REASON_INVALID_PROTO;
1720 
1721 	if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1722 		return SKB_DROP_REASON_IP_LOCALNET;
1723 
1724 	if (ipv4_is_zeronet(saddr)) {
1725 		if (!ipv4_is_local_multicast(daddr) &&
1726 		    ip_hdr(skb)->protocol != IPPROTO_IGMP)
1727 			return SKB_DROP_REASON_IP_INVALID_SOURCE;
1728 	} else {
1729 		reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
1730 						    dev, in_dev, itag);
1731 		if (reason)
1732 			return reason;
1733 	}
1734 	return SKB_NOT_DROPPED_YET;
1735 }
1736 
1737 /* called in rcu_read_lock() section */
1738 static enum skb_drop_reason
1739 ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1740 		  dscp_t dscp, struct net_device *dev, int our)
1741 {
1742 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1743 	unsigned int flags = RTCF_MULTICAST;
1744 	enum skb_drop_reason reason;
1745 	struct rtable *rth;
1746 	u32 itag = 0;
1747 
1748 	reason = ip_mc_validate_source(skb, daddr, saddr, dscp, dev, in_dev,
1749 				       &itag);
1750 	if (reason)
1751 		return reason;
1752 
1753 	if (our)
1754 		flags |= RTCF_LOCAL;
1755 
1756 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1757 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
1758 
1759 	rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1760 			   false);
1761 	if (!rth)
1762 		return SKB_DROP_REASON_NOMEM;
1763 
1764 #ifdef CONFIG_IP_ROUTE_CLASSID
1765 	rth->dst.tclassid = itag;
1766 #endif
1767 	rth->dst.output = ip_rt_bug;
1768 	rth->rt_is_input= 1;
1769 
1770 #ifdef CONFIG_IP_MROUTE
1771 	if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1772 		rth->dst.input = ip_mr_input;
1773 #endif
1774 	RT_CACHE_STAT_INC(in_slow_mc);
1775 
1776 	skb_dst_drop(skb);
1777 	skb_dst_set(skb, &rth->dst);
1778 	return SKB_NOT_DROPPED_YET;
1779 }
1780 
1781 
1782 static void ip_handle_martian_source(struct net_device *dev,
1783 				     struct in_device *in_dev,
1784 				     struct sk_buff *skb,
1785 				     __be32 daddr,
1786 				     __be32 saddr)
1787 {
1788 	RT_CACHE_STAT_INC(in_martian_src);
1789 #ifdef CONFIG_IP_ROUTE_VERBOSE
1790 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1791 		/*
1792 		 *	RFC1812 recommendation, if source is martian,
1793 		 *	the only hint is MAC header.
1794 		 */
1795 		pr_warn("martian source (src=%pI4, dst=%pI4, dev=%s)\n",
1796 			&saddr, &daddr, dev->name);
1797 		if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1798 			print_hex_dump(KERN_WARNING, "ll header: ",
1799 				       DUMP_PREFIX_OFFSET, 16, 1,
1800 				       skb_mac_header(skb),
1801 				       dev->hard_header_len, false);
1802 		}
1803 	}
1804 #endif
1805 }
1806 
1807 /* called in rcu_read_lock() section */
1808 static enum skb_drop_reason
1809 __mkroute_input(struct sk_buff *skb, const struct fib_result *res,
1810 		struct in_device *in_dev, __be32 daddr,
1811 		__be32 saddr, dscp_t dscp)
1812 {
1813 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1814 	struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1815 	struct net_device *dev = nhc->nhc_dev;
1816 	struct fib_nh_exception *fnhe;
1817 	struct rtable *rth;
1818 	int err;
1819 	struct in_device *out_dev;
1820 	bool do_cache;
1821 	u32 itag = 0;
1822 
1823 	/* get a working reference to the output device */
1824 	out_dev = __in_dev_get_rcu(dev);
1825 	if (!out_dev) {
1826 		net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1827 		return reason;
1828 	}
1829 
1830 	err = fib_validate_source(skb, saddr, daddr, dscp, FIB_RES_OIF(*res),
1831 				  in_dev->dev, in_dev, &itag);
1832 	if (err < 0) {
1833 		reason = -err;
1834 		ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1835 					 saddr);
1836 
1837 		goto cleanup;
1838 	}
1839 
1840 	do_cache = res->fi && !itag;
1841 	if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1842 	    skb->protocol == htons(ETH_P_IP)) {
1843 		__be32 gw;
1844 
1845 		gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1846 		if (IN_DEV_SHARED_MEDIA(out_dev) ||
1847 		    inet_addr_onlink(out_dev, saddr, gw))
1848 			IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1849 	}
1850 
1851 	if (skb->protocol != htons(ETH_P_IP)) {
1852 		/* Not IP (i.e. ARP). Do not create route, if it is
1853 		 * invalid for proxy arp. DNAT routes are always valid.
1854 		 *
1855 		 * Proxy arp feature have been extended to allow, ARP
1856 		 * replies back to the same interface, to support
1857 		 * Private VLAN switch technologies. See arp.c.
1858 		 */
1859 		if (out_dev == in_dev &&
1860 		    IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1861 			reason = SKB_DROP_REASON_ARP_PVLAN_DISABLE;
1862 			goto cleanup;
1863 		}
1864 	}
1865 
1866 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1867 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
1868 
1869 	fnhe = find_exception(nhc, daddr);
1870 	if (do_cache) {
1871 		if (fnhe)
1872 			rth = rcu_dereference(fnhe->fnhe_rth_input);
1873 		else
1874 			rth = rcu_dereference(nhc->nhc_rth_input);
1875 		if (rt_cache_valid(rth)) {
1876 			skb_dst_set_noref(skb, &rth->dst);
1877 			goto out;
1878 		}
1879 	}
1880 
1881 	rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1882 			   IN_DEV_ORCONF(out_dev, NOXFRM));
1883 	if (!rth) {
1884 		reason = SKB_DROP_REASON_NOMEM;
1885 		goto cleanup;
1886 	}
1887 
1888 	rth->rt_is_input = 1;
1889 	RT_CACHE_STAT_INC(in_slow_tot);
1890 
1891 	rth->dst.input = ip_forward;
1892 
1893 	rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1894 		       do_cache);
1895 	lwtunnel_set_redirect(&rth->dst);
1896 	skb_dst_set(skb, &rth->dst);
1897 out:
1898 	reason = SKB_NOT_DROPPED_YET;
1899 cleanup:
1900 	return reason;
1901 }
1902 
1903 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1904 /* To make ICMP packets follow the right flow, the multipath hash is
1905  * calculated from the inner IP addresses.
1906  */
1907 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1908 				 struct flow_keys *hash_keys)
1909 {
1910 	const struct iphdr *outer_iph = ip_hdr(skb);
1911 	const struct iphdr *key_iph = outer_iph;
1912 	const struct iphdr *inner_iph;
1913 	const struct icmphdr *icmph;
1914 	struct iphdr _inner_iph;
1915 	struct icmphdr _icmph;
1916 
1917 	if (likely(outer_iph->protocol != IPPROTO_ICMP))
1918 		goto out;
1919 
1920 	if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1921 		goto out;
1922 
1923 	icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1924 				   &_icmph);
1925 	if (!icmph)
1926 		goto out;
1927 
1928 	if (!icmp_is_err(icmph->type))
1929 		goto out;
1930 
1931 	inner_iph = skb_header_pointer(skb,
1932 				       outer_iph->ihl * 4 + sizeof(_icmph),
1933 				       sizeof(_inner_iph), &_inner_iph);
1934 	if (!inner_iph)
1935 		goto out;
1936 
1937 	key_iph = inner_iph;
1938 out:
1939 	hash_keys->addrs.v4addrs.src = key_iph->saddr;
1940 	hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1941 }
1942 
1943 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1944 					   const struct sk_buff *skb,
1945 					   bool *p_has_inner)
1946 {
1947 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1948 	struct flow_keys keys, hash_keys;
1949 
1950 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1951 		return 0;
1952 
1953 	memset(&hash_keys, 0, sizeof(hash_keys));
1954 	skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1955 
1956 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1957 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1958 		hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1959 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1960 		hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1961 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1962 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
1963 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1964 		hash_keys.ports.src = keys.ports.src;
1965 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1966 		hash_keys.ports.dst = keys.ports.dst;
1967 
1968 	*p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1969 	return fib_multipath_hash_from_keys(net, &hash_keys);
1970 }
1971 
1972 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1973 					   const struct sk_buff *skb,
1974 					   bool has_inner)
1975 {
1976 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1977 	struct flow_keys keys, hash_keys;
1978 
1979 	/* We assume the packet carries an encapsulation, but if none was
1980 	 * encountered during dissection of the outer flow, then there is no
1981 	 * point in calling the flow dissector again.
