xref: /linux/net/ipv4/route.c (revision 2dcb8e8782d8e4c38903bf37b1a24d3ffd193da7)
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/inet_dscp.h>
88 #include <net/net_namespace.h>
89 #include <net/ip.h>
90 #include <net/route.h>
91 #include <net/inetpeer.h>
92 #include <net/sock.h>
93 #include <net/ip_fib.h>
94 #include <net/nexthop.h>
95 #include <net/tcp.h>
96 #include <net/icmp.h>
97 #include <net/xfrm.h>
98 #include <net/lwtunnel.h>
99 #include <net/netevent.h>
100 #include <net/rtnetlink.h>
101 #ifdef CONFIG_SYSCTL
102 #include <linux/sysctl.h>
103 #endif
104 #include <net/secure_seq.h>
105 #include <net/ip_tunnels.h>
106 
107 #include "fib_lookup.h"
108 
109 #define RT_FL_TOS(oldflp4) \
110 	((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
111 
112 #define RT_GC_TIMEOUT (300*HZ)
113 
114 #define DEFAULT_MIN_PMTU (512 + 20 + 20)
115 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
116 #define DEFAULT_MIN_ADVMSS 256
117 static int ip_rt_max_size;
118 static int ip_rt_redirect_number __read_mostly	= 9;
119 static int ip_rt_redirect_load __read_mostly	= HZ / 50;
120 static int ip_rt_redirect_silence __read_mostly	= ((HZ / 50) << (9 + 1));
121 static int ip_rt_error_cost __read_mostly	= HZ;
122 static int ip_rt_error_burst __read_mostly	= 5 * HZ;
123 
124 static int ip_rt_gc_timeout __read_mostly	= RT_GC_TIMEOUT;
125 
126 /*
127  *	Interface to generic destination cache.
128  */
129 
130 INDIRECT_CALLABLE_SCOPE
131 struct dst_entry	*ipv4_dst_check(struct dst_entry *dst, u32 cookie);
132 static unsigned int	 ipv4_default_advmss(const struct dst_entry *dst);
133 INDIRECT_CALLABLE_SCOPE
134 unsigned int		ipv4_mtu(const struct dst_entry *dst);
135 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
136 static void		 ipv4_link_failure(struct sk_buff *skb);
137 static void		 ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
138 					   struct sk_buff *skb, u32 mtu,
139 					   bool confirm_neigh);
140 static void		 ip_do_redirect(struct dst_entry *dst, struct sock *sk,
141 					struct sk_buff *skb);
142 static void		ipv4_dst_destroy(struct dst_entry *dst);
143 
144 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
145 {
146 	WARN_ON(1);
147 	return NULL;
148 }
149 
150 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
151 					   struct sk_buff *skb,
152 					   const void *daddr);
153 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
154 
155 static struct dst_ops ipv4_dst_ops = {
156 	.family =		AF_INET,
157 	.check =		ipv4_dst_check,
158 	.default_advmss =	ipv4_default_advmss,
159 	.mtu =			ipv4_mtu,
160 	.cow_metrics =		ipv4_cow_metrics,
161 	.destroy =		ipv4_dst_destroy,
162 	.negative_advice =	ipv4_negative_advice,
163 	.link_failure =		ipv4_link_failure,
164 	.update_pmtu =		ip_rt_update_pmtu,
165 	.redirect =		ip_do_redirect,
166 	.local_out =		__ip_local_out,
167 	.neigh_lookup =		ipv4_neigh_lookup,
168 	.confirm_neigh =	ipv4_confirm_neigh,
169 };
170 
171 #define ECN_OR_COST(class)	TC_PRIO_##class
172 
173 const __u8 ip_tos2prio[16] = {
174 	TC_PRIO_BESTEFFORT,
175 	ECN_OR_COST(BESTEFFORT),
176 	TC_PRIO_BESTEFFORT,
177 	ECN_OR_COST(BESTEFFORT),
178 	TC_PRIO_BULK,
179 	ECN_OR_COST(BULK),
180 	TC_PRIO_BULK,
181 	ECN_OR_COST(BULK),
182 	TC_PRIO_INTERACTIVE,
183 	ECN_OR_COST(INTERACTIVE),
184 	TC_PRIO_INTERACTIVE,
185 	ECN_OR_COST(INTERACTIVE),
186 	TC_PRIO_INTERACTIVE_BULK,
187 	ECN_OR_COST(INTERACTIVE_BULK),
188 	TC_PRIO_INTERACTIVE_BULK,
189 	ECN_OR_COST(INTERACTIVE_BULK)
190 };
191 EXPORT_SYMBOL(ip_tos2prio);
192 
193 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
194 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
195 
196 #ifdef CONFIG_PROC_FS
197 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
198 {
199 	if (*pos)
200 		return NULL;
201 	return SEQ_START_TOKEN;
202 }
203 
204 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
205 {
206 	++*pos;
207 	return NULL;
208 }
209 
210 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
211 {
212 }
213 
214 static int rt_cache_seq_show(struct seq_file *seq, void *v)
215 {
216 	if (v == SEQ_START_TOKEN)
217 		seq_printf(seq, "%-127s\n",
218 			   "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
219 			   "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
220 			   "HHUptod\tSpecDst");
221 	return 0;
222 }
223 
224 static const struct seq_operations rt_cache_seq_ops = {
225 	.start  = rt_cache_seq_start,
226 	.next   = rt_cache_seq_next,
227 	.stop   = rt_cache_seq_stop,
228 	.show   = rt_cache_seq_show,
229 };
230 
231 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
232 {
233 	int cpu;
234 
235 	if (*pos == 0)
236 		return SEQ_START_TOKEN;
237 
238 	for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
239 		if (!cpu_possible(cpu))
240 			continue;
241 		*pos = cpu+1;
242 		return &per_cpu(rt_cache_stat, cpu);
243 	}
244 	return NULL;
245 }
246 
247 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
248 {
249 	int cpu;
250 
251 	for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
252 		if (!cpu_possible(cpu))
253 			continue;
254 		*pos = cpu+1;
255 		return &per_cpu(rt_cache_stat, cpu);
256 	}
257 	(*pos)++;
258 	return NULL;
259 
260 }
261 
262 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
263 {
264 
265 }
266 
267 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
268 {
269 	struct rt_cache_stat *st = v;
270 
271 	if (v == SEQ_START_TOKEN) {
272 		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");
273 		return 0;
274 	}
275 
276 	seq_printf(seq, "%08x %08x %08x    %08x   %08x    %08x %08x       "
277 			"%08x       %08x %08x     %08x    %08x %08x   "
278 			"%08x     %08x        %08x        %08x\n",
279 		   dst_entries_get_slow(&ipv4_dst_ops),
280 		   0, /* st->in_hit */
281 		   st->in_slow_tot,
282 		   st->in_slow_mc,
283 		   st->in_no_route,
284 		   st->in_brd,
285 		   st->in_martian_dst,
286 		   st->in_martian_src,
287 
288 		   0, /* st->out_hit */
289 		   st->out_slow_tot,
290 		   st->out_slow_mc,
291 
292 		   0, /* st->gc_total */
293 		   0, /* st->gc_ignored */
294 		   0, /* st->gc_goal_miss */
295 		   0, /* st->gc_dst_overflow */
296 		   0, /* st->in_hlist_search */
297 		   0  /* st->out_hlist_search */
298 		);
299 	return 0;
300 }
301 
302 static const struct seq_operations rt_cpu_seq_ops = {
303 	.start  = rt_cpu_seq_start,
304 	.next   = rt_cpu_seq_next,
305 	.stop   = rt_cpu_seq_stop,
306 	.show   = rt_cpu_seq_show,
307 };
308 
309 #ifdef CONFIG_IP_ROUTE_CLASSID
310 static int rt_acct_proc_show(struct seq_file *m, void *v)
311 {
312 	struct ip_rt_acct *dst, *src;
313 	unsigned int i, j;
314 
315 	dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
316 	if (!dst)
317 		return -ENOMEM;
318 
319 	for_each_possible_cpu(i) {
320 		src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
321 		for (j = 0; j < 256; j++) {
322 			dst[j].o_bytes   += src[j].o_bytes;
323 			dst[j].o_packets += src[j].o_packets;
324 			dst[j].i_bytes   += src[j].i_bytes;
325 			dst[j].i_packets += src[j].i_packets;
326 		}
327 	}
328 
329 	seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
330 	kfree(dst);
331 	return 0;
332 }
333 #endif
334 
335 static int __net_init ip_rt_do_proc_init(struct net *net)
336 {
337 	struct proc_dir_entry *pde;
338 
339 	pde = proc_create_seq("rt_cache", 0444, net->proc_net,
340 			      &rt_cache_seq_ops);
341 	if (!pde)
342 		goto err1;
343 
344 	pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
345 			      &rt_cpu_seq_ops);
346 	if (!pde)
347 		goto err2;
348 
349 #ifdef CONFIG_IP_ROUTE_CLASSID
350 	pde = proc_create_single("rt_acct", 0, net->proc_net,
351 			rt_acct_proc_show);
352 	if (!pde)
353 		goto err3;
354 #endif
355 	return 0;
356 
357 #ifdef CONFIG_IP_ROUTE_CLASSID
358 err3:
359 	remove_proc_entry("rt_cache", net->proc_net_stat);
360 #endif
361 err2:
362 	remove_proc_entry("rt_cache", net->proc_net);
363 err1:
364 	return -ENOMEM;
365 }
366 
367 static void __net_exit ip_rt_do_proc_exit(struct net *net)
368 {
369 	remove_proc_entry("rt_cache", net->proc_net_stat);
370 	remove_proc_entry("rt_cache", net->proc_net);
371 #ifdef CONFIG_IP_ROUTE_CLASSID
372 	remove_proc_entry("rt_acct", net->proc_net);
373 #endif
374 }
375 
376 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
377 	.init = ip_rt_do_proc_init,
378 	.exit = ip_rt_do_proc_exit,
379 };
380 
381 static int __init ip_rt_proc_init(void)
382 {
383 	return register_pernet_subsys(&ip_rt_proc_ops);
384 }
385 
386 #else
387 static inline int ip_rt_proc_init(void)
388 {
389 	return 0;
390 }
391 #endif /* CONFIG_PROC_FS */
392 
393 static inline bool rt_is_expired(const struct rtable *rth)
394 {
395 	return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
396 }
397 
398 void rt_cache_flush(struct net *net)
399 {
400 	rt_genid_bump_ipv4(net);
401 }
402 
403 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
404 					   struct sk_buff *skb,
405 					   const void *daddr)
406 {
407 	const struct rtable *rt = container_of(dst, struct rtable, dst);
408 	struct net_device *dev = dst->dev;
409 	struct neighbour *n;
410 
411 	rcu_read_lock_bh();
412 
413 	if (likely(rt->rt_gw_family == AF_INET)) {
414 		n = ip_neigh_gw4(dev, rt->rt_gw4);
415 	} else if (rt->rt_gw_family == AF_INET6) {
416 		n = ip_neigh_gw6(dev, &rt->rt_gw6);
417         } else {
418 		__be32 pkey;
419 
420 		pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
421 		n = ip_neigh_gw4(dev, pkey);
422 	}
423 
424 	if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
425 		n = NULL;
426 
427 	rcu_read_unlock_bh();
428 
429 	return n;
430 }
431 
432 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
433 {
434 	const struct rtable *rt = container_of(dst, struct rtable, dst);
435 	struct net_device *dev = dst->dev;
436 	const __be32 *pkey = daddr;
437 
438 	if (rt->rt_gw_family == AF_INET) {
439 		pkey = (const __be32 *)&rt->rt_gw4;
440 	} else if (rt->rt_gw_family == AF_INET6) {
441 		return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
442 	} else if (!daddr ||
443 		 (rt->rt_flags &
444 		  (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
445 		return;
446 	}
447 	__ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
448 }
449 
450 /* Hash tables of size 2048..262144 depending on RAM size.
451  * Each bucket uses 8 bytes.
452  */
453 static u32 ip_idents_mask __read_mostly;
454 static atomic_t *ip_idents __read_mostly;
455 static u32 *ip_tstamps __read_mostly;
456 
457 /* In order to protect privacy, we add a perturbation to identifiers
458  * if one generator is seldom used. This makes hard for an attacker
459  * to infer how many packets were sent between two points in time.
460  */
461 static u32 ip_idents_reserve(u32 hash, int segs)
462 {
463 	u32 bucket, old, now = (u32)jiffies;
464 	atomic_t *p_id;
465 	u32 *p_tstamp;
466 	u32 delta = 0;
467 
468 	bucket = hash & ip_idents_mask;
469 	p_tstamp = ip_tstamps + bucket;
470 	p_id = ip_idents + bucket;
471 	old = READ_ONCE(*p_tstamp);
472 
473 	if (old != now && cmpxchg(p_tstamp, old, now) == old)
474 		delta = prandom_u32_max(now - old);
475 
476 	/* If UBSAN reports an error there, please make sure your compiler
477 	 * supports -fno-strict-overflow before reporting it that was a bug
478 	 * in UBSAN, and it has been fixed in GCC-8.
479 	 */
480 	return atomic_add_return(segs + delta, p_id) - segs;
481 }
482 
483 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
484 {
485 	u32 hash, id;
486 
487 	/* Note the following code is not safe, but this is okay. */
488 	if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
489 		get_random_bytes(&net->ipv4.ip_id_key,
490 				 sizeof(net->ipv4.ip_id_key));
491 
492 	hash = siphash_3u32((__force u32)iph->daddr,
493 			    (__force u32)iph->saddr,
494 			    iph->protocol,
495 			    &net->ipv4.ip_id_key);
496 	id = ip_idents_reserve(hash, segs);
497 	iph->id = htons(id);
498 }
499 EXPORT_SYMBOL(__ip_select_ident);
500 
501 static void ip_rt_fix_tos(struct flowi4 *fl4)
502 {
503 	__u8 tos = RT_FL_TOS(fl4);
504 
505 	fl4->flowi4_tos = tos & IPTOS_RT_MASK;
506 	fl4->flowi4_scope = tos & RTO_ONLINK ?
