xref: /linux/net/ipv4/route.c (revision fe259a1bb26ec78842c975d992331705b0c2c2e8)
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 	bool res;
394 
395 	rcu_read_lock();
396 	res = rth->rt_genid != rt_genid_ipv4(dev_net_rcu(rth->dst.dev));
397 	rcu_read_unlock();
398 
399 	return res;
400 }
401 
402 void rt_cache_flush(struct net *net)
403 {
404 	rt_genid_bump_ipv4(net);
405 }
406 
407 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
408 					   struct sk_buff *skb,
409 					   const void *daddr)
410 {
411 	const struct rtable *rt = container_of(dst, struct rtable, dst);
412 	struct net_device *dev = dst->dev;
413 	struct neighbour *n;
414 
415 	rcu_read_lock();
416 
417 	if (likely(rt->rt_gw_family == AF_INET)) {
418 		n = ip_neigh_gw4(dev, rt->rt_gw4);
419 	} else if (rt->rt_gw_family == AF_INET6) {
420 		n = ip_neigh_gw6(dev, &rt->rt_gw6);
421         } else {
422 		__be32 pkey;
423 
424 		pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
425 		n = ip_neigh_gw4(dev, pkey);
426 	}
427 
428 	if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
429 		n = NULL;
430 
431 	rcu_read_unlock();
432 
433 	return n;
434 }
435 
436 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
437 {
438 	const struct rtable *rt = container_of(dst, struct rtable, dst);
439 	struct net_device *dev = dst->dev;
440 	const __be32 *pkey = daddr;
441 
442 	if (rt->rt_gw_family == AF_INET) {
443 		pkey = (const __be32 *)&rt->rt_gw4;
444 	} else if (rt->rt_gw_family == AF_INET6) {
445 		return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
446 	} else if (!daddr ||
447 		 (rt->rt_flags &
448 		  (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
449 		return;
450 	}
451 	__ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
452 }
453 
454 /* Hash tables of size 2048..262144 depending on RAM size.
455  * Each bucket uses 8 bytes.
456  */
457 static u32 ip_idents_mask __read_mostly;
458 static atomic_t *ip_idents __read_mostly;
459 static u32 *ip_tstamps __read_mostly;
460 
461 /* In order to protect privacy, we add a perturbation to identifiers
462  * if one generator is seldom used. This makes hard for an attacker
463  * to infer how many packets were sent between two points in time.
464  */
465 static u32 ip_idents_reserve(u32 hash, int segs)
466 {
467 	u32 bucket, old, now = (u32)jiffies;
468 	atomic_t *p_id;
469 	u32 *p_tstamp;
470 	u32 delta = 0;
471 
472 	bucket = hash & ip_idents_mask;
473 	p_tstamp = ip_tstamps + bucket;
474 	p_id = ip_idents + bucket;
475 	old = READ_ONCE(*p_tstamp);
476 
477 	if (old != now && cmpxchg(p_tstamp, old, now) == old)
478 		delta = get_random_u32_below(now - old);
479 
480 	/* If UBSAN reports an error there, please make sure your compiler
481 	 * supports -fno-strict-overflow before reporting it that was a bug
482 	 * in UBSAN, and it has been fixed in GCC-8.
483 	 */
484 	return atomic_add_return(segs + delta, p_id) - segs;
485 }
486 
487 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
488 {
489 	u32 hash, id;
490 
491 	/* Note the following code is not safe, but this is okay. */
492 	if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
493 		get_random_bytes(&net->ipv4.ip_id_key,
494 				 sizeof(net->ipv4.ip_id_key));
495 
496 	hash = siphash_3u32((__force u32)iph->daddr,
497 			    (__force u32)iph->saddr,
498 			    iph->protocol,
499 			    &net->ipv4.ip_id_key);
500 	id = ip_idents_reserve(hash, segs);
501 	iph->id = htons(id);
502 }
503 EXPORT_SYMBOL(__ip_select_ident);
504 
505 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
506 			     const struct sock *sk, const struct iphdr *iph,
507 			     int oif, __u8 tos, u8 prot, u32 mark,
508 			     int flow_flags)
509 {
510 	__u8 scope = RT_SCOPE_UNIVERSE;
511 
512 	if (sk) {
513 		oif = sk->sk_bound_dev_if;
514 		mark = READ_ONCE(sk->sk_mark);
515 		tos = ip_sock_rt_tos(sk);
516 		scope = ip_sock_rt_scope(sk);
517 		prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW :
518 						    sk->sk_protocol;
519 	}
520 
521 	flowi4_init_output(fl4, oif, mark, tos & INET_DSCP_MASK, scope,
522 			   prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
523 			   sock_net_uid(net, sk));
524 }
525 
526 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
527 			       const struct sock *sk)
528 {
529 	const struct net *net = dev_net(skb->dev);
530 	const struct iphdr *iph = ip_hdr(skb);
531 	int oif = skb->dev->ifindex;
532 	u8 prot = iph->protocol;
533 	u32 mark = skb->mark;
534 	__u8 tos = iph->tos;
535 
536 	__build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
537 }
538 
539 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
540 {
541 	const struct inet_sock *inet = inet_sk(sk);
542 	const struct ip_options_rcu *inet_opt;
543 	__be32 daddr = inet->inet_daddr;
544 
545 	rcu_read_lock();
546 	inet_opt = rcu_dereference(inet->inet_opt);
547 	if (inet_opt && inet_opt->opt.srr)
548 		daddr = inet_opt->opt.faddr;
549 	flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark),
550 			   ip_sock_rt_tos(sk),
551 			   ip_sock_rt_scope(sk),
552 			   inet_test_bit(HDRINCL, sk) ?
553 				IPPROTO_RAW : sk->sk_protocol,
554 			   inet_sk_flowi_flags(sk),
555 			   daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
556 	rcu_read_unlock();
557 }
558 
559 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
560 				 const struct sk_buff *skb)
561 {
562 	if (skb)
563 		build_skb_flow_key(fl4, skb, sk);
564 	else
565 		build_sk_flow_key(fl4, sk);
566 }
567 
568 static DEFINE_SPINLOCK(fnhe_lock);
569 
570 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
571 {
572 	struct rtable *rt;
573 
574 	rt = rcu_dereference(fnhe->fnhe_rth_input);
575 	if (rt) {
576 		RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
577 		dst_dev_put(&rt->dst);
578 		dst_release(&rt->dst);
579 	}
580 	rt = rcu_dereference(fnhe->fnhe_rth_output);
581 	if (rt) {
582 		RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
583 		dst_dev_put(&rt->dst);
584 		dst_release(&rt->dst);
585 	}
586 }
587 
588 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
589 {
590 	struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
591 	struct fib_nh_exception *fnhe, *oldest = NULL;
592 
593 	for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
594 		fnhe = rcu_dereference_protected(*fnhe_p,
595 						 lockdep_is_held(&fnhe_lock));
596 		if (!fnhe)
597 			break;
598 		if (!oldest ||
599 		    time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
600 			oldest = fnhe;
601 			oldest_p = fnhe_p;
602 		}
603 	}
604 	fnhe_flush_routes(oldest);
605 	*oldest_p = oldest->fnhe_next;
606 	kfree_rcu(oldest, rcu);
607 }
608 
609 static u32 fnhe_hashfun(__be32 daddr)
610 {
611 	static siphash_aligned_key_t fnhe_hash_key;
612 	u64 hval;
613 
614 	net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
615 	hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
616 	return hash_64(hval, FNHE_HASH_SHIFT);
617 }
618 
619 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
620 {
621 	rt->rt_pmtu = fnhe->fnhe_pmtu;
622 	rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
623 	rt->dst.expires = fnhe->fnhe_expires;
624 
625 	if (fnhe->fnhe_gw) {
626 		rt->rt_flags |= RTCF_REDIRECTED;
627 		rt->rt_uses_gateway = 1;
628 		rt->rt_gw_family = AF_INET;
629 		rt->rt_gw4 = fnhe->fnhe_gw;
630 	}
631 }
632 
633 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
634 				  __be32 gw, u32 pmtu, bool lock,
635 				  unsigned long expires)
636 {
637 	struct fnhe_hash_bucket *hash;
638 	struct fib_nh_exception *fnhe;
639 	struct rtable *rt;
640 	u32 genid, hval;
641 	unsigned int i;
642 	int depth;
643 
644 	genid = fnhe_genid(dev_net(nhc->nhc_dev));
645 	hval = fnhe_hashfun(daddr);
646 
647 	spin_lock_bh(&fnhe_lock);
648 
649 	hash = rcu_dereference(nhc->nhc_exceptions);
650 	if (!hash) {
651 		hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
652 		if (!hash)
653 			goto out_unlock;
654 		rcu_assign_pointer(nhc->nhc_exceptions, hash);
655 	}
656 
657 	hash += hval;
658 
659 	depth = 0;
660 	for (fnhe = rcu_dereference(hash->chain); fnhe;
661 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
662 		if (fnhe->fnhe_daddr == daddr)
663 			break;
664 		depth++;
665 	}
666 
667 	if (fnhe) {
668 		if (fnhe->fnhe_genid != genid)
669 			fnhe->fnhe_genid = genid;
670 		if (gw)
671 			fnhe->fnhe_gw = gw;
672 		if (pmtu) {
673 			fnhe->fnhe_pmtu = pmtu;
674 			fnhe->fnhe_mtu_locked = lock;
675 		}
676 		fnhe->fnhe_expires = max(1UL, expires);
677 		/* Update all cached dsts too */
678 		rt = rcu_dereference(fnhe->fnhe_rth_input);
679 		if (rt)
680 			fill_route_from_fnhe(rt, fnhe);
681 		rt = rcu_dereference(fnhe->fnhe_rth_output);
682 		if (rt)
683 			fill_route_from_fnhe(rt, fnhe);
684 	} else {
685 		/* Randomize max depth to avoid some side channels attacks. */
686 		int max_depth = FNHE_RECLAIM_DEPTH +
687 				get_random_u32_below(FNHE_RECLAIM_DEPTH);
688 
689 		while (depth > max_depth) {
690 			fnhe_remove_oldest(hash);
691 			depth--;
692 		}
693 
694 		fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
695 		if (!fnhe)
696 			goto out_unlock;
697 
698 		fnhe->fnhe_next = hash->chain;
699 
700 		fnhe->fnhe_genid = genid;
701 		fnhe->fnhe_daddr = daddr;
702 		fnhe->fnhe_gw = gw;
703 		fnhe->fnhe_pmtu = pmtu;
704 		fnhe->fnhe_mtu_locked = lock;
705 		fnhe->fnhe_expires = max(1UL, expires);
706 
707 		rcu_assign_pointer(hash->chain, fnhe);
708 
709 		/* Exception created; mark the cached routes for the nexthop
710 		 * stale, so anyone caching it rechecks if this exception
711 		 * applies to them.
712 		 */
713 		rt = rcu_dereference(nhc->nhc_rth_input);
714 		if (rt)
715 			rt->dst.obsolete = DST_OBSOLETE_KILL;
716 
717 		for_each_possible_cpu(i) {
718 			struct rtable __rcu **prt;
719 
720 			prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
721 			rt = rcu_dereference(*prt);
722 			if (rt)
723 				rt->dst.obsolete = DST_OBSOLETE_KILL;
724 		}
725 	}
726 
727 	fnhe->fnhe_stamp = jiffies;
728 
729 out_unlock:
730 	spin_unlock_bh(&fnhe_lock);
731 }
732 
733 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
734 			     bool kill_route)
735 {
736 	__be32 new_gw = icmp_hdr(skb)->un.gateway;
737 	__be32 old_gw = ip_hdr(skb)->saddr;
738 	struct net_device *dev = skb->dev;
739 	struct in_device *in_dev;
740 	struct fib_result res;
741 	struct neighbour *n;
742 	struct net *net;
743 
744 	switch (icmp_hdr(skb)->code & 7) {
745 	case ICMP_REDIR_NET:
746 	case ICMP_REDIR_NETTOS:
747 	case ICMP_REDIR_HOST:
748 	case ICMP_REDIR_HOSTTOS:
749 		break;
750 
751 	default:
752 		return;
753 	}
754 
755 	if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
756 		return;
757 
758 	in_dev = __in_dev_get_rcu(dev);
759 	if (!in_dev)
760 		return;
761 
762 	net = dev_net(dev);
763 	if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
764 	    ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
765 	    ipv4_is_zeronet(new_gw))
766 		goto reject_redirect;
767 
768 	if (!IN_DEV_SHARED_MEDIA(in_dev)) {
769 		if (!inet_addr_onlink(in_dev, new_gw, old_gw))
770 			goto reject_redirect;
771 		if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
772 			goto reject_redirect;
773 	} else {
774 		if (inet_addr_type(net, new_gw) != RTN_UNICAST)
775 			goto reject_redirect;
776 	}
777 
778 	n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw);
779 	if (!n)
780 		n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
781 	if (!IS_ERR(n)) {
782 		if (!(READ_ONCE(n->nud_state) & NUD_VALID)) {
783 			neigh_event_send(n, NULL);
784 		} else {
785 			if (fib_lookup(net, fl4, &res, 0) == 0) {
786 				struct fib_nh_common *nhc;
787 
788 				fib_select_path(net, &res, fl4, skb);
789 				nhc = FIB_RES_NHC(res);
790 				update_or_create_fnhe(nhc, fl4->daddr, new_gw,
791 						0, false,
792 						jiffies + ip_rt_gc_timeout);
793 			}
794 			if (kill_route)
795 				rt->dst.obsolete = DST_OBSOLETE_KILL;
796 			call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
797 		}
798 		neigh_release(n);
799 	}
800 	return;
801 
802 reject_redirect:
803 #ifdef CONFIG_IP_ROUTE_VERBOSE
804 	if (IN_DEV_LOG_MARTIANS(in_dev)) {
805 		const struct iphdr *iph = (const struct iphdr *) skb->data;
806 		__be32 daddr = iph->daddr;
807 		__be32 saddr = iph->saddr;
808 
809 		net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
810 				     "  Advised path = %pI4 -> %pI4\n",
811 				     &old_gw, dev->name, &new_gw,
812 				     &saddr, &daddr);
813 	}
814 #endif
815 	;
816 }
817 
818 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
819 {
820 	struct rtable *rt;
821 	struct flowi4 fl4;
822 	const struct iphdr *iph = (const struct iphdr *) skb->data;
823 	struct net *net = dev_net(skb->dev);
824 	int oif = skb->dev->ifindex;
825 	u8 prot = iph->protocol;
826 	u32 mark = skb->mark;
827 	__u8 tos = iph->tos;
828 
829 	rt = dst_rtable(dst);
830 
831 	__build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
832 	__ip_do_redirect(rt, skb, &fl4, true);
833 }
834 
835 static void ipv4_negative_advice(struct sock *sk,
836 				 struct dst_entry *dst)
837 {
838 	struct rtable *rt = dst_rtable(dst);
839 
840 	if ((dst->obsolete > 0) ||
841 	    (rt->rt_flags & RTCF_REDIRECTED) ||
842 	    rt->dst.expires)
843 		sk_dst_reset(sk);
844 }
845 
846 /*
847  * Algorithm:
848  *	1. The first ip_rt_redirect_number redirects are sent
849  *	   with exponential backoff, then we stop sending them at all,
850  *	   assuming that the host ignores our redirects.