1982 	 */
1983 	if (!has_inner)
1984 		return 0;
1985 
1986 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1987 		return 0;
1988 
1989 	memset(&hash_keys, 0, sizeof(hash_keys));
1990 	skb_flow_dissect_flow_keys(skb, &keys, 0);
1991 
1992 	if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1993 		return 0;
1994 
1995 	if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1996 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1997 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1998 			hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1999 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2000 			hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2001 	} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2002 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2003 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2004 			hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2005 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2006 			hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2007 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2008 			hash_keys.tags.flow_label = keys.tags.flow_label;
2009 	}
2010 
2011 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2012 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
2013 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2014 		hash_keys.ports.src = keys.ports.src;
2015 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2016 		hash_keys.ports.dst = keys.ports.dst;
2017 
2018 	return fib_multipath_hash_from_keys(net, &hash_keys);
2019 }
2020 
2021 static u32 fib_multipath_custom_hash_skb(const struct net *net,
2022 					 const struct sk_buff *skb)
2023 {
2024 	u32 mhash, mhash_inner;
2025 	bool has_inner = true;
2026 
2027 	mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2028 	mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2029 
2030 	return jhash_2words(mhash, mhash_inner, 0);
2031 }
2032 
2033 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2034 					 const struct flowi4 *fl4)
2035 {
2036 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2037 	struct flow_keys hash_keys;
2038 
2039 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2040 		return 0;
2041 
2042 	memset(&hash_keys, 0, sizeof(hash_keys));
2043 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2044 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2045 		hash_keys.addrs.v4addrs.src = fl4->saddr;
2046 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2047 		hash_keys.addrs.v4addrs.dst = fl4->daddr;
2048 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2049 		hash_keys.basic.ip_proto = fl4->flowi4_proto;
2050 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) {
2051 		if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT)
2052 			hash_keys.ports.src = (__force __be16)get_random_u16();
2053 		else
2054 			hash_keys.ports.src = fl4->fl4_sport;
2055 	}
2056 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2057 		hash_keys.ports.dst = fl4->fl4_dport;
2058 
2059 	return fib_multipath_hash_from_keys(net, &hash_keys);
2060 }
2061 
2062 /* if skb is set it will be used and fl4 can be NULL */
2063 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2064 		       const struct sk_buff *skb, struct flow_keys *flkeys)
2065 {
2066 	u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2067 	struct flow_keys hash_keys;
2068 	u32 mhash = 0;
2069 
2070 	switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2071 	case 0:
2072 		memset(&hash_keys, 0, sizeof(hash_keys));
2073 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2074 		if (skb) {
2075 			ip_multipath_l3_keys(skb, &hash_keys);
2076 		} else {
2077 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2078 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2079 		}
2080 		mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2081 		break;
2082 	case 1:
2083 		/* skb is currently provided only when forwarding */
2084 		if (skb) {
2085 			unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2086 			struct flow_keys keys;
2087 
2088 			/* short-circuit if we already have L4 hash present */
2089 			if (skb->l4_hash)
2090 				return skb_get_hash_raw(skb) >> 1;
2091 
2092 			memset(&hash_keys, 0, sizeof(hash_keys));
2093 
2094 			if (!flkeys) {
2095 				skb_flow_dissect_flow_keys(skb, &keys, flag);
2096 				flkeys = &keys;
2097 			}
2098 
2099 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2100 			hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2101 			hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2102 			hash_keys.ports.src = flkeys->ports.src;
2103 			hash_keys.ports.dst = flkeys->ports.dst;
2104 			hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2105 		} else {
2106 			memset(&hash_keys, 0, sizeof(hash_keys));
2107 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2108 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2109 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2110 			if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT)
2111 				hash_keys.ports.src = (__force __be16)get_random_u16();
2112 			else
2113 				hash_keys.ports.src = fl4->fl4_sport;
2114 			hash_keys.ports.dst = fl4->fl4_dport;
2115 			hash_keys.basic.ip_proto = fl4->flowi4_proto;
2116 		}
2117 		mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2118 		break;
2119 	case 2:
2120 		memset(&hash_keys, 0, sizeof(hash_keys));
2121 		/* skb is currently provided only when forwarding */
2122 		if (skb) {
2123 			struct flow_keys keys;
2124 
2125 			skb_flow_dissect_flow_keys(skb, &keys, 0);
2126 			/* Inner can be v4 or v6 */
2127 			if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2128 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2129 				hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2130 				hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2131 			} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2132 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2133 				hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2134 				hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2135 				hash_keys.tags.flow_label = keys.tags.flow_label;
2136 				hash_keys.basic.ip_proto = keys.basic.ip_proto;
2137 			} else {
2138 				/* Same as case 0 */
2139 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2140 				ip_multipath_l3_keys(skb, &hash_keys);
2141 			}
2142 		} else {
2143 			/* Same as case 0 */
2144 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2145 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2146 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2147 		}
2148 		mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2149 		break;
2150 	case 3:
2151 		if (skb)
2152 			mhash = fib_multipath_custom_hash_skb(net, skb);
2153 		else
2154 			mhash = fib_multipath_custom_hash_fl4(net, fl4);
2155 		break;
2156 	}
2157 
2158 	if (multipath_hash)
2159 		mhash = jhash_2words(mhash, multipath_hash, 0);
2160 
2161 	return mhash >> 1;
2162 }
2163 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2164 
2165 static enum skb_drop_reason
2166 ip_mkroute_input(struct sk_buff *skb, struct fib_result *res,
2167 		 struct in_device *in_dev, __be32 daddr,
2168 		 __be32 saddr, dscp_t dscp, struct flow_keys *hkeys)
2169 {
2170 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2171 	if (res->fi && fib_info_num_path(res->fi) > 1) {
2172 		int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2173 
2174 		fib_select_multipath(res, h, NULL);
2175 		IPCB(skb)->flags |= IPSKB_MULTIPATH;
2176 	}
2177 #endif
2178 
2179 	/* create a routing cache entry */
2180 	return __mkroute_input(skb, res, in_dev, daddr, saddr, dscp);
2181 }
2182 
2183 /* Implements all the saddr-related checks as ip_route_input_slow(),
2184  * assuming daddr is valid and the destination is not a local broadcast one.
2185  * Uses the provided hint instead of performing a route lookup.
2186  */
2187 enum skb_drop_reason
2188 ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2189 		  dscp_t dscp, struct net_device *dev,
2190 		  const struct sk_buff *hint)
2191 {
2192 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2193 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2194 	struct rtable *rt = skb_rtable(hint);
2195 	struct net *net = dev_net(dev);
2196 	u32 tag = 0;
2197 
2198 	if (!in_dev)
2199 		return reason;
2200 
2201 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2202 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2203 		goto martian_source;
2204 	}
2205 
2206 	if (ipv4_is_zeronet(saddr)) {
2207 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2208 		goto martian_source;
2209 	}
2210 
2211 	if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2212 		reason = SKB_DROP_REASON_IP_LOCALNET;
2213 		goto martian_source;
2214 	}
2215 
2216 	if (!(rt->rt_flags & RTCF_LOCAL))
2217 		goto skip_validate_source;
2218 
2219 	reason = fib_validate_source_reason(skb, saddr, daddr, dscp, 0, dev,
2220 					    in_dev, &tag);
2221 	if (reason)
2222 		goto martian_source;
2223 
2224 skip_validate_source:
2225 	skb_dst_copy(skb, hint);
2226 	return SKB_NOT_DROPPED_YET;
2227 
2228 martian_source:
2229 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2230 	return reason;
2231 }
2232 
2233 /* get device for dst_alloc with local routes */
2234 static struct net_device *ip_rt_get_dev(struct net *net,
2235 					const struct fib_result *res)
2236 {
2237 	struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2238 	struct net_device *dev = NULL;
2239 
2240 	if (nhc)
2241 		dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2242 
2243 	return dev ? : net->loopback_dev;
2244 }
2245 
2246 /*
2247  *	NOTE. We drop all the packets that has local source
2248  *	addresses, because every properly looped back packet
2249  *	must have correct destination already attached by output routine.
2250  *	Changes in the enforced policies must be applied also to
2251  *	ip_route_use_hint().
2252  *
2253  *	Such approach solves two big problems:
2254  *	1. Not simplex devices are handled properly.
2255  *	2. IP spoofing attempts are filtered with 100% of guarantee.
2256  *	called with rcu_read_lock()
2257  */
2258 
2259 static enum skb_drop_reason
2260 ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2261 		    dscp_t dscp, struct net_device *dev,
2262 		    struct fib_result *res)
2263 {
2264 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2265 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2266 	struct flow_keys *flkeys = NULL, _flkeys;
2267 	struct net    *net = dev_net(dev);
2268 	struct ip_tunnel_info *tun_info;
2269 	int		err = -EINVAL;
2270 	unsigned int	flags = 0;
2271 	u32		itag = 0;
2272 	struct rtable	*rth;
2273 	struct flowi4	fl4;
2274 	bool do_cache = true;
2275 
2276 	/* IP on this device is disabled. */
2277 
2278 	if (!in_dev)
2279 		goto out;
2280 
2281 	/* Check for the most weird martians, which can be not detected
2282 	 * by fib_lookup.
