xref: /linux/net/ipv4/route.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
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 void		ipv4_negative_advice(struct sock *sk,
133 					     struct dst_entry *dst);
134 static void		 ipv4_link_failure(struct sk_buff *skb);
135 static void		 ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
136 					   struct sk_buff *skb, u32 mtu,
137 					   bool confirm_neigh);
138 static void		 ip_do_redirect(struct dst_entry *dst, struct sock *sk,
139 					struct sk_buff *skb);
140 static void		ipv4_dst_destroy(struct dst_entry *dst);
141 
142 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
143 {
144 	WARN_ON(1);
145 	return NULL;
146 }
147 
148 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
149 					   struct sk_buff *skb,
150 					   const void *daddr);
151 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
152 
153 static struct dst_ops ipv4_dst_ops = {
154 	.family =		AF_INET,
155 	.check =		ipv4_dst_check,
156 	.default_advmss =	ipv4_default_advmss,
157 	.mtu =			ipv4_mtu,
158 	.cow_metrics =		ipv4_cow_metrics,
159 	.destroy =		ipv4_dst_destroy,
160 	.negative_advice =	ipv4_negative_advice,
161 	.link_failure =		ipv4_link_failure,
162 	.update_pmtu =		ip_rt_update_pmtu,
163 	.redirect =		ip_do_redirect,
164 	.local_out =		__ip_local_out,
165 	.neigh_lookup =		ipv4_neigh_lookup,
166 	.confirm_neigh =	ipv4_confirm_neigh,
167 };
168 
169 #define ECN_OR_COST(class)	TC_PRIO_##class
170 
171 const __u8 ip_tos2prio[16] = {
172 	TC_PRIO_BESTEFFORT,
173 	ECN_OR_COST(BESTEFFORT),
174 	TC_PRIO_BESTEFFORT,
175 	ECN_OR_COST(BESTEFFORT),
176 	TC_PRIO_BULK,
177 	ECN_OR_COST(BULK),
178 	TC_PRIO_BULK,
179 	ECN_OR_COST(BULK),
180 	TC_PRIO_INTERACTIVE,
181 	ECN_OR_COST(INTERACTIVE),
182 	TC_PRIO_INTERACTIVE,
183 	ECN_OR_COST(INTERACTIVE),
184 	TC_PRIO_INTERACTIVE_BULK,
185 	ECN_OR_COST(INTERACTIVE_BULK),
186 	TC_PRIO_INTERACTIVE_BULK,
187 	ECN_OR_COST(INTERACTIVE_BULK)
188 };
189 EXPORT_SYMBOL(ip_tos2prio);
190 
191 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
192 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
193 
194 #ifdef CONFIG_PROC_FS
195 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
196 {
197 	if (*pos)
198 		return NULL;
199 	return SEQ_START_TOKEN;
200 }
201 
202 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
203 {
204 	++*pos;
205 	return NULL;
206 }
207 
208 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
209 {
210 }
211 
212 static int rt_cache_seq_show(struct seq_file *seq, void *v)
213 {
214 	if (v == SEQ_START_TOKEN)
215 		seq_printf(seq, "%-127s\n",
216 			   "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
217 			   "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
218 			   "HHUptod\tSpecDst");
219 	return 0;
220 }
221 
222 static const struct seq_operations rt_cache_seq_ops = {
223 	.start  = rt_cache_seq_start,
224 	.next   = rt_cache_seq_next,
225 	.stop   = rt_cache_seq_stop,
226 	.show   = rt_cache_seq_show,
227 };
228 
229 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
230 {
231 	int cpu;
232 
233 	if (*pos == 0)
234 		return SEQ_START_TOKEN;
235 
236 	for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
237 		if (!cpu_possible(cpu))
238 			continue;
239 		*pos = cpu+1;
240 		return &per_cpu(rt_cache_stat, cpu);
241 	}
242 	return NULL;
243 }
244 
245 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
246 {
247 	int cpu;
248 
249 	for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
250 		if (!cpu_possible(cpu))
251 			continue;
252 		*pos = cpu+1;
253 		return &per_cpu(rt_cache_stat, cpu);
254 	}
255 	(*pos)++;
256 	return NULL;
257 
258 }
259 
260 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
261 {
262 
263 }
264 
265 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
266 {
267 	struct rt_cache_stat *st = v;
268 
269 	if (v == SEQ_START_TOKEN) {
270 		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");
271 		return 0;
272 	}
273 
274 	seq_printf(seq, "%08x %08x %08x    %08x   %08x    %08x %08x       "
275 			"%08x       %08x %08x     %08x    %08x %08x   "
276 			"%08x     %08x        %08x        %08x\n",
277 		   dst_entries_get_slow(&ipv4_dst_ops),
278 		   0, /* st->in_hit */
279 		   st->in_slow_tot,
280 		   st->in_slow_mc,
281 		   st->in_no_route,
282 		   st->in_brd,
283 		   st->in_martian_dst,
284 		   st->in_martian_src,
285 
286 		   0, /* st->out_hit */
287 		   st->out_slow_tot,
288 		   st->out_slow_mc,
289 
290 		   0, /* st->gc_total */
291 		   0, /* st->gc_ignored */
292 		   0, /* st->gc_goal_miss */
293 		   0, /* st->gc_dst_overflow */
294 		   0, /* st->in_hlist_search */
295 		   0  /* st->out_hlist_search */
296 		);
297 	return 0;
298 }
299 
300 static const struct seq_operations rt_cpu_seq_ops = {
301 	.start  = rt_cpu_seq_start,
302 	.next   = rt_cpu_seq_next,
303 	.stop   = rt_cpu_seq_stop,
304 	.show   = rt_cpu_seq_show,
305 };
306 
307 #ifdef CONFIG_IP_ROUTE_CLASSID
308 static int rt_acct_proc_show(struct seq_file *m, void *v)
309 {
310 	struct ip_rt_acct *dst, *src;
311 	unsigned int i, j;
312 
313 	dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
314 	if (!dst)
315 		return -ENOMEM;
316 
317 	for_each_possible_cpu(i) {
318 		src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
319 		for (j = 0; j < 256; j++) {
320 			dst[j].o_bytes   += src[j].o_bytes;
321 			dst[j].o_packets += src[j].o_packets;
322 			dst[j].i_bytes   += src[j].i_bytes;
323 			dst[j].i_packets += src[j].i_packets;
324 		}
325 	}
326 
327 	seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
328 	kfree(dst);
329 	return 0;
330 }
331 #endif
332 
333 static int __net_init ip_rt_do_proc_init(struct net *net)
334 {
335 	struct proc_dir_entry *pde;
336 
337 	pde = proc_create_seq("rt_cache", 0444, net->proc_net,
338 			      &rt_cache_seq_ops);
339 	if (!pde)
340 		goto err1;
341 
342 	pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
343 			      &rt_cpu_seq_ops);
344 	if (!pde)
345 		goto err2;
346 
347 #ifdef CONFIG_IP_ROUTE_CLASSID
348 	pde = proc_create_single("rt_acct", 0, net->proc_net,
349 			rt_acct_proc_show);
350 	if (!pde)
351 		goto err3;
352 #endif
353 	return 0;
354 
355 #ifdef CONFIG_IP_ROUTE_CLASSID
356 err3:
357 	remove_proc_entry("rt_cache", net->proc_net_stat);
358 #endif
359 err2:
360 	remove_proc_entry("rt_cache", net->proc_net);
361 err1:
362 	return -ENOMEM;
363 }
364 
365 static void __net_exit ip_rt_do_proc_exit(struct net *net)
366 {
367 	remove_proc_entry("rt_cache", net->proc_net_stat);
368 	remove_proc_entry("rt_cache", net->proc_net);
369 #ifdef CONFIG_IP_ROUTE_CLASSID
370 	remove_proc_entry("rt_acct", net->proc_net);
371 #endif
372 }
373 
374 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
375 	.init = ip_rt_do_proc_init,
376 	.exit = ip_rt_do_proc_exit,
377 };
378 
379 static int __init ip_rt_proc_init(void)
380 {
381 	return register_pernet_subsys(&ip_rt_proc_ops);
382 }
383 
384 #else
385 static inline int ip_rt_proc_init(void)
386 {
387 	return 0;
388 }
389 #endif /* CONFIG_PROC_FS */
390 
391 static inline bool rt_is_expired(const struct rtable *rth)
392 {
393 	return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
394 }
395 
396 void rt_cache_flush(struct net *net)
397 {
398 	rt_genid_bump_ipv4(net);
399 }
400 
401 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
402 					   struct sk_buff *skb,
403 					   const void *daddr)
404 {
405 	const struct rtable *rt = container_of(dst, struct rtable, dst);
406 	struct net_device *dev = dst->dev;
407 	struct neighbour *n;
408 
409 	rcu_read_lock();
410 
411 	if (likely(rt->rt_gw_family == AF_INET)) {
412 		n = ip_neigh_gw4(dev, rt->rt_gw4);
413 	} else if (rt->rt_gw_family == AF_INET6) {
414 		n = ip_neigh_gw6(dev, &rt->rt_gw6);
415         } else {
416 		__be32 pkey;
417 
418 		pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
419 		n = ip_neigh_gw4(dev, pkey);
420 	}
421 
422 	if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
423 		n = NULL;
424 
425 	rcu_read_unlock();
426 
427 	return n;
428 }
429 
430 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
431 {
432 	const struct rtable *rt = container_of(dst, struct rtable, dst);
433 	struct net_device *dev = dst->dev;
434 	const __be32 *pkey = daddr;
435 
436 	if (rt->rt_gw_family == AF_INET) {
437 		pkey = (const __be32 *)&rt->rt_gw4;
438 	} else if (rt->rt_gw_family == AF_INET6) {
439 		return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
440 	} else if (!daddr ||
441 		 (rt->rt_flags &
442 		  (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
443 		return;
444 	}
445 	__ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
446 }
447 
448 /* Hash tables of size 2048..262144 depending on RAM size.
449  * Each bucket uses 8 bytes.
450  */
451 static u32 ip_idents_mask __read_mostly;
452 static atomic_t *ip_idents __read_mostly;
453 static u32 *ip_tstamps __read_mostly;
454 
455 /* In order to protect privacy, we add a perturbation to identifiers
456  * if one generator is seldom used. This makes hard for an attacker
457  * to infer how many packets were sent between two points in time.
458  */
459 static u32 ip_idents_reserve(u32 hash, int segs)
460 {
461 	u32 bucket, old, now = (u32)jiffies;
462 	atomic_t *p_id;
463 	u32 *p_tstamp;
464 	u32 delta = 0;
465 
466 	bucket = hash & ip_idents_mask;
467 	p_tstamp = ip_tstamps + bucket;
468 	p_id = ip_idents + bucket;
469 	old = READ_ONCE(*p_tstamp);
470 
471 	if (old != now && cmpxchg(p_tstamp, old, now) == old)
472 		delta = get_random_u32_below(now - old);
473 
474 	/* If UBSAN reports an error there, please make sure your compiler
475 	 * supports -fno-strict-overflow before reporting it that was a bug
476 	 * in UBSAN, and it has been fixed in GCC-8.
477 	 */
478 	return atomic_add_return(segs + delta, p_id) - segs;
479 }
480 
481 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
482 {
483 	u32 hash, id;
484 
485 	/* Note the following code is not safe, but this is okay. */
486 	if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
487 		get_random_bytes(&net->ipv4.ip_id_key,
488 				 sizeof(net->ipv4.ip_id_key));
489 
490 	hash = siphash_3u32((__force u32)iph->daddr,
491 			    (__force u32)iph->saddr,
492 			    iph->protocol,
493 			    &net->ipv4.ip_id_key);
494 	id = ip_idents_reserve(hash, segs);
495 	iph->id = htons(id);
496 }
497 EXPORT_SYMBOL(__ip_select_ident);
498 
499 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
500 			     const struct sock *sk, const struct iphdr *iph,
501 			     int oif, __u8 tos, u8 prot, u32 mark,
502 			     int flow_flags)
503 {
504 	__u8 scope = RT_SCOPE_UNIVERSE;
505 
506 	if (sk) {
507 		oif = sk->sk_bound_dev_if;
508 		mark = READ_ONCE(sk->sk_mark);
509 		tos = ip_sock_rt_tos(sk);
510 		scope = ip_sock_rt_scope(sk);
511 		prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW :
512 						    sk->sk_protocol;
513 	}
514 
515 	flowi4_init_output(fl4, oif, mark, tos & INET_DSCP_MASK, scope,
516 			   prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
517 			   sock_net_uid(net, sk));
518 }
519 
520 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
521 			       const struct sock *sk)
522 {
523 	const struct net *net = dev_net(skb->dev);
524 	const struct iphdr *iph = ip_hdr(skb);
525 	int oif = skb->dev->ifindex;
526 	u8 prot = iph->protocol;
527 	u32 mark = skb->mark;
528 	__u8 tos = iph->tos;
529 
530 	__build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
531 }
532 
533 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
534 {
535 	const struct inet_sock *inet = inet_sk(sk);
536 	const struct ip_options_rcu *inet_opt;
537 	__be32 daddr = inet->inet_daddr;
538 
539 	rcu_read_lock();
540 	inet_opt = rcu_dereference(inet->inet_opt);
541 	if (inet_opt && inet_opt->opt.srr)
542 		daddr = inet_opt->opt.faddr;
543 	flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark),
544 			   ip_sock_rt_tos(sk),
545 			   ip_sock_rt_scope(sk),
546 			   inet_test_bit(HDRINCL, sk) ?
