xref: /linux/net/ipv4/route.c (revision 860a9bed265146b10311bcadbbcef59c3af4454d)
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_GC_TIMEOUT (300*HZ)
110 
111 #define DEFAULT_MIN_PMTU (512 + 20 + 20)
112 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
113 #define DEFAULT_MIN_ADVMSS 256
114 static int ip_rt_max_size;
115 static int ip_rt_redirect_number __read_mostly	= 9;
116 static int ip_rt_redirect_load __read_mostly	= HZ / 50;
117 static int ip_rt_redirect_silence __read_mostly	= ((HZ / 50) << (9 + 1));
118 static int ip_rt_error_cost __read_mostly	= HZ;
119 static int ip_rt_error_burst __read_mostly	= 5 * HZ;
120 
121 static int ip_rt_gc_timeout __read_mostly	= RT_GC_TIMEOUT;
122 
123 /*
124  *	Interface to generic destination cache.
125  */
126 
127 INDIRECT_CALLABLE_SCOPE
128 struct dst_entry	*ipv4_dst_check(struct dst_entry *dst, u32 cookie);
129 static unsigned int	 ipv4_default_advmss(const struct dst_entry *dst);
130 INDIRECT_CALLABLE_SCOPE
131 unsigned int		ipv4_mtu(const struct dst_entry *dst);
132 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
133 static void		 ipv4_link_failure(struct sk_buff *skb);
134 static void		 ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
135 					   struct sk_buff *skb, u32 mtu,
136 					   bool confirm_neigh);
137 static void		 ip_do_redirect(struct dst_entry *dst, struct sock *sk,
138 					struct sk_buff *skb);
139 static void		ipv4_dst_destroy(struct dst_entry *dst);
140 
141 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
142 {
143 	WARN_ON(1);
144 	return NULL;
145 }
146 
147 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
148 					   struct sk_buff *skb,
149 					   const void *daddr);
150 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
151 
152 static struct dst_ops ipv4_dst_ops = {
153 	.family =		AF_INET,
154 	.check =		ipv4_dst_check,
155 	.default_advmss =	ipv4_default_advmss,
156 	.mtu =			ipv4_mtu,
157 	.cow_metrics =		ipv4_cow_metrics,
158 	.destroy =		ipv4_dst_destroy,
159 	.negative_advice =	ipv4_negative_advice,
160 	.link_failure =		ipv4_link_failure,
161 	.update_pmtu =		ip_rt_update_pmtu,
162 	.redirect =		ip_do_redirect,
163 	.local_out =		__ip_local_out,
164 	.neigh_lookup =		ipv4_neigh_lookup,
165 	.confirm_neigh =	ipv4_confirm_neigh,
166 };
167 
168 #define ECN_OR_COST(class)	TC_PRIO_##class
169 
170 const __u8 ip_tos2prio[16] = {
171 	TC_PRIO_BESTEFFORT,
172 	ECN_OR_COST(BESTEFFORT),
173 	TC_PRIO_BESTEFFORT,
174 	ECN_OR_COST(BESTEFFORT),
175 	TC_PRIO_BULK,
176 	ECN_OR_COST(BULK),
177 	TC_PRIO_BULK,
178 	ECN_OR_COST(BULK),
179 	TC_PRIO_INTERACTIVE,
180 	ECN_OR_COST(INTERACTIVE),
181 	TC_PRIO_INTERACTIVE,
182 	ECN_OR_COST(INTERACTIVE),
183 	TC_PRIO_INTERACTIVE_BULK,
184 	ECN_OR_COST(INTERACTIVE_BULK),
185 	TC_PRIO_INTERACTIVE_BULK,
186 	ECN_OR_COST(INTERACTIVE_BULK)
187 };
188 EXPORT_SYMBOL(ip_tos2prio);
189 
190 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
191 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
192 
193 #ifdef CONFIG_PROC_FS
194 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
195 {
196 	if (*pos)
197 		return NULL;
198 	return SEQ_START_TOKEN;
199 }
200 
201 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
202 {
203 	++*pos;
204 	return NULL;
205 }
206 
207 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
208 {
209 }
210 
211 static int rt_cache_seq_show(struct seq_file *seq, void *v)
212 {
213 	if (v == SEQ_START_TOKEN)
214 		seq_printf(seq, "%-127s\n",
215 			   "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
216 			   "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
217 			   "HHUptod\tSpecDst");
218 	return 0;
219 }
220 
221 static const struct seq_operations rt_cache_seq_ops = {
222 	.start  = rt_cache_seq_start,
223 	.next   = rt_cache_seq_next,
224 	.stop   = rt_cache_seq_stop,
225 	.show   = rt_cache_seq_show,
226 };
227 
228 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
229 {
230 	int cpu;
231 
232 	if (*pos == 0)
233 		return SEQ_START_TOKEN;
234 
235 	for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
236 		if (!cpu_possible(cpu))
237 			continue;
238 		*pos = cpu+1;
239 		return &per_cpu(rt_cache_stat, cpu);
240 	}
241 	return NULL;
242 }
243 
244 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
245 {
246 	int cpu;
247 
248 	for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
249 		if (!cpu_possible(cpu))
250 			continue;
251 		*pos = cpu+1;
252 		return &per_cpu(rt_cache_stat, cpu);
253 	}
254 	(*pos)++;
255 	return NULL;
256 
257 }
258 
259 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
260 {
261 
262 }
263 
264 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
265 {
266 	struct rt_cache_stat *st = v;
267 
268 	if (v == SEQ_START_TOKEN) {
269 		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");
270 		return 0;
271 	}
272 
273 	seq_printf(seq, "%08x %08x %08x    %08x   %08x    %08x %08x       "
274 			"%08x       %08x %08x     %08x    %08x %08x   "
275 			"%08x     %08x        %08x        %08x\n",
276 		   dst_entries_get_slow(&ipv4_dst_ops),
277 		   0, /* st->in_hit */
278 		   st->in_slow_tot,
279 		   st->in_slow_mc,
280 		   st->in_no_route,
281 		   st->in_brd,
282 		   st->in_martian_dst,
283 		   st->in_martian_src,
284 
285 		   0, /* st->out_hit */
286 		   st->out_slow_tot,
287 		   st->out_slow_mc,
288 
289 		   0, /* st->gc_total */
290 		   0, /* st->gc_ignored */
291 		   0, /* st->gc_goal_miss */
292 		   0, /* st->gc_dst_overflow */
293 		   0, /* st->in_hlist_search */
294 		   0  /* st->out_hlist_search */
295 		);
296 	return 0;
297 }
298 
299 static const struct seq_operations rt_cpu_seq_ops = {
300 	.start  = rt_cpu_seq_start,
301 	.next   = rt_cpu_seq_next,
302 	.stop   = rt_cpu_seq_stop,
303 	.show   = rt_cpu_seq_show,
304 };
305 
306 #ifdef CONFIG_IP_ROUTE_CLASSID
307 static int rt_acct_proc_show(struct seq_file *m, void *v)
308 {
309 	struct ip_rt_acct *dst, *src;
310 	unsigned int i, j;
311 
312 	dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
313 	if (!dst)
314 		return -ENOMEM;
315 
316 	for_each_possible_cpu(i) {
317 		src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
318 		for (j = 0; j < 256; j++) {
319 			dst[j].o_bytes   += src[j].o_bytes;
320 			dst[j].o_packets += src[j].o_packets;
321 			dst[j].i_bytes   += src[j].i_bytes;
322 			dst[j].i_packets += src[j].i_packets;
323 		}
324 	}
325 
326 	seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
327 	kfree(dst);
328 	return 0;
329 }
330 #endif
331 
332 static int __net_init ip_rt_do_proc_init(struct net *net)
333 {
334 	struct proc_dir_entry *pde;
335 
336 	pde = proc_create_seq("rt_cache", 0444, net->proc_net,
337 			      &rt_cache_seq_ops);
338 	if (!pde)
339 		goto err1;
340 
341 	pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
342 			      &rt_cpu_seq_ops);
343 	if (!pde)
344 		goto err2;
345 
346 #ifdef CONFIG_IP_ROUTE_CLASSID
347 	pde = proc_create_single("rt_acct", 0, net->proc_net,
348 			rt_acct_proc_show);
349 	if (!pde)
350 		goto err3;
351 #endif
352 	return 0;
353 
354 #ifdef CONFIG_IP_ROUTE_CLASSID
355 err3:
356 	remove_proc_entry("rt_cache", net->proc_net_stat);
357 #endif
358 err2:
359 	remove_proc_entry("rt_cache", net->proc_net);
360 err1:
361 	return -ENOMEM;
362 }
363 
364 static void __net_exit ip_rt_do_proc_exit(struct net *net)
365 {
366 	remove_proc_entry("rt_cache", net->proc_net_stat);
367 	remove_proc_entry("rt_cache", net->proc_net);
368 #ifdef CONFIG_IP_ROUTE_CLASSID
369 	remove_proc_entry("rt_acct", net->proc_net);
370 #endif
371 }
372 
373 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
374 	.init = ip_rt_do_proc_init,
375 	.exit = ip_rt_do_proc_exit,
376 };
377 
378 static int __init ip_rt_proc_init(void)
379 {
380 	return register_pernet_subsys(&ip_rt_proc_ops);
381 }
382 
383 #else
384 static inline int ip_rt_proc_init(void)
385 {
386 	return 0;
387 }
388 #endif /* CONFIG_PROC_FS */
389 
390 static inline bool rt_is_expired(const struct rtable *rth)
391 {
392 	return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
393 }
394 
395 void rt_cache_flush(struct net *net)
396 {
397 	rt_genid_bump_ipv4(net);
398 }
399 
400 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
401 					   struct sk_buff *skb,
402 					   const void *daddr)
403 {
404 	const struct rtable *rt = container_of(dst, struct rtable, dst);
405 	struct net_device *dev = dst->dev;
406 	struct neighbour *n;
407 
408 	rcu_read_lock();
409 
410 	if (likely(rt->rt_gw_family == AF_INET)) {
411 		n = ip_neigh_gw4(dev, rt->rt_gw4);
412 	} else if (rt->rt_gw_family == AF_INET6) {
413 		n = ip_neigh_gw6(dev, &rt->rt_gw6);
414         } else {
415 		__be32 pkey;
416 
417 		pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
418 		n = ip_neigh_gw4(dev, pkey);
419 	}
420 
421 	if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
422 		n = NULL;
423 
424 	rcu_read_unlock();
425 
426 	return n;
427 }
428 
429 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
430 {
431 	const struct rtable *rt = container_of(dst, struct rtable, dst);
432 	struct net_device *dev = dst->dev;
433 	const __be32 *pkey = daddr;
434 
435 	if (rt->rt_gw_family == AF_INET) {
436 		pkey = (const __be32 *)&rt->rt_gw4;
437 	} else if (rt->rt_gw_family == AF_INET6) {
438 		return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
439 	} else if (!daddr ||
440 		 (rt->rt_flags &
441 		  (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
442 		return;
443 	}
444 	__ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
445 }
446 
447 /* Hash tables of size 2048..262144 depending on RAM size.
448  * Each bucket uses 8 bytes.
449  */
450 static u32 ip_idents_mask __read_mostly;
451 static atomic_t *ip_idents __read_mostly;
452 static u32 *ip_tstamps __read_mostly;
453 
454 /* In order to protect privacy, we add a perturbation to identifiers
455  * if one generator is seldom used. This makes hard for an attacker
456  * to infer how many packets were sent between two points in time.
457  */
458 static u32 ip_idents_reserve(u32 hash, int segs)
459 {
460 	u32 bucket, old, now = (u32)jiffies;
461 	atomic_t *p_id;
462 	u32 *p_tstamp;
463 	u32 delta = 0;
464 
465 	bucket = hash & ip_idents_mask;
466 	p_tstamp = ip_tstamps + bucket;
467 	p_id = ip_idents + bucket;
468 	old = READ_ONCE(*p_tstamp);
469 
470 	if (old != now && cmpxchg(p_tstamp, old, now) == old)
471 		delta = get_random_u32_below(now - old);
472 
473 	/* If UBSAN reports an error there, please make sure your compiler
474 	 * supports -fno-strict-overflow before reporting it that was a bug
475 	 * in UBSAN, and it has been fixed in GCC-8.
476 	 */
477 	return atomic_add_return(segs + delta, p_id) - segs;
478 }
479 
480 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
481 {
482 	u32 hash, id;
483 
484 	/* Note the following code is not safe, but this is okay. */
485 	if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
486 		get_random_bytes(&net->ipv4.ip_id_key,
487 				 sizeof(net->ipv4.ip_id_key));
488 
489 	hash = siphash_3u32((__force u32)iph->daddr,
490 			    (__force u32)iph->saddr,
491 			    iph->protocol,
492 			    &net->ipv4.ip_id_key);
493 	id = ip_idents_reserve(hash, segs);
494 	iph->id = htons(id);
495 }
496 EXPORT_SYMBOL(__ip_select_ident);
497 
498 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
499 			     const struct sock *sk, const struct iphdr *iph,
500 			     int oif, __u8 tos, u8 prot, u32 mark,
501 			     int flow_flags)
502 {
503 	__u8 scope = RT_SCOPE_UNIVERSE;
504 
505 	if (sk) {
506 		oif = sk->sk_bound_dev_if;
507 		mark = READ_ONCE(sk->sk_mark);
508 		tos = ip_sock_rt_tos(sk);
509 		scope = ip_sock_rt_scope(sk);
510 		prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW :
511 						    sk->sk_protocol;
512 	}
513 
514 	flowi4_init_output(fl4, oif, mark, tos & IPTOS_RT_MASK, scope,
515 			   prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
516 			   sock_net_uid(net, sk));
517 }
518 
519 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
520 			       const struct sock *sk)
521 {
522 	const struct net *net = dev_net(skb->dev);
523 	const struct iphdr *iph = ip_hdr(skb);
524 	int oif = skb->dev->ifindex;
525 	u8 prot = iph->protocol;
526 	u32 mark = skb->mark;
527 	__u8 tos = iph->tos;
528 
529 	__build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
530 }
531 
532 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
533 {
534 	const struct inet_sock *inet = inet_sk(sk);
535 	const struct ip_options_rcu *inet_opt;
536 	__be32 daddr = inet->inet_daddr;
537 
538 	rcu_read_lock();
539 	inet_opt = rcu_dereference(inet->inet_opt);
540 	if (inet_opt && inet_opt->opt.srr)
541 		daddr = inet_opt->opt.faddr;
542 	flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark),
543 			   ip_sock_rt_tos(sk) & IPTOS_RT_MASK,
544 			   ip_sock_rt_scope(sk),
545 			   inet_test_bit(HDRINCL, sk) ?