2283 	 */
2284 
2285 	tun_info = skb_tunnel_info(skb);
2286 	if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2287 		fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2288 	else
2289 		fl4.flowi4_tun_key.tun_id = 0;
2290 	skb_dst_drop(skb);
2291 
2292 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2293 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2294 		goto martian_source;
2295 	}
2296 
2297 	res->fi = NULL;
2298 	res->table = NULL;
2299 	if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2300 		goto brd_input;
2301 
2302 	/* Accept zero addresses only to limited broadcast;
2303 	 * I even do not know to fix it or not. Waiting for complains :-)
2304 	 */
2305 	if (ipv4_is_zeronet(saddr)) {
2306 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2307 		goto martian_source;
2308 	}
2309 
2310 	if (ipv4_is_zeronet(daddr)) {
2311 		reason = SKB_DROP_REASON_IP_INVALID_DEST;
2312 		goto martian_destination;
2313 	}
2314 
2315 	/* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2316 	 * and call it once if daddr or/and saddr are loopback addresses
2317 	 */
2318 	if (ipv4_is_loopback(daddr)) {
2319 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2320 			reason = SKB_DROP_REASON_IP_LOCALNET;
2321 			goto martian_destination;
2322 		}
2323 	} else if (ipv4_is_loopback(saddr)) {
2324 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2325 			reason = SKB_DROP_REASON_IP_LOCALNET;
2326 			goto martian_source;
2327 		}
2328 	}
2329 
2330 	/*
2331 	 *	Now we are ready to route packet.
2332 	 */
2333 	fl4.flowi4_l3mdev = 0;
2334 	fl4.flowi4_oif = 0;
2335 	fl4.flowi4_iif = dev->ifindex;
2336 	fl4.flowi4_mark = skb->mark;
2337 	fl4.flowi4_dscp = dscp;
2338 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2339 	fl4.flowi4_flags = 0;
2340 	fl4.daddr = daddr;
2341 	fl4.saddr = saddr;
2342 	fl4.flowi4_uid = sock_net_uid(net, NULL);
2343 	fl4.flowi4_multipath_hash = 0;
2344 
2345 	if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2346 		flkeys = &_flkeys;
2347 	} else {
2348 		fl4.flowi4_proto = 0;
2349 		fl4.fl4_sport = 0;
2350 		fl4.fl4_dport = 0;
2351 	}
2352 
2353 	err = fib_lookup(net, &fl4, res, 0);
2354 	if (err != 0) {
2355 		if (!IN_DEV_FORWARD(in_dev))
2356 			err = -EHOSTUNREACH;
2357 		goto no_route;
2358 	}
2359 
2360 	if (res->type == RTN_BROADCAST) {
2361 		if (IN_DEV_BFORWARD(in_dev))
2362 			goto make_route;
2363 		/* not do cache if bc_forwarding is enabled */
2364 		if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING))
2365 			do_cache = false;
2366 		goto brd_input;
2367 	}
2368 
2369 	err = -EINVAL;
2370 	if (res->type == RTN_LOCAL) {
2371 		reason = fib_validate_source_reason(skb, saddr, daddr, dscp,
2372 						    0, dev, in_dev, &itag);
2373 		if (reason)
2374 			goto martian_source;
2375 		goto local_input;
2376 	}
2377 
2378 	if (!IN_DEV_FORWARD(in_dev)) {
2379 		err = -EHOSTUNREACH;
2380 		goto no_route;
2381 	}
2382 	if (res->type != RTN_UNICAST) {
2383 		reason = SKB_DROP_REASON_IP_INVALID_DEST;
2384 		goto martian_destination;
2385 	}
2386 
2387 make_route:
2388 	reason = ip_mkroute_input(skb, res, in_dev, daddr, saddr, dscp,
2389 				  flkeys);
2390 
2391 out:
2392 	return reason;
2393 
2394 brd_input:
2395 	if (skb->protocol != htons(ETH_P_IP)) {
2396 		reason = SKB_DROP_REASON_INVALID_PROTO;
2397 		goto out;
2398 	}
2399 
2400 	if (!ipv4_is_zeronet(saddr)) {
2401 		reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
2402 						    dev, in_dev, &itag);
2403 		if (reason)
2404 			goto martian_source;
2405 	}
2406 	flags |= RTCF_BROADCAST;
2407 	res->type = RTN_BROADCAST;
2408 	RT_CACHE_STAT_INC(in_brd);
2409 
2410 local_input:
2411 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2412 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
2413 
2414 	do_cache &= res->fi && !itag;
2415 	if (do_cache) {
2416 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2417 
2418 		rth = rcu_dereference(nhc->nhc_rth_input);
2419 		if (rt_cache_valid(rth)) {
2420 			skb_dst_set_noref(skb, &rth->dst);
2421 			reason = SKB_NOT_DROPPED_YET;
2422 			goto out;
2423 		}
2424 	}
2425 
2426 	rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2427 			   flags | RTCF_LOCAL, res->type, false);
2428 	if (!rth)
2429 		goto e_nobufs;
2430 
2431 	rth->dst.output= ip_rt_bug;
2432 #ifdef CONFIG_IP_ROUTE_CLASSID
2433 	rth->dst.tclassid = itag;
2434 #endif
2435 	rth->rt_is_input = 1;
2436 
2437 	RT_CACHE_STAT_INC(in_slow_tot);
2438 	if (res->type == RTN_UNREACHABLE) {
2439 		rth->dst.input= ip_error;
2440 		rth->dst.error= -err;
2441 		rth->rt_flags	&= ~RTCF_LOCAL;
2442 	}
2443 
2444 	if (do_cache) {
2445 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2446 
2447 		rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2448 		if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2449 			WARN_ON(rth->dst.input == lwtunnel_input);
2450 			rth->dst.lwtstate->orig_input = rth->dst.input;
2451 			rth->dst.input = lwtunnel_input;
2452 		}
2453 
2454 		if (unlikely(!rt_cache_route(nhc, rth)))
2455 			rt_add_uncached_list(rth);
2456 	}
2457 	skb_dst_set(skb, &rth->dst);
2458 	reason = SKB_NOT_DROPPED_YET;
2459 	goto out;
2460 
2461 no_route:
2462 	RT_CACHE_STAT_INC(in_no_route);
2463 	res->type = RTN_UNREACHABLE;
2464 	res->fi = NULL;
2465 	res->table = NULL;
2466 	goto local_input;
2467 
2468 	/*
2469 	 *	Do not cache martian addresses: they should be logged (RFC1812)
2470 	 */
2471 martian_destination:
2472 	RT_CACHE_STAT_INC(in_martian_dst);
2473 #ifdef CONFIG_IP_ROUTE_VERBOSE
2474 	if (IN_DEV_LOG_MARTIANS(in_dev))
2475 		net_warn_ratelimited("martian destination (src=%pI4, dst=%pI4, dev=%s)\n",
2476 				     &saddr, &daddr, dev->name);
2477 #endif
2478 	goto out;
2479 
2480 e_nobufs:
2481 	reason = SKB_DROP_REASON_NOMEM;
2482 	goto out;
2483 
2484 martian_source:
2485 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2486 	goto out;
2487 }
2488 
2489 /* called with rcu_read_lock held */
2490 static enum skb_drop_reason
2491 ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2492 		   dscp_t dscp, struct net_device *dev,
2493 		   struct fib_result *res)
2494 {
2495 	/* Multicast recognition logic is moved from route cache to here.
2496 	 * The problem was that too many Ethernet cards have broken/missing
2497 	 * hardware multicast filters :-( As result the host on multicasting
2498 	 * network acquires a lot of useless route cache entries, sort of
2499 	 * SDR messages from all the world. Now we try to get rid of them.
2500 	 * Really, provided software IP multicast filter is organized
2501 	 * reasonably (at least, hashed), it does not result in a slowdown
2502 	 * comparing with route cache reject entries.
2503 	 * Note, that multicast routers are not affected, because
2504 	 * route cache entry is created eventually.