547 				IPPROTO_RAW : sk->sk_protocol,
548 			   inet_sk_flowi_flags(sk),
549 			   daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
550 	rcu_read_unlock();
551 }
552 
553 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
554 				 const struct sk_buff *skb)
555 {
556 	if (skb)
557 		build_skb_flow_key(fl4, skb, sk);
558 	else
559 		build_sk_flow_key(fl4, sk);
560 }
561 
562 static DEFINE_SPINLOCK(fnhe_lock);
563 
564 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
565 {
566 	struct rtable *rt;
567 
568 	rt = rcu_dereference(fnhe->fnhe_rth_input);
569 	if (rt) {
570 		RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
571 		dst_dev_put(&rt->dst);
572 		dst_release(&rt->dst);
573 	}
574 	rt = rcu_dereference(fnhe->fnhe_rth_output);
575 	if (rt) {
576 		RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
577 		dst_dev_put(&rt->dst);
578 		dst_release(&rt->dst);
579 	}
580 }
581 
582 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
583 {
584 	struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
585 	struct fib_nh_exception *fnhe, *oldest = NULL;
586 
587 	for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
588 		fnhe = rcu_dereference_protected(*fnhe_p,
589 						 lockdep_is_held(&fnhe_lock));
590 		if (!fnhe)
591 			break;
592 		if (!oldest ||
593 		    time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
594 			oldest = fnhe;
595 			oldest_p = fnhe_p;
596 		}
597 	}
598 	fnhe_flush_routes(oldest);
599 	*oldest_p = oldest->fnhe_next;
600 	kfree_rcu(oldest, rcu);
601 }
602 
603 static u32 fnhe_hashfun(__be32 daddr)
604 {
605 	static siphash_aligned_key_t fnhe_hash_key;
606 	u64 hval;
607 
608 	net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
609 	hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
610 	return hash_64(hval, FNHE_HASH_SHIFT);
611 }
612 
613 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
614 {
615 	rt->rt_pmtu = fnhe->fnhe_pmtu;
616 	rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
617 	rt->dst.expires = fnhe->fnhe_expires;
618 
619 	if (fnhe->fnhe_gw) {
620 		rt->rt_flags |= RTCF_REDIRECTED;
621 		rt->rt_uses_gateway = 1;
622 		rt->rt_gw_family = AF_INET;
623 		rt->rt_gw4 = fnhe->fnhe_gw;
624 	}
625 }
626 
627 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
628 				  __be32 gw, u32 pmtu, bool lock,
629 				  unsigned long expires)
630 {
631 	struct fnhe_hash_bucket *hash;
632 	struct fib_nh_exception *fnhe;
633 	struct rtable *rt;
634 	u32 genid, hval;
635 	unsigned int i;
636 	int depth;
637 
638 	genid = fnhe_genid(dev_net(nhc->nhc_dev));
639 	hval = fnhe_hashfun(daddr);
640 
641 	spin_lock_bh(&fnhe_lock);
642 
643 	hash = rcu_dereference(nhc->nhc_exceptions);
644 	if (!hash) {
645 		hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
646 		if (!hash)
647 			goto out_unlock;
648 		rcu_assign_pointer(nhc->nhc_exceptions, hash);
649 	}
650 
651 	hash += hval;
652 
653 	depth = 0;
654 	for (fnhe = rcu_dereference(hash->chain); fnhe;
655 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
656 		if (fnhe->fnhe_daddr == daddr)
657 			break;
658 		depth++;
659 	}
660 
661 	if (fnhe) {
662 		if (fnhe->fnhe_genid != genid)
663 			fnhe->fnhe_genid = genid;
664 		if (gw)
665 			fnhe->fnhe_gw = gw;
666 		if (pmtu) {
667 			fnhe->fnhe_pmtu = pmtu;
668 			fnhe->fnhe_mtu_locked = lock;
669 		}
670 		fnhe->fnhe_expires = max(1UL, expires);
671 		/* Update all cached dsts too */
672 		rt = rcu_dereference(fnhe->fnhe_rth_input);
673 		if (rt)
674 			fill_route_from_fnhe(rt, fnhe);
675 		rt = rcu_dereference(fnhe->fnhe_rth_output);
676 		if (rt)
677 			fill_route_from_fnhe(rt, fnhe);
678 	} else {
679 		/* Randomize max depth to avoid some side channels attacks. */
680 		int max_depth = FNHE_RECLAIM_DEPTH +
681 				get_random_u32_below(FNHE_RECLAIM_DEPTH);
682 
683 		while (depth > max_depth) {
684 			fnhe_remove_oldest(hash);
685 			depth--;
686 		}
687 
688 		fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
689 		if (!fnhe)
690 			goto out_unlock;
691 
692 		fnhe->fnhe_next = hash->chain;
693 
694 		fnhe->fnhe_genid = genid;
695 		fnhe->fnhe_daddr = daddr;
696 		fnhe->fnhe_gw = gw;
697 		fnhe->fnhe_pmtu = pmtu;
698 		fnhe->fnhe_mtu_locked = lock;
699 		fnhe->fnhe_expires = max(1UL, expires);
700 
701 		rcu_assign_pointer(hash->chain, fnhe);
702 
703 		/* Exception created; mark the cached routes for the nexthop
704 		 * stale, so anyone caching it rechecks if this exception
705 		 * applies to them.
706 		 */
707 		rt = rcu_dereference(nhc->nhc_rth_input);
708 		if (rt)
709 			rt->dst.obsolete = DST_OBSOLETE_KILL;
710 
711 		for_each_possible_cpu(i) {
712 			struct rtable __rcu **prt;
713 
714 			prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
715 			rt = rcu_dereference(*prt);
716 			if (rt)
717 				rt->dst.obsolete = DST_OBSOLETE_KILL;
718 		}
719 	}
720 
721 	fnhe->fnhe_stamp = jiffies;
722 
723 out_unlock:
724 	spin_unlock_bh(&fnhe_lock);
725 }
726 
727 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
728 			     bool kill_route)
729 {
730 	__be32 new_gw = icmp_hdr(skb)->un.gateway;
731 	__be32 old_gw = ip_hdr(skb)->saddr;
732 	struct net_device *dev = skb->dev;
733 	struct in_device *in_dev;
734 	struct fib_result res;
735 	struct neighbour *n;
736 	struct net *net;
737 
738 	switch (icmp_hdr(skb)->code & 7) {
739 	case ICMP_REDIR_NET:
740 	case ICMP_REDIR_NETTOS:
741 	case ICMP_REDIR_HOST:
742 	case ICMP_REDIR_HOSTTOS:
743 		break;
744 
745 	default:
746 		return;
747 	}
748 
749 	if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
750 		return;
751 
752 	in_dev = __in_dev_get_rcu(dev);
753 	if (!in_dev)
754 		return;
755 
756 	net = dev_net(dev);
757 	if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
758 	    ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
759 	    ipv4_is_zeronet(new_gw))
760 		goto reject_redirect;
761 
762 	if (!IN_DEV_SHARED_MEDIA(in_dev)) {
763 		if (!inet_addr_onlink(in_dev, new_gw, old_gw))
764 			goto reject_redirect;
765 		if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
766 			goto reject_redirect;
767 	} else {
768 		if (inet_addr_type(net, new_gw) != RTN_UNICAST)
769 			goto reject_redirect;
770 	}
771 
772 	n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw);
773 	if (!n)
774 		n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
775 	if (!IS_ERR(n)) {
776 		if (!(READ_ONCE(n->nud_state) & NUD_VALID)) {
777 			neigh_event_send(n, NULL);
778 		} else {
779 			if (fib_lookup(net, fl4, &res, 0) == 0) {
780 				struct fib_nh_common *nhc;
781 
782 				fib_select_path(net, &res, fl4, skb);
783 				nhc = FIB_RES_NHC(res);
784 				update_or_create_fnhe(nhc, fl4->daddr, new_gw,
785 						0, false,
786 						jiffies + ip_rt_gc_timeout);
787 			}
788 			if (kill_route)
789 				rt->dst.obsolete = DST_OBSOLETE_KILL;
790 			call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
791 		}
792 		neigh_release(n);
793 	}
794 	return;
795 
796 reject_redirect:
797 #ifdef CONFIG_IP_ROUTE_VERBOSE
798 	if (IN_DEV_LOG_MARTIANS(in_dev)) {
799 		const struct iphdr *iph = (const struct iphdr *) skb->data;
800 		__be32 daddr = iph->daddr;
801 		__be32 saddr = iph->saddr;
802 
803 		net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
804 				     "  Advised path = %pI4 -> %pI4\n",
805 				     &old_gw, dev->name, &new_gw,
806 				     &saddr, &daddr);
807 	}
808 #endif
809 	;
810 }
811 
812 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
813 {
814 	struct rtable *rt;
815 	struct flowi4 fl4;
816 	const struct iphdr *iph = (const struct iphdr *) skb->data;
817 	struct net *net = dev_net(skb->dev);
818 	int oif = skb->dev->ifindex;
819 	u8 prot = iph->protocol;
820 	u32 mark = skb->mark;
821 	__u8 tos = iph->tos;
822 
823 	rt = dst_rtable(dst);
824 
825 	__build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
826 	__ip_do_redirect(rt, skb, &fl4, true);
827 }
828 
829 static void ipv4_negative_advice(struct sock *sk,
830 				 struct dst_entry *dst)
831 {
832 	struct rtable *rt = dst_rtable(dst);
833 
834 	if ((dst->obsolete > 0) ||
835 	    (rt->rt_flags & RTCF_REDIRECTED) ||
836 	    rt->dst.expires)
837 		sk_dst_reset(sk);
838 }
839 
840 /*
841  * Algorithm:
842  *	1. The first ip_rt_redirect_number redirects are sent
843  *	   with exponential backoff, then we stop sending them at all,
844  *	   assuming that the host ignores our redirects.
845  *	2. If we did not see packets requiring redirects
846  *	   during ip_rt_redirect_silence, we assume that the host
847  *	   forgot redirected route and start to send redirects again.
848  *
849  * This algorithm is much cheaper and more intelligent than dumb load limiting
850  * in icmp.c.
851  *
852  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
853  * and "frag. need" (breaks PMTU discovery) in icmp.c.
854  */
855 
856 void ip_rt_send_redirect(struct sk_buff *skb)
857 {
858 	struct rtable *rt = skb_rtable(skb);
859 	struct in_device *in_dev;
860 	struct inet_peer *peer;
861 	struct net *net;
862 	int log_martians;
863 	int vif;
864 
865 	rcu_read_lock();
866 	in_dev = __in_dev_get_rcu(rt->dst.dev);
867 	if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
868 		rcu_read_unlock();
869 		return;
870 	}
871 	log_martians = IN_DEV_LOG_MARTIANS(in_dev);
872 	vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
873 	rcu_read_unlock();
874 
875 	net = dev_net(rt->dst.dev);
876 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
877 	if (!peer) {
878 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
879 			  rt_nexthop(rt, ip_hdr(skb)->daddr));
880 		return;
881 	}
882 
883 	/* No redirected packets during ip_rt_redirect_silence;
884 	 * reset the algorithm.
885 	 */
886 	if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
887 		peer->rate_tokens = 0;
888 		peer->n_redirects = 0;
889 	}
890 
891 	/* Too many ignored redirects; do not send anything
892 	 * set dst.rate_last to the last seen redirected packet.
893 	 */
894 	if (peer->n_redirects >= ip_rt_redirect_number) {
895 		peer->rate_last = jiffies;
896 		goto out_put_peer;
897 	}
898 
899 	/* Check for load limit; set rate_last to the latest sent
900 	 * redirect.
901 	 */
902 	if (peer->n_redirects == 0 ||
903 	    time_after(jiffies,
904 		       (peer->rate_last +
905 			(ip_rt_redirect_load << peer->n_redirects)))) {
906 		__be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
907 
908 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
909 		peer->rate_last = jiffies;
910 		++peer->n_redirects;
911 		if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians &&
912 		    peer->n_redirects == ip_rt_redirect_number)
913 			net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
914 					     &ip_hdr(skb)->saddr, inet_iif(skb),
915 					     &ip_hdr(skb)->daddr, &gw);
916 	}
917 out_put_peer:
918 	inet_putpeer(peer);
919 }
920 
921 static int ip_error(struct sk_buff *skb)
922 {
923 	struct rtable *rt = skb_rtable(skb);
924 	struct net_device *dev = skb->dev;
925 	struct in_device *in_dev;
926 	struct inet_peer *peer;
927 	unsigned long now;
928 	struct net *net;
929 	SKB_DR(reason);
930 	bool send;
931 	int code;
932 
933 	if (netif_is_l3_master(skb->dev)) {
934 		dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
935 		if (!dev)
936 			goto out;
937 	}
938 
939 	in_dev = __in_dev_get_rcu(dev);
940 
941 	/* IP on this device is disabled. */
942 	if (!in_dev)
943 		goto out;
944 
945 	net = dev_net(rt->dst.dev);
946 	if (!IN_DEV_FORWARD(in_dev)) {
947 		switch (rt->dst.error) {
948 		case EHOSTUNREACH:
949 			SKB_DR_SET(reason, IP_INADDRERRORS);
950 			__IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
951 			break;
952 
953 		case ENETUNREACH:
954 			SKB_DR_SET(reason, IP_INNOROUTES);
955 			__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
956 			break;
957 		}
958 		goto out;
959 	}
960 
961 	switch (rt->dst.error) {
962 	case EINVAL:
963 	default:
964 		goto out;
965 	case EHOSTUNREACH:
966 		code = ICMP_HOST_UNREACH;
967 		break;
968 	case ENETUNREACH:
969 		code = ICMP_NET_UNREACH;
970 		SKB_DR_SET(reason, IP_INNOROUTES);
971 		__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
972 		break;
973 	case EACCES:
974 		code = ICMP_PKT_FILTERED;
975 		break;
976 	}
977 
978 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
979 			       l3mdev_master_ifindex(skb->dev), 1);
980 
981 	send = true;
982 	if (peer) {
983 		now = jiffies;
984 		peer->rate_tokens += now - peer->rate_last;
985 		if (peer->rate_tokens > ip_rt_error_burst)
986 			peer->rate_tokens = ip_rt_error_burst;
987 		peer->rate_last = now;
988 		if (peer->rate_tokens >= ip_rt_error_cost)
989 			peer->rate_tokens -= ip_rt_error_cost;
990 		else
991 			send = false;
992 		inet_putpeer(peer);
993 	}
994 	if (send)
995 		icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
996 
997 out:	kfree_skb_reason(skb, reason);
998 	return 0;
999 }
1000 
1001 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1002 {
1003 	struct dst_entry *dst = &rt->dst;
1004 	struct net *net = dev_net(dst->dev);
1005 	struct fib_result res;
1006 	bool lock = false;
1007 	u32 old_mtu;
1008 
1009 	if (ip_mtu_locked(dst))
1010 		return;
1011 
1012 	old_mtu = ipv4_mtu(dst);
1013 	if (old_mtu < mtu)
1014 		return;
1015 
1016 	if (mtu < net->ipv4.ip_rt_min_pmtu) {
1017 		lock = true;
1018 		mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1019 	}
1020 
1021 	if (rt->rt_pmtu == mtu && !lock &&
1022 	    time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2))
1023 		return;
1024 
1025 	rcu_read_lock();
1026 	if (fib_lookup(net, fl4, &res, 0) == 0) {
1027 		struct fib_nh_common *nhc;
1028 
1029 		fib_select_path(net, &res, fl4, NULL);
1030 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1031 		if (fib_info_num_path(res.fi) > 1) {
1032 			int nhsel;
1033 
1034 			for (nhsel = 0; nhsel < fib_info_num_path(res.fi); nhsel++) {
1035 				nhc = fib_info_nhc(res.fi, nhsel);
1036 				update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1037 						      jiffies + net->ipv4.ip_rt_mtu_expires);
1038 			}
1039 			rcu_read_unlock();
1040 			return;
1041 		}
1042 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1043 		nhc = FIB_RES_NHC(res);
1044 		update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1045 				      jiffies + net->ipv4.ip_rt_mtu_expires);
1046 	}
1047 	rcu_read_unlock();
1048 }
1049 
1050 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1051 			      struct sk_buff *skb, u32 mtu,
1052 			      bool confirm_neigh)
1053 {
1054 	struct rtable *rt = dst_rtable(dst);
1055 	struct flowi4 fl4;
1056 
1057 	ip_rt_build_flow_key(&fl4, sk, skb);
1058 
1059 	/* Don't make lookup fail for bridged encapsulations */
1060 	if (skb && netif_is_any_bridge_port(skb->dev))
1061 		fl4.flowi4_oif = 0;
1062 
1063 	__ip_rt_update_pmtu(rt, &fl4, mtu);
1064 }
1065 
1066 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1067 		      int oif, u8 protocol)
1068 {
1069 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1070 	struct flowi4 fl4;
1071 	struct rtable *rt;
1072 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
1073 
1074 	__build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1075 			 0);
1076 	rt = __ip_route_output_key(net, &fl4);
1077 	if (!IS_ERR(rt)) {
1078 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1079 		ip_rt_put(rt);
1080 	}
1081 }
1082 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1083 
1084 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1085 {
1086 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1087 	struct flowi4 fl4;
1088 	struct rtable *rt;
1089 
1090 	__build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1091 
1092 	if (!fl4.flowi4_mark)
1093 		fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1094 
1095 	rt = __ip_route_output_key(sock_net(sk), &fl4);
1096 	if (!IS_ERR(rt)) {
1097 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1098 		ip_rt_put(rt);
1099 	}
1100 }
1101 
1102 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1103 {
1104 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1105 	struct flowi4 fl4;
1106 	struct rtable *rt;
1107 	struct dst_entry *odst = NULL;
1108 	bool new = false;
1109 	struct net *net = sock_net(sk);
1110 
1111 	bh_lock_sock(sk);
1112 
1113 	if (!ip_sk_accept_pmtu(sk))
1114 		goto out;
1115 
1116 	odst = sk_dst_get(sk);
1117 
1118 	if (sock_owned_by_user(sk) || !odst) {
1119 		__ipv4_sk_update_pmtu(skb, sk, mtu);
1120 		goto out;
1121 	}
1122 
1123 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1124 
1125 	rt = dst_rtable(odst);
1126 	if (odst->obsolete && !odst->ops->check(odst, 0)) {
1127 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1128 		if (IS_ERR(rt))
1129 			goto out;
1130 
1131 		new = true;
1132 	}
1133 
1134 	__ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu);
1135 
1136 	if (!dst_check(&rt->dst, 0)) {
1137 		if (new)
1138 			dst_release(&rt->dst);
1139 
1140 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1141 		if (IS_ERR(rt))
1142 			goto out;
1143 
1144 		new = true;
1145 	}
1146 
1147 	if (new)
1148 		sk_dst_set(sk, &rt->dst);
1149 
1150 out:
1151 	bh_unlock_sock(sk);
1152 	dst_release(odst);
1153 }
1154 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1155 
1156 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1157 		   int oif, u8 protocol)
1158 {
1159 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1160 	struct flowi4 fl4;
1161 	struct rtable *rt;
1162 
1163 	__build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1164 	rt = __ip_route_output_key(net, &fl4);
1165 	if (!IS_ERR(rt)) {
1166 		__ip_do_redirect(rt, skb, &fl4, false);
1167 		ip_rt_put(rt);
1168 	}
1169 }
1170 EXPORT_SYMBOL_GPL(ipv4_redirect);
1171 
1172 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1173 {
1174 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1175 	struct flowi4 fl4;
1176 	struct rtable *rt;
1177 	struct net *net = sock_net(sk);
1178 
1179 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1180 	rt = __ip_route_output_key(net, &fl4);
1181 	if (!IS_ERR(rt)) {
1182 		__ip_do_redirect(rt, skb, &fl4, false);
1183 		ip_rt_put(rt);
1184 	}
1185 }
1186 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1187 
1188 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1189 							 u32 cookie)
1190 {
1191 	struct rtable *rt = dst_rtable(dst);
1192 
1193 	/* All IPV4 dsts are created with ->obsolete set to the value
1194 	 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1195 	 * into this function always.