507 			    RT_SCOPE_LINK : RT_SCOPE_UNIVERSE;
508 }
509 
510 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
511 			     const struct sock *sk,
512 			     const struct iphdr *iph,
513 			     int oif, u8 tos,
514 			     u8 prot, u32 mark, int flow_flags)
515 {
516 	if (sk) {
517 		const struct inet_sock *inet = inet_sk(sk);
518 
519 		oif = sk->sk_bound_dev_if;
520 		mark = sk->sk_mark;
521 		tos = RT_CONN_FLAGS(sk);
522 		prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
523 	}
524 	flowi4_init_output(fl4, oif, mark, tos,
525 			   RT_SCOPE_UNIVERSE, prot,
526 			   flow_flags,
527 			   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 tos = RT_TOS(iph->tos);
538 	u8 prot = iph->protocol;
539 	u32 mark = skb->mark;
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, sk->sk_mark,
555 			   RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
556 			   inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
557 			   inet_sk_flowi_flags(sk),
558 			   daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
559 	rcu_read_unlock();
560 }
561 
562 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
563 				 const struct sk_buff *skb)
564 {
565 	if (skb)
566 		build_skb_flow_key(fl4, skb, sk);
567 	else
568 		build_sk_flow_key(fl4, sk);
569 }
570 
571 static DEFINE_SPINLOCK(fnhe_lock);
572 
573 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
574 {
575 	struct rtable *rt;
576 
577 	rt = rcu_dereference(fnhe->fnhe_rth_input);
578 	if (rt) {
579 		RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
580 		dst_dev_put(&rt->dst);
581 		dst_release(&rt->dst);
582 	}
583 	rt = rcu_dereference(fnhe->fnhe_rth_output);
584 	if (rt) {
585 		RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
586 		dst_dev_put(&rt->dst);
587 		dst_release(&rt->dst);
588 	}
589 }
590 
591 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
592 {
593 	struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
594 	struct fib_nh_exception *fnhe, *oldest = NULL;
595 
596 	for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
597 		fnhe = rcu_dereference_protected(*fnhe_p,
598 						 lockdep_is_held(&fnhe_lock));
599 		if (!fnhe)
600 			break;
601 		if (!oldest ||
602 		    time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
603 			oldest = fnhe;
604 			oldest_p = fnhe_p;
605 		}
606 	}
607 	fnhe_flush_routes(oldest);
608 	*oldest_p = oldest->fnhe_next;
609 	kfree_rcu(oldest, rcu);
610 }
611 
612 static u32 fnhe_hashfun(__be32 daddr)
613 {
614 	static siphash_aligned_key_t fnhe_hash_key;
615 	u64 hval;
616 
617 	net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
618 	hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
619 	return hash_64(hval, FNHE_HASH_SHIFT);
620 }
621 
622 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
623 {
624 	rt->rt_pmtu = fnhe->fnhe_pmtu;
625 	rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
626 	rt->dst.expires = fnhe->fnhe_expires;
627 
628 	if (fnhe->fnhe_gw) {
629 		rt->rt_flags |= RTCF_REDIRECTED;
630 		rt->rt_uses_gateway = 1;
631 		rt->rt_gw_family = AF_INET;
632 		rt->rt_gw4 = fnhe->fnhe_gw;
633 	}
634 }
635 
636 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
637 				  __be32 gw, u32 pmtu, bool lock,
638 				  unsigned long expires)
639 {
640 	struct fnhe_hash_bucket *hash;
641 	struct fib_nh_exception *fnhe;
642 	struct rtable *rt;
643 	u32 genid, hval;
644 	unsigned int i;
645 	int depth;
646 
647 	genid = fnhe_genid(dev_net(nhc->nhc_dev));
648 	hval = fnhe_hashfun(daddr);
649 
650 	spin_lock_bh(&fnhe_lock);
651 
652 	hash = rcu_dereference(nhc->nhc_exceptions);
653 	if (!hash) {
654 		hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
655 		if (!hash)
656 			goto out_unlock;
657 		rcu_assign_pointer(nhc->nhc_exceptions, hash);
658 	}
659 
660 	hash += hval;
661 
662 	depth = 0;
663 	for (fnhe = rcu_dereference(hash->chain); fnhe;
664 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
665 		if (fnhe->fnhe_daddr == daddr)
666 			break;
667 		depth++;
668 	}
669 
670 	if (fnhe) {
671 		if (fnhe->fnhe_genid != genid)
672 			fnhe->fnhe_genid = genid;
673 		if (gw)
674 			fnhe->fnhe_gw = gw;
675 		if (pmtu) {
676 			fnhe->fnhe_pmtu = pmtu;
677 			fnhe->fnhe_mtu_locked = lock;
678 		}
679 		fnhe->fnhe_expires = max(1UL, expires);
680 		/* Update all cached dsts too */
681 		rt = rcu_dereference(fnhe->fnhe_rth_input);
682 		if (rt)
683 			fill_route_from_fnhe(rt, fnhe);
684 		rt = rcu_dereference(fnhe->fnhe_rth_output);
685 		if (rt)
686 			fill_route_from_fnhe(rt, fnhe);
687 	} else {
688 		/* Randomize max depth to avoid some side channels attacks. */
689 		int max_depth = FNHE_RECLAIM_DEPTH +
690 				prandom_u32_max(FNHE_RECLAIM_DEPTH);
691 
692 		while (depth > max_depth) {
693 			fnhe_remove_oldest(hash);
694 			depth--;
695 		}
696 
697 		fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
698 		if (!fnhe)
699 			goto out_unlock;
700 
701 		fnhe->fnhe_next = hash->chain;
702 
703 		fnhe->fnhe_genid = genid;
704 		fnhe->fnhe_daddr = daddr;
705 		fnhe->fnhe_gw = gw;
706 		fnhe->fnhe_pmtu = pmtu;
707 		fnhe->fnhe_mtu_locked = lock;
708 		fnhe->fnhe_expires = max(1UL, expires);
709 
710 		rcu_assign_pointer(hash->chain, fnhe);
711 
712 		/* Exception created; mark the cached routes for the nexthop
713 		 * stale, so anyone caching it rechecks if this exception
714 		 * applies to them.
715 		 */
716 		rt = rcu_dereference(nhc->nhc_rth_input);
717 		if (rt)
718 			rt->dst.obsolete = DST_OBSOLETE_KILL;
719 
720 		for_each_possible_cpu(i) {
721 			struct rtable __rcu **prt;
722 
723 			prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
724 			rt = rcu_dereference(*prt);
725 			if (rt)
726 				rt->dst.obsolete = DST_OBSOLETE_KILL;
727 		}
728 	}
729 
730 	fnhe->fnhe_stamp = jiffies;
731 
732 out_unlock:
733 	spin_unlock_bh(&fnhe_lock);
734 }
735 
736 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
737 			     bool kill_route)
738 {
739 	__be32 new_gw = icmp_hdr(skb)->un.gateway;
740 	__be32 old_gw = ip_hdr(skb)->saddr;
741 	struct net_device *dev = skb->dev;
742 	struct in_device *in_dev;
743 	struct fib_result res;
744 	struct neighbour *n;
745 	struct net *net;
746 
747 	switch (icmp_hdr(skb)->code & 7) {
748 	case ICMP_REDIR_NET:
749 	case ICMP_REDIR_NETTOS:
750 	case ICMP_REDIR_HOST:
751 	case ICMP_REDIR_HOSTTOS:
752 		break;
753 
754 	default:
755 		return;
756 	}
757 
758 	if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
759 		return;
760 
761 	in_dev = __in_dev_get_rcu(dev);
762 	if (!in_dev)
763 		return;
764 
765 	net = dev_net(dev);
766 	if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
767 	    ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
768 	    ipv4_is_zeronet(new_gw))
769 		goto reject_redirect;
770 
771 	if (!IN_DEV_SHARED_MEDIA(in_dev)) {
772 		if (!inet_addr_onlink(in_dev, new_gw, old_gw))
773 			goto reject_redirect;
774 		if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
775 			goto reject_redirect;
776 	} else {
777 		if (inet_addr_type(net, new_gw) != RTN_UNICAST)
778 			goto reject_redirect;
779 	}
780 
781 	n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
782 	if (!n)
783 		n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
784 	if (!IS_ERR(n)) {
785 		if (!(n->nud_state & NUD_VALID)) {
786 			neigh_event_send(n, NULL);
787 		} else {
788 			if (fib_lookup(net, fl4, &res, 0) == 0) {
789 				struct fib_nh_common *nhc;
790 
791 				fib_select_path(net, &res, fl4, skb);
792 				nhc = FIB_RES_NHC(res);
793 				update_or_create_fnhe(nhc, fl4->daddr, new_gw,
794 						0, false,
795 						jiffies + ip_rt_gc_timeout);
796 			}
797 			if (kill_route)
798 				rt->dst.obsolete = DST_OBSOLETE_KILL;
799 			call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
800 		}
801 		neigh_release(n);
802 	}
803 	return;
804 
805 reject_redirect:
806 #ifdef CONFIG_IP_ROUTE_VERBOSE
807 	if (IN_DEV_LOG_MARTIANS(in_dev)) {
808 		const struct iphdr *iph = (const struct iphdr *) skb->data;
809 		__be32 daddr = iph->daddr;
810 		__be32 saddr = iph->saddr;
811 
812 		net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
813 				     "  Advised path = %pI4 -> %pI4\n",
814 				     &old_gw, dev->name, &new_gw,
815 				     &saddr, &daddr);
816 	}
817 #endif
818 	;
819 }
820 
821 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
822 {
823 	struct rtable *rt;
824 	struct flowi4 fl4;
825 	const struct iphdr *iph = (const struct iphdr *) skb->data;
826 	struct net *net = dev_net(skb->dev);
827 	int oif = skb->dev->ifindex;
828 	u8 tos = RT_TOS(iph->tos);
829 	u8 prot = iph->protocol;
830 	u32 mark = skb->mark;
831 
832 	rt = (struct rtable *) dst;
833 
834 	__build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
835 	ip_rt_fix_tos(&fl4);
836 	__ip_do_redirect(rt, skb, &fl4, true);
837 }
838 
839 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
840 {
841 	struct rtable *rt = (struct rtable *)dst;
842 	struct dst_entry *ret = dst;
843 
844 	if (rt) {
845 		if (dst->obsolete > 0) {
846 			ip_rt_put(rt);
847 			ret = NULL;
848 		} else if ((rt->rt_flags & RTCF_REDIRECTED) ||
849 			   rt->dst.expires) {
850 			ip_rt_put(rt);
851 			ret = NULL;
852 		}
853 	}
854 	return ret;
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 	rcu_read_unlock();
891 
892 	net = dev_net(rt->dst.dev);
893 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
894 	if (!peer) {
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_put_peer;
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 #ifdef CONFIG_IP_ROUTE_VERBOSE
929 		if (log_martians &&
930 		    peer->n_redirects == ip_rt_redirect_number)
931 			net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
932 					     &ip_hdr(skb)->saddr, inet_iif(skb),
933 					     &ip_hdr(skb)->daddr, &gw);
934 #endif
935 	}
936 out_put_peer:
937 	inet_putpeer(peer);
938 }
939 
940 static int ip_error(struct sk_buff *skb)
941 {
942 	struct rtable *rt = skb_rtable(skb);
943 	struct net_device *dev = skb->dev;
944 	struct in_device *in_dev;
945 	struct inet_peer *peer;
946 	unsigned long now;
947 	struct net *net;
948 	SKB_DR(reason);
949 	bool send;
950 	int code;
951 
952 	if (netif_is_l3_master(skb->dev)) {
953 		dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
954 		if (!dev)
955 			goto out;
956 	}
957 
958 	in_dev = __in_dev_get_rcu(dev);
959 
960 	/* IP on this device is disabled. */
961 	if (!in_dev)
962 		goto out;
963 
964 	net = dev_net(rt->dst.dev);
965 	if (!IN_DEV_FORWARD(in_dev)) {
966 		switch (rt->dst.error) {
967 		case EHOSTUNREACH:
968 			SKB_DR_SET(reason, IP_INADDRERRORS);
969 			__IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
970 			break;
971 
972 		case ENETUNREACH:
973 			SKB_DR_SET(reason, IP_INNOROUTES);
974 			__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
975 			break;
976 		}
977 		goto out;
978 	}
979 
980 	switch (rt->dst.error) {
981 	case EINVAL:
982 	default:
983 		goto out;
984 	case EHOSTUNREACH:
985 		code = ICMP_HOST_UNREACH;
986 		break;
987 	case ENETUNREACH:
988 		code = ICMP_NET_UNREACH;
989 		SKB_DR_SET(reason, IP_INNOROUTES);
990 		__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
991 		break;
992 	case EACCES:
993 		code = ICMP_PKT_FILTERED;
994 		break;
995 	}
996 
997 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
998 			       l3mdev_master_ifindex(skb->dev), 1);
999 
1000 	send = true;
1001 	if (peer) {
1002 		now = jiffies;
1003 		peer->rate_tokens += now - peer->rate_last;
1004 		if (peer->rate_tokens > ip_rt_error_burst)
1005 			peer->rate_tokens = ip_rt_error_burst;
1006 		peer->rate_last = now;
1007 		if (peer->rate_tokens >= ip_rt_error_cost)
1008 			peer->rate_tokens -= ip_rt_error_cost;
1009 		else
1010 			send = false;
1011 		inet_putpeer(peer);
1012 	}
1013 	if (send)
1014 		icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1015 
1016 out:	kfree_skb_reason(skb, reason);
1017 	return 0;
1018 }
1019 
1020 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1021 {
1022 	struct dst_entry *dst = &rt->dst;
1023 	struct net *net = dev_net(dst->dev);
1024 	struct fib_result res;
1025 	bool lock = false;
1026 	u32 old_mtu;
1027 
1028 	if (ip_mtu_locked(dst))
1029 		return;
1030 
1031 	old_mtu = ipv4_mtu(dst);
1032 	if (old_mtu < mtu)
1033 		return;
1034 
1035 	if (mtu < net->ipv4.ip_rt_min_pmtu) {
1036 		lock = true;
1037 		mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1038 	}
1039 
1040 	if (rt->rt_pmtu == mtu && !lock &&
1041 	    time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2))
1042 		return;
1043 
1044 	rcu_read_lock();
1045 	if (fib_lookup(net, fl4, &res, 0) == 0) {
1046 		struct fib_nh_common *nhc;
1047 
1048 		fib_select_path(net, &res, fl4, NULL);
1049 		nhc = FIB_RES_NHC(res);
1050 		update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1051 				      jiffies + net->ipv4.ip_rt_mtu_expires);
1052 	}
1053 	rcu_read_unlock();
1054 }
1055 
1056 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1057 			      struct sk_buff *skb, u32 mtu,
1058 			      bool confirm_neigh)
1059 {
1060 	struct rtable *rt = (struct rtable *) dst;
1061 	struct flowi4 fl4;
1062 
1063 	ip_rt_build_flow_key(&fl4, sk, skb);
1064 	ip_rt_fix_tos(&fl4);
1065 
1066 	/* Don't make lookup fail for bridged encapsulations */
1067 	if (skb && netif_is_any_bridge_port(skb->dev))
1068 		fl4.flowi4_oif = 0;
1069 
1070 	__ip_rt_update_pmtu(rt, &fl4, mtu);
1071 }
1072 
1073 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1074 		      int oif, u8 protocol)
1075 {
1076 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1077 	struct flowi4 fl4;
1078 	struct rtable *rt;
1079 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
1080 
1081 	__build_flow_key(net, &fl4, NULL, iph, oif,
1082 			 RT_TOS(iph->tos), protocol, mark, 0);
1083 	rt = __ip_route_output_key(net, &fl4);
1084 	if (!IS_ERR(rt)) {
1085 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1086 		ip_rt_put(rt);
1087 	}
1088 }
1089 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1090 
1091 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1092 {
1093 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1094 	struct flowi4 fl4;
1095 	struct rtable *rt;
1096 
1097 	__build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1098 
1099 	if (!fl4.flowi4_mark)
1100 		fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1101 
1102 	rt = __ip_route_output_key(sock_net(sk), &fl4);
1103 	if (!IS_ERR(rt)) {
1104 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1105 		ip_rt_put(rt);
1106 	}
1107 }
1108 
1109 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1110 {
1111 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1112 	struct flowi4 fl4;
1113 	struct rtable *rt;
1114 	struct dst_entry *odst = NULL;
1115 	bool new = false;
1116 	struct net *net = sock_net(sk);
1117 
1118 	bh_lock_sock(sk);
1119 
1120 	if (!ip_sk_accept_pmtu(sk))
1121 		goto out;
1122 
1123 	odst = sk_dst_get(sk);
1124 
1125 	if (sock_owned_by_user(sk) || !odst) {
1126 		__ipv4_sk_update_pmtu(skb, sk, mtu);
1127 		goto out;
1128 	}
1129 
1130 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1131 
1132 	rt = (struct rtable *)odst;
1133 	if (odst->obsolete && !odst->ops->check(odst, 0)) {
1134 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1135 		if (IS_ERR(rt))
1136 			goto out;
1137 
1138 		new = true;
1139 	} else {
1140 		ip_rt_fix_tos(&fl4);
1141 	}
1142 
1143 	__ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu);
1144 
1145 	if (!dst_check(&rt->dst, 0)) {
1146 		if (new)
1147 			dst_release(&rt->dst);
1148 
1149 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1150 		if (IS_ERR(rt))
1151 			goto out;
1152 
1153 		new = true;
1154 	}
1155 
1156 	if (new)
1157 		sk_dst_set(sk, &rt->dst);
1158 
1159 out:
1160 	bh_unlock_sock(sk);
1161 	dst_release(odst);
1162 }
1163 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1164 
1165 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1166 		   int oif, u8 protocol)
1167 {
1168 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1169 	struct flowi4 fl4;
1170 	struct rtable *rt;
1171 
1172 	__build_flow_key(net, &fl4, NULL, iph, oif,
1173 			 RT_TOS(iph->tos), protocol, 0, 0);
1174 	rt = __ip_route_output_key(net, &fl4);
1175 	if (!IS_ERR(rt)) {
1176 		__ip_do_redirect(rt, skb, &fl4, false);
1177 		ip_rt_put(rt);
1178 	}
1179 }
1180 EXPORT_SYMBOL_GPL(ipv4_redirect);
1181 
1182 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1183 {
1184 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1185 	struct flowi4 fl4;
1186 	struct rtable *rt;
1187 	struct net *net = sock_net(sk);
1188 
1189 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1190 	rt = __ip_route_output_key(net, &fl4);
1191 	if (!IS_ERR(rt)) {
1192 		__ip_do_redirect(rt, skb, &fl4, false);
1193 		ip_rt_put(rt);
1194 	}
1195 }
1196 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1197 
1198 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1199 							 u32 cookie)
1200 {
1201 	struct rtable *rt = (struct rtable *) dst;
1202 
1203 	/* All IPV4 dsts are created with ->obsolete set to the value
1204 	 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1205 	 * into this function always.