2505 	 */
2506 	if (ipv4_is_multicast(daddr)) {
2507 		enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2508 		struct in_device *in_dev = __in_dev_get_rcu(dev);
2509 		int our = 0;
2510 
2511 		if (!in_dev)
2512 			return reason;
2513 
2514 		our = ip_check_mc_rcu(in_dev, daddr, saddr,
2515 				      ip_hdr(skb)->protocol);
2516 
2517 		/* check l3 master if no match yet */
2518 		if (!our && netif_is_l3_slave(dev)) {
2519 			struct in_device *l3_in_dev;
2520 
2521 			l3_in_dev = __in_dev_get_rcu(skb->dev);
2522 			if (l3_in_dev)
2523 				our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2524 						      ip_hdr(skb)->protocol);
2525 		}
2526 
2527 		if (our
2528 #ifdef CONFIG_IP_MROUTE
2529 			||
2530 		    (!ipv4_is_local_multicast(daddr) &&
2531 		     IN_DEV_MFORWARD(in_dev))
2532 #endif
2533 		   ) {
2534 			reason = ip_route_input_mc(skb, daddr, saddr, dscp,
2535 						   dev, our);
2536 		}
2537 		return reason;
2538 	}
2539 
2540 	return ip_route_input_slow(skb, daddr, saddr, dscp, dev, res);
2541 }
2542 
2543 enum skb_drop_reason ip_route_input_noref(struct sk_buff *skb, __be32 daddr,
2544 					  __be32 saddr, dscp_t dscp,
2545 					  struct net_device *dev)
2546 {
2547 	enum skb_drop_reason reason;
2548 	struct fib_result res;
2549 
2550 	rcu_read_lock();
2551 	reason = ip_route_input_rcu(skb, daddr, saddr, dscp, dev, &res);
2552 	rcu_read_unlock();
2553 
2554 	return reason;
2555 }
2556 EXPORT_SYMBOL(ip_route_input_noref);
2557 
2558 /* called with rcu_read_lock() */
2559 static struct rtable *__mkroute_output(const struct fib_result *res,
2560 				       const struct flowi4 *fl4, int orig_oif,
2561 				       struct net_device *dev_out,
2562 				       unsigned int flags)
2563 {
2564 	struct fib_info *fi = res->fi;
2565 	struct fib_nh_exception *fnhe;
2566 	struct in_device *in_dev;
2567 	u16 type = res->type;
2568 	struct rtable *rth;
2569 	bool do_cache;
2570 
2571 	in_dev = __in_dev_get_rcu(dev_out);
2572 	if (!in_dev)
2573 		return ERR_PTR(-EINVAL);
2574 
2575 	if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2576 		if (ipv4_is_loopback(fl4->saddr) &&
2577 		    !(dev_out->flags & IFF_LOOPBACK) &&
2578 		    !netif_is_l3_master(dev_out))
2579 			return ERR_PTR(-EINVAL);
2580 
2581 	if (ipv4_is_lbcast(fl4->daddr)) {
2582 		type = RTN_BROADCAST;
2583 
2584 		/* reset fi to prevent gateway resolution */
2585 		fi = NULL;
2586 	} else if (ipv4_is_multicast(fl4->daddr)) {
2587 		type = RTN_MULTICAST;
2588 	} else if (ipv4_is_zeronet(fl4->daddr)) {
2589 		return ERR_PTR(-EINVAL);
2590 	}
2591 
2592 	if (dev_out->flags & IFF_LOOPBACK)
2593 		flags |= RTCF_LOCAL;
2594 
2595 	do_cache = true;
2596 	if (type == RTN_BROADCAST) {
2597 		flags |= RTCF_BROADCAST | RTCF_LOCAL;
2598 	} else if (type == RTN_MULTICAST) {
2599 		flags |= RTCF_MULTICAST | RTCF_LOCAL;
2600 		if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2601 				     fl4->flowi4_proto))
2602 			flags &= ~RTCF_LOCAL;
2603 		else
2604 			do_cache = false;
2605 		/* If multicast route do not exist use
2606 		 * default one, but do not gateway in this case.
2607 		 * Yes, it is hack.
2608 		 */
2609 		if (fi && res->prefixlen < 4)
2610 			fi = NULL;
2611 	} else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2612 		   (orig_oif != dev_out->ifindex)) {
2613 		/* For local routes that require a particular output interface
2614 		 * we do not want to cache the result.  Caching the result
2615 		 * causes incorrect behaviour when there are multiple source
2616 		 * addresses on the interface, the end result being that if the
2617 		 * intended recipient is waiting on that interface for the
2618 		 * packet he won't receive it because it will be delivered on
2619 		 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2620 		 * be set to the loopback interface as well.
2621 		 */
2622 		do_cache = false;
2623 	}
2624 
2625 	fnhe = NULL;
2626 	do_cache &= fi != NULL;
2627 	if (fi) {
2628 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2629 		struct rtable __rcu **prth;
2630 
2631 		fnhe = find_exception(nhc, fl4->daddr);
2632 		if (!do_cache)
2633 			goto add;
2634 		if (fnhe) {
2635 			prth = &fnhe->fnhe_rth_output;
2636 		} else {
2637 			if (unlikely(fl4->flowi4_flags &
2638 				     FLOWI_FLAG_KNOWN_NH &&
2639 				     !(nhc->nhc_gw_family &&
2640 				       nhc->nhc_scope == RT_SCOPE_LINK))) {
2641 				do_cache = false;
2642 				goto add;
2643 			}
2644 			prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2645 		}
2646 		rth = rcu_dereference(*prth);
2647 		if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2648 			return rth;
2649 	}
2650 
2651 add:
2652 	rth = rt_dst_alloc(dev_out, flags, type,
2653 			   IN_DEV_ORCONF(in_dev, NOXFRM));
2654 	if (!rth)
2655 		return ERR_PTR(-ENOBUFS);
2656 
2657 	rth->rt_iif = orig_oif;
2658 
2659 	RT_CACHE_STAT_INC(out_slow_tot);
2660 
2661 	if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2662 		if (flags & RTCF_LOCAL &&
2663 		    !(dev_out->flags & IFF_LOOPBACK)) {
2664 			rth->dst.output = ip_mc_output;
2665 			RT_CACHE_STAT_INC(out_slow_mc);
2666 		}
2667 #ifdef CONFIG_IP_MROUTE
2668 		if (type == RTN_MULTICAST) {
2669 			if (IN_DEV_MFORWARD(in_dev) &&
2670 			    !ipv4_is_local_multicast(fl4->daddr)) {
2671 				rth->dst.input = ip_mr_input;
2672 				rth->dst.output = ip_mr_output;
2673 			}
2674 		}
2675 #endif
2676 	}
2677 
2678 	rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2679 	lwtunnel_set_redirect(&rth->dst);
2680 
2681 	return rth;
2682 }
2683 
2684 /*
2685  * Major route resolver routine.
2686  */
2687 
2688 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2689 					const struct sk_buff *skb)
2690 {
2691 	struct fib_result res = {
2692 		.type		= RTN_UNSPEC,
2693 		.fi		= NULL,
2694 		.table		= NULL,
2695 		.tclassid	= 0,
2696 	};
2697 	struct rtable *rth;
2698 
2699 	fl4->flowi4_iif = LOOPBACK_IFINDEX;
2700 
2701 	rcu_read_lock();
2702 	rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2703 	rcu_read_unlock();
2704 
2705 	return rth;
2706 }
2707 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2708 
2709 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2710 					    struct fib_result *res,
2711 					    const struct sk_buff *skb)
2712 {
2713 	struct net_device *dev_out = NULL;
2714 	int orig_oif = fl4->flowi4_oif;
2715 	unsigned int flags = 0;
2716 	struct rtable *rth;
2717 	int err;
2718 
2719 	if (fl4->saddr) {
2720 		if (ipv4_is_multicast(fl4->saddr) ||
2721 		    ipv4_is_lbcast(fl4->saddr)) {
2722 			rth = ERR_PTR(-EINVAL);
2723 			goto out;
2724 		}
2725 
2726 		rth = ERR_PTR(-ENETUNREACH);
2727 
2728 		/* I removed check for oif == dev_out->oif here.
2729 		 * It was wrong for two reasons:
2730 		 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2731 		 *    is assigned to multiple interfaces.
2732 		 * 2. Moreover, we are allowed to send packets with saddr
2733 		 *    of another iface. --ANK
2734 		 */
2735 
2736 		if (fl4->flowi4_oif == 0 &&
2737 		    (ipv4_is_multicast(fl4->daddr) ||
2738 		     ipv4_is_lbcast(fl4->daddr))) {
2739 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2740 			dev_out = __ip_dev_find(net, fl4->saddr, false);
2741 			if (!dev_out)
2742 				goto out;
2743 
2744 			/* Special hack: user can direct multicasts
2745 			 * and limited broadcast via necessary interface
2746 			 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2747 			 * This hack is not just for fun, it allows
2748 			 * vic,vat and friends to work.
2749 			 * They bind socket to loopback, set ttl to zero
2750 			 * and expect that it will work.
2751 			 * From the viewpoint of routing cache they are broken,
2752 			 * because we are not allowed to build multicast path
2753 			 * with loopback source addr (look, routing cache
2754 			 * cannot know, that ttl is zero, so that packet
2755 			 * will not leave this host and route is valid).
2756 			 * Luckily, this hack is good workaround.