1196 	 *
1197 	 * When a PMTU/redirect information update invalidates a route,
1198 	 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1199 	 * DST_OBSOLETE_DEAD.
1200 	 */
1201 	if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1202 		return NULL;
1203 	return dst;
1204 }
1205 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1206 
1207 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1208 {
1209 	struct net_device *dev;
1210 	struct ip_options opt;
1211 	int res;
1212 
1213 	/* Recompile ip options since IPCB may not be valid anymore.
1214 	 * Also check we have a reasonable ipv4 header.
1215 	 */
1216 	if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1217 	    ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1218 		return;
1219 
1220 	memset(&opt, 0, sizeof(opt));
1221 	if (ip_hdr(skb)->ihl > 5) {
1222 		if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1223 			return;
1224 		opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1225 
1226 		rcu_read_lock();
1227 		dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1228 		res = __ip_options_compile(dev_net(dev), &opt, skb, NULL);
1229 		rcu_read_unlock();
1230 
1231 		if (res)
1232 			return;
1233 	}
1234 	__icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1235 }
1236 
1237 static void ipv4_link_failure(struct sk_buff *skb)
1238 {
1239 	struct rtable *rt;
1240 
1241 	ipv4_send_dest_unreach(skb);
1242 
1243 	rt = skb_rtable(skb);
1244 	if (rt)
1245 		dst_set_expires(&rt->dst, 0);
1246 }
1247 
1248 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1249 {
1250 	pr_debug("%s: %pI4 -> %pI4, %s\n",
1251 		 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1252 		 skb->dev ? skb->dev->name : "?");
1253 	kfree_skb(skb);
1254 	WARN_ON(1);
1255 	return 0;
1256 }
1257 
1258 /*
1259  * We do not cache source address of outgoing interface,
1260  * because it is used only by IP RR, TS and SRR options,
1261  * so that it out of fast path.
1262  *
1263  * BTW remember: "addr" is allowed to be not aligned
1264  * in IP options!
1265  */
1266 
1267 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1268 {
1269 	__be32 src;
1270 
1271 	if (rt_is_output_route(rt))
1272 		src = ip_hdr(skb)->saddr;
1273 	else {
1274 		struct fib_result res;
1275 		struct iphdr *iph = ip_hdr(skb);
1276 		struct flowi4 fl4 = {
1277 			.daddr = iph->daddr,
1278 			.saddr = iph->saddr,
1279 			.flowi4_tos = inet_dscp_to_dsfield(ip4h_dscp(iph)),
1280 			.flowi4_oif = rt->dst.dev->ifindex,
1281 			.flowi4_iif = skb->dev->ifindex,
1282 			.flowi4_mark = skb->mark,
1283 		};
1284 
1285 		rcu_read_lock();
1286 		if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1287 			src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1288 		else
1289 			src = inet_select_addr(rt->dst.dev,
1290 					       rt_nexthop(rt, iph->daddr),
1291 					       RT_SCOPE_UNIVERSE);
1292 		rcu_read_unlock();
1293 	}
1294 	memcpy(addr, &src, 4);
1295 }
1296 
1297 #ifdef CONFIG_IP_ROUTE_CLASSID
1298 static void set_class_tag(struct rtable *rt, u32 tag)
1299 {
1300 	if (!(rt->dst.tclassid & 0xFFFF))
1301 		rt->dst.tclassid |= tag & 0xFFFF;
1302 	if (!(rt->dst.tclassid & 0xFFFF0000))
1303 		rt->dst.tclassid |= tag & 0xFFFF0000;
1304 }
1305 #endif
1306 
1307 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1308 {
1309 	struct net *net = dev_net(dst->dev);
1310 	unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1311 	unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1312 				    net->ipv4.ip_rt_min_advmss);
1313 
1314 	return min(advmss, IPV4_MAX_PMTU - header_size);
1315 }
1316 
1317 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1318 {
1319 	return ip_dst_mtu_maybe_forward(dst, false);
1320 }
1321 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1322 
1323 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1324 {
1325 	struct fnhe_hash_bucket *hash;
1326 	struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1327 	u32 hval = fnhe_hashfun(daddr);
1328 
1329 	spin_lock_bh(&fnhe_lock);
1330 
1331 	hash = rcu_dereference_protected(nhc->nhc_exceptions,
1332 					 lockdep_is_held(&fnhe_lock));
1333 	hash += hval;
1334 
1335 	fnhe_p = &hash->chain;
1336 	fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1337 	while (fnhe) {
1338 		if (fnhe->fnhe_daddr == daddr) {
1339 			rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1340 				fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1341 			/* set fnhe_daddr to 0 to ensure it won't bind with
1342 			 * new dsts in rt_bind_exception().
1343 			 */
1344 			fnhe->fnhe_daddr = 0;
1345 			fnhe_flush_routes(fnhe);
1346 			kfree_rcu(fnhe, rcu);
1347 			break;
1348 		}
1349 		fnhe_p = &fnhe->fnhe_next;
1350 		fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1351 						 lockdep_is_held(&fnhe_lock));
1352 	}
1353 
1354 	spin_unlock_bh(&fnhe_lock);
1355 }
1356 
1357 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1358 					       __be32 daddr)
1359 {
1360 	struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1361 	struct fib_nh_exception *fnhe;
1362 	u32 hval;
1363 
1364 	if (!hash)
1365 		return NULL;
1366 
1367 	hval = fnhe_hashfun(daddr);
1368 
1369 	for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1370 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
1371 		if (fnhe->fnhe_daddr == daddr) {
1372 			if (fnhe->fnhe_expires &&
1373 			    time_after(jiffies, fnhe->fnhe_expires)) {
1374 				ip_del_fnhe(nhc, daddr);
1375 				break;
1376 			}
1377 			return fnhe;
1378 		}
1379 	}
1380 	return NULL;
1381 }
1382 
1383 /* MTU selection:
1384  * 1. mtu on route is locked - use it
1385  * 2. mtu from nexthop exception
1386  * 3. mtu from egress device
1387  */
1388 
1389 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1390 {
1391 	struct fib_nh_common *nhc = res->nhc;
1392 	struct net_device *dev = nhc->nhc_dev;
1393 	struct fib_info *fi = res->fi;
1394 	u32 mtu = 0;
1395 
1396 	if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1397 	    fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1398 		mtu = fi->fib_mtu;
1399 
1400 	if (likely(!mtu)) {
1401 		struct fib_nh_exception *fnhe;
1402 
1403 		fnhe = find_exception(nhc, daddr);
1404 		if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1405 			mtu = fnhe->fnhe_pmtu;
1406 	}
1407 
1408 	if (likely(!mtu))
1409 		mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1410 
1411 	return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1412 }
1413 
1414 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1415 			      __be32 daddr, const bool do_cache)
1416 {
1417 	bool ret = false;
1418 
1419 	spin_lock_bh(&fnhe_lock);
1420 
1421 	if (daddr == fnhe->fnhe_daddr) {
1422 		struct rtable __rcu **porig;
1423 		struct rtable *orig;
1424 		int genid = fnhe_genid(dev_net(rt->dst.dev));
1425 
1426 		if (rt_is_input_route(rt))
1427 			porig = &fnhe->fnhe_rth_input;
1428 		else
1429 			porig = &fnhe->fnhe_rth_output;
1430 		orig = rcu_dereference(*porig);
1431 
1432 		if (fnhe->fnhe_genid != genid) {
1433 			fnhe->fnhe_genid = genid;
1434 			fnhe->fnhe_gw = 0;
1435 			fnhe->fnhe_pmtu = 0;
1436 			fnhe->fnhe_expires = 0;
1437 			fnhe->fnhe_mtu_locked = false;
1438 			fnhe_flush_routes(fnhe);
1439 			orig = NULL;
1440 		}
1441 		fill_route_from_fnhe(rt, fnhe);
1442 		if (!rt->rt_gw4) {
1443 			rt->rt_gw4 = daddr;
1444 			rt->rt_gw_family = AF_INET;
1445 		}
1446 
1447 		if (do_cache) {
1448 			dst_hold(&rt->dst);
1449 			rcu_assign_pointer(*porig, rt);
1450 			if (orig) {
1451 				dst_dev_put(&orig->dst);
1452 				dst_release(&orig->dst);
1453 			}
1454 			ret = true;
1455 		}
1456 
1457 		fnhe->fnhe_stamp = jiffies;
1458 	}
1459 	spin_unlock_bh(&fnhe_lock);
1460 
1461 	return ret;
1462 }
1463 
1464 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1465 {
1466 	struct rtable *orig, *prev, **p;
1467 	bool ret = true;
1468 
1469 	if (rt_is_input_route(rt)) {
1470 		p = (struct rtable **)&nhc->nhc_rth_input;
1471 	} else {
1472 		p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1473 	}
1474 	orig = *p;
1475 
1476 	/* hold dst before doing cmpxchg() to avoid race condition
1477 	 * on this dst
1478 	 */
1479 	dst_hold(&rt->dst);
1480 	prev = cmpxchg(p, orig, rt);
1481 	if (prev == orig) {
1482 		if (orig) {
1483 			rt_add_uncached_list(orig);
1484 			dst_release(&orig->dst);
1485 		}
1486 	} else {
1487 		dst_release(&rt->dst);
1488 		ret = false;
1489 	}
1490 
1491 	return ret;
1492 }
1493 
1494 struct uncached_list {
1495 	spinlock_t		lock;
1496 	struct list_head	head;
1497 };
1498 
1499 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1500 
1501 void rt_add_uncached_list(struct rtable *rt)
1502 {
1503 	struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1504 
1505 	rt->dst.rt_uncached_list = ul;
1506 
1507 	spin_lock_bh(&ul->lock);
1508 	list_add_tail(&rt->dst.rt_uncached, &ul->head);
1509 	spin_unlock_bh(&ul->lock);
1510 }
1511 
1512 void rt_del_uncached_list(struct rtable *rt)
1513 {
1514 	if (!list_empty(&rt->dst.rt_uncached)) {
1515 		struct uncached_list *ul = rt->dst.rt_uncached_list;
1516 
1517 		spin_lock_bh(&ul->lock);
1518 		list_del_init(&rt->dst.rt_uncached);
1519 		spin_unlock_bh(&ul->lock);
1520 	}
1521 }
1522 
1523 static void ipv4_dst_destroy(struct dst_entry *dst)
1524 {
1525 	ip_dst_metrics_put(dst);
1526 	rt_del_uncached_list(dst_rtable(dst));
1527 }
1528 
1529 void rt_flush_dev(struct net_device *dev)
1530 {
1531 	struct rtable *rt, *safe;
1532 	int cpu;
1533 
1534 	for_each_possible_cpu(cpu) {
1535 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1536 
1537 		if (list_empty(&ul->head))
1538 			continue;
1539 
1540 		spin_lock_bh(&ul->lock);
1541 		list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1542 			if (rt->dst.dev != dev)
1543 				continue;
1544 			rt->dst.dev = blackhole_netdev;
1545 			netdev_ref_replace(dev, blackhole_netdev,
1546 					   &rt->dst.dev_tracker, GFP_ATOMIC);
1547 			list_del_init(&rt->dst.rt_uncached);
1548 		}
1549 		spin_unlock_bh(&ul->lock);
1550 	}
1551 }
1552 
1553 static bool rt_cache_valid(const struct rtable *rt)
1554 {
1555 	return	rt &&
1556 		rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1557 		!rt_is_expired(rt);
1558 }
1559 
1560 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1561 			   const struct fib_result *res,
1562 			   struct fib_nh_exception *fnhe,
1563 			   struct fib_info *fi, u16 type, u32 itag,
1564 			   const bool do_cache)
1565 {
1566 	bool cached = false;
1567 
1568 	if (fi) {
1569 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1570 
1571 		if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1572 			rt->rt_uses_gateway = 1;
1573 			rt->rt_gw_family = nhc->nhc_gw_family;
1574 			/* only INET and INET6 are supported */
1575 			if (likely(nhc->nhc_gw_family == AF_INET))
1576 				rt->rt_gw4 = nhc->nhc_gw.ipv4;
1577 			else
1578 				rt->rt_gw6 = nhc->nhc_gw.ipv6;
1579 		}
1580 
1581 		ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1582 
1583 #ifdef CONFIG_IP_ROUTE_CLASSID
1584 		if (nhc->nhc_family == AF_INET) {
1585 			struct fib_nh *nh;
1586 
1587 			nh = container_of(nhc, struct fib_nh, nh_common);
1588 			rt->dst.tclassid = nh->nh_tclassid;
1589 		}
1590 #endif
1591 		rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1592 		if (unlikely(fnhe))
1593 			cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1594 		else if (do_cache)
1595 			cached = rt_cache_route(nhc, rt);
1596 		if (unlikely(!cached)) {
1597 			/* Routes we intend to cache in nexthop exception or
1598 			 * FIB nexthop have the DST_NOCACHE bit clear.