851  *	2. If we did not see packets requiring redirects
852  *	   during ip_rt_redirect_silence, we assume that the host
853  *	   forgot redirected route and start to send redirects again.
854  *
855  * This algorithm is much cheaper and more intelligent than dumb load limiting
856  * in icmp.c.
857  *
858  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
859  * and "frag. need" (breaks PMTU discovery) in icmp.c.
860  */
861 
862 void ip_rt_send_redirect(struct sk_buff *skb)
863 {
864 	struct rtable *rt = skb_rtable(skb);
865 	struct in_device *in_dev;
866 	struct inet_peer *peer;
867 	struct net *net;
868 	int log_martians;
869 	int vif;
870 
871 	rcu_read_lock();
872 	in_dev = __in_dev_get_rcu(rt->dst.dev);
873 	if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
874 		rcu_read_unlock();
875 		return;
876 	}
877 	log_martians = IN_DEV_LOG_MARTIANS(in_dev);
878 	vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
879 
880 	net = dev_net(rt->dst.dev);
881 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif);
882 	if (!peer) {
883 		rcu_read_unlock();
884 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
885 			  rt_nexthop(rt, ip_hdr(skb)->daddr));
886 		return;
887 	}
888 
889 	/* No redirected packets during ip_rt_redirect_silence;
890 	 * reset the algorithm.
891 	 */
892 	if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
893 		peer->rate_tokens = 0;
894 		peer->n_redirects = 0;
895 	}
896 
897 	/* Too many ignored redirects; do not send anything
898 	 * set dst.rate_last to the last seen redirected packet.
899 	 */
900 	if (peer->n_redirects >= ip_rt_redirect_number) {
901 		peer->rate_last = jiffies;
902 		goto out_unlock;
903 	}
904 
905 	/* Check for load limit; set rate_last to the latest sent
906 	 * redirect.
907 	 */
908 	if (peer->n_redirects == 0 ||
909 	    time_after(jiffies,
910 		       (peer->rate_last +
911 			(ip_rt_redirect_load << peer->n_redirects)))) {
912 		__be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
913 
914 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
915 		peer->rate_last = jiffies;
916 		++peer->n_redirects;
917 		if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians &&
918 		    peer->n_redirects == ip_rt_redirect_number)
919 			net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
920 					     &ip_hdr(skb)->saddr, inet_iif(skb),
921 					     &ip_hdr(skb)->daddr, &gw);
922 	}
923 out_unlock:
924 	rcu_read_unlock();
925 }
926 
927 static int ip_error(struct sk_buff *skb)
928 {
929 	struct rtable *rt = skb_rtable(skb);
930 	struct net_device *dev = skb->dev;
931 	struct in_device *in_dev;
932 	struct inet_peer *peer;
933 	unsigned long now;
934 	struct net *net;
935 	SKB_DR(reason);
936 	bool send;
937 	int code;
938 
939 	if (netif_is_l3_master(skb->dev)) {
940 		dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
941 		if (!dev)
942 			goto out;
943 	}
944 
945 	in_dev = __in_dev_get_rcu(dev);
946 
947 	/* IP on this device is disabled. */
948 	if (!in_dev)
949 		goto out;
950 
951 	net = dev_net(rt->dst.dev);
952 	if (!IN_DEV_FORWARD(in_dev)) {
953 		switch (rt->dst.error) {
954 		case EHOSTUNREACH:
955 			SKB_DR_SET(reason, IP_INADDRERRORS);
956 			__IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
957 			break;
958 
959 		case ENETUNREACH:
960 			SKB_DR_SET(reason, IP_INNOROUTES);
961 			__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
962 			break;
963 		}
964 		goto out;
965 	}
966 
967 	switch (rt->dst.error) {
968 	case EINVAL:
969 	default:
970 		goto out;
971 	case EHOSTUNREACH:
972 		code = ICMP_HOST_UNREACH;
973 		break;
974 	case ENETUNREACH:
975 		code = ICMP_NET_UNREACH;
976 		SKB_DR_SET(reason, IP_INNOROUTES);
977 		__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
978 		break;
979 	case EACCES:
980 		code = ICMP_PKT_FILTERED;
981 		break;
982 	}
983 
984 	rcu_read_lock();
985 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
986 			       l3mdev_master_ifindex_rcu(skb->dev));
987 	send = true;
988 	if (peer) {
989 		now = jiffies;
990 		peer->rate_tokens += now - peer->rate_last;
991 		if (peer->rate_tokens > ip_rt_error_burst)
992 			peer->rate_tokens = ip_rt_error_burst;
993 		peer->rate_last = now;
994 		if (peer->rate_tokens >= ip_rt_error_cost)
995 			peer->rate_tokens -= ip_rt_error_cost;
996 		else
997 			send = false;
998 	}
999 	rcu_read_unlock();
1000 
1001 	if (send)
1002 		icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1003 
1004 out:	kfree_skb_reason(skb, reason);
1005 	return 0;
1006 }
1007 
1008 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1009 {
1010 	struct dst_entry *dst = &rt->dst;
1011 	struct fib_result res;
1012 	bool lock = false;
1013 	struct net *net;
1014 	u32 old_mtu;
1015 
1016 	if (ip_mtu_locked(dst))
1017 		return;
1018 
1019 	old_mtu = ipv4_mtu(dst);
1020 	if (old_mtu < mtu)
1021 		return;
1022 
1023 	rcu_read_lock();
1024 	net = dev_net_rcu(dst->dev);
1025 	if (mtu < net->ipv4.ip_rt_min_pmtu) {
1026 		lock = true;
1027 		mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1028 	}
1029 
1030 	if (rt->rt_pmtu == mtu && !lock &&
1031 	    time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2))
1032 		goto out;
1033 
1034 	if (fib_lookup(net, fl4, &res, 0) == 0) {
1035 		struct fib_nh_common *nhc;
1036 
1037 		fib_select_path(net, &res, fl4, NULL);
1038 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1039 		if (fib_info_num_path(res.fi) > 1) {
1040 			int nhsel;
1041 
1042 			for (nhsel = 0; nhsel < fib_info_num_path(res.fi); nhsel++) {
1043 				nhc = fib_info_nhc(res.fi, nhsel);
1044 				update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1045 						      jiffies + net->ipv4.ip_rt_mtu_expires);
1046 			}
1047 			goto out;
1048 		}
1049 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1050 		nhc = FIB_RES_NHC(res);
1051 		update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1052 				      jiffies + net->ipv4.ip_rt_mtu_expires);
1053 	}
1054 out:
1055 	rcu_read_unlock();
1056 }
1057 
1058 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1059 			      struct sk_buff *skb, u32 mtu,
1060 			      bool confirm_neigh)
1061 {
1062 	struct rtable *rt = dst_rtable(dst);
1063 	struct flowi4 fl4;
1064 
1065 	ip_rt_build_flow_key(&fl4, sk, skb);
1066 
1067 	/* Don't make lookup fail for bridged encapsulations */
1068 	if (skb && netif_is_any_bridge_port(skb->dev))
1069 		fl4.flowi4_oif = 0;
1070 
1071 	__ip_rt_update_pmtu(rt, &fl4, mtu);
1072 }
1073 
1074 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1075 		      int oif, u8 protocol)
1076 {
1077 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1078 	struct flowi4 fl4;
1079 	struct rtable *rt;
1080 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
1081 
1082 	__build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1083 			 0);
1084 	rt = __ip_route_output_key(net, &fl4);
1085 	if (!IS_ERR(rt)) {
1086 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1087 		ip_rt_put(rt);
1088 	}
1089 }
1090 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1091 
1092 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1093 {
1094 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1095 	struct flowi4 fl4;
1096 	struct rtable *rt;
1097 
1098 	__build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1099 
1100 	if (!fl4.flowi4_mark)
1101 		fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1102 
1103 	rt = __ip_route_output_key(sock_net(sk), &fl4);
1104 	if (!IS_ERR(rt)) {
1105 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1106 		ip_rt_put(rt);
1107 	}
1108 }
1109 
1110 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1111 {
1112 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1113 	struct flowi4 fl4;
1114 	struct rtable *rt;
1115 	struct dst_entry *odst = NULL;
1116 	bool new = false;
1117 	struct net *net = sock_net(sk);
1118 
1119 	bh_lock_sock(sk);
1120 
1121 	if (!ip_sk_accept_pmtu(sk))
1122 		goto out;
1123 
1124 	odst = sk_dst_get(sk);
1125 
1126 	if (sock_owned_by_user(sk) || !odst) {
1127 		__ipv4_sk_update_pmtu(skb, sk, mtu);
1128 		goto out;
1129 	}
1130 
1131 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1132 
1133 	rt = dst_rtable(odst);
1134 	if (odst->obsolete && !odst->ops->check(odst, 0)) {
1135 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1136 		if (IS_ERR(rt))
1137 			goto out;
1138 
1139 		new = true;
1140 	}
1141 
1142 	__ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu);
1143 
1144 	if (!dst_check(&rt->dst, 0)) {
1145 		if (new)
1146 			dst_release(&rt->dst);
1147 
1148 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1149 		if (IS_ERR(rt))
1150 			goto out;
1151 
1152 		new = true;
1153 	}
1154 
1155 	if (new)
1156 		sk_dst_set(sk, &rt->dst);
1157 
1158 out:
1159 	bh_unlock_sock(sk);
1160 	dst_release(odst);
1161 }
1162 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1163 
1164 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1165 		   int oif, u8 protocol)
1166 {
1167 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1168 	struct flowi4 fl4;
1169 	struct rtable *rt;
1170 
1171 	__build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1172 	rt = __ip_route_output_key(net, &fl4);
1173 	if (!IS_ERR(rt)) {
1174 		__ip_do_redirect(rt, skb, &fl4, false);
1175 		ip_rt_put(rt);
1176 	}
1177 }
1178 EXPORT_SYMBOL_GPL(ipv4_redirect);
1179 
1180 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1181 {
1182 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1183 	struct flowi4 fl4;
1184 	struct rtable *rt;
1185 	struct net *net = sock_net(sk);
1186 
1187 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1188 	rt = __ip_route_output_key(net, &fl4);
1189 	if (!IS_ERR(rt)) {
1190 		__ip_do_redirect(rt, skb, &fl4, false);
1191 		ip_rt_put(rt);
1192 	}
1193 }
1194 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1195 
1196 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1197 							 u32 cookie)
1198 {
1199 	struct rtable *rt = dst_rtable(dst);
1200 
1201 	/* All IPV4 dsts are created with ->obsolete set to the value
1202 	 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1203 	 * into this function always.
1204 	 *
1205 	 * When a PMTU/redirect information update invalidates a route,
1206 	 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1207 	 * DST_OBSOLETE_DEAD.
1208 	 */
1209 	if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1210 		return NULL;
1211 	return dst;
1212 }
1213 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1214 
1215 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1216 {
1217 	struct net_device *dev;
1218 	struct ip_options opt;
1219 	int res;
1220 
1221 	/* Recompile ip options since IPCB may not be valid anymore.
1222 	 * Also check we have a reasonable ipv4 header.
1223 	 */
1224 	if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1225 	    ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1226 		return;
1227 
1228 	memset(&opt, 0, sizeof(opt));
1229 	if (ip_hdr(skb)->ihl > 5) {
1230 		if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1231 			return;
1232 		opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1233 
1234 		rcu_read_lock();
1235 		dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1236 		res = __ip_options_compile(dev_net(dev), &opt, skb, NULL);
1237 		rcu_read_unlock();
1238 
1239 		if (res)
1240 			return;
1241 	}
1242 	__icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1243 }
1244 
1245 static void ipv4_link_failure(struct sk_buff *skb)
1246 {
1247 	struct rtable *rt;
1248 
1249 	ipv4_send_dest_unreach(skb);
1250 
1251 	rt = skb_rtable(skb);
1252 	if (rt)
1253 		dst_set_expires(&rt->dst, 0);
1254 }
1255 
1256 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1257 {
1258 	pr_debug("%s: %pI4 -> %pI4, %s\n",
1259 		 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1260 		 skb->dev ? skb->dev->name : "?");
1261 	kfree_skb(skb);
1262 	WARN_ON(1);
1263 	return 0;
1264 }
1265 
1266 /*
1267  * We do not cache source address of outgoing interface,
1268  * because it is used only by IP RR, TS and SRR options,
1269  * so that it out of fast path.
1270  *
1271  * BTW remember: "addr" is allowed to be not aligned
1272  * in IP options!