1206 	 *
1207 	 * When a PMTU/redirect information update invalidates a route,
1208 	 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1209 	 * DST_OBSOLETE_DEAD.
1210 	 */
1211 	if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1212 		return NULL;
1213 	return dst;
1214 }
1215 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1216 
1217 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1218 {
1219 	struct ip_options opt;
1220 	int res;
1221 
1222 	/* Recompile ip options since IPCB may not be valid anymore.
1223 	 * Also check we have a reasonable ipv4 header.
1224 	 */
1225 	if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1226 	    ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1227 		return;
1228 
1229 	memset(&opt, 0, sizeof(opt));
1230 	if (ip_hdr(skb)->ihl > 5) {
1231 		if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1232 			return;
1233 		opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1234 
1235 		rcu_read_lock();
1236 		res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1237 		rcu_read_unlock();
1238 
1239 		if (res)
1240 			return;
1241 	}
1242 	__icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1243 }
1244 
1245 static void ipv4_link_failure(struct sk_buff *skb)
1246 {
1247 	struct rtable *rt;
1248 
1249 	ipv4_send_dest_unreach(skb);
1250 
1251 	rt = skb_rtable(skb);
1252 	if (rt)
1253 		dst_set_expires(&rt->dst, 0);
1254 }
1255 
1256 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1257 {
1258 	pr_debug("%s: %pI4 -> %pI4, %s\n",
1259 		 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1260 		 skb->dev ? skb->dev->name : "?");
1261 	kfree_skb(skb);
1262 	WARN_ON(1);
1263 	return 0;
1264 }
1265 
1266 /*
1267  * We do not cache source address of outgoing interface,
1268  * because it is used only by IP RR, TS and SRR options,
1269  * so that it out of fast path.
1270  *
1271  * BTW remember: "addr" is allowed to be not aligned
1272  * in IP options!
1273  */
1274 
1275 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1276 {
1277 	__be32 src;
1278 
1279 	if (rt_is_output_route(rt))
1280 		src = ip_hdr(skb)->saddr;
1281 	else {
1282 		struct fib_result res;
1283 		struct iphdr *iph = ip_hdr(skb);
1284 		struct flowi4 fl4 = {
1285 			.daddr = iph->daddr,
1286 			.saddr = iph->saddr,
1287 			.flowi4_tos = RT_TOS(iph->tos),
1288 			.flowi4_oif = rt->dst.dev->ifindex,
1289 			.flowi4_iif = skb->dev->ifindex,
1290 			.flowi4_mark = skb->mark,
1291 		};
1292 
1293 		rcu_read_lock();
1294 		if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1295 			src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1296 		else
1297 			src = inet_select_addr(rt->dst.dev,
1298 					       rt_nexthop(rt, iph->daddr),
1299 					       RT_SCOPE_UNIVERSE);
1300 		rcu_read_unlock();
1301 	}
1302 	memcpy(addr, &src, 4);
1303 }
1304 
1305 #ifdef CONFIG_IP_ROUTE_CLASSID
1306 static void set_class_tag(struct rtable *rt, u32 tag)
1307 {
1308 	if (!(rt->dst.tclassid & 0xFFFF))
1309 		rt->dst.tclassid |= tag & 0xFFFF;
1310 	if (!(rt->dst.tclassid & 0xFFFF0000))
1311 		rt->dst.tclassid |= tag & 0xFFFF0000;
1312 }
1313 #endif
1314 
1315 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1316 {
1317 	struct net *net = dev_net(dst->dev);
1318 	unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1319 	unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1320 				    net->ipv4.ip_rt_min_advmss);
1321 
1322 	return min(advmss, IPV4_MAX_PMTU - header_size);
1323 }
1324 
1325 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1326 {
1327 	return ip_dst_mtu_maybe_forward(dst, false);
1328 }
1329 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1330 
1331 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1332 {
1333 	struct fnhe_hash_bucket *hash;
1334 	struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1335 	u32 hval = fnhe_hashfun(daddr);
1336 
1337 	spin_lock_bh(&fnhe_lock);
1338 
1339 	hash = rcu_dereference_protected(nhc->nhc_exceptions,
1340 					 lockdep_is_held(&fnhe_lock));
1341 	hash += hval;
1342 
1343 	fnhe_p = &hash->chain;
1344 	fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1345 	while (fnhe) {
1346 		if (fnhe->fnhe_daddr == daddr) {
1347 			rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1348 				fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1349 			/* set fnhe_daddr to 0 to ensure it won't bind with
1350 			 * new dsts in rt_bind_exception().
1351 			 */
1352 			fnhe->fnhe_daddr = 0;
1353 			fnhe_flush_routes(fnhe);
1354 			kfree_rcu(fnhe, rcu);
1355 			break;
1356 		}
1357 		fnhe_p = &fnhe->fnhe_next;
1358 		fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1359 						 lockdep_is_held(&fnhe_lock));
1360 	}
1361 
1362 	spin_unlock_bh(&fnhe_lock);
1363 }
1364 
1365 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1366 					       __be32 daddr)
1367 {
1368 	struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1369 	struct fib_nh_exception *fnhe;
1370 	u32 hval;
1371 
1372 	if (!hash)
1373 		return NULL;
1374 
1375 	hval = fnhe_hashfun(daddr);
1376 
1377 	for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1378 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
1379 		if (fnhe->fnhe_daddr == daddr) {
1380 			if (fnhe->fnhe_expires &&
1381 			    time_after(jiffies, fnhe->fnhe_expires)) {
1382 				ip_del_fnhe(nhc, daddr);
1383 				break;
1384 			}
1385 			return fnhe;
1386 		}
1387 	}
1388 	return NULL;
1389 }
1390 
1391 /* MTU selection:
1392  * 1. mtu on route is locked - use it
1393  * 2. mtu from nexthop exception
1394  * 3. mtu from egress device
1395  */
1396 
1397 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1398 {
1399 	struct fib_nh_common *nhc = res->nhc;
1400 	struct net_device *dev = nhc->nhc_dev;
1401 	struct fib_info *fi = res->fi;
1402 	u32 mtu = 0;
1403 
1404 	if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu ||
1405 	    fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1406 		mtu = fi->fib_mtu;
1407 
1408 	if (likely(!mtu)) {
1409 		struct fib_nh_exception *fnhe;
1410 
1411 		fnhe = find_exception(nhc, daddr);
1412 		if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1413 			mtu = fnhe->fnhe_pmtu;
1414 	}
1415 
1416 	if (likely(!mtu))
1417 		mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1418 
1419 	return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1420 }
1421 
1422 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1423 			      __be32 daddr, const bool do_cache)
1424 {
1425 	bool ret = false;
1426 
1427 	spin_lock_bh(&fnhe_lock);
1428 
1429 	if (daddr == fnhe->fnhe_daddr) {
1430 		struct rtable __rcu **porig;
1431 		struct rtable *orig;
1432 		int genid = fnhe_genid(dev_net(rt->dst.dev));
1433 
1434 		if (rt_is_input_route(rt))
1435 			porig = &fnhe->fnhe_rth_input;
1436 		else
1437 			porig = &fnhe->fnhe_rth_output;
1438 		orig = rcu_dereference(*porig);
1439 
1440 		if (fnhe->fnhe_genid != genid) {
1441 			fnhe->fnhe_genid = genid;
1442 			fnhe->fnhe_gw = 0;
1443 			fnhe->fnhe_pmtu = 0;
1444 			fnhe->fnhe_expires = 0;
1445 			fnhe->fnhe_mtu_locked = false;
1446 			fnhe_flush_routes(fnhe);
1447 			orig = NULL;
1448 		}
1449 		fill_route_from_fnhe(rt, fnhe);
1450 		if (!rt->rt_gw4) {
1451 			rt->rt_gw4 = daddr;
1452 			rt->rt_gw_family = AF_INET;
1453 		}
1454 
1455 		if (do_cache) {
1456 			dst_hold(&rt->dst);
1457 			rcu_assign_pointer(*porig, rt);
1458 			if (orig) {
1459 				dst_dev_put(&orig->dst);
1460 				dst_release(&orig->dst);
1461 			}
1462 			ret = true;
1463 		}
1464 
1465 		fnhe->fnhe_stamp = jiffies;
1466 	}
1467 	spin_unlock_bh(&fnhe_lock);
1468 
1469 	return ret;
1470 }
1471 
1472 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1473 {
1474 	struct rtable *orig, *prev, **p;
1475 	bool ret = true;
1476 
1477 	if (rt_is_input_route(rt)) {
1478 		p = (struct rtable **)&nhc->nhc_rth_input;
1479 	} else {
1480 		p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1481 	}
1482 	orig = *p;
1483 
1484 	/* hold dst before doing cmpxchg() to avoid race condition
1485 	 * on this dst
1486 	 */
1487 	dst_hold(&rt->dst);
1488 	prev = cmpxchg(p, orig, rt);
1489 	if (prev == orig) {
1490 		if (orig) {
1491 			rt_add_uncached_list(orig);
1492 			dst_release(&orig->dst);
1493 		}
1494 	} else {
1495 		dst_release(&rt->dst);
1496 		ret = false;
1497 	}
1498 
1499 	return ret;
1500 }
1501 
1502 struct uncached_list {
1503 	spinlock_t		lock;
1504 	struct list_head	head;
1505 	struct list_head	quarantine;
1506 };
1507 
1508 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1509 
1510 void rt_add_uncached_list(struct rtable *rt)
1511 {
1512 	struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1513 
1514 	rt->rt_uncached_list = ul;
1515 
1516 	spin_lock_bh(&ul->lock);
1517 	list_add_tail(&rt->rt_uncached, &ul->head);
1518 	spin_unlock_bh(&ul->lock);
1519 }
1520 
1521 void rt_del_uncached_list(struct rtable *rt)
1522 {
1523 	if (!list_empty(&rt->rt_uncached)) {
1524 		struct uncached_list *ul = rt->rt_uncached_list;
1525 
1526 		spin_lock_bh(&ul->lock);
1527 		list_del_init(&rt->rt_uncached);
1528 		spin_unlock_bh(&ul->lock);
1529 	}
1530 }
1531 
1532 static void ipv4_dst_destroy(struct dst_entry *dst)
1533 {
1534 	struct rtable *rt = (struct rtable *)dst;
1535 
1536 	ip_dst_metrics_put(dst);
1537 	rt_del_uncached_list(rt);
1538 }
1539 
1540 void rt_flush_dev(struct net_device *dev)
1541 {
1542 	struct rtable *rt, *safe;
1543 	int cpu;
1544 
1545 	for_each_possible_cpu(cpu) {
1546 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1547 
1548 		if (list_empty(&ul->head))
1549 			continue;
1550 
1551 		spin_lock_bh(&ul->lock);
1552 		list_for_each_entry_safe(rt, safe, &ul->head, rt_uncached) {
1553 			if (rt->dst.dev != dev)
1554 				continue;
1555 			rt->dst.dev = blackhole_netdev;
1556 			dev_replace_track(dev, blackhole_netdev,
1557 					  &rt->dst.dev_tracker,
1558 					  GFP_ATOMIC);
1559 			list_move(&rt->rt_uncached, &ul->quarantine);
1560 		}
1561 		spin_unlock_bh(&ul->lock);
1562 	}
1563 }
1564 
1565 static bool rt_cache_valid(const struct rtable *rt)
1566 {
1567 	return	rt &&
1568 		rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1569 		!rt_is_expired(rt);
1570 }
1571 
1572 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1573 			   const struct fib_result *res,
1574 			   struct fib_nh_exception *fnhe,
1575 			   struct fib_info *fi, u16 type, u32 itag,
1576 			   const bool do_cache)
1577 {
1578 	bool cached = false;
1579 
1580 	if (fi) {
1581 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1582 
1583 		if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1584 			rt->rt_uses_gateway = 1;
1585 			rt->rt_gw_family = nhc->nhc_gw_family;
1586 			/* only INET and INET6 are supported */
1587 			if (likely(nhc->nhc_gw_family == AF_INET))
1588 				rt->rt_gw4 = nhc->nhc_gw.ipv4;
1589 			else
1590 				rt->rt_gw6 = nhc->nhc_gw.ipv6;
1591 		}
1592 
1593 		ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1594 
1595 #ifdef CONFIG_IP_ROUTE_CLASSID
1596 		if (nhc->nhc_family == AF_INET) {
1597 			struct fib_nh *nh;
1598 
1599 			nh = container_of(nhc, struct fib_nh, nh_common);
1600 			rt->dst.tclassid = nh->nh_tclassid;
1601 		}
1602 #endif
1603 		rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1604 		if (unlikely(fnhe))
1605 			cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1606 		else if (do_cache)
1607 			cached = rt_cache_route(nhc, rt);
1608 		if (unlikely(!cached)) {
1609 			/* Routes we intend to cache in nexthop exception or
1610 			 * FIB nexthop have the DST_NOCACHE bit clear.