2757 			 */
2758 
2759 			fl4->flowi4_oif = dev_out->ifindex;
2760 			goto make_route;
2761 		}
2762 
2763 		if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2764 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2765 			if (!__ip_dev_find(net, fl4->saddr, false))
2766 				goto out;
2767 		}
2768 	}
2769 
2770 
2771 	if (fl4->flowi4_oif) {
2772 		dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2773 		rth = ERR_PTR(-ENODEV);
2774 		if (!dev_out)
2775 			goto out;
2776 
2777 		/* RACE: Check return value of inet_select_addr instead. */
2778 		if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2779 			rth = ERR_PTR(-ENETUNREACH);
2780 			goto out;
2781 		}
2782 		if (ipv4_is_local_multicast(fl4->daddr) ||
2783 		    ipv4_is_lbcast(fl4->daddr) ||
2784 		    fl4->flowi4_proto == IPPROTO_IGMP) {
2785 			if (!fl4->saddr)
2786 				fl4->saddr = inet_select_addr(dev_out, 0,
2787 							      RT_SCOPE_LINK);
2788 			goto make_route;
2789 		}
2790 		if (!fl4->saddr) {
2791 			if (ipv4_is_multicast(fl4->daddr))
2792 				fl4->saddr = inet_select_addr(dev_out, 0,
2793 							      fl4->flowi4_scope);
2794 			else if (!fl4->daddr)
2795 				fl4->saddr = inet_select_addr(dev_out, 0,
2796 							      RT_SCOPE_HOST);
2797 		}
2798 	}
2799 
2800 	if (!fl4->daddr) {
2801 		fl4->daddr = fl4->saddr;
2802 		if (!fl4->daddr)
2803 			fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2804 		dev_out = net->loopback_dev;
2805 		fl4->flowi4_oif = LOOPBACK_IFINDEX;
2806 		res->type = RTN_LOCAL;
2807 		flags |= RTCF_LOCAL;
2808 		goto make_route;
2809 	}
2810 
2811 	err = fib_lookup(net, fl4, res, 0);
2812 	if (err) {
2813 		res->fi = NULL;
2814 		res->table = NULL;
2815 		if (fl4->flowi4_oif &&
2816 		    (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2817 			/* Apparently, routing tables are wrong. Assume,
2818 			 * that the destination is on link.
2819 			 *
2820 			 * WHY? DW.
2821 			 * Because we are allowed to send to iface
2822 			 * even if it has NO routes and NO assigned
2823 			 * addresses. When oif is specified, routing
2824 			 * tables are looked up with only one purpose:
2825 			 * to catch if destination is gatewayed, rather than
2826 			 * direct. Moreover, if MSG_DONTROUTE is set,
2827 			 * we send packet, ignoring both routing tables
2828 			 * and ifaddr state. --ANK
2829 			 *
2830 			 *
2831 			 * We could make it even if oif is unknown,
2832 			 * likely IPv6, but we do not.
2833 			 */
2834 
2835 			if (fl4->saddr == 0)
2836 				fl4->saddr = inet_select_addr(dev_out, 0,
2837 							      RT_SCOPE_LINK);
2838 			res->type = RTN_UNICAST;
2839 			goto make_route;
2840 		}
2841 		rth = ERR_PTR(err);
2842 		goto out;
2843 	}
2844 
2845 	if (res->type == RTN_LOCAL) {
2846 		if (!fl4->saddr) {
2847 			if (res->fi->fib_prefsrc)
2848 				fl4->saddr = res->fi->fib_prefsrc;
2849 			else
2850 				fl4->saddr = fl4->daddr;
2851 		}
2852 
2853 		/* L3 master device is the loopback for that domain */
2854 		dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2855 			net->loopback_dev;
2856 
2857 		/* make sure orig_oif points to fib result device even
2858 		 * though packet rx/tx happens over loopback or l3mdev
2859 		 */
2860 		orig_oif = FIB_RES_OIF(*res);
2861 
2862 		fl4->flowi4_oif = dev_out->ifindex;
2863 		flags |= RTCF_LOCAL;
2864 		goto make_route;
2865 	}
2866 
2867 	fib_select_path(net, res, fl4, skb);
2868 
2869 	dev_out = FIB_RES_DEV(*res);
2870 
2871 make_route:
2872 	rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2873 
2874 out:
2875 	return rth;
2876 }
2877 
2878 static struct dst_ops ipv4_dst_blackhole_ops = {
2879 	.family			= AF_INET,
2880 	.default_advmss		= ipv4_default_advmss,
2881 	.neigh_lookup		= ipv4_neigh_lookup,
2882 	.check			= dst_blackhole_check,
2883 	.cow_metrics		= dst_blackhole_cow_metrics,
2884 	.update_pmtu		= dst_blackhole_update_pmtu,
2885 	.redirect		= dst_blackhole_redirect,
2886 	.mtu			= dst_blackhole_mtu,
2887 };
2888 
2889 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2890 {
2891 	struct rtable *ort = dst_rtable(dst_orig);
2892 	struct rtable *rt;
2893 
2894 	rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0);
2895 	if (rt) {
2896 		struct dst_entry *new = &rt->dst;
2897 
2898 		new->__use = 1;
2899 		new->input = dst_discard;
2900 		new->output = dst_discard_out;
2901 
2902 		new->dev = net->loopback_dev;
2903 		netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2904 
2905 		rt->rt_is_input = ort->rt_is_input;
2906 		rt->rt_iif = ort->rt_iif;
2907 		rt->rt_pmtu = ort->rt_pmtu;
2908 		rt->rt_mtu_locked = ort->rt_mtu_locked;
2909 
2910 		rt->rt_genid = rt_genid_ipv4(net);
2911 		rt->rt_flags = ort->rt_flags;
2912 		rt->rt_type = ort->rt_type;
2913 		rt->rt_uses_gateway = ort->rt_uses_gateway;
2914 		rt->rt_gw_family = ort->rt_gw_family;
2915 		if (rt->rt_gw_family == AF_INET)
2916 			rt->rt_gw4 = ort->rt_gw4;
2917 		else if (rt->rt_gw_family == AF_INET6)
2918 			rt->rt_gw6 = ort->rt_gw6;
2919 	}
2920 
2921 	dst_release(dst_orig);
2922 
2923 	return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2924 }
2925 
2926 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2927 				    const struct sock *sk)
2928 {
2929 	struct rtable *rt = __ip_route_output_key(net, flp4);
2930 
2931 	if (IS_ERR(rt))
2932 		return rt;
2933 
2934 	if (flp4->flowi4_proto) {
2935 		flp4->flowi4_oif = rt->dst.dev->ifindex;
2936 		rt = dst_rtable(xfrm_lookup_route(net, &rt->dst,
2937 						  flowi4_to_flowi(flp4),
2938 						  sk, 0));
2939 	}
2940 
2941 	return rt;
2942 }
2943 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2944 
2945 /* called with rcu_read_lock held */
2946 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2947 			struct rtable *rt, u32 table_id, dscp_t dscp,
2948 			struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2949 			u32 seq, unsigned int flags)
2950 {
2951 	struct rtmsg *r;
2952 	struct nlmsghdr *nlh;
2953 	unsigned long expires = 0;
2954 	u32 error;
2955 	u32 metrics[RTAX_MAX];
2956 
2957 	nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2958 	if (!nlh)
2959 		return -EMSGSIZE;
2960 
2961 	r = nlmsg_data(nlh);
2962 	r->rtm_family	 = AF_INET;
2963 	r->rtm_dst_len	= 32;
2964 	r->rtm_src_len	= 0;
2965 	r->rtm_tos	= inet_dscp_to_dsfield(dscp);
2966 	r->rtm_table	= table_id < 256 ? table_id : RT_TABLE_COMPAT;
2967 	if (nla_put_u32(skb, RTA_TABLE, table_id))
2968 		goto nla_put_failure;
2969 	r->rtm_type	= rt->rt_type;
2970 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2971 	r->rtm_protocol = RTPROT_UNSPEC;
2972 	r->rtm_flags	= (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2973 	if (rt->rt_flags & RTCF_NOTIFY)
2974 		r->rtm_flags |= RTM_F_NOTIFY;
2975 	if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2976 		r->rtm_flags |= RTCF_DOREDIRECT;
2977 
2978 	if (nla_put_in_addr(skb, RTA_DST, dst))
2979 		goto nla_put_failure;
2980 	if (src) {
2981 		r->rtm_src_len = 32;
2982 		if (nla_put_in_addr(skb, RTA_SRC, src))
2983 			goto nla_put_failure;
2984 	}
2985 	if (rt->dst.dev &&
2986 	    nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2987 		goto nla_put_failure;
2988 	if (lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2989 		goto nla_put_failure;
2990 #ifdef CONFIG_IP_ROUTE_CLASSID
2991 	if (rt->dst.tclassid &&
2992 	    nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2993 		goto nla_put_failure;
2994 #endif
2995 	if (fl4 && !rt_is_input_route(rt) &&
2996 	    fl4->saddr != src) {
2997 		if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2998 			goto nla_put_failure;