1273  */
1274 
1275 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1276 {
1277 	__be32 src;
1278 
1279 	if (rt_is_output_route(rt))
1280 		src = ip_hdr(skb)->saddr;
1281 	else {
1282 		struct fib_result res;
1283 		struct iphdr *iph = ip_hdr(skb);
1284 		struct flowi4 fl4 = {
1285 			.daddr = iph->daddr,
1286 			.saddr = iph->saddr,
1287 			.flowi4_tos = inet_dscp_to_dsfield(ip4h_dscp(iph)),
1288 			.flowi4_oif = rt->dst.dev->ifindex,
1289 			.flowi4_iif = skb->dev->ifindex,
1290 			.flowi4_mark = skb->mark,
1291 		};
1292 
1293 		rcu_read_lock();
1294 		if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1295 			src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1296 		else
1297 			src = inet_select_addr(rt->dst.dev,
1298 					       rt_nexthop(rt, iph->daddr),
1299 					       RT_SCOPE_UNIVERSE);
1300 		rcu_read_unlock();
1301 	}
1302 	memcpy(addr, &src, 4);
1303 }
1304 
1305 #ifdef CONFIG_IP_ROUTE_CLASSID
1306 static void set_class_tag(struct rtable *rt, u32 tag)
1307 {
1308 	if (!(rt->dst.tclassid & 0xFFFF))
1309 		rt->dst.tclassid |= tag & 0xFFFF;
1310 	if (!(rt->dst.tclassid & 0xFFFF0000))
1311 		rt->dst.tclassid |= tag & 0xFFFF0000;
1312 }
1313 #endif
1314 
1315 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1316 {
1317 	unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1318 	unsigned int advmss;
1319 	struct net *net;
1320 
1321 	rcu_read_lock();
1322 	net = dev_net_rcu(dst->dev);
1323 	advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1324 				   net->ipv4.ip_rt_min_advmss);
1325 	rcu_read_unlock();
1326 
1327 	return min(advmss, IPV4_MAX_PMTU - header_size);
1328 }
1329 
1330 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1331 {
1332 	return ip_dst_mtu_maybe_forward(dst, false);
1333 }
1334 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1335 
1336 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1337 {
1338 	struct fnhe_hash_bucket *hash;
1339 	struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1340 	u32 hval = fnhe_hashfun(daddr);
1341 
1342 	spin_lock_bh(&fnhe_lock);
1343 
1344 	hash = rcu_dereference_protected(nhc->nhc_exceptions,
1345 					 lockdep_is_held(&fnhe_lock));
1346 	hash += hval;
1347 
1348 	fnhe_p = &hash->chain;
1349 	fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1350 	while (fnhe) {
1351 		if (fnhe->fnhe_daddr == daddr) {
1352 			rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1353 				fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1354 			/* set fnhe_daddr to 0 to ensure it won't bind with
1355 			 * new dsts in rt_bind_exception().
1356 			 */
1357 			fnhe->fnhe_daddr = 0;
1358 			fnhe_flush_routes(fnhe);
1359 			kfree_rcu(fnhe, rcu);
1360 			break;
1361 		}
1362 		fnhe_p = &fnhe->fnhe_next;
1363 		fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1364 						 lockdep_is_held(&fnhe_lock));
1365 	}
1366 
1367 	spin_unlock_bh(&fnhe_lock);
1368 }
1369 
1370 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1371 					       __be32 daddr)
1372 {
1373 	struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1374 	struct fib_nh_exception *fnhe;
1375 	u32 hval;
1376 
1377 	if (!hash)
1378 		return NULL;
1379 
1380 	hval = fnhe_hashfun(daddr);
1381 
1382 	for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1383 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
1384 		if (fnhe->fnhe_daddr == daddr) {
1385 			if (fnhe->fnhe_expires &&
1386 			    time_after(jiffies, fnhe->fnhe_expires)) {
1387 				ip_del_fnhe(nhc, daddr);
1388 				break;
1389 			}
1390 			return fnhe;
1391 		}
1392 	}
1393 	return NULL;
1394 }
1395 
1396 /* MTU selection:
1397  * 1. mtu on route is locked - use it
1398  * 2. mtu from nexthop exception
1399  * 3. mtu from egress device
1400  */
1401 
1402 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1403 {
1404 	struct fib_nh_common *nhc = res->nhc;
1405 	struct net_device *dev = nhc->nhc_dev;
1406 	struct fib_info *fi = res->fi;
1407 	u32 mtu = 0;
1408 
1409 	if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1410 	    fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1411 		mtu = fi->fib_mtu;
1412 
1413 	if (likely(!mtu)) {
1414 		struct fib_nh_exception *fnhe;
1415 
1416 		fnhe = find_exception(nhc, daddr);
1417 		if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1418 			mtu = fnhe->fnhe_pmtu;
1419 	}
1420 
1421 	if (likely(!mtu))
1422 		mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1423 
1424 	return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1425 }
1426 
1427 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1428 			      __be32 daddr, const bool do_cache)
1429 {
1430 	bool ret = false;
1431 
1432 	spin_lock_bh(&fnhe_lock);
1433 
1434 	if (daddr == fnhe->fnhe_daddr) {
1435 		struct rtable __rcu **porig;
1436 		struct rtable *orig;
1437 		int genid = fnhe_genid(dev_net(rt->dst.dev));
1438 
1439 		if (rt_is_input_route(rt))
1440 			porig = &fnhe->fnhe_rth_input;
1441 		else
1442 			porig = &fnhe->fnhe_rth_output;
1443 		orig = rcu_dereference(*porig);
1444 
1445 		if (fnhe->fnhe_genid != genid) {
1446 			fnhe->fnhe_genid = genid;
1447 			fnhe->fnhe_gw = 0;
1448 			fnhe->fnhe_pmtu = 0;
1449 			fnhe->fnhe_expires = 0;
1450 			fnhe->fnhe_mtu_locked = false;
1451 			fnhe_flush_routes(fnhe);
1452 			orig = NULL;
1453 		}
1454 		fill_route_from_fnhe(rt, fnhe);
1455 		if (!rt->rt_gw4) {
1456 			rt->rt_gw4 = daddr;
1457 			rt->rt_gw_family = AF_INET;
1458 		}
1459 
1460 		if (do_cache) {
1461 			dst_hold(&rt->dst);
1462 			rcu_assign_pointer(*porig, rt);
1463 			if (orig) {
1464 				dst_dev_put(&orig->dst);
1465 				dst_release(&orig->dst);
1466 			}
1467 			ret = true;
1468 		}
1469 
1470 		fnhe->fnhe_stamp = jiffies;
1471 	}
1472 	spin_unlock_bh(&fnhe_lock);
1473 
1474 	return ret;
1475 }
1476 
1477 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1478 {
1479 	struct rtable *orig, *prev, **p;
1480 	bool ret = true;
1481 
1482 	if (rt_is_input_route(rt)) {
1483 		p = (struct rtable **)&nhc->nhc_rth_input;
1484 	} else {
1485 		p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1486 	}
1487 	orig = *p;
1488 
1489 	/* hold dst before doing cmpxchg() to avoid race condition
1490 	 * on this dst
1491 	 */
1492 	dst_hold(&rt->dst);
1493 	prev = cmpxchg(p, orig, rt);
1494 	if (prev == orig) {
1495 		if (orig) {
1496 			rt_add_uncached_list(orig);
1497 			dst_release(&orig->dst);
1498 		}
1499 	} else {
1500 		dst_release(&rt->dst);
1501 		ret = false;
1502 	}
1503 
1504 	return ret;
1505 }
1506 
1507 struct uncached_list {
1508 	spinlock_t		lock;
1509 	struct list_head	head;
1510 };
1511 
1512 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1513 
1514 void rt_add_uncached_list(struct rtable *rt)
1515 {
1516 	struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1517 
1518 	rt->dst.rt_uncached_list = ul;
1519 
1520 	spin_lock_bh(&ul->lock);
1521 	list_add_tail(&rt->dst.rt_uncached, &ul->head);
1522 	spin_unlock_bh(&ul->lock);
1523 }
1524 
1525 void rt_del_uncached_list(struct rtable *rt)
1526 {
1527 	if (!list_empty(&rt->dst.rt_uncached)) {
1528 		struct uncached_list *ul = rt->dst.rt_uncached_list;
1529 
1530 		spin_lock_bh(&ul->lock);
1531 		list_del_init(&rt->dst.rt_uncached);
1532 		spin_unlock_bh(&ul->lock);
1533 	}
1534 }
1535 
1536 static void ipv4_dst_destroy(struct dst_entry *dst)
1537 {
1538 	ip_dst_metrics_put(dst);
1539 	rt_del_uncached_list(dst_rtable(dst));
1540 }
1541 
1542 void rt_flush_dev(struct net_device *dev)
1543 {
1544 	struct rtable *rt, *safe;
1545 	int cpu;
1546 
1547 	for_each_possible_cpu(cpu) {
1548 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1549 
1550 		if (list_empty(&ul->head))
1551 			continue;
1552 
1553 		spin_lock_bh(&ul->lock);
1554 		list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1555 			if (rt->dst.dev != dev)
1556 				continue;
1557 			rt->dst.dev = blackhole_netdev;
1558 			netdev_ref_replace(dev, blackhole_netdev,
1559 					   &rt->dst.dev_tracker, GFP_ATOMIC);
1560 			list_del_init(&rt->dst.rt_uncached);
1561 		}
1562 		spin_unlock_bh(&ul->lock);
1563 	}
1564 }
1565 
1566 static bool rt_cache_valid(const struct rtable *rt)
1567 {
1568 	return	rt &&
1569 		rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1570 		!rt_is_expired(rt);
1571 }
1572 
1573 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1574 			   const struct fib_result *res,
1575 			   struct fib_nh_exception *fnhe,
1576 			   struct fib_info *fi, u16 type, u32 itag,
1577 			   const bool do_cache)
1578 {
1579 	bool cached = false;
1580 
1581 	if (fi) {
1582 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1583 
1584 		if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1585 			rt->rt_uses_gateway = 1;
1586 			rt->rt_gw_family = nhc->nhc_gw_family;
1587 			/* only INET and INET6 are supported */
1588 			if (likely(nhc->nhc_gw_family == AF_INET))
1589 				rt->rt_gw4 = nhc->nhc_gw.ipv4;
1590 			else
1591 				rt->rt_gw6 = nhc->nhc_gw.ipv6;
1592 		}
1593 
1594 		ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1595 
1596 #ifdef CONFIG_IP_ROUTE_CLASSID
1597 		if (nhc->nhc_family == AF_INET) {
1598 			struct fib_nh *nh;
1599 
1600 			nh = container_of(nhc, struct fib_nh, nh_common);
1601 			rt->dst.tclassid = nh->nh_tclassid;
1602 		}
1603 #endif
1604 		rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1605 		if (unlikely(fnhe))
1606 			cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1607 		else if (do_cache)
1608 			cached = rt_cache_route(nhc, rt);
1609 		if (unlikely(!cached)) {
1610 			/* Routes we intend to cache in nexthop exception or
1611 			 * FIB nexthop have the DST_NOCACHE bit clear.
1612 			 * However, if we are unsuccessful at storing this
1613 			 * route into the cache we really need to set it.
1614 			 */
1615 			if (!rt->rt_gw4) {
1616 				rt->rt_gw_family = AF_INET;
1617 				rt->rt_gw4 = daddr;
1618 			}
1619 			rt_add_uncached_list(rt);
1620 		}
1621 	} else
1622 		rt_add_uncached_list(rt);
1623 
1624 #ifdef CONFIG_IP_ROUTE_CLASSID
1625 #ifdef CONFIG_IP_MULTIPLE_TABLES
1626 	set_class_tag(rt, res->tclassid);
1627 #endif
1628 	set_class_tag(rt, itag);
1629 #endif
1630 }
1631 
1632 struct rtable *rt_dst_alloc(struct net_device *dev,
1633 			    unsigned int flags, u16 type,
1634 			    bool noxfrm)
1635 {
1636 	struct rtable *rt;
1637 
1638 	rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1639 		       (noxfrm ? DST_NOXFRM : 0));
1640 
1641 	if (rt) {
1642 		rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1643 		rt->rt_flags = flags;
1644 		rt->rt_type = type;
1645 		rt->rt_is_input = 0;
1646 		rt->rt_iif = 0;
1647 		rt->rt_pmtu = 0;
1648 		rt->rt_mtu_locked = 0;
1649 		rt->rt_uses_gateway = 0;
1650 		rt->rt_gw_family = 0;
1651 		rt->rt_gw4 = 0;
1652 
1653 		rt->dst.output = ip_output;
1654 		if (flags & RTCF_LOCAL)
1655 			rt->dst.input = ip_local_deliver;
1656 	}
1657 
1658 	return rt;
1659 }
1660 EXPORT_SYMBOL(rt_dst_alloc);
1661 
1662 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1663 {
1664 	struct rtable *new_rt;
1665 
1666 	new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1667 			   rt->dst.flags);
1668 
1669 	if (new_rt) {
1670 		new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1671 		new_rt->rt_flags = rt->rt_flags;
1672 		new_rt->rt_type = rt->rt_type;
1673 		new_rt->rt_is_input = rt->rt_is_input;
1674 		new_rt->rt_iif = rt->rt_iif;
1675 		new_rt->rt_pmtu = rt->rt_pmtu;
1676 		new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1677 		new_rt->rt_gw_family = rt->rt_gw_family;
1678 		if (rt->rt_gw_family == AF_INET)
1679 			new_rt->rt_gw4 = rt->rt_gw4;
1680 		else if (rt->rt_gw_family == AF_INET6)
1681 			new_rt->rt_gw6 = rt->rt_gw6;
1682 
1683 		new_rt->dst.input = rt->dst.input;
1684 		new_rt->dst.output = rt->dst.output;
1685 		new_rt->dst.error = rt->dst.error;
1686 		new_rt->dst.lastuse = jiffies;
1687 		new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1688 	}
1689 	return new_rt;
1690 }
1691 EXPORT_SYMBOL(rt_dst_clone);
1692 
1693 /* called in rcu_read_lock() section */
1694 enum skb_drop_reason
1695 ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1696 		      dscp_t dscp, struct net_device *dev,
1697 		      struct in_device *in_dev, u32 *itag)
1698 {
1699 	enum skb_drop_reason reason;
1700 
1701 	/* Primary sanity checks. */
1702 	if (!in_dev)
1703 		return SKB_DROP_REASON_NOT_SPECIFIED;
1704 
1705 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1706 		return SKB_DROP_REASON_IP_INVALID_SOURCE;
1707 
1708 	if (skb->protocol != htons(ETH_P_IP))
1709 		return SKB_DROP_REASON_INVALID_PROTO;
1710 
1711 	if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1712 		return SKB_DROP_REASON_IP_LOCALNET;
1713 
1714 	if (ipv4_is_zeronet(saddr)) {
1715 		if (!ipv4_is_local_multicast(daddr) &&
1716 		    ip_hdr(skb)->protocol != IPPROTO_IGMP)
1717 			return SKB_DROP_REASON_IP_INVALID_SOURCE;
1718 	} else {
1719 		reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
1720 						    dev, in_dev, itag);
1721 		if (reason)
1722 			return reason;
1723 	}
1724 	return SKB_NOT_DROPPED_YET;
1725 }
1726 
1727 /* called in rcu_read_lock() section */
1728 static enum skb_drop_reason
1729 ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1730 		  dscp_t dscp, struct net_device *dev, int our)
1731 {
1732 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1733 	unsigned int flags = RTCF_MULTICAST;
1734 	enum skb_drop_reason reason;
1735 	struct rtable *rth;
1736 	u32 itag = 0;
1737 
1738 	reason = ip_mc_validate_source(skb, daddr, saddr, dscp, dev, in_dev,
1739 				       &itag);
1740 	if (reason)
1741 		return reason;
1742 
1743 	if (our)
1744 		flags |= RTCF_LOCAL;
1745 
1746 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1747 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
1748 
1749 	rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1750 			   false);
1751 	if (!rth)
1752 		return SKB_DROP_REASON_NOMEM;
1753 
1754 #ifdef CONFIG_IP_ROUTE_CLASSID
1755 	rth->dst.tclassid = itag;
1756 #endif
1757 	rth->dst.output = ip_rt_bug;
1758 	rth->rt_is_input= 1;
1759 
1760 #ifdef CONFIG_IP_MROUTE
1761 	if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1762 		rth->dst.input = ip_mr_input;
1763 #endif
1764 	RT_CACHE_STAT_INC(in_slow_mc);
1765 
1766 	skb_dst_drop(skb);
1767 	skb_dst_set(skb, &rth->dst);
1768 	return SKB_NOT_DROPPED_YET;
1769 }
1770 
1771 
1772 static void ip_handle_martian_source(struct net_device *dev,
1773 				     struct in_device *in_dev,
1774 				     struct sk_buff *skb,
1775 				     __be32 daddr,
1776 				     __be32 saddr)
1777 {
1778 	RT_CACHE_STAT_INC(in_martian_src);
1779 #ifdef CONFIG_IP_ROUTE_VERBOSE
1780 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1781 		/*
1782 		 *	RFC1812 recommendation, if source is martian,
1783 		 *	the only hint is MAC header.