546 				IPPROTO_RAW : sk->sk_protocol,
547 			   inet_sk_flowi_flags(sk),
548 			   daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
549 	rcu_read_unlock();
550 }
551 
552 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
553 				 const struct sk_buff *skb)
554 {
555 	if (skb)
556 		build_skb_flow_key(fl4, skb, sk);
557 	else
558 		build_sk_flow_key(fl4, sk);
559 }
560 
561 static DEFINE_SPINLOCK(fnhe_lock);
562 
563 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
564 {
565 	struct rtable *rt;
566 
567 	rt = rcu_dereference(fnhe->fnhe_rth_input);
568 	if (rt) {
569 		RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
570 		dst_dev_put(&rt->dst);
571 		dst_release(&rt->dst);
572 	}
573 	rt = rcu_dereference(fnhe->fnhe_rth_output);
574 	if (rt) {
575 		RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
576 		dst_dev_put(&rt->dst);
577 		dst_release(&rt->dst);
578 	}
579 }
580 
581 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
582 {
583 	struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
584 	struct fib_nh_exception *fnhe, *oldest = NULL;
585 
586 	for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
587 		fnhe = rcu_dereference_protected(*fnhe_p,
588 						 lockdep_is_held(&fnhe_lock));
589 		if (!fnhe)
590 			break;
591 		if (!oldest ||
592 		    time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
593 			oldest = fnhe;
594 			oldest_p = fnhe_p;
595 		}
596 	}
597 	fnhe_flush_routes(oldest);
598 	*oldest_p = oldest->fnhe_next;
599 	kfree_rcu(oldest, rcu);
600 }
601 
602 static u32 fnhe_hashfun(__be32 daddr)
603 {
604 	static siphash_aligned_key_t fnhe_hash_key;
605 	u64 hval;
606 
607 	net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
608 	hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
609 	return hash_64(hval, FNHE_HASH_SHIFT);
610 }
611 
612 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
613 {
614 	rt->rt_pmtu = fnhe->fnhe_pmtu;
615 	rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
616 	rt->dst.expires = fnhe->fnhe_expires;
617 
618 	if (fnhe->fnhe_gw) {
619 		rt->rt_flags |= RTCF_REDIRECTED;
620 		rt->rt_uses_gateway = 1;
621 		rt->rt_gw_family = AF_INET;
622 		rt->rt_gw4 = fnhe->fnhe_gw;
623 	}
624 }
625 
626 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
627 				  __be32 gw, u32 pmtu, bool lock,
628 				  unsigned long expires)
629 {
630 	struct fnhe_hash_bucket *hash;
631 	struct fib_nh_exception *fnhe;
632 	struct rtable *rt;
633 	u32 genid, hval;
634 	unsigned int i;
635 	int depth;
636 
637 	genid = fnhe_genid(dev_net(nhc->nhc_dev));
638 	hval = fnhe_hashfun(daddr);
639 
640 	spin_lock_bh(&fnhe_lock);
641 
642 	hash = rcu_dereference(nhc->nhc_exceptions);
643 	if (!hash) {
644 		hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
645 		if (!hash)
646 			goto out_unlock;
647 		rcu_assign_pointer(nhc->nhc_exceptions, hash);
648 	}
649 
650 	hash += hval;
651 
652 	depth = 0;
653 	for (fnhe = rcu_dereference(hash->chain); fnhe;
654 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
655 		if (fnhe->fnhe_daddr == daddr)
656 			break;
657 		depth++;
658 	}
659 
660 	if (fnhe) {
661 		if (fnhe->fnhe_genid != genid)
662 			fnhe->fnhe_genid = genid;
663 		if (gw)
664 			fnhe->fnhe_gw = gw;
665 		if (pmtu) {
666 			fnhe->fnhe_pmtu = pmtu;
667 			fnhe->fnhe_mtu_locked = lock;
668 		}
669 		fnhe->fnhe_expires = max(1UL, expires);
670 		/* Update all cached dsts too */
671 		rt = rcu_dereference(fnhe->fnhe_rth_input);
672 		if (rt)
673 			fill_route_from_fnhe(rt, fnhe);
674 		rt = rcu_dereference(fnhe->fnhe_rth_output);
675 		if (rt)
676 			fill_route_from_fnhe(rt, fnhe);
677 	} else {
678 		/* Randomize max depth to avoid some side channels attacks. */
679 		int max_depth = FNHE_RECLAIM_DEPTH +
680 				get_random_u32_below(FNHE_RECLAIM_DEPTH);
681 
682 		while (depth > max_depth) {
683 			fnhe_remove_oldest(hash);
684 			depth--;
685 		}
686 
687 		fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
688 		if (!fnhe)
689 			goto out_unlock;
690 
691 		fnhe->fnhe_next = hash->chain;
692 
693 		fnhe->fnhe_genid = genid;
694 		fnhe->fnhe_daddr = daddr;
695 		fnhe->fnhe_gw = gw;
696 		fnhe->fnhe_pmtu = pmtu;
697 		fnhe->fnhe_mtu_locked = lock;
698 		fnhe->fnhe_expires = max(1UL, expires);
699 
700 		rcu_assign_pointer(hash->chain, fnhe);
701 
702 		/* Exception created; mark the cached routes for the nexthop
703 		 * stale, so anyone caching it rechecks if this exception
704 		 * applies to them.
705 		 */
706 		rt = rcu_dereference(nhc->nhc_rth_input);
707 		if (rt)
708 			rt->dst.obsolete = DST_OBSOLETE_KILL;
709 
710 		for_each_possible_cpu(i) {
711 			struct rtable __rcu **prt;
712 
713 			prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
714 			rt = rcu_dereference(*prt);
715 			if (rt)
716 				rt->dst.obsolete = DST_OBSOLETE_KILL;
717 		}
718 	}
719 
720 	fnhe->fnhe_stamp = jiffies;
721 
722 out_unlock:
723 	spin_unlock_bh(&fnhe_lock);
724 }
725 
726 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
727 			     bool kill_route)
728 {
729 	__be32 new_gw = icmp_hdr(skb)->un.gateway;
730 	__be32 old_gw = ip_hdr(skb)->saddr;
731 	struct net_device *dev = skb->dev;
732 	struct in_device *in_dev;
733 	struct fib_result res;
734 	struct neighbour *n;
735 	struct net *net;
736 
737 	switch (icmp_hdr(skb)->code & 7) {
738 	case ICMP_REDIR_NET:
739 	case ICMP_REDIR_NETTOS:
740 	case ICMP_REDIR_HOST:
741 	case ICMP_REDIR_HOSTTOS:
742 		break;
743 
744 	default:
745 		return;
746 	}
747 
748 	if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
749 		return;
750 
751 	in_dev = __in_dev_get_rcu(dev);
752 	if (!in_dev)
753 		return;
754 
755 	net = dev_net(dev);
756 	if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
757 	    ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
758 	    ipv4_is_zeronet(new_gw))
759 		goto reject_redirect;
760 
761 	if (!IN_DEV_SHARED_MEDIA(in_dev)) {
762 		if (!inet_addr_onlink(in_dev, new_gw, old_gw))
763 			goto reject_redirect;
764 		if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
765 			goto reject_redirect;
766 	} else {
767 		if (inet_addr_type(net, new_gw) != RTN_UNICAST)
768 			goto reject_redirect;
769 	}
770 
771 	n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw);
772 	if (!n)
773 		n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
774 	if (!IS_ERR(n)) {
775 		if (!(READ_ONCE(n->nud_state) & NUD_VALID)) {
776 			neigh_event_send(n, NULL);
777 		} else {
778 			if (fib_lookup(net, fl4, &res, 0) == 0) {
779 				struct fib_nh_common *nhc;
780 
781 				fib_select_path(net, &res, fl4, skb);
782 				nhc = FIB_RES_NHC(res);
783 				update_or_create_fnhe(nhc, fl4->daddr, new_gw,
784 						0, false,
785 						jiffies + ip_rt_gc_timeout);
786 			}
787 			if (kill_route)
788 				rt->dst.obsolete = DST_OBSOLETE_KILL;
789 			call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
790 		}
791 		neigh_release(n);
792 	}
793 	return;
794 
795 reject_redirect:
796 #ifdef CONFIG_IP_ROUTE_VERBOSE
797 	if (IN_DEV_LOG_MARTIANS(in_dev)) {
798 		const struct iphdr *iph = (const struct iphdr *) skb->data;
799 		__be32 daddr = iph->daddr;
800 		__be32 saddr = iph->saddr;
801 
802 		net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
803 				     "  Advised path = %pI4 -> %pI4\n",
804 				     &old_gw, dev->name, &new_gw,
805 				     &saddr, &daddr);
806 	}
807 #endif
808 	;
809 }
810 
811 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
812 {
813 	struct rtable *rt;
814 	struct flowi4 fl4;
815 	const struct iphdr *iph = (const struct iphdr *) skb->data;
816 	struct net *net = dev_net(skb->dev);
817 	int oif = skb->dev->ifindex;
818 	u8 prot = iph->protocol;
819 	u32 mark = skb->mark;
820 	__u8 tos = iph->tos;
821 
822 	rt = (struct rtable *) dst;
823 
824 	__build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
825 	__ip_do_redirect(rt, skb, &fl4, true);
826 }
827 
828 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
829 {
830 	struct rtable *rt = (struct rtable *)dst;
831 	struct dst_entry *ret = dst;
832 
833 	if (rt) {
834 		if (dst->obsolete > 0) {
835 			ip_rt_put(rt);
836 			ret = NULL;
837 		} else if ((rt->rt_flags & RTCF_REDIRECTED) ||
838 			   rt->dst.expires) {
839 			ip_rt_put(rt);
840 			ret = NULL;
841 		}
842 	}
843 	return ret;
844 }
845 
846 /*
847  * Algorithm:
848  *	1. The first ip_rt_redirect_number redirects are sent
849  *	   with exponential backoff, then we stop sending them at all,
850  *	   assuming that the host ignores our redirects.
851  *	2. If we did not see packets requiring redirects
852  *	   during ip_rt_redirect_silence, we assume that the host
853  *	   forgot redirected route and start to send redirects again.
854  *
855  * This algorithm is much cheaper and more intelligent than dumb load limiting
856  * in icmp.c.
857  *
858  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
859  * and "frag. need" (breaks PMTU discovery) in icmp.c.
860  */
861 
862 void ip_rt_send_redirect(struct sk_buff *skb)
863 {
864 	struct rtable *rt = skb_rtable(skb);
865 	struct in_device *in_dev;
866 	struct inet_peer *peer;
867 	struct net *net;
868 	int log_martians;
869 	int vif;
870 
871 	rcu_read_lock();
872 	in_dev = __in_dev_get_rcu(rt->dst.dev);
873 	if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
874 		rcu_read_unlock();
875 		return;
876 	}
877 	log_martians = IN_DEV_LOG_MARTIANS(in_dev);
878 	vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
879 	rcu_read_unlock();
880 
881 	net = dev_net(rt->dst.dev);
882 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
883 	if (!peer) {
884 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
885 			  rt_nexthop(rt, ip_hdr(skb)->daddr));
886 		return;
887 	}
888 
889 	/* No redirected packets during ip_rt_redirect_silence;
890 	 * reset the algorithm.
891 	 */
892 	if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
893 		peer->rate_tokens = 0;
894 		peer->n_redirects = 0;
895 	}
896 
897 	/* Too many ignored redirects; do not send anything
898 	 * set dst.rate_last to the last seen redirected packet.
899 	 */
900 	if (peer->n_redirects >= ip_rt_redirect_number) {
901 		peer->rate_last = jiffies;
902 		goto out_put_peer;
903 	}
904 
905 	/* Check for load limit; set rate_last to the latest sent
906 	 * redirect.