2999 	}
3000 	if (rt->rt_uses_gateway) {
3001 		if (rt->rt_gw_family == AF_INET &&
3002 		    nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
3003 			goto nla_put_failure;
3004 		} else if (rt->rt_gw_family == AF_INET6) {
3005 			int alen = sizeof(struct in6_addr);
3006 			struct nlattr *nla;
3007 			struct rtvia *via;
3008 
3009 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
3010 			if (!nla)
3011 				goto nla_put_failure;
3012 
3013 			via = nla_data(nla);
3014 			via->rtvia_family = AF_INET6;
3015 			memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
3016 		}
3017 	}
3018 
3019 	expires = READ_ONCE(rt->dst.expires);
3020 	if (expires) {
3021 		unsigned long now = jiffies;
3022 
3023 		if (time_before(now, expires))
3024 			expires -= now;
3025 		else
3026 			expires = 0;
3027 	}
3028 
3029 	memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3030 	if (rt->rt_pmtu && expires)
3031 		metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3032 	if (rt->rt_mtu_locked && expires)
3033 		metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3034 	if (rtnetlink_put_metrics(skb, metrics) < 0)
3035 		goto nla_put_failure;
3036 
3037 	if (fl4) {
3038 		if (fl4->flowi4_mark &&
3039 		    nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3040 			goto nla_put_failure;
3041 
3042 		if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3043 		    nla_put_u32(skb, RTA_UID,
3044 				from_kuid_munged(current_user_ns(),
3045 						 fl4->flowi4_uid)))
3046 			goto nla_put_failure;
3047 
3048 		if (rt_is_input_route(rt)) {
3049 #ifdef CONFIG_IP_MROUTE
3050 			if (ipv4_is_multicast(dst) &&
3051 			    !ipv4_is_local_multicast(dst) &&
3052 			    IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) {
3053 				int err = ipmr_get_route(net, skb,
3054 							 fl4->saddr, fl4->daddr,
3055 							 r, portid);
3056 
3057 				if (err <= 0) {
3058 					if (err == 0)
3059 						return 0;
3060 					goto nla_put_failure;
3061 				}
3062 			} else
3063 #endif
3064 				if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3065 					goto nla_put_failure;
3066 		}
3067 	}
3068 
3069 	error = rt->dst.error;
3070 
3071 	if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3072 		goto nla_put_failure;
3073 
3074 	nlmsg_end(skb, nlh);
3075 	return 0;
3076 
3077 nla_put_failure:
3078 	nlmsg_cancel(skb, nlh);
3079 	return -EMSGSIZE;
3080 }
3081 
3082 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3083 			    struct netlink_callback *cb, u32 table_id,
3084 			    struct fnhe_hash_bucket *bucket, int genid,
3085 			    int *fa_index, int fa_start, unsigned int flags)
3086 {
3087 	int i;
3088 
3089 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
3090 		struct fib_nh_exception *fnhe;
3091 
3092 		for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3093 		     fnhe = rcu_dereference(fnhe->fnhe_next)) {
3094 			struct rtable *rt;
3095 			int err;
3096 
3097 			if (*fa_index < fa_start)
3098 				goto next;
3099 
3100 			if (fnhe->fnhe_genid != genid)
3101 				goto next;
3102 
3103 			if (fnhe->fnhe_expires &&
3104 			    time_after(jiffies, fnhe->fnhe_expires))
3105 				goto next;
3106 
3107 			rt = rcu_dereference(fnhe->fnhe_rth_input);
3108 			if (!rt)
3109 				rt = rcu_dereference(fnhe->fnhe_rth_output);
3110 			if (!rt)
3111 				goto next;
3112 
3113 			err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3114 					   table_id, 0, NULL, skb,
3115 					   NETLINK_CB(cb->skb).portid,
3116 					   cb->nlh->nlmsg_seq, flags);
3117 			if (err)
3118 				return err;
3119 next:
3120 			(*fa_index)++;
3121 		}
3122 	}
3123 
3124 	return 0;
3125 }
3126 
3127 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3128 		       u32 table_id, struct fib_info *fi,
3129 		       int *fa_index, int fa_start, unsigned int flags)
3130 {
3131 	struct net *net = sock_net(cb->skb->sk);
3132 	int nhsel, genid = fnhe_genid(net);
3133 
3134 	for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3135 		struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3136 		struct fnhe_hash_bucket *bucket;
3137 		int err;
3138 
3139 		if (nhc->nhc_flags & RTNH_F_DEAD)
3140 			continue;
3141 
3142 		rcu_read_lock();
3143 		bucket = rcu_dereference(nhc->nhc_exceptions);
3144 		err = 0;
3145 		if (bucket)
3146 			err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3147 					       genid, fa_index, fa_start,
3148 					       flags);
3149 		rcu_read_unlock();
3150 		if (err)
3151 			return err;
3152 	}
3153 
3154 	return 0;
3155 }
3156 
3157 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3158 						   u8 ip_proto, __be16 sport,
3159 						   __be16 dport)
3160 {
3161 	struct sk_buff *skb;
3162 	struct iphdr *iph;
3163 
3164 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3165 	if (!skb)
3166 		return NULL;
3167 
3168 	/* Reserve room for dummy headers, this skb can pass
3169 	 * through good chunk of routing engine.
3170 	 */
3171 	skb_reset_mac_header(skb);
3172 	skb_reset_network_header(skb);
3173 	skb->protocol = htons(ETH_P_IP);
3174 	iph = skb_put(skb, sizeof(struct iphdr));
3175 	iph->protocol = ip_proto;
3176 	iph->saddr = src;
3177 	iph->daddr = dst;
3178 	iph->version = 0x4;
3179 	iph->frag_off = 0;
3180 	iph->ihl = 0x5;
3181 	skb_set_transport_header(skb, skb->len);
3182 
3183 	switch (iph->protocol) {
3184 	case IPPROTO_UDP: {
3185 		struct udphdr *udph;
3186 
3187 		udph = skb_put_zero(skb, sizeof(struct udphdr));
3188 		udph->source = sport;
3189 		udph->dest = dport;
3190 		udph->len = htons(sizeof(struct udphdr));
3191 		udph->check = 0;
3192 		break;
3193 	}
3194 	case IPPROTO_TCP: {
3195 		struct tcphdr *tcph;
3196 
3197 		tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3198 		tcph->source	= sport;
3199 		tcph->dest	= dport;
3200 		tcph->doff	= sizeof(struct tcphdr) / 4;
3201 		tcph->rst = 1;
3202 		tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3203 					    src, dst, 0);
3204 		break;
3205 	}
3206 	case IPPROTO_ICMP: {
3207 		struct icmphdr *icmph;
3208 
3209 		icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3210 		icmph->type = ICMP_ECHO;
3211 		icmph->code = 0;
3212 	}
3213 	}
3214 
3215 	return skb;
3216 }
3217 
3218 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3219 				       const struct nlmsghdr *nlh,
3220 				       struct nlattr **tb,
3221 				       struct netlink_ext_ack *extack)
3222 {
3223 	struct rtmsg *rtm;
3224 	int i, err;
3225 
3226 	rtm = nlmsg_payload(nlh, sizeof(*rtm));
3227 	if (!rtm) {
3228 		NL_SET_ERR_MSG(extack,
3229 			       "ipv4: Invalid header for route get request");
3230 		return -EINVAL;
3231 	}
3232 
3233 	if (!netlink_strict_get_check(skb))
3234 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3235 					      rtm_ipv4_policy, extack);
3236 
3237 	if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3238 	    (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3239 	    rtm->rtm_table || rtm->rtm_protocol ||
3240 	    rtm->rtm_scope || rtm->rtm_type) {
3241 		NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3242 		return -EINVAL;
3243 	}
3244 
3245 	if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3246 			       RTM_F_LOOKUP_TABLE |
3247 			       RTM_F_FIB_MATCH)) {
3248 		NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3249 		return -EINVAL;
3250 	}
3251 
3252 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3253 					    rtm_ipv4_policy, extack);
3254 	if (err)
3255 		return err;
3256 
3257 	if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3258 	    (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3259 		NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3260 		return -EINVAL;
