xref: /openbmc/linux/net/ipv4/route.c (revision 2ae1beb3)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		ROUTE - implementation of the IP router.
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
12  *		Linus Torvalds, <Linus.Torvalds@helsinki.fi>
13  *		Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
14  *
15  * Fixes:
16  *		Alan Cox	:	Verify area fixes.
17  *		Alan Cox	:	cli() protects routing changes
18  *		Rui Oliveira	:	ICMP routing table updates
19  *		(rco@di.uminho.pt)	Routing table insertion and update
20  *		Linus Torvalds	:	Rewrote bits to be sensible
21  *		Alan Cox	:	Added BSD route gw semantics
22  *		Alan Cox	:	Super /proc >4K
23  *		Alan Cox	:	MTU in route table
24  *		Alan Cox	:	MSS actually. Also added the window
25  *					clamper.
26  *		Sam Lantinga	:	Fixed route matching in rt_del()
27  *		Alan Cox	:	Routing cache support.
28  *		Alan Cox	:	Removed compatibility cruft.
29  *		Alan Cox	:	RTF_REJECT support.
30  *		Alan Cox	:	TCP irtt support.
31  *		Jonathan Naylor	:	Added Metric support.
32  *	Miquel van Smoorenburg	:	BSD API fixes.
33  *	Miquel van Smoorenburg	:	Metrics.
34  *		Alan Cox	:	Use __u32 properly
35  *		Alan Cox	:	Aligned routing errors more closely with BSD
36  *					our system is still very different.
37  *		Alan Cox	:	Faster /proc handling
38  *	Alexey Kuznetsov	:	Massive rework to support tree based routing,
39  *					routing caches and better behaviour.
40  *
41  *		Olaf Erb	:	irtt wasn't being copied right.
42  *		Bjorn Ekwall	:	Kerneld route support.
43  *		Alan Cox	:	Multicast fixed (I hope)
44  *		Pavel Krauz	:	Limited broadcast fixed
45  *		Mike McLagan	:	Routing by source
46  *	Alexey Kuznetsov	:	End of old history. Split to fib.c and
47  *					route.c and rewritten from scratch.
48  *		Andi Kleen	:	Load-limit warning messages.
49  *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
50  *	Vitaly E. Lavrov	:	Race condition in ip_route_input_slow.
51  *	Tobias Ringstrom	:	Uninitialized res.type in ip_route_output_slow.
52  *	Vladimir V. Ivanov	:	IP rule info (flowid) is really useful.
53  *		Marc Boucher	:	routing by fwmark
54  *	Robert Olsson		:	Added rt_cache statistics
55  *	Arnaldo C. Melo		:	Convert proc stuff to seq_file
56  *	Eric Dumazet		:	hashed spinlocks and rt_check_expire() fixes.
57  *	Ilia Sotnikov		:	Ignore TOS on PMTUD and Redirect
58  *	Ilia Sotnikov		:	Removed TOS from hash calculations
59  */
60 
61 #define pr_fmt(fmt) "IPv4: " fmt
62 
63 #include <linux/module.h>
64 #include <linux/bitops.h>
65 #include <linux/kernel.h>
66 #include <linux/mm.h>
67 #include <linux/memblock.h>
68 #include <linux/socket.h>
69 #include <linux/errno.h>
70 #include <linux/in.h>
71 #include <linux/inet.h>
72 #include <linux/netdevice.h>
73 #include <linux/proc_fs.h>
74 #include <linux/init.h>
75 #include <linux/skbuff.h>
76 #include <linux/inetdevice.h>
77 #include <linux/igmp.h>
78 #include <linux/pkt_sched.h>
79 #include <linux/mroute.h>
80 #include <linux/netfilter_ipv4.h>
81 #include <linux/random.h>
82 #include <linux/rcupdate.h>
83 #include <linux/slab.h>
84 #include <linux/jhash.h>
85 #include <net/dst.h>
86 #include <net/dst_metadata.h>
87 #include <net/inet_dscp.h>
88 #include <net/net_namespace.h>
89 #include <net/ip.h>
90 #include <net/route.h>
91 #include <net/inetpeer.h>
92 #include <net/sock.h>
93 #include <net/ip_fib.h>
94 #include <net/nexthop.h>
95 #include <net/tcp.h>
96 #include <net/icmp.h>
97 #include <net/xfrm.h>
98 #include <net/lwtunnel.h>
99 #include <net/netevent.h>
100 #include <net/rtnetlink.h>
101 #ifdef CONFIG_SYSCTL
102 #include <linux/sysctl.h>
103 #endif
104 #include <net/secure_seq.h>
105 #include <net/ip_tunnels.h>
106 
107 #include "fib_lookup.h"
108 
109 #define RT_FL_TOS(oldflp4) \
110 	((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
111 
112 #define RT_GC_TIMEOUT (300*HZ)
113 
114 #define DEFAULT_MIN_PMTU (512 + 20 + 20)
115 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
116 #define DEFAULT_MIN_ADVMSS 256
117 static int ip_rt_max_size;
118 static int ip_rt_redirect_number __read_mostly	= 9;
119 static int ip_rt_redirect_load __read_mostly	= HZ / 50;
120 static int ip_rt_redirect_silence __read_mostly	= ((HZ / 50) << (9 + 1));
121 static int ip_rt_error_cost __read_mostly	= HZ;
122 static int ip_rt_error_burst __read_mostly	= 5 * HZ;
123 
124 static int ip_rt_gc_timeout __read_mostly	= RT_GC_TIMEOUT;
125 
126 /*
127  *	Interface to generic destination cache.
128  */
129 
130 INDIRECT_CALLABLE_SCOPE
131 struct dst_entry	*ipv4_dst_check(struct dst_entry *dst, u32 cookie);
132 static unsigned int	 ipv4_default_advmss(const struct dst_entry *dst);
133 INDIRECT_CALLABLE_SCOPE
134 unsigned int		ipv4_mtu(const struct dst_entry *dst);
135 static void		ipv4_negative_advice(struct sock *sk,
136 					     struct dst_entry *dst);
137 static void		 ipv4_link_failure(struct sk_buff *skb);
138 static void		 ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
139 					   struct sk_buff *skb, u32 mtu,
140 					   bool confirm_neigh);
141 static void		 ip_do_redirect(struct dst_entry *dst, struct sock *sk,
142 					struct sk_buff *skb);
143 static void		ipv4_dst_destroy(struct dst_entry *dst);
144 
145 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
146 {
147 	WARN_ON(1);
148 	return NULL;
149 }
150 
151 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
152 					   struct sk_buff *skb,
153 					   const void *daddr);
154 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
155 
156 static struct dst_ops ipv4_dst_ops = {
157 	.family =		AF_INET,
158 	.check =		ipv4_dst_check,
159 	.default_advmss =	ipv4_default_advmss,
160 	.mtu =			ipv4_mtu,
161 	.cow_metrics =		ipv4_cow_metrics,
162 	.destroy =		ipv4_dst_destroy,
163 	.negative_advice =	ipv4_negative_advice,
164 	.link_failure =		ipv4_link_failure,
165 	.update_pmtu =		ip_rt_update_pmtu,
166 	.redirect =		ip_do_redirect,
167 	.local_out =		__ip_local_out,
168 	.neigh_lookup =		ipv4_neigh_lookup,
169 	.confirm_neigh =	ipv4_confirm_neigh,
170 };
171 
172 #define ECN_OR_COST(class)	TC_PRIO_##class
173 
174 const __u8 ip_tos2prio[16] = {
175 	TC_PRIO_BESTEFFORT,
176 	ECN_OR_COST(BESTEFFORT),
177 	TC_PRIO_BESTEFFORT,
178 	ECN_OR_COST(BESTEFFORT),
179 	TC_PRIO_BULK,
180 	ECN_OR_COST(BULK),
181 	TC_PRIO_BULK,
182 	ECN_OR_COST(BULK),
183 	TC_PRIO_INTERACTIVE,
184 	ECN_OR_COST(INTERACTIVE),
185 	TC_PRIO_INTERACTIVE,
186 	ECN_OR_COST(INTERACTIVE),
187 	TC_PRIO_INTERACTIVE_BULK,
188 	ECN_OR_COST(INTERACTIVE_BULK),
189 	TC_PRIO_INTERACTIVE_BULK,
190 	ECN_OR_COST(INTERACTIVE_BULK)
191 };
192 EXPORT_SYMBOL(ip_tos2prio);
193 
194 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
195 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
196 
197 #ifdef CONFIG_PROC_FS
198 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
199 {
200 	if (*pos)
201 		return NULL;
202 	return SEQ_START_TOKEN;
203 }
204 
205 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
206 {
207 	++*pos;
208 	return NULL;
209 }
210 
211 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
212 {
213 }
214 
215 static int rt_cache_seq_show(struct seq_file *seq, void *v)
216 {
217 	if (v == SEQ_START_TOKEN)
218 		seq_printf(seq, "%-127s\n",
219 			   "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
220 			   "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
221 			   "HHUptod\tSpecDst");
222 	return 0;
223 }
224 
225 static const struct seq_operations rt_cache_seq_ops = {
226 	.start  = rt_cache_seq_start,
227 	.next   = rt_cache_seq_next,
228 	.stop   = rt_cache_seq_stop,
229 	.show   = rt_cache_seq_show,
230 };
231 
232 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
233 {
234 	int cpu;
235 
236 	if (*pos == 0)
237 		return SEQ_START_TOKEN;
238 
239 	for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
240 		if (!cpu_possible(cpu))
241 			continue;
242 		*pos = cpu+1;
243 		return &per_cpu(rt_cache_stat, cpu);
244 	}
245 	return NULL;
246 }
247 
248 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
249 {
250 	int cpu;
251 
252 	for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
253 		if (!cpu_possible(cpu))
254 			continue;
255 		*pos = cpu+1;
256 		return &per_cpu(rt_cache_stat, cpu);
257 	}
258 	(*pos)++;
259 	return NULL;
260 
261 }
262 
263 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
264 {
265 
266 }
267 
268 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
269 {
270 	struct rt_cache_stat *st = v;
271 
272 	if (v == SEQ_START_TOKEN) {
273 		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");
274 		return 0;
275 	}
276 
277 	seq_printf(seq, "%08x %08x %08x    %08x   %08x    %08x %08x       "
278 			"%08x       %08x %08x     %08x    %08x %08x   "
279 			"%08x     %08x        %08x        %08x\n",
280 		   dst_entries_get_slow(&ipv4_dst_ops),
281 		   0, /* st->in_hit */
282 		   st->in_slow_tot,
283 		   st->in_slow_mc,
284 		   st->in_no_route,
285 		   st->in_brd,
286 		   st->in_martian_dst,
287 		   st->in_martian_src,
288 
289 		   0, /* st->out_hit */
290 		   st->out_slow_tot,
291 		   st->out_slow_mc,
292 
293 		   0, /* st->gc_total */
294 		   0, /* st->gc_ignored */
295 		   0, /* st->gc_goal_miss */
296 		   0, /* st->gc_dst_overflow */
297 		   0, /* st->in_hlist_search */
298 		   0  /* st->out_hlist_search */
299 		);
300 	return 0;
301 }
302 
303 static const struct seq_operations rt_cpu_seq_ops = {
304 	.start  = rt_cpu_seq_start,
305 	.next   = rt_cpu_seq_next,
306 	.stop   = rt_cpu_seq_stop,
307 	.show   = rt_cpu_seq_show,
308 };
309 
310 #ifdef CONFIG_IP_ROUTE_CLASSID
311 static int rt_acct_proc_show(struct seq_file *m, void *v)
312 {
313 	struct ip_rt_acct *dst, *src;
314 	unsigned int i, j;
315 
316 	dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
317 	if (!dst)
318 		return -ENOMEM;
319 
320 	for_each_possible_cpu(i) {
321 		src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
322 		for (j = 0; j < 256; j++) {
323 			dst[j].o_bytes   += src[j].o_bytes;
324 			dst[j].o_packets += src[j].o_packets;
325 			dst[j].i_bytes   += src[j].i_bytes;
326 			dst[j].i_packets += src[j].i_packets;
327 		}
328 	}
329 
330 	seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
331 	kfree(dst);
332 	return 0;
333 }
334 #endif
335 
336 static int __net_init ip_rt_do_proc_init(struct net *net)
337 {
338 	struct proc_dir_entry *pde;
339 
340 	pde = proc_create_seq("rt_cache", 0444, net->proc_net,
341 			      &rt_cache_seq_ops);
342 	if (!pde)
343 		goto err1;
344 
345 	pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
346 			      &rt_cpu_seq_ops);
347 	if (!pde)
348 		goto err2;
349 
350 #ifdef CONFIG_IP_ROUTE_CLASSID
351 	pde = proc_create_single("rt_acct", 0, net->proc_net,
352 			rt_acct_proc_show);
353 	if (!pde)
354 		goto err3;
355 #endif
356 	return 0;
357 
358 #ifdef CONFIG_IP_ROUTE_CLASSID
359 err3:
360 	remove_proc_entry("rt_cache", net->proc_net_stat);
361 #endif
362 err2:
363 	remove_proc_entry("rt_cache", net->proc_net);
364 err1:
365 	return -ENOMEM;
366 }
367 
368 static void __net_exit ip_rt_do_proc_exit(struct net *net)
369 {
370 	remove_proc_entry("rt_cache", net->proc_net_stat);
371 	remove_proc_entry("rt_cache", net->proc_net);
372 #ifdef CONFIG_IP_ROUTE_CLASSID
373 	remove_proc_entry("rt_acct", net->proc_net);
374 #endif
375 }
376 
377 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
378 	.init = ip_rt_do_proc_init,
379 	.exit = ip_rt_do_proc_exit,
380 };
381 
382 static int __init ip_rt_proc_init(void)
383 {
384 	return register_pernet_subsys(&ip_rt_proc_ops);
385 }
386 
387 #else
388 static inline int ip_rt_proc_init(void)
389 {
390 	return 0;
391 }
392 #endif /* CONFIG_PROC_FS */
393 
394 static inline bool rt_is_expired(const struct rtable *rth)
395 {
396 	return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
397 }
398 
399 void rt_cache_flush(struct net *net)
400 {
401 	rt_genid_bump_ipv4(net);
402 }
403 
404 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
405 					   struct sk_buff *skb,
406 					   const void *daddr)
407 {
408 	const struct rtable *rt = container_of(dst, struct rtable, dst);
409 	struct net_device *dev = dst->dev;
410 	struct neighbour *n;
411 
412 	rcu_read_lock();
413 
414 	if (likely(rt->rt_gw_family == AF_INET)) {
415 		n = ip_neigh_gw4(dev, rt->rt_gw4);
416 	} else if (rt->rt_gw_family == AF_INET6) {
417 		n = ip_neigh_gw6(dev, &rt->rt_gw6);
418         } else {
419 		__be32 pkey;
420 
421 		pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
422 		n = ip_neigh_gw4(dev, pkey);
423 	}
424 
425 	if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
426 		n = NULL;
427 
428 	rcu_read_unlock();
429 
430 	return n;
431 }
432 
433 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
434 {
435 	const struct rtable *rt = container_of(dst, struct rtable, dst);
436 	struct net_device *dev = dst->dev;
437 	const __be32 *pkey = daddr;
438 
439 	if (rt->rt_gw_family == AF_INET) {
440 		pkey = (const __be32 *)&rt->rt_gw4;
441 	} else if (rt->rt_gw_family == AF_INET6) {
442 		return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
443 	} else if (!daddr ||
444 		 (rt->rt_flags &
445 		  (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
446 		return;
447 	}
448 	__ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
449 }
450 
451 /* Hash tables of size 2048..262144 depending on RAM size.