1784 		 */
1785 		pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1786 			&daddr, &saddr, dev->name);
1787 		if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1788 			print_hex_dump(KERN_WARNING, "ll header: ",
1789 				       DUMP_PREFIX_OFFSET, 16, 1,
1790 				       skb_mac_header(skb),
1791 				       dev->hard_header_len, false);
1792 		}
1793 	}
1794 #endif
1795 }
1796 
1797 /* called in rcu_read_lock() section */
1798 static enum skb_drop_reason
1799 __mkroute_input(struct sk_buff *skb, const struct fib_result *res,
1800 		struct in_device *in_dev, __be32 daddr,
1801 		__be32 saddr, dscp_t dscp)
1802 {
1803 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1804 	struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1805 	struct net_device *dev = nhc->nhc_dev;
1806 	struct fib_nh_exception *fnhe;
1807 	struct rtable *rth;
1808 	int err;
1809 	struct in_device *out_dev;
1810 	bool do_cache;
1811 	u32 itag = 0;
1812 
1813 	/* get a working reference to the output device */
1814 	out_dev = __in_dev_get_rcu(dev);
1815 	if (!out_dev) {
1816 		net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1817 		return reason;
1818 	}
1819 
1820 	err = fib_validate_source(skb, saddr, daddr, dscp, FIB_RES_OIF(*res),
1821 				  in_dev->dev, in_dev, &itag);
1822 	if (err < 0) {
1823 		reason = -err;
1824 		ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1825 					 saddr);
1826 
1827 		goto cleanup;
1828 	}
1829 
1830 	do_cache = res->fi && !itag;
1831 	if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1832 	    skb->protocol == htons(ETH_P_IP)) {
1833 		__be32 gw;
1834 
1835 		gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1836 		if (IN_DEV_SHARED_MEDIA(out_dev) ||
1837 		    inet_addr_onlink(out_dev, saddr, gw))
1838 			IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1839 	}
1840 
1841 	if (skb->protocol != htons(ETH_P_IP)) {
1842 		/* Not IP (i.e. ARP). Do not create route, if it is
1843 		 * invalid for proxy arp. DNAT routes are always valid.
1844 		 *
1845 		 * Proxy arp feature have been extended to allow, ARP
1846 		 * replies back to the same interface, to support
1847 		 * Private VLAN switch technologies. See arp.c.
1848 		 */
1849 		if (out_dev == in_dev &&
1850 		    IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1851 			reason = SKB_DROP_REASON_ARP_PVLAN_DISABLE;
1852 			goto cleanup;
1853 		}
1854 	}
1855 
1856 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1857 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
1858 
1859 	fnhe = find_exception(nhc, daddr);
1860 	if (do_cache) {
1861 		if (fnhe)
1862 			rth = rcu_dereference(fnhe->fnhe_rth_input);
1863 		else
1864 			rth = rcu_dereference(nhc->nhc_rth_input);
1865 		if (rt_cache_valid(rth)) {
1866 			skb_dst_set_noref(skb, &rth->dst);
1867 			goto out;
1868 		}
1869 	}
1870 
1871 	rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1872 			   IN_DEV_ORCONF(out_dev, NOXFRM));
1873 	if (!rth) {
1874 		reason = SKB_DROP_REASON_NOMEM;
1875 		goto cleanup;
1876 	}
1877 
1878 	rth->rt_is_input = 1;
1879 	RT_CACHE_STAT_INC(in_slow_tot);
1880 
1881 	rth->dst.input = ip_forward;
1882 
1883 	rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1884 		       do_cache);
1885 	lwtunnel_set_redirect(&rth->dst);
1886 	skb_dst_set(skb, &rth->dst);
1887 out:
1888 	reason = SKB_NOT_DROPPED_YET;
1889 cleanup:
1890 	return reason;
1891 }
1892 
1893 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1894 /* To make ICMP packets follow the right flow, the multipath hash is
1895  * calculated from the inner IP addresses.
1896  */
1897 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1898 				 struct flow_keys *hash_keys)
1899 {
1900 	const struct iphdr *outer_iph = ip_hdr(skb);
1901 	const struct iphdr *key_iph = outer_iph;
1902 	const struct iphdr *inner_iph;
1903 	const struct icmphdr *icmph;
1904 	struct iphdr _inner_iph;
1905 	struct icmphdr _icmph;
1906 
1907 	if (likely(outer_iph->protocol != IPPROTO_ICMP))
1908 		goto out;
1909 
1910 	if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1911 		goto out;
1912 
1913 	icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1914 				   &_icmph);
1915 	if (!icmph)
1916 		goto out;
1917 
1918 	if (!icmp_is_err(icmph->type))
1919 		goto out;
1920 
1921 	inner_iph = skb_header_pointer(skb,
1922 				       outer_iph->ihl * 4 + sizeof(_icmph),
1923 				       sizeof(_inner_iph), &_inner_iph);
1924 	if (!inner_iph)
1925 		goto out;
1926 
1927 	key_iph = inner_iph;
1928 out:
1929 	hash_keys->addrs.v4addrs.src = key_iph->saddr;
1930 	hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1931 }
1932 
1933 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1934 					   const struct sk_buff *skb,
1935 					   bool *p_has_inner)
1936 {
1937 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1938 	struct flow_keys keys, hash_keys;
1939 
1940 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1941 		return 0;
1942 
1943 	memset(&hash_keys, 0, sizeof(hash_keys));
1944 	skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1945 
1946 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1947 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1948 		hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1949 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1950 		hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1951 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1952 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
1953 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1954 		hash_keys.ports.src = keys.ports.src;
1955 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1956 		hash_keys.ports.dst = keys.ports.dst;
1957 
1958 	*p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1959 	return fib_multipath_hash_from_keys(net, &hash_keys);
1960 }
1961 
1962 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1963 					   const struct sk_buff *skb,
1964 					   bool has_inner)
1965 {
1966 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1967 	struct flow_keys keys, hash_keys;
1968 
1969 	/* We assume the packet carries an encapsulation, but if none was
1970 	 * encountered during dissection of the outer flow, then there is no
1971 	 * point in calling the flow dissector again.
1972 	 */
1973 	if (!has_inner)
1974 		return 0;
1975 
1976 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1977 		return 0;
1978 
1979 	memset(&hash_keys, 0, sizeof(hash_keys));
1980 	skb_flow_dissect_flow_keys(skb, &keys, 0);
1981 
1982 	if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1983 		return 0;
1984 
1985 	if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1986 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1987 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1988 			hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1989 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1990 			hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1991 	} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1992 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1993 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1994 			hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1995 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1996 			hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1997 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1998 			hash_keys.tags.flow_label = keys.tags.flow_label;
1999 	}
2000 
2001 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2002 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
2003 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2004 		hash_keys.ports.src = keys.ports.src;
2005 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2006 		hash_keys.ports.dst = keys.ports.dst;
2007 
2008 	return fib_multipath_hash_from_keys(net, &hash_keys);
2009 }
2010 
2011 static u32 fib_multipath_custom_hash_skb(const struct net *net,
2012 					 const struct sk_buff *skb)
2013 {
2014 	u32 mhash, mhash_inner;
2015 	bool has_inner = true;
2016 
2017 	mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2018 	mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2019 
2020 	return jhash_2words(mhash, mhash_inner, 0);
2021 }
2022 
2023 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2024 					 const struct flowi4 *fl4)
2025 {
2026 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2027 	struct flow_keys hash_keys;
2028 
2029 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2030 		return 0;
2031 
2032 	memset(&hash_keys, 0, sizeof(hash_keys));
2033 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2034 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2035 		hash_keys.addrs.v4addrs.src = fl4->saddr;
2036 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2037 		hash_keys.addrs.v4addrs.dst = fl4->daddr;
2038 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2039 		hash_keys.basic.ip_proto = fl4->flowi4_proto;
2040 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) {
2041 		if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT)
2042 			hash_keys.ports.src = (__force __be16)get_random_u16();
2043 		else
2044 			hash_keys.ports.src = fl4->fl4_sport;
2045 	}
2046 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2047 		hash_keys.ports.dst = fl4->fl4_dport;
2048 
2049 	return fib_multipath_hash_from_keys(net, &hash_keys);
2050 }
2051 
2052 /* if skb is set it will be used and fl4 can be NULL */
2053 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2054 		       const struct sk_buff *skb, struct flow_keys *flkeys)
2055 {
2056 	u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2057 	struct flow_keys hash_keys;
2058 	u32 mhash = 0;
2059 
2060 	switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2061 	case 0:
2062 		memset(&hash_keys, 0, sizeof(hash_keys));
2063 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2064 		if (skb) {
2065 			ip_multipath_l3_keys(skb, &hash_keys);
2066 		} else {
2067 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2068 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2069 		}
2070 		mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2071 		break;
2072 	case 1:
2073 		/* skb is currently provided only when forwarding */
2074 		if (skb) {
2075 			unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2076 			struct flow_keys keys;
2077 
2078 			/* short-circuit if we already have L4 hash present */
2079 			if (skb->l4_hash)
2080 				return skb_get_hash_raw(skb) >> 1;
2081 
2082 			memset(&hash_keys, 0, sizeof(hash_keys));
2083 
2084 			if (!flkeys) {
2085 				skb_flow_dissect_flow_keys(skb, &keys, flag);
2086 				flkeys = &keys;
2087 			}
2088 
2089 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2090 			hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2091 			hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2092 			hash_keys.ports.src = flkeys->ports.src;
2093 			hash_keys.ports.dst = flkeys->ports.dst;
2094 			hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2095 		} else {
2096 			memset(&hash_keys, 0, sizeof(hash_keys));
2097 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2098 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2099 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2100 			if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT)
2101 				hash_keys.ports.src = (__force __be16)get_random_u16();
2102 			else
2103 				hash_keys.ports.src = fl4->fl4_sport;
2104 			hash_keys.ports.dst = fl4->fl4_dport;
2105 			hash_keys.basic.ip_proto = fl4->flowi4_proto;
2106 		}
2107 		mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2108 		break;
2109 	case 2:
2110 		memset(&hash_keys, 0, sizeof(hash_keys));
2111 		/* skb is currently provided only when forwarding */
2112 		if (skb) {
2113 			struct flow_keys keys;
2114 
2115 			skb_flow_dissect_flow_keys(skb, &keys, 0);
2116 			/* Inner can be v4 or v6 */
2117 			if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2118 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2119 				hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2120 				hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2121 			} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2122 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2123 				hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2124 				hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2125 				hash_keys.tags.flow_label = keys.tags.flow_label;
2126 				hash_keys.basic.ip_proto = keys.basic.ip_proto;
2127 			} else {
2128 				/* Same as case 0 */
2129 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2130 				ip_multipath_l3_keys(skb, &hash_keys);
2131 			}
2132 		} else {
2133 			/* Same as case 0 */
2134 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2135 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2136 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2137 		}
2138 		mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2139 		break;
2140 	case 3:
2141 		if (skb)
2142 			mhash = fib_multipath_custom_hash_skb(net, skb);
2143 		else
2144 			mhash = fib_multipath_custom_hash_fl4(net, fl4);
2145 		break;
2146 	}
2147 
2148 	if (multipath_hash)
2149 		mhash = jhash_2words(mhash, multipath_hash, 0);
2150 
2151 	return mhash >> 1;
2152 }
2153 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2154 
2155 static enum skb_drop_reason
2156 ip_mkroute_input(struct sk_buff *skb, struct fib_result *res,
2157 		 struct in_device *in_dev, __be32 daddr,
2158 		 __be32 saddr, dscp_t dscp, struct flow_keys *hkeys)
2159 {
2160 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2161 	if (res->fi && fib_info_num_path(res->fi) > 1) {
2162 		int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2163 
2164 		fib_select_multipath(res, h, NULL);
2165 		IPCB(skb)->flags |= IPSKB_MULTIPATH;
2166 	}
2167 #endif
2168 
2169 	/* create a routing cache entry */
2170 	return __mkroute_input(skb, res, in_dev, daddr, saddr, dscp);
2171 }
2172 
2173 /* Implements all the saddr-related checks as ip_route_input_slow(),
2174  * assuming daddr is valid and the destination is not a local broadcast one.