1599 			 * However, if we are unsuccessful at storing this
1600 			 * route into the cache we really need to set it.
1601 			 */
1602 			if (!rt->rt_gw4) {
1603 				rt->rt_gw_family = AF_INET;
1604 				rt->rt_gw4 = daddr;
1605 			}
1606 			rt_add_uncached_list(rt);
1607 		}
1608 	} else
1609 		rt_add_uncached_list(rt);
1610 
1611 #ifdef CONFIG_IP_ROUTE_CLASSID
1612 #ifdef CONFIG_IP_MULTIPLE_TABLES
1613 	set_class_tag(rt, res->tclassid);
1614 #endif
1615 	set_class_tag(rt, itag);
1616 #endif
1617 }
1618 
1619 struct rtable *rt_dst_alloc(struct net_device *dev,
1620 			    unsigned int flags, u16 type,
1621 			    bool noxfrm)
1622 {
1623 	struct rtable *rt;
1624 
1625 	rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1626 		       (noxfrm ? DST_NOXFRM : 0));
1627 
1628 	if (rt) {
1629 		rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1630 		rt->rt_flags = flags;
1631 		rt->rt_type = type;
1632 		rt->rt_is_input = 0;
1633 		rt->rt_iif = 0;
1634 		rt->rt_pmtu = 0;
1635 		rt->rt_mtu_locked = 0;
1636 		rt->rt_uses_gateway = 0;
1637 		rt->rt_gw_family = 0;
1638 		rt->rt_gw4 = 0;
1639 
1640 		rt->dst.output = ip_output;
1641 		if (flags & RTCF_LOCAL)
1642 			rt->dst.input = ip_local_deliver;
1643 	}
1644 
1645 	return rt;
1646 }
1647 EXPORT_SYMBOL(rt_dst_alloc);
1648 
1649 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1650 {
1651 	struct rtable *new_rt;
1652 
1653 	new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1654 			   rt->dst.flags);
1655 
1656 	if (new_rt) {
1657 		new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1658 		new_rt->rt_flags = rt->rt_flags;
1659 		new_rt->rt_type = rt->rt_type;
1660 		new_rt->rt_is_input = rt->rt_is_input;
1661 		new_rt->rt_iif = rt->rt_iif;
1662 		new_rt->rt_pmtu = rt->rt_pmtu;
1663 		new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1664 		new_rt->rt_gw_family = rt->rt_gw_family;
1665 		if (rt->rt_gw_family == AF_INET)
1666 			new_rt->rt_gw4 = rt->rt_gw4;
1667 		else if (rt->rt_gw_family == AF_INET6)
1668 			new_rt->rt_gw6 = rt->rt_gw6;
1669 
1670 		new_rt->dst.input = rt->dst.input;
1671 		new_rt->dst.output = rt->dst.output;
1672 		new_rt->dst.error = rt->dst.error;
1673 		new_rt->dst.lastuse = jiffies;
1674 		new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1675 	}
1676 	return new_rt;
1677 }
1678 EXPORT_SYMBOL(rt_dst_clone);
1679 
1680 /* called in rcu_read_lock() section */
1681 enum skb_drop_reason
1682 ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1683 		      dscp_t dscp, struct net_device *dev,
1684 		      struct in_device *in_dev, u32 *itag)
1685 {
1686 	enum skb_drop_reason reason;
1687 
1688 	/* Primary sanity checks. */
1689 	if (!in_dev)
1690 		return SKB_DROP_REASON_NOT_SPECIFIED;
1691 
1692 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1693 		return SKB_DROP_REASON_IP_INVALID_SOURCE;
1694 
1695 	if (skb->protocol != htons(ETH_P_IP))
1696 		return SKB_DROP_REASON_INVALID_PROTO;
1697 
1698 	if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1699 		return SKB_DROP_REASON_IP_LOCALNET;
1700 
1701 	if (ipv4_is_zeronet(saddr)) {
1702 		if (!ipv4_is_local_multicast(daddr) &&
1703 		    ip_hdr(skb)->protocol != IPPROTO_IGMP)
1704 			return SKB_DROP_REASON_IP_INVALID_SOURCE;
1705 	} else {
1706 		reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
1707 						    dev, in_dev, itag);
1708 		if (reason)
1709 			return reason;
1710 	}
1711 	return SKB_NOT_DROPPED_YET;
1712 }
1713 
1714 /* called in rcu_read_lock() section */
1715 static enum skb_drop_reason
1716 ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1717 		  dscp_t dscp, struct net_device *dev, int our)
1718 {
1719 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1720 	unsigned int flags = RTCF_MULTICAST;
1721 	enum skb_drop_reason reason;
1722 	struct rtable *rth;
1723 	u32 itag = 0;
1724 
1725 	reason = ip_mc_validate_source(skb, daddr, saddr, dscp, dev, in_dev,
1726 				       &itag);
1727 	if (reason)
1728 		return reason;
1729 
1730 	if (our)
1731 		flags |= RTCF_LOCAL;
1732 
1733 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1734 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
1735 
1736 	rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1737 			   false);
1738 	if (!rth)
1739 		return SKB_DROP_REASON_NOMEM;
1740 
1741 #ifdef CONFIG_IP_ROUTE_CLASSID
1742 	rth->dst.tclassid = itag;
1743 #endif
1744 	rth->dst.output = ip_rt_bug;
1745 	rth->rt_is_input= 1;
1746 
1747 #ifdef CONFIG_IP_MROUTE
1748 	if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1749 		rth->dst.input = ip_mr_input;
1750 #endif
1751 	RT_CACHE_STAT_INC(in_slow_mc);
1752 
1753 	skb_dst_drop(skb);
1754 	skb_dst_set(skb, &rth->dst);
1755 	return SKB_NOT_DROPPED_YET;
1756 }
1757 
1758 
1759 static void ip_handle_martian_source(struct net_device *dev,
1760 				     struct in_device *in_dev,
1761 				     struct sk_buff *skb,
1762 				     __be32 daddr,
1763 				     __be32 saddr)
1764 {
1765 	RT_CACHE_STAT_INC(in_martian_src);
1766 #ifdef CONFIG_IP_ROUTE_VERBOSE
1767 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1768 		/*
1769 		 *	RFC1812 recommendation, if source is martian,
1770 		 *	the only hint is MAC header.
1771 		 */
1772 		pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1773 			&daddr, &saddr, dev->name);
1774 		if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1775 			print_hex_dump(KERN_WARNING, "ll header: ",
1776 				       DUMP_PREFIX_OFFSET, 16, 1,
1777 				       skb_mac_header(skb),
1778 				       dev->hard_header_len, false);
1779 		}
1780 	}
1781 #endif
1782 }
1783 
1784 /* called in rcu_read_lock() section */
1785 static enum skb_drop_reason
1786 __mkroute_input(struct sk_buff *skb, const struct fib_result *res,
1787 		struct in_device *in_dev, __be32 daddr,
1788 		__be32 saddr, dscp_t dscp)
1789 {
1790 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1791 	struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1792 	struct net_device *dev = nhc->nhc_dev;
1793 	struct fib_nh_exception *fnhe;
1794 	struct rtable *rth;
1795 	int err;
1796 	struct in_device *out_dev;
1797 	bool do_cache;
1798 	u32 itag = 0;
1799 
1800 	/* get a working reference to the output device */
1801 	out_dev = __in_dev_get_rcu(dev);
1802 	if (!out_dev) {
1803 		net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1804 		return reason;
1805 	}
1806 
1807 	err = fib_validate_source(skb, saddr, daddr, dscp, FIB_RES_OIF(*res),
1808 				  in_dev->dev, in_dev, &itag);
1809 	if (err < 0) {
1810 		reason = -err;
1811 		ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1812 					 saddr);
1813 
1814 		goto cleanup;
1815 	}
1816 
1817 	do_cache = res->fi && !itag;
1818 	if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1819 	    skb->protocol == htons(ETH_P_IP)) {
1820 		__be32 gw;
1821 
1822 		gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1823 		if (IN_DEV_SHARED_MEDIA(out_dev) ||
1824 		    inet_addr_onlink(out_dev, saddr, gw))
1825 			IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1826 	}
1827 
1828 	if (skb->protocol != htons(ETH_P_IP)) {
1829 		/* Not IP (i.e. ARP). Do not create route, if it is
1830 		 * invalid for proxy arp. DNAT routes are always valid.
1831 		 *
1832 		 * Proxy arp feature have been extended to allow, ARP
1833 		 * replies back to the same interface, to support
1834 		 * Private VLAN switch technologies. See arp.c.
1835 		 */
1836 		if (out_dev == in_dev &&
1837 		    IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1838 			reason = SKB_DROP_REASON_ARP_PVLAN_DISABLE;
1839 			goto cleanup;
1840 		}
1841 	}
1842 
1843 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1844 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
1845 
1846 	fnhe = find_exception(nhc, daddr);
1847 	if (do_cache) {
1848 		if (fnhe)
1849 			rth = rcu_dereference(fnhe->fnhe_rth_input);
1850 		else
1851 			rth = rcu_dereference(nhc->nhc_rth_input);
1852 		if (rt_cache_valid(rth)) {
1853 			skb_dst_set_noref(skb, &rth->dst);
1854 			goto out;
1855 		}
1856 	}
1857 
1858 	rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1859 			   IN_DEV_ORCONF(out_dev, NOXFRM));
1860 	if (!rth) {
1861 		reason = SKB_DROP_REASON_NOMEM;
1862 		goto cleanup;
1863 	}
1864 
1865 	rth->rt_is_input = 1;
1866 	RT_CACHE_STAT_INC(in_slow_tot);
1867 
1868 	rth->dst.input = ip_forward;
1869 
1870 	rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1871 		       do_cache);
1872 	lwtunnel_set_redirect(&rth->dst);
1873 	skb_dst_set(skb, &rth->dst);
1874 out:
1875 	reason = SKB_NOT_DROPPED_YET;
1876 cleanup:
1877 	return reason;
1878 }
1879 
1880 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1881 /* To make ICMP packets follow the right flow, the multipath hash is
1882  * calculated from the inner IP addresses.
1883  */
1884 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1885 				 struct flow_keys *hash_keys)
1886 {
1887 	const struct iphdr *outer_iph = ip_hdr(skb);
1888 	const struct iphdr *key_iph = outer_iph;
1889 	const struct iphdr *inner_iph;
1890 	const struct icmphdr *icmph;
1891 	struct iphdr _inner_iph;
1892 	struct icmphdr _icmph;
1893 
1894 	if (likely(outer_iph->protocol != IPPROTO_ICMP))
1895 		goto out;
1896 
1897 	if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1898 		goto out;
1899 
1900 	icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1901 				   &_icmph);
1902 	if (!icmph)
1903 		goto out;
1904 
1905 	if (!icmp_is_err(icmph->type))
1906 		goto out;
1907 
1908 	inner_iph = skb_header_pointer(skb,
1909 				       outer_iph->ihl * 4 + sizeof(_icmph),
1910 				       sizeof(_inner_iph), &_inner_iph);
1911 	if (!inner_iph)
1912 		goto out;
1913 
1914 	key_iph = inner_iph;
1915 out:
1916 	hash_keys->addrs.v4addrs.src = key_iph->saddr;
1917 	hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1918 }
1919 
1920 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1921 					   const struct sk_buff *skb,
1922 					   bool *p_has_inner)
1923 {
1924 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1925 	struct flow_keys keys, hash_keys;
1926 
1927 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1928 		return 0;
1929 
1930 	memset(&hash_keys, 0, sizeof(hash_keys));
1931 	skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1932 
1933 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1934 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1935 		hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1936 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1937 		hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1938 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1939 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
1940 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1941 		hash_keys.ports.src = keys.ports.src;
1942 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1943 		hash_keys.ports.dst = keys.ports.dst;
1944 
1945 	*p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1946 	return fib_multipath_hash_from_keys(net, &hash_keys);
1947 }
1948 
1949 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1950 					   const struct sk_buff *skb,
1951 					   bool has_inner)
1952 {
1953 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1954 	struct flow_keys keys, hash_keys;
1955 
1956 	/* We assume the packet carries an encapsulation, but if none was
1957 	 * encountered during dissection of the outer flow, then there is no
1958 	 * point in calling the flow dissector again.