452  * Each bucket uses 8 bytes.
453  */
454 static u32 ip_idents_mask __read_mostly;
455 static atomic_t *ip_idents __read_mostly;
456 static u32 *ip_tstamps __read_mostly;
457 
458 /* In order to protect privacy, we add a perturbation to identifiers
459  * if one generator is seldom used. This makes hard for an attacker
460  * to infer how many packets were sent between two points in time.
461  */
462 static u32 ip_idents_reserve(u32 hash, int segs)
463 {
464 	u32 bucket, old, now = (u32)jiffies;
465 	atomic_t *p_id;
466 	u32 *p_tstamp;
467 	u32 delta = 0;
468 
469 	bucket = hash & ip_idents_mask;
470 	p_tstamp = ip_tstamps + bucket;
471 	p_id = ip_idents + bucket;
472 	old = READ_ONCE(*p_tstamp);
473 
474 	if (old != now && cmpxchg(p_tstamp, old, now) == old)
475 		delta = get_random_u32_below(now - old);
476 
477 	/* If UBSAN reports an error there, please make sure your compiler
478 	 * supports -fno-strict-overflow before reporting it that was a bug
479 	 * in UBSAN, and it has been fixed in GCC-8.
480 	 */
481 	return atomic_add_return(segs + delta, p_id) - segs;
482 }
483 
484 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
485 {
486 	u32 hash, id;
487 
488 	/* Note the following code is not safe, but this is okay. */
489 	if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
490 		get_random_bytes(&net->ipv4.ip_id_key,
491 				 sizeof(net->ipv4.ip_id_key));
492 
493 	hash = siphash_3u32((__force u32)iph->daddr,
494 			    (__force u32)iph->saddr,
495 			    iph->protocol,
496 			    &net->ipv4.ip_id_key);
497 	id = ip_idents_reserve(hash, segs);
498 	iph->id = htons(id);
499 }
500 EXPORT_SYMBOL(__ip_select_ident);
501 
502 static void ip_rt_fix_tos(struct flowi4 *fl4)
503 {
504 	__u8 tos = RT_FL_TOS(fl4);
505 
506 	fl4->flowi4_tos = tos & IPTOS_RT_MASK;
507 	if (tos & RTO_ONLINK)
508 		fl4->flowi4_scope = RT_SCOPE_LINK;
509 }
510 
511 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
512 			     const struct sock *sk, const struct iphdr *iph,
513 			     int oif, __u8 tos, u8 prot, u32 mark,
514 			     int flow_flags)
515 {
516 	__u8 scope = RT_SCOPE_UNIVERSE;
517 
518 	if (sk) {
519 		oif = sk->sk_bound_dev_if;
520 		mark = READ_ONCE(sk->sk_mark);
521 		tos = ip_sock_rt_tos(sk);
522 		scope = ip_sock_rt_scope(sk);
523 		prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW :
524 						    sk->sk_protocol;
525 	}
526 
527 	flowi4_init_output(fl4, oif, mark, tos & IPTOS_RT_MASK, scope,
528 			   prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
529 			   sock_net_uid(net, sk));
530 }
531 
532 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
533 			       const struct sock *sk)
534 {
535 	const struct net *net = dev_net(skb->dev);
536 	const struct iphdr *iph = ip_hdr(skb);
537 	int oif = skb->dev->ifindex;
538 	u8 prot = iph->protocol;
539 	u32 mark = skb->mark;
540 	__u8 tos = iph->tos;
541 
542 	__build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
543 }
544 
545 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
546 {
547 	const struct inet_sock *inet = inet_sk(sk);
548 	const struct ip_options_rcu *inet_opt;
549 	__be32 daddr = inet->inet_daddr;
550 
551 	rcu_read_lock();
552 	inet_opt = rcu_dereference(inet->inet_opt);
553 	if (inet_opt && inet_opt->opt.srr)
554 		daddr = inet_opt->opt.faddr;
555 	flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark),
556 			   ip_sock_rt_tos(sk) & IPTOS_RT_MASK,
557 			   ip_sock_rt_scope(sk),
558 			   inet_test_bit(HDRINCL, sk) ?
559 				IPPROTO_RAW : sk->sk_protocol,
560 			   inet_sk_flowi_flags(sk),
561 			   daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
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 	fnhe_flush_routes(oldest);
611 	*oldest_p = oldest->fnhe_next;
612 	kfree_rcu(oldest, rcu);
613 }
614 
615 static u32 fnhe_hashfun(__be32 daddr)
616 {
617 	static siphash_aligned_key_t fnhe_hash_key;
618 	u64 hval;
619 
620 	net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
621 	hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
622 	return hash_64(hval, FNHE_HASH_SHIFT);
623 }
624 
625 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
626 {
627 	rt->rt_pmtu = fnhe->fnhe_pmtu;
628 	rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
629 	rt->dst.expires = fnhe->fnhe_expires;
630 
631 	if (fnhe->fnhe_gw) {
632 		rt->rt_flags |= RTCF_REDIRECTED;
633 		rt->rt_uses_gateway = 1;
634 		rt->rt_gw_family = AF_INET;
635 		rt->rt_gw4 = fnhe->fnhe_gw;
636 	}
637 }
638 
639 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
640 				  __be32 gw, u32 pmtu, bool lock,
641 				  unsigned long expires)
642 {
643 	struct fnhe_hash_bucket *hash;
644 	struct fib_nh_exception *fnhe;
645 	struct rtable *rt;
646 	u32 genid, hval;
647 	unsigned int i;
648 	int depth;
649 
650 	genid = fnhe_genid(dev_net(nhc->nhc_dev));
651 	hval = fnhe_hashfun(daddr);
652 
653 	spin_lock_bh(&fnhe_lock);
654 
655 	hash = rcu_dereference(nhc->nhc_exceptions);
656 	if (!hash) {
657 		hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
658 		if (!hash)
659 			goto out_unlock;
660 		rcu_assign_pointer(nhc->nhc_exceptions, hash);
661 	}
662 
663 	hash += hval;
664 
665 	depth = 0;
666 	for (fnhe = rcu_dereference(hash->chain); fnhe;
667 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
668 		if (fnhe->fnhe_daddr == daddr)
669 			break;
670 		depth++;
671 	}
672 
673 	if (fnhe) {
674 		if (fnhe->fnhe_genid != genid)
675 			fnhe->fnhe_genid = genid;
676 		if (gw)
677 			fnhe->fnhe_gw = gw;
678 		if (pmtu) {
679 			fnhe->fnhe_pmtu = pmtu;
680 			fnhe->fnhe_mtu_locked = lock;
681 		}
682 		fnhe->fnhe_expires = max(1UL, expires);
683 		/* Update all cached dsts too */
684 		rt = rcu_dereference(fnhe->fnhe_rth_input);
685 		if (rt)
686 			fill_route_from_fnhe(rt, fnhe);
687 		rt = rcu_dereference(fnhe->fnhe_rth_output);
688 		if (rt)
689 			fill_route_from_fnhe(rt, fnhe);
690 	} else {
691 		/* Randomize max depth to avoid some side channels attacks. */
692 		int max_depth = FNHE_RECLAIM_DEPTH +
693 				get_random_u32_below(FNHE_RECLAIM_DEPTH);
694 
695 		while (depth > max_depth) {
696 			fnhe_remove_oldest(hash);
697 			depth--;
698 		}
699 
700 		fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
701 		if (!fnhe)
702 			goto out_unlock;
703 
704 		fnhe->fnhe_next = hash->chain;
705 
706 		fnhe->fnhe_genid = genid;
707 		fnhe->fnhe_daddr = daddr;
708 		fnhe->fnhe_gw = gw;
709 		fnhe->fnhe_pmtu = pmtu;
710 		fnhe->fnhe_mtu_locked = lock;
711 		fnhe->fnhe_expires = max(1UL, expires);
712 
713 		rcu_assign_pointer(hash->chain, fnhe);
714 
715 		/* Exception created; mark the cached routes for the nexthop
716 		 * stale, so anyone caching it rechecks if this exception
717 		 * applies to them.
718 		 */
719 		rt = rcu_dereference(nhc->nhc_rth_input);
720 		if (rt)
721 			rt->dst.obsolete = DST_OBSOLETE_KILL;
722 
723 		for_each_possible_cpu(i) {
724 			struct rtable __rcu **prt;
725 
726 			prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
727 			rt = rcu_dereference(*prt);
728 			if (rt)
729 				rt->dst.obsolete = DST_OBSOLETE_KILL;
730 		}
731 	}
732 
733 	fnhe->fnhe_stamp = jiffies;
734 
735 out_unlock:
736 	spin_unlock_bh(&fnhe_lock);
737 }
738 
739 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
740 			     bool kill_route)
741 {
742 	__be32 new_gw = icmp_hdr(skb)->un.gateway;
743 	__be32 old_gw = ip_hdr(skb)->saddr;
744 	struct net_device *dev = skb->dev;
745 	struct in_device *in_dev;
746 	struct fib_result res;
747 	struct neighbour *n;
748 	struct net *net;
749 
750 	switch (icmp_hdr(skb)->code & 7) {
751 	case ICMP_REDIR_NET:
752 	case ICMP_REDIR_NETTOS:
753 	case ICMP_REDIR_HOST:
754 	case ICMP_REDIR_HOSTTOS:
755 		break;
756 
757 	default:
758 		return;
759 	}
760 
761 	if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
762 		return;
763 
764 	in_dev = __in_dev_get_rcu(dev);
765 	if (!in_dev)
766 		return;
767 
768 	net = dev_net(dev);
769 	if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
770 	    ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
771 	    ipv4_is_zeronet(new_gw))
772 		goto reject_redirect;
773 
774 	if (!IN_DEV_SHARED_MEDIA(in_dev)) {
775 		if (!inet_addr_onlink(in_dev, new_gw, old_gw))
776 			goto reject_redirect;
777 		if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
778 			goto reject_redirect;
779 	} else {
780 		if (inet_addr_type(net, new_gw) != RTN_UNICAST)
781 			goto reject_redirect;
782 	}
783 
784 	n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw);
785 	if (!n)
786 		n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
787 	if (!IS_ERR(n)) {
788 		if (!(READ_ONCE(n->nud_state) & NUD_VALID)) {
789 			neigh_event_send(n, NULL);
790 		} else {
791 			if (fib_lookup(net, fl4, &res, 0) == 0) {
792 				struct fib_nh_common *nhc;
793 
794 				fib_select_path(net, &res, fl4, skb);
795 				nhc = FIB_RES_NHC(res);
796 				update_or_create_fnhe(nhc, fl4->daddr, new_gw,
797 						0, false,
798 						jiffies + ip_rt_gc_timeout);
799 			}
800 			if (kill_route)
801 				rt->dst.obsolete = DST_OBSOLETE_KILL;
802 			call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
803 		}
804 		neigh_release(n);
805 	}
806 	return;
807 
808 reject_redirect:
809 #ifdef CONFIG_IP_ROUTE_VERBOSE
810 	if (IN_DEV_LOG_MARTIANS(in_dev)) {
811 		const struct iphdr *iph = (const struct iphdr *) skb->data;
812 		__be32 daddr = iph->daddr;
813 		__be32 saddr = iph->saddr;
814 
815 		net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
816 				     "  Advised path = %pI4 -> %pI4\n",
817 				     &old_gw, dev->name, &new_gw,
818 				     &saddr, &daddr);
819 	}
820 #endif
821 	;
822 }
823 
824 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
825 {
826 	struct rtable *rt;
827 	struct flowi4 fl4;
828 	const struct iphdr *iph = (const struct iphdr *) skb->data;
829 	struct net *net = dev_net(skb->dev);
830 	int oif = skb->dev->ifindex;
831 	u8 prot = iph->protocol;
832 	u32 mark = skb->mark;
833 	__u8 tos = iph->tos;
834 
835 	rt = (struct rtable *) dst;
836 
837 	__build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
838 	__ip_do_redirect(rt, skb, &fl4, true);
839 }
840 
841 static void ipv4_negative_advice(struct sock *sk,
842 				 struct dst_entry *dst)
843 {
844 	struct rtable *rt = (struct rtable *)dst;
845 
846 	if ((dst->obsolete > 0) ||
847 	    (rt->rt_flags & RTCF_REDIRECTED) ||
848 	    rt->dst.expires)
849 		sk_dst_reset(sk);
850 }
851 
852 /*
853  * Algorithm:
854  *	1. The first ip_rt_redirect_number redirects are sent
855  *	   with exponential backoff, then we stop sending them at all,
856  *	   assuming that the host ignores our redirects.
857  *	2. If we did not see packets requiring redirects
858  *	   during ip_rt_redirect_silence, we assume that the host
859  *	   forgot redirected route and start to send redirects again.
860  *
861  * This algorithm is much cheaper and more intelligent than dumb load limiting
862  * in icmp.c.
863  *
864  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
865  * and "frag. need" (breaks PMTU discovery) in icmp.c.
866  */
867 
868 void ip_rt_send_redirect(struct sk_buff *skb)
869 {
870 	struct rtable *rt = skb_rtable(skb);
871 	struct in_device *in_dev;
872 	struct inet_peer *peer;
873 	struct net *net;
874 	int log_martians;
875 	int vif;
876 
877 	rcu_read_lock();
878 	in_dev = __in_dev_get_rcu(rt->dst.dev);
879 	if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
880 		rcu_read_unlock();
881 		return;
882 	}
883 	log_martians = IN_DEV_LOG_MARTIANS(in_dev);
884 	vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
885 	rcu_read_unlock();
886 
887 	net = dev_net(rt->dst.dev);
888 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
889 	if (!peer) {
890 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
891 			  rt_nexthop(rt, ip_hdr(skb)->daddr));
892 		return;
893 	}
894 
895 	/* No redirected packets during ip_rt_redirect_silence;
896 	 * reset the algorithm.
897 	 */
898 	if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
899 		peer->rate_tokens = 0;
900 		peer->n_redirects = 0;
901 	}
902 
903 	/* Too many ignored redirects; do not send anything
904 	 * set dst.rate_last to the last seen redirected packet.
905 	 */
906 	if (peer->n_redirects >= ip_rt_redirect_number) {
907 		peer->rate_last = jiffies;
908 		goto out_put_peer;
909 	}
910 
911 	/* Check for load limit; set rate_last to the latest sent
912 	 * redirect.