907 	 */
908 	if (peer->n_redirects == 0 ||
909 	    time_after(jiffies,
910 		       (peer->rate_last +
911 			(ip_rt_redirect_load << peer->n_redirects)))) {
912 		__be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
913 
914 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
915 		peer->rate_last = jiffies;
916 		++peer->n_redirects;
917 		if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians &&
918 		    peer->n_redirects == ip_rt_redirect_number)
919 			net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
920 					     &ip_hdr(skb)->saddr, inet_iif(skb),
921 					     &ip_hdr(skb)->daddr, &gw);
922 	}
923 out_put_peer:
924 	inet_putpeer(peer);
925 }
926 
927 static int ip_error(struct sk_buff *skb)
928 {
929 	struct rtable *rt = skb_rtable(skb);
930 	struct net_device *dev = skb->dev;
931 	struct in_device *in_dev;
932 	struct inet_peer *peer;
933 	unsigned long now;
934 	struct net *net;
935 	SKB_DR(reason);
936 	bool send;
937 	int code;
938 
939 	if (netif_is_l3_master(skb->dev)) {
940 		dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
941 		if (!dev)
942 			goto out;
943 	}
944 
945 	in_dev = __in_dev_get_rcu(dev);
946 
947 	/* IP on this device is disabled. */
948 	if (!in_dev)
949 		goto out;
950 
951 	net = dev_net(rt->dst.dev);
952 	if (!IN_DEV_FORWARD(in_dev)) {
953 		switch (rt->dst.error) {
954 		case EHOSTUNREACH:
955 			SKB_DR_SET(reason, IP_INADDRERRORS);
956 			__IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
957 			break;
958 
959 		case ENETUNREACH:
960 			SKB_DR_SET(reason, IP_INNOROUTES);
961 			__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
962 			break;
963 		}
964 		goto out;
965 	}
966 
967 	switch (rt->dst.error) {
968 	case EINVAL:
969 	default:
970 		goto out;
971 	case EHOSTUNREACH:
972 		code = ICMP_HOST_UNREACH;
973 		break;
974 	case ENETUNREACH:
975 		code = ICMP_NET_UNREACH;
976 		SKB_DR_SET(reason, IP_INNOROUTES);
977 		__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
978 		break;
979 	case EACCES:
980 		code = ICMP_PKT_FILTERED;
981 		break;
982 	}
983 
984 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
985 			       l3mdev_master_ifindex(skb->dev), 1);
986 
987 	send = true;
988 	if (peer) {
989 		now = jiffies;
990 		peer->rate_tokens += now - peer->rate_last;
991 		if (peer->rate_tokens > ip_rt_error_burst)
992 			peer->rate_tokens = ip_rt_error_burst;
993 		peer->rate_last = now;
994 		if (peer->rate_tokens >= ip_rt_error_cost)
995 			peer->rate_tokens -= ip_rt_error_cost;
996 		else
997 			send = false;
998 		inet_putpeer(peer);
999 	}
1000 	if (send)
1001 		icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1002 
1003 out:	kfree_skb_reason(skb, reason);
1004 	return 0;
1005 }
1006 
1007 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1008 {
1009 	struct dst_entry *dst = &rt->dst;
1010 	struct net *net = dev_net(dst->dev);
1011 	struct fib_result res;
1012 	bool lock = false;
1013 	u32 old_mtu;
1014 
1015 	if (ip_mtu_locked(dst))
1016 		return;
1017 
1018 	old_mtu = ipv4_mtu(dst);
1019 	if (old_mtu < mtu)
1020 		return;
1021 
1022 	if (mtu < net->ipv4.ip_rt_min_pmtu) {
1023 		lock = true;
1024 		mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1025 	}
1026 
1027 	if (rt->rt_pmtu == mtu && !lock &&
1028 	    time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2))
1029 		return;
1030 
1031 	rcu_read_lock();
1032 	if (fib_lookup(net, fl4, &res, 0) == 0) {
1033 		struct fib_nh_common *nhc;
1034 
1035 		fib_select_path(net, &res, fl4, NULL);
1036 		nhc = FIB_RES_NHC(res);
1037 		update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1038 				      jiffies + net->ipv4.ip_rt_mtu_expires);
1039 	}
1040 	rcu_read_unlock();
1041 }
1042 
1043 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1044 			      struct sk_buff *skb, u32 mtu,
1045 			      bool confirm_neigh)
1046 {
1047 	struct rtable *rt = (struct rtable *) dst;
1048 	struct flowi4 fl4;
1049 
1050 	ip_rt_build_flow_key(&fl4, sk, skb);
1051 
1052 	/* Don't make lookup fail for bridged encapsulations */
1053 	if (skb && netif_is_any_bridge_port(skb->dev))
1054 		fl4.flowi4_oif = 0;
1055 
1056 	__ip_rt_update_pmtu(rt, &fl4, mtu);
1057 }
1058 
1059 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1060 		      int oif, u8 protocol)
1061 {
1062 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1063 	struct flowi4 fl4;
1064 	struct rtable *rt;
1065 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
1066 
1067 	__build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1068 			 0);
1069 	rt = __ip_route_output_key(net, &fl4);
1070 	if (!IS_ERR(rt)) {
1071 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1072 		ip_rt_put(rt);
1073 	}
1074 }
1075 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1076 
1077 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1078 {
1079 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1080 	struct flowi4 fl4;
1081 	struct rtable *rt;
1082 
1083 	__build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1084 
1085 	if (!fl4.flowi4_mark)
1086 		fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1087 
1088 	rt = __ip_route_output_key(sock_net(sk), &fl4);
1089 	if (!IS_ERR(rt)) {
1090 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1091 		ip_rt_put(rt);
1092 	}
1093 }
1094 
1095 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1096 {
1097 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1098 	struct flowi4 fl4;
1099 	struct rtable *rt;
1100 	struct dst_entry *odst = NULL;
1101 	bool new = false;
1102 	struct net *net = sock_net(sk);
1103 
1104 	bh_lock_sock(sk);
1105 
1106 	if (!ip_sk_accept_pmtu(sk))
1107 		goto out;
1108 
1109 	odst = sk_dst_get(sk);
1110 
1111 	if (sock_owned_by_user(sk) || !odst) {
1112 		__ipv4_sk_update_pmtu(skb, sk, mtu);
1113 		goto out;
1114 	}
1115 
1116 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1117 
1118 	rt = (struct rtable *)odst;
1119 	if (odst->obsolete && !odst->ops->check(odst, 0)) {
1120 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1121 		if (IS_ERR(rt))
1122 			goto out;
1123 
1124 		new = true;
1125 	}
1126 
1127 	__ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu);
1128 
1129 	if (!dst_check(&rt->dst, 0)) {
1130 		if (new)
1131 			dst_release(&rt->dst);
1132 
1133 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1134 		if (IS_ERR(rt))
1135 			goto out;
1136 
1137 		new = true;
1138 	}
1139 
1140 	if (new)
1141 		sk_dst_set(sk, &rt->dst);
1142 
1143 out:
1144 	bh_unlock_sock(sk);
1145 	dst_release(odst);
1146 }
1147 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1148 
1149 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1150 		   int oif, u8 protocol)
1151 {
1152 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1153 	struct flowi4 fl4;
1154 	struct rtable *rt;
1155 
1156 	__build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1157 	rt = __ip_route_output_key(net, &fl4);
1158 	if (!IS_ERR(rt)) {
1159 		__ip_do_redirect(rt, skb, &fl4, false);
1160 		ip_rt_put(rt);
1161 	}
1162 }
1163 EXPORT_SYMBOL_GPL(ipv4_redirect);
1164 
1165 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1166 {
1167 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1168 	struct flowi4 fl4;
1169 	struct rtable *rt;
1170 	struct net *net = sock_net(sk);
1171 
1172 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1173 	rt = __ip_route_output_key(net, &fl4);
1174 	if (!IS_ERR(rt)) {
1175 		__ip_do_redirect(rt, skb, &fl4, false);
1176 		ip_rt_put(rt);
1177 	}
1178 }
1179 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1180 
1181 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1182 							 u32 cookie)
1183 {
1184 	struct rtable *rt = (struct rtable *) dst;
1185 
1186 	/* All IPV4 dsts are created with ->obsolete set to the value
1187 	 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1188 	 * into this function always.
1189 	 *
1190 	 * When a PMTU/redirect information update invalidates a route,
1191 	 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1192 	 * DST_OBSOLETE_DEAD.
1193 	 */
1194 	if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1195 		return NULL;
1196 	return dst;
1197 }
1198 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1199 
1200 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1201 {
1202 	struct net_device *dev;
1203 	struct ip_options opt;
1204 	int res;
1205 
1206 	/* Recompile ip options since IPCB may not be valid anymore.
1207 	 * Also check we have a reasonable ipv4 header.
1208 	 */
1209 	if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1210 	    ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1211 		return;
1212 
1213 	memset(&opt, 0, sizeof(opt));
1214 	if (ip_hdr(skb)->ihl > 5) {
1215 		if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1216 			return;
1217 		opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1218 
1219 		rcu_read_lock();
1220 		dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1221 		res = __ip_options_compile(dev_net(dev), &opt, skb, NULL);
1222 		rcu_read_unlock();
1223 
1224 		if (res)
1225 			return;
1226 	}
1227 	__icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1228 }
1229 
1230 static void ipv4_link_failure(struct sk_buff *skb)
1231 {
1232 	struct rtable *rt;
1233 
1234 	ipv4_send_dest_unreach(skb);
1235 
1236 	rt = skb_rtable(skb);
1237 	if (rt)
1238 		dst_set_expires(&rt->dst, 0);
1239 }
1240 
1241 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1242 {
1243 	pr_debug("%s: %pI4 -> %pI4, %s\n",
1244 		 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1245 		 skb->dev ? skb->dev->name : "?");
1246 	kfree_skb(skb);
1247 	WARN_ON(1);
1248 	return 0;
1249 }
1250 
1251 /*
1252  * We do not cache source address of outgoing interface,
1253  * because it is used only by IP RR, TS and SRR options,
1254  * so that it out of fast path.
1255  *
1256  * BTW remember: "addr" is allowed to be not aligned
1257  * in IP options!
1258  */
1259 
1260 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1261 {
1262 	__be32 src;
1263 
1264 	if (rt_is_output_route(rt))
1265 		src = ip_hdr(skb)->saddr;
1266 	else {
1267 		struct fib_result res;
1268 		struct iphdr *iph = ip_hdr(skb);
1269 		struct flowi4 fl4 = {
1270 			.daddr = iph->daddr,
1271 			.saddr = iph->saddr,
1272 			.flowi4_tos = RT_TOS(iph->tos),
1273 			.flowi4_oif = rt->dst.dev->ifindex,
1274 			.flowi4_iif = skb->dev->ifindex,
1275 			.flowi4_mark = skb->mark,
1276 		};
1277 
1278 		rcu_read_lock();
1279 		if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1280 			src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1281 		else
1282 			src = inet_select_addr(rt->dst.dev,
1283 					       rt_nexthop(rt, iph->daddr),
1284 					       RT_SCOPE_UNIVERSE);
1285 		rcu_read_unlock();
1286 	}
1287 	memcpy(addr, &src, 4);
1288 }
1289 
1290 #ifdef CONFIG_IP_ROUTE_CLASSID
1291 static void set_class_tag(struct rtable *rt, u32 tag)
1292 {
1293 	if (!(rt->dst.tclassid & 0xFFFF))
1294 		rt->dst.tclassid |= tag & 0xFFFF;
1295 	if (!(rt->dst.tclassid & 0xFFFF0000))
1296 		rt->dst.tclassid |= tag & 0xFFFF0000;
1297 }
1298 #endif
1299 
1300 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1301 {
1302 	struct net *net = dev_net(dst->dev);
1303 	unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1304 	unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1305 				    net->ipv4.ip_rt_min_advmss);
1306 
1307 	return min(advmss, IPV4_MAX_PMTU - header_size);
1308 }
1309 
1310 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1311 {
1312 	return ip_dst_mtu_maybe_forward(dst, false);
1313 }
1314 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1315 
1316 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1317 {
1318 	struct fnhe_hash_bucket *hash;
1319 	struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1320 	u32 hval = fnhe_hashfun(daddr);
1321 
1322 	spin_lock_bh(&fnhe_lock);
1323 
1324 	hash = rcu_dereference_protected(nhc->nhc_exceptions,
1325 					 lockdep_is_held(&fnhe_lock));
1326 	hash += hval;
1327 
1328 	fnhe_p = &hash->chain;
1329 	fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1330 	while (fnhe) {
1331 		if (fnhe->fnhe_daddr == daddr) {
1332 			rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1333 				fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1334 			/* set fnhe_daddr to 0 to ensure it won't bind with
1335 			 * new dsts in rt_bind_exception().
1336 			 */
1337 			fnhe->fnhe_daddr = 0;
1338 			fnhe_flush_routes(fnhe);
1339 			kfree_rcu(fnhe, rcu);
1340 			break;
1341 		}
1342 		fnhe_p = &fnhe->fnhe_next;
1343 		fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1344 						 lockdep_is_held(&fnhe_lock));
1345 	}
1346 
1347 	spin_unlock_bh(&fnhe_lock);
1348 }
1349 
1350 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1351 					       __be32 daddr)
1352 {
1353 	struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1354 	struct fib_nh_exception *fnhe;
1355 	u32 hval;
1356 
1357 	if (!hash)
1358 		return NULL;
1359 
1360 	hval = fnhe_hashfun(daddr);
1361 
1362 	for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1363 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
1364 		if (fnhe->fnhe_daddr == daddr) {
1365 			if (fnhe->fnhe_expires &&
1366 			    time_after(jiffies, fnhe->fnhe_expires)) {
1367 				ip_del_fnhe(nhc, daddr);
1368 				break;
1369 			}
1370 			return fnhe;
1371 		}
1372 	}
1373 	return NULL;
1374 }
1375 
1376 /* MTU selection:
1377  * 1. mtu on route is locked - use it
1378  * 2. mtu from nexthop exception
1379  * 3. mtu from egress device
1380  */
1381 
1382 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1383 {
1384 	struct fib_nh_common *nhc = res->nhc;
1385 	struct net_device *dev = nhc->nhc_dev;
1386 	struct fib_info *fi = res->fi;
1387 	u32 mtu = 0;
1388 
1389 	if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1390 	    fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1391 		mtu = fi->fib_mtu;
1392 
1393 	if (likely(!mtu)) {
1394 		struct fib_nh_exception *fnhe;
1395 
1396 		fnhe = find_exception(nhc, daddr);
1397 		if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1398 			mtu = fnhe->fnhe_pmtu;
1399 	}
1400 
1401 	if (likely(!mtu))
1402 		mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1403 
1404 	return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1405 }
1406 
1407 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1408 			      __be32 daddr, const bool do_cache)
1409 {
1410 	bool ret = false;
1411 
1412 	spin_lock_bh(&fnhe_lock);
1413 
1414 	if (daddr == fnhe->fnhe_daddr) {
1415 		struct rtable __rcu **porig;
1416 		struct rtable *orig;
1417 		int genid = fnhe_genid(dev_net(rt->dst.dev));
1418 
1419 		if (rt_is_input_route(rt))
1420 			porig = &fnhe->fnhe_rth_input;
1421 		else
1422 			porig = &fnhe->fnhe_rth_output;
1423 		orig = rcu_dereference(*porig);
1424 
1425 		if (fnhe->fnhe_genid != genid) {
1426 			fnhe->fnhe_genid = genid;
1427 			fnhe->fnhe_gw = 0;
1428 			fnhe->fnhe_pmtu = 0;
1429 			fnhe->fnhe_expires = 0;
1430 			fnhe->fnhe_mtu_locked = false;
1431 			fnhe_flush_routes(fnhe);
1432 			orig = NULL;
1433 		}
1434 		fill_route_from_fnhe(rt, fnhe);
1435 		if (!rt->rt_gw4) {
1436 			rt->rt_gw4 = daddr;
1437 			rt->rt_gw_family = AF_INET;
1438 		}
1439 
1440 		if (do_cache) {
1441 			dst_hold(&rt->dst);
1442 			rcu_assign_pointer(*porig, rt);
1443 			if (orig) {
1444 				dst_dev_put(&orig->dst);
1445 				dst_release(&orig->dst);
1446 			}
1447 			ret = true;
1448 		}
1449 
1450 		fnhe->fnhe_stamp = jiffies;
1451 	}
1452 	spin_unlock_bh(&fnhe_lock);
1453 
1454 	return ret;
1455 }
1456 
1457 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1458 {
1459 	struct rtable *orig, *prev, **p;
1460 	bool ret = true;
1461 
1462 	if (rt_is_input_route(rt)) {
1463 		p = (struct rtable **)&nhc->nhc_rth_input;
1464 	} else {
1465 		p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1466 	}
1467 	orig = *p;
1468 
1469 	/* hold dst before doing cmpxchg() to avoid race condition
1470 	 * on this dst
1471 	 */
1472 	dst_hold(&rt->dst);
1473 	prev = cmpxchg(p, orig, rt);
1474 	if (prev == orig) {
1475 		if (orig) {
1476 			rt_add_uncached_list(orig);
1477 			dst_release(&orig->dst);
1478 		}
1479 	} else {
1480 		dst_release(&rt->dst);
1481 		ret = false;
1482 	}
1483 
1484 	return ret;
1485 }
1486 
1487 struct uncached_list {
1488 	spinlock_t		lock;
1489 	struct list_head	head;
1490 	struct list_head	quarantine;
1491 };
1492 
1493 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1494 
1495 void rt_add_uncached_list(struct rtable *rt)
1496 {
1497 	struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1498 
1499 	rt->dst.rt_uncached_list = ul;
1500 
1501 	spin_lock_bh(&ul->lock);
1502 	list_add_tail(&rt->dst.rt_uncached, &ul->head);
1503 	spin_unlock_bh(&ul->lock);
1504 }
1505 
1506 void rt_del_uncached_list(struct rtable *rt)
1507 {
1508 	if (!list_empty(&rt->dst.rt_uncached)) {
1509 		struct uncached_list *ul = rt->dst.rt_uncached_list;
1510 
1511 		spin_lock_bh(&ul->lock);
1512 		list_del_init(&rt->dst.rt_uncached);
1513 		spin_unlock_bh(&ul->lock);
1514 	}
1515 }
1516 
1517 static void ipv4_dst_destroy(struct dst_entry *dst)
1518 {
1519 	struct rtable *rt = (struct rtable *)dst;
1520 
1521 	ip_dst_metrics_put(dst);
1522 	rt_del_uncached_list(rt);
1523 }
1524 
1525 void rt_flush_dev(struct net_device *dev)
1526 {
1527 	struct rtable *rt, *safe;
1528 	int cpu;
1529 
1530 	for_each_possible_cpu(cpu) {
1531 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1532 
1533 		if (list_empty(&ul->head))
1534 			continue;
1535 
1536 		spin_lock_bh(&ul->lock);
1537 		list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1538 			if (rt->dst.dev != dev)
1539 				continue;
1540 			rt->dst.dev = blackhole_netdev;
1541 			netdev_ref_replace(dev, blackhole_netdev,
1542 					   &rt->dst.dev_tracker, GFP_ATOMIC);
1543 			list_move(&rt->dst.rt_uncached, &ul->quarantine);
1544 		}
1545 		spin_unlock_bh(&ul->lock);
1546 	}
1547 }
1548 
1549 static bool rt_cache_valid(const struct rtable *rt)
1550 {
1551 	return	rt &&
1552 		rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1553 		!rt_is_expired(rt);
1554 }
1555 
1556 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1557 			   const struct fib_result *res,
1558 			   struct fib_nh_exception *fnhe,
1559 			   struct fib_info *fi, u16 type, u32 itag,
1560 			   const bool do_cache)
1561 {
1562 	bool cached = false;
1563 
1564 	if (fi) {
1565 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1566 
1567 		if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1568 			rt->rt_uses_gateway = 1;
1569 			rt->rt_gw_family = nhc->nhc_gw_family;
1570 			/* only INET and INET6 are supported */
1571 			if (likely(nhc->nhc_gw_family == AF_INET))
1572 				rt->rt_gw4 = nhc->nhc_gw.ipv4;
1573 			else
1574 				rt->rt_gw6 = nhc->nhc_gw.ipv6;
1575 		}
1576 
1577 		ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1578 
1579 #ifdef CONFIG_IP_ROUTE_CLASSID
1580 		if (nhc->nhc_family == AF_INET) {
1581 			struct fib_nh *nh;
1582 
1583 			nh = container_of(nhc, struct fib_nh, nh_common);
1584 			rt->dst.tclassid = nh->nh_tclassid;
1585 		}
1586 #endif
1587 		rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1588 		if (unlikely(fnhe))
1589 			cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1590 		else if (do_cache)
1591 			cached = rt_cache_route(nhc, rt);
1592 		if (unlikely(!cached)) {
1593 			/* Routes we intend to cache in nexthop exception or
1594 			 * FIB nexthop have the DST_NOCACHE bit clear.
