xref: /linux/net/ipv4/route.c (revision 87579b8cda9ec6ecba8327c59d8505d3b83fda98)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		ROUTE - implementation of the IP router.
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
12  *		Linus Torvalds, <Linus.Torvalds@helsinki.fi>
13  *		Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
14  *
15  * Fixes:
16  *		Alan Cox	:	Verify area fixes.
17  *		Alan Cox	:	cli() protects routing changes
18  *		Rui Oliveira	:	ICMP routing table updates
19  *		(rco@di.uminho.pt)	Routing table insertion and update
20  *		Linus Torvalds	:	Rewrote bits to be sensible
21  *		Alan Cox	:	Added BSD route gw semantics
22  *		Alan Cox	:	Super /proc >4K
23  *		Alan Cox	:	MTU in route table
24  *		Alan Cox	:	MSS actually. Also added the window
25  *					clamper.
26  *		Sam Lantinga	:	Fixed route matching in rt_del()
27  *		Alan Cox	:	Routing cache support.
28  *		Alan Cox	:	Removed compatibility cruft.
29  *		Alan Cox	:	RTF_REJECT support.
30  *		Alan Cox	:	TCP irtt support.
31  *		Jonathan Naylor	:	Added Metric support.
32  *	Miquel van Smoorenburg	:	BSD API fixes.
33  *	Miquel van Smoorenburg	:	Metrics.
34  *		Alan Cox	:	Use __u32 properly
35  *		Alan Cox	:	Aligned routing errors more closely with BSD
36  *					our system is still very different.
37  *		Alan Cox	:	Faster /proc handling
38  *	Alexey Kuznetsov	:	Massive rework to support tree based routing,
39  *					routing caches and better behaviour.
40  *
41  *		Olaf Erb	:	irtt wasn't being copied right.
42  *		Bjorn Ekwall	:	Kerneld route support.
43  *		Alan Cox	:	Multicast fixed (I hope)
44  *		Pavel Krauz	:	Limited broadcast fixed
45  *		Mike McLagan	:	Routing by source
46  *	Alexey Kuznetsov	:	End of old history. Split to fib.c and
47  *					route.c and rewritten from scratch.
48  *		Andi Kleen	:	Load-limit warning messages.
49  *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
50  *	Vitaly E. Lavrov	:	Race condition in ip_route_input_slow.
51  *	Tobias Ringstrom	:	Uninitialized res.type in ip_route_output_slow.
52  *	Vladimir V. Ivanov	:	IP rule info (flowid) is really useful.
53  *		Marc Boucher	:	routing by fwmark
54  *	Robert Olsson		:	Added rt_cache statistics
55  *	Arnaldo C. Melo		:	Convert proc stuff to seq_file
56  *	Eric Dumazet		:	hashed spinlocks and rt_check_expire() fixes.
57  *	Ilia Sotnikov		:	Ignore TOS on PMTUD and Redirect
58  *	Ilia Sotnikov		:	Removed TOS from hash calculations
59  */
60 
61 #define pr_fmt(fmt) "IPv4: " fmt
62 
63 #include <linux/module.h>
64 #include <linux/bitops.h>
65 #include <linux/kernel.h>
66 #include <linux/mm.h>
67 #include <linux/memblock.h>
68 #include <linux/socket.h>
69 #include <linux/errno.h>
70 #include <linux/in.h>
71 #include <linux/inet.h>
72 #include <linux/netdevice.h>
73 #include <linux/proc_fs.h>
74 #include <linux/init.h>
75 #include <linux/skbuff.h>
76 #include <linux/inetdevice.h>
77 #include <linux/igmp.h>
78 #include <linux/pkt_sched.h>
79 #include <linux/mroute.h>
80 #include <linux/netfilter_ipv4.h>
81 #include <linux/random.h>
82 #include <linux/rcupdate.h>
83 #include <linux/slab.h>
84 #include <linux/jhash.h>
85 #include <net/dst.h>
86 #include <net/dst_metadata.h>
87 #include <net/flow.h>
88 #include <net/inet_dscp.h>
89 #include <net/net_namespace.h>
90 #include <net/ip.h>
91 #include <net/route.h>
92 #include <net/inetpeer.h>
93 #include <net/sock.h>
94 #include <net/ip_fib.h>
95 #include <net/nexthop.h>
96 #include <net/tcp.h>
97 #include <net/icmp.h>
98 #include <net/xfrm.h>
99 #include <net/lwtunnel.h>
100 #include <net/netevent.h>
101 #include <net/rtnetlink.h>
102 #ifdef CONFIG_SYSCTL
103 #include <linux/sysctl.h>
104 #endif
105 #include <net/secure_seq.h>
106 #include <net/ip_tunnels.h>
107 
108 #include "fib_lookup.h"
109 
110 #define RT_GC_TIMEOUT (300*HZ)
111 
112 #define DEFAULT_MIN_PMTU (512 + 20 + 20)
113 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
114 #define DEFAULT_MIN_ADVMSS 256
115 static int ip_rt_max_size;
116 static int ip_rt_redirect_number __read_mostly	= 9;
117 static int ip_rt_redirect_load __read_mostly	= HZ / 50;
118 static int ip_rt_redirect_silence __read_mostly	= ((HZ / 50) << (9 + 1));
119 static int ip_rt_error_cost __read_mostly	= HZ;
120 static int ip_rt_error_burst __read_mostly	= 5 * HZ;
121 
122 static int ip_rt_gc_timeout __read_mostly	= RT_GC_TIMEOUT;
123 
124 /*
125  *	Interface to generic destination cache.
126  */
127 
128 INDIRECT_CALLABLE_SCOPE
129 struct dst_entry	*ipv4_dst_check(struct dst_entry *dst, u32 cookie);
130 static unsigned int	 ipv4_default_advmss(const struct dst_entry *dst);
131 INDIRECT_CALLABLE_SCOPE
132 unsigned int		ipv4_mtu(const struct dst_entry *dst);
133 static void		ipv4_negative_advice(struct sock *sk,
134 					     struct dst_entry *dst);
135 static void		 ipv4_link_failure(struct sk_buff *skb);
136 static void		 ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
137 					   struct sk_buff *skb, u32 mtu,
138 					   bool confirm_neigh);
139 static void		 ip_do_redirect(struct dst_entry *dst, struct sock *sk,
140 					struct sk_buff *skb);
141 static void		ipv4_dst_destroy(struct dst_entry *dst);
142 
143 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
144 {
145 	WARN_ON(1);
146 	return NULL;
147 }
148 
149 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
150 					   struct sk_buff *skb,
151 					   const void *daddr);
152 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
153 
154 static struct dst_ops ipv4_dst_ops = {
155 	.family =		AF_INET,
156 	.check =		ipv4_dst_check,
157 	.default_advmss =	ipv4_default_advmss,
158 	.mtu =			ipv4_mtu,
159 	.cow_metrics =		ipv4_cow_metrics,
160 	.destroy =		ipv4_dst_destroy,
161 	.negative_advice =	ipv4_negative_advice,
162 	.link_failure =		ipv4_link_failure,
163 	.update_pmtu =		ip_rt_update_pmtu,
164 	.redirect =		ip_do_redirect,
165 	.local_out =		__ip_local_out,
166 	.neigh_lookup =		ipv4_neigh_lookup,
167 	.confirm_neigh =	ipv4_confirm_neigh,
168 };
169 
170 #define ECN_OR_COST(class)	TC_PRIO_##class
171 
172 const __u8 ip_tos2prio[16] = {
173 	TC_PRIO_BESTEFFORT,
174 	ECN_OR_COST(BESTEFFORT),
175 	TC_PRIO_BESTEFFORT,
176 	ECN_OR_COST(BESTEFFORT),
177 	TC_PRIO_BULK,
178 	ECN_OR_COST(BULK),
179 	TC_PRIO_BULK,
180 	ECN_OR_COST(BULK),
181 	TC_PRIO_INTERACTIVE,
182 	ECN_OR_COST(INTERACTIVE),
183 	TC_PRIO_INTERACTIVE,
184 	ECN_OR_COST(INTERACTIVE),
185 	TC_PRIO_INTERACTIVE_BULK,
186 	ECN_OR_COST(INTERACTIVE_BULK),
187 	TC_PRIO_INTERACTIVE_BULK,
188 	ECN_OR_COST(INTERACTIVE_BULK)
189 };
190 EXPORT_SYMBOL(ip_tos2prio);
191 
192 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
193 #ifndef CONFIG_PREEMPT_RT
194 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
195 #else
196 #define RT_CACHE_STAT_INC(field) this_cpu_inc(rt_cache_stat.field)
197 #endif
198 
199 #ifdef CONFIG_PROC_FS
200 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
201 {
202 	if (*pos)
203 		return NULL;
204 	return SEQ_START_TOKEN;
205 }
206 
207 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
208 {
209 	++*pos;
210 	return NULL;
211 }
212 
213 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
214 {
215 }
216 
217 static int rt_cache_seq_show(struct seq_file *seq, void *v)
218 {
219 	if (v == SEQ_START_TOKEN)
220 		seq_printf(seq, "%-127s\n",
221 			   "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
222 			   "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
223 			   "HHUptod\tSpecDst");
224 	return 0;
225 }
226 
227 static const struct seq_operations rt_cache_seq_ops = {
228 	.start  = rt_cache_seq_start,
229 	.next   = rt_cache_seq_next,
230 	.stop   = rt_cache_seq_stop,
231 	.show   = rt_cache_seq_show,
232 };
233 
234 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
235 {
236 	int cpu;
237 
238 	if (*pos == 0)
239 		return SEQ_START_TOKEN;
240 
241 	for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
242 		if (!cpu_possible(cpu))
243 			continue;
244 		*pos = cpu+1;
245 		return &per_cpu(rt_cache_stat, cpu);
246 	}
247 	return NULL;
248 }
249 
250 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
251 {
252 	int cpu;
253 
254 	for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
255 		if (!cpu_possible(cpu))
256 			continue;
257 		*pos = cpu+1;
258 		return &per_cpu(rt_cache_stat, cpu);
259 	}
260 	(*pos)++;
261 	return NULL;
262 
263 }
264 
265 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
266 {
267 
268 }
269 
270 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
271 {
272 	struct rt_cache_stat *st = v;
273 
274 	if (v == SEQ_START_TOKEN) {
275 		seq_puts(seq, "entries  in_hit   in_slow_tot in_slow_mc in_no_route in_brd   in_martian_dst in_martian_src out_hit  out_slow_tot out_slow_mc gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
276 		return 0;
277 	}
278 
279 	seq_printf(seq, "%08x %08x %08x    %08x   %08x    %08x %08x       "
280 			"%08x       %08x %08x     %08x    %08x %08x   "
281 			"%08x     %08x        %08x        %08x\n",
282 		   dst_entries_get_slow(&ipv4_dst_ops),
283 		   0, /* st->in_hit */
284 		   st->in_slow_tot,
285 		   st->in_slow_mc,
286 		   st->in_no_route,
287 		   st->in_brd,
288 		   st->in_martian_dst,
289 		   st->in_martian_src,
290 
291 		   0, /* st->out_hit */
292 		   st->out_slow_tot,
293 		   st->out_slow_mc,
294 
295 		   0, /* st->gc_total */
296 		   0, /* st->gc_ignored */
297 		   0, /* st->gc_goal_miss */
298 		   0, /* st->gc_dst_overflow */
299 		   0, /* st->in_hlist_search */
300 		   0  /* st->out_hlist_search */
301 		);
302 	return 0;
303 }
304 
305 static const struct seq_operations rt_cpu_seq_ops = {
306 	.start  = rt_cpu_seq_start,
307 	.next   = rt_cpu_seq_next,
308 	.stop   = rt_cpu_seq_stop,
309 	.show   = rt_cpu_seq_show,
310 };
311 
312 #ifdef CONFIG_IP_ROUTE_CLASSID
313 static int rt_acct_proc_show(struct seq_file *m, void *v)
314 {
315 	struct ip_rt_acct *dst, *src;
316 	unsigned int i, j;
317 
318 	dst = kzalloc_objs(struct ip_rt_acct, 256);
319 	if (!dst)
320 		return -ENOMEM;
321 
322 	for_each_possible_cpu(i) {
323 		src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
324 		for (j = 0; j < 256; j++) {
325 			dst[j].o_bytes   += src[j].o_bytes;
326 			dst[j].o_packets += src[j].o_packets;
327 			dst[j].i_bytes   += src[j].i_bytes;
328 			dst[j].i_packets += src[j].i_packets;
329 		}
330 	}
331 
332 	seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
333 	kfree(dst);
334 	return 0;
335 }
336 #endif
337 
338 static int __net_init ip_rt_do_proc_init(struct net *net)
339 {
340 	struct proc_dir_entry *pde;
341 
342 	pde = proc_create_seq("rt_cache", 0444, net->proc_net,
343 			      &rt_cache_seq_ops);
344 	if (!pde)
345 		goto err1;
346 
347 	pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
348 			      &rt_cpu_seq_ops);
349 	if (!pde)
350 		goto err2;
351 
352 #ifdef CONFIG_IP_ROUTE_CLASSID
353 	pde = proc_create_single("rt_acct", 0, net->proc_net,
354 			rt_acct_proc_show);
355 	if (!pde)
356 		goto err3;
357 #endif
358 	return 0;
359 
360 #ifdef CONFIG_IP_ROUTE_CLASSID
361 err3:
362 	remove_proc_entry("rt_cache", net->proc_net_stat);
363 #endif
364 err2:
365 	remove_proc_entry("rt_cache", net->proc_net);
366 err1:
367 	return -ENOMEM;
368 }
369 
370 static void __net_exit ip_rt_do_proc_exit(struct net *net)
371 {
372 	remove_proc_entry("rt_cache", net->proc_net_stat);
373 	remove_proc_entry("rt_cache", net->proc_net);
374 #ifdef CONFIG_IP_ROUTE_CLASSID
375 	remove_proc_entry("rt_acct", net->proc_net);
376 #endif
377 }
378 
379 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
380 	.init = ip_rt_do_proc_init,
381 	.exit = ip_rt_do_proc_exit,
382 };
383 
384 static int __init ip_rt_proc_init(void)
385 {
386 	return register_pernet_subsys(&ip_rt_proc_ops);
387 }
388 
389 #else
390 static inline int ip_rt_proc_init(void)
391 {
392 	return 0;
393 }
394 #endif /* CONFIG_PROC_FS */
395 
396 static inline bool rt_is_expired(const struct rtable *rth)
397 {
398 	bool res;
399 
400 	rcu_read_lock();
401 	res = rth->rt_genid != rt_genid_ipv4(dev_net_rcu(rth->dst.dev));
402 	rcu_read_unlock();
403 
404 	return res;
405 }
406 
407 void rt_cache_flush(struct net *net)
408 {
409 	rt_genid_bump_ipv4(net);
410 }
411 
412 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
413 					   struct sk_buff *skb,
414 					   const void *daddr)
415 {
416 	const struct rtable *rt = container_of(dst, struct rtable, dst);
417 	struct net_device *dev;
418 	struct neighbour *n;
419 
420 	rcu_read_lock();
421 	dev = dst_dev_rcu(dst);
422 	if (likely(rt->rt_gw_family == AF_INET)) {
423 		n = ip_neigh_gw4(dev, rt->rt_gw4);
424 	} else if (rt->rt_gw_family == AF_INET6) {
425 		n = ip_neigh_gw6(dev, &rt->rt_gw6);
426         } else {
427 		__be32 pkey;
428 
429 		pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
430 		n = ip_neigh_gw4(dev, pkey);
431 	}
432 
433 	if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
434 		n = NULL;
435 
436 	rcu_read_unlock();
437 
438 	return n;
439 }
440 
441 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
442 {
443 	const struct rtable *rt = container_of(dst, struct rtable, dst);
444 	struct net_device *dev = dst_dev(dst);
445 	const __be32 *pkey = daddr;
446 
447 	if (rt->rt_gw_family == AF_INET) {
448 		pkey = (const __be32 *)&rt->rt_gw4;
449 	} else if (IS_ENABLED(CONFIG_IPV6) && rt->rt_gw_family == AF_INET6) {
450 		return __ipv6_confirm_neigh(dev, &rt->rt_gw6);
451 	} else if (!daddr ||
452 		 (rt->rt_flags &
453 		  (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
454 		return;
455 	}
456 	__ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
457 }
458 
459 /* Hash tables of size 2048..262144 depending on RAM size.