2175  * Uses the provided hint instead of performing a route lookup.
2176  */
2177 enum skb_drop_reason
2178 ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2179 		  dscp_t dscp, struct net_device *dev,
2180 		  const struct sk_buff *hint)
2181 {
2182 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2183 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2184 	struct rtable *rt = skb_rtable(hint);
2185 	struct net *net = dev_net(dev);
2186 	u32 tag = 0;
2187 
2188 	if (!in_dev)
2189 		return reason;
2190 
2191 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2192 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2193 		goto martian_source;
2194 	}
2195 
2196 	if (ipv4_is_zeronet(saddr)) {
2197 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2198 		goto martian_source;
2199 	}
2200 
2201 	if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2202 		reason = SKB_DROP_REASON_IP_LOCALNET;
2203 		goto martian_source;
2204 	}
2205 
2206 	if (rt->rt_type != RTN_LOCAL)
2207 		goto skip_validate_source;
2208 
2209 	reason = fib_validate_source_reason(skb, saddr, daddr, dscp, 0, dev,
2210 					    in_dev, &tag);
2211 	if (reason)
2212 		goto martian_source;
2213 
2214 skip_validate_source:
2215 	skb_dst_copy(skb, hint);
2216 	return SKB_NOT_DROPPED_YET;
2217 
2218 martian_source:
2219 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2220 	return reason;
2221 }
2222 
2223 /* get device for dst_alloc with local routes */
2224 static struct net_device *ip_rt_get_dev(struct net *net,
2225 					const struct fib_result *res)
2226 {
2227 	struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2228 	struct net_device *dev = NULL;
2229 
2230 	if (nhc)
2231 		dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2232 
2233 	return dev ? : net->loopback_dev;
2234 }
2235 
2236 /*
2237  *	NOTE. We drop all the packets that has local source
2238  *	addresses, because every properly looped back packet
2239  *	must have correct destination already attached by output routine.
2240  *	Changes in the enforced policies must be applied also to
2241  *	ip_route_use_hint().
2242  *
2243  *	Such approach solves two big problems:
2244  *	1. Not simplex devices are handled properly.
2245  *	2. IP spoofing attempts are filtered with 100% of guarantee.
2246  *	called with rcu_read_lock()
2247  */
2248 
2249 static enum skb_drop_reason
2250 ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2251 		    dscp_t dscp, struct net_device *dev,
2252 		    struct fib_result *res)
2253 {
2254 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2255 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2256 	struct flow_keys *flkeys = NULL, _flkeys;
2257 	struct net    *net = dev_net(dev);
2258 	struct ip_tunnel_info *tun_info;
2259 	int		err = -EINVAL;
2260 	unsigned int	flags = 0;
2261 	u32		itag = 0;
2262 	struct rtable	*rth;
2263 	struct flowi4	fl4;
2264 	bool do_cache = true;
2265 
2266 	/* IP on this device is disabled. */
2267 
2268 	if (!in_dev)
2269 		goto out;
2270 
2271 	/* Check for the most weird martians, which can be not detected
2272 	 * by fib_lookup.
2273 	 */
2274 
2275 	tun_info = skb_tunnel_info(skb);
2276 	if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2277 		fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2278 	else
2279 		fl4.flowi4_tun_key.tun_id = 0;
2280 	skb_dst_drop(skb);
2281 
2282 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2283 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2284 		goto martian_source;
2285 	}
2286 
2287 	res->fi = NULL;
2288 	res->table = NULL;
2289 	if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2290 		goto brd_input;
2291 
2292 	/* Accept zero addresses only to limited broadcast;
2293 	 * I even do not know to fix it or not. Waiting for complains :-)
2294 	 */
2295 	if (ipv4_is_zeronet(saddr)) {
2296 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2297 		goto martian_source;
2298 	}
2299 
2300 	if (ipv4_is_zeronet(daddr)) {
2301 		reason = SKB_DROP_REASON_IP_INVALID_DEST;
2302 		goto martian_destination;
2303 	}
2304 
2305 	/* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2306 	 * and call it once if daddr or/and saddr are loopback addresses
2307 	 */
2308 	if (ipv4_is_loopback(daddr)) {
2309 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2310 			reason = SKB_DROP_REASON_IP_LOCALNET;
2311 			goto martian_destination;
2312 		}
2313 	} else if (ipv4_is_loopback(saddr)) {
2314 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2315 			reason = SKB_DROP_REASON_IP_LOCALNET;
2316 			goto martian_source;
2317 		}
2318 	}
2319 
2320 	/*
2321 	 *	Now we are ready to route packet.
2322 	 */
2323 	fl4.flowi4_l3mdev = 0;
2324 	fl4.flowi4_oif = 0;
2325 	fl4.flowi4_iif = dev->ifindex;
2326 	fl4.flowi4_mark = skb->mark;
2327 	fl4.flowi4_tos = inet_dscp_to_dsfield(dscp);
2328 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2329 	fl4.flowi4_flags = 0;
2330 	fl4.daddr = daddr;
2331 	fl4.saddr = saddr;
2332 	fl4.flowi4_uid = sock_net_uid(net, NULL);
2333 	fl4.flowi4_multipath_hash = 0;
2334 
2335 	if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2336 		flkeys = &_flkeys;
2337 	} else {
2338 		fl4.flowi4_proto = 0;
2339 		fl4.fl4_sport = 0;
2340 		fl4.fl4_dport = 0;
2341 	}
2342 
2343 	err = fib_lookup(net, &fl4, res, 0);
2344 	if (err != 0) {
2345 		if (!IN_DEV_FORWARD(in_dev))
2346 			err = -EHOSTUNREACH;
2347 		goto no_route;
2348 	}
2349 
2350 	if (res->type == RTN_BROADCAST) {
2351 		if (IN_DEV_BFORWARD(in_dev))
2352 			goto make_route;
2353 		/* not do cache if bc_forwarding is enabled */
2354 		if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING))
2355 			do_cache = false;
2356 		goto brd_input;
2357 	}
2358 
2359 	err = -EINVAL;
2360 	if (res->type == RTN_LOCAL) {
2361 		reason = fib_validate_source_reason(skb, saddr, daddr, dscp,
2362 						    0, dev, in_dev, &itag);
2363 		if (reason)
2364 			goto martian_source;
2365 		goto local_input;
2366 	}
2367 
2368 	if (!IN_DEV_FORWARD(in_dev)) {
2369 		err = -EHOSTUNREACH;
2370 		goto no_route;
2371 	}
2372 	if (res->type != RTN_UNICAST) {
2373 		reason = SKB_DROP_REASON_IP_INVALID_DEST;
2374 		goto martian_destination;
2375 	}
2376 
2377 make_route:
2378 	reason = ip_mkroute_input(skb, res, in_dev, daddr, saddr, dscp,
2379 				  flkeys);
2380 
2381 out:
2382 	return reason;
2383 
2384 brd_input:
2385 	if (skb->protocol != htons(ETH_P_IP)) {
2386 		reason = SKB_DROP_REASON_INVALID_PROTO;
2387 		goto out;
2388 	}
2389 
2390 	if (!ipv4_is_zeronet(saddr)) {
2391 		reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
2392 						    dev, in_dev, &itag);
2393 		if (reason)
2394 			goto martian_source;
2395 	}
2396 	flags |= RTCF_BROADCAST;
2397 	res->type = RTN_BROADCAST;
2398 	RT_CACHE_STAT_INC(in_brd);
2399 
2400 local_input:
2401 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2402 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
2403 
2404 	do_cache &= res->fi && !itag;
2405 	if (do_cache) {
2406 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2407 
2408 		rth = rcu_dereference(nhc->nhc_rth_input);
2409 		if (rt_cache_valid(rth)) {
2410 			skb_dst_set_noref(skb, &rth->dst);
2411 			reason = SKB_NOT_DROPPED_YET;
2412 			goto out;
2413 		}
2414 	}
2415 
2416 	rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2417 			   flags | RTCF_LOCAL, res->type, false);
2418 	if (!rth)
2419 		goto e_nobufs;
2420 
2421 	rth->dst.output= ip_rt_bug;
2422 #ifdef CONFIG_IP_ROUTE_CLASSID
2423 	rth->dst.tclassid = itag;
2424 #endif
2425 	rth->rt_is_input = 1;
2426 
2427 	RT_CACHE_STAT_INC(in_slow_tot);
2428 	if (res->type == RTN_UNREACHABLE) {
2429 		rth->dst.input= ip_error;
2430 		rth->dst.error= -err;
2431 		rth->rt_flags	&= ~RTCF_LOCAL;
2432 	}
2433 
2434 	if (do_cache) {
2435 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2436 
2437 		rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2438 		if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2439 			WARN_ON(rth->dst.input == lwtunnel_input);
2440 			rth->dst.lwtstate->orig_input = rth->dst.input;
2441 			rth->dst.input = lwtunnel_input;
2442 		}
2443 
2444 		if (unlikely(!rt_cache_route(nhc, rth)))
2445 			rt_add_uncached_list(rth);
2446 	}
2447 	skb_dst_set(skb, &rth->dst);
2448 	reason = SKB_NOT_DROPPED_YET;
2449 	goto out;
2450 
2451 no_route:
2452 	RT_CACHE_STAT_INC(in_no_route);
2453 	res->type = RTN_UNREACHABLE;
2454 	res->fi = NULL;
2455 	res->table = NULL;
2456 	goto local_input;
2457 
2458 	/*
2459 	 *	Do not cache martian addresses: they should be logged (RFC1812)
2460 	 */
2461 martian_destination:
2462 	RT_CACHE_STAT_INC(in_martian_dst);
2463 #ifdef CONFIG_IP_ROUTE_VERBOSE
2464 	if (IN_DEV_LOG_MARTIANS(in_dev))
2465 		net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2466 				     &daddr, &saddr, dev->name);
2467 #endif
2468 	goto out;
2469 
2470 e_nobufs:
2471 	reason = SKB_DROP_REASON_NOMEM;
2472 	goto out;
2473 
2474 martian_source:
2475 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2476 	goto out;
2477 }
2478 
2479 /* called with rcu_read_lock held */
2480 static enum skb_drop_reason
2481 ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2482 		   dscp_t dscp, struct net_device *dev,
2483 		   struct fib_result *res)
2484 {
2485 	/* Multicast recognition logic is moved from route cache to here.
2486 	 * The problem was that too many Ethernet cards have broken/missing
2487 	 * hardware multicast filters :-( As result the host on multicasting
2488 	 * network acquires a lot of useless route cache entries, sort of
2489 	 * SDR messages from all the world. Now we try to get rid of them.
2490 	 * Really, provided software IP multicast filter is organized
2491 	 * reasonably (at least, hashed), it does not result in a slowdown
2492 	 * comparing with route cache reject entries.
2493 	 * Note, that multicast routers are not affected, because
2494 	 * route cache entry is created eventually.
2495 	 */
2496 	if (ipv4_is_multicast(daddr)) {
2497 		enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2498 		struct in_device *in_dev = __in_dev_get_rcu(dev);
2499 		int our = 0;
2500 
2501 		if (!in_dev)
2502 			return reason;
2503 
2504 		our = ip_check_mc_rcu(in_dev, daddr, saddr,
2505 				      ip_hdr(skb)->protocol);
2506 
2507 		/* check l3 master if no match yet */
2508 		if (!our && netif_is_l3_slave(dev)) {
2509 			struct in_device *l3_in_dev;
2510 
2511 			l3_in_dev = __in_dev_get_rcu(skb->dev);
2512 			if (l3_in_dev)
2513 				our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2514 						      ip_hdr(skb)->protocol);
2515 		}
2516 
2517 		if (our
2518 #ifdef CONFIG_IP_MROUTE
2519 			||
2520 		    (!ipv4_is_local_multicast(daddr) &&
2521 		     IN_DEV_MFORWARD(in_dev))
2522 #endif
2523 		   ) {
2524 			reason = ip_route_input_mc(skb, daddr, saddr, dscp,
2525 						   dev, our);
2526 		}
2527 		return reason;
2528 	}
2529 
2530 	return ip_route_input_slow(skb, daddr, saddr, dscp, dev, res);
2531 }
2532 
2533 enum skb_drop_reason ip_route_input_noref(struct sk_buff *skb, __be32 daddr,
2534 					  __be32 saddr, dscp_t dscp,
2535 					  struct net_device *dev)
2536 {
2537 	enum skb_drop_reason reason;
2538 	struct fib_result res;
2539 
2540 	rcu_read_lock();
2541 	reason = ip_route_input_rcu(skb, daddr, saddr, dscp, dev, &res);
2542 	rcu_read_unlock();
2543 
2544 	return reason;
2545 }
2546 EXPORT_SYMBOL(ip_route_input_noref);
2547 
2548 /* called with rcu_read_lock() */
2549 static struct rtable *__mkroute_output(const struct fib_result *res,
2550 				       const struct flowi4 *fl4, int orig_oif,
2551 				       struct net_device *dev_out,
2552 				       unsigned int flags)
2553 {
2554 	struct fib_info *fi = res->fi;
2555 	struct fib_nh_exception *fnhe;
2556 	struct in_device *in_dev;
2557 	u16 type = res->type;
2558 	struct rtable *rth;
2559 	bool do_cache;
2560 
2561 	in_dev = __in_dev_get_rcu(dev_out);
2562 	if (!in_dev)
2563 		return ERR_PTR(-EINVAL);
2564 
2565 	if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2566 		if (ipv4_is_loopback(fl4->saddr) &&
2567 		    !(dev_out->flags & IFF_LOOPBACK) &&
2568 		    !netif_is_l3_master(dev_out))
2569 			return ERR_PTR(-EINVAL);
2570 
2571 	if (ipv4_is_lbcast(fl4->daddr))
2572 		type = RTN_BROADCAST;
2573 	else if (ipv4_is_multicast(fl4->daddr))
2574 		type = RTN_MULTICAST;
2575 	else if (ipv4_is_zeronet(fl4->daddr))
2576 		return ERR_PTR(-EINVAL);
2577 
2578 	if (dev_out->flags & IFF_LOOPBACK)
2579 		flags |= RTCF_LOCAL;
2580 
2581 	do_cache = true;
2582 	if (type == RTN_BROADCAST) {
2583 		flags |= RTCF_BROADCAST | RTCF_LOCAL;
2584 		fi = NULL;
2585 	} else if (type == RTN_MULTICAST) {
2586 		flags |= RTCF_MULTICAST | RTCF_LOCAL;
2587 		if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2588 				     fl4->flowi4_proto))
2589 			flags &= ~RTCF_LOCAL;
2590 		else
2591 			do_cache = false;
2592 		/* If multicast route do not exist use
2593 		 * default one, but do not gateway in this case.