1595 			 * However, if we are unsuccessful at storing this
1596 			 * route into the cache we really need to set it.
1597 			 */
1598 			if (!rt->rt_gw4) {
1599 				rt->rt_gw_family = AF_INET;
1600 				rt->rt_gw4 = daddr;
1601 			}
1602 			rt_add_uncached_list(rt);
1603 		}
1604 	} else
1605 		rt_add_uncached_list(rt);
1606 
1607 #ifdef CONFIG_IP_ROUTE_CLASSID
1608 #ifdef CONFIG_IP_MULTIPLE_TABLES
1609 	set_class_tag(rt, res->tclassid);
1610 #endif
1611 	set_class_tag(rt, itag);
1612 #endif
1613 }
1614 
1615 struct rtable *rt_dst_alloc(struct net_device *dev,
1616 			    unsigned int flags, u16 type,
1617 			    bool noxfrm)
1618 {
1619 	struct rtable *rt;
1620 
1621 	rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1622 		       (noxfrm ? DST_NOXFRM : 0));
1623 
1624 	if (rt) {
1625 		rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1626 		rt->rt_flags = flags;
1627 		rt->rt_type = type;
1628 		rt->rt_is_input = 0;
1629 		rt->rt_iif = 0;
1630 		rt->rt_pmtu = 0;
1631 		rt->rt_mtu_locked = 0;
1632 		rt->rt_uses_gateway = 0;
1633 		rt->rt_gw_family = 0;
1634 		rt->rt_gw4 = 0;
1635 
1636 		rt->dst.output = ip_output;
1637 		if (flags & RTCF_LOCAL)
1638 			rt->dst.input = ip_local_deliver;
1639 	}
1640 
1641 	return rt;
1642 }
1643 EXPORT_SYMBOL(rt_dst_alloc);
1644 
1645 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1646 {
1647 	struct rtable *new_rt;
1648 
1649 	new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1650 			   rt->dst.flags);
1651 
1652 	if (new_rt) {
1653 		new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1654 		new_rt->rt_flags = rt->rt_flags;
1655 		new_rt->rt_type = rt->rt_type;
1656 		new_rt->rt_is_input = rt->rt_is_input;
1657 		new_rt->rt_iif = rt->rt_iif;
1658 		new_rt->rt_pmtu = rt->rt_pmtu;
1659 		new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1660 		new_rt->rt_gw_family = rt->rt_gw_family;
1661 		if (rt->rt_gw_family == AF_INET)
1662 			new_rt->rt_gw4 = rt->rt_gw4;
1663 		else if (rt->rt_gw_family == AF_INET6)
1664 			new_rt->rt_gw6 = rt->rt_gw6;
1665 
1666 		new_rt->dst.input = rt->dst.input;
1667 		new_rt->dst.output = rt->dst.output;
1668 		new_rt->dst.error = rt->dst.error;
1669 		new_rt->dst.lastuse = jiffies;
1670 		new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1671 	}
1672 	return new_rt;
1673 }
1674 EXPORT_SYMBOL(rt_dst_clone);
1675 
1676 /* called in rcu_read_lock() section */
1677 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1678 			  u8 tos, struct net_device *dev,
1679 			  struct in_device *in_dev, u32 *itag)
1680 {
1681 	int err;
1682 
1683 	/* Primary sanity checks. */
1684 	if (!in_dev)
1685 		return -EINVAL;
1686 
1687 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1688 	    skb->protocol != htons(ETH_P_IP))
1689 		return -EINVAL;
1690 
1691 	if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1692 		return -EINVAL;
1693 
1694 	if (ipv4_is_zeronet(saddr)) {
1695 		if (!ipv4_is_local_multicast(daddr) &&
1696 		    ip_hdr(skb)->protocol != IPPROTO_IGMP)
1697 			return -EINVAL;
1698 	} else {
1699 		err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1700 					  in_dev, itag);
1701 		if (err < 0)
1702 			return err;
1703 	}
1704 	return 0;
1705 }
1706 
1707 /* called in rcu_read_lock() section */
1708 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1709 			     u8 tos, struct net_device *dev, int our)
1710 {
1711 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1712 	unsigned int flags = RTCF_MULTICAST;
1713 	struct rtable *rth;
1714 	u32 itag = 0;
1715 	int err;
1716 
1717 	err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1718 	if (err)
1719 		return err;
1720 
1721 	if (our)
1722 		flags |= RTCF_LOCAL;
1723 
1724 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1725 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
1726 
1727 	rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1728 			   false);
1729 	if (!rth)
1730 		return -ENOBUFS;
1731 
1732 #ifdef CONFIG_IP_ROUTE_CLASSID
1733 	rth->dst.tclassid = itag;
1734 #endif
1735 	rth->dst.output = ip_rt_bug;
1736 	rth->rt_is_input= 1;
1737 
1738 #ifdef CONFIG_IP_MROUTE
1739 	if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1740 		rth->dst.input = ip_mr_input;
1741 #endif
1742 	RT_CACHE_STAT_INC(in_slow_mc);
1743 
1744 	skb_dst_drop(skb);
1745 	skb_dst_set(skb, &rth->dst);
1746 	return 0;
1747 }
1748 
1749 
1750 static void ip_handle_martian_source(struct net_device *dev,
1751 				     struct in_device *in_dev,
1752 				     struct sk_buff *skb,
1753 				     __be32 daddr,
1754 				     __be32 saddr)
1755 {
1756 	RT_CACHE_STAT_INC(in_martian_src);
1757 #ifdef CONFIG_IP_ROUTE_VERBOSE
1758 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1759 		/*
1760 		 *	RFC1812 recommendation, if source is martian,
1761 		 *	the only hint is MAC header.
1762 		 */
1763 		pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1764 			&daddr, &saddr, dev->name);
1765 		if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1766 			print_hex_dump(KERN_WARNING, "ll header: ",
1767 				       DUMP_PREFIX_OFFSET, 16, 1,
1768 				       skb_mac_header(skb),
1769 				       dev->hard_header_len, false);
1770 		}
1771 	}
1772 #endif
1773 }
1774 
1775 /* called in rcu_read_lock() section */
1776 static int __mkroute_input(struct sk_buff *skb,
1777 			   const struct fib_result *res,
1778 			   struct in_device *in_dev,
1779 			   __be32 daddr, __be32 saddr, u32 tos)
1780 {
1781 	struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1782 	struct net_device *dev = nhc->nhc_dev;
1783 	struct fib_nh_exception *fnhe;
1784 	struct rtable *rth;
1785 	int err;
1786 	struct in_device *out_dev;
1787 	bool do_cache;
1788 	u32 itag = 0;
1789 
1790 	/* get a working reference to the output device */
1791 	out_dev = __in_dev_get_rcu(dev);
1792 	if (!out_dev) {
1793 		net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1794 		return -EINVAL;
1795 	}
1796 
1797 	err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1798 				  in_dev->dev, in_dev, &itag);
1799 	if (err < 0) {
1800 		ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1801 					 saddr);
1802 
1803 		goto cleanup;
1804 	}
1805 
1806 	do_cache = res->fi && !itag;
1807 	if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1808 	    skb->protocol == htons(ETH_P_IP)) {
1809 		__be32 gw;
1810 
1811 		gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1812 		if (IN_DEV_SHARED_MEDIA(out_dev) ||
1813 		    inet_addr_onlink(out_dev, saddr, gw))
1814 			IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1815 	}
1816 
1817 	if (skb->protocol != htons(ETH_P_IP)) {
1818 		/* Not IP (i.e. ARP). Do not create route, if it is
1819 		 * invalid for proxy arp. DNAT routes are always valid.
1820 		 *
1821 		 * Proxy arp feature have been extended to allow, ARP
1822 		 * replies back to the same interface, to support
1823 		 * Private VLAN switch technologies. See arp.c.
1824 		 */
1825 		if (out_dev == in_dev &&
1826 		    IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1827 			err = -EINVAL;
1828 			goto cleanup;
1829 		}
1830 	}
1831 
1832 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1833 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
1834 
1835 	fnhe = find_exception(nhc, daddr);
1836 	if (do_cache) {
1837 		if (fnhe)
1838 			rth = rcu_dereference(fnhe->fnhe_rth_input);
1839 		else
1840 			rth = rcu_dereference(nhc->nhc_rth_input);
1841 		if (rt_cache_valid(rth)) {
1842 			skb_dst_set_noref(skb, &rth->dst);
1843 			goto out;
1844 		}
1845 	}
1846 
1847 	rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1848 			   IN_DEV_ORCONF(out_dev, NOXFRM));
1849 	if (!rth) {
1850 		err = -ENOBUFS;
1851 		goto cleanup;
1852 	}
1853 
1854 	rth->rt_is_input = 1;
1855 	RT_CACHE_STAT_INC(in_slow_tot);
1856 
1857 	rth->dst.input = ip_forward;
1858 
1859 	rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1860 		       do_cache);
1861 	lwtunnel_set_redirect(&rth->dst);
1862 	skb_dst_set(skb, &rth->dst);
1863 out:
1864 	err = 0;
1865  cleanup:
1866 	return err;
1867 }
1868 
1869 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1870 /* To make ICMP packets follow the right flow, the multipath hash is
1871  * calculated from the inner IP addresses.
1872  */
1873 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1874 				 struct flow_keys *hash_keys)
1875 {
1876 	const struct iphdr *outer_iph = ip_hdr(skb);
1877 	const struct iphdr *key_iph = outer_iph;
1878 	const struct iphdr *inner_iph;
1879 	const struct icmphdr *icmph;
1880 	struct iphdr _inner_iph;
1881 	struct icmphdr _icmph;
1882 
1883 	if (likely(outer_iph->protocol != IPPROTO_ICMP))
1884 		goto out;
1885 
1886 	if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1887 		goto out;
1888 
1889 	icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1890 				   &_icmph);
1891 	if (!icmph)
1892 		goto out;
1893 
1894 	if (!icmp_is_err(icmph->type))
1895 		goto out;
1896 
1897 	inner_iph = skb_header_pointer(skb,
1898 				       outer_iph->ihl * 4 + sizeof(_icmph),
1899 				       sizeof(_inner_iph), &_inner_iph);
1900 	if (!inner_iph)
1901 		goto out;
1902 
1903 	key_iph = inner_iph;
1904 out:
1905 	hash_keys->addrs.v4addrs.src = key_iph->saddr;
1906 	hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1907 }
1908 
1909 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1910 					   const struct sk_buff *skb,
1911 					   bool *p_has_inner)
1912 {
1913 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1914 	struct flow_keys keys, hash_keys;
1915 
1916 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1917 		return 0;
1918 
1919 	memset(&hash_keys, 0, sizeof(hash_keys));
1920 	skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1921 
1922 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1923 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1924 		hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1925 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1926 		hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1927 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1928 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
1929 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1930 		hash_keys.ports.src = keys.ports.src;
1931 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1932 		hash_keys.ports.dst = keys.ports.dst;
1933 
1934 	*p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1935 	return flow_hash_from_keys(&hash_keys);
1936 }
1937 
1938 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1939 					   const struct sk_buff *skb,
1940 					   bool has_inner)
1941 {
1942 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1943 	struct flow_keys keys, hash_keys;
1944 
1945 	/* We assume the packet carries an encapsulation, but if none was
1946 	 * encountered during dissection of the outer flow, then there is no
1947 	 * point in calling the flow dissector again.