460  * Each bucket uses 8 bytes.
461  */
462 static u32 ip_idents_mask __read_mostly;
463 static atomic_t *ip_idents __read_mostly;
464 static u32 *ip_tstamps __read_mostly;
465 
466 /* In order to protect privacy, we add a perturbation to identifiers
467  * if one generator is seldom used. This makes hard for an attacker
468  * to infer how many packets were sent between two points in time.
469  */
470 static u32 ip_idents_reserve(u32 hash, int segs)
471 {
472 	u32 bucket, old, now = (u32)jiffies;
473 	atomic_t *p_id;
474 	u32 *p_tstamp;
475 	u32 delta = 0;
476 
477 	bucket = hash & ip_idents_mask;
478 	p_tstamp = ip_tstamps + bucket;
479 	p_id = ip_idents + bucket;
480 	old = READ_ONCE(*p_tstamp);
481 
482 	if (old != now && cmpxchg(p_tstamp, old, now) == old)
483 		delta = get_random_u32_below(now - old);
484 
485 	/* If UBSAN reports an error there, please make sure your compiler
486 	 * supports -fno-strict-overflow before reporting it that was a bug
487 	 * in UBSAN, and it has been fixed in GCC-8.
488 	 */
489 	return atomic_add_return(segs + delta, p_id) - segs;
490 }
491 
492 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
493 {
494 	u32 hash, id;
495 
496 	/* Note the following code is not safe, but this is okay. */
497 	if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
498 		get_random_bytes(&net->ipv4.ip_id_key,
499 				 sizeof(net->ipv4.ip_id_key));
500 
501 	hash = siphash_3u32((__force u32)iph->daddr,
502 			    (__force u32)iph->saddr,
503 			    iph->protocol,
504 			    &net->ipv4.ip_id_key);
505 	id = ip_idents_reserve(hash, segs);
506 	iph->id = htons(id);
507 }
508 EXPORT_SYMBOL(__ip_select_ident);
509 
510 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
511 			     const struct sock *sk, const struct iphdr *iph,
512 			     int oif, __u8 tos, u8 prot, u32 mark,
513 			     int flow_flags)
514 {
515 	__u8 scope = RT_SCOPE_UNIVERSE;
516 
517 	if (sk) {
518 		oif = sk->sk_bound_dev_if;
519 		mark = READ_ONCE(sk->sk_mark);
520 		tos = ip_sock_rt_tos(sk);
521 		scope = ip_sock_rt_scope(sk);
522 		prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW :
523 						    sk->sk_protocol;
524 	}
525 
526 	flowi4_init_output(fl4, oif, mark, tos & INET_DSCP_MASK, scope,
527 			   prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
528 			   sock_net_uid(net, sk));
529 }
530 
531 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
532 			       const struct sock *sk)
533 {
534 	const struct net *net = dev_net(skb->dev);
535 	const struct iphdr *iph = ip_hdr(skb);
536 	int oif = skb->dev->ifindex;
537 	u8 prot = iph->protocol;
538 	u32 mark = skb->mark;
539 	__u8 tos = iph->tos;
540 
541 	__build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
542 }
543 
544 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
545 {
546 	const struct inet_sock *inet = inet_sk(sk);
547 	const struct ip_options_rcu *inet_opt;
548 	__be32 daddr = inet->inet_daddr;
549 
550 	rcu_read_lock();
551 	inet_opt = rcu_dereference(inet->inet_opt);
552 	if (inet_opt && inet_opt->opt.srr)
553 		daddr = inet_opt->opt.faddr;
554 	flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark),
555 			   ip_sock_rt_tos(sk),
556 			   ip_sock_rt_scope(sk),
557 			   inet_test_bit(HDRINCL, sk) ?
558 				IPPROTO_RAW : sk->sk_protocol,
559 			   inet_sk_flowi_flags(sk),
560 			   daddr, inet->inet_saddr, 0, 0,
561 			   sk_uid(sk));
562 	rcu_read_unlock();
563 }
564 
565 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
566 				 const struct sk_buff *skb)
567 {
568 	if (skb)
569 		build_skb_flow_key(fl4, skb, sk);
570 	else
571 		build_sk_flow_key(fl4, sk);
572 }
573 
574 static DEFINE_SPINLOCK(fnhe_lock);
575 
576 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
577 {
578 	struct rtable *rt;
579 
580 	rt = rcu_dereference(fnhe->fnhe_rth_input);
581 	if (rt) {
582 		RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
583 		dst_dev_put(&rt->dst);
584 		dst_release(&rt->dst);
585 	}
586 	rt = rcu_dereference(fnhe->fnhe_rth_output);
587 	if (rt) {
588 		RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
589 		dst_dev_put(&rt->dst);
590 		dst_release(&rt->dst);
591 	}
592 }
593 
594 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
595 {
596 	struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
597 	struct fib_nh_exception *fnhe, *oldest = NULL;
598 
599 	for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
600 		fnhe = rcu_dereference_protected(*fnhe_p,
601 						 lockdep_is_held(&fnhe_lock));
602 		if (!fnhe)
603 			break;
604 		if (!oldest ||
605 		    time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
606 			oldest = fnhe;
607 			oldest_p = fnhe_p;
608 		}
609 	}
610 
611 	/* Clear oldest->fnhe_daddr to prevent this fnhe from being
612 	 * rebound with new dsts in rt_bind_exception().
613 	 */
614 	oldest->fnhe_daddr = 0;
615 	fnhe_flush_routes(oldest);
616 	*oldest_p = oldest->fnhe_next;
617 	kfree_rcu(oldest, rcu);
618 }
619 
620 static u32 fnhe_hashfun(__be32 daddr)
621 {
622 	static siphash_aligned_key_t fnhe_hash_key;
623 	u64 hval;
624 
625 	net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
626 	hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
627 	return hash_64(hval, FNHE_HASH_SHIFT);
628 }
629 
630 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
631 {
632 	rt->rt_pmtu = fnhe->fnhe_pmtu;
633 	rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
634 	rt->dst.expires = fnhe->fnhe_expires;
635 
636 	if (fnhe->fnhe_gw) {
637 		rt->rt_flags |= RTCF_REDIRECTED;
638 		rt->rt_uses_gateway = 1;
639 		rt->rt_gw_family = AF_INET;
640 		rt->rt_gw4 = fnhe->fnhe_gw;
641 	}
642 }
643 
644 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
645 				  __be32 gw, u32 pmtu, bool lock,
646 				  unsigned long expires)
647 {
648 	struct fnhe_hash_bucket *hash;
649 	struct fib_nh_exception *fnhe;
650 	struct rtable *rt;
651 	u32 genid, hval;
652 	unsigned int i;
653 	int depth;
654 
655 	genid = fnhe_genid(dev_net(nhc->nhc_dev));
656 	hval = fnhe_hashfun(daddr);
657 
658 	spin_lock_bh(&fnhe_lock);
659 
660 	hash = rcu_dereference(nhc->nhc_exceptions);
661 	if (!hash) {
662 		hash = kzalloc_objs(*hash, FNHE_HASH_SIZE, GFP_ATOMIC);
663 		if (!hash)
664 			goto out_unlock;
665 		rcu_assign_pointer(nhc->nhc_exceptions, hash);
666 	}
667 
668 	hash += hval;
669 
670 	depth = 0;
671 	for (fnhe = rcu_dereference(hash->chain); fnhe;
672 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
673 		if (fnhe->fnhe_daddr == daddr)
674 			break;
675 		depth++;
676 	}
677 
678 	if (fnhe) {
679 		if (fnhe->fnhe_genid != genid)
680 			fnhe->fnhe_genid = genid;
681 		if (gw)
682 			fnhe->fnhe_gw = gw;
683 		if (pmtu) {
684 			fnhe->fnhe_pmtu = pmtu;
685 			fnhe->fnhe_mtu_locked = lock;
686 		}
687 		fnhe->fnhe_expires = max(1UL, expires);
688 		/* Update all cached dsts too */
689 		rt = rcu_dereference(fnhe->fnhe_rth_input);
690 		if (rt)
691 			fill_route_from_fnhe(rt, fnhe);
692 		rt = rcu_dereference(fnhe->fnhe_rth_output);
693 		if (rt)
694 			fill_route_from_fnhe(rt, fnhe);
695 	} else {
696 		/* Randomize max depth to avoid some side channels attacks. */
697 		int max_depth = FNHE_RECLAIM_DEPTH +
698 				get_random_u32_below(FNHE_RECLAIM_DEPTH);
699 
700 		while (depth > max_depth) {
701 			fnhe_remove_oldest(hash);
702 			depth--;
703 		}
704 
705 		fnhe = kzalloc_obj(*fnhe, GFP_ATOMIC);
706 		if (!fnhe)
707 			goto out_unlock;
708 
709 		fnhe->fnhe_next = hash->chain;
710 
711 		fnhe->fnhe_genid = genid;
712 		fnhe->fnhe_daddr = daddr;
713 		fnhe->fnhe_gw = gw;
714 		fnhe->fnhe_pmtu = pmtu;
715 		fnhe->fnhe_mtu_locked = lock;
716 		fnhe->fnhe_expires = max(1UL, expires);
717 
718 		rcu_assign_pointer(hash->chain, fnhe);
719 
720 		/* Exception created; mark the cached routes for the nexthop
721 		 * stale, so anyone caching it rechecks if this exception
722 		 * applies to them.
723 		 */
724 		rt = rcu_dereference(nhc->nhc_rth_input);
725 		if (rt)
726 			WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL);
727 
728 		for_each_possible_cpu(i) {
729 			struct rtable __rcu **prt;
730 
731 			prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
732 			rt = rcu_dereference(*prt);
733 			if (rt)
734 				WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL);
735 		}
736 	}
737 
738 	fnhe->fnhe_stamp = jiffies;
739 
740 out_unlock:
741 	spin_unlock_bh(&fnhe_lock);
742 }
743 
744 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
745 			     bool kill_route)
746 {
747 	__be32 new_gw = icmp_hdr(skb)->un.gateway;
748 	__be32 old_gw = ip_hdr(skb)->saddr;
749 	struct net_device *dev = skb->dev;
750 	struct in_device *in_dev;
751 	struct fib_result res;
752 	struct neighbour *n;
753 	struct net *net;
754 
755 	switch (icmp_hdr(skb)->code & 7) {
756 	case ICMP_REDIR_NET:
757 	case ICMP_REDIR_NETTOS:
758 	case ICMP_REDIR_HOST:
759 	case ICMP_REDIR_HOSTTOS:
760 		break;
761 
762 	default:
763 		return;
764 	}
765 
766 	if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
767 		return;
768 
769 	in_dev = __in_dev_get_rcu(dev);
770 	if (!in_dev)
771 		return;
772 
773 	net = dev_net(dev);
774 	if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
775 	    ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
776 	    ipv4_is_zeronet(new_gw))
777 		goto reject_redirect;
778 
779 	if (!IN_DEV_SHARED_MEDIA(in_dev)) {
780 		if (!inet_addr_onlink(in_dev, new_gw, old_gw))
781 			goto reject_redirect;
782 		if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
783 			goto reject_redirect;
784 	} else {
785 		if (inet_addr_type(net, new_gw) != RTN_UNICAST)
786 			goto reject_redirect;
787 	}
788 
789 	n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw);
790 	if (!n)
791 		n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
792 	if (!IS_ERR(n)) {
793 		if (!(READ_ONCE(n->nud_state) & NUD_VALID)) {
794 			neigh_event_send(n, NULL);
795 		} else {
796 			if (fib_lookup(net, fl4, &res, 0) == 0) {
797 				struct fib_nh_common *nhc;
798 
799 				fib_select_path(net, &res, fl4, skb);
800 				nhc = FIB_RES_NHC(res);
801 				update_or_create_fnhe(nhc, fl4->daddr, new_gw,
802 						0, false,
803 						jiffies + ip_rt_gc_timeout);
804 			}
805 			if (kill_route)
806 				WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL);
807 			call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
808 		}
809 		neigh_release(n);
810 	}
811 	return;
812 
813 reject_redirect:
814 #ifdef CONFIG_IP_ROUTE_VERBOSE
815 	if (IN_DEV_LOG_MARTIANS(in_dev)) {
816 		const struct iphdr *iph = (const struct iphdr *) skb->data;
817 		__be32 daddr = iph->daddr;
818 		__be32 saddr = iph->saddr;
819 
820 		net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
821 				     "  Advised path = %pI4 -> %pI4\n",
822 				     &old_gw, dev->name, &new_gw,
823 				     &saddr, &daddr);
824 	}
825 #endif
826 	;
827 }
828 
829 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
830 {
831 	struct rtable *rt;
832 	struct flowi4 fl4;
833 	const struct iphdr *iph = (const struct iphdr *) skb->data;
834 	struct net *net = dev_net(skb->dev);
835 	int oif = skb->dev->ifindex;
836 	u8 prot = iph->protocol;
837 	u32 mark = skb->mark;
838 	__u8 tos = iph->tos;
839 
840 	rt = dst_rtable(dst);
841 
842 	__build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
843 	__ip_do_redirect(rt, skb, &fl4, true);
844 }
845 
846 static void ipv4_negative_advice(struct sock *sk,
847 				 struct dst_entry *dst)
848 {
849 	struct rtable *rt = dst_rtable(dst);
850 
851 	if ((READ_ONCE(dst->obsolete) > 0) ||
852 	    (rt->rt_flags & RTCF_REDIRECTED) ||
853 	    READ_ONCE(rt->dst.expires))
854 		sk_dst_reset(sk);
855 }
856 
857 /*
858  * Algorithm:
859  *	1. The first ip_rt_redirect_number redirects are sent
860  *	   with exponential backoff, then we stop sending them at all,
861  *	   assuming that the host ignores our redirects.
862  *	2. If we did not see packets requiring redirects
863  *	   during ip_rt_redirect_silence, we assume that the host
864  *	   forgot redirected route and start to send redirects again.
865  *
866  * This algorithm is much cheaper and more intelligent than dumb load limiting
867  * in icmp.c.
868  *
869  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
870  * and "frag. need" (breaks PMTU discovery) in icmp.c.
871  */
872 
873 void ip_rt_send_redirect(struct sk_buff *skb)
874 {
875 	struct rtable *rt = skb_rtable(skb);
876 	struct in_device *in_dev;
877 	struct net_device *dev;
878 	struct inet_peer *peer;
879 	int log_martians;
880 	struct net *net;
881 	int vif;
882 
883 	rcu_read_lock();
884 	dev = dst_dev_rcu(&rt->dst);
885 	in_dev = __in_dev_get_rcu(dev);
886 	if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
887 		rcu_read_unlock();
888 		return;
889 	}
890 	log_martians = IN_DEV_LOG_MARTIANS(in_dev);
891 	vif = l3mdev_master_ifindex_rcu(dev);
892 
893 	net = dev_net_rcu(dev);
894 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif);
895 	if (!peer) {
896 		rcu_read_unlock();
897 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
898 			  rt_nexthop(rt, ip_hdr(skb)->daddr));
899 		return;
900 	}
901 
902 	/* No redirected packets during ip_rt_redirect_silence;
903 	 * reset the algorithm.
904 	 */
905 	if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
906 		peer->rate_tokens = 0;
907 		peer->n_redirects = 0;
908 	}
909 
910 	/* Too many ignored redirects; do not send anything
911 	 * set dst.rate_last to the last seen redirected packet.
912 	 */
913 	if (peer->n_redirects >= ip_rt_redirect_number) {
914 		peer->rate_last = jiffies;
915 		goto out_unlock;
916 	}
917 
918 	/* Check for load limit; set rate_last to the latest sent
919 	 * redirect.