1959 	 */
1960 	if (!has_inner)
1961 		return 0;
1962 
1963 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1964 		return 0;
1965 
1966 	memset(&hash_keys, 0, sizeof(hash_keys));
1967 	skb_flow_dissect_flow_keys(skb, &keys, 0);
1968 
1969 	if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1970 		return 0;
1971 
1972 	if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1973 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1974 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1975 			hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1976 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1977 			hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1978 	} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1979 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1980 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1981 			hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1982 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1983 			hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1984 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1985 			hash_keys.tags.flow_label = keys.tags.flow_label;
1986 	}
1987 
1988 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1989 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
1990 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1991 		hash_keys.ports.src = keys.ports.src;
1992 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
1993 		hash_keys.ports.dst = keys.ports.dst;
1994 
1995 	return fib_multipath_hash_from_keys(net, &hash_keys);
1996 }
1997 
1998 static u32 fib_multipath_custom_hash_skb(const struct net *net,
1999 					 const struct sk_buff *skb)
2000 {
2001 	u32 mhash, mhash_inner;
2002 	bool has_inner = true;
2003 
2004 	mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2005 	mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2006 
2007 	return jhash_2words(mhash, mhash_inner, 0);
2008 }
2009 
2010 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2011 					 const struct flowi4 *fl4)
2012 {
2013 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2014 	struct flow_keys hash_keys;
2015 
2016 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2017 		return 0;
2018 
2019 	memset(&hash_keys, 0, sizeof(hash_keys));
2020 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2021 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2022 		hash_keys.addrs.v4addrs.src = fl4->saddr;
2023 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2024 		hash_keys.addrs.v4addrs.dst = fl4->daddr;
2025 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2026 		hash_keys.basic.ip_proto = fl4->flowi4_proto;
2027 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2028 		hash_keys.ports.src = fl4->fl4_sport;
2029 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2030 		hash_keys.ports.dst = fl4->fl4_dport;
2031 
2032 	return fib_multipath_hash_from_keys(net, &hash_keys);
2033 }
2034 
2035 /* if skb is set it will be used and fl4 can be NULL */
2036 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2037 		       const struct sk_buff *skb, struct flow_keys *flkeys)
2038 {
2039 	u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2040 	struct flow_keys hash_keys;
2041 	u32 mhash = 0;
2042 
2043 	switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2044 	case 0:
2045 		memset(&hash_keys, 0, sizeof(hash_keys));
2046 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2047 		if (skb) {
2048 			ip_multipath_l3_keys(skb, &hash_keys);
2049 		} else {
2050 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2051 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2052 		}
2053 		mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2054 		break;
2055 	case 1:
2056 		/* skb is currently provided only when forwarding */
2057 		if (skb) {
2058 			unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2059 			struct flow_keys keys;
2060 
2061 			/* short-circuit if we already have L4 hash present */
2062 			if (skb->l4_hash)
2063 				return skb_get_hash_raw(skb) >> 1;
2064 
2065 			memset(&hash_keys, 0, sizeof(hash_keys));
2066 
2067 			if (!flkeys) {
2068 				skb_flow_dissect_flow_keys(skb, &keys, flag);
2069 				flkeys = &keys;
2070 			}
2071 
2072 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2073 			hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2074 			hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2075 			hash_keys.ports.src = flkeys->ports.src;
2076 			hash_keys.ports.dst = flkeys->ports.dst;
2077 			hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2078 		} else {
2079 			memset(&hash_keys, 0, sizeof(hash_keys));
2080 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2081 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2082 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2083 			hash_keys.ports.src = fl4->fl4_sport;
2084 			hash_keys.ports.dst = fl4->fl4_dport;
2085 			hash_keys.basic.ip_proto = fl4->flowi4_proto;
2086 		}
2087 		mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2088 		break;
2089 	case 2:
2090 		memset(&hash_keys, 0, sizeof(hash_keys));
2091 		/* skb is currently provided only when forwarding */
2092 		if (skb) {
2093 			struct flow_keys keys;
2094 
2095 			skb_flow_dissect_flow_keys(skb, &keys, 0);
2096 			/* Inner can be v4 or v6 */
2097 			if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2098 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2099 				hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2100 				hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2101 			} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2102 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2103 				hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2104 				hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2105 				hash_keys.tags.flow_label = keys.tags.flow_label;
2106 				hash_keys.basic.ip_proto = keys.basic.ip_proto;
2107 			} else {
2108 				/* Same as case 0 */
2109 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2110 				ip_multipath_l3_keys(skb, &hash_keys);
2111 			}
2112 		} else {
2113 			/* Same as case 0 */
2114 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2115 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2116 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2117 		}
2118 		mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2119 		break;
2120 	case 3:
2121 		if (skb)
2122 			mhash = fib_multipath_custom_hash_skb(net, skb);
2123 		else
2124 			mhash = fib_multipath_custom_hash_fl4(net, fl4);
2125 		break;
2126 	}
2127 
2128 	if (multipath_hash)
2129 		mhash = jhash_2words(mhash, multipath_hash, 0);
2130 
2131 	return mhash >> 1;
2132 }
2133 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2134 
2135 static enum skb_drop_reason
2136 ip_mkroute_input(struct sk_buff *skb, struct fib_result *res,
2137 		 struct in_device *in_dev, __be32 daddr,
2138 		 __be32 saddr, dscp_t dscp, struct flow_keys *hkeys)
2139 {
2140 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2141 	if (res->fi && fib_info_num_path(res->fi) > 1) {
2142 		int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2143 
2144 		fib_select_multipath(res, h);
2145 		IPCB(skb)->flags |= IPSKB_MULTIPATH;
2146 	}
2147 #endif
2148 
2149 	/* create a routing cache entry */
2150 	return __mkroute_input(skb, res, in_dev, daddr, saddr, dscp);
2151 }
2152 
2153 /* Implements all the saddr-related checks as ip_route_input_slow(),
2154  * assuming daddr is valid and the destination is not a local broadcast one.
2155  * Uses the provided hint instead of performing a route lookup.
2156  */
2157 enum skb_drop_reason
2158 ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2159 		  dscp_t dscp, struct net_device *dev,
2160 		  const struct sk_buff *hint)
2161 {
2162 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2163 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2164 	struct rtable *rt = skb_rtable(hint);
2165 	struct net *net = dev_net(dev);
2166 	u32 tag = 0;
2167 
2168 	if (!in_dev)
2169 		return reason;
2170 
2171 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2172 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2173 		goto martian_source;
2174 	}
2175 
2176 	if (ipv4_is_zeronet(saddr)) {
2177 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2178 		goto martian_source;
2179 	}
2180 
2181 	if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2182 		reason = SKB_DROP_REASON_IP_LOCALNET;
2183 		goto martian_source;
2184 	}
2185 
2186 	if (rt->rt_type != RTN_LOCAL)
2187 		goto skip_validate_source;
2188 
2189 	reason = fib_validate_source_reason(skb, saddr, daddr, dscp, 0, dev,
2190 					    in_dev, &tag);
2191 	if (reason)
2192 		goto martian_source;
2193 
2194 skip_validate_source:
2195 	skb_dst_copy(skb, hint);
2196 	return SKB_NOT_DROPPED_YET;
2197 
2198 martian_source:
2199 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2200 	return reason;
2201 }
2202 
2203 /* get device for dst_alloc with local routes */
2204 static struct net_device *ip_rt_get_dev(struct net *net,
2205 					const struct fib_result *res)
2206 {
2207 	struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2208 	struct net_device *dev = NULL;
2209 
2210 	if (nhc)
2211 		dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2212 
2213 	return dev ? : net->loopback_dev;
2214 }
2215 
2216 /*
2217  *	NOTE. We drop all the packets that has local source
2218  *	addresses, because every properly looped back packet
2219  *	must have correct destination already attached by output routine.
2220  *	Changes in the enforced policies must be applied also to
2221  *	ip_route_use_hint().
2222  *
2223  *	Such approach solves two big problems:
2224  *	1. Not simplex devices are handled properly.
2225  *	2. IP spoofing attempts are filtered with 100% of guarantee.
2226  *	called with rcu_read_lock()
2227  */
2228 
2229 static enum skb_drop_reason
2230 ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2231 		    dscp_t dscp, struct net_device *dev,
2232 		    struct fib_result *res)
2233 {
2234 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2235 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2236 	struct flow_keys *flkeys = NULL, _flkeys;
2237 	struct net    *net = dev_net(dev);
2238 	struct ip_tunnel_info *tun_info;
2239 	int		err = -EINVAL;
2240 	unsigned int	flags = 0;
2241 	u32		itag = 0;
2242 	struct rtable	*rth;
2243 	struct flowi4	fl4;
2244 	bool do_cache = true;
2245 
2246 	/* IP on this device is disabled. */
2247 
2248 	if (!in_dev)
2249 		goto out;
2250 
2251 	/* Check for the most weird martians, which can be not detected
2252 	 * by fib_lookup.
2253 	 */
2254 
2255 	tun_info = skb_tunnel_info(skb);
2256 	if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2257 		fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2258 	else
2259 		fl4.flowi4_tun_key.tun_id = 0;
2260 	skb_dst_drop(skb);
2261 
2262 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2263 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2264 		goto martian_source;
2265 	}
2266 
2267 	res->fi = NULL;
2268 	res->table = NULL;
2269 	if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2270 		goto brd_input;
2271 
2272 	/* Accept zero addresses only to limited broadcast;
2273 	 * I even do not know to fix it or not. Waiting for complains :-)
2274 	 */
2275 	if (ipv4_is_zeronet(saddr)) {
2276 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2277 		goto martian_source;
2278 	}
2279 
2280 	if (ipv4_is_zeronet(daddr)) {
2281 		reason = SKB_DROP_REASON_IP_INVALID_DEST;
2282 		goto martian_destination;
2283 	}
2284 
2285 	/* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2286 	 * and call it once if daddr or/and saddr are loopback addresses
2287 	 */
2288 	if (ipv4_is_loopback(daddr)) {
2289 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2290 			reason = SKB_DROP_REASON_IP_LOCALNET;
2291 			goto martian_destination;
2292 		}
2293 	} else if (ipv4_is_loopback(saddr)) {
2294 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2295 			reason = SKB_DROP_REASON_IP_LOCALNET;
2296 			goto martian_source;
2297 		}
2298 	}
2299 
2300 	/*
2301 	 *	Now we are ready to route packet.
2302 	 */
2303 	fl4.flowi4_l3mdev = 0;
2304 	fl4.flowi4_oif = 0;
2305 	fl4.flowi4_iif = dev->ifindex;
2306 	fl4.flowi4_mark = skb->mark;
2307 	fl4.flowi4_tos = inet_dscp_to_dsfield(dscp);
2308 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2309 	fl4.flowi4_flags = 0;
2310 	fl4.daddr = daddr;
2311 	fl4.saddr = saddr;
2312 	fl4.flowi4_uid = sock_net_uid(net, NULL);
2313 	fl4.flowi4_multipath_hash = 0;
2314 
2315 	if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2316 		flkeys = &_flkeys;
2317 	} else {
2318 		fl4.flowi4_proto = 0;
2319 		fl4.fl4_sport = 0;
2320 		fl4.fl4_dport = 0;
2321 	}
2322 
2323 	err = fib_lookup(net, &fl4, res, 0);
2324 	if (err != 0) {
2325 		if (!IN_DEV_FORWARD(in_dev))
2326 			err = -EHOSTUNREACH;
2327 		goto no_route;
2328 	}
2329 
2330 	if (res->type == RTN_BROADCAST) {
2331 		if (IN_DEV_BFORWARD(in_dev))
2332 			goto make_route;
2333 		/* not do cache if bc_forwarding is enabled */
2334 		if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING))
2335 			do_cache = false;
2336 		goto brd_input;
2337 	}
2338 
2339 	err = -EINVAL;
2340 	if (res->type == RTN_LOCAL) {
2341 		reason = fib_validate_source_reason(skb, saddr, daddr, dscp,
2342 						    0, dev, in_dev, &itag);
2343 		if (reason)
2344 			goto martian_source;
2345 		goto local_input;
2346 	}
2347 
2348 	if (!IN_DEV_FORWARD(in_dev)) {
2349 		err = -EHOSTUNREACH;
2350 		goto no_route;
2351 	}
2352 	if (res->type != RTN_UNICAST) {
2353 		reason = SKB_DROP_REASON_IP_INVALID_DEST;
2354 		goto martian_destination;
2355 	}
2356 
2357 make_route:
2358 	reason = ip_mkroute_input(skb, res, in_dev, daddr, saddr, dscp,
2359 				  flkeys);
2360 
2361 out:
2362 	return reason;
2363 
2364 brd_input:
2365 	if (skb->protocol != htons(ETH_P_IP)) {
2366 		reason = SKB_DROP_REASON_INVALID_PROTO;
2367 		goto out;
2368 	}
2369 
2370 	if (!ipv4_is_zeronet(saddr)) {
2371 		reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
2372 						    dev, in_dev, &itag);
2373 		if (reason)
2374 			goto martian_source;
2375 	}
2376 	flags |= RTCF_BROADCAST;
2377 	res->type = RTN_BROADCAST;
2378 	RT_CACHE_STAT_INC(in_brd);
2379 
2380 local_input:
2381 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2382 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
2383 
2384 	do_cache &= res->fi && !itag;
2385 	if (do_cache) {
2386 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2387 
2388 		rth = rcu_dereference(nhc->nhc_rth_input);
2389 		if (rt_cache_valid(rth)) {
2390 			skb_dst_set_noref(skb, &rth->dst);
2391 			reason = SKB_NOT_DROPPED_YET;
2392 			goto out;
2393 		}
2394 	}
2395 
2396 	rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2397 			   flags | RTCF_LOCAL, res->type, false);
2398 	if (!rth)
2399 		goto e_nobufs;
2400 
2401 	rth->dst.output= ip_rt_bug;
2402 #ifdef CONFIG_IP_ROUTE_CLASSID
2403 	rth->dst.tclassid = itag;
2404 #endif
2405 	rth->rt_is_input = 1;
2406 
2407 	RT_CACHE_STAT_INC(in_slow_tot);
2408 	if (res->type == RTN_UNREACHABLE) {
2409 		rth->dst.input= ip_error;
2410 		rth->dst.error= -err;
2411 		rth->rt_flags	&= ~RTCF_LOCAL;
2412 	}
2413 
2414 	if (do_cache) {
2415 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2416 
2417 		rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2418 		if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2419 			WARN_ON(rth->dst.input == lwtunnel_input);
2420 			rth->dst.lwtstate->orig_input = rth->dst.input;
2421 			rth->dst.input = lwtunnel_input;
2422 		}
2423 
2424 		if (unlikely(!rt_cache_route(nhc, rth)))
2425 			rt_add_uncached_list(rth);
2426 	}
2427 	skb_dst_set(skb, &rth->dst);
2428 	reason = SKB_NOT_DROPPED_YET;
2429 	goto out;
2430 
2431 no_route:
2432 	RT_CACHE_STAT_INC(in_no_route);
2433 	res->type = RTN_UNREACHABLE;
2434 	res->fi = NULL;
2435 	res->table = NULL;
2436 	goto local_input;
2437 
2438 	/*
2439 	 *	Do not cache martian addresses: they should be logged (RFC1812)
2440 	 */
2441 martian_destination:
2442 	RT_CACHE_STAT_INC(in_martian_dst);
2443 #ifdef CONFIG_IP_ROUTE_VERBOSE
2444 	if (IN_DEV_LOG_MARTIANS(in_dev))
2445 		net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2446 				     &daddr, &saddr, dev->name);
2447 #endif
2448 
2449 e_nobufs:
2450 	reason = SKB_DROP_REASON_NOMEM;
2451 	goto out;
2452 
2453 martian_source:
2454 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2455 	goto out;
2456 }
2457 
2458 /* called with rcu_read_lock held */
2459 static enum skb_drop_reason
2460 ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2461 		   dscp_t dscp, struct net_device *dev,
2462 		   struct fib_result *res)
2463 {
2464 	/* Multicast recognition logic is moved from route cache to here.
2465 	 * The problem was that too many Ethernet cards have broken/missing
2466 	 * hardware multicast filters :-( As result the host on multicasting
2467 	 * network acquires a lot of useless route cache entries, sort of
2468 	 * SDR messages from all the world. Now we try to get rid of them.
2469 	 * Really, provided software IP multicast filter is organized
2470 	 * reasonably (at least, hashed), it does not result in a slowdown
2471 	 * comparing with route cache reject entries.
2472 	 * Note, that multicast routers are not affected, because
2473 	 * route cache entry is created eventually.