1948 	 */
1949 	if (!has_inner)
1950 		return 0;
1951 
1952 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1953 		return 0;
1954 
1955 	memset(&hash_keys, 0, sizeof(hash_keys));
1956 	skb_flow_dissect_flow_keys(skb, &keys, 0);
1957 
1958 	if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1959 		return 0;
1960 
1961 	if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1962 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1963 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1964 			hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1965 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1966 			hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1967 	} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1968 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1969 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1970 			hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1971 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1972 			hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1973 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1974 			hash_keys.tags.flow_label = keys.tags.flow_label;
1975 	}
1976 
1977 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1978 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
1979 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1980 		hash_keys.ports.src = keys.ports.src;
1981 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
1982 		hash_keys.ports.dst = keys.ports.dst;
1983 
1984 	return flow_hash_from_keys(&hash_keys);
1985 }
1986 
1987 static u32 fib_multipath_custom_hash_skb(const struct net *net,
1988 					 const struct sk_buff *skb)
1989 {
1990 	u32 mhash, mhash_inner;
1991 	bool has_inner = true;
1992 
1993 	mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
1994 	mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
1995 
1996 	return jhash_2words(mhash, mhash_inner, 0);
1997 }
1998 
1999 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2000 					 const struct flowi4 *fl4)
2001 {
2002 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2003 	struct flow_keys hash_keys;
2004 
2005 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2006 		return 0;
2007 
2008 	memset(&hash_keys, 0, sizeof(hash_keys));
2009 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2010 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2011 		hash_keys.addrs.v4addrs.src = fl4->saddr;
2012 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2013 		hash_keys.addrs.v4addrs.dst = fl4->daddr;
2014 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2015 		hash_keys.basic.ip_proto = fl4->flowi4_proto;
2016 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2017 		hash_keys.ports.src = fl4->fl4_sport;
2018 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2019 		hash_keys.ports.dst = fl4->fl4_dport;
2020 
2021 	return flow_hash_from_keys(&hash_keys);
2022 }
2023 
2024 /* if skb is set it will be used and fl4 can be NULL */
2025 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2026 		       const struct sk_buff *skb, struct flow_keys *flkeys)
2027 {
2028 	u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2029 	struct flow_keys hash_keys;
2030 	u32 mhash = 0;
2031 
2032 	switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2033 	case 0:
2034 		memset(&hash_keys, 0, sizeof(hash_keys));
2035 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2036 		if (skb) {
2037 			ip_multipath_l3_keys(skb, &hash_keys);
2038 		} else {
2039 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2040 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2041 		}
2042 		mhash = flow_hash_from_keys(&hash_keys);
2043 		break;
2044 	case 1:
2045 		/* skb is currently provided only when forwarding */
2046 		if (skb) {
2047 			unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2048 			struct flow_keys keys;
2049 
2050 			/* short-circuit if we already have L4 hash present */
2051 			if (skb->l4_hash)
2052 				return skb_get_hash_raw(skb) >> 1;
2053 
2054 			memset(&hash_keys, 0, sizeof(hash_keys));
2055 
2056 			if (!flkeys) {
2057 				skb_flow_dissect_flow_keys(skb, &keys, flag);
2058 				flkeys = &keys;
2059 			}
2060 
2061 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2062 			hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2063 			hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2064 			hash_keys.ports.src = flkeys->ports.src;
2065 			hash_keys.ports.dst = flkeys->ports.dst;
2066 			hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2067 		} else {
2068 			memset(&hash_keys, 0, sizeof(hash_keys));
2069 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2070 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2071 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2072 			hash_keys.ports.src = fl4->fl4_sport;
2073 			hash_keys.ports.dst = fl4->fl4_dport;
2074 			hash_keys.basic.ip_proto = fl4->flowi4_proto;
2075 		}
2076 		mhash = flow_hash_from_keys(&hash_keys);
2077 		break;
2078 	case 2:
2079 		memset(&hash_keys, 0, sizeof(hash_keys));
2080 		/* skb is currently provided only when forwarding */
2081 		if (skb) {
2082 			struct flow_keys keys;
2083 
2084 			skb_flow_dissect_flow_keys(skb, &keys, 0);
2085 			/* Inner can be v4 or v6 */
2086 			if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2087 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2088 				hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2089 				hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2090 			} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2091 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2092 				hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2093 				hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2094 				hash_keys.tags.flow_label = keys.tags.flow_label;
2095 				hash_keys.basic.ip_proto = keys.basic.ip_proto;
2096 			} else {
2097 				/* Same as case 0 */
2098 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2099 				ip_multipath_l3_keys(skb, &hash_keys);
2100 			}
2101 		} else {
2102 			/* Same as case 0 */
2103 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2104 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2105 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2106 		}
2107 		mhash = flow_hash_from_keys(&hash_keys);
2108 		break;
2109 	case 3:
2110 		if (skb)
2111 			mhash = fib_multipath_custom_hash_skb(net, skb);
2112 		else
2113 			mhash = fib_multipath_custom_hash_fl4(net, fl4);
2114 		break;
2115 	}
2116 
2117 	if (multipath_hash)
2118 		mhash = jhash_2words(mhash, multipath_hash, 0);
2119 
2120 	return mhash >> 1;
2121 }
2122 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2123 
2124 static int ip_mkroute_input(struct sk_buff *skb,
2125 			    struct fib_result *res,
2126 			    struct in_device *in_dev,
2127 			    __be32 daddr, __be32 saddr, u32 tos,
2128 			    struct flow_keys *hkeys)
2129 {
2130 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2131 	if (res->fi && fib_info_num_path(res->fi) > 1) {
2132 		int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2133 
2134 		fib_select_multipath(res, h);
2135 		IPCB(skb)->flags |= IPSKB_MULTIPATH;
2136 	}
2137 #endif
2138 
2139 	/* create a routing cache entry */
2140 	return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2141 }
2142 
2143 /* Implements all the saddr-related checks as ip_route_input_slow(),
2144  * assuming daddr is valid and the destination is not a local broadcast one.
2145  * Uses the provided hint instead of performing a route lookup.
2146  */
2147 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2148 		      u8 tos, struct net_device *dev,
2149 		      const struct sk_buff *hint)
2150 {
2151 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2152 	struct rtable *rt = skb_rtable(hint);
2153 	struct net *net = dev_net(dev);
2154 	int err = -EINVAL;
2155 	u32 tag = 0;
2156 
2157 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2158 		goto martian_source;
2159 
2160 	if (ipv4_is_zeronet(saddr))
2161 		goto martian_source;
2162 
2163 	if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2164 		goto martian_source;
2165 
2166 	if (rt->rt_type != RTN_LOCAL)
2167 		goto skip_validate_source;
2168 
2169 	tos &= IPTOS_RT_MASK;
2170 	err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2171 	if (err < 0)
2172 		goto martian_source;
2173 
2174 skip_validate_source:
2175 	skb_dst_copy(skb, hint);
2176 	return 0;
2177 
2178 martian_source:
2179 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2180 	return err;
2181 }
2182 
2183 /* get device for dst_alloc with local routes */
2184 static struct net_device *ip_rt_get_dev(struct net *net,
2185 					const struct fib_result *res)
2186 {
2187 	struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2188 	struct net_device *dev = NULL;
2189 
2190 	if (nhc)
2191 		dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2192 
2193 	return dev ? : net->loopback_dev;
2194 }
2195 
2196 /*
2197  *	NOTE. We drop all the packets that has local source
2198  *	addresses, because every properly looped back packet
2199  *	must have correct destination already attached by output routine.
2200  *	Changes in the enforced policies must be applied also to
2201  *	ip_route_use_hint().
2202  *
2203  *	Such approach solves two big problems:
2204  *	1. Not simplex devices are handled properly.
2205  *	2. IP spoofing attempts are filtered with 100% of guarantee.
2206  *	called with rcu_read_lock()
2207  */
2208 
2209 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2210 			       u8 tos, struct net_device *dev,
2211 			       struct fib_result *res)
2212 {
2213 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2214 	struct flow_keys *flkeys = NULL, _flkeys;
2215 	struct net    *net = dev_net(dev);
2216 	struct ip_tunnel_info *tun_info;
2217 	int		err = -EINVAL;
2218 	unsigned int	flags = 0;
2219 	u32		itag = 0;
2220 	struct rtable	*rth;
2221 	struct flowi4	fl4;
2222 	bool do_cache = true;
2223 
2224 	/* IP on this device is disabled. */
2225 
2226 	if (!in_dev)
2227 		goto out;
2228 
2229 	/* Check for the most weird martians, which can be not detected
2230 	 * by fib_lookup.
2231 	 */
2232 
2233 	tun_info = skb_tunnel_info(skb);
2234 	if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2235 		fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2236 	else
2237 		fl4.flowi4_tun_key.tun_id = 0;
2238 	skb_dst_drop(skb);
2239 
2240 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2241 		goto martian_source;
2242 
2243 	res->fi = NULL;
2244 	res->table = NULL;
2245 	if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2246 		goto brd_input;
2247 
2248 	/* Accept zero addresses only to limited broadcast;
2249 	 * I even do not know to fix it or not. Waiting for complains :-)
2250 	 */
2251 	if (ipv4_is_zeronet(saddr))
2252 		goto martian_source;
2253 
2254 	if (ipv4_is_zeronet(daddr))
2255 		goto martian_destination;
2256 
2257 	/* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2258 	 * and call it once if daddr or/and saddr are loopback addresses
2259 	 */
2260 	if (ipv4_is_loopback(daddr)) {
2261 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2262 			goto martian_destination;
2263 	} else if (ipv4_is_loopback(saddr)) {
2264 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2265 			goto martian_source;
2266 	}
2267 
2268 	/*
2269 	 *	Now we are ready to route packet.
2270 	 */
2271 	fl4.flowi4_l3mdev = 0;
2272 	fl4.flowi4_oif = 0;
2273 	fl4.flowi4_iif = dev->ifindex;
2274 	fl4.flowi4_mark = skb->mark;
2275 	fl4.flowi4_tos = tos;
2276 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2277 	fl4.flowi4_flags = 0;
2278 	fl4.daddr = daddr;
2279 	fl4.saddr = saddr;
2280 	fl4.flowi4_uid = sock_net_uid(net, NULL);
2281 	fl4.flowi4_multipath_hash = 0;
2282 
2283 	if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2284 		flkeys = &_flkeys;
2285 	} else {
2286 		fl4.flowi4_proto = 0;
2287 		fl4.fl4_sport = 0;
2288 		fl4.fl4_dport = 0;
2289 	}
2290 
2291 	err = fib_lookup(net, &fl4, res, 0);
2292 	if (err != 0) {
2293 		if (!IN_DEV_FORWARD(in_dev))
2294 			err = -EHOSTUNREACH;
2295 		goto no_route;
2296 	}
2297 
2298 	if (res->type == RTN_BROADCAST) {
2299 		if (IN_DEV_BFORWARD(in_dev))
2300 			goto make_route;
2301 		/* not do cache if bc_forwarding is enabled */
2302 		if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING))
2303 			do_cache = false;
2304 		goto brd_input;
2305 	}
2306 
2307 	if (res->type == RTN_LOCAL) {
2308 		err = fib_validate_source(skb, saddr, daddr, tos,
2309 					  0, dev, in_dev, &itag);
2310 		if (err < 0)
2311 			goto martian_source;
2312 		goto local_input;
2313 	}
2314 
2315 	if (!IN_DEV_FORWARD(in_dev)) {
2316 		err = -EHOSTUNREACH;
2317 		goto no_route;
2318 	}
2319 	if (res->type != RTN_UNICAST)
2320 		goto martian_destination;
2321 
2322 make_route:
2323 	err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2324 out:	return err;
2325 
2326 brd_input:
2327 	if (skb->protocol != htons(ETH_P_IP))
2328 		goto e_inval;
2329 
2330 	if (!ipv4_is_zeronet(saddr)) {
2331 		err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2332 					  in_dev, &itag);
2333 		if (err < 0)
2334 			goto martian_source;
2335 	}
2336 	flags |= RTCF_BROADCAST;
2337 	res->type = RTN_BROADCAST;
2338 	RT_CACHE_STAT_INC(in_brd);
2339 
2340 local_input:
2341 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2342 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
2343 
2344 	do_cache &= res->fi && !itag;
2345 	if (do_cache) {
2346 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2347 
2348 		rth = rcu_dereference(nhc->nhc_rth_input);
2349 		if (rt_cache_valid(rth)) {
2350 			skb_dst_set_noref(skb, &rth->dst);
2351 			err = 0;
2352 			goto out;
2353 		}
2354 	}
2355 
2356 	rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2357 			   flags | RTCF_LOCAL, res->type, false);
2358 	if (!rth)
2359 		goto e_nobufs;
2360 
2361 	rth->dst.output= ip_rt_bug;
2362 #ifdef CONFIG_IP_ROUTE_CLASSID
2363 	rth->dst.tclassid = itag;
2364 #endif
2365 	rth->rt_is_input = 1;
2366 
2367 	RT_CACHE_STAT_INC(in_slow_tot);
2368 	if (res->type == RTN_UNREACHABLE) {
2369 		rth->dst.input= ip_error;
2370 		rth->dst.error= -err;
2371 		rth->rt_flags	&= ~RTCF_LOCAL;
2372 	}
2373 
2374 	if (do_cache) {
2375 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2376 
2377 		rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2378 		if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2379 			WARN_ON(rth->dst.input == lwtunnel_input);
2380 			rth->dst.lwtstate->orig_input = rth->dst.input;
2381 			rth->dst.input = lwtunnel_input;
2382 		}
2383 
2384 		if (unlikely(!rt_cache_route(nhc, rth)))
2385 			rt_add_uncached_list(rth);
2386 	}
2387 	skb_dst_set(skb, &rth->dst);
2388 	err = 0;
2389 	goto out;
2390 
2391 no_route:
2392 	RT_CACHE_STAT_INC(in_no_route);
2393 	res->type = RTN_UNREACHABLE;
2394 	res->fi = NULL;
2395 	res->table = NULL;
2396 	goto local_input;
2397 
2398 	/*
2399 	 *	Do not cache martian addresses: they should be logged (RFC1812)
2400 	 */
2401 martian_destination:
2402 	RT_CACHE_STAT_INC(in_martian_dst);
2403 #ifdef CONFIG_IP_ROUTE_VERBOSE
2404 	if (IN_DEV_LOG_MARTIANS(in_dev))
2405 		net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2406 				     &daddr, &saddr, dev->name);
2407 #endif
2408 
2409 e_inval:
2410 	err = -EINVAL;
2411 	goto out;
2412 
2413 e_nobufs:
2414 	err = -ENOBUFS;
2415 	goto out;
2416 
2417 martian_source:
2418 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2419 	goto out;
2420 }
2421 
2422 /* called with rcu_read_lock held */
2423 static int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2424 			      u8 tos, struct net_device *dev, struct fib_result *res)
2425 {
2426 	/* Multicast recognition logic is moved from route cache to here.
2427 	 * The problem was that too many Ethernet cards have broken/missing
2428 	 * hardware multicast filters :-( As result the host on multicasting
2429 	 * network acquires a lot of useless route cache entries, sort of
2430 	 * SDR messages from all the world. Now we try to get rid of them.