920 	 */
921 	if (peer->n_redirects == 0 ||
922 	    time_after(jiffies,
923 		       (peer->rate_last +
924 			(ip_rt_redirect_load << peer->n_redirects)))) {
925 		__be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
926 
927 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
928 		peer->rate_last = jiffies;
929 		++peer->n_redirects;
930 		if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians &&
931 		    peer->n_redirects == ip_rt_redirect_number)
932 			net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
933 					     &ip_hdr(skb)->saddr, inet_iif(skb),
934 					     &ip_hdr(skb)->daddr, &gw);
935 	}
936 out_unlock:
937 	rcu_read_unlock();
938 }
939 
940 static int ip_error(struct sk_buff *skb)
941 {
942 	struct rtable *rt = skb_rtable(skb);
943 	struct net_device *dev = skb->dev;
944 	struct in_device *in_dev;
945 	struct inet_peer *peer;
946 	unsigned long now;
947 	struct net *net;
948 	SKB_DR(reason);
949 	bool send;
950 	int code;
951 
952 	if (netif_is_l3_master(skb->dev)) {
953 		dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
954 		if (!dev)
955 			goto out;
956 	}
957 
958 	in_dev = __in_dev_get_rcu(dev);
959 
960 	/* IP on this device is disabled. */
961 	if (!in_dev)
962 		goto out;
963 
964 	net = dev_net(rt->dst.dev);
965 	if (!IN_DEV_FORWARD(in_dev)) {
966 		switch (rt->dst.error) {
967 		case EHOSTUNREACH:
968 			SKB_DR_SET(reason, IP_INADDRERRORS);
969 			__IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
970 			break;
971 
972 		case ENETUNREACH:
973 			SKB_DR_SET(reason, IP_INNOROUTES);
974 			__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
975 			break;
976 		}
977 		goto out;
978 	}
979 
980 	switch (rt->dst.error) {
981 	case EINVAL:
982 	default:
983 		goto out;
984 	case EHOSTUNREACH:
985 		code = ICMP_HOST_UNREACH;
986 		break;
987 	case ENETUNREACH:
988 		code = ICMP_NET_UNREACH;
989 		SKB_DR_SET(reason, IP_INNOROUTES);
990 		__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
991 		break;
992 	case EACCES:
993 		code = ICMP_PKT_FILTERED;
994 		break;
995 	}
996 
997 	rcu_read_lock();
998 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
999 			       l3mdev_master_ifindex_rcu(skb->dev));
1000 	send = true;
1001 	if (peer) {
1002 		now = jiffies;
1003 		peer->rate_tokens += now - peer->rate_last;
1004 		if (peer->rate_tokens > ip_rt_error_burst)
1005 			peer->rate_tokens = ip_rt_error_burst;
1006 		peer->rate_last = now;
1007 		if (peer->rate_tokens >= ip_rt_error_cost)
1008 			peer->rate_tokens -= ip_rt_error_cost;
1009 		else
1010 			send = false;
1011 	}
1012 	rcu_read_unlock();
1013 
1014 	if (send)
1015 		icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1016 
1017 out:	kfree_skb_reason(skb, reason);
1018 	return 0;
1019 }
1020 
1021 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1022 {
1023 	struct dst_entry *dst = &rt->dst;
1024 	struct fib_result res;
1025 	bool lock = false;
1026 	struct net *net;
1027 	u32 old_mtu;
1028 
1029 	if (ip_mtu_locked(dst))
1030 		return;
1031 
1032 	old_mtu = ipv4_mtu(dst);
1033 	if (old_mtu < mtu)
1034 		return;
1035 
1036 	rcu_read_lock();
1037 	net = dst_dev_net_rcu(dst);
1038 	if (mtu < net->ipv4.ip_rt_min_pmtu) {
1039 		lock = true;
1040 		mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1041 	}
1042 
1043 	if (rt->rt_pmtu == mtu && !lock &&
1044 	    time_before(jiffies, READ_ONCE(dst->expires) -
1045 				 net->ipv4.ip_rt_mtu_expires / 2))
1046 		goto out;
1047 
1048 	if (fib_lookup(net, fl4, &res, 0) == 0) {
1049 		struct fib_nh_common *nhc;
1050 
1051 		fib_select_path(net, &res, fl4, NULL);
1052 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1053 		if (fib_info_num_path(res.fi) > 1) {
1054 			int nhsel;
1055 
1056 			for (nhsel = 0; nhsel < fib_info_num_path(res.fi); nhsel++) {
1057 				nhc = fib_info_nhc(res.fi, nhsel);
1058 				update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1059 						      jiffies + net->ipv4.ip_rt_mtu_expires);
1060 			}
1061 			goto out;
1062 		}
1063 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1064 		nhc = FIB_RES_NHC(res);
1065 		update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1066 				      jiffies + net->ipv4.ip_rt_mtu_expires);
1067 	}
1068 out:
1069 	rcu_read_unlock();
1070 }
1071 
1072 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1073 			      struct sk_buff *skb, u32 mtu,
1074 			      bool confirm_neigh)
1075 {
1076 	struct rtable *rt = dst_rtable(dst);
1077 	struct flowi4 fl4;
1078 
1079 	ip_rt_build_flow_key(&fl4, sk, skb);
1080 
1081 	/* Don't make lookup fail for bridged encapsulations */
1082 	if (skb && netif_is_any_bridge_port(skb->dev))
1083 		fl4.flowi4_oif = 0;
1084 
1085 	__ip_rt_update_pmtu(rt, &fl4, mtu);
1086 }
1087 
1088 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1089 		      int oif, u8 protocol)
1090 {
1091 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1092 	struct flowi4 fl4;
1093 	struct rtable *rt;
1094 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
1095 
1096 	__build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1097 			 0);
1098 	rt = __ip_route_output_key(net, &fl4);
1099 	if (!IS_ERR(rt)) {
1100 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1101 		ip_rt_put(rt);
1102 	}
1103 }
1104 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1105 
1106 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1107 {
1108 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1109 	struct flowi4 fl4;
1110 	struct rtable *rt;
1111 
1112 	__build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1113 
1114 	if (!fl4.flowi4_mark)
1115 		fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1116 
1117 	rt = __ip_route_output_key(sock_net(sk), &fl4);
1118 	if (!IS_ERR(rt)) {
1119 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1120 		ip_rt_put(rt);
1121 	}
1122 }
1123 
1124 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1125 {
1126 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1127 	struct flowi4 fl4;
1128 	struct rtable *rt;
1129 	struct dst_entry *odst = NULL;
1130 	bool new = false;
1131 	struct net *net = sock_net(sk);
1132 
1133 	bh_lock_sock(sk);
1134 
1135 	if (!ip_sk_accept_pmtu(sk))
1136 		goto out;
1137 
1138 	odst = sk_dst_get(sk);
1139 
1140 	if (sock_owned_by_user(sk) || !odst) {
1141 		__ipv4_sk_update_pmtu(skb, sk, mtu);
1142 		goto out;
1143 	}
1144 
1145 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1146 
1147 	rt = dst_rtable(odst);
1148 	if (READ_ONCE(odst->obsolete) && !odst->ops->check(odst, 0)) {
1149 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1150 		if (IS_ERR(rt))
1151 			goto out;
1152 
1153 		new = true;
1154 	}
1155 
1156 	__ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu);
1157 
1158 	if (!dst_check(&rt->dst, 0)) {
1159 		if (new)
1160 			dst_release(&rt->dst);
1161 
1162 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1163 		if (IS_ERR(rt))
1164 			goto out;
1165 
1166 		new = true;
1167 	}
1168 
1169 	if (new)
1170 		sk_dst_set(sk, &rt->dst);
1171 
1172 out:
1173 	bh_unlock_sock(sk);
1174 	dst_release(odst);
1175 }
1176 
1177 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1178 		   int oif, u8 protocol)
1179 {
1180 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1181 	struct flowi4 fl4;
1182 	struct rtable *rt;
1183 
1184 	__build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1185 	rt = __ip_route_output_key(net, &fl4);
1186 	if (!IS_ERR(rt)) {
1187 		__ip_do_redirect(rt, skb, &fl4, false);
1188 		ip_rt_put(rt);
1189 	}
1190 }
1191 EXPORT_SYMBOL_GPL(ipv4_redirect);
1192 
1193 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1194 {
1195 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1196 	struct flowi4 fl4;
1197 	struct rtable *rt;
1198 	struct net *net = sock_net(sk);
1199 
1200 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1201 	rt = __ip_route_output_key(net, &fl4);
1202 	if (!IS_ERR(rt)) {
1203 		__ip_do_redirect(rt, skb, &fl4, false);
1204 		ip_rt_put(rt);
1205 	}
1206 }
1207 
1208 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1209 							 u32 cookie)
1210 {
1211 	struct rtable *rt = dst_rtable(dst);
1212 
1213 	/* All IPV4 dsts are created with ->obsolete set to the value
1214 	 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1215 	 * into this function always.
1216 	 *
1217 	 * When a PMTU/redirect information update invalidates a route,
1218 	 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1219 	 * DST_OBSOLETE_DEAD.
1220 	 */
1221 	if (READ_ONCE(dst->obsolete) != DST_OBSOLETE_FORCE_CHK ||
1222 	    rt_is_expired(rt))
1223 		return NULL;
1224 	return dst;
1225 }
1226 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1227 
1228 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1229 {
1230 	struct inet_skb_parm parm;
1231 	struct net_device *dev;
1232 	int res;
1233 
1234 	/* Recompile ip options since IPCB may not be valid anymore.
1235 	 * Also check we have a reasonable ipv4 header.
1236 	 */
1237 	if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1238 	    ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1239 		return;
1240 
1241 	memset(&parm, 0, sizeof(parm));
1242 	if (ip_hdr(skb)->ihl > 5) {
1243 		if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1244 			return;
1245 		parm.opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1246 
1247 		rcu_read_lock();
1248 		dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1249 		res = __ip_options_compile(dev_net(dev), &parm.opt, skb, NULL);
1250 		rcu_read_unlock();
1251 
1252 		if (res)
1253 			return;
1254 	}
1255 	__icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &parm);
1256 }
1257 
1258 static void ipv4_link_failure(struct sk_buff *skb)
1259 {
1260 	struct rtable *rt;
1261 
1262 	ipv4_send_dest_unreach(skb);
1263 
1264 	rt = skb_rtable(skb);
1265 	if (rt)
1266 		dst_set_expires(&rt->dst, 0);
1267 }
1268 
1269 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1270 {
1271 	pr_debug("%s: %pI4 -> %pI4, %s\n",
1272 		 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1273 		 skb->dev ? skb->dev->name : "?");
1274 	kfree_skb(skb);
1275 	WARN_ON_ONCE(1);
1276 	return 0;
1277 }
1278 
1279 /*
1280  * We do not cache source address of outgoing interface,
1281  * because it is used only by IP RR, TS and SRR options,
1282  * so that it out of fast path.
1283  *
1284  * BTW remember: "addr" is allowed to be not aligned
1285  * in IP options!
1286  */
1287 
1288 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1289 {
1290 	__be32 src;
1291 
1292 	rcu_read_lock();
1293 	if (rt_is_output_route(rt)) {
1294 		src = ip_hdr(skb)->saddr;
1295 	} else {
1296 		struct net_device *dev = dst_dev_rcu(&rt->dst);
1297 		struct net *net = dev_net_rcu(dev);
1298 		struct iphdr *iph = ip_hdr(skb);
1299 		struct fib_result res;
1300 		struct flowi4 fl4 = {
1301 			.daddr = iph->daddr,
1302 			.saddr = iph->saddr,
1303 			.flowi4_dscp = ip4h_dscp(iph),
1304 			.flowi4_oif = dev->ifindex,
1305 			.flowi4_iif = skb->dev->ifindex,
1306 			.flowi4_mark = skb->mark,
1307 		};
1308 
1309 		if (fib_lookup(net, &fl4, &res, 0) == 0)
1310 			src = fib_result_prefsrc(net, &res);
1311 		else
1312 			src = inet_select_addr(dev,
1313 					       rt_nexthop(rt, iph->daddr),
1314 					       RT_SCOPE_UNIVERSE);
1315 	}
1316 	rcu_read_unlock();
1317 
1318 	memcpy(addr, &src, 4);
1319 }
1320 
1321 #ifdef CONFIG_IP_ROUTE_CLASSID
1322 static void set_class_tag(struct rtable *rt, u32 tag)
1323 {
1324 	if (!(rt->dst.tclassid & 0xFFFF))
1325 		rt->dst.tclassid |= tag & 0xFFFF;
1326 	if (!(rt->dst.tclassid & 0xFFFF0000))
1327 		rt->dst.tclassid |= tag & 0xFFFF0000;
1328 }
1329 #endif
1330 
1331 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1332 {
1333 	unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1334 	unsigned int advmss;
1335 	struct net *net;
1336 
1337 	rcu_read_lock();
1338 	net = dst_dev_net_rcu(dst);
1339 	advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1340 				   net->ipv4.ip_rt_min_advmss);
1341 	rcu_read_unlock();
1342 
1343 	return min(advmss, IPV4_MAX_PMTU - header_size);
1344 }
1345 
1346 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1347 {
1348 	return ip_dst_mtu_maybe_forward(dst, false);
1349 }
1350 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1351 
1352 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1353 {
1354 	struct fnhe_hash_bucket *hash;
1355 	struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1356 	u32 hval = fnhe_hashfun(daddr);
1357 
1358 	spin_lock_bh(&fnhe_lock);
1359 
1360 	hash = rcu_dereference_protected(nhc->nhc_exceptions,
1361 					 lockdep_is_held(&fnhe_lock));
1362 	hash += hval;
1363 
1364 	fnhe_p = &hash->chain;
1365 	fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1366 	while (fnhe) {
1367 		if (fnhe->fnhe_daddr == daddr) {
1368 			rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1369 				fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1370 			/* set fnhe_daddr to 0 to ensure it won't bind with
1371 			 * new dsts in rt_bind_exception().