3261 	}
3262 
3263 	for (i = 0; i <= RTA_MAX; i++) {
3264 		if (!tb[i])
3265 			continue;
3266 
3267 		switch (i) {
3268 		case RTA_IIF:
3269 		case RTA_OIF:
3270 		case RTA_SRC:
3271 		case RTA_DST:
3272 		case RTA_IP_PROTO:
3273 		case RTA_SPORT:
3274 		case RTA_DPORT:
3275 		case RTA_MARK:
3276 		case RTA_UID:
3277 			break;
3278 		default:
3279 			NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3280 			return -EINVAL;
3281 		}
3282 	}
3283 
3284 	return 0;
3285 }
3286 
3287 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3288 			     struct netlink_ext_ack *extack)
3289 {
3290 	struct net *net = sock_net(in_skb->sk);
3291 	struct nlattr *tb[RTA_MAX+1];
3292 	u32 table_id = RT_TABLE_MAIN;
3293 	__be16 sport = 0, dport = 0;
3294 	struct fib_result res = {};
3295 	u8 ip_proto = IPPROTO_UDP;
3296 	struct rtable *rt = NULL;
3297 	struct sk_buff *skb;
3298 	struct rtmsg *rtm;
3299 	struct flowi4 fl4 = {};
3300 	__be32 dst = 0;
3301 	__be32 src = 0;
3302 	dscp_t dscp;
3303 	kuid_t uid;
3304 	u32 iif;
3305 	int err;
3306 	int mark;
3307 
3308 	err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3309 	if (err < 0)
3310 		return err;
3311 
3312 	rtm = nlmsg_data(nlh);
3313 	src = nla_get_in_addr_default(tb[RTA_SRC], 0);
3314 	dst = nla_get_in_addr_default(tb[RTA_DST], 0);
3315 	iif = nla_get_u32_default(tb[RTA_IIF], 0);
3316 	mark = nla_get_u32_default(tb[RTA_MARK], 0);
3317 	dscp = inet_dsfield_to_dscp(rtm->rtm_tos);
3318 	if (tb[RTA_UID])
3319 		uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3320 	else
3321 		uid = (iif ? INVALID_UID : current_uid());
3322 
3323 	if (tb[RTA_IP_PROTO]) {
3324 		err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3325 						  &ip_proto, AF_INET, extack);
3326 		if (err)
3327 			return err;
3328 	}
3329 
3330 	if (tb[RTA_SPORT])
3331 		sport = nla_get_be16(tb[RTA_SPORT]);
3332 
3333 	if (tb[RTA_DPORT])
3334 		dport = nla_get_be16(tb[RTA_DPORT]);
3335 
3336 	skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3337 	if (!skb)
3338 		return -ENOBUFS;
3339 
3340 	fl4.daddr = dst;
3341 	fl4.saddr = src;
3342 	fl4.flowi4_dscp = dscp;
3343 	fl4.flowi4_oif = nla_get_u32_default(tb[RTA_OIF], 0);
3344 	fl4.flowi4_mark = mark;
3345 	fl4.flowi4_uid = uid;
3346 	if (sport)
3347 		fl4.fl4_sport = sport;
3348 	if (dport)
3349 		fl4.fl4_dport = dport;
3350 	fl4.flowi4_proto = ip_proto;
3351 
3352 	rcu_read_lock();
3353 
3354 	if (iif) {
3355 		struct net_device *dev;
3356 
3357 		dev = dev_get_by_index_rcu(net, iif);
3358 		if (!dev) {
3359 			err = -ENODEV;
3360 			goto errout_rcu;
3361 		}
3362 
3363 		fl4.flowi4_iif = iif; /* for rt_fill_info */
3364 		skb->dev	= dev;
3365 		skb->mark	= mark;
3366 		err = ip_route_input_rcu(skb, dst, src, dscp, dev,
3367 					 &res) ? -EINVAL : 0;
3368 
3369 		rt = skb_rtable(skb);
3370 		if (err == 0 && rt->dst.error)
3371 			err = -rt->dst.error;
3372 	} else {
3373 		fl4.flowi4_iif = LOOPBACK_IFINDEX;
3374 		skb->dev = net->loopback_dev;
3375 		rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3376 		err = 0;
3377 		if (IS_ERR(rt))
3378 			err = PTR_ERR(rt);
3379 		else
3380 			skb_dst_set(skb, &rt->dst);
3381 	}
3382 
3383 	if (err)
3384 		goto errout_rcu;
3385 
3386 	if (rtm->rtm_flags & RTM_F_NOTIFY)
3387 		rt->rt_flags |= RTCF_NOTIFY;
3388 
3389 	if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3390 		table_id = res.table ? res.table->tb_id : 0;
3391 
3392 	/* reset skb for netlink reply msg */
3393 	skb_trim(skb, 0);
3394 	skb_reset_network_header(skb);
3395 	skb_reset_transport_header(skb);
3396 	skb_reset_mac_header(skb);
3397 
3398 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3399 		struct fib_rt_info fri;
3400 
3401 		if (!res.fi) {
3402 			err = fib_props[res.type].error;
3403 			if (!err)
3404 				err = -EHOSTUNREACH;
3405 			goto errout_rcu;
3406 		}
3407 		fri.fi = res.fi;
3408 		fri.tb_id = table_id;
3409 		fri.dst = res.prefix;
3410 		fri.dst_len = res.prefixlen;
3411 		fri.dscp = res.dscp;
3412 		fri.type = rt->rt_type;
3413 		fri.offload = 0;
3414 		fri.trap = 0;
3415 		fri.offload_failed = 0;
3416 		if (res.fa_head) {
3417 			struct fib_alias *fa;
3418 
3419 			hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3420 				u8 slen = 32 - fri.dst_len;
3421 
3422 				if (fa->fa_slen == slen &&
3423 				    fa->tb_id == fri.tb_id &&
3424 				    fa->fa_dscp == fri.dscp &&
3425 				    fa->fa_info == res.fi &&
3426 				    fa->fa_type == fri.type) {
3427 					fri.offload = READ_ONCE(fa->offload);
3428 					fri.trap = READ_ONCE(fa->trap);
3429 					fri.offload_failed =
3430 						READ_ONCE(fa->offload_failed);
3431 					break;
3432 				}
3433 			}
3434 		}
3435 		err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3436 				    nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3437 	} else {
3438 		err = rt_fill_info(net, dst, src, rt, table_id, res.dscp, &fl4,
3439 				   skb, NETLINK_CB(in_skb).portid,
3440 				   nlh->nlmsg_seq, 0);
3441 	}
3442 	if (err < 0)
3443 		goto errout_rcu;
3444 
3445 	rcu_read_unlock();
3446 
3447 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3448 
3449 errout_free:
3450 	return err;
3451 errout_rcu:
3452 	rcu_read_unlock();
3453 	kfree_skb(skb);
3454 	goto errout_free;
3455 }
3456 
3457 void ip_rt_multicast_event(struct in_device *in_dev)
3458 {
3459 	rt_cache_flush(dev_net(in_dev->dev));
3460 }
3461 
3462 #ifdef CONFIG_SYSCTL
3463 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3464 static int ip_rt_gc_min_interval __read_mostly	= HZ / 2;
3465 static int ip_rt_gc_elasticity __read_mostly	= 8;
3466 static int ip_min_valid_pmtu __read_mostly	= IPV4_MIN_MTU;
3467 
3468 static int ipv4_sysctl_rtcache_flush(const struct ctl_table *__ctl, int write,
3469 		void *buffer, size_t *lenp, loff_t *ppos)
3470 {
3471 	struct net *net = (struct net *)__ctl->extra1;
3472 
3473 	if (write) {
3474 		rt_cache_flush(net);
3475 		fnhe_genid_bump(net);
3476 		return 0;
3477 	}
3478 
3479 	return -EINVAL;
3480 }
3481 
3482 static struct ctl_table ipv4_route_table[] = {
3483 	{
3484 		.procname	= "gc_thresh",
3485 		.data		= &ipv4_dst_ops.gc_thresh,
3486 		.maxlen		= sizeof(int),
3487 		.mode		= 0644,
3488 		.proc_handler	= proc_dointvec,
3489 	},
3490 	{
3491 		.procname	= "max_size",
3492 		.data		= &ip_rt_max_size,
3493 		.maxlen		= sizeof(int),
3494 		.mode		= 0644,
3495 		.proc_handler	= proc_dointvec,
3496 	},
3497 	{
3498 		/*  Deprecated. Use gc_min_interval_ms */
3499 
3500 		.procname	= "gc_min_interval",
3501 		.data		= &ip_rt_gc_min_interval,
3502 		.maxlen		= sizeof(int),
3503 		.mode		= 0644,
3504 		.proc_handler	= proc_dointvec_jiffies,
3505 	},
3506 	{
3507 		.procname	= "gc_min_interval_ms",
3508 		.data		= &ip_rt_gc_min_interval,
3509 		.maxlen		= sizeof(int),
3510 		.mode		= 0644,
3511 		.proc_handler	= proc_dointvec_ms_jiffies,
3512 	},
3513 	{
3514 		.procname	= "gc_timeout",
3515 		.data		= &ip_rt_gc_timeout,
3516 		.maxlen		= sizeof(int),
3517 		.mode		= 0644,
3518 		.proc_handler	= proc_dointvec_jiffies,
3519 	},
3520 	{
3521 		.procname	= "gc_interval",
3522 		.data		= &ip_rt_gc_interval,
3523 		.maxlen		= sizeof(int),
3524 		.mode		= 0644,
3525 		.proc_handler	= proc_dointvec_jiffies,
3526 	},
3527 	{
3528 		.procname	= "redirect_load",
3529 		.data		= &ip_rt_redirect_load,
3530 		.maxlen		= sizeof(int),
3531 		.mode		= 0644,
3532 		.proc_handler	= proc_dointvec,
3533 	},
3534 	{
3535 		.procname	= "redirect_number",
3536 		.data		= &ip_rt_redirect_number,