1611 			 * However, if we are unsuccessful at storing this
1612 			 * route into the cache we really need to set it.
1613 			 */
1614 			if (!rt->rt_gw4) {
1615 				rt->rt_gw_family = AF_INET;
1616 				rt->rt_gw4 = daddr;
1617 			}
1618 			rt_add_uncached_list(rt);
1619 		}
1620 	} else
1621 		rt_add_uncached_list(rt);
1622 
1623 #ifdef CONFIG_IP_ROUTE_CLASSID
1624 #ifdef CONFIG_IP_MULTIPLE_TABLES
1625 	set_class_tag(rt, res->tclassid);
1626 #endif
1627 	set_class_tag(rt, itag);
1628 #endif
1629 }
1630 
1631 struct rtable *rt_dst_alloc(struct net_device *dev,
1632 			    unsigned int flags, u16 type,
1633 			    bool nopolicy, bool noxfrm)
1634 {
1635 	struct rtable *rt;
1636 
1637 	rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1638 		       (nopolicy ? DST_NOPOLICY : 0) |
1639 		       (noxfrm ? DST_NOXFRM : 0));
1640 
1641 	if (rt) {
1642 		rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1643 		rt->rt_flags = flags;
1644 		rt->rt_type = type;
1645 		rt->rt_is_input = 0;
1646 		rt->rt_iif = 0;
1647 		rt->rt_pmtu = 0;
1648 		rt->rt_mtu_locked = 0;
1649 		rt->rt_uses_gateway = 0;
1650 		rt->rt_gw_family = 0;
1651 		rt->rt_gw4 = 0;
1652 		INIT_LIST_HEAD(&rt->rt_uncached);
1653 
1654 		rt->dst.output = ip_output;
1655 		if (flags & RTCF_LOCAL)
1656 			rt->dst.input = ip_local_deliver;
1657 	}
1658 
1659 	return rt;
1660 }
1661 EXPORT_SYMBOL(rt_dst_alloc);
1662 
1663 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1664 {
1665 	struct rtable *new_rt;
1666 
1667 	new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1668 			   rt->dst.flags);
1669 
1670 	if (new_rt) {
1671 		new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1672 		new_rt->rt_flags = rt->rt_flags;
1673 		new_rt->rt_type = rt->rt_type;
1674 		new_rt->rt_is_input = rt->rt_is_input;
1675 		new_rt->rt_iif = rt->rt_iif;
1676 		new_rt->rt_pmtu = rt->rt_pmtu;
1677 		new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1678 		new_rt->rt_gw_family = rt->rt_gw_family;
1679 		if (rt->rt_gw_family == AF_INET)
1680 			new_rt->rt_gw4 = rt->rt_gw4;
1681 		else if (rt->rt_gw_family == AF_INET6)
1682 			new_rt->rt_gw6 = rt->rt_gw6;
1683 		INIT_LIST_HEAD(&new_rt->rt_uncached);
1684 
1685 		new_rt->dst.input = rt->dst.input;
1686 		new_rt->dst.output = rt->dst.output;
1687 		new_rt->dst.error = rt->dst.error;
1688 		new_rt->dst.lastuse = jiffies;
1689 		new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1690 	}
1691 	return new_rt;
1692 }
1693 EXPORT_SYMBOL(rt_dst_clone);
1694 
1695 /* called in rcu_read_lock() section */
1696 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1697 			  u8 tos, struct net_device *dev,
1698 			  struct in_device *in_dev, u32 *itag)
1699 {
1700 	int err;
1701 
1702 	/* Primary sanity checks. */
1703 	if (!in_dev)
1704 		return -EINVAL;
1705 
1706 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1707 	    skb->protocol != htons(ETH_P_IP))
1708 		return -EINVAL;
1709 
1710 	if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1711 		return -EINVAL;
1712 
1713 	if (ipv4_is_zeronet(saddr)) {
1714 		if (!ipv4_is_local_multicast(daddr) &&
1715 		    ip_hdr(skb)->protocol != IPPROTO_IGMP)
1716 			return -EINVAL;
1717 	} else {
1718 		err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1719 					  in_dev, itag);
1720 		if (err < 0)
1721 			return err;
1722 	}
1723 	return 0;
1724 }
1725 
1726 /* called in rcu_read_lock() section */
1727 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1728 			     u8 tos, struct net_device *dev, int our)
1729 {
1730 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1731 	unsigned int flags = RTCF_MULTICAST;
1732 	struct rtable *rth;
1733 	u32 itag = 0;
1734 	int err;
1735 
1736 	err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1737 	if (err)
1738 		return err;
1739 
1740 	if (our)
1741 		flags |= RTCF_LOCAL;
1742 
1743 	rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1744 			   IN_DEV_ORCONF(in_dev, NOPOLICY), false);
1745 	if (!rth)
1746 		return -ENOBUFS;
1747 
1748 #ifdef CONFIG_IP_ROUTE_CLASSID
1749 	rth->dst.tclassid = itag;
1750 #endif
1751 	rth->dst.output = ip_rt_bug;
1752 	rth->rt_is_input= 1;
1753 
1754 #ifdef CONFIG_IP_MROUTE
1755 	if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1756 		rth->dst.input = ip_mr_input;
1757 #endif
1758 	RT_CACHE_STAT_INC(in_slow_mc);
1759 
1760 	skb_dst_set(skb, &rth->dst);
1761 	return 0;
1762 }
1763 
1764 
1765 static void ip_handle_martian_source(struct net_device *dev,
1766 				     struct in_device *in_dev,
1767 				     struct sk_buff *skb,
1768 				     __be32 daddr,
1769 				     __be32 saddr)
1770 {
1771 	RT_CACHE_STAT_INC(in_martian_src);
1772 #ifdef CONFIG_IP_ROUTE_VERBOSE
1773 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1774 		/*
1775 		 *	RFC1812 recommendation, if source is martian,
1776 		 *	the only hint is MAC header.
1777 		 */
1778 		pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1779 			&daddr, &saddr, dev->name);
1780 		if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1781 			print_hex_dump(KERN_WARNING, "ll header: ",
1782 				       DUMP_PREFIX_OFFSET, 16, 1,
1783 				       skb_mac_header(skb),
1784 				       dev->hard_header_len, false);
1785 		}
1786 	}
1787 #endif
1788 }
1789 
1790 /* called in rcu_read_lock() section */
1791 static int __mkroute_input(struct sk_buff *skb,
1792 			   const struct fib_result *res,
1793 			   struct in_device *in_dev,
1794 			   __be32 daddr, __be32 saddr, u32 tos)
1795 {
1796 	struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1797 	struct net_device *dev = nhc->nhc_dev;
1798 	struct fib_nh_exception *fnhe;
1799 	struct rtable *rth;
1800 	int err;
1801 	struct in_device *out_dev;
1802 	bool do_cache;
1803 	u32 itag = 0;
1804 
1805 	/* get a working reference to the output device */
1806 	out_dev = __in_dev_get_rcu(dev);
1807 	if (!out_dev) {
1808 		net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1809 		return -EINVAL;
1810 	}
1811 
1812 	err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1813 				  in_dev->dev, in_dev, &itag);
1814 	if (err < 0) {
1815 		ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1816 					 saddr);
1817 
1818 		goto cleanup;
1819 	}
1820 
1821 	do_cache = res->fi && !itag;
1822 	if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1823 	    skb->protocol == htons(ETH_P_IP)) {
1824 		__be32 gw;
1825 
1826 		gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1827 		if (IN_DEV_SHARED_MEDIA(out_dev) ||
1828 		    inet_addr_onlink(out_dev, saddr, gw))
1829 			IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1830 	}
1831 
1832 	if (skb->protocol != htons(ETH_P_IP)) {
1833 		/* Not IP (i.e. ARP). Do not create route, if it is
1834 		 * invalid for proxy arp. DNAT routes are always valid.
1835 		 *
1836 		 * Proxy arp feature have been extended to allow, ARP
1837 		 * replies back to the same interface, to support
1838 		 * Private VLAN switch technologies. See arp.c.
1839 		 */
1840 		if (out_dev == in_dev &&
1841 		    IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1842 			err = -EINVAL;
1843 			goto cleanup;
1844 		}
1845 	}
1846 
1847 	fnhe = find_exception(nhc, daddr);
1848 	if (do_cache) {
1849 		if (fnhe)
1850 			rth = rcu_dereference(fnhe->fnhe_rth_input);
1851 		else
1852 			rth = rcu_dereference(nhc->nhc_rth_input);
1853 		if (rt_cache_valid(rth)) {
1854 			skb_dst_set_noref(skb, &rth->dst);
1855 			goto out;
1856 		}
1857 	}
1858 
1859 	rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1860 			   IN_DEV_ORCONF(in_dev, NOPOLICY),
1861 			   IN_DEV_ORCONF(out_dev, NOXFRM));
1862 	if (!rth) {
1863 		err = -ENOBUFS;
1864 		goto cleanup;
1865 	}
1866 
1867 	rth->rt_is_input = 1;
1868 	RT_CACHE_STAT_INC(in_slow_tot);
1869 
1870 	rth->dst.input = ip_forward;
1871 
1872 	rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1873 		       do_cache);
1874 	lwtunnel_set_redirect(&rth->dst);
1875 	skb_dst_set(skb, &rth->dst);
1876 out:
1877 	err = 0;
1878  cleanup:
1879 	return err;
1880 }
1881 
1882 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1883 /* To make ICMP packets follow the right flow, the multipath hash is
1884  * calculated from the inner IP addresses.
1885  */
1886 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1887 				 struct flow_keys *hash_keys)
1888 {
1889 	const struct iphdr *outer_iph = ip_hdr(skb);
1890 	const struct iphdr *key_iph = outer_iph;
1891 	const struct iphdr *inner_iph;
1892 	const struct icmphdr *icmph;
1893 	struct iphdr _inner_iph;
1894 	struct icmphdr _icmph;
1895 
1896 	if (likely(outer_iph->protocol != IPPROTO_ICMP))
1897 		goto out;
1898 
1899 	if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1900 		goto out;
1901 
1902 	icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1903 				   &_icmph);
1904 	if (!icmph)
1905 		goto out;
1906 
1907 	if (!icmp_is_err(icmph->type))
1908 		goto out;
1909 
1910 	inner_iph = skb_header_pointer(skb,
1911 				       outer_iph->ihl * 4 + sizeof(_icmph),
1912 				       sizeof(_inner_iph), &_inner_iph);
1913 	if (!inner_iph)
1914 		goto out;
1915 
1916 	key_iph = inner_iph;
1917 out:
1918 	hash_keys->addrs.v4addrs.src = key_iph->saddr;
1919 	hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1920 }
1921 
1922 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1923 					   const struct sk_buff *skb,
1924 					   bool *p_has_inner)
1925 {
1926 	u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
1927 	struct flow_keys keys, hash_keys;
1928 
1929 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1930 		return 0;
1931 
1932 	memset(&hash_keys, 0, sizeof(hash_keys));
1933 	skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1934 
1935 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1936 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1937 		hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1938 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1939 		hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1940 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1941 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
1942 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1943 		hash_keys.ports.src = keys.ports.src;
1944 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1945 		hash_keys.ports.dst = keys.ports.dst;
1946 
1947 	*p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1948 	return flow_hash_from_keys(&hash_keys);
1949 }
1950 
1951 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1952 					   const struct sk_buff *skb,
1953 					   bool has_inner)
1954 {
1955 	u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
1956 	struct flow_keys keys, hash_keys;
1957 
1958 	/* We assume the packet carries an encapsulation, but if none was
1959 	 * encountered during dissection of the outer flow, then there is no
1960 	 * point in calling the flow dissector again.