2431 	 * Really, provided software IP multicast filter is organized
2432 	 * reasonably (at least, hashed), it does not result in a slowdown
2433 	 * comparing with route cache reject entries.
2434 	 * Note, that multicast routers are not affected, because
2435 	 * route cache entry is created eventually.
2436 	 */
2437 	if (ipv4_is_multicast(daddr)) {
2438 		struct in_device *in_dev = __in_dev_get_rcu(dev);
2439 		int our = 0;
2440 		int err = -EINVAL;
2441 
2442 		if (!in_dev)
2443 			return err;
2444 		our = ip_check_mc_rcu(in_dev, daddr, saddr,
2445 				      ip_hdr(skb)->protocol);
2446 
2447 		/* check l3 master if no match yet */
2448 		if (!our && netif_is_l3_slave(dev)) {
2449 			struct in_device *l3_in_dev;
2450 
2451 			l3_in_dev = __in_dev_get_rcu(skb->dev);
2452 			if (l3_in_dev)
2453 				our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2454 						      ip_hdr(skb)->protocol);
2455 		}
2456 
2457 		if (our
2458 #ifdef CONFIG_IP_MROUTE
2459 			||
2460 		    (!ipv4_is_local_multicast(daddr) &&
2461 		     IN_DEV_MFORWARD(in_dev))
2462 #endif
2463 		   ) {
2464 			err = ip_route_input_mc(skb, daddr, saddr,
2465 						tos, dev, our);
2466 		}
2467 		return err;
2468 	}
2469 
2470 	return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2471 }
2472 
2473 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2474 			 u8 tos, struct net_device *dev)
2475 {
2476 	struct fib_result res;
2477 	int err;
2478 
2479 	tos &= IPTOS_RT_MASK;
2480 	rcu_read_lock();
2481 	err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2482 	rcu_read_unlock();
2483 
2484 	return err;
2485 }
2486 EXPORT_SYMBOL(ip_route_input_noref);
2487 
2488 /* called with rcu_read_lock() */
2489 static struct rtable *__mkroute_output(const struct fib_result *res,
2490 				       const struct flowi4 *fl4, int orig_oif,
2491 				       struct net_device *dev_out,
2492 				       unsigned int flags)
2493 {
2494 	struct fib_info *fi = res->fi;
2495 	struct fib_nh_exception *fnhe;
2496 	struct in_device *in_dev;
2497 	u16 type = res->type;
2498 	struct rtable *rth;
2499 	bool do_cache;
2500 
2501 	in_dev = __in_dev_get_rcu(dev_out);
2502 	if (!in_dev)
2503 		return ERR_PTR(-EINVAL);
2504 
2505 	if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2506 		if (ipv4_is_loopback(fl4->saddr) &&
2507 		    !(dev_out->flags & IFF_LOOPBACK) &&
2508 		    !netif_is_l3_master(dev_out))
2509 			return ERR_PTR(-EINVAL);
2510 
2511 	if (ipv4_is_lbcast(fl4->daddr))
2512 		type = RTN_BROADCAST;
2513 	else if (ipv4_is_multicast(fl4->daddr))
2514 		type = RTN_MULTICAST;
2515 	else if (ipv4_is_zeronet(fl4->daddr))
2516 		return ERR_PTR(-EINVAL);
2517 
2518 	if (dev_out->flags & IFF_LOOPBACK)
2519 		flags |= RTCF_LOCAL;
2520 
2521 	do_cache = true;
2522 	if (type == RTN_BROADCAST) {
2523 		flags |= RTCF_BROADCAST | RTCF_LOCAL;
2524 		fi = NULL;
2525 	} else if (type == RTN_MULTICAST) {
2526 		flags |= RTCF_MULTICAST | RTCF_LOCAL;
2527 		if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2528 				     fl4->flowi4_proto))
2529 			flags &= ~RTCF_LOCAL;
2530 		else
2531 			do_cache = false;
2532 		/* If multicast route do not exist use
2533 		 * default one, but do not gateway in this case.
2534 		 * Yes, it is hack.
2535 		 */
2536 		if (fi && res->prefixlen < 4)
2537 			fi = NULL;
2538 	} else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2539 		   (orig_oif != dev_out->ifindex)) {
2540 		/* For local routes that require a particular output interface
2541 		 * we do not want to cache the result.  Caching the result
2542 		 * causes incorrect behaviour when there are multiple source
2543 		 * addresses on the interface, the end result being that if the
2544 		 * intended recipient is waiting on that interface for the
2545 		 * packet he won't receive it because it will be delivered on
2546 		 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2547 		 * be set to the loopback interface as well.
2548 		 */
2549 		do_cache = false;
2550 	}
2551 
2552 	fnhe = NULL;
2553 	do_cache &= fi != NULL;
2554 	if (fi) {
2555 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2556 		struct rtable __rcu **prth;
2557 
2558 		fnhe = find_exception(nhc, fl4->daddr);
2559 		if (!do_cache)
2560 			goto add;
2561 		if (fnhe) {
2562 			prth = &fnhe->fnhe_rth_output;
2563 		} else {
2564 			if (unlikely(fl4->flowi4_flags &
2565 				     FLOWI_FLAG_KNOWN_NH &&
2566 				     !(nhc->nhc_gw_family &&
2567 				       nhc->nhc_scope == RT_SCOPE_LINK))) {
2568 				do_cache = false;
2569 				goto add;
2570 			}
2571 			prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2572 		}
2573 		rth = rcu_dereference(*prth);
2574 		if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2575 			return rth;
2576 	}
2577 
2578 add:
2579 	rth = rt_dst_alloc(dev_out, flags, type,
2580 			   IN_DEV_ORCONF(in_dev, NOXFRM));
2581 	if (!rth)
2582 		return ERR_PTR(-ENOBUFS);
2583 
2584 	rth->rt_iif = orig_oif;
2585 
2586 	RT_CACHE_STAT_INC(out_slow_tot);
2587 
2588 	if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2589 		if (flags & RTCF_LOCAL &&
2590 		    !(dev_out->flags & IFF_LOOPBACK)) {
2591 			rth->dst.output = ip_mc_output;
2592 			RT_CACHE_STAT_INC(out_slow_mc);
2593 		}
2594 #ifdef CONFIG_IP_MROUTE
2595 		if (type == RTN_MULTICAST) {
2596 			if (IN_DEV_MFORWARD(in_dev) &&
2597 			    !ipv4_is_local_multicast(fl4->daddr)) {
2598 				rth->dst.input = ip_mr_input;
2599 				rth->dst.output = ip_mc_output;
2600 			}
2601 		}
2602 #endif
2603 	}
2604 
2605 	rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2606 	lwtunnel_set_redirect(&rth->dst);
2607 
2608 	return rth;
2609 }
2610 
2611 /*
2612  * Major route resolver routine.
2613  */
2614 
2615 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2616 					const struct sk_buff *skb)
2617 {
2618 	struct fib_result res = {
2619 		.type		= RTN_UNSPEC,
2620 		.fi		= NULL,
2621 		.table		= NULL,
2622 		.tclassid	= 0,
2623 	};
2624 	struct rtable *rth;
2625 
2626 	fl4->flowi4_iif = LOOPBACK_IFINDEX;
2627 	fl4->flowi4_tos &= IPTOS_RT_MASK;
2628 
2629 	rcu_read_lock();
2630 	rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2631 	rcu_read_unlock();
2632 
2633 	return rth;
2634 }
2635 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2636 
2637 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2638 					    struct fib_result *res,
2639 					    const struct sk_buff *skb)
2640 {
2641 	struct net_device *dev_out = NULL;
2642 	int orig_oif = fl4->flowi4_oif;
2643 	unsigned int flags = 0;
2644 	struct rtable *rth;
2645 	int err;
2646 
2647 	if (fl4->saddr) {
2648 		if (ipv4_is_multicast(fl4->saddr) ||
2649 		    ipv4_is_lbcast(fl4->saddr) ||
2650 		    ipv4_is_zeronet(fl4->saddr)) {
2651 			rth = ERR_PTR(-EINVAL);
2652 			goto out;
2653 		}
2654 
2655 		rth = ERR_PTR(-ENETUNREACH);
2656 
2657 		/* I removed check for oif == dev_out->oif here.
2658 		 * It was wrong for two reasons:
2659 		 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2660 		 *    is assigned to multiple interfaces.
2661 		 * 2. Moreover, we are allowed to send packets with saddr
2662 		 *    of another iface. --ANK
2663 		 */
2664 
2665 		if (fl4->flowi4_oif == 0 &&
2666 		    (ipv4_is_multicast(fl4->daddr) ||
2667 		     ipv4_is_lbcast(fl4->daddr))) {
2668 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2669 			dev_out = __ip_dev_find(net, fl4->saddr, false);
2670 			if (!dev_out)
2671 				goto out;
2672 
2673 			/* Special hack: user can direct multicasts
2674 			 * and limited broadcast via necessary interface
2675 			 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2676 			 * This hack is not just for fun, it allows
2677 			 * vic,vat and friends to work.
2678 			 * They bind socket to loopback, set ttl to zero
2679 			 * and expect that it will work.
2680 			 * From the viewpoint of routing cache they are broken,
2681 			 * because we are not allowed to build multicast path
2682 			 * with loopback source addr (look, routing cache
2683 			 * cannot know, that ttl is zero, so that packet
2684 			 * will not leave this host and route is valid).
2685 			 * Luckily, this hack is good workaround.
2686 			 */
2687 
2688 			fl4->flowi4_oif = dev_out->ifindex;
2689 			goto make_route;
2690 		}
2691 
2692 		if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2693 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2694 			if (!__ip_dev_find(net, fl4->saddr, false))
2695 				goto out;
2696 		}
2697 	}
2698 
2699 
2700 	if (fl4->flowi4_oif) {
2701 		dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2702 		rth = ERR_PTR(-ENODEV);
2703 		if (!dev_out)
2704 			goto out;
2705 
2706 		/* RACE: Check return value of inet_select_addr instead. */
2707 		if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2708 			rth = ERR_PTR(-ENETUNREACH);
2709 			goto out;
2710 		}
2711 		if (ipv4_is_local_multicast(fl4->daddr) ||
2712 		    ipv4_is_lbcast(fl4->daddr) ||
2713 		    fl4->flowi4_proto == IPPROTO_IGMP) {
2714 			if (!fl4->saddr)
2715 				fl4->saddr = inet_select_addr(dev_out, 0,
2716 							      RT_SCOPE_LINK);
2717 			goto make_route;
2718 		}
2719 		if (!fl4->saddr) {
2720 			if (ipv4_is_multicast(fl4->daddr))
2721 				fl4->saddr = inet_select_addr(dev_out, 0,
2722 							      fl4->flowi4_scope);
2723 			else if (!fl4->daddr)
2724 				fl4->saddr = inet_select_addr(dev_out, 0,
2725 							      RT_SCOPE_HOST);
2726 		}
2727 	}
2728 
2729 	if (!fl4->daddr) {
2730 		fl4->daddr = fl4->saddr;
2731 		if (!fl4->daddr)
2732 			fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2733 		dev_out = net->loopback_dev;
2734 		fl4->flowi4_oif = LOOPBACK_IFINDEX;
2735 		res->type = RTN_LOCAL;
2736 		flags |= RTCF_LOCAL;
2737 		goto make_route;
2738 	}
2739 
2740 	err = fib_lookup(net, fl4, res, 0);
2741 	if (err) {
2742 		res->fi = NULL;
2743 		res->table = NULL;
2744 		if (fl4->flowi4_oif &&
2745 		    (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2746 			/* Apparently, routing tables are wrong. Assume,
2747 			 * that the destination is on link.
2748 			 *
2749 			 * WHY? DW.
2750 			 * Because we are allowed to send to iface
2751 			 * even if it has NO routes and NO assigned
2752 			 * addresses. When oif is specified, routing
2753 			 * tables are looked up with only one purpose:
2754 			 * to catch if destination is gatewayed, rather than
2755 			 * direct. Moreover, if MSG_DONTROUTE is set,
2756 			 * we send packet, ignoring both routing tables
2757 			 * and ifaddr state. --ANK
2758 			 *
2759 			 *
2760 			 * We could make it even if oif is unknown,
2761 			 * likely IPv6, but we do not.