1372 			 */
1373 			fnhe->fnhe_daddr = 0;
1374 			fnhe_flush_routes(fnhe);
1375 			kfree_rcu(fnhe, rcu);
1376 			break;
1377 		}
1378 		fnhe_p = &fnhe->fnhe_next;
1379 		fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1380 						 lockdep_is_held(&fnhe_lock));
1381 	}
1382 
1383 	spin_unlock_bh(&fnhe_lock);
1384 }
1385 
1386 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1387 					       __be32 daddr)
1388 {
1389 	struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1390 	struct fib_nh_exception *fnhe;
1391 	u32 hval;
1392 
1393 	if (!hash)
1394 		return NULL;
1395 
1396 	hval = fnhe_hashfun(daddr);
1397 
1398 	for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1399 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
1400 		if (fnhe->fnhe_daddr == daddr) {
1401 			if (fnhe->fnhe_expires &&
1402 			    time_after(jiffies, fnhe->fnhe_expires)) {
1403 				ip_del_fnhe(nhc, daddr);
1404 				break;
1405 			}
1406 			return fnhe;
1407 		}
1408 	}
1409 	return NULL;
1410 }
1411 
1412 /* MTU selection:
1413  * 1. mtu on route is locked - use it
1414  * 2. mtu from nexthop exception
1415  * 3. mtu from egress device
1416  */
1417 
1418 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1419 {
1420 	struct fib_nh_common *nhc = res->nhc;
1421 	struct net_device *dev = nhc->nhc_dev;
1422 	struct fib_info *fi = res->fi;
1423 	u32 mtu = 0;
1424 
1425 	if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1426 	    fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1427 		mtu = fi->fib_mtu;
1428 
1429 	if (likely(!mtu)) {
1430 		struct fib_nh_exception *fnhe;
1431 
1432 		fnhe = find_exception(nhc, daddr);
1433 		if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1434 			mtu = fnhe->fnhe_pmtu;
1435 	}
1436 
1437 	if (likely(!mtu))
1438 		mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1439 
1440 	return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1441 }
1442 
1443 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1444 			      __be32 daddr, const bool do_cache)
1445 {
1446 	bool ret = false;
1447 
1448 	spin_lock_bh(&fnhe_lock);
1449 
1450 	if (daddr == fnhe->fnhe_daddr) {
1451 		struct rtable __rcu **porig;
1452 		struct rtable *orig;
1453 		int genid = fnhe_genid(dev_net(rt->dst.dev));
1454 
1455 		if (rt_is_input_route(rt))
1456 			porig = &fnhe->fnhe_rth_input;
1457 		else
1458 			porig = &fnhe->fnhe_rth_output;
1459 		orig = rcu_dereference(*porig);
1460 
1461 		if (fnhe->fnhe_genid != genid) {
1462 			fnhe->fnhe_genid = genid;
1463 			fnhe->fnhe_gw = 0;
1464 			fnhe->fnhe_pmtu = 0;
1465 			fnhe->fnhe_expires = 0;
1466 			fnhe->fnhe_mtu_locked = false;
1467 			fnhe_flush_routes(fnhe);
1468 			orig = NULL;
1469 		}
1470 		fill_route_from_fnhe(rt, fnhe);
1471 		if (!rt->rt_gw4) {
1472 			rt->rt_gw4 = daddr;
1473 			rt->rt_gw_family = AF_INET;
1474 		}
1475 
1476 		if (do_cache) {
1477 			dst_hold(&rt->dst);
1478 			rcu_assign_pointer(*porig, rt);
1479 			if (orig) {
1480 				dst_dev_put(&orig->dst);
1481 				dst_release(&orig->dst);
1482 			}
1483 			ret = true;
1484 		}
1485 
1486 		fnhe->fnhe_stamp = jiffies;
1487 	}
1488 	spin_unlock_bh(&fnhe_lock);
1489 
1490 	return ret;
1491 }
1492 
1493 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1494 {
1495 	struct rtable *orig, *prev, **p;
1496 	bool ret = true;
1497 
1498 	if (rt_is_input_route(rt)) {
1499 		p = (struct rtable **)&nhc->nhc_rth_input;
1500 	} else {
1501 		p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1502 	}
1503 	orig = *p;
1504 
1505 	/* hold dst before doing cmpxchg() to avoid race condition
1506 	 * on this dst
1507 	 */
1508 	dst_hold(&rt->dst);
1509 	prev = cmpxchg(p, orig, rt);
1510 	if (prev == orig) {
1511 		if (orig) {
1512 			rt_add_uncached_list(orig);
1513 			dst_release(&orig->dst);
1514 		}
1515 	} else {
1516 		dst_release(&rt->dst);
1517 		ret = false;
1518 	}
1519 
1520 	return ret;
1521 }
1522 
1523 struct uncached_list {
1524 	spinlock_t		lock;
1525 	struct list_head	head;
1526 };
1527 
1528 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1529 
1530 void rt_add_uncached_list(struct rtable *rt)
1531 {
1532 	struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1533 
1534 	rt->dst.rt_uncached_list = ul;
1535 
1536 	spin_lock_bh(&ul->lock);
1537 	list_add_tail(&rt->dst.rt_uncached, &ul->head);
1538 	spin_unlock_bh(&ul->lock);
1539 }
1540 
1541 void rt_del_uncached_list(struct rtable *rt)
1542 {
1543 	struct uncached_list *ul = rt->dst.rt_uncached_list;
1544 
1545 	if (ul) {
1546 		spin_lock_bh(&ul->lock);
1547 		list_del_init(&rt->dst.rt_uncached);
1548 		spin_unlock_bh(&ul->lock);
1549 	}
1550 }
1551 
1552 static void ipv4_dst_destroy(struct dst_entry *dst)
1553 {
1554 	ip_dst_metrics_put(dst);
1555 	rt_del_uncached_list(dst_rtable(dst));
1556 }
1557 
1558 void rt_flush_dev(struct net_device *dev)
1559 {
1560 	struct rtable *rt, *safe;
1561 	int cpu;
1562 
1563 	for_each_possible_cpu(cpu) {
1564 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1565 
1566 		if (list_empty(&ul->head))
1567 			continue;
1568 
1569 		spin_lock_bh(&ul->lock);
1570 		list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1571 			if (rt->dst.dev != dev)
1572 				continue;
1573 			rcu_assign_pointer(rt->dst.dev_rcu, blackhole_netdev);
1574 			netdev_ref_replace(dev, blackhole_netdev,
1575 					   &rt->dst.dev_tracker, GFP_ATOMIC);
1576 			list_del_init(&rt->dst.rt_uncached);
1577 		}
1578 		spin_unlock_bh(&ul->lock);
1579 	}
1580 }
1581 
1582 static bool rt_cache_valid(const struct rtable *rt)
1583 {
1584 	return	rt &&
1585 		READ_ONCE(rt->dst.obsolete) == DST_OBSOLETE_FORCE_CHK &&
1586 		!rt_is_expired(rt);
1587 }
1588 
1589 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1590 			   const struct fib_result *res,
1591 			   struct fib_nh_exception *fnhe,
1592 			   struct fib_info *fi, u16 type, u32 itag,
1593 			   const bool do_cache)
1594 {
1595 	bool cached = false;
1596 
1597 	if (fi) {
1598 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1599 
1600 		if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1601 			rt->rt_uses_gateway = 1;
1602 			rt->rt_gw_family = nhc->nhc_gw_family;
1603 			/* only INET and INET6 are supported */
1604 			if (likely(nhc->nhc_gw_family == AF_INET))
1605 				rt->rt_gw4 = nhc->nhc_gw.ipv4;
1606 			else
1607 				rt->rt_gw6 = nhc->nhc_gw.ipv6;
1608 		}
1609 
1610 		ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1611 
1612 #ifdef CONFIG_IP_ROUTE_CLASSID
1613 		if (nhc->nhc_family == AF_INET) {
1614 			struct fib_nh *nh;
1615 
1616 			nh = container_of(nhc, struct fib_nh, nh_common);
1617 			rt->dst.tclassid = nh->nh_tclassid;
1618 		}
1619 #endif
1620 		rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1621 		if (unlikely(fnhe))
1622 			cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1623 		else if (do_cache)
1624 			cached = rt_cache_route(nhc, rt);
1625 		if (unlikely(!cached)) {
1626 			/* Routes we intend to cache in nexthop exception or
1627 			 * FIB nexthop have the DST_NOCACHE bit clear.
1628 			 * However, if we are unsuccessful at storing this
1629 			 * route into the cache we really need to set it.
1630 			 */
1631 			if (!rt->rt_gw4) {
1632 				rt->rt_gw_family = AF_INET;
1633 				rt->rt_gw4 = daddr;
1634 			}
1635 			rt_add_uncached_list(rt);
1636 		}
1637 	} else
1638 		rt_add_uncached_list(rt);
1639 
1640 #ifdef CONFIG_IP_ROUTE_CLASSID
1641 #ifdef CONFIG_IP_MULTIPLE_TABLES
1642 	set_class_tag(rt, res->tclassid);
1643 #endif
1644 	set_class_tag(rt, itag);
1645 #endif
1646 }
1647 
1648 struct rtable *rt_dst_alloc(struct net_device *dev,
1649 			    unsigned int flags, u16 type,
1650 			    bool noxfrm)
1651 {
1652 	struct rtable *rt;
1653 
1654 	rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1655 		       (noxfrm ? DST_NOXFRM : 0));
1656 
1657 	if (rt) {
1658 		rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1659 		rt->rt_flags = flags;
1660 		rt->rt_type = type;
1661 		rt->rt_is_input = 0;
1662 		rt->rt_iif = 0;
1663 		rt->rt_pmtu = 0;
1664 		rt->rt_mtu_locked = 0;
1665 		rt->rt_uses_gateway = 0;
1666 		rt->rt_gw_family = 0;
1667 		rt->rt_gw4 = 0;
1668 
1669 		rt->dst.output = ip_output;
1670 		if (flags & RTCF_LOCAL)
1671 			rt->dst.input = ip_local_deliver;
1672 	}
1673 
1674 	return rt;
1675 }
1676 EXPORT_SYMBOL(rt_dst_alloc);
1677 
1678 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1679 {
1680 	struct rtable *new_rt;
1681 
1682 	new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1683 			   rt->dst.flags);
1684 
1685 	if (new_rt) {
1686 		new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1687 		new_rt->rt_flags = rt->rt_flags;
1688 		new_rt->rt_type = rt->rt_type;
1689 		new_rt->rt_is_input = rt->rt_is_input;
1690 		new_rt->rt_iif = rt->rt_iif;
1691 		new_rt->rt_pmtu = rt->rt_pmtu;
1692 		new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1693 		new_rt->rt_gw_family = rt->rt_gw_family;
1694 		if (rt->rt_gw_family == AF_INET)
1695 			new_rt->rt_gw4 = rt->rt_gw4;
1696 		else if (rt->rt_gw_family == AF_INET6)
1697 			new_rt->rt_gw6 = rt->rt_gw6;
1698 
1699 		new_rt->dst.input = READ_ONCE(rt->dst.input);
1700 		new_rt->dst.output = READ_ONCE(rt->dst.output);
1701 		new_rt->dst.error = rt->dst.error;
1702 		new_rt->dst.lastuse = jiffies;
1703 		new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1704 	}
1705 	return new_rt;
1706 }
1707 
1708 /* called in rcu_read_lock() section */
1709 enum skb_drop_reason
1710 ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1711 		      dscp_t dscp, struct net_device *dev,
1712 		      struct in_device *in_dev, u32 *itag)
1713 {
1714 	enum skb_drop_reason reason;
1715 
1716 	/* Primary sanity checks. */
1717 	if (!in_dev)
1718 		return SKB_DROP_REASON_NOT_SPECIFIED;
1719 
1720 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1721 		return SKB_DROP_REASON_IP_INVALID_SOURCE;
1722 
1723 	if (skb->protocol != htons(ETH_P_IP))
1724 		return SKB_DROP_REASON_INVALID_PROTO;
1725 
1726 	if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1727 		return SKB_DROP_REASON_IP_LOCALNET;
1728 
1729 	if (ipv4_is_zeronet(saddr)) {
1730 		if (!ipv4_is_local_multicast(daddr) &&
1731 		    ip_hdr(skb)->protocol != IPPROTO_IGMP)
1732 			return SKB_DROP_REASON_IP_INVALID_SOURCE;
1733 	} else {
1734 		reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
1735 						    dev, in_dev, itag);
1736 		if (reason)
1737 			return reason;
1738 	}
1739 	return SKB_NOT_DROPPED_YET;
1740 }
1741 
1742 /* called in rcu_read_lock() section */
1743 static enum skb_drop_reason
1744 ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1745 		  dscp_t dscp, struct net_device *dev, int our)
1746 {
1747 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1748 	unsigned int flags = RTCF_MULTICAST;
1749 	enum skb_drop_reason reason;
1750 	struct rtable *rth;
1751 	u32 itag = 0;
1752 
1753 	reason = ip_mc_validate_source(skb, daddr, saddr, dscp, dev, in_dev,
1754 				       &itag);
1755 	if (reason)
1756 		return reason;
1757 
1758 	if (our)
1759 		flags |= RTCF_LOCAL;
1760 
1761 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1762 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
1763 
1764 	rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1765 			   false);
1766 	if (!rth)
1767 		return SKB_DROP_REASON_NOMEM;
1768 
1769 #ifdef CONFIG_IP_ROUTE_CLASSID
1770 	rth->dst.tclassid = itag;
1771 #endif
1772 	rth->dst.output = ip_rt_bug;
1773 	rth->rt_is_input= 1;
1774 
1775 #ifdef CONFIG_IP_MROUTE
1776 	if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1777 		rth->dst.input = ip_mr_input;
1778 #endif
1779 	RT_CACHE_STAT_INC(in_slow_mc);
1780 
1781 	skb_dst_drop(skb);
1782 	skb_dst_set(skb, &rth->dst);
1783 	return SKB_NOT_DROPPED_YET;
1784 }
1785 
1786 
1787 static void ip_handle_martian_source(struct net_device *dev,
1788 				     struct in_device *in_dev,
1789 				     struct sk_buff *skb,
1790 				     __be32 daddr,
1791 				     __be32 saddr)
1792 {
1793 	RT_CACHE_STAT_INC(in_martian_src);
1794 #ifdef CONFIG_IP_ROUTE_VERBOSE
1795 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1796 		/*
1797 		 *	RFC1812 recommendation, if source is martian,
1798 		 *	the only hint is MAC header.
1799 		 */
1800 		pr_warn("martian source (src=%pI4, dst=%pI4, dev=%s)\n",
1801 			&saddr, &daddr, dev->name);
1802 		if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1803 			print_hex_dump(KERN_WARNING, "ll header: ",
1804 				       DUMP_PREFIX_OFFSET, 16, 1,
1805 				       skb_mac_header(skb),
1806 				       dev->hard_header_len, false);
1807 		}
1808 	}
1809 #endif
1810 }
1811 
1812 /* called in rcu_read_lock() section */
1813 static enum skb_drop_reason
1814 __mkroute_input(struct sk_buff *skb, const struct fib_result *res,
1815 		struct in_device *in_dev, __be32 daddr,
1816 		__be32 saddr, dscp_t dscp)
1817 {
1818 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1819 	struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1820 	struct net_device *dev = nhc->nhc_dev;
1821 	struct fib_nh_exception *fnhe;
1822 	struct rtable *rth;
1823 	int err;
1824 	struct in_device *out_dev;
1825 	bool do_cache;
1826 	u32 itag = 0;
1827 
1828 	/* get a working reference to the output device */
1829 	out_dev = __in_dev_get_rcu(dev);
1830 	if (!out_dev) {
1831 		net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1832 		return reason;
1833 	}
1834 
1835 	err = fib_validate_source(skb, saddr, daddr, dscp, FIB_RES_OIF(*res),
1836 				  in_dev->dev, in_dev, &itag);
1837 	if (err < 0) {
1838 		reason = -err;
1839 		ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1840 					 saddr);
1841 
1842 		goto cleanup;
1843 	}
1844 
1845 	do_cache = res->fi && !itag;
1846 	if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1847 	    skb->protocol == htons(ETH_P_IP)) {
1848 		__be32 gw;
1849 
1850 		gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1851 		if (IN_DEV_SHARED_MEDIA(out_dev) ||
1852 		    inet_addr_onlink(out_dev, saddr, gw))
1853 			IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1854 	}
1855 
1856 	if (skb->protocol != htons(ETH_P_IP)) {
1857 		/* Not IP (i.e. ARP). Do not create route, if it is
1858 		 * invalid for proxy arp. DNAT routes are always valid.
1859 		 *
1860 		 * Proxy arp feature have been extended to allow, ARP
1861 		 * replies back to the same interface, to support
1862 		 * Private VLAN switch technologies. See arp.c.
1863 		 */
1864 		if (out_dev == in_dev &&
1865 		    IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1866 			reason = SKB_DROP_REASON_ARP_PVLAN_DISABLE;
1867 			goto cleanup;
1868 		}
1869 	}
1870 
1871 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1872 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
1873 
1874 	fnhe = find_exception(nhc, daddr);
1875 	if (do_cache) {
1876 		if (fnhe)
1877 			rth = rcu_dereference(fnhe->fnhe_rth_input);
1878 		else
1879 			rth = rcu_dereference(nhc->nhc_rth_input);
1880 		if (rt_cache_valid(rth)) {
1881 			skb_dst_set_noref(skb, &rth->dst);
1882 			goto out;
1883 		}
1884 	}
1885 
1886 	rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1887 			   IN_DEV_ORCONF(out_dev, NOXFRM));
1888 	if (!rth) {
1889 		reason = SKB_DROP_REASON_NOMEM;
1890 		goto cleanup;
1891 	}
1892 
1893 	rth->rt_is_input = 1;
1894 	RT_CACHE_STAT_INC(in_slow_tot);
1895 
1896 	rth->dst.input = ip_forward;
1897 
1898 	rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1899 		       do_cache);
1900 	lwtunnel_set_redirect(&rth->dst);
1901 	skb_dst_set(skb, &rth->dst);
1902 out:
1903 	reason = SKB_NOT_DROPPED_YET;
1904 cleanup:
1905 	return reason;
1906 }
1907 
1908 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1909 /* To make ICMP packets follow the right flow, the multipath hash is
1910  * calculated from the inner IP addresses.