913 	 */
914 	if (peer->n_redirects == 0 ||
915 	    time_after(jiffies,
916 		       (peer->rate_last +
917 			(ip_rt_redirect_load << peer->n_redirects)))) {
918 		__be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
919 
920 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
921 		peer->rate_last = jiffies;
922 		++peer->n_redirects;
923 		if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians &&
924 		    peer->n_redirects == ip_rt_redirect_number)
925 			net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
926 					     &ip_hdr(skb)->saddr, inet_iif(skb),
927 					     &ip_hdr(skb)->daddr, &gw);
928 	}
929 out_put_peer:
930 	inet_putpeer(peer);
931 }
932 
933 static int ip_error(struct sk_buff *skb)
934 {
935 	struct rtable *rt = skb_rtable(skb);
936 	struct net_device *dev = skb->dev;
937 	struct in_device *in_dev;
938 	struct inet_peer *peer;
939 	unsigned long now;
940 	struct net *net;
941 	SKB_DR(reason);
942 	bool send;
943 	int code;
944 
945 	if (netif_is_l3_master(skb->dev)) {
946 		dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
947 		if (!dev)
948 			goto out;
949 	}
950 
951 	in_dev = __in_dev_get_rcu(dev);
952 
953 	/* IP on this device is disabled. */
954 	if (!in_dev)
955 		goto out;
956 
957 	net = dev_net(rt->dst.dev);
958 	if (!IN_DEV_FORWARD(in_dev)) {
959 		switch (rt->dst.error) {
960 		case EHOSTUNREACH:
961 			SKB_DR_SET(reason, IP_INADDRERRORS);
962 			__IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
963 			break;
964 
965 		case ENETUNREACH:
966 			SKB_DR_SET(reason, IP_INNOROUTES);
967 			__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
968 			break;
969 		}
970 		goto out;
971 	}
972 
973 	switch (rt->dst.error) {
974 	case EINVAL:
975 	default:
976 		goto out;
977 	case EHOSTUNREACH:
978 		code = ICMP_HOST_UNREACH;
979 		break;
980 	case ENETUNREACH:
981 		code = ICMP_NET_UNREACH;
982 		SKB_DR_SET(reason, IP_INNOROUTES);
983 		__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
984 		break;
985 	case EACCES:
986 		code = ICMP_PKT_FILTERED;
987 		break;
988 	}
989 
990 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
991 			       l3mdev_master_ifindex(skb->dev), 1);
992 
993 	send = true;
994 	if (peer) {
995 		now = jiffies;
996 		peer->rate_tokens += now - peer->rate_last;
997 		if (peer->rate_tokens > ip_rt_error_burst)
998 			peer->rate_tokens = ip_rt_error_burst;
999 		peer->rate_last = now;
1000 		if (peer->rate_tokens >= ip_rt_error_cost)
1001 			peer->rate_tokens -= ip_rt_error_cost;
1002 		else
1003 			send = false;
1004 		inet_putpeer(peer);
1005 	}
1006 	if (send)
1007 		icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1008 
1009 out:	kfree_skb_reason(skb, reason);
1010 	return 0;
1011 }
1012 
1013 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1014 {
1015 	struct dst_entry *dst = &rt->dst;
1016 	struct net *net = dev_net(dst->dev);
1017 	struct fib_result res;
1018 	bool lock = false;
1019 	u32 old_mtu;
1020 
1021 	if (ip_mtu_locked(dst))
1022 		return;
1023 
1024 	old_mtu = ipv4_mtu(dst);
1025 	if (old_mtu < mtu)
1026 		return;
1027 
1028 	if (mtu < net->ipv4.ip_rt_min_pmtu) {
1029 		lock = true;
1030 		mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1031 	}
1032 
1033 	if (rt->rt_pmtu == mtu && !lock &&
1034 	    time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2))
1035 		return;
1036 
1037 	rcu_read_lock();
1038 	if (fib_lookup(net, fl4, &res, 0) == 0) {
1039 		struct fib_nh_common *nhc;
1040 
1041 		fib_select_path(net, &res, fl4, NULL);
1042 		nhc = FIB_RES_NHC(res);
1043 		update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1044 				      jiffies + net->ipv4.ip_rt_mtu_expires);
1045 	}
1046 	rcu_read_unlock();
1047 }
1048 
1049 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1050 			      struct sk_buff *skb, u32 mtu,
1051 			      bool confirm_neigh)
1052 {
1053 	struct rtable *rt = (struct rtable *) dst;
1054 	struct flowi4 fl4;
1055 
1056 	ip_rt_build_flow_key(&fl4, sk, skb);
1057 
1058 	/* Don't make lookup fail for bridged encapsulations */
1059 	if (skb && netif_is_any_bridge_port(skb->dev))
1060 		fl4.flowi4_oif = 0;
1061 
1062 	__ip_rt_update_pmtu(rt, &fl4, mtu);
1063 }
1064 
1065 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1066 		      int oif, u8 protocol)
1067 {
1068 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1069 	struct flowi4 fl4;
1070 	struct rtable *rt;
1071 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
1072 
1073 	__build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1074 			 0);
1075 	rt = __ip_route_output_key(net, &fl4);
1076 	if (!IS_ERR(rt)) {
1077 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1078 		ip_rt_put(rt);
1079 	}
1080 }
1081 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1082 
1083 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1084 {
1085 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1086 	struct flowi4 fl4;
1087 	struct rtable *rt;
1088 
1089 	__build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1090 
1091 	if (!fl4.flowi4_mark)
1092 		fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1093 
1094 	rt = __ip_route_output_key(sock_net(sk), &fl4);
1095 	if (!IS_ERR(rt)) {
1096 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1097 		ip_rt_put(rt);
1098 	}
1099 }
1100 
1101 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1102 {
1103 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1104 	struct flowi4 fl4;
1105 	struct rtable *rt;
1106 	struct dst_entry *odst = NULL;
1107 	bool new = false;
1108 	struct net *net = sock_net(sk);
1109 
1110 	bh_lock_sock(sk);
1111 
1112 	if (!ip_sk_accept_pmtu(sk))
1113 		goto out;
1114 
1115 	odst = sk_dst_get(sk);
1116 
1117 	if (sock_owned_by_user(sk) || !odst) {
1118 		__ipv4_sk_update_pmtu(skb, sk, mtu);
1119 		goto out;
1120 	}
1121 
1122 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1123 
1124 	rt = (struct rtable *)odst;
1125 	if (odst->obsolete && !odst->ops->check(odst, 0)) {
1126 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1127 		if (IS_ERR(rt))
1128 			goto out;
1129 
1130 		new = true;
1131 	}
1132 
1133 	__ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu);
1134 
1135 	if (!dst_check(&rt->dst, 0)) {
1136 		if (new)
1137 			dst_release(&rt->dst);
1138 
1139 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1140 		if (IS_ERR(rt))
1141 			goto out;
1142 
1143 		new = true;
1144 	}
1145 
1146 	if (new)
1147 		sk_dst_set(sk, &rt->dst);
1148 
1149 out:
1150 	bh_unlock_sock(sk);
1151 	dst_release(odst);
1152 }
1153 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1154 
1155 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1156 		   int oif, u8 protocol)
1157 {
1158 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1159 	struct flowi4 fl4;
1160 	struct rtable *rt;
1161 
1162 	__build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1163 	rt = __ip_route_output_key(net, &fl4);
1164 	if (!IS_ERR(rt)) {
1165 		__ip_do_redirect(rt, skb, &fl4, false);
1166 		ip_rt_put(rt);
1167 	}
1168 }
1169 EXPORT_SYMBOL_GPL(ipv4_redirect);
1170 
1171 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1172 {
1173 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1174 	struct flowi4 fl4;
1175 	struct rtable *rt;
1176 	struct net *net = sock_net(sk);
1177 
1178 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1179 	rt = __ip_route_output_key(net, &fl4);
1180 	if (!IS_ERR(rt)) {
1181 		__ip_do_redirect(rt, skb, &fl4, false);
1182 		ip_rt_put(rt);
1183 	}
1184 }
1185 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1186 
1187 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1188 							 u32 cookie)
1189 {
1190 	struct rtable *rt = (struct rtable *) dst;
1191 
1192 	/* All IPV4 dsts are created with ->obsolete set to the value
1193 	 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1194 	 * into this function always.
1195 	 *
1196 	 * When a PMTU/redirect information update invalidates a route,
1197 	 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1198 	 * DST_OBSOLETE_DEAD.
1199 	 */
1200 	if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1201 		return NULL;
1202 	return dst;
1203 }
1204 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1205 
1206 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1207 {
1208 	struct net_device *dev;
1209 	struct ip_options opt;
1210 	int res;
1211 
1212 	/* Recompile ip options since IPCB may not be valid anymore.
1213 	 * Also check we have a reasonable ipv4 header.
1214 	 */
1215 	if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1216 	    ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1217 		return;
1218 
1219 	memset(&opt, 0, sizeof(opt));
1220 	if (ip_hdr(skb)->ihl > 5) {
1221 		if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1222 			return;
1223 		opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1224 
1225 		rcu_read_lock();
1226 		dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1227 		res = __ip_options_compile(dev_net(dev), &opt, skb, NULL);
1228 		rcu_read_unlock();
1229 
1230 		if (res)
1231 			return;
1232 	}
1233 	__icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1234 }
1235 
1236 static void ipv4_link_failure(struct sk_buff *skb)
1237 {
1238 	struct rtable *rt;
1239 
1240 	ipv4_send_dest_unreach(skb);
1241 
1242 	rt = skb_rtable(skb);
1243 	if (rt)
1244 		dst_set_expires(&rt->dst, 0);
1245 }
1246 
1247 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1248 {
1249 	pr_debug("%s: %pI4 -> %pI4, %s\n",
1250 		 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1251 		 skb->dev ? skb->dev->name : "?");
1252 	kfree_skb(skb);
1253 	WARN_ON(1);
1254 	return 0;
1255 }
1256 
1257 /*
1258  * We do not cache source address of outgoing interface,
1259  * because it is used only by IP RR, TS and SRR options,
1260  * so that it out of fast path.
1261  *
1262  * BTW remember: "addr" is allowed to be not aligned
1263  * in IP options!
1264  */
1265 
1266 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1267 {
1268 	__be32 src;
1269 
1270 	if (rt_is_output_route(rt))
1271 		src = ip_hdr(skb)->saddr;
1272 	else {
1273 		struct fib_result res;
1274 		struct iphdr *iph = ip_hdr(skb);
1275 		struct flowi4 fl4 = {
1276 			.daddr = iph->daddr,
1277 			.saddr = iph->saddr,
1278 			.flowi4_tos = iph->tos & IPTOS_RT_MASK,
1279 			.flowi4_oif = rt->dst.dev->ifindex,
1280 			.flowi4_iif = skb->dev->ifindex,
1281 			.flowi4_mark = skb->mark,
1282 		};
1283 
1284 		rcu_read_lock();
1285 		if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1286 			src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1287 		else
1288 			src = inet_select_addr(rt->dst.dev,
1289 					       rt_nexthop(rt, iph->daddr),
1290 					       RT_SCOPE_UNIVERSE);
1291 		rcu_read_unlock();
1292 	}
1293 	memcpy(addr, &src, 4);
1294 }
1295 
1296 #ifdef CONFIG_IP_ROUTE_CLASSID
1297 static void set_class_tag(struct rtable *rt, u32 tag)
1298 {
1299 	if (!(rt->dst.tclassid & 0xFFFF))
1300 		rt->dst.tclassid |= tag & 0xFFFF;
1301 	if (!(rt->dst.tclassid & 0xFFFF0000))
1302 		rt->dst.tclassid |= tag & 0xFFFF0000;
1303 }
1304 #endif
1305 
1306 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1307 {
1308 	struct net *net = dev_net(dst->dev);
1309 	unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1310 	unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1311 				    net->ipv4.ip_rt_min_advmss);
1312 
1313 	return min(advmss, IPV4_MAX_PMTU - header_size);
1314 }
1315 
1316 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1317 {
1318 	return ip_dst_mtu_maybe_forward(dst, false);
1319 }
1320 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1321 
1322 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1323 {
1324 	struct fnhe_hash_bucket *hash;
1325 	struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1326 	u32 hval = fnhe_hashfun(daddr);
1327 
1328 	spin_lock_bh(&fnhe_lock);
1329 
1330 	hash = rcu_dereference_protected(nhc->nhc_exceptions,
1331 					 lockdep_is_held(&fnhe_lock));
1332 	hash += hval;
1333 
1334 	fnhe_p = &hash->chain;
1335 	fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1336 	while (fnhe) {
1337 		if (fnhe->fnhe_daddr == daddr) {
1338 			rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1339 				fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1340 			/* set fnhe_daddr to 0 to ensure it won't bind with
1341 			 * new dsts in rt_bind_exception().