2594 		 * Yes, it is hack.
2595 		 */
2596 		if (fi && res->prefixlen < 4)
2597 			fi = NULL;
2598 	} else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2599 		   (orig_oif != dev_out->ifindex)) {
2600 		/* For local routes that require a particular output interface
2601 		 * we do not want to cache the result.  Caching the result
2602 		 * causes incorrect behaviour when there are multiple source
2603 		 * addresses on the interface, the end result being that if the
2604 		 * intended recipient is waiting on that interface for the
2605 		 * packet he won't receive it because it will be delivered on
2606 		 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2607 		 * be set to the loopback interface as well.
2608 		 */
2609 		do_cache = false;
2610 	}
2611 
2612 	fnhe = NULL;
2613 	do_cache &= fi != NULL;
2614 	if (fi) {
2615 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2616 		struct rtable __rcu **prth;
2617 
2618 		fnhe = find_exception(nhc, fl4->daddr);
2619 		if (!do_cache)
2620 			goto add;
2621 		if (fnhe) {
2622 			prth = &fnhe->fnhe_rth_output;
2623 		} else {
2624 			if (unlikely(fl4->flowi4_flags &
2625 				     FLOWI_FLAG_KNOWN_NH &&
2626 				     !(nhc->nhc_gw_family &&
2627 				       nhc->nhc_scope == RT_SCOPE_LINK))) {
2628 				do_cache = false;
2629 				goto add;
2630 			}
2631 			prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2632 		}
2633 		rth = rcu_dereference(*prth);
2634 		if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2635 			return rth;
2636 	}
2637 
2638 add:
2639 	rth = rt_dst_alloc(dev_out, flags, type,
2640 			   IN_DEV_ORCONF(in_dev, NOXFRM));
2641 	if (!rth)
2642 		return ERR_PTR(-ENOBUFS);
2643 
2644 	rth->rt_iif = orig_oif;
2645 
2646 	RT_CACHE_STAT_INC(out_slow_tot);
2647 
2648 	if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2649 		if (flags & RTCF_LOCAL &&
2650 		    !(dev_out->flags & IFF_LOOPBACK)) {
2651 			rth->dst.output = ip_mc_output;
2652 			RT_CACHE_STAT_INC(out_slow_mc);
2653 		}
2654 #ifdef CONFIG_IP_MROUTE
2655 		if (type == RTN_MULTICAST) {
2656 			if (IN_DEV_MFORWARD(in_dev) &&
2657 			    !ipv4_is_local_multicast(fl4->daddr)) {
2658 				rth->dst.input = ip_mr_input;
2659 				rth->dst.output = ip_mc_output;
2660 			}
2661 		}
2662 #endif
2663 	}
2664 
2665 	rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2666 	lwtunnel_set_redirect(&rth->dst);
2667 
2668 	return rth;
2669 }
2670 
2671 /*
2672  * Major route resolver routine.
2673  */
2674 
2675 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2676 					const struct sk_buff *skb)
2677 {
2678 	struct fib_result res = {
2679 		.type		= RTN_UNSPEC,
2680 		.fi		= NULL,
2681 		.table		= NULL,
2682 		.tclassid	= 0,
2683 	};
2684 	struct rtable *rth;
2685 
2686 	fl4->flowi4_iif = LOOPBACK_IFINDEX;
2687 	fl4->flowi4_tos &= INET_DSCP_MASK;
2688 
2689 	rcu_read_lock();
2690 	rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2691 	rcu_read_unlock();
2692 
2693 	return rth;
2694 }
2695 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2696 
2697 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2698 					    struct fib_result *res,
2699 					    const struct sk_buff *skb)
2700 {
2701 	struct net_device *dev_out = NULL;
2702 	int orig_oif = fl4->flowi4_oif;
2703 	unsigned int flags = 0;
2704 	struct rtable *rth;
2705 	int err;
2706 
2707 	if (fl4->saddr) {
2708 		if (ipv4_is_multicast(fl4->saddr) ||
2709 		    ipv4_is_lbcast(fl4->saddr)) {
2710 			rth = ERR_PTR(-EINVAL);
2711 			goto out;
2712 		}
2713 
2714 		rth = ERR_PTR(-ENETUNREACH);
2715 
2716 		/* I removed check for oif == dev_out->oif here.
2717 		 * It was wrong for two reasons:
2718 		 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2719 		 *    is assigned to multiple interfaces.
2720 		 * 2. Moreover, we are allowed to send packets with saddr
2721 		 *    of another iface. --ANK
2722 		 */
2723 
2724 		if (fl4->flowi4_oif == 0 &&
2725 		    (ipv4_is_multicast(fl4->daddr) ||
2726 		     ipv4_is_lbcast(fl4->daddr))) {
2727 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2728 			dev_out = __ip_dev_find(net, fl4->saddr, false);
2729 			if (!dev_out)
2730 				goto out;
2731 
2732 			/* Special hack: user can direct multicasts
2733 			 * and limited broadcast via necessary interface
2734 			 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2735 			 * This hack is not just for fun, it allows
2736 			 * vic,vat and friends to work.
2737 			 * They bind socket to loopback, set ttl to zero
2738 			 * and expect that it will work.
2739 			 * From the viewpoint of routing cache they are broken,
2740 			 * because we are not allowed to build multicast path
2741 			 * with loopback source addr (look, routing cache
2742 			 * cannot know, that ttl is zero, so that packet
2743 			 * will not leave this host and route is valid).
2744 			 * Luckily, this hack is good workaround.
2745 			 */
2746 
2747 			fl4->flowi4_oif = dev_out->ifindex;
2748 			goto make_route;
2749 		}
2750 
2751 		if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2752 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2753 			if (!__ip_dev_find(net, fl4->saddr, false))
2754 				goto out;
2755 		}
2756 	}
2757 
2758 
2759 	if (fl4->flowi4_oif) {
2760 		dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2761 		rth = ERR_PTR(-ENODEV);
2762 		if (!dev_out)
2763 			goto out;
2764 
2765 		/* RACE: Check return value of inet_select_addr instead. */
2766 		if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2767 			rth = ERR_PTR(-ENETUNREACH);
2768 			goto out;
2769 		}
2770 		if (ipv4_is_local_multicast(fl4->daddr) ||
2771 		    ipv4_is_lbcast(fl4->daddr) ||
2772 		    fl4->flowi4_proto == IPPROTO_IGMP) {
2773 			if (!fl4->saddr)
2774 				fl4->saddr = inet_select_addr(dev_out, 0,
2775 							      RT_SCOPE_LINK);
2776 			goto make_route;
2777 		}
2778 		if (!fl4->saddr) {
2779 			if (ipv4_is_multicast(fl4->daddr))
2780 				fl4->saddr = inet_select_addr(dev_out, 0,
2781 							      fl4->flowi4_scope);
2782 			else if (!fl4->daddr)
2783 				fl4->saddr = inet_select_addr(dev_out, 0,
2784 							      RT_SCOPE_HOST);
2785 		}
2786 	}
2787 
2788 	if (!fl4->daddr) {
2789 		fl4->daddr = fl4->saddr;
2790 		if (!fl4->daddr)
2791 			fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2792 		dev_out = net->loopback_dev;
2793 		fl4->flowi4_oif = LOOPBACK_IFINDEX;
2794 		res->type = RTN_LOCAL;
2795 		flags |= RTCF_LOCAL;
2796 		goto make_route;
2797 	}
2798 
2799 	err = fib_lookup(net, fl4, res, 0);
2800 	if (err) {
2801 		res->fi = NULL;
2802 		res->table = NULL;
2803 		if (fl4->flowi4_oif &&
2804 		    (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2805 			/* Apparently, routing tables are wrong. Assume,
2806 			 * that the destination is on link.
2807 			 *
2808 			 * WHY? DW.
2809 			 * Because we are allowed to send to iface
2810 			 * even if it has NO routes and NO assigned
2811 			 * addresses. When oif is specified, routing
2812 			 * tables are looked up with only one purpose:
2813 			 * to catch if destination is gatewayed, rather than
2814 			 * direct. Moreover, if MSG_DONTROUTE is set,
2815 			 * we send packet, ignoring both routing tables
2816 			 * and ifaddr state. --ANK
2817 			 *
2818 			 *
2819 			 * We could make it even if oif is unknown,
2820 			 * likely IPv6, but we do not.
2821 			 */
2822 
2823 			if (fl4->saddr == 0)
2824 				fl4->saddr = inet_select_addr(dev_out, 0,
2825 							      RT_SCOPE_LINK);
2826 			res->type = RTN_UNICAST;
2827 			goto make_route;
2828 		}
2829 		rth = ERR_PTR(err);
2830 		goto out;
2831 	}
2832 
2833 	if (res->type == RTN_LOCAL) {
2834 		if (!fl4->saddr) {
2835 			if (res->fi->fib_prefsrc)
2836 				fl4->saddr = res->fi->fib_prefsrc;
2837 			else
2838 				fl4->saddr = fl4->daddr;
2839 		}
2840 
2841 		/* L3 master device is the loopback for that domain */
2842 		dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2843 			net->loopback_dev;
2844 
2845 		/* make sure orig_oif points to fib result device even
2846 		 * though packet rx/tx happens over loopback or l3mdev
2847 		 */
2848 		orig_oif = FIB_RES_OIF(*res);
2849 
2850 		fl4->flowi4_oif = dev_out->ifindex;
2851 		flags |= RTCF_LOCAL;
2852 		goto make_route;
2853 	}
2854 
2855 	fib_select_path(net, res, fl4, skb);
2856 
2857 	dev_out = FIB_RES_DEV(*res);
2858 
2859 make_route:
2860 	rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2861 
2862 out:
2863 	return rth;
2864 }
2865 
2866 static struct dst_ops ipv4_dst_blackhole_ops = {
2867 	.family			= AF_INET,
2868 	.default_advmss		= ipv4_default_advmss,
2869 	.neigh_lookup		= ipv4_neigh_lookup,
2870 	.check			= dst_blackhole_check,
2871 	.cow_metrics		= dst_blackhole_cow_metrics,
2872 	.update_pmtu		= dst_blackhole_update_pmtu,
2873 	.redirect		= dst_blackhole_redirect,
2874 	.mtu			= dst_blackhole_mtu,
2875 };
2876 
2877 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2878 {
2879 	struct rtable *ort = dst_rtable(dst_orig);
2880 	struct rtable *rt;
2881 
2882 	rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0);
2883 	if (rt) {
2884 		struct dst_entry *new = &rt->dst;
2885 
2886 		new->__use = 1;
2887 		new->input = dst_discard;
2888 		new->output = dst_discard_out;
2889 
2890 		new->dev = net->loopback_dev;
2891 		netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2892 
2893 		rt->rt_is_input = ort->rt_is_input;
2894 		rt->rt_iif = ort->rt_iif;
2895 		rt->rt_pmtu = ort->rt_pmtu;
2896 		rt->rt_mtu_locked = ort->rt_mtu_locked;
2897 
2898 		rt->rt_genid = rt_genid_ipv4(net);
2899 		rt->rt_flags = ort->rt_flags;
2900 		rt->rt_type = ort->rt_type;
2901 		rt->rt_uses_gateway = ort->rt_uses_gateway;
2902 		rt->rt_gw_family = ort->rt_gw_family;
2903 		if (rt->rt_gw_family == AF_INET)
2904 			rt->rt_gw4 = ort->rt_gw4;
2905 		else if (rt->rt_gw_family == AF_INET6)
2906 			rt->rt_gw6 = ort->rt_gw6;
2907 	}
2908 
2909 	dst_release(dst_orig);
2910 
2911 	return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2912 }
2913 
2914 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2915 				    const struct sock *sk)
2916 {
2917 	struct rtable *rt = __ip_route_output_key(net, flp4);
2918 
2919 	if (IS_ERR(rt))
2920 		return rt;
2921 
2922 	if (flp4->flowi4_proto) {
2923 		flp4->flowi4_oif = rt->dst.dev->ifindex;
2924 		rt = dst_rtable(xfrm_lookup_route(net, &rt->dst,
2925 						  flowi4_to_flowi(flp4),
2926 						  sk, 0));
2927 	}
2928 
2929 	return rt;
2930 }
2931 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2932 
2933 /* called with rcu_read_lock held */
2934 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2935 			struct rtable *rt, u32 table_id, dscp_t dscp,
2936 			struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2937 			u32 seq, unsigned int flags)
2938 {
2939 	struct rtmsg *r;
2940 	struct nlmsghdr *nlh;
2941 	unsigned long expires = 0;
2942 	u32 error;
2943 	u32 metrics[RTAX_MAX];
2944 
2945 	nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2946 	if (!nlh)
2947 		return -EMSGSIZE;
2948 
2949 	r = nlmsg_data(nlh);
2950 	r->rtm_family	 = AF_INET;
2951 	r->rtm_dst_len	= 32;
2952 	r->rtm_src_len	= 0;
2953 	r->rtm_tos	= inet_dscp_to_dsfield(dscp);
2954 	r->rtm_table	= table_id < 256 ? table_id : RT_TABLE_COMPAT;
2955 	if (nla_put_u32(skb, RTA_TABLE, table_id))
2956 		goto nla_put_failure;
2957 	r->rtm_type	= rt->rt_type;
2958 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2959 	r->rtm_protocol = RTPROT_UNSPEC;
2960 	r->rtm_flags	= (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2961 	if (rt->rt_flags & RTCF_NOTIFY)
2962 		r->rtm_flags |= RTM_F_NOTIFY;
2963 	if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2964 		r->rtm_flags |= RTCF_DOREDIRECT;
2965 
2966 	if (nla_put_in_addr(skb, RTA_DST, dst))