1911  */
1912 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1913 				 struct flow_keys *hash_keys)
1914 {
1915 	const struct iphdr *outer_iph = ip_hdr(skb);
1916 	const struct iphdr *key_iph = outer_iph;
1917 	const struct iphdr *inner_iph;
1918 	const struct icmphdr *icmph;
1919 	struct iphdr _inner_iph;
1920 	struct icmphdr _icmph;
1921 
1922 	if (likely(outer_iph->protocol != IPPROTO_ICMP))
1923 		goto out;
1924 
1925 	if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1926 		goto out;
1927 
1928 	icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1929 				   &_icmph);
1930 	if (!icmph)
1931 		goto out;
1932 
1933 	if (!icmp_is_err(icmph->type))
1934 		goto out;
1935 
1936 	inner_iph = skb_header_pointer(skb,
1937 				       outer_iph->ihl * 4 + sizeof(_icmph),
1938 				       sizeof(_inner_iph), &_inner_iph);
1939 	if (!inner_iph)
1940 		goto out;
1941 
1942 	key_iph = inner_iph;
1943 out:
1944 	hash_keys->addrs.v4addrs.src = key_iph->saddr;
1945 	hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1946 }
1947 
1948 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1949 					   const struct sk_buff *skb,
1950 					   bool *p_has_inner)
1951 {
1952 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1953 	struct flow_keys keys, hash_keys;
1954 
1955 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1956 		return 0;
1957 
1958 	memset(&hash_keys, 0, sizeof(hash_keys));
1959 	skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1960 
1961 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1962 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1963 		hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1964 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1965 		hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1966 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1967 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
1968 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1969 		hash_keys.ports.src = keys.ports.src;
1970 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1971 		hash_keys.ports.dst = keys.ports.dst;
1972 
1973 	*p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1974 	return fib_multipath_hash_from_keys(net, &hash_keys);
1975 }
1976 
1977 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1978 					   const struct sk_buff *skb,
1979 					   bool has_inner)
1980 {
1981 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1982 	struct flow_keys keys, hash_keys;
1983 
1984 	/* We assume the packet carries an encapsulation, but if none was
1985 	 * encountered during dissection of the outer flow, then there is no
1986 	 * point in calling the flow dissector again.
1987 	 */
1988 	if (!has_inner)
1989 		return 0;
1990 
1991 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1992 		return 0;
1993 
1994 	memset(&hash_keys, 0, sizeof(hash_keys));
1995 	skb_flow_dissect_flow_keys(skb, &keys, 0);
1996 
1997 	if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1998 		return 0;
1999 
2000 	if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2001 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2002 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2003 			hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2004 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2005 			hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2006 	} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2007 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2008 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2009 			hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2010 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2011 			hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2012 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2013 			hash_keys.tags.flow_label = keys.tags.flow_label;
2014 	}
2015 
2016 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2017 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
2018 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2019 		hash_keys.ports.src = keys.ports.src;
2020 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2021 		hash_keys.ports.dst = keys.ports.dst;
2022 
2023 	return fib_multipath_hash_from_keys(net, &hash_keys);
2024 }
2025 
2026 static u32 fib_multipath_custom_hash_skb(const struct net *net,
2027 					 const struct sk_buff *skb)
2028 {
2029 	u32 mhash, mhash_inner;
2030 	bool has_inner = true;
2031 
2032 	mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2033 	mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2034 
2035 	return jhash_2words(mhash, mhash_inner, 0);
2036 }
2037 
2038 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2039 					 const struct flowi4 *fl4)
2040 {
2041 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2042 	struct flow_keys hash_keys;
2043 
2044 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2045 		return 0;
2046 
2047 	memset(&hash_keys, 0, sizeof(hash_keys));
2048 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2049 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2050 		hash_keys.addrs.v4addrs.src = fl4->saddr;
2051 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2052 		hash_keys.addrs.v4addrs.dst = fl4->daddr;
2053 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2054 		hash_keys.basic.ip_proto = fl4->flowi4_proto;
2055 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) {
2056 		if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT)
2057 			hash_keys.ports.src = (__force __be16)get_random_u16();
2058 		else
2059 			hash_keys.ports.src = fl4->fl4_sport;
2060 	}
2061 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2062 		hash_keys.ports.dst = fl4->fl4_dport;
2063 
2064 	return fib_multipath_hash_from_keys(net, &hash_keys);
2065 }
2066 
2067 /* if skb is set it will be used and fl4 can be NULL */
2068 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2069 		       const struct sk_buff *skb, struct flow_keys *flkeys)
2070 {
2071 	u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2072 	struct flow_keys hash_keys;
2073 	u32 mhash = 0;
2074 
2075 	switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2076 	case 0:
2077 		memset(&hash_keys, 0, sizeof(hash_keys));
2078 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2079 		if (skb) {
2080 			ip_multipath_l3_keys(skb, &hash_keys);
2081 		} else {
2082 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2083 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2084 		}
2085 		mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2086 		break;
2087 	case 1:
2088 		/* skb is currently provided only when forwarding */
2089 		if (skb) {
2090 			unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2091 			struct flow_keys keys;
2092 
2093 			/* short-circuit if we already have L4 hash present */
2094 			if (skb->l4_hash)
2095 				return skb_get_hash_raw(skb) >> 1;
2096 
2097 			memset(&hash_keys, 0, sizeof(hash_keys));
2098 
2099 			if (!flkeys) {
2100 				skb_flow_dissect_flow_keys(skb, &keys, flag);
2101 				flkeys = &keys;
2102 			}
2103 
2104 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2105 			hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2106 			hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2107 			hash_keys.ports.src = flkeys->ports.src;
2108 			hash_keys.ports.dst = flkeys->ports.dst;
2109 			hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2110 		} else {
2111 			memset(&hash_keys, 0, sizeof(hash_keys));
2112 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2113 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2114 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2115 			if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT)
2116 				hash_keys.ports.src = (__force __be16)get_random_u16();
2117 			else
2118 				hash_keys.ports.src = fl4->fl4_sport;
2119 			hash_keys.ports.dst = fl4->fl4_dport;
2120 			hash_keys.basic.ip_proto = fl4->flowi4_proto;
2121 		}
2122 		mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2123 		break;
2124 	case 2:
2125 		memset(&hash_keys, 0, sizeof(hash_keys));
2126 		/* skb is currently provided only when forwarding */
2127 		if (skb) {
2128 			struct flow_keys keys;
2129 
2130 			skb_flow_dissect_flow_keys(skb, &keys, 0);
2131 			/* Inner can be v4 or v6 */
2132 			if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2133 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2134 				hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2135 				hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2136 			} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2137 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2138 				hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2139 				hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2140 				hash_keys.tags.flow_label = keys.tags.flow_label;
2141 				hash_keys.basic.ip_proto = keys.basic.ip_proto;
2142 			} else {
2143 				/* Same as case 0 */
2144 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2145 				ip_multipath_l3_keys(skb, &hash_keys);
2146 			}
2147 		} else {
2148 			/* Same as case 0 */
2149 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2150 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2151 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2152 		}
2153 		mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2154 		break;
2155 	case 3:
2156 		if (skb)
2157 			mhash = fib_multipath_custom_hash_skb(net, skb);
2158 		else
2159 			mhash = fib_multipath_custom_hash_fl4(net, fl4);
2160 		break;
2161 	}
2162 
2163 	if (multipath_hash)
2164 		mhash = jhash_2words(mhash, multipath_hash, 0);
2165 
2166 	return mhash >> 1;
2167 }
2168 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2169 
2170 static enum skb_drop_reason
2171 ip_mkroute_input(struct sk_buff *skb, struct fib_result *res,
2172 		 struct in_device *in_dev, __be32 daddr,
2173 		 __be32 saddr, dscp_t dscp, struct flow_keys *hkeys)
2174 {
2175 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2176 	if (res->fi && fib_info_num_path(res->fi) > 1) {
2177 		int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2178 
2179 		fib_select_multipath(res, h, NULL);
2180 		IPCB(skb)->flags |= IPSKB_MULTIPATH;
2181 	}
2182 #endif
2183 
2184 	/* create a routing cache entry */
2185 	return __mkroute_input(skb, res, in_dev, daddr, saddr, dscp);
2186 }
2187 
2188 /* Implements all the saddr-related checks as ip_route_input_slow(),
2189  * assuming daddr is valid and the destination is not a local broadcast one.
2190  * Uses the provided hint instead of performing a route lookup.
2191  */
2192 enum skb_drop_reason
2193 ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2194 		  dscp_t dscp, struct net_device *dev,
2195 		  const struct sk_buff *hint)
2196 {
2197 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2198 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2199 	struct rtable *rt = skb_rtable(hint);
2200 	struct net *net = dev_net(dev);
2201 	u32 tag = 0;
2202 
2203 	if (!in_dev)
2204 		return reason;
2205 
2206 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2207 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2208 		goto martian_source;
2209 	}
2210 
2211 	if (ipv4_is_zeronet(saddr)) {
2212 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2213 		goto martian_source;
2214 	}
2215 
2216 	if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2217 		reason = SKB_DROP_REASON_IP_LOCALNET;
2218 		goto martian_source;
2219 	}
2220 
2221 	if (!(rt->rt_flags & RTCF_LOCAL))
2222 		goto skip_validate_source;
2223 
2224 	reason = fib_validate_source_reason(skb, saddr, daddr, dscp, 0, dev,
2225 					    in_dev, &tag);
2226 	if (reason)
2227 		goto martian_source;
2228 
2229 skip_validate_source:
2230 	skb_dst_copy(skb, hint);
2231 	return SKB_NOT_DROPPED_YET;
2232 
2233 martian_source:
2234 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2235 	return reason;
2236 }
2237 
2238 /* get device for dst_alloc with local routes */
2239 static struct net_device *ip_rt_get_dev(struct net *net,
2240 					const struct fib_result *res)
2241 {
2242 	struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2243 	struct net_device *dev = NULL;
2244 
2245 	if (nhc)
2246 		dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2247 
2248 	return dev ? : net->loopback_dev;
2249 }
2250 
2251 /*
2252  *	NOTE. We drop all the packets that has local source
2253  *	addresses, because every properly looped back packet
2254  *	must have correct destination already attached by output routine.
2255  *	Changes in the enforced policies must be applied also to
2256  *	ip_route_use_hint().
2257  *
2258  *	Such approach solves two big problems:
2259  *	1. Not simplex devices are handled properly.
2260  *	2. IP spoofing attempts are filtered with 100% of guarantee.
2261  *	called with rcu_read_lock()
2262  */
2263 
2264 static enum skb_drop_reason
2265 ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2266 		    dscp_t dscp, struct net_device *dev,
2267 		    struct fib_result *res)
2268 {
2269 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2270 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2271 	struct flow_keys *flkeys = NULL, _flkeys;
2272 	struct net    *net = dev_net(dev);
2273 	struct ip_tunnel_info *tun_info;
2274 	int		err = -EINVAL;
2275 	unsigned int	flags = 0;
2276 	u32		itag = 0;
2277 	struct rtable	*rth;
2278 	struct flowi4	fl4;
2279 	bool do_cache = true;
2280 
2281 	/* IP on this device is disabled. */
2282 
2283 	if (!in_dev)
2284 		goto out;
2285 
2286 	/* Check for the most weird martians, which can be not detected
2287 	 * by fib_lookup.
2288 	 */
2289 
2290 	tun_info = skb_tunnel_info(skb);
2291 	if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2292 		fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2293 	else
2294 		fl4.flowi4_tun_key.tun_id = 0;
2295 	skb_dst_drop(skb);
2296 
2297 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2298 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2299 		goto martian_source;
2300 	}
2301 
2302 	res->fi = NULL;
2303 	res->table = NULL;
2304 	if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2305 		goto brd_input;
2306 
2307 	/* Accept zero addresses only to limited broadcast;
2308 	 * I even do not know to fix it or not. Waiting for complains :-)
2309 	 */
2310 	if (ipv4_is_zeronet(saddr)) {
2311 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2312 		goto martian_source;
2313 	}
2314 
2315 	if (ipv4_is_zeronet(daddr)) {
2316 		reason = SKB_DROP_REASON_IP_INVALID_DEST;
2317 		goto martian_destination;
2318 	}
2319 
2320 	/* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2321 	 * and call it once if daddr or/and saddr are loopback addresses
2322 	 */
2323 	if (ipv4_is_loopback(daddr)) {
2324 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2325 			reason = SKB_DROP_REASON_IP_LOCALNET;
2326 			goto martian_destination;
2327 		}
2328 	} else if (ipv4_is_loopback(saddr)) {
2329 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2330 			reason = SKB_DROP_REASON_IP_LOCALNET;
2331 			goto martian_source;
2332 		}
2333 	}
2334 
2335 	/*
2336 	 *	Now we are ready to route packet.
2337 	 */
2338 	fl4.flowi4_l3mdev = 0;
2339 	fl4.flowi4_oif = 0;
2340 	fl4.flowi4_iif = dev->ifindex;
2341 	fl4.flowi4_mark = skb->mark;
2342 	fl4.flowi4_dscp = dscp;
2343 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2344 	fl4.flowi4_flags = 0;
2345 	fl4.daddr = daddr;
2346 	fl4.saddr = saddr;
2347 	fl4.flowi4_uid = sock_net_uid(net, NULL);
2348 	fl4.flowi4_multipath_hash = 0;
2349 
2350 	if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2351 		flkeys = &_flkeys;
2352 	} else {
2353 		fl4.flowi4_proto = 0;
2354 		fl4.fl4_sport = 0;
2355 		fl4.fl4_dport = 0;
2356 	}
2357 
2358 	err = fib_lookup(net, &fl4, res, 0);
2359 	if (err != 0) {
2360 		if (!IN_DEV_FORWARD(in_dev))
2361 			err = -EHOSTUNREACH;
2362 		goto no_route;
2363 	}
2364 
2365 	if (res->type == RTN_BROADCAST) {
2366 		if (IN_DEV_BFORWARD(in_dev))
2367 			goto make_route;
2368 		/* not do cache if bc_forwarding is enabled */
2369 		if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING))
2370 			do_cache = false;
2371 		goto brd_input;
2372 	}
2373 
2374 	err = -EINVAL;
2375 	if (res->type == RTN_LOCAL) {
2376 		reason = fib_validate_source_reason(skb, saddr, daddr, dscp,
2377 						    0, dev, in_dev, &itag);
2378 		if (reason)
2379 			goto martian_source;
2380 		goto local_input;
2381 	}
2382 
2383 	if (!IN_DEV_FORWARD(in_dev)) {
2384 		err = -EHOSTUNREACH;
2385 		goto no_route;
2386 	}
2387 	if (res->type != RTN_UNICAST) {
2388 		reason = SKB_DROP_REASON_IP_INVALID_DEST;
2389 		goto martian_destination;
2390 	}
2391 
2392 make_route:
2393 	reason = ip_mkroute_input(skb, res, in_dev, daddr, saddr, dscp,
2394 				  flkeys);
2395 
2396 out:
2397 	return reason;
2398 
2399 brd_input:
2400 	if (skb->protocol != htons(ETH_P_IP)) {
2401 		reason = SKB_DROP_REASON_INVALID_PROTO;
2402 		goto out;
2403 	}
2404 
2405 	if (!ipv4_is_zeronet(saddr)) {
2406 		reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
2407 						    dev, in_dev, &itag);
2408 		if (reason)
2409 			goto martian_source;
2410 	}
2411 	flags |= RTCF_BROADCAST;
2412 	res->type = RTN_BROADCAST;
2413 	RT_CACHE_STAT_INC(in_brd);
2414 
2415 local_input:
2416 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2417 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
2418 
2419 	do_cache &= res->fi && !itag;
2420 	if (do_cache) {
2421 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2422 
2423 		rth = rcu_dereference(nhc->nhc_rth_input);
2424 		if (rt_cache_valid(rth)) {
2425 			skb_dst_set_noref(skb, &rth->dst);
2426 			reason = SKB_NOT_DROPPED_YET;
2427 			goto out;
2428 		}
2429 	}
2430 
2431 	rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2432 			   flags | RTCF_LOCAL, res->type, false);
2433 	if (!rth)
2434 		goto e_nobufs;
2435 
2436 	rth->dst.output= ip_rt_bug;
2437 #ifdef CONFIG_IP_ROUTE_CLASSID
2438 	rth->dst.tclassid = itag;
2439 #endif
2440 	rth->rt_is_input = 1;
2441 
2442 	RT_CACHE_STAT_INC(in_slow_tot);
2443 	if (res->type == RTN_UNREACHABLE) {
2444 		rth->dst.input= ip_error;
2445 		rth->dst.error= -err;
2446 		rth->rt_flags	&= ~RTCF_LOCAL;
2447 	}
2448 
2449 	if (do_cache) {
2450 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2451 
2452 		rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2453 		if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2454 			WARN_ON(rth->dst.input == lwtunnel_input);
2455 			rth->dst.lwtstate->orig_input = rth->dst.input;
2456 			rth->dst.input = lwtunnel_input;
2457 		}
2458 
2459 		if (unlikely(!rt_cache_route(nhc, rth)))
2460 			rt_add_uncached_list(rth);
2461 	}
2462 	skb_dst_set(skb, &rth->dst);
2463 	reason = SKB_NOT_DROPPED_YET;
2464 	goto out;
2465 
2466 no_route:
2467 	RT_CACHE_STAT_INC(in_no_route);
2468 	res->type = RTN_UNREACHABLE;
2469 	res->fi = NULL;
2470 	res->table = NULL;
2471 	goto local_input;
2472 
2473 	/*
2474 	 *	Do not cache martian addresses: they should be logged (RFC1812)
2475 	 */
2476 martian_destination:
2477 	RT_CACHE_STAT_INC(in_martian_dst);
2478 #ifdef CONFIG_IP_ROUTE_VERBOSE
2479 	if (IN_DEV_LOG_MARTIANS(in_dev))
2480 		net_warn_ratelimited("martian destination (src=%pI4, dst=%pI4, dev=%s)\n",
2481 				     &saddr, &daddr, dev->name);
2482 #endif
2483 	goto out;
2484 
2485 e_nobufs:
2486 	reason = SKB_DROP_REASON_NOMEM;
2487 	goto out;
2488 
2489 martian_source:
2490 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2491 	goto out;
2492 }
2493 
2494 /* called with rcu_read_lock held */
2495 static enum skb_drop_reason
2496 ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2497 		   dscp_t dscp, struct net_device *dev,
2498 		   struct fib_result *res)
2499 {
2500 	/* Multicast recognition logic is moved from route cache to here.