1342 			 */
1343 			fnhe->fnhe_daddr = 0;
1344 			fnhe_flush_routes(fnhe);
1345 			kfree_rcu(fnhe, rcu);
1346 			break;
1347 		}
1348 		fnhe_p = &fnhe->fnhe_next;
1349 		fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1350 						 lockdep_is_held(&fnhe_lock));
1351 	}
1352 
1353 	spin_unlock_bh(&fnhe_lock);
1354 }
1355 
1356 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1357 					       __be32 daddr)
1358 {
1359 	struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1360 	struct fib_nh_exception *fnhe;
1361 	u32 hval;
1362 
1363 	if (!hash)
1364 		return NULL;
1365 
1366 	hval = fnhe_hashfun(daddr);
1367 
1368 	for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1369 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
1370 		if (fnhe->fnhe_daddr == daddr) {
1371 			if (fnhe->fnhe_expires &&
1372 			    time_after(jiffies, fnhe->fnhe_expires)) {
1373 				ip_del_fnhe(nhc, daddr);
1374 				break;
1375 			}
1376 			return fnhe;
1377 		}
1378 	}
1379 	return NULL;
1380 }
1381 
1382 /* MTU selection:
1383  * 1. mtu on route is locked - use it
1384  * 2. mtu from nexthop exception
1385  * 3. mtu from egress device
1386  */
1387 
1388 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1389 {
1390 	struct fib_nh_common *nhc = res->nhc;
1391 	struct net_device *dev = nhc->nhc_dev;
1392 	struct fib_info *fi = res->fi;
1393 	u32 mtu = 0;
1394 
1395 	if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1396 	    fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1397 		mtu = fi->fib_mtu;
1398 
1399 	if (likely(!mtu)) {
1400 		struct fib_nh_exception *fnhe;
1401 
1402 		fnhe = find_exception(nhc, daddr);
1403 		if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1404 			mtu = fnhe->fnhe_pmtu;
1405 	}
1406 
1407 	if (likely(!mtu))
1408 		mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1409 
1410 	return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1411 }
1412 
1413 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1414 			      __be32 daddr, const bool do_cache)
1415 {
1416 	bool ret = false;
1417 
1418 	spin_lock_bh(&fnhe_lock);
1419 
1420 	if (daddr == fnhe->fnhe_daddr) {
1421 		struct rtable __rcu **porig;
1422 		struct rtable *orig;
1423 		int genid = fnhe_genid(dev_net(rt->dst.dev));
1424 
1425 		if (rt_is_input_route(rt))
1426 			porig = &fnhe->fnhe_rth_input;
1427 		else
1428 			porig = &fnhe->fnhe_rth_output;
1429 		orig = rcu_dereference(*porig);
1430 
1431 		if (fnhe->fnhe_genid != genid) {
1432 			fnhe->fnhe_genid = genid;
1433 			fnhe->fnhe_gw = 0;
1434 			fnhe->fnhe_pmtu = 0;
1435 			fnhe->fnhe_expires = 0;
1436 			fnhe->fnhe_mtu_locked = false;
1437 			fnhe_flush_routes(fnhe);
1438 			orig = NULL;
1439 		}
1440 		fill_route_from_fnhe(rt, fnhe);
1441 		if (!rt->rt_gw4) {
1442 			rt->rt_gw4 = daddr;
1443 			rt->rt_gw_family = AF_INET;
1444 		}
1445 
1446 		if (do_cache) {
1447 			dst_hold(&rt->dst);
1448 			rcu_assign_pointer(*porig, rt);
1449 			if (orig) {
1450 				dst_dev_put(&orig->dst);
1451 				dst_release(&orig->dst);
1452 			}
1453 			ret = true;
1454 		}
1455 
1456 		fnhe->fnhe_stamp = jiffies;
1457 	}
1458 	spin_unlock_bh(&fnhe_lock);
1459 
1460 	return ret;
1461 }
1462 
1463 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1464 {
1465 	struct rtable *orig, *prev, **p;
1466 	bool ret = true;
1467 
1468 	if (rt_is_input_route(rt)) {
1469 		p = (struct rtable **)&nhc->nhc_rth_input;
1470 	} else {
1471 		p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1472 	}
1473 	orig = *p;
1474 
1475 	/* hold dst before doing cmpxchg() to avoid race condition
1476 	 * on this dst
1477 	 */
1478 	dst_hold(&rt->dst);
1479 	prev = cmpxchg(p, orig, rt);
1480 	if (prev == orig) {
1481 		if (orig) {
1482 			rt_add_uncached_list(orig);
1483 			dst_release(&orig->dst);
1484 		}
1485 	} else {
1486 		dst_release(&rt->dst);
1487 		ret = false;
1488 	}
1489 
1490 	return ret;
1491 }
1492 
1493 struct uncached_list {
1494 	spinlock_t		lock;
1495 	struct list_head	head;
1496 	struct list_head	quarantine;
1497 };
1498 
1499 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1500 
1501 void rt_add_uncached_list(struct rtable *rt)
1502 {
1503 	struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1504 
1505 	rt->dst.rt_uncached_list = ul;
1506 
1507 	spin_lock_bh(&ul->lock);
1508 	list_add_tail(&rt->dst.rt_uncached, &ul->head);
1509 	spin_unlock_bh(&ul->lock);
1510 }
1511 
1512 void rt_del_uncached_list(struct rtable *rt)
1513 {
1514 	if (!list_empty(&rt->dst.rt_uncached)) {
1515 		struct uncached_list *ul = rt->dst.rt_uncached_list;
1516 
1517 		spin_lock_bh(&ul->lock);
1518 		list_del_init(&rt->dst.rt_uncached);
1519 		spin_unlock_bh(&ul->lock);
1520 	}
1521 }
1522 
1523 static void ipv4_dst_destroy(struct dst_entry *dst)
1524 {
1525 	struct rtable *rt = (struct rtable *)dst;
1526 
1527 	ip_dst_metrics_put(dst);
1528 	rt_del_uncached_list(rt);
1529 }
1530 
1531 void rt_flush_dev(struct net_device *dev)
1532 {
1533 	struct rtable *rt, *safe;
1534 	int cpu;
1535 
1536 	for_each_possible_cpu(cpu) {
1537 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1538 
1539 		if (list_empty(&ul->head))
1540 			continue;
1541 
1542 		spin_lock_bh(&ul->lock);
1543 		list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1544 			if (rt->dst.dev != dev)
1545 				continue;
1546 			rt->dst.dev = blackhole_netdev;
1547 			netdev_ref_replace(dev, blackhole_netdev,
1548 					   &rt->dst.dev_tracker, GFP_ATOMIC);
1549 			list_move(&rt->dst.rt_uncached, &ul->quarantine);
1550 		}
1551 		spin_unlock_bh(&ul->lock);
1552 	}
1553 }
1554 
1555 static bool rt_cache_valid(const struct rtable *rt)
1556 {
1557 	return	rt &&
1558 		rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1559 		!rt_is_expired(rt);
1560 }
1561 
1562 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1563 			   const struct fib_result *res,
1564 			   struct fib_nh_exception *fnhe,
1565 			   struct fib_info *fi, u16 type, u32 itag,
1566 			   const bool do_cache)
1567 {
1568 	bool cached = false;
1569 
1570 	if (fi) {
1571 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1572 
1573 		if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1574 			rt->rt_uses_gateway = 1;
1575 			rt->rt_gw_family = nhc->nhc_gw_family;
1576 			/* only INET and INET6 are supported */
1577 			if (likely(nhc->nhc_gw_family == AF_INET))
1578 				rt->rt_gw4 = nhc->nhc_gw.ipv4;
1579 			else
1580 				rt->rt_gw6 = nhc->nhc_gw.ipv6;
1581 		}
1582 
1583 		ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1584 
1585 #ifdef CONFIG_IP_ROUTE_CLASSID
1586 		if (nhc->nhc_family == AF_INET) {
1587 			struct fib_nh *nh;
1588 
1589 			nh = container_of(nhc, struct fib_nh, nh_common);
1590 			rt->dst.tclassid = nh->nh_tclassid;
1591 		}
1592 #endif
1593 		rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1594 		if (unlikely(fnhe))
1595 			cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1596 		else if (do_cache)
1597 			cached = rt_cache_route(nhc, rt);
1598 		if (unlikely(!cached)) {
1599 			/* Routes we intend to cache in nexthop exception or
1600 			 * FIB nexthop have the DST_NOCACHE bit clear.
1601 			 * However, if we are unsuccessful at storing this
1602 			 * route into the cache we really need to set it.
1603 			 */
1604 			if (!rt->rt_gw4) {
1605 				rt->rt_gw_family = AF_INET;
1606 				rt->rt_gw4 = daddr;
1607 			}
1608 			rt_add_uncached_list(rt);
1609 		}
1610 	} else
1611 		rt_add_uncached_list(rt);
1612 
1613 #ifdef CONFIG_IP_ROUTE_CLASSID
1614 #ifdef CONFIG_IP_MULTIPLE_TABLES
1615 	set_class_tag(rt, res->tclassid);
1616 #endif
1617 	set_class_tag(rt, itag);
1618 #endif
1619 }
1620 
1621 struct rtable *rt_dst_alloc(struct net_device *dev,
1622 			    unsigned int flags, u16 type,
1623 			    bool noxfrm)
1624 {
1625 	struct rtable *rt;
1626 
1627 	rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1628 		       (noxfrm ? DST_NOXFRM : 0));
1629 
1630 	if (rt) {
1631 		rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1632 		rt->rt_flags = flags;
1633 		rt->rt_type = type;
1634 		rt->rt_is_input = 0;
1635 		rt->rt_iif = 0;
1636 		rt->rt_pmtu = 0;
1637 		rt->rt_mtu_locked = 0;
1638 		rt->rt_uses_gateway = 0;
1639 		rt->rt_gw_family = 0;
1640 		rt->rt_gw4 = 0;
1641 
1642 		rt->dst.output = ip_output;
1643 		if (flags & RTCF_LOCAL)
1644 			rt->dst.input = ip_local_deliver;
1645 	}
1646 
1647 	return rt;
1648 }
1649 EXPORT_SYMBOL(rt_dst_alloc);
1650 
1651 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1652 {
1653 	struct rtable *new_rt;
1654 
1655 	new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1656 			   rt->dst.flags);
1657 
1658 	if (new_rt) {
1659 		new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1660 		new_rt->rt_flags = rt->rt_flags;
1661 		new_rt->rt_type = rt->rt_type;
1662 		new_rt->rt_is_input = rt->rt_is_input;
1663 		new_rt->rt_iif = rt->rt_iif;
1664 		new_rt->rt_pmtu = rt->rt_pmtu;
1665 		new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1666 		new_rt->rt_gw_family = rt->rt_gw_family;
1667 		if (rt->rt_gw_family == AF_INET)
1668 			new_rt->rt_gw4 = rt->rt_gw4;
1669 		else if (rt->rt_gw_family == AF_INET6)
1670 			new_rt->rt_gw6 = rt->rt_gw6;
1671 
1672 		new_rt->dst.input = rt->dst.input;
1673 		new_rt->dst.output = rt->dst.output;
1674 		new_rt->dst.error = rt->dst.error;
1675 		new_rt->dst.lastuse = jiffies;
1676 		new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1677 	}
1678 	return new_rt;
1679 }
1680 EXPORT_SYMBOL(rt_dst_clone);
1681 
1682 /* called in rcu_read_lock() section */
1683 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1684 			  u8 tos, struct net_device *dev,
1685 			  struct in_device *in_dev, u32 *itag)
1686 {
1687 	int err;
1688 
1689 	/* Primary sanity checks. */
1690 	if (!in_dev)
1691 		return -EINVAL;
1692 
1693 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1694 	    skb->protocol != htons(ETH_P_IP))
1695 		return -EINVAL;
1696 
1697 	if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1698 		return -EINVAL;
1699 
1700 	if (ipv4_is_zeronet(saddr)) {
1701 		if (!ipv4_is_local_multicast(daddr) &&
1702 		    ip_hdr(skb)->protocol != IPPROTO_IGMP)
1703 			return -EINVAL;
1704 	} else {
1705 		err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1706 					  in_dev, itag);
1707 		if (err < 0)
1708 			return err;
1709 	}
1710 	return 0;
1711 }
1712 
1713 /* called in rcu_read_lock() section */
1714 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1715 			     u8 tos, struct net_device *dev, int our)
1716 {
1717 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1718 	unsigned int flags = RTCF_MULTICAST;
1719 	struct rtable *rth;
1720 	u32 itag = 0;
1721 	int err;
1722 
1723 	err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1724 	if (err)
1725 		return err;
1726 
1727 	if (our)
1728 		flags |= RTCF_LOCAL;
1729 
1730 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1731 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
1732 
1733 	rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1734 			   false);
1735 	if (!rth)
1736 		return -ENOBUFS;
1737 
1738 #ifdef CONFIG_IP_ROUTE_CLASSID
1739 	rth->dst.tclassid = itag;
1740 #endif
1741 	rth->dst.output = ip_rt_bug;
1742 	rth->rt_is_input= 1;
1743 
1744 #ifdef CONFIG_IP_MROUTE
1745 	if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1746 		rth->dst.input = ip_mr_input;
1747 #endif
1748 	RT_CACHE_STAT_INC(in_slow_mc);
1749 
1750 	skb_dst_drop(skb);
1751 	skb_dst_set(skb, &rth->dst);
1752 	return 0;
1753 }
1754 
1755 
1756 static void ip_handle_martian_source(struct net_device *dev,
1757 				     struct in_device *in_dev,
1758 				     struct sk_buff *skb,
1759 				     __be32 daddr,
1760 				     __be32 saddr)
1761 {
1762 	RT_CACHE_STAT_INC(in_martian_src);
1763 #ifdef CONFIG_IP_ROUTE_VERBOSE
1764 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1765 		/*
1766 		 *	RFC1812 recommendation, if source is martian,
1767 		 *	the only hint is MAC header.
1768 		 */
1769 		pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1770 			&daddr, &saddr, dev->name);
1771 		if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1772 			print_hex_dump(KERN_WARNING, "ll header: ",
1773 				       DUMP_PREFIX_OFFSET, 16, 1,
1774 				       skb_mac_header(skb),
1775 				       dev->hard_header_len, false);
1776 		}
1777 	}
1778 #endif
1779 }
1780 
1781 /* called in rcu_read_lock() section */
1782 static int __mkroute_input(struct sk_buff *skb,
1783 			   const struct fib_result *res,
1784 			   struct in_device *in_dev,
1785 			   __be32 daddr, __be32 saddr, u32 tos)
1786 {
1787 	struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1788 	struct net_device *dev = nhc->nhc_dev;
1789 	struct fib_nh_exception *fnhe;
1790 	struct rtable *rth;
1791 	int err;
1792 	struct in_device *out_dev;
1793 	bool do_cache;
1794 	u32 itag = 0;
1795 
1796 	/* get a working reference to the output device */
1797 	out_dev = __in_dev_get_rcu(dev);
1798 	if (!out_dev) {
1799 		net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1800 		return -EINVAL;
1801 	}
1802 
1803 	err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1804 				  in_dev->dev, in_dev, &itag);
1805 	if (err < 0) {
1806 		ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1807 					 saddr);
1808 
1809 		goto cleanup;
1810 	}
1811 
1812 	do_cache = res->fi && !itag;
1813 	if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1814 	    skb->protocol == htons(ETH_P_IP)) {
1815 		__be32 gw;
1816 
1817 		gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1818 		if (IN_DEV_SHARED_MEDIA(out_dev) ||
1819 		    inet_addr_onlink(out_dev, saddr, gw))
1820 			IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1821 	}
1822 
1823 	if (skb->protocol != htons(ETH_P_IP)) {
1824 		/* Not IP (i.e. ARP). Do not create route, if it is
1825 		 * invalid for proxy arp. DNAT routes are always valid.
1826 		 *
1827 		 * Proxy arp feature have been extended to allow, ARP
1828 		 * replies back to the same interface, to support
1829 		 * Private VLAN switch technologies. See arp.c.
1830 		 */
1831 		if (out_dev == in_dev &&
1832 		    IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1833 			err = -EINVAL;
1834 			goto cleanup;
1835 		}
1836 	}
1837 
1838 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1839 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
1840 
1841 	fnhe = find_exception(nhc, daddr);
1842 	if (do_cache) {
1843 		if (fnhe)
1844 			rth = rcu_dereference(fnhe->fnhe_rth_input);
1845 		else
1846 			rth = rcu_dereference(nhc->nhc_rth_input);
1847 		if (rt_cache_valid(rth)) {
1848 			skb_dst_set_noref(skb, &rth->dst);
1849 			goto out;
1850 		}
1851 	}
1852 
1853 	rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1854 			   IN_DEV_ORCONF(out_dev, NOXFRM));
1855 	if (!rth) {
1856 		err = -ENOBUFS;
1857 		goto cleanup;
1858 	}
1859 
1860 	rth->rt_is_input = 1;
1861 	RT_CACHE_STAT_INC(in_slow_tot);
1862 
1863 	rth->dst.input = ip_forward;
1864 
1865 	rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1866 		       do_cache);
1867 	lwtunnel_set_redirect(&rth->dst);
1868 	skb_dst_set(skb, &rth->dst);
1869 out:
1870 	err = 0;
1871  cleanup:
1872 	return err;
1873 }
1874 
1875 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1876 /* To make ICMP packets follow the right flow, the multipath hash is
1877  * calculated from the inner IP addresses.