1961 	 */
1962 	if (!has_inner)
1963 		return 0;
1964 
1965 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1966 		return 0;
1967 
1968 	memset(&hash_keys, 0, sizeof(hash_keys));
1969 	skb_flow_dissect_flow_keys(skb, &keys, 0);
1970 
1971 	if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1972 		return 0;
1973 
1974 	if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1975 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1976 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1977 			hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1978 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1979 			hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1980 	} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1981 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1982 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1983 			hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1984 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1985 			hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1986 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1987 			hash_keys.tags.flow_label = keys.tags.flow_label;
1988 	}
1989 
1990 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1991 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
1992 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1993 		hash_keys.ports.src = keys.ports.src;
1994 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
1995 		hash_keys.ports.dst = keys.ports.dst;
1996 
1997 	return flow_hash_from_keys(&hash_keys);
1998 }
1999 
2000 static u32 fib_multipath_custom_hash_skb(const struct net *net,
2001 					 const struct sk_buff *skb)
2002 {
2003 	u32 mhash, mhash_inner;
2004 	bool has_inner = true;
2005 
2006 	mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2007 	mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2008 
2009 	return jhash_2words(mhash, mhash_inner, 0);
2010 }
2011 
2012 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2013 					 const struct flowi4 *fl4)
2014 {
2015 	u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
2016 	struct flow_keys hash_keys;
2017 
2018 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2019 		return 0;
2020 
2021 	memset(&hash_keys, 0, sizeof(hash_keys));
2022 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2023 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2024 		hash_keys.addrs.v4addrs.src = fl4->saddr;
2025 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2026 		hash_keys.addrs.v4addrs.dst = fl4->daddr;
2027 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2028 		hash_keys.basic.ip_proto = fl4->flowi4_proto;
2029 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2030 		hash_keys.ports.src = fl4->fl4_sport;
2031 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2032 		hash_keys.ports.dst = fl4->fl4_dport;
2033 
2034 	return flow_hash_from_keys(&hash_keys);
2035 }
2036 
2037 /* if skb is set it will be used and fl4 can be NULL */
2038 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2039 		       const struct sk_buff *skb, struct flow_keys *flkeys)
2040 {
2041 	u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2042 	struct flow_keys hash_keys;
2043 	u32 mhash = 0;
2044 
2045 	switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
2046 	case 0:
2047 		memset(&hash_keys, 0, sizeof(hash_keys));
2048 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2049 		if (skb) {
2050 			ip_multipath_l3_keys(skb, &hash_keys);
2051 		} else {
2052 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2053 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2054 		}
2055 		mhash = flow_hash_from_keys(&hash_keys);
2056 		break;
2057 	case 1:
2058 		/* skb is currently provided only when forwarding */
2059 		if (skb) {
2060 			unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2061 			struct flow_keys keys;
2062 
2063 			/* short-circuit if we already have L4 hash present */
2064 			if (skb->l4_hash)
2065 				return skb_get_hash_raw(skb) >> 1;
2066 
2067 			memset(&hash_keys, 0, sizeof(hash_keys));
2068 
2069 			if (!flkeys) {
2070 				skb_flow_dissect_flow_keys(skb, &keys, flag);
2071 				flkeys = &keys;
2072 			}
2073 
2074 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2075 			hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2076 			hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2077 			hash_keys.ports.src = flkeys->ports.src;
2078 			hash_keys.ports.dst = flkeys->ports.dst;
2079 			hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2080 		} else {
2081 			memset(&hash_keys, 0, sizeof(hash_keys));
2082 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2083 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2084 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2085 			hash_keys.ports.src = fl4->fl4_sport;
2086 			hash_keys.ports.dst = fl4->fl4_dport;
2087 			hash_keys.basic.ip_proto = fl4->flowi4_proto;
2088 		}
2089 		mhash = flow_hash_from_keys(&hash_keys);
2090 		break;
2091 	case 2:
2092 		memset(&hash_keys, 0, sizeof(hash_keys));
2093 		/* skb is currently provided only when forwarding */
2094 		if (skb) {
2095 			struct flow_keys keys;
2096 
2097 			skb_flow_dissect_flow_keys(skb, &keys, 0);
2098 			/* Inner can be v4 or v6 */
2099 			if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2100 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2101 				hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2102 				hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2103 			} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2104 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2105 				hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2106 				hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2107 				hash_keys.tags.flow_label = keys.tags.flow_label;
2108 				hash_keys.basic.ip_proto = keys.basic.ip_proto;
2109 			} else {
2110 				/* Same as case 0 */
2111 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2112 				ip_multipath_l3_keys(skb, &hash_keys);
2113 			}
2114 		} else {
2115 			/* Same as case 0 */
2116 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2117 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2118 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2119 		}
2120 		mhash = flow_hash_from_keys(&hash_keys);
2121 		break;
2122 	case 3:
2123 		if (skb)
2124 			mhash = fib_multipath_custom_hash_skb(net, skb);
2125 		else
2126 			mhash = fib_multipath_custom_hash_fl4(net, fl4);
2127 		break;
2128 	}
2129 
2130 	if (multipath_hash)
2131 		mhash = jhash_2words(mhash, multipath_hash, 0);
2132 
2133 	return mhash >> 1;
2134 }
2135 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2136 
2137 static int ip_mkroute_input(struct sk_buff *skb,
2138 			    struct fib_result *res,
2139 			    struct in_device *in_dev,
2140 			    __be32 daddr, __be32 saddr, u32 tos,
2141 			    struct flow_keys *hkeys)
2142 {
2143 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2144 	if (res->fi && fib_info_num_path(res->fi) > 1) {
2145 		int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2146 
2147 		fib_select_multipath(res, h);
2148 	}
2149 #endif
2150 
2151 	/* create a routing cache entry */
2152 	return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2153 }
2154 
2155 /* Implements all the saddr-related checks as ip_route_input_slow(),
2156  * assuming daddr is valid and the destination is not a local broadcast one.
2157  * Uses the provided hint instead of performing a route lookup.
2158  */
2159 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2160 		      u8 tos, struct net_device *dev,
2161 		      const struct sk_buff *hint)
2162 {
2163 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2164 	struct rtable *rt = skb_rtable(hint);
2165 	struct net *net = dev_net(dev);
2166 	int err = -EINVAL;
2167 	u32 tag = 0;
2168 
2169 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2170 		goto martian_source;
2171 
2172 	if (ipv4_is_zeronet(saddr))
2173 		goto martian_source;
2174 
2175 	if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2176 		goto martian_source;
2177 
2178 	if (rt->rt_type != RTN_LOCAL)
2179 		goto skip_validate_source;
2180 
2181 	tos &= IPTOS_RT_MASK;
2182 	err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2183 	if (err < 0)
2184 		goto martian_source;
2185 
2186 skip_validate_source:
2187 	skb_dst_copy(skb, hint);
2188 	return 0;
2189 
2190 martian_source:
2191 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2192 	return err;
2193 }
2194 
2195 /* get device for dst_alloc with local routes */
2196 static struct net_device *ip_rt_get_dev(struct net *net,
2197 					const struct fib_result *res)
2198 {
2199 	struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2200 	struct net_device *dev = NULL;
2201 
2202 	if (nhc)
2203 		dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2204 
2205 	return dev ? : net->loopback_dev;
2206 }
2207 
2208 /*
2209  *	NOTE. We drop all the packets that has local source
2210  *	addresses, because every properly looped back packet
2211  *	must have correct destination already attached by output routine.
2212  *	Changes in the enforced policies must be applied also to
2213  *	ip_route_use_hint().
2214  *
2215  *	Such approach solves two big problems:
2216  *	1. Not simplex devices are handled properly.
2217  *	2. IP spoofing attempts are filtered with 100% of guarantee.
2218  *	called with rcu_read_lock()
2219  */
2220 
2221 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2222 			       u8 tos, struct net_device *dev,
2223 			       struct fib_result *res)
2224 {
2225 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2226 	struct flow_keys *flkeys = NULL, _flkeys;
2227 	struct net    *net = dev_net(dev);
2228 	struct ip_tunnel_info *tun_info;
2229 	int		err = -EINVAL;
2230 	unsigned int	flags = 0;
2231 	u32		itag = 0;
2232 	struct rtable	*rth;
2233 	struct flowi4	fl4;
2234 	bool do_cache = true;
2235 
2236 	/* IP on this device is disabled. */
2237 
2238 	if (!in_dev)
2239 		goto out;
2240 
2241 	/* Check for the most weird martians, which can be not detected
2242 	 * by fib_lookup.
2243 	 */
2244 
2245 	tun_info = skb_tunnel_info(skb);
2246 	if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2247 		fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2248 	else
2249 		fl4.flowi4_tun_key.tun_id = 0;
2250 	skb_dst_drop(skb);
2251 
2252 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2253 		goto martian_source;
2254 
2255 	res->fi = NULL;
2256 	res->table = NULL;
2257 	if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2258 		goto brd_input;
2259 
2260 	/* Accept zero addresses only to limited broadcast;
2261 	 * I even do not know to fix it or not. Waiting for complains :-)
2262 	 */
2263 	if (ipv4_is_zeronet(saddr))
2264 		goto martian_source;
2265 
2266 	if (ipv4_is_zeronet(daddr))
2267 		goto martian_destination;
2268 
2269 	/* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2270 	 * and call it once if daddr or/and saddr are loopback addresses
2271 	 */
2272 	if (ipv4_is_loopback(daddr)) {
2273 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2274 			goto martian_destination;
2275 	} else if (ipv4_is_loopback(saddr)) {
2276 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2277 			goto martian_source;
2278 	}
2279 
2280 	/*
2281 	 *	Now we are ready to route packet.
2282 	 */
2283 	fl4.flowi4_l3mdev = 0;
2284 	fl4.flowi4_oif = 0;
2285 	fl4.flowi4_iif = dev->ifindex;
2286 	fl4.flowi4_mark = skb->mark;
2287 	fl4.flowi4_tos = tos;
2288 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2289 	fl4.flowi4_flags = 0;
2290 	fl4.daddr = daddr;
2291 	fl4.saddr = saddr;
2292 	fl4.flowi4_uid = sock_net_uid(net, NULL);
2293 	fl4.flowi4_multipath_hash = 0;
2294 
2295 	if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2296 		flkeys = &_flkeys;
2297 	} else {
2298 		fl4.flowi4_proto = 0;
2299 		fl4.fl4_sport = 0;
2300 		fl4.fl4_dport = 0;
2301 	}
2302 
2303 	err = fib_lookup(net, &fl4, res, 0);
2304 	if (err != 0) {
2305 		if (!IN_DEV_FORWARD(in_dev))
2306 			err = -EHOSTUNREACH;
2307 		goto no_route;
2308 	}
2309 
2310 	if (res->type == RTN_BROADCAST) {
2311 		if (IN_DEV_BFORWARD(in_dev))
2312 			goto make_route;
2313 		/* not do cache if bc_forwarding is enabled */
2314 		if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2315 			do_cache = false;
2316 		goto brd_input;
2317 	}
2318 
2319 	if (res->type == RTN_LOCAL) {
2320 		err = fib_validate_source(skb, saddr, daddr, tos,
2321 					  0, dev, in_dev, &itag);
2322 		if (err < 0)
2323 			goto martian_source;
2324 		goto local_input;
2325 	}
2326 
2327 	if (!IN_DEV_FORWARD(in_dev)) {
2328 		err = -EHOSTUNREACH;
2329 		goto no_route;
2330 	}
2331 	if (res->type != RTN_UNICAST)
2332 		goto martian_destination;
2333 
2334 make_route:
2335 	err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2336 out:	return err;
2337 
2338 brd_input:
2339 	if (skb->protocol != htons(ETH_P_IP))
2340 		goto e_inval;
2341 
2342 	if (!ipv4_is_zeronet(saddr)) {
2343 		err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2344 					  in_dev, &itag);
2345 		if (err < 0)
2346 			goto martian_source;
2347 	}
2348 	flags |= RTCF_BROADCAST;
2349 	res->type = RTN_BROADCAST;
2350 	RT_CACHE_STAT_INC(in_brd);
2351 
2352 local_input:
2353 	do_cache &= res->fi && !itag;
2354 	if (do_cache) {
2355 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2356 
2357 		rth = rcu_dereference(nhc->nhc_rth_input);
2358 		if (rt_cache_valid(rth)) {
2359 			skb_dst_set_noref(skb, &rth->dst);
2360 			err = 0;
2361 			goto out;
2362 		}
2363 	}
2364 
2365 	rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2366 			   flags | RTCF_LOCAL, res->type,
2367 			   IN_DEV_ORCONF(in_dev, NOPOLICY), false);
2368 	if (!rth)
2369 		goto e_nobufs;
2370 
2371 	rth->dst.output= ip_rt_bug;
2372 #ifdef CONFIG_IP_ROUTE_CLASSID
2373 	rth->dst.tclassid = itag;
2374 #endif
2375 	rth->rt_is_input = 1;
2376 
2377 	RT_CACHE_STAT_INC(in_slow_tot);
2378 	if (res->type == RTN_UNREACHABLE) {
2379 		rth->dst.input= ip_error;
2380 		rth->dst.error= -err;
2381 		rth->rt_flags	&= ~RTCF_LOCAL;
2382 	}
2383 
2384 	if (do_cache) {
2385 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2386 
2387 		rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2388 		if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2389 			WARN_ON(rth->dst.input == lwtunnel_input);
2390 			rth->dst.lwtstate->orig_input = rth->dst.input;
2391 			rth->dst.input = lwtunnel_input;
2392 		}
2393 
2394 		if (unlikely(!rt_cache_route(nhc, rth)))
2395 			rt_add_uncached_list(rth);
2396 	}
2397 	skb_dst_set(skb, &rth->dst);
2398 	err = 0;
2399 	goto out;
2400 
2401 no_route:
2402 	RT_CACHE_STAT_INC(in_no_route);
2403 	res->type = RTN_UNREACHABLE;
2404 	res->fi = NULL;
2405 	res->table = NULL;
2406 	goto local_input;
2407 
2408 	/*
2409 	 *	Do not cache martian addresses: they should be logged (RFC1812)
2410 	 */
2411 martian_destination:
2412 	RT_CACHE_STAT_INC(in_martian_dst);
2413 #ifdef CONFIG_IP_ROUTE_VERBOSE
2414 	if (IN_DEV_LOG_MARTIANS(in_dev))
2415 		net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2416 				     &daddr, &saddr, dev->name);
2417 #endif
2418 
2419 e_inval:
2420 	err = -EINVAL;
2421 	goto out;
2422 
2423 e_nobufs:
2424 	err = -ENOBUFS;
2425 	goto out;
2426 
2427 martian_source:
2428 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2429 	goto out;
2430 }
2431 
2432 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2433 			 u8 tos, struct net_device *dev)
2434 {
2435 	struct fib_result res;
2436 	int err;
2437 
2438 	tos &= IPTOS_RT_MASK;
2439 	rcu_read_lock();
2440 	err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2441 	rcu_read_unlock();
2442 
2443 	return err;
2444 }
2445 EXPORT_SYMBOL(ip_route_input_noref);
2446 
2447 /* called with rcu_read_lock held */
2448 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2449 		       u8 tos, struct net_device *dev, struct fib_result *res)
2450 {
2451 	/* Multicast recognition logic is moved from route cache to here.
2452 	 * The problem was that too many Ethernet cards have broken/missing
2453 	 * hardware multicast filters :-( As result the host on multicasting
2454 	 * network acquires a lot of useless route cache entries, sort of
2455 	 * SDR messages from all the world. Now we try to get rid of them.
2456 	 * Really, provided software IP multicast filter is organized
2457 	 * reasonably (at least, hashed), it does not result in a slowdown
2458 	 * comparing with route cache reject entries.
2459 	 * Note, that multicast routers are not affected, because
2460 	 * route cache entry is created eventually.