2501 	 * The problem was that too many Ethernet cards have broken/missing
2502 	 * hardware multicast filters :-( As result the host on multicasting
2503 	 * network acquires a lot of useless route cache entries, sort of
2504 	 * SDR messages from all the world. Now we try to get rid of them.
2505 	 * Really, provided software IP multicast filter is organized
2506 	 * reasonably (at least, hashed), it does not result in a slowdown
2507 	 * comparing with route cache reject entries.
2508 	 * Note, that multicast routers are not affected, because
2509 	 * route cache entry is created eventually.
2510 	 */
2511 	if (ipv4_is_multicast(daddr)) {
2512 		enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2513 		struct in_device *in_dev = __in_dev_get_rcu(dev);
2514 		int our = 0;
2515 
2516 		if (!in_dev)
2517 			return reason;
2518 
2519 		our = ip_check_mc_rcu(in_dev, daddr, saddr,
2520 				      ip_hdr(skb)->protocol);
2521 
2522 		/* check l3 master if no match yet */
2523 		if (!our && netif_is_l3_slave(dev)) {
2524 			struct in_device *l3_in_dev;
2525 
2526 			l3_in_dev = __in_dev_get_rcu(skb->dev);
2527 			if (l3_in_dev)
2528 				our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2529 						      ip_hdr(skb)->protocol);
2530 		}
2531 
2532 		if (our
2533 #ifdef CONFIG_IP_MROUTE
2534 			||
2535 		    (!ipv4_is_local_multicast(daddr) &&
2536 		     IN_DEV_MFORWARD(in_dev))
2537 #endif
2538 		   ) {
2539 			reason = ip_route_input_mc(skb, daddr, saddr, dscp,
2540 						   dev, our);
2541 		}
2542 		return reason;
2543 	}
2544 
2545 	return ip_route_input_slow(skb, daddr, saddr, dscp, dev, res);
2546 }
2547 
2548 enum skb_drop_reason ip_route_input_noref(struct sk_buff *skb, __be32 daddr,
2549 					  __be32 saddr, dscp_t dscp,
2550 					  struct net_device *dev)
2551 {
2552 	enum skb_drop_reason reason;
2553 	struct fib_result res;
2554 
2555 	rcu_read_lock();
2556 	reason = ip_route_input_rcu(skb, daddr, saddr, dscp, dev, &res);
2557 	rcu_read_unlock();
2558 
2559 	return reason;
2560 }
2561 EXPORT_SYMBOL(ip_route_input_noref);
2562 
2563 /* called with rcu_read_lock() */
2564 static struct rtable *__mkroute_output(const struct fib_result *res,
2565 				       const struct flowi4 *fl4, int orig_oif,
2566 				       struct net_device *dev_out,
2567 				       unsigned int flags)
2568 {
2569 	struct fib_info *fi = res->fi;
2570 	struct fib_nh_exception *fnhe;
2571 	struct in_device *in_dev;
2572 	u16 type = res->type;
2573 	struct rtable *rth;
2574 	bool do_cache;
2575 
2576 	in_dev = __in_dev_get_rcu(dev_out);
2577 	if (!in_dev)
2578 		return ERR_PTR(-EINVAL);
2579 
2580 	if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2581 		if (ipv4_is_loopback(fl4->saddr) &&
2582 		    !(dev_out->flags & IFF_LOOPBACK) &&
2583 		    !netif_is_l3_master(dev_out))
2584 			return ERR_PTR(-EINVAL);
2585 
2586 	if (ipv4_is_lbcast(fl4->daddr)) {
2587 		type = RTN_BROADCAST;
2588 
2589 		/* reset fi to prevent gateway resolution */
2590 		fi = NULL;
2591 	} else if (ipv4_is_multicast(fl4->daddr)) {
2592 		type = RTN_MULTICAST;
2593 	} else if (ipv4_is_zeronet(fl4->daddr)) {
2594 		return ERR_PTR(-EINVAL);
2595 	}
2596 
2597 	if (dev_out->flags & IFF_LOOPBACK)
2598 		flags |= RTCF_LOCAL;
2599 
2600 	do_cache = true;
2601 	if (type == RTN_BROADCAST) {
2602 		flags |= RTCF_BROADCAST | RTCF_LOCAL;
2603 	} else if (type == RTN_MULTICAST) {
2604 		flags |= RTCF_MULTICAST | RTCF_LOCAL;
2605 		if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2606 				     fl4->flowi4_proto))
2607 			flags &= ~RTCF_LOCAL;
2608 		else
2609 			do_cache = false;
2610 		/* If multicast route do not exist use
2611 		 * default one, but do not gateway in this case.
2612 		 * Yes, it is hack.
2613 		 */
2614 		if (fi && res->prefixlen < 4)
2615 			fi = NULL;
2616 	} else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2617 		   (orig_oif != dev_out->ifindex)) {
2618 		/* For local routes that require a particular output interface
2619 		 * we do not want to cache the result.  Caching the result
2620 		 * causes incorrect behaviour when there are multiple source
2621 		 * addresses on the interface, the end result being that if the
2622 		 * intended recipient is waiting on that interface for the
2623 		 * packet he won't receive it because it will be delivered on
2624 		 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2625 		 * be set to the loopback interface as well.
2626 		 */
2627 		do_cache = false;
2628 	}
2629 
2630 	fnhe = NULL;
2631 	do_cache &= fi != NULL;
2632 	if (fi) {
2633 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2634 		struct rtable __rcu **prth;
2635 
2636 		fnhe = find_exception(nhc, fl4->daddr);
2637 		if (!do_cache)
2638 			goto add;
2639 		if (fnhe) {
2640 			prth = &fnhe->fnhe_rth_output;
2641 		} else {
2642 			if (unlikely(fl4->flowi4_flags &
2643 				     FLOWI_FLAG_KNOWN_NH &&
2644 				     !(nhc->nhc_gw_family &&
2645 				       nhc->nhc_scope == RT_SCOPE_LINK))) {
2646 				do_cache = false;
2647 				goto add;
2648 			}
2649 			prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2650 		}
2651 		rth = rcu_dereference(*prth);
2652 		if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2653 			return rth;
2654 	}
2655 
2656 add:
2657 	rth = rt_dst_alloc(dev_out, flags, type,
2658 			   IN_DEV_ORCONF(in_dev, NOXFRM));
2659 	if (!rth)
2660 		return ERR_PTR(-ENOBUFS);
2661 
2662 	rth->rt_iif = orig_oif;
2663 
2664 	RT_CACHE_STAT_INC(out_slow_tot);
2665 
2666 	if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2667 		if (flags & RTCF_LOCAL &&
2668 		    !(dev_out->flags & IFF_LOOPBACK)) {
2669 			rth->dst.output = ip_mc_output;
2670 			RT_CACHE_STAT_INC(out_slow_mc);
2671 		}
2672 #ifdef CONFIG_IP_MROUTE
2673 		if (type == RTN_MULTICAST) {
2674 			if (IN_DEV_MFORWARD(in_dev) &&
2675 			    !ipv4_is_local_multicast(fl4->daddr)) {
2676 				rth->dst.input = ip_mr_input;
2677 				rth->dst.output = ip_mr_output;
2678 			}
2679 		}
2680 #endif
2681 	}
2682 
2683 	rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2684 	lwtunnel_set_redirect(&rth->dst);
2685 
2686 	return rth;
2687 }
2688 
2689 /*
2690  * Major route resolver routine.
2691  */
2692 
2693 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2694 					const struct sk_buff *skb)
2695 {
2696 	struct fib_result res = {
2697 		.type		= RTN_UNSPEC,
2698 		.fi		= NULL,
2699 		.table		= NULL,
2700 		.tclassid	= 0,
2701 	};
2702 	struct rtable *rth;
2703 
2704 	fl4->flowi4_iif = LOOPBACK_IFINDEX;
2705 
2706 	rcu_read_lock();
2707 	rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2708 	rcu_read_unlock();
2709 
2710 	return rth;
2711 }
2712 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2713 
2714 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2715 					    struct fib_result *res,
2716 					    const struct sk_buff *skb)
2717 {
2718 	struct net_device *dev_out = NULL;
2719 	int orig_oif = fl4->flowi4_oif;
2720 	unsigned int flags = 0;
2721 	struct rtable *rth;
2722 	int err;
2723 
2724 	if (fl4->saddr) {
2725 		if (ipv4_is_multicast(fl4->saddr) ||
2726 		    ipv4_is_lbcast(fl4->saddr)) {
2727 			rth = ERR_PTR(-EINVAL);
2728 			goto out;
2729 		}
2730 
2731 		rth = ERR_PTR(-ENETUNREACH);
2732 
2733 		/* I removed check for oif == dev_out->oif here.
2734 		 * It was wrong for two reasons:
2735 		 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2736 		 *    is assigned to multiple interfaces.
2737 		 * 2. Moreover, we are allowed to send packets with saddr
2738 		 *    of another iface. --ANK
2739 		 */
2740 
2741 		if (fl4->flowi4_oif == 0 &&
2742 		    (ipv4_is_multicast(fl4->daddr) ||
2743 		     ipv4_is_lbcast(fl4->daddr))) {
2744 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2745 			dev_out = __ip_dev_find(net, fl4->saddr, false);
2746 			if (!dev_out)
2747 				goto out;
2748 
2749 			/* Special hack: user can direct multicasts
2750 			 * and limited broadcast via necessary interface
2751 			 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2752 			 * This hack is not just for fun, it allows
2753 			 * vic,vat and friends to work.
2754 			 * They bind socket to loopback, set ttl to zero
2755 			 * and expect that it will work.
2756 			 * From the viewpoint of routing cache they are broken,
2757 			 * because we are not allowed to build multicast path
2758 			 * with loopback source addr (look, routing cache
2759 			 * cannot know, that ttl is zero, so that packet
2760 			 * will not leave this host and route is valid).
2761 			 * Luckily, this hack is good workaround.
2762 			 */
2763 
2764 			fl4->flowi4_oif = dev_out->ifindex;
2765 			goto make_route;
2766 		}
2767 
2768 		if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2769 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2770 			if (!__ip_dev_find(net, fl4->saddr, false))
2771 				goto out;
2772 		}
2773 	}
2774 
2775 
2776 	if (fl4->flowi4_oif) {
2777 		dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2778 		rth = ERR_PTR(-ENODEV);
2779 		if (!dev_out)
2780 			goto out;
2781 
2782 		/* RACE: Check return value of inet_select_addr instead. */
2783 		if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2784 			rth = ERR_PTR(-ENETUNREACH);
2785 			goto out;
2786 		}
2787 		if (ipv4_is_local_multicast(fl4->daddr) ||
2788 		    ipv4_is_lbcast(fl4->daddr) ||
2789 		    fl4->flowi4_proto == IPPROTO_IGMP) {
2790 			if (!fl4->saddr)
2791 				fl4->saddr = inet_select_addr(dev_out, 0,
2792 							      RT_SCOPE_LINK);
2793 			goto make_route;
2794 		}
2795 		if (!fl4->saddr) {
2796 			if (ipv4_is_multicast(fl4->daddr))
2797 				fl4->saddr = inet_select_addr(dev_out, 0,
2798 							      fl4->flowi4_scope);
2799 			else if (!fl4->daddr)
2800 				fl4->saddr = inet_select_addr(dev_out, 0,
2801 							      RT_SCOPE_HOST);
2802 		}
2803 	}
2804 
2805 	if (!fl4->daddr) {
2806 		fl4->daddr = fl4->saddr;
2807 		if (!fl4->daddr)
2808 			fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2809 		dev_out = net->loopback_dev;
2810 		fl4->flowi4_oif = LOOPBACK_IFINDEX;
2811 		res->type = RTN_LOCAL;
2812 		flags |= RTCF_LOCAL;
2813 		goto make_route;
2814 	}
2815 
2816 	err = fib_lookup(net, fl4, res, 0);
2817 	if (err) {
2818 		res->fi = NULL;
2819 		res->table = NULL;
2820 		if (fl4->flowi4_oif &&
2821 		    (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2822 			/* Apparently, routing tables are wrong. Assume,
2823 			 * that the destination is on link.
2824 			 *
2825 			 * WHY? DW.
2826 			 * Because we are allowed to send to iface
2827 			 * even if it has NO routes and NO assigned
2828 			 * addresses. When oif is specified, routing
2829 			 * tables are looked up with only one purpose:
2830 			 * to catch if destination is gatewayed, rather than
2831 			 * direct. Moreover, if MSG_DONTROUTE is set,
2832 			 * we send packet, ignoring both routing tables
2833 			 * and ifaddr state. --ANK
2834 			 *
2835 			 *
2836 			 * We could make it even if oif is unknown,
2837 			 * likely IPv6, but we do not.