2762 			 */
2763 
2764 			if (fl4->saddr == 0)
2765 				fl4->saddr = inet_select_addr(dev_out, 0,
2766 							      RT_SCOPE_LINK);
2767 			res->type = RTN_UNICAST;
2768 			goto make_route;
2769 		}
2770 		rth = ERR_PTR(err);
2771 		goto out;
2772 	}
2773 
2774 	if (res->type == RTN_LOCAL) {
2775 		if (!fl4->saddr) {
2776 			if (res->fi->fib_prefsrc)
2777 				fl4->saddr = res->fi->fib_prefsrc;
2778 			else
2779 				fl4->saddr = fl4->daddr;
2780 		}
2781 
2782 		/* L3 master device is the loopback for that domain */
2783 		dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2784 			net->loopback_dev;
2785 
2786 		/* make sure orig_oif points to fib result device even
2787 		 * though packet rx/tx happens over loopback or l3mdev
2788 		 */
2789 		orig_oif = FIB_RES_OIF(*res);
2790 
2791 		fl4->flowi4_oif = dev_out->ifindex;
2792 		flags |= RTCF_LOCAL;
2793 		goto make_route;
2794 	}
2795 
2796 	fib_select_path(net, res, fl4, skb);
2797 
2798 	dev_out = FIB_RES_DEV(*res);
2799 
2800 make_route:
2801 	rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2802 
2803 out:
2804 	return rth;
2805 }
2806 
2807 static struct dst_ops ipv4_dst_blackhole_ops = {
2808 	.family			= AF_INET,
2809 	.default_advmss		= ipv4_default_advmss,
2810 	.neigh_lookup		= ipv4_neigh_lookup,
2811 	.check			= dst_blackhole_check,
2812 	.cow_metrics		= dst_blackhole_cow_metrics,
2813 	.update_pmtu		= dst_blackhole_update_pmtu,
2814 	.redirect		= dst_blackhole_redirect,
2815 	.mtu			= dst_blackhole_mtu,
2816 };
2817 
2818 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2819 {
2820 	struct rtable *ort = (struct rtable *) dst_orig;
2821 	struct rtable *rt;
2822 
2823 	rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0);
2824 	if (rt) {
2825 		struct dst_entry *new = &rt->dst;
2826 
2827 		new->__use = 1;
2828 		new->input = dst_discard;
2829 		new->output = dst_discard_out;
2830 
2831 		new->dev = net->loopback_dev;
2832 		netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2833 
2834 		rt->rt_is_input = ort->rt_is_input;
2835 		rt->rt_iif = ort->rt_iif;
2836 		rt->rt_pmtu = ort->rt_pmtu;
2837 		rt->rt_mtu_locked = ort->rt_mtu_locked;
2838 
2839 		rt->rt_genid = rt_genid_ipv4(net);
2840 		rt->rt_flags = ort->rt_flags;
2841 		rt->rt_type = ort->rt_type;
2842 		rt->rt_uses_gateway = ort->rt_uses_gateway;
2843 		rt->rt_gw_family = ort->rt_gw_family;
2844 		if (rt->rt_gw_family == AF_INET)
2845 			rt->rt_gw4 = ort->rt_gw4;
2846 		else if (rt->rt_gw_family == AF_INET6)
2847 			rt->rt_gw6 = ort->rt_gw6;
2848 	}
2849 
2850 	dst_release(dst_orig);
2851 
2852 	return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2853 }
2854 
2855 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2856 				    const struct sock *sk)
2857 {
2858 	struct rtable *rt = __ip_route_output_key(net, flp4);
2859 
2860 	if (IS_ERR(rt))
2861 		return rt;
2862 
2863 	if (flp4->flowi4_proto) {
2864 		flp4->flowi4_oif = rt->dst.dev->ifindex;
2865 		rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2866 							flowi4_to_flowi(flp4),
2867 							sk, 0);
2868 	}
2869 
2870 	return rt;
2871 }
2872 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2873 
2874 /* called with rcu_read_lock held */
2875 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2876 			struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2877 			struct sk_buff *skb, u32 portid, u32 seq,
2878 			unsigned int flags)
2879 {
2880 	struct rtmsg *r;
2881 	struct nlmsghdr *nlh;
2882 	unsigned long expires = 0;
2883 	u32 error;
2884 	u32 metrics[RTAX_MAX];
2885 
2886 	nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2887 	if (!nlh)
2888 		return -EMSGSIZE;
2889 
2890 	r = nlmsg_data(nlh);
2891 	r->rtm_family	 = AF_INET;
2892 	r->rtm_dst_len	= 32;
2893 	r->rtm_src_len	= 0;
2894 	r->rtm_tos	= fl4 ? fl4->flowi4_tos : 0;
2895 	r->rtm_table	= table_id < 256 ? table_id : RT_TABLE_COMPAT;
2896 	if (nla_put_u32(skb, RTA_TABLE, table_id))
2897 		goto nla_put_failure;
2898 	r->rtm_type	= rt->rt_type;
2899 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2900 	r->rtm_protocol = RTPROT_UNSPEC;
2901 	r->rtm_flags	= (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2902 	if (rt->rt_flags & RTCF_NOTIFY)
2903 		r->rtm_flags |= RTM_F_NOTIFY;
2904 	if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2905 		r->rtm_flags |= RTCF_DOREDIRECT;
2906 
2907 	if (nla_put_in_addr(skb, RTA_DST, dst))
2908 		goto nla_put_failure;
2909 	if (src) {
2910 		r->rtm_src_len = 32;
2911 		if (nla_put_in_addr(skb, RTA_SRC, src))
2912 			goto nla_put_failure;
2913 	}
2914 	if (rt->dst.dev &&
2915 	    nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2916 		goto nla_put_failure;
2917 	if (rt->dst.lwtstate &&
2918 	    lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2919 		goto nla_put_failure;
2920 #ifdef CONFIG_IP_ROUTE_CLASSID
2921 	if (rt->dst.tclassid &&
2922 	    nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2923 		goto nla_put_failure;
2924 #endif
2925 	if (fl4 && !rt_is_input_route(rt) &&
2926 	    fl4->saddr != src) {
2927 		if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2928 			goto nla_put_failure;
2929 	}
2930 	if (rt->rt_uses_gateway) {
2931 		if (rt->rt_gw_family == AF_INET &&
2932 		    nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2933 			goto nla_put_failure;
2934 		} else if (rt->rt_gw_family == AF_INET6) {
2935 			int alen = sizeof(struct in6_addr);
2936 			struct nlattr *nla;
2937 			struct rtvia *via;
2938 
2939 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
2940 			if (!nla)
2941 				goto nla_put_failure;
2942 
2943 			via = nla_data(nla);
2944 			via->rtvia_family = AF_INET6;
2945 			memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2946 		}
2947 	}
2948 
2949 	expires = rt->dst.expires;
2950 	if (expires) {
2951 		unsigned long now = jiffies;
2952 
2953 		if (time_before(now, expires))
2954 			expires -= now;
2955 		else
2956 			expires = 0;
2957 	}
2958 
2959 	memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2960 	if (rt->rt_pmtu && expires)
2961 		metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2962 	if (rt->rt_mtu_locked && expires)
2963 		metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2964 	if (rtnetlink_put_metrics(skb, metrics) < 0)
2965 		goto nla_put_failure;
2966 
2967 	if (fl4) {
2968 		if (fl4->flowi4_mark &&
2969 		    nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2970 			goto nla_put_failure;
2971 
2972 		if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2973 		    nla_put_u32(skb, RTA_UID,
2974 				from_kuid_munged(current_user_ns(),
2975 						 fl4->flowi4_uid)))
2976 			goto nla_put_failure;
2977 
2978 		if (rt_is_input_route(rt)) {
2979 #ifdef CONFIG_IP_MROUTE
2980 			if (ipv4_is_multicast(dst) &&
2981 			    !ipv4_is_local_multicast(dst) &&
2982 			    IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) {
2983 				int err = ipmr_get_route(net, skb,
2984 							 fl4->saddr, fl4->daddr,
2985 							 r, portid);
2986 
2987 				if (err <= 0) {
2988 					if (err == 0)
2989 						return 0;
2990 					goto nla_put_failure;
2991 				}
2992 			} else
2993 #endif
2994 				if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
2995 					goto nla_put_failure;
2996 		}
2997 	}
2998 
2999 	error = rt->dst.error;
3000 
3001 	if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3002 		goto nla_put_failure;
3003 
3004 	nlmsg_end(skb, nlh);
3005 	return 0;
3006 
3007 nla_put_failure:
3008 	nlmsg_cancel(skb, nlh);
3009 	return -EMSGSIZE;
3010 }
3011 
3012 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3013 			    struct netlink_callback *cb, u32 table_id,
3014 			    struct fnhe_hash_bucket *bucket, int genid,
3015 			    int *fa_index, int fa_start, unsigned int flags)
3016 {
3017 	int i;
3018 
3019 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
3020 		struct fib_nh_exception *fnhe;
3021 
3022 		for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3023 		     fnhe = rcu_dereference(fnhe->fnhe_next)) {
3024 			struct rtable *rt;
3025 			int err;
3026 
3027 			if (*fa_index < fa_start)
3028 				goto next;
3029 
3030 			if (fnhe->fnhe_genid != genid)
3031 				goto next;
3032 
3033 			if (fnhe->fnhe_expires &&
3034 			    time_after(jiffies, fnhe->fnhe_expires))
3035 				goto next;
3036 
3037 			rt = rcu_dereference(fnhe->fnhe_rth_input);
3038 			if (!rt)
3039 				rt = rcu_dereference(fnhe->fnhe_rth_output);
3040 			if (!rt)
3041 				goto next;
3042 
3043 			err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3044 					   table_id, NULL, skb,
3045 					   NETLINK_CB(cb->skb).portid,
3046 					   cb->nlh->nlmsg_seq, flags);
3047 			if (err)
3048 				return err;
3049 next:
3050 			(*fa_index)++;
3051 		}
3052 	}
3053 
3054 	return 0;
3055 }
3056 
3057 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3058 		       u32 table_id, struct fib_info *fi,
3059 		       int *fa_index, int fa_start, unsigned int flags)
3060 {
3061 	struct net *net = sock_net(cb->skb->sk);
3062 	int nhsel, genid = fnhe_genid(net);
3063 
3064 	for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3065 		struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3066 		struct fnhe_hash_bucket *bucket;
3067 		int err;
3068 
3069 		if (nhc->nhc_flags & RTNH_F_DEAD)
3070 			continue;
3071 
3072 		rcu_read_lock();
3073 		bucket = rcu_dereference(nhc->nhc_exceptions);
3074 		err = 0;
3075 		if (bucket)
3076 			err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3077 					       genid, fa_index, fa_start,
3078 					       flags);
3079 		rcu_read_unlock();
3080 		if (err)
3081 			return err;
3082 	}
3083 
3084 	return 0;
3085 }
3086 
3087 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3088 						   u8 ip_proto, __be16 sport,
3089 						   __be16 dport)
3090 {
3091 	struct sk_buff *skb;
3092 	struct iphdr *iph;
3093 
3094 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3095 	if (!skb)
3096 		return NULL;
3097 
3098 	/* Reserve room for dummy headers, this skb can pass
3099 	 * through good chunk of routing engine.
3100 	 */
3101 	skb_reset_mac_header(skb);
3102 	skb_reset_network_header(skb);
3103 	skb->protocol = htons(ETH_P_IP);
3104 	iph = skb_put(skb, sizeof(struct iphdr));
3105 	iph->protocol = ip_proto;
3106 	iph->saddr = src;
3107 	iph->daddr = dst;
3108 	iph->version = 0x4;
3109 	iph->frag_off = 0;
3110 	iph->ihl = 0x5;
3111 	skb_set_transport_header(skb, skb->len);
3112 
3113 	switch (iph->protocol) {
3114 	case IPPROTO_UDP: {
3115 		struct udphdr *udph;
3116 
3117 		udph = skb_put_zero(skb, sizeof(struct udphdr));
3118 		udph->source = sport;
3119 		udph->dest = dport;
3120 		udph->len = htons(sizeof(struct udphdr));
3121 		udph->check = 0;
3122 		break;
3123 	}
3124 	case IPPROTO_TCP: {
3125 		struct tcphdr *tcph;
3126 
3127 		tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3128 		tcph->source	= sport;
3129 		tcph->dest	= dport;
3130 		tcph->doff	= sizeof(struct tcphdr) / 4;
3131 		tcph->rst = 1;
3132 		tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3133 					    src, dst, 0);
3134 		break;
3135 	}
3136 	case IPPROTO_ICMP: {
3137 		struct icmphdr *icmph;
3138 
3139 		icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3140 		icmph->type = ICMP_ECHO;
3141 		icmph->code = 0;
3142 	}
3143 	}
3144 
3145 	return skb;
3146 }
3147 
3148 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3149 				       const struct nlmsghdr *nlh,
3150 				       struct nlattr **tb,
3151 				       struct netlink_ext_ack *extack)
3152 {
3153 	struct rtmsg *rtm;
3154 	int i, err;
3155 
3156 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3157 		NL_SET_ERR_MSG(extack,
3158 			       "ipv4: Invalid header for route get request");
3159 		return -EINVAL;
3160 	}
3161 
3162 	if (!netlink_strict_get_check(skb))
3163 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3164 					      rtm_ipv4_policy, extack);
3165 
3166 	rtm = nlmsg_data(nlh);
3167 	if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3168 	    (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3169 	    rtm->rtm_table || rtm->rtm_protocol ||
3170 	    rtm->rtm_scope || rtm->rtm_type) {
3171 		NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3172 		return -EINVAL;
3173 	}
3174 
3175 	if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3176 			       RTM_F_LOOKUP_TABLE |
3177 			       RTM_F_FIB_MATCH)) {
3178 		NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3179 		return -EINVAL;
3180 	}
3181 
3182 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3183 					    rtm_ipv4_policy, extack);
3184 	if (err)
3185 		return err;
3186 
3187 	if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3188 	    (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3189 		NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3190 		return -EINVAL;
3191 	}
3192 
3193 	for (i = 0; i <= RTA_MAX; i++) {
3194 		if (!tb[i])
3195 			continue;
3196 
3197 		switch (i) {
3198 		case RTA_IIF:
3199 		case RTA_OIF:
3200 		case RTA_SRC:
3201 		case RTA_DST:
3202 		case RTA_IP_PROTO:
3203 		case RTA_SPORT:
3204 		case RTA_DPORT:
3205 		case RTA_MARK:
3206 		case RTA_UID:
3207 			break;
3208 		default:
3209 			NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3210 			return -EINVAL;
3211 		}
3212 	}
3213 
3214 	return 0;
3215 }
3216 
3217 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3218 			     struct netlink_ext_ack *extack)
3219 {
3220 	struct net *net = sock_net(in_skb->sk);
3221 	struct nlattr *tb[RTA_MAX+1];
3222 	u32 table_id = RT_TABLE_MAIN;
3223 	__be16 sport = 0, dport = 0;
3224 	struct fib_result res = {};
3225 	u8 ip_proto = IPPROTO_UDP;
3226 	struct rtable *rt = NULL;
3227 	struct sk_buff *skb;
3228 	struct rtmsg *rtm;
3229 	struct flowi4 fl4 = {};
3230 	__be32 dst = 0;
3231 	__be32 src = 0;
3232 	kuid_t uid;
3233 	u32 iif;
3234 	int err;
3235 	int mark;
3236 
3237 	err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3238 	if (err < 0)