2461 	 */
2462 	if (ipv4_is_multicast(daddr)) {
2463 		struct in_device *in_dev = __in_dev_get_rcu(dev);
2464 		int our = 0;
2465 		int err = -EINVAL;
2466 
2467 		if (!in_dev)
2468 			return err;
2469 		our = ip_check_mc_rcu(in_dev, daddr, saddr,
2470 				      ip_hdr(skb)->protocol);
2471 
2472 		/* check l3 master if no match yet */
2473 		if (!our && netif_is_l3_slave(dev)) {
2474 			struct in_device *l3_in_dev;
2475 
2476 			l3_in_dev = __in_dev_get_rcu(skb->dev);
2477 			if (l3_in_dev)
2478 				our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2479 						      ip_hdr(skb)->protocol);
2480 		}
2481 
2482 		if (our
2483 #ifdef CONFIG_IP_MROUTE
2484 			||
2485 		    (!ipv4_is_local_multicast(daddr) &&
2486 		     IN_DEV_MFORWARD(in_dev))
2487 #endif
2488 		   ) {
2489 			err = ip_route_input_mc(skb, daddr, saddr,
2490 						tos, dev, our);
2491 		}
2492 		return err;
2493 	}
2494 
2495 	return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2496 }
2497 
2498 /* called with rcu_read_lock() */
2499 static struct rtable *__mkroute_output(const struct fib_result *res,
2500 				       const struct flowi4 *fl4, int orig_oif,
2501 				       struct net_device *dev_out,
2502 				       unsigned int flags)
2503 {
2504 	struct fib_info *fi = res->fi;
2505 	struct fib_nh_exception *fnhe;
2506 	struct in_device *in_dev;
2507 	u16 type = res->type;
2508 	struct rtable *rth;
2509 	bool do_cache;
2510 
2511 	in_dev = __in_dev_get_rcu(dev_out);
2512 	if (!in_dev)
2513 		return ERR_PTR(-EINVAL);
2514 
2515 	if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2516 		if (ipv4_is_loopback(fl4->saddr) &&
2517 		    !(dev_out->flags & IFF_LOOPBACK) &&
2518 		    !netif_is_l3_master(dev_out))
2519 			return ERR_PTR(-EINVAL);
2520 
2521 	if (ipv4_is_lbcast(fl4->daddr))
2522 		type = RTN_BROADCAST;
2523 	else if (ipv4_is_multicast(fl4->daddr))
2524 		type = RTN_MULTICAST;
2525 	else if (ipv4_is_zeronet(fl4->daddr))
2526 		return ERR_PTR(-EINVAL);
2527 
2528 	if (dev_out->flags & IFF_LOOPBACK)
2529 		flags |= RTCF_LOCAL;
2530 
2531 	do_cache = true;
2532 	if (type == RTN_BROADCAST) {
2533 		flags |= RTCF_BROADCAST | RTCF_LOCAL;
2534 		fi = NULL;
2535 	} else if (type == RTN_MULTICAST) {
2536 		flags |= RTCF_MULTICAST | RTCF_LOCAL;
2537 		if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2538 				     fl4->flowi4_proto))
2539 			flags &= ~RTCF_LOCAL;
2540 		else
2541 			do_cache = false;
2542 		/* If multicast route do not exist use
2543 		 * default one, but do not gateway in this case.
2544 		 * Yes, it is hack.
2545 		 */
2546 		if (fi && res->prefixlen < 4)
2547 			fi = NULL;
2548 	} else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2549 		   (orig_oif != dev_out->ifindex)) {
2550 		/* For local routes that require a particular output interface
2551 		 * we do not want to cache the result.  Caching the result
2552 		 * causes incorrect behaviour when there are multiple source
2553 		 * addresses on the interface, the end result being that if the
2554 		 * intended recipient is waiting on that interface for the
2555 		 * packet he won't receive it because it will be delivered on
2556 		 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2557 		 * be set to the loopback interface as well.
2558 		 */
2559 		do_cache = false;
2560 	}
2561 
2562 	fnhe = NULL;
2563 	do_cache &= fi != NULL;
2564 	if (fi) {
2565 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2566 		struct rtable __rcu **prth;
2567 
2568 		fnhe = find_exception(nhc, fl4->daddr);
2569 		if (!do_cache)
2570 			goto add;
2571 		if (fnhe) {
2572 			prth = &fnhe->fnhe_rth_output;
2573 		} else {
2574 			if (unlikely(fl4->flowi4_flags &
2575 				     FLOWI_FLAG_KNOWN_NH &&
2576 				     !(nhc->nhc_gw_family &&
2577 				       nhc->nhc_scope == RT_SCOPE_LINK))) {
2578 				do_cache = false;
2579 				goto add;
2580 			}
2581 			prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2582 		}
2583 		rth = rcu_dereference(*prth);
2584 		if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2585 			return rth;
2586 	}
2587 
2588 add:
2589 	rth = rt_dst_alloc(dev_out, flags, type,
2590 			   IN_DEV_ORCONF(in_dev, NOPOLICY),
2591 			   IN_DEV_ORCONF(in_dev, NOXFRM));
2592 	if (!rth)
2593 		return ERR_PTR(-ENOBUFS);
2594 
2595 	rth->rt_iif = orig_oif;
2596 
2597 	RT_CACHE_STAT_INC(out_slow_tot);
2598 
2599 	if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2600 		if (flags & RTCF_LOCAL &&
2601 		    !(dev_out->flags & IFF_LOOPBACK)) {
2602 			rth->dst.output = ip_mc_output;
2603 			RT_CACHE_STAT_INC(out_slow_mc);
2604 		}
2605 #ifdef CONFIG_IP_MROUTE
2606 		if (type == RTN_MULTICAST) {
2607 			if (IN_DEV_MFORWARD(in_dev) &&
2608 			    !ipv4_is_local_multicast(fl4->daddr)) {
2609 				rth->dst.input = ip_mr_input;
2610 				rth->dst.output = ip_mc_output;
2611 			}
2612 		}
2613 #endif
2614 	}
2615 
2616 	rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2617 	lwtunnel_set_redirect(&rth->dst);
2618 
2619 	return rth;
2620 }
2621 
2622 /*
2623  * Major route resolver routine.
2624  */
2625 
2626 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2627 					const struct sk_buff *skb)
2628 {
2629 	struct fib_result res = {
2630 		.type		= RTN_UNSPEC,
2631 		.fi		= NULL,
2632 		.table		= NULL,
2633 		.tclassid	= 0,
2634 	};
2635 	struct rtable *rth;
2636 
2637 	fl4->flowi4_iif = LOOPBACK_IFINDEX;
2638 	ip_rt_fix_tos(fl4);
2639 
2640 	rcu_read_lock();
2641 	rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2642 	rcu_read_unlock();
2643 
2644 	return rth;
2645 }
2646 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2647 
2648 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2649 					    struct fib_result *res,
2650 					    const struct sk_buff *skb)
2651 {
2652 	struct net_device *dev_out = NULL;
2653 	int orig_oif = fl4->flowi4_oif;
2654 	unsigned int flags = 0;
2655 	struct rtable *rth;
2656 	int err;
2657 
2658 	if (fl4->saddr) {
2659 		if (ipv4_is_multicast(fl4->saddr) ||
2660 		    ipv4_is_lbcast(fl4->saddr) ||
2661 		    ipv4_is_zeronet(fl4->saddr)) {
2662 			rth = ERR_PTR(-EINVAL);
2663 			goto out;
2664 		}
2665 
2666 		rth = ERR_PTR(-ENETUNREACH);
2667 
2668 		/* I removed check for oif == dev_out->oif here.
2669 		 * It was wrong for two reasons:
2670 		 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2671 		 *    is assigned to multiple interfaces.
2672 		 * 2. Moreover, we are allowed to send packets with saddr
2673 		 *    of another iface. --ANK
2674 		 */
2675 
2676 		if (fl4->flowi4_oif == 0 &&
2677 		    (ipv4_is_multicast(fl4->daddr) ||
2678 		     ipv4_is_lbcast(fl4->daddr))) {
2679 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2680 			dev_out = __ip_dev_find(net, fl4->saddr, false);
2681 			if (!dev_out)
2682 				goto out;
2683 
2684 			/* Special hack: user can direct multicasts
2685 			 * and limited broadcast via necessary interface
2686 			 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2687 			 * This hack is not just for fun, it allows
2688 			 * vic,vat and friends to work.
2689 			 * They bind socket to loopback, set ttl to zero
2690 			 * and expect that it will work.
2691 			 * From the viewpoint of routing cache they are broken,
2692 			 * because we are not allowed to build multicast path
2693 			 * with loopback source addr (look, routing cache
2694 			 * cannot know, that ttl is zero, so that packet
2695 			 * will not leave this host and route is valid).
2696 			 * Luckily, this hack is good workaround.
2697 			 */
2698 
2699 			fl4->flowi4_oif = dev_out->ifindex;
2700 			goto make_route;
2701 		}
2702 
2703 		if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2704 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2705 			if (!__ip_dev_find(net, fl4->saddr, false))
2706 				goto out;
2707 		}
2708 	}
2709 
2710 
2711 	if (fl4->flowi4_oif) {
2712 		dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2713 		rth = ERR_PTR(-ENODEV);
2714 		if (!dev_out)
2715 			goto out;
2716 
2717 		/* RACE: Check return value of inet_select_addr instead. */
2718 		if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2719 			rth = ERR_PTR(-ENETUNREACH);
2720 			goto out;
2721 		}
2722 		if (ipv4_is_local_multicast(fl4->daddr) ||
2723 		    ipv4_is_lbcast(fl4->daddr) ||
2724 		    fl4->flowi4_proto == IPPROTO_IGMP) {
2725 			if (!fl4->saddr)
2726 				fl4->saddr = inet_select_addr(dev_out, 0,
2727 							      RT_SCOPE_LINK);
2728 			goto make_route;
2729 		}
2730 		if (!fl4->saddr) {
2731 			if (ipv4_is_multicast(fl4->daddr))
2732 				fl4->saddr = inet_select_addr(dev_out, 0,
2733 							      fl4->flowi4_scope);
2734 			else if (!fl4->daddr)
2735 				fl4->saddr = inet_select_addr(dev_out, 0,
2736 							      RT_SCOPE_HOST);
2737 		}
2738 	}
2739 
2740 	if (!fl4->daddr) {
2741 		fl4->daddr = fl4->saddr;
2742 		if (!fl4->daddr)
2743 			fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2744 		dev_out = net->loopback_dev;
2745 		fl4->flowi4_oif = LOOPBACK_IFINDEX;
2746 		res->type = RTN_LOCAL;
2747 		flags |= RTCF_LOCAL;
2748 		goto make_route;
2749 	}
2750 
2751 	err = fib_lookup(net, fl4, res, 0);
2752 	if (err) {
2753 		res->fi = NULL;
2754 		res->table = NULL;
2755 		if (fl4->flowi4_oif &&
2756 		    (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2757 			/* Apparently, routing tables are wrong. Assume,
2758 			 * that the destination is on link.
2759 			 *
2760 			 * WHY? DW.
2761 			 * Because we are allowed to send to iface
2762 			 * even if it has NO routes and NO assigned
2763 			 * addresses. When oif is specified, routing
2764 			 * tables are looked up with only one purpose:
2765 			 * to catch if destination is gatewayed, rather than
2766 			 * direct. Moreover, if MSG_DONTROUTE is set,
2767 			 * we send packet, ignoring both routing tables
2768 			 * and ifaddr state. --ANK
2769 			 *
2770 			 *
2771 			 * We could make it even if oif is unknown,
2772 			 * likely IPv6, but we do not.