2838 			 */
2839 
2840 			if (fl4->saddr == 0)
2841 				fl4->saddr = inet_select_addr(dev_out, 0,
2842 							      RT_SCOPE_LINK);
2843 			res->type = RTN_UNICAST;
2844 			goto make_route;
2845 		}
2846 		rth = ERR_PTR(err);
2847 		goto out;
2848 	}
2849 
2850 	if (res->type == RTN_LOCAL) {
2851 		if (!fl4->saddr) {
2852 			if (res->fi->fib_prefsrc)
2853 				fl4->saddr = res->fi->fib_prefsrc;
2854 			else
2855 				fl4->saddr = fl4->daddr;
2856 		}
2857 
2858 		/* L3 master device is the loopback for that domain */
2859 		dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2860 			net->loopback_dev;
2861 
2862 		/* make sure orig_oif points to fib result device even
2863 		 * though packet rx/tx happens over loopback or l3mdev
2864 		 */
2865 		orig_oif = FIB_RES_OIF(*res);
2866 
2867 		fl4->flowi4_oif = dev_out->ifindex;
2868 		flags |= RTCF_LOCAL;
2869 		goto make_route;
2870 	}
2871 
2872 	fib_select_path(net, res, fl4, skb);
2873 
2874 	dev_out = FIB_RES_DEV(*res);
2875 
2876 make_route:
2877 	rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2878 
2879 out:
2880 	return rth;
2881 }
2882 
2883 static struct dst_ops ipv4_dst_blackhole_ops = {
2884 	.family			= AF_INET,
2885 	.default_advmss		= ipv4_default_advmss,
2886 	.neigh_lookup		= ipv4_neigh_lookup,
2887 	.check			= dst_blackhole_check,
2888 	.cow_metrics		= dst_blackhole_cow_metrics,
2889 	.update_pmtu		= dst_blackhole_update_pmtu,
2890 	.redirect		= dst_blackhole_redirect,
2891 	.mtu			= dst_blackhole_mtu,
2892 };
2893 
2894 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2895 {
2896 	struct rtable *ort = dst_rtable(dst_orig);
2897 	struct rtable *rt;
2898 
2899 	rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0);
2900 	if (rt) {
2901 		struct dst_entry *new = &rt->dst;
2902 
2903 		new->__use = 1;
2904 		new->input = dst_discard;
2905 		new->output = dst_discard_out;
2906 
2907 		new->dev = net->loopback_dev;
2908 		netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2909 
2910 		rt->rt_is_input = ort->rt_is_input;
2911 		rt->rt_iif = ort->rt_iif;
2912 		rt->rt_pmtu = ort->rt_pmtu;
2913 		rt->rt_mtu_locked = ort->rt_mtu_locked;
2914 
2915 		rt->rt_genid = rt_genid_ipv4(net);
2916 		rt->rt_flags = ort->rt_flags;
2917 		rt->rt_type = ort->rt_type;
2918 		rt->rt_uses_gateway = ort->rt_uses_gateway;
2919 		rt->rt_gw_family = ort->rt_gw_family;
2920 		if (rt->rt_gw_family == AF_INET)
2921 			rt->rt_gw4 = ort->rt_gw4;
2922 		else if (rt->rt_gw_family == AF_INET6)
2923 			rt->rt_gw6 = ort->rt_gw6;
2924 	}
2925 
2926 	dst_release(dst_orig);
2927 
2928 	return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2929 }
2930 
2931 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2932 				    const struct sock *sk)
2933 {
2934 	struct rtable *rt = __ip_route_output_key(net, flp4);
2935 
2936 	if (IS_ERR(rt))
2937 		return rt;
2938 
2939 	if (flp4->flowi4_proto) {
2940 		flp4->flowi4_oif = rt->dst.dev->ifindex;
2941 		rt = dst_rtable(xfrm_lookup_route(net, &rt->dst,
2942 						  flowi4_to_flowi(flp4),
2943 						  sk, 0));
2944 	}
2945 
2946 	return rt;
2947 }
2948 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2949 
2950 /* called with rcu_read_lock held */
2951 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2952 			struct rtable *rt, u32 table_id, dscp_t dscp,
2953 			struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2954 			u32 seq, unsigned int flags)
2955 {
2956 	struct rtmsg *r;
2957 	struct nlmsghdr *nlh;
2958 	unsigned long expires = 0;
2959 	u32 error;
2960 	u32 metrics[RTAX_MAX];
2961 
2962 	nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2963 	if (!nlh)
2964 		return -EMSGSIZE;
2965 
2966 	r = nlmsg_data(nlh);
2967 	r->rtm_family	 = AF_INET;
2968 	r->rtm_dst_len	= 32;
2969 	r->rtm_src_len	= 0;
2970 	r->rtm_tos	= inet_dscp_to_dsfield(dscp);
2971 	r->rtm_table	= table_id < 256 ? table_id : RT_TABLE_COMPAT;
2972 	if (nla_put_u32(skb, RTA_TABLE, table_id))
2973 		goto nla_put_failure;
2974 	r->rtm_type	= rt->rt_type;
2975 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2976 	r->rtm_protocol = RTPROT_UNSPEC;
2977 	r->rtm_flags	= (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2978 	if (rt->rt_flags & RTCF_NOTIFY)
2979 		r->rtm_flags |= RTM_F_NOTIFY;
2980 	if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2981 		r->rtm_flags |= RTCF_DOREDIRECT;
2982 
2983 	if (nla_put_in_addr(skb, RTA_DST, dst))
2984 		goto nla_put_failure;
2985 	if (src) {
2986 		r->rtm_src_len = 32;
2987 		if (nla_put_in_addr(skb, RTA_SRC, src))
2988 			goto nla_put_failure;
2989 	}
2990 	if (rt->dst.dev &&
2991 	    nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2992 		goto nla_put_failure;
2993 	if (lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2994 		goto nla_put_failure;
2995 #ifdef CONFIG_IP_ROUTE_CLASSID
2996 	if (rt->dst.tclassid &&
2997 	    nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2998 		goto nla_put_failure;
2999 #endif
3000 	if (fl4 && !rt_is_input_route(rt) &&
3001 	    fl4->saddr != src) {
3002 		if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
3003 			goto nla_put_failure;
3004 	}
3005 	if (rt->rt_uses_gateway) {
3006 		if (rt->rt_gw_family == AF_INET &&
3007 		    nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
3008 			goto nla_put_failure;
3009 		} else if (rt->rt_gw_family == AF_INET6) {
3010 			int alen = sizeof(struct in6_addr);
3011 			struct nlattr *nla;
3012 			struct rtvia *via;
3013 
3014 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
3015 			if (!nla)
3016 				goto nla_put_failure;
3017 
3018 			via = nla_data(nla);
3019 			via->rtvia_family = AF_INET6;
3020 			memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
3021 		}
3022 	}
3023 
3024 	expires = READ_ONCE(rt->dst.expires);
3025 	if (expires) {
3026 		unsigned long now = jiffies;
3027 
3028 		if (time_before(now, expires))
3029 			expires -= now;
3030 		else
3031 			expires = 0;
3032 	}
3033 
3034 	memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3035 	if (rt->rt_pmtu && expires)
3036 		metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3037 	if (rt->rt_mtu_locked && expires)
3038 		metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3039 	if (rtnetlink_put_metrics(skb, metrics) < 0)
3040 		goto nla_put_failure;
3041 
3042 	if (fl4) {
3043 		if (fl4->flowi4_mark &&
3044 		    nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3045 			goto nla_put_failure;
3046 
3047 		if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3048 		    nla_put_u32(skb, RTA_UID,
3049 				from_kuid_munged(current_user_ns(),
3050 						 fl4->flowi4_uid)))
3051 			goto nla_put_failure;
3052 
3053 		if (rt_is_input_route(rt)) {
3054 #ifdef CONFIG_IP_MROUTE
3055 			if (ipv4_is_multicast(dst) &&
3056 			    !ipv4_is_local_multicast(dst) &&
3057 			    IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) {
3058 				int err = ipmr_get_route(net, skb,
3059 							 fl4->saddr, fl4->daddr,
3060 							 r, portid);
3061 
3062 				if (err <= 0) {
3063 					if (err == 0)
3064 						return 0;
3065 					goto nla_put_failure;
3066 				}
3067 			} else
3068 #endif
3069 				if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3070 					goto nla_put_failure;
3071 		}
3072 	}
3073 
3074 	error = rt->dst.error;
3075 
3076 	if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3077 		goto nla_put_failure;
3078 
3079 	nlmsg_end(skb, nlh);
3080 	return 0;
3081 
3082 nla_put_failure:
3083 	nlmsg_cancel(skb, nlh);
3084 	return -EMSGSIZE;
3085 }
3086 
3087 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3088 			    struct netlink_callback *cb, u32 table_id,
3089 			    struct fnhe_hash_bucket *bucket, int genid,
3090 			    int *fa_index, int fa_start, unsigned int flags)
3091 {
3092 	int i;
3093 
3094 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
3095 		struct fib_nh_exception *fnhe;
3096 
3097 		for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3098 		     fnhe = rcu_dereference(fnhe->fnhe_next)) {
3099 			struct rtable *rt;
3100 			int err;
3101 
3102 			if (*fa_index < fa_start)
3103 				goto next;
3104 
3105 			if (fnhe->fnhe_genid != genid)
3106 				goto next;
3107 
3108 			if (fnhe->fnhe_expires &&
3109 			    time_after(jiffies, fnhe->fnhe_expires))
3110 				goto next;
3111 
3112 			rt = rcu_dereference(fnhe->fnhe_rth_input);
3113 			if (!rt)
3114 				rt = rcu_dereference(fnhe->fnhe_rth_output);
3115 			if (!rt)
3116 				goto next;
3117 
3118 			err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3119 					   table_id, 0, NULL, skb,
3120 					   NETLINK_CB(cb->skb).portid,
3121 					   cb->nlh->nlmsg_seq, flags);
3122 			if (err)
3123 				return err;
3124 next:
3125 			(*fa_index)++;
3126 		}
3127 	}
3128 
3129 	return 0;
3130 }
3131 
3132 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3133 		       u32 table_id, struct fib_info *fi,
3134 		       int *fa_index, int fa_start, unsigned int flags)
3135 {
3136 	struct net *net = sock_net(cb->skb->sk);
3137 	int nhsel, genid = fnhe_genid(net);
3138 
3139 	for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3140 		struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3141 		struct fnhe_hash_bucket *bucket;
3142 		int err;
3143 
3144 		if (nhc->nhc_flags & RTNH_F_DEAD)
3145 			continue;
3146 
3147 		rcu_read_lock();
3148 		bucket = rcu_dereference(nhc->nhc_exceptions);
3149 		err = 0;
3150 		if (bucket)
3151 			err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3152 					       genid, fa_index, fa_start,
3153 					       flags);
3154 		rcu_read_unlock();
3155 		if (err)
3156 			return err;
3157 	}
3158 
3159 	return 0;
3160 }
3161 
3162 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3163 						   u8 ip_proto, __be16 sport,
3164 						   __be16 dport)
3165 {
3166 	struct sk_buff *skb;
3167 	struct iphdr *iph;
3168 
3169 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3170 	if (!skb)
3171 		return NULL;
3172 
3173 	/* Reserve room for dummy headers, this skb can pass
3174 	 * through good chunk of routing engine.
3175 	 */
3176 	skb_reset_mac_header(skb);
3177 	skb_reset_network_header(skb);
3178 	skb->protocol = htons(ETH_P_IP);
3179 	iph = skb_put(skb, sizeof(struct iphdr));
3180 	iph->protocol = ip_proto;
3181 	iph->saddr = src;
3182 	iph->daddr = dst;
3183 	iph->version = 0x4;
3184 	iph->frag_off = 0;
3185 	iph->ihl = 0x5;
3186 	skb_set_transport_header(skb, skb->len);
3187 
3188 	switch (iph->protocol) {
3189 	case IPPROTO_UDP: {
3190 		struct udphdr *udph;
3191 
3192 		udph = skb_put_zero(skb, sizeof(struct udphdr));
3193 		udph->source = sport;
3194 		udph->dest = dport;
3195 		udp_set_len_short(udph, sizeof(struct udphdr));
3196 		udph->check = 0;
3197 		break;
3198 	}
3199 	case IPPROTO_TCP: {
3200 		struct tcphdr *tcph;
3201 
3202 		tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3203 		tcph->source	= sport;
3204 		tcph->dest	= dport;
3205 		tcph->doff	= sizeof(struct tcphdr) / 4;
3206 		tcph->rst = 1;
3207 		tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3208 					    src, dst, 0);
3209 		break;
3210 	}
3211 	case IPPROTO_ICMP: {
3212 		struct icmphdr *icmph;
3213 
3214 		icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3215 		icmph->type = ICMP_ECHO;
3216 		icmph->code = 0;
3217 	}
3218 	}
3219 
3220 	return skb;
3221 }
3222 
3223 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3224 				       const struct nlmsghdr *nlh,
3225 				       struct nlattr **tb,
3226 				       struct netlink_ext_ack *extack)
3227 {
3228 	struct rtmsg *rtm;
3229 	int i, err;
3230 
3231 	rtm = nlmsg_payload(nlh, sizeof(*rtm));
3232 	if (!rtm) {
3233 		NL_SET_ERR_MSG(extack,
3234 			       "ipv4: Invalid header for route get request");
3235 		return -EINVAL;
3236 	}
3237 
3238 	if (!netlink_strict_get_check(skb))
3239 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3240 					      rtm_ipv4_policy, extack);
3241 
3242 	if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3243 	    (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3244 	    rtm->rtm_table || rtm->rtm_protocol ||
3245 	    rtm->rtm_scope || rtm->rtm_type) {
3246 		NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3247 		return -EINVAL;
3248 	}
3249 
3250 	if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3251 			       RTM_F_LOOKUP_TABLE |
3252 			       RTM_F_FIB_MATCH)) {
3253 		NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3254 		return -EINVAL;
3255 	}
3256 
3257 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3258 					    rtm_ipv4_policy, extack);
3259 	if (err)
3260 		return err;
3261 
3262 	if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3263 	    (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3264 		NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3265 		return -EINVAL;
3266 	}
3267 
3268 	for (i = 0; i <= RTA_MAX; i++) {
3269 		if (!tb[i])
3270 			continue;
3271 
3272 		switch (i) {
3273 		case RTA_IIF:
3274 		case RTA_OIF:
3275 		case RTA_SRC:
3276 		case RTA_DST:
3277 		case RTA_IP_PROTO:
3278 		case RTA_SPORT:
3279 		case RTA_DPORT:
3280 		case RTA_MARK:
3281 		case RTA_UID:
3282 			break;
3283 		default:
3284 			NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3285 			return -EINVAL;
3286 		}
3287 	}
3288 
3289 	return 0;
3290 }
3291 
3292 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3293 			     struct netlink_ext_ack *extack)
3294 {
3295 	struct net *net = sock_net(in_skb->sk);
3296 	struct nlattr *tb[RTA_MAX+1];
3297 	u32 table_id = RT_TABLE_MAIN;
3298 	__be16 sport = 0, dport = 0;
3299 	struct fib_result res = {};
3300 	u8 ip_proto = IPPROTO_UDP;
3301 	struct rtable *rt = NULL;
3302 	struct sk_buff *skb;
3303 	struct rtmsg *rtm;
3304 	struct flowi4 fl4 = {};
3305 	__be32 dst = 0;
3306 	__be32 src = 0;
3307 	dscp_t dscp;
3308 	kuid_t uid;
3309 	u32 iif;
3310 	int err;
3311 	int mark;
3312 
3313 	err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3314 	if (err < 0)