1878  */
1879 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1880 				 struct flow_keys *hash_keys)
1881 {
1882 	const struct iphdr *outer_iph = ip_hdr(skb);
1883 	const struct iphdr *key_iph = outer_iph;
1884 	const struct iphdr *inner_iph;
1885 	const struct icmphdr *icmph;
1886 	struct iphdr _inner_iph;
1887 	struct icmphdr _icmph;
1888 
1889 	if (likely(outer_iph->protocol != IPPROTO_ICMP))
1890 		goto out;
1891 
1892 	if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1893 		goto out;
1894 
1895 	icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1896 				   &_icmph);
1897 	if (!icmph)
1898 		goto out;
1899 
1900 	if (!icmp_is_err(icmph->type))
1901 		goto out;
1902 
1903 	inner_iph = skb_header_pointer(skb,
1904 				       outer_iph->ihl * 4 + sizeof(_icmph),
1905 				       sizeof(_inner_iph), &_inner_iph);
1906 	if (!inner_iph)
1907 		goto out;
1908 
1909 	key_iph = inner_iph;
1910 out:
1911 	hash_keys->addrs.v4addrs.src = key_iph->saddr;
1912 	hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1913 }
1914 
1915 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1916 					   const struct sk_buff *skb,
1917 					   bool *p_has_inner)
1918 {
1919 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1920 	struct flow_keys keys, hash_keys;
1921 
1922 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1923 		return 0;
1924 
1925 	memset(&hash_keys, 0, sizeof(hash_keys));
1926 	skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1927 
1928 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1929 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1930 		hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1931 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1932 		hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1933 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1934 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
1935 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1936 		hash_keys.ports.src = keys.ports.src;
1937 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1938 		hash_keys.ports.dst = keys.ports.dst;
1939 
1940 	*p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1941 	return flow_hash_from_keys(&hash_keys);
1942 }
1943 
1944 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1945 					   const struct sk_buff *skb,
1946 					   bool has_inner)
1947 {
1948 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1949 	struct flow_keys keys, hash_keys;
1950 
1951 	/* We assume the packet carries an encapsulation, but if none was
1952 	 * encountered during dissection of the outer flow, then there is no
1953 	 * point in calling the flow dissector again.
1954 	 */
1955 	if (!has_inner)
1956 		return 0;
1957 
1958 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1959 		return 0;
1960 
1961 	memset(&hash_keys, 0, sizeof(hash_keys));
1962 	skb_flow_dissect_flow_keys(skb, &keys, 0);
1963 
1964 	if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1965 		return 0;
1966 
1967 	if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1968 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1969 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1970 			hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1971 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1972 			hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1973 	} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1974 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1975 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1976 			hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1977 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1978 			hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1979 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1980 			hash_keys.tags.flow_label = keys.tags.flow_label;
1981 	}
1982 
1983 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1984 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
1985 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1986 		hash_keys.ports.src = keys.ports.src;
1987 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
1988 		hash_keys.ports.dst = keys.ports.dst;
1989 
1990 	return flow_hash_from_keys(&hash_keys);
1991 }
1992 
1993 static u32 fib_multipath_custom_hash_skb(const struct net *net,
1994 					 const struct sk_buff *skb)
1995 {
1996 	u32 mhash, mhash_inner;
1997 	bool has_inner = true;
1998 
1999 	mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2000 	mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2001 
2002 	return jhash_2words(mhash, mhash_inner, 0);
2003 }
2004 
2005 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2006 					 const struct flowi4 *fl4)
2007 {
2008 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2009 	struct flow_keys hash_keys;
2010 
2011 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2012 		return 0;
2013 
2014 	memset(&hash_keys, 0, sizeof(hash_keys));
2015 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2016 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2017 		hash_keys.addrs.v4addrs.src = fl4->saddr;
2018 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2019 		hash_keys.addrs.v4addrs.dst = fl4->daddr;
2020 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2021 		hash_keys.basic.ip_proto = fl4->flowi4_proto;
2022 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2023 		hash_keys.ports.src = fl4->fl4_sport;
2024 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2025 		hash_keys.ports.dst = fl4->fl4_dport;
2026 
2027 	return flow_hash_from_keys(&hash_keys);
2028 }
2029 
2030 /* if skb is set it will be used and fl4 can be NULL */
2031 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2032 		       const struct sk_buff *skb, struct flow_keys *flkeys)
2033 {
2034 	u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2035 	struct flow_keys hash_keys;
2036 	u32 mhash = 0;
2037 
2038 	switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2039 	case 0:
2040 		memset(&hash_keys, 0, sizeof(hash_keys));
2041 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2042 		if (skb) {
2043 			ip_multipath_l3_keys(skb, &hash_keys);
2044 		} else {
2045 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2046 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2047 		}
2048 		mhash = flow_hash_from_keys(&hash_keys);
2049 		break;
2050 	case 1:
2051 		/* skb is currently provided only when forwarding */
2052 		if (skb) {
2053 			unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2054 			struct flow_keys keys;
2055 
2056 			/* short-circuit if we already have L4 hash present */
2057 			if (skb->l4_hash)
2058 				return skb_get_hash_raw(skb) >> 1;
2059 
2060 			memset(&hash_keys, 0, sizeof(hash_keys));
2061 
2062 			if (!flkeys) {
2063 				skb_flow_dissect_flow_keys(skb, &keys, flag);
2064 				flkeys = &keys;
2065 			}
2066 
2067 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2068 			hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2069 			hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2070 			hash_keys.ports.src = flkeys->ports.src;
2071 			hash_keys.ports.dst = flkeys->ports.dst;
2072 			hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2073 		} else {
2074 			memset(&hash_keys, 0, sizeof(hash_keys));
2075 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2076 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2077 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2078 			hash_keys.ports.src = fl4->fl4_sport;
2079 			hash_keys.ports.dst = fl4->fl4_dport;
2080 			hash_keys.basic.ip_proto = fl4->flowi4_proto;
2081 		}
2082 		mhash = flow_hash_from_keys(&hash_keys);
2083 		break;
2084 	case 2:
2085 		memset(&hash_keys, 0, sizeof(hash_keys));
2086 		/* skb is currently provided only when forwarding */
2087 		if (skb) {
2088 			struct flow_keys keys;
2089 
2090 			skb_flow_dissect_flow_keys(skb, &keys, 0);
2091 			/* Inner can be v4 or v6 */
2092 			if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2093 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2094 				hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2095 				hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2096 			} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2097 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2098 				hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2099 				hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2100 				hash_keys.tags.flow_label = keys.tags.flow_label;
2101 				hash_keys.basic.ip_proto = keys.basic.ip_proto;
2102 			} else {
2103 				/* Same as case 0 */
2104 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2105 				ip_multipath_l3_keys(skb, &hash_keys);
2106 			}
2107 		} else {
2108 			/* Same as case 0 */
2109 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2110 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2111 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2112 		}
2113 		mhash = flow_hash_from_keys(&hash_keys);
2114 		break;
2115 	case 3:
2116 		if (skb)
2117 			mhash = fib_multipath_custom_hash_skb(net, skb);
2118 		else
2119 			mhash = fib_multipath_custom_hash_fl4(net, fl4);
2120 		break;
2121 	}
2122 
2123 	if (multipath_hash)
2124 		mhash = jhash_2words(mhash, multipath_hash, 0);
2125 
2126 	return mhash >> 1;
2127 }
2128 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2129 
2130 static int ip_mkroute_input(struct sk_buff *skb,
2131 			    struct fib_result *res,
2132 			    struct in_device *in_dev,
2133 			    __be32 daddr, __be32 saddr, u32 tos,
2134 			    struct flow_keys *hkeys)
2135 {
2136 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2137 	if (res->fi && fib_info_num_path(res->fi) > 1) {
2138 		int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2139 
2140 		fib_select_multipath(res, h);
2141 		IPCB(skb)->flags |= IPSKB_MULTIPATH;
2142 	}
2143 #endif
2144 
2145 	/* create a routing cache entry */
2146 	return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2147 }
2148 
2149 /* Implements all the saddr-related checks as ip_route_input_slow(),
2150  * assuming daddr is valid and the destination is not a local broadcast one.
2151  * Uses the provided hint instead of performing a route lookup.
2152  */
2153 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2154 		      u8 tos, struct net_device *dev,
2155 		      const struct sk_buff *hint)
2156 {
2157 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2158 	struct rtable *rt = skb_rtable(hint);
2159 	struct net *net = dev_net(dev);
2160 	int err = -EINVAL;
2161 	u32 tag = 0;
2162 
2163 	if (!in_dev)
2164 		return -EINVAL;
2165 
2166 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2167 		goto martian_source;
2168 
2169 	if (ipv4_is_zeronet(saddr))
2170 		goto martian_source;
2171 
2172 	if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2173 		goto martian_source;
2174 
2175 	if (rt->rt_type != RTN_LOCAL)
2176 		goto skip_validate_source;
2177 
2178 	tos &= IPTOS_RT_MASK;
2179 	err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2180 	if (err < 0)
2181 		goto martian_source;
2182 
2183 skip_validate_source:
2184 	skb_dst_copy(skb, hint);
2185 	return 0;
2186 
2187 martian_source:
2188 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2189 	return err;
2190 }
2191 
2192 /* get device for dst_alloc with local routes */
2193 static struct net_device *ip_rt_get_dev(struct net *net,
2194 					const struct fib_result *res)
2195 {
2196 	struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2197 	struct net_device *dev = NULL;
2198 
2199 	if (nhc)
2200 		dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2201 
2202 	return dev ? : net->loopback_dev;
2203 }
2204 
2205 /*
2206  *	NOTE. We drop all the packets that has local source
2207  *	addresses, because every properly looped back packet
2208  *	must have correct destination already attached by output routine.
2209  *	Changes in the enforced policies must be applied also to
2210  *	ip_route_use_hint().
2211  *
2212  *	Such approach solves two big problems:
2213  *	1. Not simplex devices are handled properly.
2214  *	2. IP spoofing attempts are filtered with 100% of guarantee.
2215  *	called with rcu_read_lock()
2216  */
2217 
2218 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2219 			       u8 tos, struct net_device *dev,
2220 			       struct fib_result *res)
2221 {
2222 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2223 	struct flow_keys *flkeys = NULL, _flkeys;
2224 	struct net    *net = dev_net(dev);
2225 	struct ip_tunnel_info *tun_info;
2226 	int		err = -EINVAL;
2227 	unsigned int	flags = 0;
2228 	u32		itag = 0;
2229 	struct rtable	*rth;
2230 	struct flowi4	fl4;
2231 	bool do_cache = true;
2232 
2233 	/* IP on this device is disabled. */
2234 
2235 	if (!in_dev)
2236 		goto out;
2237 
2238 	/* Check for the most weird martians, which can be not detected
2239 	 * by fib_lookup.
2240 	 */
2241 
2242 	tun_info = skb_tunnel_info(skb);
2243 	if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2244 		fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2245 	else
2246 		fl4.flowi4_tun_key.tun_id = 0;
2247 	skb_dst_drop(skb);
2248 
2249 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2250 		goto martian_source;
2251 
2252 	res->fi = NULL;
2253 	res->table = NULL;
2254 	if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2255 		goto brd_input;
2256 
2257 	/* Accept zero addresses only to limited broadcast;
2258 	 * I even do not know to fix it or not. Waiting for complains :-)
2259 	 */
2260 	if (ipv4_is_zeronet(saddr))
2261 		goto martian_source;
2262 
2263 	if (ipv4_is_zeronet(daddr))
2264 		goto martian_destination;
2265 
2266 	/* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2267 	 * and call it once if daddr or/and saddr are loopback addresses
2268 	 */
2269 	if (ipv4_is_loopback(daddr)) {
2270 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2271 			goto martian_destination;
2272 	} else if (ipv4_is_loopback(saddr)) {
2273 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2274 			goto martian_source;
2275 	}
2276 
2277 	/*
2278 	 *	Now we are ready to route packet.