2967 		goto nla_put_failure;
2968 	if (src) {
2969 		r->rtm_src_len = 32;
2970 		if (nla_put_in_addr(skb, RTA_SRC, src))
2971 			goto nla_put_failure;
2972 	}
2973 	if (rt->dst.dev &&
2974 	    nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2975 		goto nla_put_failure;
2976 	if (rt->dst.lwtstate &&
2977 	    lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2978 		goto nla_put_failure;
2979 #ifdef CONFIG_IP_ROUTE_CLASSID
2980 	if (rt->dst.tclassid &&
2981 	    nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2982 		goto nla_put_failure;
2983 #endif
2984 	if (fl4 && !rt_is_input_route(rt) &&
2985 	    fl4->saddr != src) {
2986 		if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2987 			goto nla_put_failure;
2988 	}
2989 	if (rt->rt_uses_gateway) {
2990 		if (rt->rt_gw_family == AF_INET &&
2991 		    nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2992 			goto nla_put_failure;
2993 		} else if (rt->rt_gw_family == AF_INET6) {
2994 			int alen = sizeof(struct in6_addr);
2995 			struct nlattr *nla;
2996 			struct rtvia *via;
2997 
2998 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
2999 			if (!nla)
3000 				goto nla_put_failure;
3001 
3002 			via = nla_data(nla);
3003 			via->rtvia_family = AF_INET6;
3004 			memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
3005 		}
3006 	}
3007 
3008 	expires = rt->dst.expires;
3009 	if (expires) {
3010 		unsigned long now = jiffies;
3011 
3012 		if (time_before(now, expires))
3013 			expires -= now;
3014 		else
3015 			expires = 0;
3016 	}
3017 
3018 	memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3019 	if (rt->rt_pmtu && expires)
3020 		metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3021 	if (rt->rt_mtu_locked && expires)
3022 		metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3023 	if (rtnetlink_put_metrics(skb, metrics) < 0)
3024 		goto nla_put_failure;
3025 
3026 	if (fl4) {
3027 		if (fl4->flowi4_mark &&
3028 		    nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3029 			goto nla_put_failure;
3030 
3031 		if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3032 		    nla_put_u32(skb, RTA_UID,
3033 				from_kuid_munged(current_user_ns(),
3034 						 fl4->flowi4_uid)))
3035 			goto nla_put_failure;
3036 
3037 		if (rt_is_input_route(rt)) {
3038 #ifdef CONFIG_IP_MROUTE
3039 			if (ipv4_is_multicast(dst) &&
3040 			    !ipv4_is_local_multicast(dst) &&
3041 			    IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) {
3042 				int err = ipmr_get_route(net, skb,
3043 							 fl4->saddr, fl4->daddr,
3044 							 r, portid);
3045 
3046 				if (err <= 0) {
3047 					if (err == 0)
3048 						return 0;
3049 					goto nla_put_failure;
3050 				}
3051 			} else
3052 #endif
3053 				if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3054 					goto nla_put_failure;
3055 		}
3056 	}
3057 
3058 	error = rt->dst.error;
3059 
3060 	if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3061 		goto nla_put_failure;
3062 
3063 	nlmsg_end(skb, nlh);
3064 	return 0;
3065 
3066 nla_put_failure:
3067 	nlmsg_cancel(skb, nlh);
3068 	return -EMSGSIZE;
3069 }
3070 
3071 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3072 			    struct netlink_callback *cb, u32 table_id,
3073 			    struct fnhe_hash_bucket *bucket, int genid,
3074 			    int *fa_index, int fa_start, unsigned int flags)
3075 {
3076 	int i;
3077 
3078 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
3079 		struct fib_nh_exception *fnhe;
3080 
3081 		for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3082 		     fnhe = rcu_dereference(fnhe->fnhe_next)) {
3083 			struct rtable *rt;
3084 			int err;
3085 
3086 			if (*fa_index < fa_start)
3087 				goto next;
3088 
3089 			if (fnhe->fnhe_genid != genid)
3090 				goto next;
3091 
3092 			if (fnhe->fnhe_expires &&
3093 			    time_after(jiffies, fnhe->fnhe_expires))
3094 				goto next;
3095 
3096 			rt = rcu_dereference(fnhe->fnhe_rth_input);
3097 			if (!rt)
3098 				rt = rcu_dereference(fnhe->fnhe_rth_output);
3099 			if (!rt)
3100 				goto next;
3101 
3102 			err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3103 					   table_id, 0, NULL, skb,
3104 					   NETLINK_CB(cb->skb).portid,
3105 					   cb->nlh->nlmsg_seq, flags);
3106 			if (err)
3107 				return err;
3108 next:
3109 			(*fa_index)++;
3110 		}
3111 	}
3112 
3113 	return 0;
3114 }
3115 
3116 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3117 		       u32 table_id, struct fib_info *fi,
3118 		       int *fa_index, int fa_start, unsigned int flags)
3119 {
3120 	struct net *net = sock_net(cb->skb->sk);
3121 	int nhsel, genid = fnhe_genid(net);
3122 
3123 	for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3124 		struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3125 		struct fnhe_hash_bucket *bucket;
3126 		int err;
3127 
3128 		if (nhc->nhc_flags & RTNH_F_DEAD)
3129 			continue;
3130 
3131 		rcu_read_lock();
3132 		bucket = rcu_dereference(nhc->nhc_exceptions);
3133 		err = 0;
3134 		if (bucket)
3135 			err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3136 					       genid, fa_index, fa_start,
3137 					       flags);
3138 		rcu_read_unlock();
3139 		if (err)
3140 			return err;
3141 	}
3142 
3143 	return 0;
3144 }
3145 
3146 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3147 						   u8 ip_proto, __be16 sport,
3148 						   __be16 dport)
3149 {
3150 	struct sk_buff *skb;
3151 	struct iphdr *iph;
3152 
3153 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3154 	if (!skb)
3155 		return NULL;
3156 
3157 	/* Reserve room for dummy headers, this skb can pass
3158 	 * through good chunk of routing engine.
3159 	 */
3160 	skb_reset_mac_header(skb);
3161 	skb_reset_network_header(skb);
3162 	skb->protocol = htons(ETH_P_IP);
3163 	iph = skb_put(skb, sizeof(struct iphdr));
3164 	iph->protocol = ip_proto;
3165 	iph->saddr = src;
3166 	iph->daddr = dst;
3167 	iph->version = 0x4;
3168 	iph->frag_off = 0;
3169 	iph->ihl = 0x5;
3170 	skb_set_transport_header(skb, skb->len);
3171 
3172 	switch (iph->protocol) {
3173 	case IPPROTO_UDP: {
3174 		struct udphdr *udph;
3175 
3176 		udph = skb_put_zero(skb, sizeof(struct udphdr));
3177 		udph->source = sport;
3178 		udph->dest = dport;
3179 		udph->len = htons(sizeof(struct udphdr));
3180 		udph->check = 0;
3181 		break;
3182 	}
3183 	case IPPROTO_TCP: {
3184 		struct tcphdr *tcph;
3185 
3186 		tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3187 		tcph->source	= sport;
3188 		tcph->dest	= dport;
3189 		tcph->doff	= sizeof(struct tcphdr) / 4;
3190 		tcph->rst = 1;
3191 		tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3192 					    src, dst, 0);
3193 		break;
3194 	}
3195 	case IPPROTO_ICMP: {
3196 		struct icmphdr *icmph;
3197 
3198 		icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3199 		icmph->type = ICMP_ECHO;
3200 		icmph->code = 0;
3201 	}
3202 	}
3203 
3204 	return skb;
3205 }
3206 
3207 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3208 				       const struct nlmsghdr *nlh,
3209 				       struct nlattr **tb,
3210 				       struct netlink_ext_ack *extack)
3211 {
3212 	struct rtmsg *rtm;
3213 	int i, err;
3214 
3215 	rtm = nlmsg_payload(nlh, sizeof(*rtm));
3216 	if (!rtm) {
3217 		NL_SET_ERR_MSG(extack,
3218 			       "ipv4: Invalid header for route get request");
3219 		return -EINVAL;
3220 	}
3221 
3222 	if (!netlink_strict_get_check(skb))
3223 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3224 					      rtm_ipv4_policy, extack);
3225 
3226 	if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3227 	    (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3228 	    rtm->rtm_table || rtm->rtm_protocol ||
3229 	    rtm->rtm_scope || rtm->rtm_type) {
3230 		NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3231 		return -EINVAL;
3232 	}
3233 
3234 	if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3235 			       RTM_F_LOOKUP_TABLE |
3236 			       RTM_F_FIB_MATCH)) {
3237 		NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3238 		return -EINVAL;
3239 	}
3240 
3241 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3242 					    rtm_ipv4_policy, extack);
3243 	if (err)
3244 		return err;
3245 
3246 	if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3247 	    (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3248 		NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3249 		return -EINVAL;
3250 	}
3251 
3252 	for (i = 0; i <= RTA_MAX; i++) {
3253 		if (!tb[i])
3254 			continue;
3255 
3256 		switch (i) {
3257 		case RTA_IIF:
3258 		case RTA_OIF:
3259 		case RTA_SRC:
3260 		case RTA_DST:
3261 		case RTA_IP_PROTO:
3262 		case RTA_SPORT:
3263 		case RTA_DPORT:
3264 		case RTA_MARK:
3265 		case RTA_UID:
3266 			break;
3267 		default:
3268 			NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3269 			return -EINVAL;
3270 		}
3271 	}
3272 
3273 	return 0;
3274 }
3275 
3276 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3277 			     struct netlink_ext_ack *extack)
3278 {
3279 	struct net *net = sock_net(in_skb->sk);
3280 	struct nlattr *tb[RTA_MAX+1];
3281 	u32 table_id = RT_TABLE_MAIN;
3282 	__be16 sport = 0, dport = 0;
3283 	struct fib_result res = {};
3284 	u8 ip_proto = IPPROTO_UDP;
3285 	struct rtable *rt = NULL;
3286 	struct sk_buff *skb;
3287 	struct rtmsg *rtm;
3288 	struct flowi4 fl4 = {};
3289 	__be32 dst = 0;
3290 	__be32 src = 0;
3291 	dscp_t dscp;
3292 	kuid_t uid;
3293 	u32 iif;
3294 	int err;
3295 	int mark;
3296 
3297 	err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3298 	if (err < 0)
3299 		return err;
3300 
3301 	rtm = nlmsg_data(nlh);
3302 	src = nla_get_in_addr_default(tb[RTA_SRC], 0);
3303 	dst = nla_get_in_addr_default(tb[RTA_DST], 0);
3304 	iif = nla_get_u32_default(tb[RTA_IIF], 0);
3305 	mark = nla_get_u32_default(tb[RTA_MARK], 0);
3306 	dscp = inet_dsfield_to_dscp(rtm->rtm_tos);
3307 	if (tb[RTA_UID])
3308 		uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3309 	else
3310 		uid = (iif ? INVALID_UID : current_uid());
3311 
3312 	if (tb[RTA_IP_PROTO]) {
3313 		err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3314 						  &ip_proto, AF_INET, extack);
3315 		if (err)
3316 			return err;
3317 	}
3318 
3319 	if (tb[RTA_SPORT])
3320 		sport = nla_get_be16(tb[RTA_SPORT]);
3321 
3322 	if (tb[RTA_DPORT])
3323 		dport = nla_get_be16(tb[RTA_DPORT]);
3324 
3325 	skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3326 	if (!skb)
3327 		return -ENOBUFS;
3328 
3329 	fl4.daddr = dst;
3330 	fl4.saddr = src;
3331 	fl4.flowi4_tos = inet_dscp_to_dsfield(dscp);
3332 	fl4.flowi4_oif = nla_get_u32_default(tb[RTA_OIF], 0);
3333 	fl4.flowi4_mark = mark;
3334 	fl4.flowi4_uid = uid;
3335 	if (sport)
3336 		fl4.fl4_sport = sport;
3337 	if (dport)
3338 		fl4.fl4_dport = dport;
3339 	fl4.flowi4_proto = ip_proto;
3340 
3341 	rcu_read_lock();
3342 
3343 	if (iif) {
3344 		struct net_device *dev;
3345 
3346 		dev = dev_get_by_index_rcu(net, iif);
3347 		if (!dev) {
3348 			err = -ENODEV;
3349 			goto errout_rcu;
3350 		}
3351 
3352 		fl4.flowi4_iif = iif; /* for rt_fill_info */
3353 		skb->dev	= dev;
3354 		skb->mark	= mark;
3355 		err = ip_route_input_rcu(skb, dst, src, dscp, dev,
3356 					 &res) ? -EINVAL : 0;
3357 
3358 		rt = skb_rtable(skb);
3359 		if (err == 0 && rt->dst.error)
3360 			err = -rt->dst.error;
3361 	} else {
3362 		fl4.flowi4_iif = LOOPBACK_IFINDEX;
3363 		skb->dev = net->loopback_dev;
3364 		rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3365 		err = 0;
3366 		if (IS_ERR(rt))
3367 			err = PTR_ERR(rt);
3368 		else
3369 			skb_dst_set(skb, &rt->dst);
3370 	}
3371 
3372 	if (err)
3373 		goto errout_rcu;
3374 
3375 	if (rtm->rtm_flags & RTM_F_NOTIFY)
3376 		rt->rt_flags |= RTCF_NOTIFY;
3377 