3239 		return err;
3240 
3241 	rtm = nlmsg_data(nlh);
3242 	src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3243 	dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3244 	iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3245 	mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3246 	if (tb[RTA_UID])
3247 		uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3248 	else
3249 		uid = (iif ? INVALID_UID : current_uid());
3250 
3251 	if (tb[RTA_IP_PROTO]) {
3252 		err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3253 						  &ip_proto, AF_INET, extack);
3254 		if (err)
3255 			return err;
3256 	}
3257 
3258 	if (tb[RTA_SPORT])
3259 		sport = nla_get_be16(tb[RTA_SPORT]);
3260 
3261 	if (tb[RTA_DPORT])
3262 		dport = nla_get_be16(tb[RTA_DPORT]);
3263 
3264 	skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3265 	if (!skb)
3266 		return -ENOBUFS;
3267 
3268 	fl4.daddr = dst;
3269 	fl4.saddr = src;
3270 	fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3271 	fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3272 	fl4.flowi4_mark = mark;
3273 	fl4.flowi4_uid = uid;
3274 	if (sport)
3275 		fl4.fl4_sport = sport;
3276 	if (dport)
3277 		fl4.fl4_dport = dport;
3278 	fl4.flowi4_proto = ip_proto;
3279 
3280 	rcu_read_lock();
3281 
3282 	if (iif) {
3283 		struct net_device *dev;
3284 
3285 		dev = dev_get_by_index_rcu(net, iif);
3286 		if (!dev) {
3287 			err = -ENODEV;
3288 			goto errout_rcu;
3289 		}
3290 
3291 		fl4.flowi4_iif = iif; /* for rt_fill_info */
3292 		skb->dev	= dev;
3293 		skb->mark	= mark;
3294 		err = ip_route_input_rcu(skb, dst, src,
3295 					 rtm->rtm_tos & IPTOS_RT_MASK, dev,
3296 					 &res);
3297 
3298 		rt = skb_rtable(skb);
3299 		if (err == 0 && rt->dst.error)
3300 			err = -rt->dst.error;
3301 	} else {
3302 		fl4.flowi4_iif = LOOPBACK_IFINDEX;
3303 		skb->dev = net->loopback_dev;
3304 		rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3305 		err = 0;
3306 		if (IS_ERR(rt))
3307 			err = PTR_ERR(rt);
3308 		else
3309 			skb_dst_set(skb, &rt->dst);
3310 	}
3311 
3312 	if (err)
3313 		goto errout_rcu;
3314 
3315 	if (rtm->rtm_flags & RTM_F_NOTIFY)
3316 		rt->rt_flags |= RTCF_NOTIFY;
3317 
3318 	if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3319 		table_id = res.table ? res.table->tb_id : 0;
3320 
3321 	/* reset skb for netlink reply msg */
3322 	skb_trim(skb, 0);
3323 	skb_reset_network_header(skb);
3324 	skb_reset_transport_header(skb);
3325 	skb_reset_mac_header(skb);
3326 
3327 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3328 		struct fib_rt_info fri;
3329 
3330 		if (!res.fi) {
3331 			err = fib_props[res.type].error;
3332 			if (!err)
3333 				err = -EHOSTUNREACH;
3334 			goto errout_rcu;
3335 		}
3336 		fri.fi = res.fi;
3337 		fri.tb_id = table_id;
3338 		fri.dst = res.prefix;
3339 		fri.dst_len = res.prefixlen;
3340 		fri.dscp = inet_dsfield_to_dscp(fl4.flowi4_tos);
3341 		fri.type = rt->rt_type;
3342 		fri.offload = 0;
3343 		fri.trap = 0;
3344 		fri.offload_failed = 0;
3345 		if (res.fa_head) {
3346 			struct fib_alias *fa;
3347 
3348 			hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3349 				u8 slen = 32 - fri.dst_len;
3350 
3351 				if (fa->fa_slen == slen &&
3352 				    fa->tb_id == fri.tb_id &&
3353 				    fa->fa_dscp == fri.dscp &&
3354 				    fa->fa_info == res.fi &&
3355 				    fa->fa_type == fri.type) {
3356 					fri.offload = READ_ONCE(fa->offload);
3357 					fri.trap = READ_ONCE(fa->trap);
3358 					fri.offload_failed =
3359 						READ_ONCE(fa->offload_failed);
3360 					break;
3361 				}
3362 			}
3363 		}
3364 		err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3365 				    nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3366 	} else {
3367 		err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3368 				   NETLINK_CB(in_skb).portid,
3369 				   nlh->nlmsg_seq, 0);
3370 	}
3371 	if (err < 0)
3372 		goto errout_rcu;
3373 
3374 	rcu_read_unlock();
3375 
3376 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3377 
3378 errout_free:
3379 	return err;
3380 errout_rcu:
3381 	rcu_read_unlock();
3382 	kfree_skb(skb);
3383 	goto errout_free;
3384 }
3385 
3386 void ip_rt_multicast_event(struct in_device *in_dev)
3387 {
3388 	rt_cache_flush(dev_net(in_dev->dev));
3389 }
3390 
3391 #ifdef CONFIG_SYSCTL
3392 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3393 static int ip_rt_gc_min_interval __read_mostly	= HZ / 2;
3394 static int ip_rt_gc_elasticity __read_mostly	= 8;
3395 static int ip_min_valid_pmtu __read_mostly	= IPV4_MIN_MTU;
3396 
3397 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3398 		void *buffer, size_t *lenp, loff_t *ppos)
3399 {
3400 	struct net *net = (struct net *)__ctl->extra1;
3401 
3402 	if (write) {
3403 		rt_cache_flush(net);
3404 		fnhe_genid_bump(net);
3405 		return 0;
3406 	}
3407 
3408 	return -EINVAL;
3409 }
3410 
3411 static struct ctl_table ipv4_route_table[] = {
3412 	{
3413 		.procname	= "gc_thresh",
3414 		.data		= &ipv4_dst_ops.gc_thresh,
3415 		.maxlen		= sizeof(int),
3416 		.mode		= 0644,
3417 		.proc_handler	= proc_dointvec,
3418 	},
3419 	{
3420 		.procname	= "max_size",
3421 		.data		= &ip_rt_max_size,
3422 		.maxlen		= sizeof(int),
3423 		.mode		= 0644,
3424 		.proc_handler	= proc_dointvec,
3425 	},
3426 	{
3427 		/*  Deprecated. Use gc_min_interval_ms */
3428 
3429 		.procname	= "gc_min_interval",
3430 		.data		= &ip_rt_gc_min_interval,
3431 		.maxlen		= sizeof(int),
3432 		.mode		= 0644,
3433 		.proc_handler	= proc_dointvec_jiffies,
3434 	},
3435 	{
3436 		.procname	= "gc_min_interval_ms",
3437 		.data		= &ip_rt_gc_min_interval,
3438 		.maxlen		= sizeof(int),
3439 		.mode		= 0644,
3440 		.proc_handler	= proc_dointvec_ms_jiffies,
3441 	},
3442 	{
3443 		.procname	= "gc_timeout",
3444 		.data		= &ip_rt_gc_timeout,
3445 		.maxlen		= sizeof(int),
3446 		.mode		= 0644,
3447 		.proc_handler	= proc_dointvec_jiffies,
3448 	},
3449 	{
3450 		.procname	= "gc_interval",
3451 		.data		= &ip_rt_gc_interval,
3452 		.maxlen		= sizeof(int),
3453 		.mode		= 0644,
3454 		.proc_handler	= proc_dointvec_jiffies,
3455 	},
3456 	{
3457 		.procname	= "redirect_load",
3458 		.data		= &ip_rt_redirect_load,
3459 		.maxlen		= sizeof(int),
3460 		.mode		= 0644,
3461 		.proc_handler	= proc_dointvec,
3462 	},
3463 	{
3464 		.procname	= "redirect_number",
3465 		.data		= &ip_rt_redirect_number,
3466 		.maxlen		= sizeof(int),
3467 		.mode		= 0644,
3468 		.proc_handler	= proc_dointvec,
3469 	},
3470 	{
3471 		.procname	= "redirect_silence",
3472 		.data		= &ip_rt_redirect_silence,
3473 		.maxlen		= sizeof(int),
3474 		.mode		= 0644,
3475 		.proc_handler	= proc_dointvec,
3476 	},
3477 	{
3478 		.procname	= "error_cost",
3479 		.data		= &ip_rt_error_cost,
3480 		.maxlen		= sizeof(int),
3481 		.mode		= 0644,
3482 		.proc_handler	= proc_dointvec,
3483 	},
3484 	{
3485 		.procname	= "error_burst",
3486 		.data		= &ip_rt_error_burst,
3487 		.maxlen		= sizeof(int),
3488 		.mode		= 0644,
3489 		.proc_handler	= proc_dointvec,
3490 	},
3491 	{
3492 		.procname	= "gc_elasticity",
3493 		.data		= &ip_rt_gc_elasticity,
3494 		.maxlen		= sizeof(int),
3495 		.mode		= 0644,
3496 		.proc_handler	= proc_dointvec,
3497 	},
3498 	{ }
3499 };
3500 
3501 static const char ipv4_route_flush_procname[] = "flush";
3502 
3503 static struct ctl_table ipv4_route_netns_table[] = {
3504 	{
3505 		.procname	= ipv4_route_flush_procname,
3506 		.maxlen		= sizeof(int),
3507 		.mode		= 0200,
3508 		.proc_handler	= ipv4_sysctl_rtcache_flush,
3509 	},
3510 	{
3511 		.procname       = "min_pmtu",
3512 		.data           = &init_net.ipv4.ip_rt_min_pmtu,
3513 		.maxlen         = sizeof(int),
3514 		.mode           = 0644,
3515 		.proc_handler   = proc_dointvec_minmax,
3516 		.extra1         = &ip_min_valid_pmtu,
3517 	},
3518 	{
3519 		.procname       = "mtu_expires",
3520 		.data           = &init_net.ipv4.ip_rt_mtu_expires,
3521 		.maxlen         = sizeof(int),
3522 		.mode           = 0644,
3523 		.proc_handler   = proc_dointvec_jiffies,
3524 	},
3525 	{
3526 		.procname   = "min_adv_mss",
3527 		.data       = &init_net.ipv4.ip_rt_min_advmss,
3528 		.maxlen     = sizeof(int),
3529 		.mode       = 0644,
3530 		.proc_handler   = proc_dointvec,
3531 	},
3532 	{ },
3533 };
3534 
3535 static __net_init int sysctl_route_net_init(struct net *net)
3536 {
3537 	struct ctl_table *tbl;
3538 	size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3539 
3540 	tbl = ipv4_route_netns_table;
3541 	if (!net_eq(net, &init_net)) {
3542 		int i;
3543 
3544 		tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3545 		if (!tbl)
3546 			goto err_dup;
3547 
3548 		/* Don't export non-whitelisted sysctls to unprivileged users */
3549 		if (net->user_ns != &init_user_ns) {
3550 			if (tbl[0].procname != ipv4_route_flush_procname) {
3551 				tbl[0].procname = NULL;
3552 				table_size = 0;
3553 			}
3554 		}
3555 
3556 		/* Update the variables to point into the current struct net
3557 		 * except for the first element flush
3558 		 */
3559 		for (i = 1; i < ARRAY_SIZE(ipv4_route_netns_table) - 1; i++)
3560 			tbl[i].data += (void *)net - (void *)&init_net;
3561 	}
3562 	tbl[0].extra1 = net;
3563 
3564 	net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3565 						     tbl, table_size);
3566 	if (!net->ipv4.route_hdr)
3567 		goto err_reg;
3568 	return 0;
3569 
3570 err_reg:
3571 	if (tbl != ipv4_route_netns_table)
3572 		kfree(tbl);
3573 err_dup:
3574 	return -ENOMEM;
3575 }
3576 
3577 static __net_exit void sysctl_route_net_exit(struct net *net)
3578 {
3579 	struct ctl_table *tbl;
3580 
3581 	tbl = net->ipv4.route_hdr->ctl_table_arg;
3582 	unregister_net_sysctl_table(net->ipv4.route_hdr);
3583 	BUG_ON(tbl == ipv4_route_netns_table);
3584 	kfree(tbl);
3585 }
3586 
3587 static __net_initdata struct pernet_operations sysctl_route_ops = {
3588 	.init = sysctl_route_net_init,
3589 	.exit = sysctl_route_net_exit,
3590 };
3591 #endif
3592 
3593 static __net_init int netns_ip_rt_init(struct net *net)
3594 {
3595 	/* Set default value for namespaceified sysctls */
3596 	net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3597 	net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3598 	net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3599 	return 0;
3600 }
3601 
3602 static struct pernet_operations __net_initdata ip_rt_ops = {
3603 	.init = netns_ip_rt_init,
3604 };
3605 
3606 static __net_init int rt_genid_init(struct net *net)
3607 {
3608 	atomic_set(&net->ipv4.rt_genid, 0);
3609 	atomic_set(&net->fnhe_genid, 0);
3610 	atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3611 	return 0;
3612 }
3613 
3614 static __net_initdata struct pernet_operations rt_genid_ops = {
3615 	.init = rt_genid_init,
3616 };
3617 
3618 static int __net_init ipv4_inetpeer_init(struct net *net)
3619 {
3620 	struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3621 
3622 	if (!bp)
3623 		return -ENOMEM;
3624 	inet_peer_base_init(bp);
3625 	net->ipv4.peers = bp;
3626 	return 0;
3627 }
3628 
3629 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3630 {
3631 	struct inet_peer_base *bp = net->ipv4.peers;
3632 
3633 	net->ipv4.peers = NULL;
3634 	inetpeer_invalidate_tree(bp);
3635 	kfree(bp);
3636 }
3637 
3638 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3639 	.init	=	ipv4_inetpeer_init,
3640 	.exit	=	ipv4_inetpeer_exit,
3641 };
3642 
3643 #ifdef CONFIG_IP_ROUTE_CLASSID
3644 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3645 #endif /* CONFIG_IP_ROUTE_CLASSID */
3646 
3647 int __init ip_rt_init(void)
3648 {
3649 	void *idents_hash;
3650 	int cpu;
3651 
3652 	/* For modern hosts, this will use 2 MB of memory */
3653 	idents_hash = alloc_large_system_hash("IP idents",
3654 					      sizeof(*ip_idents) + sizeof(*ip_tstamps),
3655 					      0,
3656 					      16, /* one bucket per 64 KB */
3657 					      HASH_ZERO,
3658 					      NULL,
3659 					      &ip_idents_mask,
3660 					      2048,
3661 					      256*1024);
3662 
3663 	ip_idents = idents_hash;
3664 
3665 	get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3666 
3667 	ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3668 
3669 	for_each_possible_cpu(cpu) {
3670 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3671 
3672 		INIT_LIST_HEAD(&ul->head);
3673 		INIT_LIST_HEAD(&ul->quarantine);
3674 		spin_lock_init(&ul->lock);
3675 	}
3676 #ifdef CONFIG_IP_ROUTE_CLASSID
3677 	ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3678 	if (!ip_rt_acct)
3679 		panic("IP: failed to allocate ip_rt_acct\n");
3680 #endif
3681 
3682 	ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable,
3683 					      SLAB_HWCACHE_ALIGN | SLAB_PANIC);
3684 
3685 	ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3686 
3687 	if (dst_entries_init(&ipv4_dst_ops) < 0)
3688 		panic("IP: failed to allocate ipv4_dst_ops counter\n");
3689 
3690 	if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3691 		panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3692 
3693 	ipv4_dst_ops.gc_thresh = ~0;
3694 	ip_rt_max_size = INT_MAX;
3695 
3696 	devinet_init();
3697 	ip_fib_init();
3698 
3699 	if (ip_rt_proc_init())
3700 		pr_err("Unable to create route proc files\n");
3701 #ifdef CONFIG_XFRM
3702 	xfrm_init();
3703 	xfrm4_init();
3704 #endif
3705 	rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3706 		      RTNL_FLAG_DOIT_UNLOCKED);
3707 
3708 #ifdef CONFIG_SYSCTL
3709 	register_pernet_subsys(&sysctl_route_ops);
3710 #endif
3711 	register_pernet_subsys(&ip_rt_ops);
3712 	register_pernet_subsys(&rt_genid_ops);
3713 	register_pernet_subsys(&ipv4_inetpeer_ops);
3714 	return 0;
3715 }
3716 
3717 #ifdef CONFIG_SYSCTL
3718 /*
3719  * We really need to sanitize the damn ipv4 init order, then all
3720  * this nonsense will go away.
3721  */
3722 void __init ip_static_sysctl_init(void)
3723 {
3724 	register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3725 }
3726 #endif
3727