2279 	 */
2280 	fl4.flowi4_l3mdev = 0;
2281 	fl4.flowi4_oif = 0;
2282 	fl4.flowi4_iif = dev->ifindex;
2283 	fl4.flowi4_mark = skb->mark;
2284 	fl4.flowi4_tos = tos;
2285 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2286 	fl4.flowi4_flags = 0;
2287 	fl4.daddr = daddr;
2288 	fl4.saddr = saddr;
2289 	fl4.flowi4_uid = sock_net_uid(net, NULL);
2290 	fl4.flowi4_multipath_hash = 0;
2291 
2292 	if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2293 		flkeys = &_flkeys;
2294 	} else {
2295 		fl4.flowi4_proto = 0;
2296 		fl4.fl4_sport = 0;
2297 		fl4.fl4_dport = 0;
2298 	}
2299 
2300 	err = fib_lookup(net, &fl4, res, 0);
2301 	if (err != 0) {
2302 		if (!IN_DEV_FORWARD(in_dev))
2303 			err = -EHOSTUNREACH;
2304 		goto no_route;
2305 	}
2306 
2307 	if (res->type == RTN_BROADCAST) {
2308 		if (IN_DEV_BFORWARD(in_dev))
2309 			goto make_route;
2310 		/* not do cache if bc_forwarding is enabled */
2311 		if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2312 			do_cache = false;
2313 		goto brd_input;
2314 	}
2315 
2316 	if (res->type == RTN_LOCAL) {
2317 		err = fib_validate_source(skb, saddr, daddr, tos,
2318 					  0, dev, in_dev, &itag);
2319 		if (err < 0)
2320 			goto martian_source;
2321 		goto local_input;
2322 	}
2323 
2324 	if (!IN_DEV_FORWARD(in_dev)) {
2325 		err = -EHOSTUNREACH;
2326 		goto no_route;
2327 	}
2328 	if (res->type != RTN_UNICAST)
2329 		goto martian_destination;
2330 
2331 make_route:
2332 	err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2333 out:	return err;
2334 
2335 brd_input:
2336 	if (skb->protocol != htons(ETH_P_IP))
2337 		goto e_inval;
2338 
2339 	if (!ipv4_is_zeronet(saddr)) {
2340 		err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2341 					  in_dev, &itag);
2342 		if (err < 0)
2343 			goto martian_source;
2344 	}
2345 	flags |= RTCF_BROADCAST;
2346 	res->type = RTN_BROADCAST;
2347 	RT_CACHE_STAT_INC(in_brd);
2348 
2349 local_input:
2350 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2351 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
2352 
2353 	do_cache &= res->fi && !itag;
2354 	if (do_cache) {
2355 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2356 
2357 		rth = rcu_dereference(nhc->nhc_rth_input);
2358 		if (rt_cache_valid(rth)) {
2359 			skb_dst_set_noref(skb, &rth->dst);
2360 			err = 0;
2361 			goto out;
2362 		}
2363 	}
2364 
2365 	rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2366 			   flags | RTCF_LOCAL, res->type, false);
2367 	if (!rth)
2368 		goto e_nobufs;
2369 
2370 	rth->dst.output= ip_rt_bug;
2371 #ifdef CONFIG_IP_ROUTE_CLASSID
2372 	rth->dst.tclassid = itag;
2373 #endif
2374 	rth->rt_is_input = 1;
2375 
2376 	RT_CACHE_STAT_INC(in_slow_tot);
2377 	if (res->type == RTN_UNREACHABLE) {
2378 		rth->dst.input= ip_error;
2379 		rth->dst.error= -err;
2380 		rth->rt_flags	&= ~RTCF_LOCAL;
2381 	}
2382 
2383 	if (do_cache) {
2384 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2385 
2386 		rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2387 		if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2388 			WARN_ON(rth->dst.input == lwtunnel_input);
2389 			rth->dst.lwtstate->orig_input = rth->dst.input;
2390 			rth->dst.input = lwtunnel_input;
2391 		}
2392 
2393 		if (unlikely(!rt_cache_route(nhc, rth)))
2394 			rt_add_uncached_list(rth);
2395 	}
2396 	skb_dst_set(skb, &rth->dst);
2397 	err = 0;
2398 	goto out;
2399 
2400 no_route:
2401 	RT_CACHE_STAT_INC(in_no_route);
2402 	res->type = RTN_UNREACHABLE;
2403 	res->fi = NULL;
2404 	res->table = NULL;
2405 	goto local_input;
2406 
2407 	/*
2408 	 *	Do not cache martian addresses: they should be logged (RFC1812)
2409 	 */
2410 martian_destination:
2411 	RT_CACHE_STAT_INC(in_martian_dst);
2412 #ifdef CONFIG_IP_ROUTE_VERBOSE
2413 	if (IN_DEV_LOG_MARTIANS(in_dev))
2414 		net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2415 				     &daddr, &saddr, dev->name);
2416 #endif
2417 
2418 e_inval:
2419 	err = -EINVAL;
2420 	goto out;
2421 
2422 e_nobufs:
2423 	err = -ENOBUFS;
2424 	goto out;
2425 
2426 martian_source:
2427 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2428 	goto out;
2429 }
2430 
2431 /* called with rcu_read_lock held */
2432 static int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2433 			      u8 tos, struct net_device *dev, struct fib_result *res)
2434 {
2435 	/* Multicast recognition logic is moved from route cache to here.
2436 	 * The problem was that too many Ethernet cards have broken/missing
2437 	 * hardware multicast filters :-( As result the host on multicasting
2438 	 * network acquires a lot of useless route cache entries, sort of
2439 	 * SDR messages from all the world. Now we try to get rid of them.
2440 	 * Really, provided software IP multicast filter is organized
2441 	 * reasonably (at least, hashed), it does not result in a slowdown
2442 	 * comparing with route cache reject entries.
2443 	 * Note, that multicast routers are not affected, because
2444 	 * route cache entry is created eventually.
2445 	 */
2446 	if (ipv4_is_multicast(daddr)) {
2447 		struct in_device *in_dev = __in_dev_get_rcu(dev);
2448 		int our = 0;
2449 		int err = -EINVAL;
2450 
2451 		if (!in_dev)
2452 			return err;
2453 		our = ip_check_mc_rcu(in_dev, daddr, saddr,
2454 				      ip_hdr(skb)->protocol);
2455 
2456 		/* check l3 master if no match yet */
2457 		if (!our && netif_is_l3_slave(dev)) {
2458 			struct in_device *l3_in_dev;
2459 
2460 			l3_in_dev = __in_dev_get_rcu(skb->dev);
2461 			if (l3_in_dev)
2462 				our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2463 						      ip_hdr(skb)->protocol);
2464 		}
2465 
2466 		if (our
2467 #ifdef CONFIG_IP_MROUTE
2468 			||
2469 		    (!ipv4_is_local_multicast(daddr) &&
2470 		     IN_DEV_MFORWARD(in_dev))
2471 #endif
2472 		   ) {
2473 			err = ip_route_input_mc(skb, daddr, saddr,
2474 						tos, dev, our);
2475 		}
2476 		return err;
2477 	}
2478 
2479 	return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2480 }
2481 
2482 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2483 			 u8 tos, struct net_device *dev)
2484 {
2485 	struct fib_result res;
2486 	int err;
2487 
2488 	tos &= IPTOS_RT_MASK;
2489 	rcu_read_lock();
2490 	err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2491 	rcu_read_unlock();
2492 
2493 	return err;
2494 }
2495 EXPORT_SYMBOL(ip_route_input_noref);
2496 
2497 /* called with rcu_read_lock() */
2498 static struct rtable *__mkroute_output(const struct fib_result *res,
2499 				       const struct flowi4 *fl4, int orig_oif,
2500 				       struct net_device *dev_out,
2501 				       unsigned int flags)
2502 {
2503 	struct fib_info *fi = res->fi;
2504 	struct fib_nh_exception *fnhe;
2505 	struct in_device *in_dev;
2506 	u16 type = res->type;
2507 	struct rtable *rth;
2508 	bool do_cache;
2509 
2510 	in_dev = __in_dev_get_rcu(dev_out);
2511 	if (!in_dev)
2512 		return ERR_PTR(-EINVAL);
2513 
2514 	if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2515 		if (ipv4_is_loopback(fl4->saddr) &&
2516 		    !(dev_out->flags & IFF_LOOPBACK) &&
2517 		    !netif_is_l3_master(dev_out))
2518 			return ERR_PTR(-EINVAL);
2519 
2520 	if (ipv4_is_lbcast(fl4->daddr))
2521 		type = RTN_BROADCAST;
2522 	else if (ipv4_is_multicast(fl4->daddr))
2523 		type = RTN_MULTICAST;
2524 	else if (ipv4_is_zeronet(fl4->daddr))
2525 		return ERR_PTR(-EINVAL);
2526 
2527 	if (dev_out->flags & IFF_LOOPBACK)
2528 		flags |= RTCF_LOCAL;
2529 
2530 	do_cache = true;
2531 	if (type == RTN_BROADCAST) {
2532 		flags |= RTCF_BROADCAST | RTCF_LOCAL;
2533 		fi = NULL;
2534 	} else if (type == RTN_MULTICAST) {
2535 		flags |= RTCF_MULTICAST | RTCF_LOCAL;
2536 		if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2537 				     fl4->flowi4_proto))
2538 			flags &= ~RTCF_LOCAL;
2539 		else
2540 			do_cache = false;
2541 		/* If multicast route do not exist use
2542 		 * default one, but do not gateway in this case.
2543 		 * Yes, it is hack.
2544 		 */
2545 		if (fi && res->prefixlen < 4)
2546 			fi = NULL;
2547 	} else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2548 		   (orig_oif != dev_out->ifindex)) {
2549 		/* For local routes that require a particular output interface
2550 		 * we do not want to cache the result.  Caching the result
2551 		 * causes incorrect behaviour when there are multiple source
2552 		 * addresses on the interface, the end result being that if the
2553 		 * intended recipient is waiting on that interface for the
2554 		 * packet he won't receive it because it will be delivered on
2555 		 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2556 		 * be set to the loopback interface as well.
2557 		 */
2558 		do_cache = false;
2559 	}
2560 
2561 	fnhe = NULL;
2562 	do_cache &= fi != NULL;
2563 	if (fi) {
2564 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2565 		struct rtable __rcu **prth;
2566 
2567 		fnhe = find_exception(nhc, fl4->daddr);
2568 		if (!do_cache)
2569 			goto add;
2570 		if (fnhe) {
2571 			prth = &fnhe->fnhe_rth_output;
2572 		} else {
2573 			if (unlikely(fl4->flowi4_flags &
2574 				     FLOWI_FLAG_KNOWN_NH &&
2575 				     !(nhc->nhc_gw_family &&
2576 				       nhc->nhc_scope == RT_SCOPE_LINK))) {
2577 				do_cache = false;
2578 				goto add;
2579 			}
2580 			prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2581 		}
2582 		rth = rcu_dereference(*prth);
2583 		if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2584 			return rth;
2585 	}
2586 
2587 add:
2588 	rth = rt_dst_alloc(dev_out, flags, type,
2589 			   IN_DEV_ORCONF(in_dev, NOXFRM));
2590 	if (!rth)
2591 		return ERR_PTR(-ENOBUFS);
2592 
2593 	rth->rt_iif = orig_oif;
2594 
2595 	RT_CACHE_STAT_INC(out_slow_tot);
2596 
2597 	if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2598 		if (flags & RTCF_LOCAL &&
2599 		    !(dev_out->flags & IFF_LOOPBACK)) {
2600 			rth->dst.output = ip_mc_output;
2601 			RT_CACHE_STAT_INC(out_slow_mc);
2602 		}
2603 #ifdef CONFIG_IP_MROUTE
2604 		if (type == RTN_MULTICAST) {
2605 			if (IN_DEV_MFORWARD(in_dev) &&
2606 			    !ipv4_is_local_multicast(fl4->daddr)) {
2607 				rth->dst.input = ip_mr_input;
2608 				rth->dst.output = ip_mc_output;
2609 			}
2610 		}
2611 #endif
2612 	}
2613 
2614 	rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2615 	lwtunnel_set_redirect(&rth->dst);
2616 
2617 	return rth;
2618 }
2619 
2620 /*
2621  * Major route resolver routine.
2622  */
2623 
2624 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2625 					const struct sk_buff *skb)
2626 {
2627 	struct fib_result res = {
2628 		.type		= RTN_UNSPEC,
2629 		.fi		= NULL,
2630 		.table		= NULL,
2631 		.tclassid	= 0,
2632 	};
2633 	struct rtable *rth;
2634 
2635 	fl4->flowi4_iif = LOOPBACK_IFINDEX;
2636 	ip_rt_fix_tos(fl4);
2637 
2638 	rcu_read_lock();
2639 	rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2640 	rcu_read_unlock();
2641 
2642 	return rth;
2643 }
2644 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2645 
2646 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2647 					    struct fib_result *res,
2648 					    const struct sk_buff *skb)
2649 {
2650 	struct net_device *dev_out = NULL;
2651 	int orig_oif = fl4->flowi4_oif;
2652 	unsigned int flags = 0;
2653 	struct rtable *rth;
2654 	int err;
2655 
2656 	if (fl4->saddr) {
2657 		if (ipv4_is_multicast(fl4->saddr) ||
2658 		    ipv4_is_lbcast(fl4->saddr) ||
2659 		    ipv4_is_zeronet(fl4->saddr)) {
2660 			rth = ERR_PTR(-EINVAL);
2661 			goto out;
2662 		}
2663 
2664 		rth = ERR_PTR(-ENETUNREACH);
2665 
2666 		/* I removed check for oif == dev_out->oif here.
2667 		 * It was wrong for two reasons:
2668 		 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2669 		 *    is assigned to multiple interfaces.
2670 		 * 2. Moreover, we are allowed to send packets with saddr
2671 		 *    of another iface. --ANK
2672 		 */
2673 
2674 		if (fl4->flowi4_oif == 0 &&
2675 		    (ipv4_is_multicast(fl4->daddr) ||
2676 		     ipv4_is_lbcast(fl4->daddr))) {
2677 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2678 			dev_out = __ip_dev_find(net, fl4->saddr, false);
2679 			if (!dev_out)
2680 				goto out;
2681 
2682 			/* Special hack: user can direct multicasts
2683 			 * and limited broadcast via necessary interface
2684 			 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2685 			 * This hack is not just for fun, it allows
2686 			 * vic,vat and friends to work.
2687 			 * They bind socket to loopback, set ttl to zero
2688 			 * and expect that it will work.
2689 			 * From the viewpoint of routing cache they are broken,
2690 			 * because we are not allowed to build multicast path
2691 			 * with loopback source addr (look, routing cache
2692 			 * cannot know, that ttl is zero, so that packet
2693 			 * will not leave this host and route is valid).
2694 			 * Luckily, this hack is good workaround.
2695 			 */
2696 
2697 			fl4->flowi4_oif = dev_out->ifindex;
2698 			goto make_route;
2699 		}
2700 
2701 		if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2702 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2703 			if (!__ip_dev_find(net, fl4->saddr, false))
2704 				goto out;
2705 		}
2706 	}
2707 
2708 
2709 	if (fl4->flowi4_oif) {
2710 		dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2711 		rth = ERR_PTR(-ENODEV);
2712 		if (!dev_out)
2713 			goto out;
2714 
2715 		/* RACE: Check return value of inet_select_addr instead. */
2716 		if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2717 			rth = ERR_PTR(-ENETUNREACH);
2718 			goto out;
2719 		}
2720 		if (ipv4_is_local_multicast(fl4->daddr) ||
2721 		    ipv4_is_lbcast(fl4->daddr) ||
2722 		    fl4->flowi4_proto == IPPROTO_IGMP) {
2723 			if (!fl4->saddr)
2724 				fl4->saddr = inet_select_addr(dev_out, 0,
2725 							      RT_SCOPE_LINK);
2726 			goto make_route;
2727 		}
2728 		if (!fl4->saddr) {
2729 			if (ipv4_is_multicast(fl4->daddr))
2730 				fl4->saddr = inet_select_addr(dev_out, 0,
2731 							      fl4->flowi4_scope);
2732 			else if (!fl4->daddr)
2733 				fl4->saddr = inet_select_addr(dev_out, 0,
2734 							      RT_SCOPE_HOST);
2735 		}
2736 	}
2737 
2738 	if (!fl4->daddr) {
2739 		fl4->daddr = fl4->saddr;
2740 		if (!fl4->daddr)
2741 			fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2742 		dev_out = net->loopback_dev;
2743 		fl4->flowi4_oif = LOOPBACK_IFINDEX;
2744 		res->type = RTN_LOCAL;
2745 		flags |= RTCF_LOCAL;
2746 		goto make_route;
2747 	}
2748 
2749 	err = fib_lookup(net, fl4, res, 0);
2750 	if (err) {
2751 		res->fi = NULL;
2752 		res->table = NULL;
2753 		if (fl4->flowi4_oif &&
2754 		    (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2755 			/* Apparently, routing tables are wrong. Assume,
2756 			 * that the destination is on link.
2757 			 *
2758 			 * WHY? DW.
2759 			 * Because we are allowed to send to iface
2760 			 * even if it has NO routes and NO assigned
2761 			 * addresses. When oif is specified, routing
2762 			 * tables are looked up with only one purpose:
2763 			 * to catch if destination is gatewayed, rather than
2764 			 * direct. Moreover, if MSG_DONTROUTE is set,
2765 			 * we send packet, ignoring both routing tables
2766 			 * and ifaddr state. --ANK
2767 			 *
2768 			 *
2769 			 * We could make it even if oif is unknown,
2770 			 * likely IPv6, but we do not.