2773 			 */
2774 
2775 			if (fl4->saddr == 0)
2776 				fl4->saddr = inet_select_addr(dev_out, 0,
2777 							      RT_SCOPE_LINK);
2778 			res->type = RTN_UNICAST;
2779 			goto make_route;
2780 		}
2781 		rth = ERR_PTR(err);
2782 		goto out;
2783 	}
2784 
2785 	if (res->type == RTN_LOCAL) {
2786 		if (!fl4->saddr) {
2787 			if (res->fi->fib_prefsrc)
2788 				fl4->saddr = res->fi->fib_prefsrc;
2789 			else
2790 				fl4->saddr = fl4->daddr;
2791 		}
2792 
2793 		/* L3 master device is the loopback for that domain */
2794 		dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2795 			net->loopback_dev;
2796 
2797 		/* make sure orig_oif points to fib result device even
2798 		 * though packet rx/tx happens over loopback or l3mdev
2799 		 */
2800 		orig_oif = FIB_RES_OIF(*res);
2801 
2802 		fl4->flowi4_oif = dev_out->ifindex;
2803 		flags |= RTCF_LOCAL;
2804 		goto make_route;
2805 	}
2806 
2807 	fib_select_path(net, res, fl4, skb);
2808 
2809 	dev_out = FIB_RES_DEV(*res);
2810 
2811 make_route:
2812 	rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2813 
2814 out:
2815 	return rth;
2816 }
2817 
2818 static struct dst_ops ipv4_dst_blackhole_ops = {
2819 	.family			= AF_INET,
2820 	.default_advmss		= ipv4_default_advmss,
2821 	.neigh_lookup		= ipv4_neigh_lookup,
2822 	.check			= dst_blackhole_check,
2823 	.cow_metrics		= dst_blackhole_cow_metrics,
2824 	.update_pmtu		= dst_blackhole_update_pmtu,
2825 	.redirect		= dst_blackhole_redirect,
2826 	.mtu			= dst_blackhole_mtu,
2827 };
2828 
2829 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2830 {
2831 	struct rtable *ort = (struct rtable *) dst_orig;
2832 	struct rtable *rt;
2833 
2834 	rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2835 	if (rt) {
2836 		struct dst_entry *new = &rt->dst;
2837 
2838 		new->__use = 1;
2839 		new->input = dst_discard;
2840 		new->output = dst_discard_out;
2841 
2842 		new->dev = net->loopback_dev;
2843 		dev_hold_track(new->dev, &new->dev_tracker, GFP_ATOMIC);
2844 
2845 		rt->rt_is_input = ort->rt_is_input;
2846 		rt->rt_iif = ort->rt_iif;
2847 		rt->rt_pmtu = ort->rt_pmtu;
2848 		rt->rt_mtu_locked = ort->rt_mtu_locked;
2849 
2850 		rt->rt_genid = rt_genid_ipv4(net);
2851 		rt->rt_flags = ort->rt_flags;
2852 		rt->rt_type = ort->rt_type;
2853 		rt->rt_uses_gateway = ort->rt_uses_gateway;
2854 		rt->rt_gw_family = ort->rt_gw_family;
2855 		if (rt->rt_gw_family == AF_INET)
2856 			rt->rt_gw4 = ort->rt_gw4;
2857 		else if (rt->rt_gw_family == AF_INET6)
2858 			rt->rt_gw6 = ort->rt_gw6;
2859 
2860 		INIT_LIST_HEAD(&rt->rt_uncached);
2861 	}
2862 
2863 	dst_release(dst_orig);
2864 
2865 	return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2866 }
2867 
2868 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2869 				    const struct sock *sk)
2870 {
2871 	struct rtable *rt = __ip_route_output_key(net, flp4);
2872 
2873 	if (IS_ERR(rt))
2874 		return rt;
2875 
2876 	if (flp4->flowi4_proto) {
2877 		flp4->flowi4_oif = rt->dst.dev->ifindex;
2878 		rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2879 							flowi4_to_flowi(flp4),
2880 							sk, 0);
2881 	}
2882 
2883 	return rt;
2884 }
2885 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2886 
2887 struct rtable *ip_route_output_tunnel(struct sk_buff *skb,
2888 				      struct net_device *dev,
2889 				      struct net *net, __be32 *saddr,
2890 				      const struct ip_tunnel_info *info,
2891 				      u8 protocol, bool use_cache)
2892 {
2893 #ifdef CONFIG_DST_CACHE
2894 	struct dst_cache *dst_cache;
2895 #endif
2896 	struct rtable *rt = NULL;
2897 	struct flowi4 fl4;
2898 	__u8 tos;
2899 
2900 #ifdef CONFIG_DST_CACHE
2901 	dst_cache = (struct dst_cache *)&info->dst_cache;
2902 	if (use_cache) {
2903 		rt = dst_cache_get_ip4(dst_cache, saddr);
2904 		if (rt)
2905 			return rt;
2906 	}
2907 #endif
2908 	memset(&fl4, 0, sizeof(fl4));
2909 	fl4.flowi4_mark = skb->mark;
2910 	fl4.flowi4_proto = protocol;
2911 	fl4.daddr = info->key.u.ipv4.dst;
2912 	fl4.saddr = info->key.u.ipv4.src;
2913 	tos = info->key.tos;
2914 	fl4.flowi4_tos = RT_TOS(tos);
2915 
2916 	rt = ip_route_output_key(net, &fl4);
2917 	if (IS_ERR(rt)) {
2918 		netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
2919 		return ERR_PTR(-ENETUNREACH);
2920 	}
2921 	if (rt->dst.dev == dev) { /* is this necessary? */
2922 		netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr);
2923 		ip_rt_put(rt);
2924 		return ERR_PTR(-ELOOP);
2925 	}
2926 #ifdef CONFIG_DST_CACHE
2927 	if (use_cache)
2928 		dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2929 #endif
2930 	*saddr = fl4.saddr;
2931 	return rt;
2932 }
2933 EXPORT_SYMBOL_GPL(ip_route_output_tunnel);
2934 
2935 /* called with rcu_read_lock held */
2936 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2937 			struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2938 			struct sk_buff *skb, u32 portid, u32 seq,
2939 			unsigned int flags)
2940 {
2941 	struct rtmsg *r;
2942 	struct nlmsghdr *nlh;
2943 	unsigned long expires = 0;
2944 	u32 error;
2945 	u32 metrics[RTAX_MAX];
2946 
2947 	nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2948 	if (!nlh)
2949 		return -EMSGSIZE;
2950 
2951 	r = nlmsg_data(nlh);
2952 	r->rtm_family	 = AF_INET;
2953 	r->rtm_dst_len	= 32;
2954 	r->rtm_src_len	= 0;
2955 	r->rtm_tos	= fl4 ? fl4->flowi4_tos : 0;
2956 	r->rtm_table	= table_id < 256 ? table_id : RT_TABLE_COMPAT;
2957 	if (nla_put_u32(skb, RTA_TABLE, table_id))
2958 		goto nla_put_failure;
2959 	r->rtm_type	= rt->rt_type;
2960 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2961 	r->rtm_protocol = RTPROT_UNSPEC;
2962 	r->rtm_flags	= (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2963 	if (rt->rt_flags & RTCF_NOTIFY)
2964 		r->rtm_flags |= RTM_F_NOTIFY;
2965 	if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2966 		r->rtm_flags |= RTCF_DOREDIRECT;
2967 
2968 	if (nla_put_in_addr(skb, RTA_DST, dst))
2969 		goto nla_put_failure;
2970 	if (src) {
2971 		r->rtm_src_len = 32;
2972 		if (nla_put_in_addr(skb, RTA_SRC, src))
2973 			goto nla_put_failure;
2974 	}
2975 	if (rt->dst.dev &&
2976 	    nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2977 		goto nla_put_failure;
2978 	if (rt->dst.lwtstate &&
2979 	    lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2980 		goto nla_put_failure;
2981 #ifdef CONFIG_IP_ROUTE_CLASSID
2982 	if (rt->dst.tclassid &&
2983 	    nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2984 		goto nla_put_failure;
2985 #endif
2986 	if (fl4 && !rt_is_input_route(rt) &&
2987 	    fl4->saddr != src) {
2988 		if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2989 			goto nla_put_failure;
2990 	}
2991 	if (rt->rt_uses_gateway) {
2992 		if (rt->rt_gw_family == AF_INET &&
2993 		    nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2994 			goto nla_put_failure;
2995 		} else if (rt->rt_gw_family == AF_INET6) {
2996 			int alen = sizeof(struct in6_addr);
2997 			struct nlattr *nla;
2998 			struct rtvia *via;
2999 
3000 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
3001 			if (!nla)
3002 				goto nla_put_failure;
3003 
3004 			via = nla_data(nla);
3005 			via->rtvia_family = AF_INET6;
3006 			memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
3007 		}
3008 	}
3009 
3010 	expires = rt->dst.expires;
3011 	if (expires) {
3012 		unsigned long now = jiffies;
3013 
3014 		if (time_before(now, expires))
3015 			expires -= now;
3016 		else
3017 			expires = 0;
3018 	}
3019 
3020 	memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3021 	if (rt->rt_pmtu && expires)
3022 		metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3023 	if (rt->rt_mtu_locked && expires)
3024 		metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3025 	if (rtnetlink_put_metrics(skb, metrics) < 0)
3026 		goto nla_put_failure;
3027 
3028 	if (fl4) {
3029 		if (fl4->flowi4_mark &&
3030 		    nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3031 			goto nla_put_failure;
3032 
3033 		if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3034 		    nla_put_u32(skb, RTA_UID,
3035 				from_kuid_munged(current_user_ns(),
3036 						 fl4->flowi4_uid)))
3037 			goto nla_put_failure;
3038 
3039 		if (rt_is_input_route(rt)) {
3040 #ifdef CONFIG_IP_MROUTE
3041 			if (ipv4_is_multicast(dst) &&
3042 			    !ipv4_is_local_multicast(dst) &&
3043 			    IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
3044 				int err = ipmr_get_route(net, skb,
3045 							 fl4->saddr, fl4->daddr,
3046 							 r, portid);
3047 
3048 				if (err <= 0) {
3049 					if (err == 0)
3050 						return 0;
3051 					goto nla_put_failure;
3052 				}
3053 			} else
3054 #endif
3055 				if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3056 					goto nla_put_failure;
3057 		}
3058 	}
3059 
3060 	error = rt->dst.error;
3061 
3062 	if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3063 		goto nla_put_failure;
3064 
3065 	nlmsg_end(skb, nlh);
3066 	return 0;
3067 
3068 nla_put_failure:
3069 	nlmsg_cancel(skb, nlh);
3070 	return -EMSGSIZE;
3071 }
3072 
3073 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3074 			    struct netlink_callback *cb, u32 table_id,
3075 			    struct fnhe_hash_bucket *bucket, int genid,
3076 			    int *fa_index, int fa_start, unsigned int flags)
3077 {
3078 	int i;
3079 
3080 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
3081 		struct fib_nh_exception *fnhe;
3082 
3083 		for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3084 		     fnhe = rcu_dereference(fnhe->fnhe_next)) {
3085 			struct rtable *rt;
3086 			int err;
3087 
3088 			if (*fa_index < fa_start)
3089 				goto next;
3090 
3091 			if (fnhe->fnhe_genid != genid)
3092 				goto next;
3093 
3094 			if (fnhe->fnhe_expires &&
3095 			    time_after(jiffies, fnhe->fnhe_expires))
3096 				goto next;
3097 
3098 			rt = rcu_dereference(fnhe->fnhe_rth_input);
3099 			if (!rt)
3100 				rt = rcu_dereference(fnhe->fnhe_rth_output);
3101 			if (!rt)
3102 				goto next;
3103 
3104 			err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3105 					   table_id, NULL, skb,
3106 					   NETLINK_CB(cb->skb).portid,
3107 					   cb->nlh->nlmsg_seq, flags);
3108 			if (err)
3109 				return err;
3110 next:
3111 			(*fa_index)++;
3112 		}
3113 	}
3114 
3115 	return 0;
3116 }
3117 
3118 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3119 		       u32 table_id, struct fib_info *fi,
3120 		       int *fa_index, int fa_start, unsigned int flags)
3121 {
3122 	struct net *net = sock_net(cb->skb->sk);
3123 	int nhsel, genid = fnhe_genid(net);
3124 
3125 	for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3126 		struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3127 		struct fnhe_hash_bucket *bucket;
3128 		int err;
3129 
3130 		if (nhc->nhc_flags & RTNH_F_DEAD)
3131 			continue;
3132 
3133 		rcu_read_lock();
3134 		bucket = rcu_dereference(nhc->nhc_exceptions);
3135 		err = 0;
3136 		if (bucket)
3137 			err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3138 					       genid, fa_index, fa_start,
3139 					       flags);
3140 		rcu_read_unlock();
3141 		if (err)
3142 			return err;
3143 	}
3144 
3145 	return 0;
3146 }
3147 
3148 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3149 						   u8 ip_proto, __be16 sport,
3150 						   __be16 dport)
3151 {
3152 	struct sk_buff *skb;
3153 	struct iphdr *iph;
3154 
3155 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3156 	if (!skb)
3157 		return NULL;
3158 
3159 	/* Reserve room for dummy headers, this skb can pass
3160 	 * through good chunk of routing engine.
3161 	 */
3162 	skb_reset_mac_header(skb);
3163 	skb_reset_network_header(skb);
3164 	skb->protocol = htons(ETH_P_IP);
3165 	iph = skb_put(skb, sizeof(struct iphdr));
3166 	iph->protocol = ip_proto;
3167 	iph->saddr = src;
3168 	iph->daddr = dst;
3169 	iph->version = 0x4;
3170 	iph->frag_off = 0;
3171 	iph->ihl = 0x5;
3172 	skb_set_transport_header(skb, skb->len);
3173 
3174 	switch (iph->protocol) {
3175 	case IPPROTO_UDP: {
3176 		struct udphdr *udph;
3177 
3178 		udph = skb_put_zero(skb, sizeof(struct udphdr));
3179 		udph->source = sport;
3180 		udph->dest = dport;
3181 		udph->len = htons(sizeof(struct udphdr));
3182 		udph->check = 0;
3183 		break;
3184 	}
3185 	case IPPROTO_TCP: {
3186 		struct tcphdr *tcph;
3187 
3188 		tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3189 		tcph->source	= sport;
3190 		tcph->dest	= dport;
3191 		tcph->doff	= sizeof(struct tcphdr) / 4;
3192 		tcph->rst = 1;
3193 		tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3194 					    src, dst, 0);
3195 		break;
3196 	}
3197 	case IPPROTO_ICMP: {
3198 		struct icmphdr *icmph;
3199 
3200 		icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3201 		icmph->type = ICMP_ECHO;
3202 		icmph->code = 0;
3203 	}
3204 	}
3205 
3206 	return skb;
3207 }
3208 
3209 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3210 				       const struct nlmsghdr *nlh,
3211 				       struct nlattr **tb,
3212 				       struct netlink_ext_ack *extack)
3213 {
3214 	struct rtmsg *rtm;
3215 	int i, err;
3216 
3217 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3218 		NL_SET_ERR_MSG(extack,
3219 			       "ipv4: Invalid header for route get request");
3220 		return -EINVAL;
3221 	}
3222 
3223 	if (!netlink_strict_get_check(skb))
3224 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3225 					      rtm_ipv4_policy, extack);
3226 
3227 	rtm = nlmsg_data(nlh);
3228 	if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3229 	    (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3230 	    rtm->rtm_table || rtm->rtm_protocol ||
3231 	    rtm->rtm_scope || rtm->rtm_type) {
3232 		NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3233 		return -EINVAL;
3234 	}
3235 
3236 	if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3237 			       RTM_F_LOOKUP_TABLE |
3238 			       RTM_F_FIB_MATCH)) {
3239 		NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3240 		return -EINVAL;
3241 	}
3242 
3243 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3244 					    rtm_ipv4_policy, extack);
3245 	if (err)
3246 		return err;
3247 
3248 	if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3249 	    (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3250 		NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3251 		return -EINVAL;
3252 	}
3253 
3254 	for (i = 0; i <= RTA_MAX; i++) {
3255 		if (!tb[i])
3256 			continue;
3257 
3258 		switch (i) {
3259 		case RTA_IIF:
3260 		case RTA_OIF:
3261 		case RTA_SRC:
3262 		case RTA_DST:
3263 		case RTA_IP_PROTO:
3264 		case RTA_SPORT:
3265 		case RTA_DPORT:
3266 		case RTA_MARK:
3267 		case RTA_UID:
3268 			break;
3269 		default:
3270 			NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3271 			return -EINVAL;
3272 		}
3273 	}
3274 
3275 	return 0;
3276 }
3277 
3278 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3279 			     struct netlink_ext_ack *extack)
3280 {
3281 	struct net *net = sock_net(in_skb->sk);
3282 	struct nlattr *tb[RTA_MAX+1];
3283 	u32 table_id = RT_TABLE_MAIN;
3284 	__be16 sport = 0, dport = 0;
3285 	struct fib_result res = {};
3286 	u8 ip_proto = IPPROTO_UDP;
3287 	struct rtable *rt = NULL;
3288 	struct sk_buff *skb;
3289 	struct rtmsg *rtm;
3290 	struct flowi4 fl4 = {};
3291 	__be32 dst = 0;
3292 	__be32 src = 0;
3293 	kuid_t uid;
3294 	u32 iif;
3295 	int err;
3296 	int mark;
3297 
3298 	err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3299 	if (err < 0)
3300 		return err;
3301 
3302 	rtm = nlmsg_data(nlh);
3303 	src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3304 	dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3305 	iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3306 	mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3307 	if (tb[RTA_UID])
3308 		uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3309 	else
3310 		uid = (iif ? INVALID_UID : current_uid());
3311 
3312 	if (tb[RTA_IP_PROTO]) {
3313 		err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3314 						  &ip_proto, AF_INET, extack);
3315 		if (err)
3316 			return err;
3317 	}
3318 
3319 	if (tb[RTA_SPORT])
3320 		sport = nla_get_be16(tb[RTA_SPORT]);
3321 
3322 	if (tb[RTA_DPORT])
3323 		dport = nla_get_be16(tb[RTA_DPORT]);
3324 
3325 	skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3326 	if (!skb)
3327 		return -ENOBUFS;
3328 
3329 	fl4.daddr = dst;
3330 	fl4.saddr = src;
3331 	fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3332 	fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3333 	fl4.flowi4_mark = mark;
3334 	fl4.flowi4_uid = uid;
3335 	if (sport)
3336 		fl4.fl4_sport = sport;
3337 	if (dport)
3338 		fl4.fl4_dport = dport;
3339 	fl4.flowi4_proto = ip_proto;
3340 
3341 	rcu_read_lock();
3342 
3343 	if (iif) {
3344 		struct net_device *dev;
3345 
3346 		dev = dev_get_by_index_rcu(net, iif);
3347 		if (!dev) {
3348 			err = -ENODEV;
3349 			goto errout_rcu;
3350 		}
3351 
3352 		fl4.flowi4_iif = iif; /* for rt_fill_info */
3353 		skb->dev	= dev;
3354 		skb->mark	= mark;
3355 		err = ip_route_input_rcu(skb, dst, src,
3356 					 rtm->rtm_tos & IPTOS_RT_MASK, dev,
3357 					 &res);
3358 
3359 		rt = skb_rtable(skb);
3360 		if (err == 0 && rt->dst.error)
3361 			err = -rt->dst.error;
3362 	} else {
3363 		fl4.flowi4_iif = LOOPBACK_IFINDEX;
3364 		skb->dev = net->loopback_dev;
3365 		rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3366 		err = 0;
3367 		if (IS_ERR(rt))
3368 			err = PTR_ERR(rt);
3369 		else
3370 			skb_dst_set(skb, &rt->dst);
3371 	}
3372 
3373 	if (err)
3374 		goto errout_rcu;
3375 
3376 	if (rtm->rtm_flags & RTM_F_NOTIFY)
3377 		rt->rt_flags |= RTCF_NOTIFY;
3378 
3379 	if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3380 		table_id = res.table ? res.table->tb_id : 0;
3381 
3382 	/* reset skb for netlink reply msg */
3383 	skb_trim(skb, 0);
3384 	skb_reset_network_header(skb);
3385 	skb_reset_transport_header(skb);
3386 	skb_reset_mac_header(skb);
3387 
3388 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3389 		struct fib_rt_info fri;
3390 
3391 		if (!res.fi) {
3392 			err = fib_props[res.type].error;
3393 			if (!err)
3394 				err = -EHOSTUNREACH;
3395 			goto errout_rcu;
3396 		}
3397 		fri.fi = res.fi;
3398 		fri.tb_id = table_id;
3399 		fri.dst = res.prefix;
3400 		fri.dst_len = res.prefixlen;
3401 		fri.dscp = inet_dsfield_to_dscp(fl4.flowi4_tos);
3402 		fri.type = rt->rt_type;
3403 		fri.offload = 0;
3404 		fri.trap = 0;
3405 		fri.offload_failed = 0;
3406 		if (res.fa_head) {
3407 			struct fib_alias *fa;
3408 
3409 			hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3410 				u8 slen = 32 - fri.dst_len;
3411 
3412 				if (fa->fa_slen == slen &&
3413 				    fa->tb_id == fri.tb_id &&
3414 				    fa->fa_dscp == fri.dscp &&
3415 				    fa->fa_info == res.fi &&
3416 				    fa->fa_type == fri.type) {
3417 					fri.offload = READ_ONCE(fa->offload);
3418 					fri.trap = READ_ONCE(fa->trap);
3419 					break;
3420 				}
3421 			}
3422 		}
3423 		err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3424 				    nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3425 	} else {
3426 		err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3427 				   NETLINK_CB(in_skb).portid,
3428 				   nlh->nlmsg_seq, 0);
3429 	}
3430 	if (err < 0)
3431 		goto errout_rcu;
3432 
3433 	rcu_read_unlock();
3434 
3435 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3436 
3437 errout_free:
3438 	return err;
3439 errout_rcu:
3440 	rcu_read_unlock();
3441 	kfree_skb(skb);
3442 	goto errout_free;
3443 }
3444 
3445 void ip_rt_multicast_event(struct in_device *in_dev)
3446 {
3447 	rt_cache_flush(dev_net(in_dev->dev));
3448 }
3449 
3450 #ifdef CONFIG_SYSCTL
3451 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3452 static int ip_rt_gc_min_interval __read_mostly	= HZ / 2;
3453 static int ip_rt_gc_elasticity __read_mostly	= 8;
3454 static int ip_min_valid_pmtu __read_mostly	= IPV4_MIN_MTU;
3455 
3456 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3457 		void *buffer, size_t *lenp, loff_t *ppos)
3458 {
3459 	struct net *net = (struct net *)__ctl->extra1;
3460 
3461 	if (write) {
3462 		rt_cache_flush(net);
3463 		fnhe_genid_bump(net);
3464 		return 0;
3465 	}
3466 
3467 	return -EINVAL;
3468 }
3469 
3470 static struct ctl_table ipv4_route_table[] = {
3471 	{
3472 		.procname	= "gc_thresh",
3473 		.data		= &ipv4_dst_ops.gc_thresh,
3474 		.maxlen		= sizeof(int),
3475 		.mode		= 0644,
3476 		.proc_handler	= proc_dointvec,
3477 	},
3478 	{
3479 		.procname	= "max_size",
3480 		.data		= &ip_rt_max_size,
3481 		.maxlen		= sizeof(int),
3482 		.mode		= 0644,
3483 		.proc_handler	= proc_dointvec,
3484 	},
3485 	{
3486 		/*  Deprecated. Use gc_min_interval_ms */
3487 
3488 		.procname	= "gc_min_interval",
3489 		.data		= &ip_rt_gc_min_interval,
3490 		.maxlen		= sizeof(int),
3491 		.mode		= 0644,
3492 		.proc_handler	= proc_dointvec_jiffies,
3493 	},
3494 	{
3495 		.procname	= "gc_min_interval_ms",
3496 		.data		= &ip_rt_gc_min_interval,
3497 		.maxlen		= sizeof(int),
3498 		.mode		= 0644,
3499 		.proc_handler	= proc_dointvec_ms_jiffies,
3500 	},
3501 	{
3502 		.procname	= "gc_timeout",
3503 		.data		= &ip_rt_gc_timeout,
3504 		.maxlen		= sizeof(int),
3505 		.mode		= 0644,
3506 		.proc_handler	= proc_dointvec_jiffies,
3507 	},
3508 	{
3509 		.procname	= "gc_interval",
3510 		.data		= &ip_rt_gc_interval,
3511 		.maxlen		= sizeof(int),
3512 		.mode		= 0644,
3513 		.proc_handler	= proc_dointvec_jiffies,
3514 	},
3515 	{
3516 		.procname	= "redirect_load",
3517 		.data		= &ip_rt_redirect_load,
3518 		.maxlen		= sizeof(int),
3519 		.mode		= 0644,
3520 		.proc_handler	= proc_dointvec,
3521 	},
3522 	{
3523 		.procname	= "redirect_number",
3524 		.data		= &ip_rt_redirect_number,
3525 		.maxlen		= sizeof(int),
3526 		.mode		= 0644,
3527 		.proc_handler	= proc_dointvec,
3528 	},
3529 	{
3530 		.procname	= "redirect_silence",
3531 		.data		= &ip_rt_redirect_silence,
3532 		.maxlen		= sizeof(int),
3533 		.mode		= 0644,
3534 		.proc_handler	= proc_dointvec,
3535 	},
3536 	{
3537 		.procname	= "error_cost",
3538 		.data		= &ip_rt_error_cost,
3539 		.maxlen		= sizeof(int),
3540 		.mode		= 0644,
3541 		.proc_handler	= proc_dointvec,
3542 	},
3543 	{
3544 		.procname	= "error_burst",
3545 		.data		= &ip_rt_error_burst,
3546 		.maxlen		= sizeof(int),
3547 		.mode		= 0644,
3548 		.proc_handler	= proc_dointvec,
3549 	},
3550 	{
3551 		.procname	= "gc_elasticity",
3552 		.data		= &ip_rt_gc_elasticity,
3553 		.maxlen		= sizeof(int),
3554 		.mode		= 0644,
3555 		.proc_handler	= proc_dointvec,
3556 	},
3557 	{ }
3558 };
3559 
3560 static const char ipv4_route_flush_procname[] = "flush";
3561 
3562 static struct ctl_table ipv4_route_netns_table[] = {
3563 	{
3564 		.procname	= ipv4_route_flush_procname,
3565 		.maxlen		= sizeof(int),
3566 		.mode		= 0200,
3567 		.proc_handler	= ipv4_sysctl_rtcache_flush,
3568 	},
3569 	{
3570 		.procname       = "min_pmtu",
3571 		.data           = &init_net.ipv4.ip_rt_min_pmtu,
3572 		.maxlen         = sizeof(int),
3573 		.mode           = 0644,
3574 		.proc_handler   = proc_dointvec_minmax,
3575 		.extra1         = &ip_min_valid_pmtu,
3576 	},
3577 	{
3578 		.procname       = "mtu_expires",
3579 		.data           = &init_net.ipv4.ip_rt_mtu_expires,
3580 		.maxlen         = sizeof(int),
3581 		.mode           = 0644,
3582 		.proc_handler   = proc_dointvec_jiffies,
3583 	},
3584 	{
3585 		.procname   = "min_adv_mss",
3586 		.data       = &init_net.ipv4.ip_rt_min_advmss,
3587 		.maxlen     = sizeof(int),
3588 		.mode       = 0644,
3589 		.proc_handler   = proc_dointvec,
3590 	},
3591 	{ },
3592 };
3593 
3594 static __net_init int sysctl_route_net_init(struct net *net)
3595 {
3596 	struct ctl_table *tbl;
3597 
3598 	tbl = ipv4_route_netns_table;
3599 	if (!net_eq(net, &init_net)) {
3600 		int i;
3601 
3602 		tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3603 		if (!tbl)
3604 			goto err_dup;
3605 
3606 		/* Don't export non-whitelisted sysctls to unprivileged users */
3607 		if (net->user_ns != &init_user_ns) {
3608 			if (tbl[0].procname != ipv4_route_flush_procname)
3609 				tbl[0].procname = NULL;
3610 		}
3611 
3612 		/* Update the variables to point into the current struct net
3613 		 * except for the first element flush
3614 		 */
3615 		for (i = 1; i < ARRAY_SIZE(ipv4_route_netns_table) - 1; i++)
3616 			tbl[i].data += (void *)net - (void *)&init_net;
3617 	}
3618 	tbl[0].extra1 = net;
3619 
3620 	net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3621 	if (!net->ipv4.route_hdr)
3622 		goto err_reg;
3623 	return 0;
3624 
3625 err_reg:
3626 	if (tbl != ipv4_route_netns_table)
3627 		kfree(tbl);
3628 err_dup:
3629 	return -ENOMEM;
3630 }
3631 
3632 static __net_exit void sysctl_route_net_exit(struct net *net)
3633 {
3634 	struct ctl_table *tbl;
3635 
3636 	tbl = net->ipv4.route_hdr->ctl_table_arg;
3637 	unregister_net_sysctl_table(net->ipv4.route_hdr);
3638 	BUG_ON(tbl == ipv4_route_netns_table);
3639 	kfree(tbl);
3640 }
3641 
3642 static __net_initdata struct pernet_operations sysctl_route_ops = {
3643 	.init = sysctl_route_net_init,
3644 	.exit = sysctl_route_net_exit,
3645 };
3646 #endif
3647 
3648 static __net_init int netns_ip_rt_init(struct net *net)
3649 {
3650 	/* Set default value for namespaceified sysctls */
3651 	net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3652 	net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3653 	net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3654 	return 0;
3655 }
3656 
3657 static struct pernet_operations __net_initdata ip_rt_ops = {
3658 	.init = netns_ip_rt_init,
3659 };
3660 
3661 static __net_init int rt_genid_init(struct net *net)
3662 {
3663 	atomic_set(&net->ipv4.rt_genid, 0);
3664 	atomic_set(&net->fnhe_genid, 0);
3665 	atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3666 	return 0;
3667 }
3668 
3669 static __net_initdata struct pernet_operations rt_genid_ops = {
3670 	.init = rt_genid_init,
3671 };
3672 
3673 static int __net_init ipv4_inetpeer_init(struct net *net)
3674 {
3675 	struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3676 
3677 	if (!bp)
3678 		return -ENOMEM;
3679 	inet_peer_base_init(bp);
3680 	net->ipv4.peers = bp;
3681 	return 0;
3682 }
3683 
3684 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3685 {
3686 	struct inet_peer_base *bp = net->ipv4.peers;
3687 
3688 	net->ipv4.peers = NULL;
3689 	inetpeer_invalidate_tree(bp);
3690 	kfree(bp);
3691 }
3692 
3693 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3694 	.init	=	ipv4_inetpeer_init,
3695 	.exit	=	ipv4_inetpeer_exit,
3696 };
3697 
3698 #ifdef CONFIG_IP_ROUTE_CLASSID
3699 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3700 #endif /* CONFIG_IP_ROUTE_CLASSID */
3701 
3702 int __init ip_rt_init(void)
3703 {
3704 	void *idents_hash;
3705 	int cpu;
3706 
3707 	/* For modern hosts, this will use 2 MB of memory */
3708 	idents_hash = alloc_large_system_hash("IP idents",
3709 					      sizeof(*ip_idents) + sizeof(*ip_tstamps),
3710 					      0,
3711 					      16, /* one bucket per 64 KB */
3712 					      HASH_ZERO,
3713 					      NULL,
3714 					      &ip_idents_mask,
3715 					      2048,
3716 					      256*1024);
3717 
3718 	ip_idents = idents_hash;
3719 
3720 	prandom_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3721 
3722 	ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3723 
3724 	for_each_possible_cpu(cpu) {
3725 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3726 
3727 		INIT_LIST_HEAD(&ul->head);
3728 		INIT_LIST_HEAD(&ul->quarantine);
3729 		spin_lock_init(&ul->lock);
3730 	}
3731 #ifdef CONFIG_IP_ROUTE_CLASSID
3732 	ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3733 	if (!ip_rt_acct)
3734 		panic("IP: failed to allocate ip_rt_acct\n");
3735 #endif
3736 
3737 	ipv4_dst_ops.kmem_cachep =
3738 		kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3739 				  SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3740 
3741 	ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3742 
3743 	if (dst_entries_init(&ipv4_dst_ops) < 0)
3744 		panic("IP: failed to allocate ipv4_dst_ops counter\n");
3745 
3746 	if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3747 		panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3748 
3749 	ipv4_dst_ops.gc_thresh = ~0;
3750 	ip_rt_max_size = INT_MAX;
3751 
3752 	devinet_init();
3753 	ip_fib_init();
3754 
3755 	if (ip_rt_proc_init())
3756 		pr_err("Unable to create route proc files\n");
3757 #ifdef CONFIG_XFRM
3758 	xfrm_init();
3759 	xfrm4_init();
3760 #endif
3761 	rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3762 		      RTNL_FLAG_DOIT_UNLOCKED);
3763 
3764 #ifdef CONFIG_SYSCTL
3765 	register_pernet_subsys(&sysctl_route_ops);
3766 #endif
3767 	register_pernet_subsys(&ip_rt_ops);
3768 	register_pernet_subsys(&rt_genid_ops);
3769 	register_pernet_subsys(&ipv4_inetpeer_ops);
3770 	return 0;
3771 }
3772 
3773 #ifdef CONFIG_SYSCTL
3774 /*
3775  * We really need to sanitize the damn ipv4 init order, then all
3776  * this nonsense will go away.
3777  */
3778 void __init ip_static_sysctl_init(void)
3779 {
3780 	register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3781 }
3782 #endif
3783