2474 	 */
2475 	if (ipv4_is_multicast(daddr)) {
2476 		enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2477 		struct in_device *in_dev = __in_dev_get_rcu(dev);
2478 		int our = 0;
2479 
2480 		if (!in_dev)
2481 			return -EINVAL;
2482 		our = ip_check_mc_rcu(in_dev, daddr, saddr,
2483 				      ip_hdr(skb)->protocol);
2484 
2485 		/* check l3 master if no match yet */
2486 		if (!our && netif_is_l3_slave(dev)) {
2487 			struct in_device *l3_in_dev;
2488 
2489 			l3_in_dev = __in_dev_get_rcu(skb->dev);
2490 			if (l3_in_dev)
2491 				our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2492 						      ip_hdr(skb)->protocol);
2493 		}
2494 
2495 		if (our
2496 #ifdef CONFIG_IP_MROUTE
2497 			||
2498 		    (!ipv4_is_local_multicast(daddr) &&
2499 		     IN_DEV_MFORWARD(in_dev))
2500 #endif
2501 		   ) {
2502 			reason = ip_route_input_mc(skb, daddr, saddr, dscp,
2503 						   dev, our);
2504 		}
2505 		return reason;
2506 	}
2507 
2508 	return ip_route_input_slow(skb, daddr, saddr, dscp, dev, res);
2509 }
2510 
2511 enum skb_drop_reason ip_route_input_noref(struct sk_buff *skb, __be32 daddr,
2512 					  __be32 saddr, dscp_t dscp,
2513 					  struct net_device *dev)
2514 {
2515 	enum skb_drop_reason reason;
2516 	struct fib_result res;
2517 
2518 	rcu_read_lock();
2519 	reason = ip_route_input_rcu(skb, daddr, saddr, dscp, dev, &res);
2520 	rcu_read_unlock();
2521 
2522 	return reason;
2523 }
2524 EXPORT_SYMBOL(ip_route_input_noref);
2525 
2526 /* called with rcu_read_lock() */
2527 static struct rtable *__mkroute_output(const struct fib_result *res,
2528 				       const struct flowi4 *fl4, int orig_oif,
2529 				       struct net_device *dev_out,
2530 				       unsigned int flags)
2531 {
2532 	struct fib_info *fi = res->fi;
2533 	struct fib_nh_exception *fnhe;
2534 	struct in_device *in_dev;
2535 	u16 type = res->type;
2536 	struct rtable *rth;
2537 	bool do_cache;
2538 
2539 	in_dev = __in_dev_get_rcu(dev_out);
2540 	if (!in_dev)
2541 		return ERR_PTR(-EINVAL);
2542 
2543 	if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2544 		if (ipv4_is_loopback(fl4->saddr) &&
2545 		    !(dev_out->flags & IFF_LOOPBACK) &&
2546 		    !netif_is_l3_master(dev_out))
2547 			return ERR_PTR(-EINVAL);
2548 
2549 	if (ipv4_is_lbcast(fl4->daddr))
2550 		type = RTN_BROADCAST;
2551 	else if (ipv4_is_multicast(fl4->daddr))
2552 		type = RTN_MULTICAST;
2553 	else if (ipv4_is_zeronet(fl4->daddr))
2554 		return ERR_PTR(-EINVAL);
2555 
2556 	if (dev_out->flags & IFF_LOOPBACK)
2557 		flags |= RTCF_LOCAL;
2558 
2559 	do_cache = true;
2560 	if (type == RTN_BROADCAST) {
2561 		flags |= RTCF_BROADCAST | RTCF_LOCAL;
2562 		fi = NULL;
2563 	} else if (type == RTN_MULTICAST) {
2564 		flags |= RTCF_MULTICAST | RTCF_LOCAL;
2565 		if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2566 				     fl4->flowi4_proto))
2567 			flags &= ~RTCF_LOCAL;
2568 		else
2569 			do_cache = false;
2570 		/* If multicast route do not exist use
2571 		 * default one, but do not gateway in this case.
2572 		 * Yes, it is hack.
2573 		 */
2574 		if (fi && res->prefixlen < 4)
2575 			fi = NULL;
2576 	} else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2577 		   (orig_oif != dev_out->ifindex)) {
2578 		/* For local routes that require a particular output interface
2579 		 * we do not want to cache the result.  Caching the result
2580 		 * causes incorrect behaviour when there are multiple source
2581 		 * addresses on the interface, the end result being that if the
2582 		 * intended recipient is waiting on that interface for the
2583 		 * packet he won't receive it because it will be delivered on
2584 		 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2585 		 * be set to the loopback interface as well.
2586 		 */
2587 		do_cache = false;
2588 	}
2589 
2590 	fnhe = NULL;
2591 	do_cache &= fi != NULL;
2592 	if (fi) {
2593 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2594 		struct rtable __rcu **prth;
2595 
2596 		fnhe = find_exception(nhc, fl4->daddr);
2597 		if (!do_cache)
2598 			goto add;
2599 		if (fnhe) {
2600 			prth = &fnhe->fnhe_rth_output;
2601 		} else {
2602 			if (unlikely(fl4->flowi4_flags &
2603 				     FLOWI_FLAG_KNOWN_NH &&
2604 				     !(nhc->nhc_gw_family &&
2605 				       nhc->nhc_scope == RT_SCOPE_LINK))) {
2606 				do_cache = false;
2607 				goto add;
2608 			}
2609 			prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2610 		}
2611 		rth = rcu_dereference(*prth);
2612 		if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2613 			return rth;
2614 	}
2615 
2616 add:
2617 	rth = rt_dst_alloc(dev_out, flags, type,
2618 			   IN_DEV_ORCONF(in_dev, NOXFRM));
2619 	if (!rth)
2620 		return ERR_PTR(-ENOBUFS);
2621 
2622 	rth->rt_iif = orig_oif;
2623 
2624 	RT_CACHE_STAT_INC(out_slow_tot);
2625 
2626 	if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2627 		if (flags & RTCF_LOCAL &&
2628 		    !(dev_out->flags & IFF_LOOPBACK)) {
2629 			rth->dst.output = ip_mc_output;
2630 			RT_CACHE_STAT_INC(out_slow_mc);
2631 		}
2632 #ifdef CONFIG_IP_MROUTE
2633 		if (type == RTN_MULTICAST) {
2634 			if (IN_DEV_MFORWARD(in_dev) &&
2635 			    !ipv4_is_local_multicast(fl4->daddr)) {
2636 				rth->dst.input = ip_mr_input;
2637 				rth->dst.output = ip_mc_output;
2638 			}
2639 		}
2640 #endif
2641 	}
2642 
2643 	rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2644 	lwtunnel_set_redirect(&rth->dst);
2645 
2646 	return rth;
2647 }
2648 
2649 /*
2650  * Major route resolver routine.
2651  */
2652 
2653 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2654 					const struct sk_buff *skb)
2655 {
2656 	struct fib_result res = {
2657 		.type		= RTN_UNSPEC,
2658 		.fi		= NULL,
2659 		.table		= NULL,
2660 		.tclassid	= 0,
2661 	};
2662 	struct rtable *rth;
2663 
2664 	fl4->flowi4_iif = LOOPBACK_IFINDEX;
2665 	fl4->flowi4_tos &= INET_DSCP_MASK;
2666 
2667 	rcu_read_lock();
2668 	rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2669 	rcu_read_unlock();
2670 
2671 	return rth;
2672 }
2673 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2674 
2675 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2676 					    struct fib_result *res,
2677 					    const struct sk_buff *skb)
2678 {
2679 	struct net_device *dev_out = NULL;
2680 	int orig_oif = fl4->flowi4_oif;
2681 	unsigned int flags = 0;
2682 	struct rtable *rth;
2683 	int err;
2684 
2685 	if (fl4->saddr) {
2686 		if (ipv4_is_multicast(fl4->saddr) ||
2687 		    ipv4_is_lbcast(fl4->saddr) ||
2688 		    ipv4_is_zeronet(fl4->saddr)) {
2689 			rth = ERR_PTR(-EINVAL);
2690 			goto out;
2691 		}
2692 
2693 		rth = ERR_PTR(-ENETUNREACH);
2694 
2695 		/* I removed check for oif == dev_out->oif here.
2696 		 * It was wrong for two reasons:
2697 		 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2698 		 *    is assigned to multiple interfaces.
2699 		 * 2. Moreover, we are allowed to send packets with saddr
2700 		 *    of another iface. --ANK
2701 		 */
2702 
2703 		if (fl4->flowi4_oif == 0 &&
2704 		    (ipv4_is_multicast(fl4->daddr) ||
2705 		     ipv4_is_lbcast(fl4->daddr))) {
2706 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2707 			dev_out = __ip_dev_find(net, fl4->saddr, false);
2708 			if (!dev_out)
2709 				goto out;
2710 
2711 			/* Special hack: user can direct multicasts
2712 			 * and limited broadcast via necessary interface
2713 			 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2714 			 * This hack is not just for fun, it allows
2715 			 * vic,vat and friends to work.
2716 			 * They bind socket to loopback, set ttl to zero
2717 			 * and expect that it will work.
2718 			 * From the viewpoint of routing cache they are broken,
2719 			 * because we are not allowed to build multicast path
2720 			 * with loopback source addr (look, routing cache
2721 			 * cannot know, that ttl is zero, so that packet
2722 			 * will not leave this host and route is valid).
2723 			 * Luckily, this hack is good workaround.
2724 			 */
2725 
2726 			fl4->flowi4_oif = dev_out->ifindex;
2727 			goto make_route;
2728 		}
2729 
2730 		if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2731 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2732 			if (!__ip_dev_find(net, fl4->saddr, false))
2733 				goto out;
2734 		}
2735 	}
2736 
2737 
2738 	if (fl4->flowi4_oif) {
2739 		dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2740 		rth = ERR_PTR(-ENODEV);
2741 		if (!dev_out)
2742 			goto out;
2743 
2744 		/* RACE: Check return value of inet_select_addr instead. */
2745 		if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2746 			rth = ERR_PTR(-ENETUNREACH);
2747 			goto out;
2748 		}
2749 		if (ipv4_is_local_multicast(fl4->daddr) ||
2750 		    ipv4_is_lbcast(fl4->daddr) ||
2751 		    fl4->flowi4_proto == IPPROTO_IGMP) {
2752 			if (!fl4->saddr)
2753 				fl4->saddr = inet_select_addr(dev_out, 0,
2754 							      RT_SCOPE_LINK);
2755 			goto make_route;
2756 		}
2757 		if (!fl4->saddr) {
2758 			if (ipv4_is_multicast(fl4->daddr))
2759 				fl4->saddr = inet_select_addr(dev_out, 0,
2760 							      fl4->flowi4_scope);
2761 			else if (!fl4->daddr)
2762 				fl4->saddr = inet_select_addr(dev_out, 0,
2763 							      RT_SCOPE_HOST);
2764 		}
2765 	}
2766 
2767 	if (!fl4->daddr) {
2768 		fl4->daddr = fl4->saddr;
2769 		if (!fl4->daddr)
2770 			fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2771 		dev_out = net->loopback_dev;
2772 		fl4->flowi4_oif = LOOPBACK_IFINDEX;
2773 		res->type = RTN_LOCAL;
2774 		flags |= RTCF_LOCAL;
2775 		goto make_route;
2776 	}
2777 
2778 	err = fib_lookup(net, fl4, res, 0);
2779 	if (err) {
2780 		res->fi = NULL;
2781 		res->table = NULL;
2782 		if (fl4->flowi4_oif &&
2783 		    (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2784 			/* Apparently, routing tables are wrong. Assume,
2785 			 * that the destination is on link.
2786 			 *
2787 			 * WHY? DW.
2788 			 * Because we are allowed to send to iface
2789 			 * even if it has NO routes and NO assigned
2790 			 * addresses. When oif is specified, routing
2791 			 * tables are looked up with only one purpose:
2792 			 * to catch if destination is gatewayed, rather than
2793 			 * direct. Moreover, if MSG_DONTROUTE is set,
2794 			 * we send packet, ignoring both routing tables
2795 			 * and ifaddr state. --ANK
2796 			 *
2797 			 *
2798 			 * We could make it even if oif is unknown,
2799 			 * likely IPv6, but we do not.