2771 			 */
2772 
2773 			if (fl4->saddr == 0)
2774 				fl4->saddr = inet_select_addr(dev_out, 0,
2775 							      RT_SCOPE_LINK);
2776 			res->type = RTN_UNICAST;
2777 			goto make_route;
2778 		}
2779 		rth = ERR_PTR(err);
2780 		goto out;
2781 	}
2782 
2783 	if (res->type == RTN_LOCAL) {
2784 		if (!fl4->saddr) {
2785 			if (res->fi->fib_prefsrc)
2786 				fl4->saddr = res->fi->fib_prefsrc;
2787 			else
2788 				fl4->saddr = fl4->daddr;
2789 		}
2790 
2791 		/* L3 master device is the loopback for that domain */
2792 		dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2793 			net->loopback_dev;
2794 
2795 		/* make sure orig_oif points to fib result device even
2796 		 * though packet rx/tx happens over loopback or l3mdev
2797 		 */
2798 		orig_oif = FIB_RES_OIF(*res);
2799 
2800 		fl4->flowi4_oif = dev_out->ifindex;
2801 		flags |= RTCF_LOCAL;
2802 		goto make_route;
2803 	}
2804 
2805 	fib_select_path(net, res, fl4, skb);
2806 
2807 	dev_out = FIB_RES_DEV(*res);
2808 
2809 make_route:
2810 	rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2811 
2812 out:
2813 	return rth;
2814 }
2815 
2816 static struct dst_ops ipv4_dst_blackhole_ops = {
2817 	.family			= AF_INET,
2818 	.default_advmss		= ipv4_default_advmss,
2819 	.neigh_lookup		= ipv4_neigh_lookup,
2820 	.check			= dst_blackhole_check,
2821 	.cow_metrics		= dst_blackhole_cow_metrics,
2822 	.update_pmtu		= dst_blackhole_update_pmtu,
2823 	.redirect		= dst_blackhole_redirect,
2824 	.mtu			= dst_blackhole_mtu,
2825 };
2826 
2827 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2828 {
2829 	struct rtable *ort = (struct rtable *) dst_orig;
2830 	struct rtable *rt;
2831 
2832 	rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2833 	if (rt) {
2834 		struct dst_entry *new = &rt->dst;
2835 
2836 		new->__use = 1;
2837 		new->input = dst_discard;
2838 		new->output = dst_discard_out;
2839 
2840 		new->dev = net->loopback_dev;
2841 		netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2842 
2843 		rt->rt_is_input = ort->rt_is_input;
2844 		rt->rt_iif = ort->rt_iif;
2845 		rt->rt_pmtu = ort->rt_pmtu;
2846 		rt->rt_mtu_locked = ort->rt_mtu_locked;
2847 
2848 		rt->rt_genid = rt_genid_ipv4(net);
2849 		rt->rt_flags = ort->rt_flags;
2850 		rt->rt_type = ort->rt_type;
2851 		rt->rt_uses_gateway = ort->rt_uses_gateway;
2852 		rt->rt_gw_family = ort->rt_gw_family;
2853 		if (rt->rt_gw_family == AF_INET)
2854 			rt->rt_gw4 = ort->rt_gw4;
2855 		else if (rt->rt_gw_family == AF_INET6)
2856 			rt->rt_gw6 = ort->rt_gw6;
2857 	}
2858 
2859 	dst_release(dst_orig);
2860 
2861 	return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2862 }
2863 
2864 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2865 				    const struct sock *sk)
2866 {
2867 	struct rtable *rt = __ip_route_output_key(net, flp4);
2868 
2869 	if (IS_ERR(rt))
2870 		return rt;
2871 
2872 	if (flp4->flowi4_proto) {
2873 		flp4->flowi4_oif = rt->dst.dev->ifindex;
2874 		rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2875 							flowi4_to_flowi(flp4),
2876 							sk, 0);
2877 	}
2878 
2879 	return rt;
2880 }
2881 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2882 
2883 struct rtable *ip_route_output_tunnel(struct sk_buff *skb,
2884 				      struct net_device *dev,
2885 				      struct net *net, __be32 *saddr,
2886 				      const struct ip_tunnel_info *info,
2887 				      u8 protocol, bool use_cache)
2888 {
2889 #ifdef CONFIG_DST_CACHE
2890 	struct dst_cache *dst_cache;
2891 #endif
2892 	struct rtable *rt = NULL;
2893 	struct flowi4 fl4;
2894 	__u8 tos;
2895 
2896 #ifdef CONFIG_DST_CACHE
2897 	dst_cache = (struct dst_cache *)&info->dst_cache;
2898 	if (use_cache) {
2899 		rt = dst_cache_get_ip4(dst_cache, saddr);
2900 		if (rt)
2901 			return rt;
2902 	}
2903 #endif
2904 	memset(&fl4, 0, sizeof(fl4));
2905 	fl4.flowi4_mark = skb->mark;
2906 	fl4.flowi4_proto = protocol;
2907 	fl4.daddr = info->key.u.ipv4.dst;
2908 	fl4.saddr = info->key.u.ipv4.src;
2909 	tos = info->key.tos;
2910 	fl4.flowi4_tos = RT_TOS(tos);
2911 
2912 	rt = ip_route_output_key(net, &fl4);
2913 	if (IS_ERR(rt)) {
2914 		netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
2915 		return ERR_PTR(-ENETUNREACH);
2916 	}
2917 	if (rt->dst.dev == dev) { /* is this necessary? */
2918 		netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr);
2919 		ip_rt_put(rt);
2920 		return ERR_PTR(-ELOOP);
2921 	}
2922 #ifdef CONFIG_DST_CACHE
2923 	if (use_cache)
2924 		dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2925 #endif
2926 	*saddr = fl4.saddr;
2927 	return rt;
2928 }
2929 EXPORT_SYMBOL_GPL(ip_route_output_tunnel);
2930 
2931 /* called with rcu_read_lock held */
2932 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2933 			struct rtable *rt, u32 table_id, dscp_t dscp,
2934 			struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2935 			u32 seq, unsigned int flags)
2936 {
2937 	struct rtmsg *r;
2938 	struct nlmsghdr *nlh;
2939 	unsigned long expires = 0;
2940 	u32 error;
2941 	u32 metrics[RTAX_MAX];
2942 
2943 	nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2944 	if (!nlh)
2945 		return -EMSGSIZE;
2946 
2947 	r = nlmsg_data(nlh);
2948 	r->rtm_family	 = AF_INET;
2949 	r->rtm_dst_len	= 32;
2950 	r->rtm_src_len	= 0;
2951 	r->rtm_tos	= inet_dscp_to_dsfield(dscp);
2952 	r->rtm_table	= table_id < 256 ? table_id : RT_TABLE_COMPAT;
2953 	if (nla_put_u32(skb, RTA_TABLE, table_id))
2954 		goto nla_put_failure;
2955 	r->rtm_type	= rt->rt_type;
2956 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2957 	r->rtm_protocol = RTPROT_UNSPEC;
2958 	r->rtm_flags	= (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2959 	if (rt->rt_flags & RTCF_NOTIFY)
2960 		r->rtm_flags |= RTM_F_NOTIFY;
2961 	if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2962 		r->rtm_flags |= RTCF_DOREDIRECT;
2963 
2964 	if (nla_put_in_addr(skb, RTA_DST, dst))
2965 		goto nla_put_failure;
2966 	if (src) {
2967 		r->rtm_src_len = 32;
2968 		if (nla_put_in_addr(skb, RTA_SRC, src))
2969 			goto nla_put_failure;
2970 	}
2971 	if (rt->dst.dev &&
2972 	    nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2973 		goto nla_put_failure;
2974 	if (rt->dst.lwtstate &&
2975 	    lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2976 		goto nla_put_failure;
2977 #ifdef CONFIG_IP_ROUTE_CLASSID
2978 	if (rt->dst.tclassid &&
2979 	    nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2980 		goto nla_put_failure;
2981 #endif
2982 	if (fl4 && !rt_is_input_route(rt) &&
2983 	    fl4->saddr != src) {
2984 		if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2985 			goto nla_put_failure;
2986 	}
2987 	if (rt->rt_uses_gateway) {
2988 		if (rt->rt_gw_family == AF_INET &&
2989 		    nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2990 			goto nla_put_failure;
2991 		} else if (rt->rt_gw_family == AF_INET6) {
2992 			int alen = sizeof(struct in6_addr);
2993 			struct nlattr *nla;
2994 			struct rtvia *via;
2995 
2996 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
2997 			if (!nla)
2998 				goto nla_put_failure;
2999 
3000 			via = nla_data(nla);
3001 			via->rtvia_family = AF_INET6;
3002 			memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
3003 		}
3004 	}
3005 
3006 	expires = rt->dst.expires;
3007 	if (expires) {
3008 		unsigned long now = jiffies;
3009 
3010 		if (time_before(now, expires))
3011 			expires -= now;
3012 		else
3013 			expires = 0;
3014 	}
3015 
3016 	memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3017 	if (rt->rt_pmtu && expires)
3018 		metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3019 	if (rt->rt_mtu_locked && expires)
3020 		metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3021 	if (rtnetlink_put_metrics(skb, metrics) < 0)
3022 		goto nla_put_failure;
3023 
3024 	if (fl4) {
3025 		if (fl4->flowi4_mark &&
3026 		    nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3027 			goto nla_put_failure;
3028 
3029 		if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3030 		    nla_put_u32(skb, RTA_UID,
3031 				from_kuid_munged(current_user_ns(),
3032 						 fl4->flowi4_uid)))
3033 			goto nla_put_failure;
3034 
3035 		if (rt_is_input_route(rt)) {
3036 #ifdef CONFIG_IP_MROUTE
3037 			if (ipv4_is_multicast(dst) &&
3038 			    !ipv4_is_local_multicast(dst) &&
3039 			    IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
3040 				int err = ipmr_get_route(net, skb,
3041 							 fl4->saddr, fl4->daddr,
3042 							 r, portid);
3043 
3044 				if (err <= 0) {
3045 					if (err == 0)
3046 						return 0;
3047 					goto nla_put_failure;
3048 				}
3049 			} else
3050 #endif
3051 				if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3052 					goto nla_put_failure;
3053 		}
3054 	}
3055 
3056 	error = rt->dst.error;
3057 
3058 	if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3059 		goto nla_put_failure;
3060 
3061 	nlmsg_end(skb, nlh);
3062 	return 0;
3063 
3064 nla_put_failure:
3065 	nlmsg_cancel(skb, nlh);
3066 	return -EMSGSIZE;
3067 }
3068 
3069 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3070 			    struct netlink_callback *cb, u32 table_id,
3071 			    struct fnhe_hash_bucket *bucket, int genid,
3072 			    int *fa_index, int fa_start, unsigned int flags)
3073 {
3074 	int i;
3075 
3076 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
3077 		struct fib_nh_exception *fnhe;
3078 
3079 		for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3080 		     fnhe = rcu_dereference(fnhe->fnhe_next)) {
3081 			struct rtable *rt;
3082 			int err;
3083 
3084 			if (*fa_index < fa_start)
3085 				goto next;
3086 
3087 			if (fnhe->fnhe_genid != genid)
3088 				goto next;
3089 
3090 			if (fnhe->fnhe_expires &&
3091 			    time_after(jiffies, fnhe->fnhe_expires))
3092 				goto next;
3093 
3094 			rt = rcu_dereference(fnhe->fnhe_rth_input);
3095 			if (!rt)
3096 				rt = rcu_dereference(fnhe->fnhe_rth_output);
3097 			if (!rt)
3098 				goto next;
3099 
3100 			err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3101 					   table_id, 0, NULL, skb,
3102 					   NETLINK_CB(cb->skb).portid,
3103 					   cb->nlh->nlmsg_seq, flags);
3104 			if (err)
3105 				return err;
3106 next:
3107 			(*fa_index)++;
3108 		}
3109 	}
3110 
3111 	return 0;
3112 }
3113 
3114 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3115 		       u32 table_id, struct fib_info *fi,
3116 		       int *fa_index, int fa_start, unsigned int flags)
3117 {
3118 	struct net *net = sock_net(cb->skb->sk);
3119 	int nhsel, genid = fnhe_genid(net);
3120 
3121 	for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3122 		struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3123 		struct fnhe_hash_bucket *bucket;
3124 		int err;
3125 
3126 		if (nhc->nhc_flags & RTNH_F_DEAD)
3127 			continue;
3128 
3129 		rcu_read_lock();
3130 		bucket = rcu_dereference(nhc->nhc_exceptions);
3131 		err = 0;
3132 		if (bucket)
3133 			err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3134 					       genid, fa_index, fa_start,
3135 					       flags);
3136 		rcu_read_unlock();
3137 		if (err)
3138 			return err;
3139 	}
3140 
3141 	return 0;
3142 }
3143 
3144 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3145 						   u8 ip_proto, __be16 sport,
3146 						   __be16 dport)
3147 {
3148 	struct sk_buff *skb;
3149 	struct iphdr *iph;
3150 
3151 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3152 	if (!skb)
3153 		return NULL;
3154 
3155 	/* Reserve room for dummy headers, this skb can pass
3156 	 * through good chunk of routing engine.