3378 	if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3379 		table_id = res.table ? res.table->tb_id : 0;
3380 
3381 	/* reset skb for netlink reply msg */
3382 	skb_trim(skb, 0);
3383 	skb_reset_network_header(skb);
3384 	skb_reset_transport_header(skb);
3385 	skb_reset_mac_header(skb);
3386 
3387 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3388 		struct fib_rt_info fri;
3389 
3390 		if (!res.fi) {
3391 			err = fib_props[res.type].error;
3392 			if (!err)
3393 				err = -EHOSTUNREACH;
3394 			goto errout_rcu;
3395 		}
3396 		fri.fi = res.fi;
3397 		fri.tb_id = table_id;
3398 		fri.dst = res.prefix;
3399 		fri.dst_len = res.prefixlen;
3400 		fri.dscp = res.dscp;
3401 		fri.type = rt->rt_type;
3402 		fri.offload = 0;
3403 		fri.trap = 0;
3404 		fri.offload_failed = 0;
3405 		if (res.fa_head) {
3406 			struct fib_alias *fa;
3407 
3408 			hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3409 				u8 slen = 32 - fri.dst_len;
3410 
3411 				if (fa->fa_slen == slen &&
3412 				    fa->tb_id == fri.tb_id &&
3413 				    fa->fa_dscp == fri.dscp &&
3414 				    fa->fa_info == res.fi &&
3415 				    fa->fa_type == fri.type) {
3416 					fri.offload = READ_ONCE(fa->offload);
3417 					fri.trap = READ_ONCE(fa->trap);
3418 					fri.offload_failed =
3419 						READ_ONCE(fa->offload_failed);
3420 					break;
3421 				}
3422 			}
3423 		}
3424 		err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3425 				    nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3426 	} else {
3427 		err = rt_fill_info(net, dst, src, rt, table_id, res.dscp, &fl4,
3428 				   skb, NETLINK_CB(in_skb).portid,
3429 				   nlh->nlmsg_seq, 0);
3430 	}
3431 	if (err < 0)
3432 		goto errout_rcu;
3433 
3434 	rcu_read_unlock();
3435 
3436 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3437 
3438 errout_free:
3439 	return err;
3440 errout_rcu:
3441 	rcu_read_unlock();
3442 	kfree_skb(skb);
3443 	goto errout_free;
3444 }
3445 
3446 void ip_rt_multicast_event(struct in_device *in_dev)
3447 {
3448 	rt_cache_flush(dev_net(in_dev->dev));
3449 }
3450 
3451 #ifdef CONFIG_SYSCTL
3452 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3453 static int ip_rt_gc_min_interval __read_mostly	= HZ / 2;
3454 static int ip_rt_gc_elasticity __read_mostly	= 8;
3455 static int ip_min_valid_pmtu __read_mostly	= IPV4_MIN_MTU;
3456 
3457 static int ipv4_sysctl_rtcache_flush(const struct ctl_table *__ctl, int write,
3458 		void *buffer, size_t *lenp, loff_t *ppos)
3459 {
3460 	struct net *net = (struct net *)__ctl->extra1;
3461 
3462 	if (write) {
3463 		rt_cache_flush(net);
3464 		fnhe_genid_bump(net);
3465 		return 0;
3466 	}
3467 
3468 	return -EINVAL;
3469 }
3470 
3471 static struct ctl_table ipv4_route_table[] = {
3472 	{
3473 		.procname	= "gc_thresh",
3474 		.data		= &ipv4_dst_ops.gc_thresh,
3475 		.maxlen		= sizeof(int),
3476 		.mode		= 0644,
3477 		.proc_handler	= proc_dointvec,
3478 	},
3479 	{
3480 		.procname	= "max_size",
3481 		.data		= &ip_rt_max_size,
3482 		.maxlen		= sizeof(int),
3483 		.mode		= 0644,
3484 		.proc_handler	= proc_dointvec,
3485 	},
3486 	{
3487 		/*  Deprecated. Use gc_min_interval_ms */
3488 
3489 		.procname	= "gc_min_interval",
3490 		.data		= &ip_rt_gc_min_interval,
3491 		.maxlen		= sizeof(int),
3492 		.mode		= 0644,
3493 		.proc_handler	= proc_dointvec_jiffies,
3494 	},
3495 	{
3496 		.procname	= "gc_min_interval_ms",
3497 		.data		= &ip_rt_gc_min_interval,
3498 		.maxlen		= sizeof(int),
3499 		.mode		= 0644,
3500 		.proc_handler	= proc_dointvec_ms_jiffies,
3501 	},
3502 	{
3503 		.procname	= "gc_timeout",
3504 		.data		= &ip_rt_gc_timeout,
3505 		.maxlen		= sizeof(int),
3506 		.mode		= 0644,
3507 		.proc_handler	= proc_dointvec_jiffies,
3508 	},
3509 	{
3510 		.procname	= "gc_interval",
3511 		.data		= &ip_rt_gc_interval,
3512 		.maxlen		= sizeof(int),
3513 		.mode		= 0644,
3514 		.proc_handler	= proc_dointvec_jiffies,
3515 	},
3516 	{
3517 		.procname	= "redirect_load",
3518 		.data		= &ip_rt_redirect_load,
3519 		.maxlen		= sizeof(int),
3520 		.mode		= 0644,
3521 		.proc_handler	= proc_dointvec,
3522 	},
3523 	{
3524 		.procname	= "redirect_number",
3525 		.data		= &ip_rt_redirect_number,
3526 		.maxlen		= sizeof(int),
3527 		.mode		= 0644,
3528 		.proc_handler	= proc_dointvec,
3529 	},
3530 	{
3531 		.procname	= "redirect_silence",
3532 		.data		= &ip_rt_redirect_silence,
3533 		.maxlen		= sizeof(int),
3534 		.mode		= 0644,
3535 		.proc_handler	= proc_dointvec,
3536 	},
3537 	{
3538 		.procname	= "error_cost",
3539 		.data		= &ip_rt_error_cost,
3540 		.maxlen		= sizeof(int),
3541 		.mode		= 0644,
3542 		.proc_handler	= proc_dointvec,
3543 	},
3544 	{
3545 		.procname	= "error_burst",
3546 		.data		= &ip_rt_error_burst,
3547 		.maxlen		= sizeof(int),
3548 		.mode		= 0644,
3549 		.proc_handler	= proc_dointvec,
3550 	},
3551 	{
3552 		.procname	= "gc_elasticity",
3553 		.data		= &ip_rt_gc_elasticity,
3554 		.maxlen		= sizeof(int),
3555 		.mode		= 0644,
3556 		.proc_handler	= proc_dointvec,
3557 	},
3558 };
3559 
3560 static const char ipv4_route_flush_procname[] = "flush";
3561 
3562 static struct ctl_table ipv4_route_netns_table[] = {
3563 	{
3564 		.procname	= ipv4_route_flush_procname,
3565 		.maxlen		= sizeof(int),
3566 		.mode		= 0200,
3567 		.proc_handler	= ipv4_sysctl_rtcache_flush,
3568 	},
3569 	{
3570 		.procname       = "min_pmtu",
3571 		.data           = &init_net.ipv4.ip_rt_min_pmtu,
3572 		.maxlen         = sizeof(int),
3573 		.mode           = 0644,
3574 		.proc_handler   = proc_dointvec_minmax,
3575 		.extra1         = &ip_min_valid_pmtu,
3576 	},
3577 	{
3578 		.procname       = "mtu_expires",
3579 		.data           = &init_net.ipv4.ip_rt_mtu_expires,
3580 		.maxlen         = sizeof(int),
3581 		.mode           = 0644,
3582 		.proc_handler   = proc_dointvec_jiffies,
3583 	},
3584 	{
3585 		.procname   = "min_adv_mss",
3586 		.data       = &init_net.ipv4.ip_rt_min_advmss,
3587 		.maxlen     = sizeof(int),
3588 		.mode       = 0644,
3589 		.proc_handler   = proc_dointvec,
3590 	},
3591 };
3592 
3593 static __net_init int sysctl_route_net_init(struct net *net)
3594 {
3595 	struct ctl_table *tbl;
3596 	size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3597 
3598 	tbl = ipv4_route_netns_table;
3599 	if (!net_eq(net, &init_net)) {
3600 		int i;
3601 
3602 		tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3603 		if (!tbl)
3604 			goto err_dup;
3605 
3606 		/* Don't export non-whitelisted sysctls to unprivileged users */
3607 		if (net->user_ns != &init_user_ns) {
3608 			if (tbl[0].procname != ipv4_route_flush_procname)
3609 				table_size = 0;
3610 		}
3611 
3612 		/* Update the variables to point into the current struct net
3613 		 * except for the first element flush
3614 		 */
3615 		for (i = 1; i < table_size; i++)
3616 			tbl[i].data += (void *)net - (void *)&init_net;
3617 	}
3618 	tbl[0].extra1 = net;
3619 
3620 	net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3621 						     tbl, table_size);
3622 	if (!net->ipv4.route_hdr)
3623 		goto err_reg;
3624 	return 0;
3625 
3626 err_reg:
3627 	if (tbl != ipv4_route_netns_table)
3628 		kfree(tbl);
3629 err_dup:
3630 	return -ENOMEM;
3631 }
3632 
3633 static __net_exit void sysctl_route_net_exit(struct net *net)
3634 {
3635 	const struct ctl_table *tbl;
3636 
3637 	tbl = net->ipv4.route_hdr->ctl_table_arg;
3638 	unregister_net_sysctl_table(net->ipv4.route_hdr);
3639 	BUG_ON(tbl == ipv4_route_netns_table);
3640 	kfree(tbl);
3641 }
3642 
3643 static __net_initdata struct pernet_operations sysctl_route_ops = {
3644 	.init = sysctl_route_net_init,
3645 	.exit = sysctl_route_net_exit,
3646 };
3647 #endif
3648 
3649 static __net_init int netns_ip_rt_init(struct net *net)
3650 {
3651 	/* Set default value for namespaceified sysctls */
3652 	net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3653 	net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3654 	net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3655 	return 0;
3656 }
3657 
3658 static struct pernet_operations __net_initdata ip_rt_ops = {
3659 	.init = netns_ip_rt_init,
3660 };
3661 
3662 static __net_init int rt_genid_init(struct net *net)
3663 {
3664 	atomic_set(&net->ipv4.rt_genid, 0);
3665 	atomic_set(&net->fnhe_genid, 0);
3666 	atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3667 	return 0;
3668 }
3669 
3670 static __net_initdata struct pernet_operations rt_genid_ops = {
3671 	.init = rt_genid_init,
3672 };
3673 
3674 static int __net_init ipv4_inetpeer_init(struct net *net)
3675 {
3676 	struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3677 
3678 	if (!bp)
3679 		return -ENOMEM;
3680 	inet_peer_base_init(bp);
3681 	net->ipv4.peers = bp;
3682 	return 0;
3683 }
3684 
3685 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3686 {
3687 	struct inet_peer_base *bp = net->ipv4.peers;
3688 
3689 	net->ipv4.peers = NULL;
3690 	inetpeer_invalidate_tree(bp);
3691 	kfree(bp);
3692 }
3693 
3694 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3695 	.init	=	ipv4_inetpeer_init,
3696 	.exit	=	ipv4_inetpeer_exit,
3697 };
3698 
3699 #ifdef CONFIG_IP_ROUTE_CLASSID
3700 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3701 #endif /* CONFIG_IP_ROUTE_CLASSID */
3702 
3703 static const struct rtnl_msg_handler ip_rt_rtnl_msg_handlers[] __initconst = {
3704 	{.protocol = PF_INET, .msgtype = RTM_GETROUTE,
3705 	 .doit = inet_rtm_getroute, .flags = RTNL_FLAG_DOIT_UNLOCKED},
3706 };
3707 
3708 int __init ip_rt_init(void)
3709 {
3710 	void *idents_hash;
3711 	int cpu;
3712 
3713 	/* For modern hosts, this will use 2 MB of memory */
3714 	idents_hash = alloc_large_system_hash("IP idents",
3715 					      sizeof(*ip_idents) + sizeof(*ip_tstamps),
3716 					      0,
3717 					      16, /* one bucket per 64 KB */
3718 					      HASH_ZERO,
3719 					      NULL,
3720 					      &ip_idents_mask,
3721 					      2048,
3722 					      256*1024);
3723 
3724 	ip_idents = idents_hash;
3725 
3726 	get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3727 
3728 	ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3729 
3730 	for_each_possible_cpu(cpu) {
3731 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3732 
3733 		INIT_LIST_HEAD(&ul->head);
3734 		spin_lock_init(&ul->lock);
3735 	}
3736 #ifdef CONFIG_IP_ROUTE_CLASSID
3737 	ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3738 	if (!ip_rt_acct)
3739 		panic("IP: failed to allocate ip_rt_acct\n");
3740 #endif
3741 
3742 	ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable,
3743 					      SLAB_HWCACHE_ALIGN | SLAB_PANIC);
3744 
3745 	ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3746 
3747 	if (dst_entries_init(&ipv4_dst_ops) < 0)
3748 		panic("IP: failed to allocate ipv4_dst_ops counter\n");
3749 
3750 	if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3751 		panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3752 
3753 	ipv4_dst_ops.gc_thresh = ~0;
3754 	ip_rt_max_size = INT_MAX;
3755 
3756 	devinet_init();
3757 	ip_fib_init();
3758 
3759 	if (ip_rt_proc_init())
3760 		pr_err("Unable to create route proc files\n");
3761 #ifdef CONFIG_XFRM
3762 	xfrm_init();
3763 	xfrm4_init();
3764 #endif
3765 	rtnl_register_many(ip_rt_rtnl_msg_handlers);
3766 
3767 #ifdef CONFIG_SYSCTL
3768 	register_pernet_subsys(&sysctl_route_ops);
3769 #endif
3770 	register_pernet_subsys(&ip_rt_ops);
3771 	register_pernet_subsys(&rt_genid_ops);
3772 	register_pernet_subsys(&ipv4_inetpeer_ops);
3773 	return 0;
3774 }
3775 
3776 #ifdef CONFIG_SYSCTL
3777 /*
3778  * We really need to sanitize the damn ipv4 init order, then all
3779  * this nonsense will go away.
3780  */
3781 void __init ip_static_sysctl_init(void)
3782 {
3783 	register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3784 }
3785 #endif
3786