2800 			 */
2801 
2802 			if (fl4->saddr == 0)
2803 				fl4->saddr = inet_select_addr(dev_out, 0,
2804 							      RT_SCOPE_LINK);
2805 			res->type = RTN_UNICAST;
2806 			goto make_route;
2807 		}
2808 		rth = ERR_PTR(err);
2809 		goto out;
2810 	}
2811 
2812 	if (res->type == RTN_LOCAL) {
2813 		if (!fl4->saddr) {
2814 			if (res->fi->fib_prefsrc)
2815 				fl4->saddr = res->fi->fib_prefsrc;
2816 			else
2817 				fl4->saddr = fl4->daddr;
2818 		}
2819 
2820 		/* L3 master device is the loopback for that domain */
2821 		dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2822 			net->loopback_dev;
2823 
2824 		/* make sure orig_oif points to fib result device even
2825 		 * though packet rx/tx happens over loopback or l3mdev
2826 		 */
2827 		orig_oif = FIB_RES_OIF(*res);
2828 
2829 		fl4->flowi4_oif = dev_out->ifindex;
2830 		flags |= RTCF_LOCAL;
2831 		goto make_route;
2832 	}
2833 
2834 	fib_select_path(net, res, fl4, skb);
2835 
2836 	dev_out = FIB_RES_DEV(*res);
2837 
2838 make_route:
2839 	rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2840 
2841 out:
2842 	return rth;
2843 }
2844 
2845 static struct dst_ops ipv4_dst_blackhole_ops = {
2846 	.family			= AF_INET,
2847 	.default_advmss		= ipv4_default_advmss,
2848 	.neigh_lookup		= ipv4_neigh_lookup,
2849 	.check			= dst_blackhole_check,
2850 	.cow_metrics		= dst_blackhole_cow_metrics,
2851 	.update_pmtu		= dst_blackhole_update_pmtu,
2852 	.redirect		= dst_blackhole_redirect,
2853 	.mtu			= dst_blackhole_mtu,
2854 };
2855 
2856 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2857 {
2858 	struct rtable *ort = dst_rtable(dst_orig);
2859 	struct rtable *rt;
2860 
2861 	rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0);
2862 	if (rt) {
2863 		struct dst_entry *new = &rt->dst;
2864 
2865 		new->__use = 1;
2866 		new->input = dst_discard;
2867 		new->output = dst_discard_out;
2868 
2869 		new->dev = net->loopback_dev;
2870 		netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2871 
2872 		rt->rt_is_input = ort->rt_is_input;
2873 		rt->rt_iif = ort->rt_iif;
2874 		rt->rt_pmtu = ort->rt_pmtu;
2875 		rt->rt_mtu_locked = ort->rt_mtu_locked;
2876 
2877 		rt->rt_genid = rt_genid_ipv4(net);
2878 		rt->rt_flags = ort->rt_flags;
2879 		rt->rt_type = ort->rt_type;
2880 		rt->rt_uses_gateway = ort->rt_uses_gateway;
2881 		rt->rt_gw_family = ort->rt_gw_family;
2882 		if (rt->rt_gw_family == AF_INET)
2883 			rt->rt_gw4 = ort->rt_gw4;
2884 		else if (rt->rt_gw_family == AF_INET6)
2885 			rt->rt_gw6 = ort->rt_gw6;
2886 	}
2887 
2888 	dst_release(dst_orig);
2889 
2890 	return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2891 }
2892 
2893 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2894 				    const struct sock *sk)
2895 {
2896 	struct rtable *rt = __ip_route_output_key(net, flp4);
2897 
2898 	if (IS_ERR(rt))
2899 		return rt;
2900 
2901 	if (flp4->flowi4_proto) {
2902 		flp4->flowi4_oif = rt->dst.dev->ifindex;
2903 		rt = dst_rtable(xfrm_lookup_route(net, &rt->dst,
2904 						  flowi4_to_flowi(flp4),
2905 						  sk, 0));
2906 	}
2907 
2908 	return rt;
2909 }
2910 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2911 
2912 /* called with rcu_read_lock held */
2913 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2914 			struct rtable *rt, u32 table_id, dscp_t dscp,
2915 			struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2916 			u32 seq, unsigned int flags)
2917 {
2918 	struct rtmsg *r;
2919 	struct nlmsghdr *nlh;
2920 	unsigned long expires = 0;
2921 	u32 error;
2922 	u32 metrics[RTAX_MAX];
2923 
2924 	nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2925 	if (!nlh)
2926 		return -EMSGSIZE;
2927 
2928 	r = nlmsg_data(nlh);
2929 	r->rtm_family	 = AF_INET;
2930 	r->rtm_dst_len	= 32;
2931 	r->rtm_src_len	= 0;
2932 	r->rtm_tos	= inet_dscp_to_dsfield(dscp);
2933 	r->rtm_table	= table_id < 256 ? table_id : RT_TABLE_COMPAT;
2934 	if (nla_put_u32(skb, RTA_TABLE, table_id))
2935 		goto nla_put_failure;
2936 	r->rtm_type	= rt->rt_type;
2937 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2938 	r->rtm_protocol = RTPROT_UNSPEC;
2939 	r->rtm_flags	= (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2940 	if (rt->rt_flags & RTCF_NOTIFY)
2941 		r->rtm_flags |= RTM_F_NOTIFY;
2942 	if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2943 		r->rtm_flags |= RTCF_DOREDIRECT;
2944 
2945 	if (nla_put_in_addr(skb, RTA_DST, dst))
2946 		goto nla_put_failure;
2947 	if (src) {
2948 		r->rtm_src_len = 32;
2949 		if (nla_put_in_addr(skb, RTA_SRC, src))
2950 			goto nla_put_failure;
2951 	}
2952 	if (rt->dst.dev &&
2953 	    nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2954 		goto nla_put_failure;
2955 	if (rt->dst.lwtstate &&
2956 	    lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2957 		goto nla_put_failure;
2958 #ifdef CONFIG_IP_ROUTE_CLASSID
2959 	if (rt->dst.tclassid &&
2960 	    nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2961 		goto nla_put_failure;
2962 #endif
2963 	if (fl4 && !rt_is_input_route(rt) &&
2964 	    fl4->saddr != src) {
2965 		if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2966 			goto nla_put_failure;
2967 	}
2968 	if (rt->rt_uses_gateway) {
2969 		if (rt->rt_gw_family == AF_INET &&
2970 		    nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2971 			goto nla_put_failure;
2972 		} else if (rt->rt_gw_family == AF_INET6) {
2973 			int alen = sizeof(struct in6_addr);
2974 			struct nlattr *nla;
2975 			struct rtvia *via;
2976 
2977 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
2978 			if (!nla)
2979 				goto nla_put_failure;
2980 
2981 			via = nla_data(nla);
2982 			via->rtvia_family = AF_INET6;
2983 			memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2984 		}
2985 	}
2986 
2987 	expires = rt->dst.expires;
2988 	if (expires) {
2989 		unsigned long now = jiffies;
2990 
2991 		if (time_before(now, expires))
2992 			expires -= now;
2993 		else
2994 			expires = 0;
2995 	}
2996 
2997 	memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2998 	if (rt->rt_pmtu && expires)
2999 		metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3000 	if (rt->rt_mtu_locked && expires)
3001 		metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3002 	if (rtnetlink_put_metrics(skb, metrics) < 0)
3003 		goto nla_put_failure;
3004 
3005 	if (fl4) {
3006 		if (fl4->flowi4_mark &&
3007 		    nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3008 			goto nla_put_failure;
3009 
3010 		if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3011 		    nla_put_u32(skb, RTA_UID,
3012 				from_kuid_munged(current_user_ns(),
3013 						 fl4->flowi4_uid)))
3014 			goto nla_put_failure;
3015 
3016 		if (rt_is_input_route(rt)) {
3017 #ifdef CONFIG_IP_MROUTE
3018 			if (ipv4_is_multicast(dst) &&
3019 			    !ipv4_is_local_multicast(dst) &&
3020 			    IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) {
3021 				int err = ipmr_get_route(net, skb,
3022 							 fl4->saddr, fl4->daddr,
3023 							 r, portid);
3024 
3025 				if (err <= 0) {
3026 					if (err == 0)
3027 						return 0;
3028 					goto nla_put_failure;
3029 				}
3030 			} else
3031 #endif
3032 				if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3033 					goto nla_put_failure;
3034 		}
3035 	}
3036 
3037 	error = rt->dst.error;
3038 
3039 	if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3040 		goto nla_put_failure;
3041 
3042 	nlmsg_end(skb, nlh);
3043 	return 0;
3044 
3045 nla_put_failure:
3046 	nlmsg_cancel(skb, nlh);
3047 	return -EMSGSIZE;
3048 }
3049 
3050 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3051 			    struct netlink_callback *cb, u32 table_id,
3052 			    struct fnhe_hash_bucket *bucket, int genid,
3053 			    int *fa_index, int fa_start, unsigned int flags)
3054 {
3055 	int i;
3056 
3057 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
3058 		struct fib_nh_exception *fnhe;
3059 
3060 		for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3061 		     fnhe = rcu_dereference(fnhe->fnhe_next)) {
3062 			struct rtable *rt;
3063 			int err;
3064 
3065 			if (*fa_index < fa_start)
3066 				goto next;
3067 
3068 			if (fnhe->fnhe_genid != genid)
3069 				goto next;
3070 
3071 			if (fnhe->fnhe_expires &&
3072 			    time_after(jiffies, fnhe->fnhe_expires))
3073 				goto next;
3074 
3075 			rt = rcu_dereference(fnhe->fnhe_rth_input);
3076 			if (!rt)
3077 				rt = rcu_dereference(fnhe->fnhe_rth_output);
3078 			if (!rt)
3079 				goto next;
3080 
3081 			err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3082 					   table_id, 0, NULL, skb,
3083 					   NETLINK_CB(cb->skb).portid,
3084 					   cb->nlh->nlmsg_seq, flags);
3085 			if (err)
3086 				return err;
3087 next:
3088 			(*fa_index)++;
3089 		}
3090 	}
3091 
3092 	return 0;
3093 }
3094 
3095 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3096 		       u32 table_id, struct fib_info *fi,
3097 		       int *fa_index, int fa_start, unsigned int flags)
3098 {
3099 	struct net *net = sock_net(cb->skb->sk);
3100 	int nhsel, genid = fnhe_genid(net);
3101 
3102 	for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3103 		struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3104 		struct fnhe_hash_bucket *bucket;
3105 		int err;
3106 
3107 		if (nhc->nhc_flags & RTNH_F_DEAD)
3108 			continue;
3109 
3110 		rcu_read_lock();
3111 		bucket = rcu_dereference(nhc->nhc_exceptions);
3112 		err = 0;
3113 		if (bucket)
3114 			err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3115 					       genid, fa_index, fa_start,
3116 					       flags);
3117 		rcu_read_unlock();
3118 		if (err)
3119 			return err;
3120 	}
3121 
3122 	return 0;
3123 }
3124 
3125 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3126 						   u8 ip_proto, __be16 sport,
3127 						   __be16 dport)
3128 {
3129 	struct sk_buff *skb;
3130 	struct iphdr *iph;
3131 
3132 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3133 	if (!skb)
3134 		return NULL;
3135 
3136 	/* Reserve room for dummy headers, this skb can pass
3137 	 * through good chunk of routing engine.
3138 	 */
3139 	skb_reset_mac_header(skb);
3140 	skb_reset_network_header(skb);
3141 	skb->protocol = htons(ETH_P_IP);
3142 	iph = skb_put(skb, sizeof(struct iphdr));
3143 	iph->protocol = ip_proto;
3144 	iph->saddr = src;
3145 	iph->daddr = dst;
3146 	iph->version = 0x4;
3147 	iph->frag_off = 0;
3148 	iph->ihl = 0x5;
3149 	skb_set_transport_header(skb, skb->len);
3150 
3151 	switch (iph->protocol) {
3152 	case IPPROTO_UDP: {
3153 		struct udphdr *udph;
3154 
3155 		udph = skb_put_zero(skb, sizeof(struct udphdr));
3156 		udph->source = sport;
3157 		udph->dest = dport;
3158 		udph->len = htons(sizeof(struct udphdr));
3159 		udph->check = 0;
3160 		break;
3161 	}
3162 	case IPPROTO_TCP: {
3163 		struct tcphdr *tcph;
3164 
3165 		tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3166 		tcph->source	= sport;
3167 		tcph->dest	= dport;
3168 		tcph->doff	= sizeof(struct tcphdr) / 4;
3169 		tcph->rst = 1;
3170 		tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3171 					    src, dst, 0);
3172 		break;
3173 	}
3174 	case IPPROTO_ICMP: {
3175 		struct icmphdr *icmph;
3176 
3177 		icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3178 		icmph->type = ICMP_ECHO;
3179 		icmph->code = 0;
3180 	}
3181 	}
3182 
3183 	return skb;
3184 }
3185 
3186 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3187 				       const struct nlmsghdr *nlh,
3188 				       struct nlattr **tb,
3189 				       struct netlink_ext_ack *extack)
3190 {
3191 	struct rtmsg *rtm;
3192 	int i, err;
3193 
3194 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3195 		NL_SET_ERR_MSG(extack,
3196 			       "ipv4: Invalid header for route get request");
3197 		return -EINVAL;
3198 	}
3199 
3200 	if (!netlink_strict_get_check(skb))
3201 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3202 					      rtm_ipv4_policy, extack);
3203 
3204 	rtm = nlmsg_data(nlh);
3205 	if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3206 	    (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3207 	    rtm->rtm_table || rtm->rtm_protocol ||
3208 	    rtm->rtm_scope || rtm->rtm_type) {
3209 		NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3210 		return -EINVAL;
3211 	}
3212 
3213 	if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3214 			       RTM_F_LOOKUP_TABLE |
3215 			       RTM_F_FIB_MATCH)) {
3216 		NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3217 		return -EINVAL;
3218 	}
3219 
3220 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3221 					    rtm_ipv4_policy, extack);
3222 	if (err)
3223 		return err;
3224 
3225 	if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3226 	    (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3227 		NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3228 		return -EINVAL;
3229 	}
3230 
3231 	for (i = 0; i <= RTA_MAX; i++) {
3232 		if (!tb[i])
3233 			continue;
3234 
3235 		switch (i) {
3236 		case RTA_IIF:
3237 		case RTA_OIF:
3238 		case RTA_SRC:
3239 		case RTA_DST:
3240 		case RTA_IP_PROTO:
3241 		case RTA_SPORT:
3242 		case RTA_DPORT:
3243 		case RTA_MARK:
3244 		case RTA_UID:
3245 			break;
3246 		default:
3247 			NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3248 			return -EINVAL;
3249 		}
3250 	}
3251 
3252 	return 0;
3253 }
3254 
3255 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3256 			     struct netlink_ext_ack *extack)
3257 {
3258 	struct net *net = sock_net(in_skb->sk);
3259 	struct nlattr *tb[RTA_MAX+1];
3260 	u32 table_id = RT_TABLE_MAIN;
3261 	__be16 sport = 0, dport = 0;
3262 	struct fib_result res = {};
3263 	u8 ip_proto = IPPROTO_UDP;
3264 	struct rtable *rt = NULL;
3265 	struct sk_buff *skb;
3266 	struct rtmsg *rtm;
3267 	struct flowi4 fl4 = {};
3268 	__be32 dst = 0;
3269 	__be32 src = 0;
3270 	kuid_t uid;
3271 	u32 iif;
3272 	int err;
3273 	int mark;
3274 
3275 	err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3276 	if (err < 0)