3537 		.maxlen		= sizeof(int),
3538 		.mode		= 0644,
3539 		.proc_handler	= proc_dointvec,
3540 	},
3541 	{
3542 		.procname	= "redirect_silence",
3543 		.data		= &ip_rt_redirect_silence,
3544 		.maxlen		= sizeof(int),
3545 		.mode		= 0644,
3546 		.proc_handler	= proc_dointvec,
3547 	},
3548 	{
3549 		.procname	= "error_cost",
3550 		.data		= &ip_rt_error_cost,
3551 		.maxlen		= sizeof(int),
3552 		.mode		= 0644,
3553 		.proc_handler	= proc_dointvec,
3554 	},
3555 	{
3556 		.procname	= "error_burst",
3557 		.data		= &ip_rt_error_burst,
3558 		.maxlen		= sizeof(int),
3559 		.mode		= 0644,
3560 		.proc_handler	= proc_dointvec,
3561 	},
3562 	{
3563 		.procname	= "gc_elasticity",
3564 		.data		= &ip_rt_gc_elasticity,
3565 		.maxlen		= sizeof(int),
3566 		.mode		= 0644,
3567 		.proc_handler	= proc_dointvec,
3568 	},
3569 };
3570 
3571 static const char ipv4_route_flush_procname[] = "flush";
3572 
3573 static struct ctl_table ipv4_route_netns_table[] = {
3574 	{
3575 		.procname	= ipv4_route_flush_procname,
3576 		.maxlen		= sizeof(int),
3577 		.mode		= 0200,
3578 		.proc_handler	= ipv4_sysctl_rtcache_flush,
3579 	},
3580 	{
3581 		.procname       = "min_pmtu",
3582 		.data           = &init_net.ipv4.ip_rt_min_pmtu,
3583 		.maxlen         = sizeof(int),
3584 		.mode           = 0644,
3585 		.proc_handler   = proc_dointvec_minmax,
3586 		.extra1         = &ip_min_valid_pmtu,
3587 	},
3588 	{
3589 		.procname       = "mtu_expires",
3590 		.data           = &init_net.ipv4.ip_rt_mtu_expires,
3591 		.maxlen         = sizeof(int),
3592 		.mode           = 0644,
3593 		.proc_handler   = proc_dointvec_jiffies,
3594 	},
3595 	{
3596 		.procname   = "min_adv_mss",
3597 		.data       = &init_net.ipv4.ip_rt_min_advmss,
3598 		.maxlen     = sizeof(int),
3599 		.mode       = 0644,
3600 		.proc_handler   = proc_dointvec,
3601 	},
3602 };
3603 
3604 static __net_init int sysctl_route_net_init(struct net *net)
3605 {
3606 	struct ctl_table *tbl;
3607 	size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3608 
3609 	tbl = ipv4_route_netns_table;
3610 	if (!net_eq(net, &init_net)) {
3611 		int i;
3612 
3613 		tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3614 		if (!tbl)
3615 			goto err_dup;
3616 
3617 		/* Don't export non-whitelisted sysctls to unprivileged users */
3618 		if (net->user_ns != &init_user_ns) {
3619 			if (tbl[0].procname != ipv4_route_flush_procname)
3620 				table_size = 0;
3621 		}
3622 
3623 		/* Update the variables to point into the current struct net
3624 		 * except for the first element flush
3625 		 */
3626 		for (i = 1; i < table_size; i++)
3627 			tbl[i].data += (void *)net - (void *)&init_net;
3628 	}
3629 	tbl[0].extra1 = net;
3630 
3631 	net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3632 						     tbl, table_size);
3633 	if (!net->ipv4.route_hdr)
3634 		goto err_reg;
3635 	return 0;
3636 
3637 err_reg:
3638 	if (tbl != ipv4_route_netns_table)
3639 		kfree(tbl);
3640 err_dup:
3641 	return -ENOMEM;
3642 }
3643 
3644 static __net_exit void sysctl_route_net_exit(struct net *net)
3645 {
3646 	const struct ctl_table *tbl;
3647 
3648 	tbl = net->ipv4.route_hdr->ctl_table_arg;
3649 	unregister_net_sysctl_table(net->ipv4.route_hdr);
3650 	BUG_ON(tbl == ipv4_route_netns_table);
3651 	kfree(tbl);
3652 }
3653 
3654 static __net_initdata struct pernet_operations sysctl_route_ops = {
3655 	.init = sysctl_route_net_init,
3656 	.exit = sysctl_route_net_exit,
3657 };
3658 #endif
3659 
3660 static __net_init int netns_ip_rt_init(struct net *net)
3661 {
3662 	/* Set default value for namespaceified sysctls */
3663 	net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3664 	net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3665 	net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3666 	return 0;
3667 }
3668 
3669 static struct pernet_operations __net_initdata ip_rt_ops = {
3670 	.init = netns_ip_rt_init,
3671 };
3672 
3673 static __net_init int rt_genid_init(struct net *net)
3674 {
3675 	atomic_set(&net->ipv4.rt_genid, 0);
3676 	atomic_set(&net->fnhe_genid, 0);
3677 	atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3678 	return 0;
3679 }
3680 
3681 static __net_initdata struct pernet_operations rt_genid_ops = {
3682 	.init = rt_genid_init,
3683 };
3684 
3685 static int __net_init ipv4_inetpeer_init(struct net *net)
3686 {
3687 	struct inet_peer_base *bp = kmalloc_obj(*bp);
3688 
3689 	if (!bp)
3690 		return -ENOMEM;
3691 	inet_peer_base_init(bp);
3692 	net->ipv4.peers = bp;
3693 	return 0;
3694 }
3695 
3696 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3697 {
3698 	struct inet_peer_base *bp = net->ipv4.peers;
3699 
3700 	net->ipv4.peers = NULL;
3701 	inetpeer_invalidate_tree(bp);
3702 	kfree(bp);
3703 }
3704 
3705 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3706 	.init	=	ipv4_inetpeer_init,
3707 	.exit	=	ipv4_inetpeer_exit,
3708 };
3709 
3710 #ifdef CONFIG_IP_ROUTE_CLASSID
3711 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3712 #endif /* CONFIG_IP_ROUTE_CLASSID */
3713 
3714 static const struct rtnl_msg_handler ip_rt_rtnl_msg_handlers[] __initconst = {
3715 	{.protocol = PF_INET, .msgtype = RTM_GETROUTE,
3716 	 .doit = inet_rtm_getroute, .flags = RTNL_FLAG_DOIT_UNLOCKED},
3717 };
3718 
3719 int __init ip_rt_init(void)
3720 {
3721 	void *idents_hash;
3722 	int cpu;
3723 
3724 	/* For modern hosts, this will use 2 MB of memory */
3725 	idents_hash = alloc_large_system_hash("IP idents",
3726 					      sizeof(*ip_idents) + sizeof(*ip_tstamps),
3727 					      0,
3728 					      16, /* one bucket per 64 KB */
3729 					      HASH_ZERO,
3730 					      NULL,
3731 					      &ip_idents_mask,
3732 					      2048,
3733 					      256*1024);
3734 
3735 	ip_idents = idents_hash;
3736 
3737 	get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3738 
3739 	ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3740 
3741 	for_each_possible_cpu(cpu) {
3742 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3743 
3744 		INIT_LIST_HEAD(&ul->head);
3745 		spin_lock_init(&ul->lock);
3746 	}
3747 #ifdef CONFIG_IP_ROUTE_CLASSID
3748 	ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3749 	if (!ip_rt_acct)
3750 		panic("IP: failed to allocate ip_rt_acct\n");
3751 #endif
3752 
3753 	ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable,
3754 					      SLAB_HWCACHE_ALIGN | SLAB_PANIC);
3755 
3756 	ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3757 
3758 	if (dst_entries_init(&ipv4_dst_ops) < 0)
3759 		panic("IP: failed to allocate ipv4_dst_ops counter\n");
3760 
3761 	if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3762 		panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3763 
3764 	ipv4_dst_ops.gc_thresh = ~0;
3765 	ip_rt_max_size = INT_MAX;
3766 
3767 	devinet_init();
3768 	ip_fib_init();
3769 
3770 	if (ip_rt_proc_init())
3771 		pr_err("Unable to create route proc files\n");
3772 #ifdef CONFIG_XFRM
3773 	xfrm_init();
3774 	xfrm4_init();
3775 #endif
3776 	rtnl_register_many(ip_rt_rtnl_msg_handlers);
3777 
3778 #ifdef CONFIG_SYSCTL
3779 	register_pernet_subsys(&sysctl_route_ops);
3780 #endif
3781 	register_pernet_subsys(&ip_rt_ops);
3782 	register_pernet_subsys(&rt_genid_ops);
3783 	register_pernet_subsys(&ipv4_inetpeer_ops);
3784 	return 0;
3785 }
3786 
3787 #ifdef CONFIG_SYSCTL
3788 /*
3789  * We really need to sanitize the damn ipv4 init order, then all
3790  * this nonsense will go away.
3791  */
3792 void __init ip_static_sysctl_init(void)
3793 {
3794 	register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3795 }
3796 #endif
3797