3315 		return err;
3316 
3317 	rtm = nlmsg_data(nlh);
3318 	src = nla_get_in_addr_default(tb[RTA_SRC], 0);
3319 	dst = nla_get_in_addr_default(tb[RTA_DST], 0);
3320 	iif = nla_get_u32_default(tb[RTA_IIF], 0);
3321 	mark = nla_get_u32_default(tb[RTA_MARK], 0);
3322 	dscp = inet_dsfield_to_dscp(rtm->rtm_tos);
3323 	if (tb[RTA_UID])
3324 		uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3325 	else
3326 		uid = (iif ? INVALID_UID : current_uid());
3327 
3328 	if (tb[RTA_IP_PROTO]) {
3329 		err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3330 						  &ip_proto, AF_INET, extack);
3331 		if (err)
3332 			return err;
3333 	}
3334 
3335 	if (tb[RTA_SPORT])
3336 		sport = nla_get_be16(tb[RTA_SPORT]);
3337 
3338 	if (tb[RTA_DPORT])
3339 		dport = nla_get_be16(tb[RTA_DPORT]);
3340 
3341 	skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3342 	if (!skb)
3343 		return -ENOBUFS;
3344 
3345 	fl4.daddr = dst;
3346 	fl4.saddr = src;
3347 	fl4.flowi4_dscp = dscp;
3348 	fl4.flowi4_oif = nla_get_u32_default(tb[RTA_OIF], 0);
3349 	fl4.flowi4_mark = mark;
3350 	fl4.flowi4_uid = uid;
3351 	if (sport)
3352 		fl4.fl4_sport = sport;
3353 	if (dport)
3354 		fl4.fl4_dport = dport;
3355 	fl4.flowi4_proto = ip_proto;
3356 
3357 	rcu_read_lock();
3358 
3359 	if (iif) {
3360 		struct net_device *dev;
3361 
3362 		dev = dev_get_by_index_rcu(net, iif);
3363 		if (!dev) {
3364 			err = -ENODEV;
3365 			goto errout_rcu;
3366 		}
3367 
3368 		fl4.flowi4_iif = iif; /* for rt_fill_info */
3369 		skb->dev	= dev;
3370 		skb->mark	= mark;
3371 		err = ip_route_input_rcu(skb, dst, src, dscp, dev,
3372 					 &res) ? -EINVAL : 0;
3373 
3374 		rt = skb_rtable(skb);
3375 		if (err == 0 && rt->dst.error)
3376 			err = -rt->dst.error;
3377 	} else {
3378 		fl4.flowi4_iif = LOOPBACK_IFINDEX;
3379 		skb->dev = net->loopback_dev;
3380 		rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3381 		err = 0;
3382 		if (IS_ERR(rt))
3383 			err = PTR_ERR(rt);
3384 		else
3385 			skb_dst_set(skb, &rt->dst);
3386 	}
3387 
3388 	if (err)
3389 		goto errout_rcu;
3390 
3391 	if (rtm->rtm_flags & RTM_F_NOTIFY)
3392 		rt->rt_flags |= RTCF_NOTIFY;
3393 
3394 	if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3395 		table_id = res.table ? res.table->tb_id : 0;
3396 
3397 	/* reset skb for netlink reply msg */
3398 	skb_trim(skb, 0);
3399 	skb_reset_network_header(skb);
3400 	skb_reset_transport_header(skb);
3401 	skb_reset_mac_header(skb);
3402 
3403 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3404 		struct fib_rt_info fri;
3405 
3406 		if (!res.fi) {
3407 			err = fib_props[res.type].error;
3408 			if (!err)
3409 				err = -EHOSTUNREACH;
3410 			goto errout_rcu;
3411 		}
3412 		fri.fi = res.fi;
3413 		fri.tb_id = table_id;
3414 		fri.dst = res.prefix;
3415 		fri.dst_len = res.prefixlen;
3416 		fri.dscp = res.dscp;
3417 		fri.type = rt->rt_type;
3418 		fri.offload = 0;
3419 		fri.trap = 0;
3420 		fri.offload_failed = 0;
3421 		if (res.fa_head) {
3422 			struct fib_alias *fa;
3423 
3424 			hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3425 				u8 slen = 32 - fri.dst_len;
3426 
3427 				if (fa->fa_slen == slen &&
3428 				    fa->tb_id == fri.tb_id &&
3429 				    fa->fa_dscp == fri.dscp &&
3430 				    fa->fa_info == res.fi &&
3431 				    fa->fa_type == fri.type) {
3432 					fri.offload = READ_ONCE(fa->offload);
3433 					fri.trap = READ_ONCE(fa->trap);
3434 					fri.offload_failed =
3435 						READ_ONCE(fa->offload_failed);
3436 					break;
3437 				}
3438 			}
3439 		}
3440 		err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3441 				    nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3442 	} else {
3443 		err = rt_fill_info(net, dst, src, rt, table_id, res.dscp, &fl4,
3444 				   skb, NETLINK_CB(in_skb).portid,
3445 				   nlh->nlmsg_seq, 0);
3446 	}
3447 	if (err < 0)
3448 		goto errout_rcu;
3449 
3450 	rcu_read_unlock();
3451 
3452 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3453 
3454 errout_free:
3455 	return err;
3456 errout_rcu:
3457 	rcu_read_unlock();
3458 	kfree_skb(skb);
3459 	goto errout_free;
3460 }
3461 
3462 void ip_rt_multicast_event(struct in_device *in_dev)
3463 {
3464 	rt_cache_flush(dev_net(in_dev->dev));
3465 }
3466 
3467 #ifdef CONFIG_SYSCTL
3468 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3469 static int ip_rt_gc_min_interval __read_mostly	= HZ / 2;
3470 static int ip_rt_gc_elasticity __read_mostly	= 8;
3471 static int ip_min_valid_pmtu __read_mostly	= IPV4_MIN_MTU;
3472 
3473 static int ipv4_sysctl_rtcache_flush(const struct ctl_table *__ctl, int write,
3474 		void *buffer, size_t *lenp, loff_t *ppos)
3475 {
3476 	struct net *net = (struct net *)__ctl->extra1;
3477 
3478 	if (write) {
3479 		rt_cache_flush(net);
3480 		fnhe_genid_bump(net);
3481 		return 0;
3482 	}
3483 
3484 	return -EINVAL;
3485 }
3486 
3487 static struct ctl_table ipv4_route_table[] = {
3488 	{
3489 		.procname	= "gc_thresh",
3490 		.data		= &ipv4_dst_ops.gc_thresh,
3491 		.maxlen		= sizeof(int),
3492 		.mode		= 0644,
3493 		.proc_handler	= proc_dointvec,
3494 	},
3495 	{
3496 		.procname	= "max_size",
3497 		.data		= &ip_rt_max_size,
3498 		.maxlen		= sizeof(int),
3499 		.mode		= 0644,
3500 		.proc_handler	= proc_dointvec,
3501 	},
3502 	{
3503 		/*  Deprecated. Use gc_min_interval_ms */
3504 
3505 		.procname	= "gc_min_interval",
3506 		.data		= &ip_rt_gc_min_interval,
3507 		.maxlen		= sizeof(int),
3508 		.mode		= 0644,
3509 		.proc_handler	= proc_dointvec_jiffies,
3510 	},
3511 	{
3512 		.procname	= "gc_min_interval_ms",
3513 		.data		= &ip_rt_gc_min_interval,
3514 		.maxlen		= sizeof(int),
3515 		.mode		= 0644,
3516 		.proc_handler	= proc_dointvec_ms_jiffies,
3517 	},
3518 	{
3519 		.procname	= "gc_timeout",
3520 		.data		= &ip_rt_gc_timeout,
3521 		.maxlen		= sizeof(int),
3522 		.mode		= 0644,
3523 		.proc_handler	= proc_dointvec_jiffies,
3524 	},
3525 	{
3526 		.procname	= "gc_interval",
3527 		.data		= &ip_rt_gc_interval,
3528 		.maxlen		= sizeof(int),
3529 		.mode		= 0644,
3530 		.proc_handler	= proc_dointvec_jiffies,
3531 	},
3532 	{
3533 		.procname	= "redirect_load",
3534 		.data		= &ip_rt_redirect_load,
3535 		.maxlen		= sizeof(int),
3536 		.mode		= 0644,
3537 		.proc_handler	= proc_dointvec,
3538 	},
3539 	{
3540 		.procname	= "redirect_number",
3541 		.data		= &ip_rt_redirect_number,
3542 		.maxlen		= sizeof(int),
3543 		.mode		= 0644,
3544 		.proc_handler	= proc_dointvec,
3545 	},
3546 	{
3547 		.procname	= "redirect_silence",
3548 		.data		= &ip_rt_redirect_silence,
3549 		.maxlen		= sizeof(int),
3550 		.mode		= 0644,
3551 		.proc_handler	= proc_dointvec,
3552 	},
3553 	{
3554 		.procname	= "error_cost",
3555 		.data		= &ip_rt_error_cost,
3556 		.maxlen		= sizeof(int),
3557 		.mode		= 0644,
3558 		.proc_handler	= proc_dointvec,
3559 	},
3560 	{
3561 		.procname	= "error_burst",
3562 		.data		= &ip_rt_error_burst,
3563 		.maxlen		= sizeof(int),
3564 		.mode		= 0644,
3565 		.proc_handler	= proc_dointvec,
3566 	},
3567 	{
3568 		.procname	= "gc_elasticity",
3569 		.data		= &ip_rt_gc_elasticity,
3570 		.maxlen		= sizeof(int),
3571 		.mode		= 0644,
3572 		.proc_handler	= proc_dointvec,
3573 	},
3574 };
3575 
3576 static const char ipv4_route_flush_procname[] = "flush";
3577 
3578 static struct ctl_table ipv4_route_netns_table[] = {
3579 	{
3580 		.procname	= ipv4_route_flush_procname,
3581 		.maxlen		= sizeof(int),
3582 		.mode		= 0200,
3583 		.proc_handler	= ipv4_sysctl_rtcache_flush,
3584 	},
3585 	{
3586 		.procname       = "min_pmtu",
3587 		.data           = &init_net.ipv4.ip_rt_min_pmtu,
3588 		.maxlen         = sizeof(int),
3589 		.mode           = 0644,
3590 		.proc_handler   = proc_dointvec_minmax,
3591 		.extra1         = &ip_min_valid_pmtu,
3592 	},
3593 	{
3594 		.procname       = "mtu_expires",
3595 		.data           = &init_net.ipv4.ip_rt_mtu_expires,
3596 		.maxlen         = sizeof(int),
3597 		.mode           = 0644,
3598 		.proc_handler   = proc_dointvec_jiffies,
3599 	},
3600 	{
3601 		.procname   = "min_adv_mss",
3602 		.data       = &init_net.ipv4.ip_rt_min_advmss,
3603 		.maxlen     = sizeof(int),
3604 		.mode       = 0644,
3605 		.proc_handler   = proc_dointvec,
3606 	},
3607 };
3608 
3609 static __net_init int sysctl_route_net_init(struct net *net)
3610 {
3611 	struct ctl_table *tbl;
3612 	size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3613 
3614 	tbl = ipv4_route_netns_table;
3615 	if (!net_eq(net, &init_net)) {
3616 		int i;
3617 
3618 		tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3619 		if (!tbl)
3620 			goto err_dup;
3621 
3622 		/* Don't export non-whitelisted sysctls to unprivileged users */
3623 		if (net->user_ns != &init_user_ns) {
3624 			if (tbl[0].procname != ipv4_route_flush_procname)
3625 				table_size = 0;
3626 		}
3627 
3628 		/* Update the variables to point into the current struct net
3629 		 * except for the first element flush
3630 		 */
3631 		for (i = 1; i < table_size; i++)
3632 			tbl[i].data += (void *)net - (void *)&init_net;
3633 	}
3634 	tbl[0].extra1 = net;
3635 
3636 	net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3637 						     tbl, table_size);
3638 	if (!net->ipv4.route_hdr)
3639 		goto err_reg;
3640 	return 0;
3641 
3642 err_reg:
3643 	if (tbl != ipv4_route_netns_table)
3644 		kfree(tbl);
3645 err_dup:
3646 	return -ENOMEM;
3647 }
3648 
3649 static __net_exit void sysctl_route_net_exit(struct net *net)
3650 {
3651 	const struct ctl_table *tbl;
3652 
3653 	tbl = net->ipv4.route_hdr->ctl_table_arg;
3654 	unregister_net_sysctl_table(net->ipv4.route_hdr);
3655 	BUG_ON(tbl == ipv4_route_netns_table);
3656 	kfree(tbl);
3657 }
3658 
3659 static __net_initdata struct pernet_operations sysctl_route_ops = {
3660 	.init = sysctl_route_net_init,
3661 	.exit = sysctl_route_net_exit,
3662 };
3663 #endif
3664 
3665 static __net_init int netns_ip_rt_init(struct net *net)
3666 {
3667 	/* Set default value for namespaceified sysctls */
3668 	net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3669 	net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3670 	net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3671 	return 0;
3672 }
3673 
3674 static struct pernet_operations __net_initdata ip_rt_ops = {
3675 	.init = netns_ip_rt_init,
3676 };
3677 
3678 static __net_init int rt_genid_init(struct net *net)
3679 {
3680 	atomic_set(&net->ipv4.rt_genid, 0);
3681 	atomic_set(&net->fnhe_genid, 0);
3682 	atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3683 	return 0;
3684 }
3685 
3686 static __net_initdata struct pernet_operations rt_genid_ops = {
3687 	.init = rt_genid_init,
3688 };
3689 
3690 static int __net_init ipv4_inetpeer_init(struct net *net)
3691 {
3692 	struct inet_peer_base *bp = kmalloc_obj(*bp);
3693 
3694 	if (!bp)
3695 		return -ENOMEM;
3696 	inet_peer_base_init(bp);
3697 	net->ipv4.peers = bp;
3698 	return 0;
3699 }
3700 
3701 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3702 {
3703 	struct inet_peer_base *bp = net->ipv4.peers;
3704 
3705 	net->ipv4.peers = NULL;
3706 	inetpeer_invalidate_tree(bp);
3707 	kfree(bp);
3708 }
3709 
3710 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3711 	.init	=	ipv4_inetpeer_init,
3712 	.exit	=	ipv4_inetpeer_exit,
3713 };
3714 
3715 #ifdef CONFIG_IP_ROUTE_CLASSID
3716 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3717 #endif /* CONFIG_IP_ROUTE_CLASSID */
3718 
3719 static const struct rtnl_msg_handler ip_rt_rtnl_msg_handlers[] __initconst = {
3720 	{.protocol = PF_INET, .msgtype = RTM_GETROUTE,
3721 	 .doit = inet_rtm_getroute, .flags = RTNL_FLAG_DOIT_UNLOCKED},
3722 };
3723 
3724 int __init ip_rt_init(void)
3725 {
3726 	void *idents_hash;
3727 	int cpu;
3728 
3729 	/* For modern hosts, this will use 2 MB of memory */
3730 	idents_hash = alloc_large_system_hash("IP idents",
3731 					      sizeof(*ip_idents) + sizeof(*ip_tstamps),
3732 					      0,
3733 					      16, /* one bucket per 64 KB */
3734 					      HASH_ZERO,
3735 					      NULL,
3736 					      &ip_idents_mask,
3737 					      2048,
3738 					      256*1024);
3739 
3740 	ip_idents = idents_hash;
3741 
3742 	get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3743 
3744 	ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3745 
3746 	for_each_possible_cpu(cpu) {
3747 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3748 
3749 		INIT_LIST_HEAD(&ul->head);
3750 		spin_lock_init(&ul->lock);
3751 	}
3752 #ifdef CONFIG_IP_ROUTE_CLASSID
3753 	ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3754 	if (!ip_rt_acct)
3755 		panic("IP: failed to allocate ip_rt_acct\n");
3756 #endif
3757 
3758 	ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable,
3759 					      SLAB_HWCACHE_ALIGN | SLAB_PANIC);
3760 
3761 	ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3762 
3763 	if (dst_entries_init(&ipv4_dst_ops) < 0)
3764 		panic("IP: failed to allocate ipv4_dst_ops counter\n");
3765 
3766 	if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3767 		panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3768 
3769 	ipv4_dst_ops.gc_thresh = ~0;
3770 	ip_rt_max_size = INT_MAX;
3771 
3772 	devinet_init();
3773 	ip_fib_init();
3774 
3775 	if (ip_rt_proc_init())
3776 		pr_err("Unable to create route proc files\n");
3777 #ifdef CONFIG_XFRM
3778 	xfrm_init();
3779 	xfrm4_init();
3780 #endif
3781 	rtnl_register_many(ip_rt_rtnl_msg_handlers);
3782 
3783 #ifdef CONFIG_SYSCTL
3784 	register_pernet_subsys(&sysctl_route_ops);
3785 #endif
3786 	register_pernet_subsys(&ip_rt_ops);
3787 	register_pernet_subsys(&rt_genid_ops);
3788 	register_pernet_subsys(&ipv4_inetpeer_ops);
3789 	return 0;
3790 }
3791 
3792 #ifdef CONFIG_SYSCTL
3793 /*
3794  * We really need to sanitize the damn ipv4 init order, then all
3795  * this nonsense will go away.
3796  */
3797 void __init ip_static_sysctl_init(void)
3798 {
3799 	register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3800 }
3801 #endif
3802