3277 		return err;
3278 
3279 	rtm = nlmsg_data(nlh);
3280 	src = nla_get_in_addr_default(tb[RTA_SRC], 0);
3281 	dst = nla_get_in_addr_default(tb[RTA_DST], 0);
3282 	iif = nla_get_u32_default(tb[RTA_IIF], 0);
3283 	mark = nla_get_u32_default(tb[RTA_MARK], 0);
3284 	if (tb[RTA_UID])
3285 		uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3286 	else
3287 		uid = (iif ? INVALID_UID : current_uid());
3288 
3289 	if (tb[RTA_IP_PROTO]) {
3290 		err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3291 						  &ip_proto, AF_INET, extack);
3292 		if (err)
3293 			return err;
3294 	}
3295 
3296 	if (tb[RTA_SPORT])
3297 		sport = nla_get_be16(tb[RTA_SPORT]);
3298 
3299 	if (tb[RTA_DPORT])
3300 		dport = nla_get_be16(tb[RTA_DPORT]);
3301 
3302 	skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3303 	if (!skb)
3304 		return -ENOBUFS;
3305 
3306 	fl4.daddr = dst;
3307 	fl4.saddr = src;
3308 	fl4.flowi4_tos = rtm->rtm_tos & INET_DSCP_MASK;
3309 	fl4.flowi4_oif = nla_get_u32_default(tb[RTA_OIF], 0);
3310 	fl4.flowi4_mark = mark;
3311 	fl4.flowi4_uid = uid;
3312 	if (sport)
3313 		fl4.fl4_sport = sport;
3314 	if (dport)
3315 		fl4.fl4_dport = dport;
3316 	fl4.flowi4_proto = ip_proto;
3317 
3318 	rcu_read_lock();
3319 
3320 	if (iif) {
3321 		struct net_device *dev;
3322 
3323 		dev = dev_get_by_index_rcu(net, iif);
3324 		if (!dev) {
3325 			err = -ENODEV;
3326 			goto errout_rcu;
3327 		}
3328 
3329 		fl4.flowi4_iif = iif; /* for rt_fill_info */
3330 		skb->dev	= dev;
3331 		skb->mark	= mark;
3332 		err = ip_route_input_rcu(skb, dst, src,
3333 					 inet_dsfield_to_dscp(rtm->rtm_tos),
3334 					 dev, &res) ? -EINVAL : 0;
3335 
3336 		rt = skb_rtable(skb);
3337 		if (err == 0 && rt->dst.error)
3338 			err = -rt->dst.error;
3339 	} else {
3340 		fl4.flowi4_iif = LOOPBACK_IFINDEX;
3341 		skb->dev = net->loopback_dev;
3342 		rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3343 		err = 0;
3344 		if (IS_ERR(rt))
3345 			err = PTR_ERR(rt);
3346 		else
3347 			skb_dst_set(skb, &rt->dst);
3348 	}
3349 
3350 	if (err)
3351 		goto errout_rcu;
3352 
3353 	if (rtm->rtm_flags & RTM_F_NOTIFY)
3354 		rt->rt_flags |= RTCF_NOTIFY;
3355 
3356 	if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3357 		table_id = res.table ? res.table->tb_id : 0;
3358 
3359 	/* reset skb for netlink reply msg */
3360 	skb_trim(skb, 0);
3361 	skb_reset_network_header(skb);
3362 	skb_reset_transport_header(skb);
3363 	skb_reset_mac_header(skb);
3364 
3365 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3366 		struct fib_rt_info fri;
3367 
3368 		if (!res.fi) {
3369 			err = fib_props[res.type].error;
3370 			if (!err)
3371 				err = -EHOSTUNREACH;
3372 			goto errout_rcu;
3373 		}
3374 		fri.fi = res.fi;
3375 		fri.tb_id = table_id;
3376 		fri.dst = res.prefix;
3377 		fri.dst_len = res.prefixlen;
3378 		fri.dscp = res.dscp;
3379 		fri.type = rt->rt_type;
3380 		fri.offload = 0;
3381 		fri.trap = 0;
3382 		fri.offload_failed = 0;
3383 		if (res.fa_head) {
3384 			struct fib_alias *fa;
3385 
3386 			hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3387 				u8 slen = 32 - fri.dst_len;
3388 
3389 				if (fa->fa_slen == slen &&
3390 				    fa->tb_id == fri.tb_id &&
3391 				    fa->fa_dscp == fri.dscp &&
3392 				    fa->fa_info == res.fi &&
3393 				    fa->fa_type == fri.type) {
3394 					fri.offload = READ_ONCE(fa->offload);
3395 					fri.trap = READ_ONCE(fa->trap);
3396 					fri.offload_failed =
3397 						READ_ONCE(fa->offload_failed);
3398 					break;
3399 				}
3400 			}
3401 		}
3402 		err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3403 				    nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3404 	} else {
3405 		err = rt_fill_info(net, dst, src, rt, table_id, res.dscp, &fl4,
3406 				   skb, NETLINK_CB(in_skb).portid,
3407 				   nlh->nlmsg_seq, 0);
3408 	}
3409 	if (err < 0)
3410 		goto errout_rcu;
3411 
3412 	rcu_read_unlock();
3413 
3414 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3415 
3416 errout_free:
3417 	return err;
3418 errout_rcu:
3419 	rcu_read_unlock();
3420 	kfree_skb(skb);
3421 	goto errout_free;
3422 }
3423 
3424 void ip_rt_multicast_event(struct in_device *in_dev)
3425 {
3426 	rt_cache_flush(dev_net(in_dev->dev));
3427 }
3428 
3429 #ifdef CONFIG_SYSCTL
3430 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3431 static int ip_rt_gc_min_interval __read_mostly	= HZ / 2;
3432 static int ip_rt_gc_elasticity __read_mostly	= 8;
3433 static int ip_min_valid_pmtu __read_mostly	= IPV4_MIN_MTU;
3434 
3435 static int ipv4_sysctl_rtcache_flush(const struct ctl_table *__ctl, int write,
3436 		void *buffer, size_t *lenp, loff_t *ppos)
3437 {
3438 	struct net *net = (struct net *)__ctl->extra1;
3439 
3440 	if (write) {
3441 		rt_cache_flush(net);
3442 		fnhe_genid_bump(net);
3443 		return 0;
3444 	}
3445 
3446 	return -EINVAL;
3447 }
3448 
3449 static struct ctl_table ipv4_route_table[] = {
3450 	{
3451 		.procname	= "gc_thresh",
3452 		.data		= &ipv4_dst_ops.gc_thresh,
3453 		.maxlen		= sizeof(int),
3454 		.mode		= 0644,
3455 		.proc_handler	= proc_dointvec,
3456 	},
3457 	{
3458 		.procname	= "max_size",
3459 		.data		= &ip_rt_max_size,
3460 		.maxlen		= sizeof(int),
3461 		.mode		= 0644,
3462 		.proc_handler	= proc_dointvec,
3463 	},
3464 	{
3465 		/*  Deprecated. Use gc_min_interval_ms */
3466 
3467 		.procname	= "gc_min_interval",
3468 		.data		= &ip_rt_gc_min_interval,
3469 		.maxlen		= sizeof(int),
3470 		.mode		= 0644,
3471 		.proc_handler	= proc_dointvec_jiffies,
3472 	},
3473 	{
3474 		.procname	= "gc_min_interval_ms",
3475 		.data		= &ip_rt_gc_min_interval,
3476 		.maxlen		= sizeof(int),
3477 		.mode		= 0644,
3478 		.proc_handler	= proc_dointvec_ms_jiffies,
3479 	},
3480 	{
3481 		.procname	= "gc_timeout",
3482 		.data		= &ip_rt_gc_timeout,
3483 		.maxlen		= sizeof(int),
3484 		.mode		= 0644,
3485 		.proc_handler	= proc_dointvec_jiffies,
3486 	},
3487 	{
3488 		.procname	= "gc_interval",
3489 		.data		= &ip_rt_gc_interval,
3490 		.maxlen		= sizeof(int),
3491 		.mode		= 0644,
3492 		.proc_handler	= proc_dointvec_jiffies,
3493 	},
3494 	{
3495 		.procname	= "redirect_load",
3496 		.data		= &ip_rt_redirect_load,
3497 		.maxlen		= sizeof(int),
3498 		.mode		= 0644,
3499 		.proc_handler	= proc_dointvec,
3500 	},
3501 	{
3502 		.procname	= "redirect_number",
3503 		.data		= &ip_rt_redirect_number,
3504 		.maxlen		= sizeof(int),
3505 		.mode		= 0644,
3506 		.proc_handler	= proc_dointvec,
3507 	},
3508 	{
3509 		.procname	= "redirect_silence",
3510 		.data		= &ip_rt_redirect_silence,
3511 		.maxlen		= sizeof(int),
3512 		.mode		= 0644,
3513 		.proc_handler	= proc_dointvec,
3514 	},
3515 	{
3516 		.procname	= "error_cost",
3517 		.data		= &ip_rt_error_cost,
3518 		.maxlen		= sizeof(int),
3519 		.mode		= 0644,
3520 		.proc_handler	= proc_dointvec,
3521 	},
3522 	{
3523 		.procname	= "error_burst",
3524 		.data		= &ip_rt_error_burst,
3525 		.maxlen		= sizeof(int),
3526 		.mode		= 0644,
3527 		.proc_handler	= proc_dointvec,
3528 	},
3529 	{
3530 		.procname	= "gc_elasticity",
3531 		.data		= &ip_rt_gc_elasticity,
3532 		.maxlen		= sizeof(int),
3533 		.mode		= 0644,
3534 		.proc_handler	= proc_dointvec,
3535 	},
3536 };
3537 
3538 static const char ipv4_route_flush_procname[] = "flush";
3539 
3540 static struct ctl_table ipv4_route_netns_table[] = {
3541 	{
3542 		.procname	= ipv4_route_flush_procname,
3543 		.maxlen		= sizeof(int),
3544 		.mode		= 0200,
3545 		.proc_handler	= ipv4_sysctl_rtcache_flush,
3546 	},
3547 	{
3548 		.procname       = "min_pmtu",
3549 		.data           = &init_net.ipv4.ip_rt_min_pmtu,
3550 		.maxlen         = sizeof(int),
3551 		.mode           = 0644,
3552 		.proc_handler   = proc_dointvec_minmax,
3553 		.extra1         = &ip_min_valid_pmtu,
3554 	},
3555 	{
3556 		.procname       = "mtu_expires",
3557 		.data           = &init_net.ipv4.ip_rt_mtu_expires,
3558 		.maxlen         = sizeof(int),
3559 		.mode           = 0644,
3560 		.proc_handler   = proc_dointvec_jiffies,
3561 	},
3562 	{
3563 		.procname   = "min_adv_mss",
3564 		.data       = &init_net.ipv4.ip_rt_min_advmss,
3565 		.maxlen     = sizeof(int),
3566 		.mode       = 0644,
3567 		.proc_handler   = proc_dointvec,
3568 	},
3569 };
3570 
3571 static __net_init int sysctl_route_net_init(struct net *net)
3572 {
3573 	struct ctl_table *tbl;
3574 	size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3575 
3576 	tbl = ipv4_route_netns_table;
3577 	if (!net_eq(net, &init_net)) {
3578 		int i;
3579 
3580 		tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3581 		if (!tbl)
3582 			goto err_dup;
3583 
3584 		/* Don't export non-whitelisted sysctls to unprivileged users */
3585 		if (net->user_ns != &init_user_ns) {
3586 			if (tbl[0].procname != ipv4_route_flush_procname)
3587 				table_size = 0;
3588 		}
3589 
3590 		/* Update the variables to point into the current struct net
3591 		 * except for the first element flush
3592 		 */
3593 		for (i = 1; i < table_size; i++)
3594 			tbl[i].data += (void *)net - (void *)&init_net;
3595 	}
3596 	tbl[0].extra1 = net;
3597 
3598 	net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3599 						     tbl, table_size);
3600 	if (!net->ipv4.route_hdr)
3601 		goto err_reg;
3602 	return 0;
3603 
3604 err_reg:
3605 	if (tbl != ipv4_route_netns_table)
3606 		kfree(tbl);
3607 err_dup:
3608 	return -ENOMEM;
3609 }
3610 
3611 static __net_exit void sysctl_route_net_exit(struct net *net)
3612 {
3613 	const struct ctl_table *tbl;
3614 
3615 	tbl = net->ipv4.route_hdr->ctl_table_arg;
3616 	unregister_net_sysctl_table(net->ipv4.route_hdr);
3617 	BUG_ON(tbl == ipv4_route_netns_table);
3618 	kfree(tbl);
3619 }
3620 
3621 static __net_initdata struct pernet_operations sysctl_route_ops = {
3622 	.init = sysctl_route_net_init,
3623 	.exit = sysctl_route_net_exit,
3624 };
3625 #endif
3626 
3627 static __net_init int netns_ip_rt_init(struct net *net)
3628 {
3629 	/* Set default value for namespaceified sysctls */
3630 	net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3631 	net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3632 	net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3633 	return 0;
3634 }
3635 
3636 static struct pernet_operations __net_initdata ip_rt_ops = {
3637 	.init = netns_ip_rt_init,
3638 };
3639 
3640 static __net_init int rt_genid_init(struct net *net)
3641 {
3642 	atomic_set(&net->ipv4.rt_genid, 0);
3643 	atomic_set(&net->fnhe_genid, 0);
3644 	atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3645 	return 0;
3646 }
3647 
3648 static __net_initdata struct pernet_operations rt_genid_ops = {
3649 	.init = rt_genid_init,
3650 };
3651 
3652 static int __net_init ipv4_inetpeer_init(struct net *net)
3653 {
3654 	struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3655 
3656 	if (!bp)
3657 		return -ENOMEM;
3658 	inet_peer_base_init(bp);
3659 	net->ipv4.peers = bp;
3660 	return 0;
3661 }
3662 
3663 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3664 {
3665 	struct inet_peer_base *bp = net->ipv4.peers;
3666 
3667 	net->ipv4.peers = NULL;
3668 	inetpeer_invalidate_tree(bp);
3669 	kfree(bp);
3670 }
3671 
3672 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3673 	.init	=	ipv4_inetpeer_init,
3674 	.exit	=	ipv4_inetpeer_exit,
3675 };
3676 
3677 #ifdef CONFIG_IP_ROUTE_CLASSID
3678 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3679 #endif /* CONFIG_IP_ROUTE_CLASSID */
3680 
3681 static const struct rtnl_msg_handler ip_rt_rtnl_msg_handlers[] __initconst = {
3682 	{.protocol = PF_INET, .msgtype = RTM_GETROUTE,
3683 	 .doit = inet_rtm_getroute, .flags = RTNL_FLAG_DOIT_UNLOCKED},
3684 };
3685 
3686 int __init ip_rt_init(void)
3687 {
3688 	void *idents_hash;
3689 	int cpu;
3690 
3691 	/* For modern hosts, this will use 2 MB of memory */
3692 	idents_hash = alloc_large_system_hash("IP idents",
3693 					      sizeof(*ip_idents) + sizeof(*ip_tstamps),
3694 					      0,
3695 					      16, /* one bucket per 64 KB */
3696 					      HASH_ZERO,
3697 					      NULL,
3698 					      &ip_idents_mask,
3699 					      2048,
3700 					      256*1024);
3701 
3702 	ip_idents = idents_hash;
3703 
3704 	get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3705 
3706 	ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3707 
3708 	for_each_possible_cpu(cpu) {
3709 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3710 
3711 		INIT_LIST_HEAD(&ul->head);
3712 		spin_lock_init(&ul->lock);
3713 	}
3714 #ifdef CONFIG_IP_ROUTE_CLASSID
3715 	ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3716 	if (!ip_rt_acct)
3717 		panic("IP: failed to allocate ip_rt_acct\n");
3718 #endif
3719 
3720 	ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable,
3721 					      SLAB_HWCACHE_ALIGN | SLAB_PANIC);
3722 
3723 	ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3724 
3725 	if (dst_entries_init(&ipv4_dst_ops) < 0)
3726 		panic("IP: failed to allocate ipv4_dst_ops counter\n");
3727 
3728 	if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3729 		panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3730 
3731 	ipv4_dst_ops.gc_thresh = ~0;
3732 	ip_rt_max_size = INT_MAX;
3733 
3734 	devinet_init();
3735 	ip_fib_init();
3736 
3737 	if (ip_rt_proc_init())
3738 		pr_err("Unable to create route proc files\n");
3739 #ifdef CONFIG_XFRM
3740 	xfrm_init();
3741 	xfrm4_init();
3742 #endif
3743 	rtnl_register_many(ip_rt_rtnl_msg_handlers);
3744 
3745 #ifdef CONFIG_SYSCTL
3746 	register_pernet_subsys(&sysctl_route_ops);
3747 #endif
3748 	register_pernet_subsys(&ip_rt_ops);
3749 	register_pernet_subsys(&rt_genid_ops);
3750 	register_pernet_subsys(&ipv4_inetpeer_ops);
3751 	return 0;
3752 }
3753 
3754 #ifdef CONFIG_SYSCTL
3755 /*
3756  * We really need to sanitize the damn ipv4 init order, then all
3757  * this nonsense will go away.
3758  */
3759 void __init ip_static_sysctl_init(void)
3760 {
3761 	register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3762 }
3763 #endif
3764