3157 	 */
3158 	skb_reset_mac_header(skb);
3159 	skb_reset_network_header(skb);
3160 	skb->protocol = htons(ETH_P_IP);
3161 	iph = skb_put(skb, sizeof(struct iphdr));
3162 	iph->protocol = ip_proto;
3163 	iph->saddr = src;
3164 	iph->daddr = dst;
3165 	iph->version = 0x4;
3166 	iph->frag_off = 0;
3167 	iph->ihl = 0x5;
3168 	skb_set_transport_header(skb, skb->len);
3169 
3170 	switch (iph->protocol) {
3171 	case IPPROTO_UDP: {
3172 		struct udphdr *udph;
3173 
3174 		udph = skb_put_zero(skb, sizeof(struct udphdr));
3175 		udph->source = sport;
3176 		udph->dest = dport;
3177 		udph->len = htons(sizeof(struct udphdr));
3178 		udph->check = 0;
3179 		break;
3180 	}
3181 	case IPPROTO_TCP: {
3182 		struct tcphdr *tcph;
3183 
3184 		tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3185 		tcph->source	= sport;
3186 		tcph->dest	= dport;
3187 		tcph->doff	= sizeof(struct tcphdr) / 4;
3188 		tcph->rst = 1;
3189 		tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3190 					    src, dst, 0);
3191 		break;
3192 	}
3193 	case IPPROTO_ICMP: {
3194 		struct icmphdr *icmph;
3195 
3196 		icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3197 		icmph->type = ICMP_ECHO;
3198 		icmph->code = 0;
3199 	}
3200 	}
3201 
3202 	return skb;
3203 }
3204 
3205 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3206 				       const struct nlmsghdr *nlh,
3207 				       struct nlattr **tb,
3208 				       struct netlink_ext_ack *extack)
3209 {
3210 	struct rtmsg *rtm;
3211 	int i, err;
3212 
3213 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3214 		NL_SET_ERR_MSG(extack,
3215 			       "ipv4: Invalid header for route get request");
3216 		return -EINVAL;
3217 	}
3218 
3219 	if (!netlink_strict_get_check(skb))
3220 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3221 					      rtm_ipv4_policy, extack);
3222 
3223 	rtm = nlmsg_data(nlh);
3224 	if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3225 	    (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3226 	    rtm->rtm_table || rtm->rtm_protocol ||
3227 	    rtm->rtm_scope || rtm->rtm_type) {
3228 		NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3229 		return -EINVAL;
3230 	}
3231 
3232 	if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3233 			       RTM_F_LOOKUP_TABLE |
3234 			       RTM_F_FIB_MATCH)) {
3235 		NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3236 		return -EINVAL;
3237 	}
3238 
3239 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3240 					    rtm_ipv4_policy, extack);
3241 	if (err)
3242 		return err;
3243 
3244 	if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3245 	    (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3246 		NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3247 		return -EINVAL;
3248 	}
3249 
3250 	for (i = 0; i <= RTA_MAX; i++) {
3251 		if (!tb[i])
3252 			continue;
3253 
3254 		switch (i) {
3255 		case RTA_IIF:
3256 		case RTA_OIF:
3257 		case RTA_SRC:
3258 		case RTA_DST:
3259 		case RTA_IP_PROTO:
3260 		case RTA_SPORT:
3261 		case RTA_DPORT:
3262 		case RTA_MARK:
3263 		case RTA_UID:
3264 			break;
3265 		default:
3266 			NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3267 			return -EINVAL;
3268 		}
3269 	}
3270 
3271 	return 0;
3272 }
3273 
3274 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3275 			     struct netlink_ext_ack *extack)
3276 {
3277 	struct net *net = sock_net(in_skb->sk);
3278 	struct nlattr *tb[RTA_MAX+1];
3279 	u32 table_id = RT_TABLE_MAIN;
3280 	__be16 sport = 0, dport = 0;
3281 	struct fib_result res = {};
3282 	u8 ip_proto = IPPROTO_UDP;
3283 	struct rtable *rt = NULL;
3284 	struct sk_buff *skb;
3285 	struct rtmsg *rtm;
3286 	struct flowi4 fl4 = {};
3287 	__be32 dst = 0;
3288 	__be32 src = 0;
3289 	kuid_t uid;
3290 	u32 iif;
3291 	int err;
3292 	int mark;
3293 
3294 	err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3295 	if (err < 0)
3296 		return err;
3297 
3298 	rtm = nlmsg_data(nlh);
3299 	src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3300 	dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3301 	iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3302 	mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3303 	if (tb[RTA_UID])
3304 		uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3305 	else
3306 		uid = (iif ? INVALID_UID : current_uid());
3307 
3308 	if (tb[RTA_IP_PROTO]) {
3309 		err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3310 						  &ip_proto, AF_INET, extack);
3311 		if (err)
3312 			return err;
3313 	}
3314 
3315 	if (tb[RTA_SPORT])
3316 		sport = nla_get_be16(tb[RTA_SPORT]);
3317 
3318 	if (tb[RTA_DPORT])
3319 		dport = nla_get_be16(tb[RTA_DPORT]);
3320 
3321 	skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3322 	if (!skb)
3323 		return -ENOBUFS;
3324 
3325 	fl4.daddr = dst;
3326 	fl4.saddr = src;
3327 	fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3328 	fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3329 	fl4.flowi4_mark = mark;
3330 	fl4.flowi4_uid = uid;
3331 	if (sport)
3332 		fl4.fl4_sport = sport;
3333 	if (dport)
3334 		fl4.fl4_dport = dport;
3335 	fl4.flowi4_proto = ip_proto;
3336 
3337 	rcu_read_lock();
3338 
3339 	if (iif) {
3340 		struct net_device *dev;
3341 
3342 		dev = dev_get_by_index_rcu(net, iif);
3343 		if (!dev) {
3344 			err = -ENODEV;
3345 			goto errout_rcu;
3346 		}
3347 
3348 		fl4.flowi4_iif = iif; /* for rt_fill_info */
3349 		skb->dev	= dev;
3350 		skb->mark	= mark;
3351 		err = ip_route_input_rcu(skb, dst, src,
3352 					 rtm->rtm_tos & IPTOS_RT_MASK, dev,
3353 					 &res);
3354 
3355 		rt = skb_rtable(skb);
3356 		if (err == 0 && rt->dst.error)
3357 			err = -rt->dst.error;
3358 	} else {
3359 		fl4.flowi4_iif = LOOPBACK_IFINDEX;
3360 		skb->dev = net->loopback_dev;
3361 		rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3362 		err = 0;
3363 		if (IS_ERR(rt))
3364 			err = PTR_ERR(rt);
3365 		else
3366 			skb_dst_set(skb, &rt->dst);
3367 	}
3368 
3369 	if (err)
3370 		goto errout_rcu;
3371 
3372 	if (rtm->rtm_flags & RTM_F_NOTIFY)
3373 		rt->rt_flags |= RTCF_NOTIFY;
3374 
3375 	if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3376 		table_id = res.table ? res.table->tb_id : 0;
3377 
3378 	/* reset skb for netlink reply msg */
3379 	skb_trim(skb, 0);
3380 	skb_reset_network_header(skb);
3381 	skb_reset_transport_header(skb);
3382 	skb_reset_mac_header(skb);
3383 
3384 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3385 		struct fib_rt_info fri;
3386 
3387 		if (!res.fi) {
3388 			err = fib_props[res.type].error;
3389 			if (!err)
3390 				err = -EHOSTUNREACH;
3391 			goto errout_rcu;
3392 		}
3393 		fri.fi = res.fi;
3394 		fri.tb_id = table_id;
3395 		fri.dst = res.prefix;
3396 		fri.dst_len = res.prefixlen;
3397 		fri.dscp = res.dscp;
3398 		fri.type = rt->rt_type;
3399 		fri.offload = 0;
3400 		fri.trap = 0;
3401 		fri.offload_failed = 0;
3402 		if (res.fa_head) {
3403 			struct fib_alias *fa;
3404 
3405 			hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3406 				u8 slen = 32 - fri.dst_len;
3407 
3408 				if (fa->fa_slen == slen &&
3409 				    fa->tb_id == fri.tb_id &&
3410 				    fa->fa_dscp == fri.dscp &&
3411 				    fa->fa_info == res.fi &&
3412 				    fa->fa_type == fri.type) {
3413 					fri.offload = READ_ONCE(fa->offload);
3414 					fri.trap = READ_ONCE(fa->trap);
3415 					fri.offload_failed =
3416 						READ_ONCE(fa->offload_failed);
3417 					break;
3418 				}
3419 			}
3420 		}
3421 		err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3422 				    nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3423 	} else {
3424 		err = rt_fill_info(net, dst, src, rt, table_id, res.dscp, &fl4,
3425 				   skb, NETLINK_CB(in_skb).portid,
3426 				   nlh->nlmsg_seq, 0);
3427 	}
3428 	if (err < 0)
3429 		goto errout_rcu;
3430 
3431 	rcu_read_unlock();
3432 
3433 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3434 
3435 errout_free:
3436 	return err;
3437 errout_rcu:
3438 	rcu_read_unlock();
3439 	kfree_skb(skb);
3440 	goto errout_free;
3441 }
3442 
3443 void ip_rt_multicast_event(struct in_device *in_dev)
3444 {
3445 	rt_cache_flush(dev_net(in_dev->dev));
3446 }
3447 
3448 #ifdef CONFIG_SYSCTL
3449 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3450 static int ip_rt_gc_min_interval __read_mostly	= HZ / 2;
3451 static int ip_rt_gc_elasticity __read_mostly	= 8;
3452 static int ip_min_valid_pmtu __read_mostly	= IPV4_MIN_MTU;
3453 
3454 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3455 		void *buffer, size_t *lenp, loff_t *ppos)
3456 {
3457 	struct net *net = (struct net *)__ctl->extra1;
3458 
3459 	if (write) {
3460 		rt_cache_flush(net);
3461 		fnhe_genid_bump(net);
3462 		return 0;
3463 	}
3464 
3465 	return -EINVAL;
3466 }
3467 
3468 static struct ctl_table ipv4_route_table[] = {
3469 	{
3470 		.procname	= "gc_thresh",
3471 		.data		= &ipv4_dst_ops.gc_thresh,
3472 		.maxlen		= sizeof(int),
3473 		.mode		= 0644,
3474 		.proc_handler	= proc_dointvec,
3475 	},
3476 	{
3477 		.procname	= "max_size",
3478 		.data		= &ip_rt_max_size,
3479 		.maxlen		= sizeof(int),
3480 		.mode		= 0644,
3481 		.proc_handler	= proc_dointvec,
3482 	},
3483 	{
3484 		/*  Deprecated. Use gc_min_interval_ms */
3485 
3486 		.procname	= "gc_min_interval",
3487 		.data		= &ip_rt_gc_min_interval,
3488 		.maxlen		= sizeof(int),
3489 		.mode		= 0644,
3490 		.proc_handler	= proc_dointvec_jiffies,
3491 	},
3492 	{
3493 		.procname	= "gc_min_interval_ms",
3494 		.data		= &ip_rt_gc_min_interval,
3495 		.maxlen		= sizeof(int),
3496 		.mode		= 0644,
3497 		.proc_handler	= proc_dointvec_ms_jiffies,
3498 	},
3499 	{
3500 		.procname	= "gc_timeout",
3501 		.data		= &ip_rt_gc_timeout,
3502 		.maxlen		= sizeof(int),
3503 		.mode		= 0644,
3504 		.proc_handler	= proc_dointvec_jiffies,
3505 	},
3506 	{
3507 		.procname	= "gc_interval",
3508 		.data		= &ip_rt_gc_interval,
3509 		.maxlen		= sizeof(int),
3510 		.mode		= 0644,
3511 		.proc_handler	= proc_dointvec_jiffies,
3512 	},
3513 	{
3514 		.procname	= "redirect_load",
3515 		.data		= &ip_rt_redirect_load,
3516 		.maxlen		= sizeof(int),
3517 		.mode		= 0644,
3518 		.proc_handler	= proc_dointvec,
3519 	},
3520 	{
3521 		.procname	= "redirect_number",
3522 		.data		= &ip_rt_redirect_number,
3523 		.maxlen		= sizeof(int),
3524 		.mode		= 0644,
3525 		.proc_handler	= proc_dointvec,
3526 	},
3527 	{
3528 		.procname	= "redirect_silence",
3529 		.data		= &ip_rt_redirect_silence,
3530 		.maxlen		= sizeof(int),
3531 		.mode		= 0644,
3532 		.proc_handler	= proc_dointvec,
3533 	},
3534 	{
3535 		.procname	= "error_cost",
3536 		.data		= &ip_rt_error_cost,
3537 		.maxlen		= sizeof(int),
3538 		.mode		= 0644,
3539 		.proc_handler	= proc_dointvec,
3540 	},
3541 	{
3542 		.procname	= "error_burst",
3543 		.data		= &ip_rt_error_burst,
3544 		.maxlen		= sizeof(int),
3545 		.mode		= 0644,
3546 		.proc_handler	= proc_dointvec,
3547 	},
3548 	{
3549 		.procname	= "gc_elasticity",
3550 		.data		= &ip_rt_gc_elasticity,
3551 		.maxlen		= sizeof(int),
3552 		.mode		= 0644,
3553 		.proc_handler	= proc_dointvec,
3554 	},
3555 	{ }
3556 };
3557 
3558 static const char ipv4_route_flush_procname[] = "flush";
3559 
3560 static struct ctl_table ipv4_route_netns_table[] = {
3561 	{
3562 		.procname	= ipv4_route_flush_procname,
3563 		.maxlen		= sizeof(int),
3564 		.mode		= 0200,
3565 		.proc_handler	= ipv4_sysctl_rtcache_flush,
3566 	},
3567 	{
3568 		.procname       = "min_pmtu",
3569 		.data           = &init_net.ipv4.ip_rt_min_pmtu,
3570 		.maxlen         = sizeof(int),
3571 		.mode           = 0644,
3572 		.proc_handler   = proc_dointvec_minmax,
3573 		.extra1         = &ip_min_valid_pmtu,
3574 	},
3575 	{
3576 		.procname       = "mtu_expires",
3577 		.data           = &init_net.ipv4.ip_rt_mtu_expires,
3578 		.maxlen         = sizeof(int),
3579 		.mode           = 0644,
3580 		.proc_handler   = proc_dointvec_jiffies,
3581 	},
3582 	{
3583 		.procname   = "min_adv_mss",
3584 		.data       = &init_net.ipv4.ip_rt_min_advmss,
3585 		.maxlen     = sizeof(int),
3586 		.mode       = 0644,
3587 		.proc_handler   = proc_dointvec,
3588 	},
3589 	{ },
3590 };
3591 
3592 static __net_init int sysctl_route_net_init(struct net *net)
3593 {
3594 	struct ctl_table *tbl;
3595 	size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3596 
3597 	tbl = ipv4_route_netns_table;
3598 	if (!net_eq(net, &init_net)) {
3599 		int i;
3600 
3601 		tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3602 		if (!tbl)
3603 			goto err_dup;
3604 
3605 		/* Don't export non-whitelisted sysctls to unprivileged users */
3606 		if (net->user_ns != &init_user_ns) {
3607 			if (tbl[0].procname != ipv4_route_flush_procname) {
3608 				tbl[0].procname = NULL;
3609 				table_size = 0;
3610 			}
3611 		}
3612 
3613 		/* Update the variables to point into the current struct net
3614 		 * except for the first element flush
3615 		 */
3616 		for (i = 1; i < ARRAY_SIZE(ipv4_route_netns_table) - 1; i++)
3617 			tbl[i].data += (void *)net - (void *)&init_net;
3618 	}
3619 	tbl[0].extra1 = net;
3620 
3621 	net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3622 						     tbl, table_size);
3623 	if (!net->ipv4.route_hdr)
3624 		goto err_reg;
3625 	return 0;
3626 
3627 err_reg:
3628 	if (tbl != ipv4_route_netns_table)
3629 		kfree(tbl);
3630 err_dup:
3631 	return -ENOMEM;
3632 }
3633 
3634 static __net_exit void sysctl_route_net_exit(struct net *net)
3635 {
3636 	struct ctl_table *tbl;
3637 
3638 	tbl = net->ipv4.route_hdr->ctl_table_arg;
3639 	unregister_net_sysctl_table(net->ipv4.route_hdr);
3640 	BUG_ON(tbl == ipv4_route_netns_table);
3641 	kfree(tbl);
3642 }
3643 
3644 static __net_initdata struct pernet_operations sysctl_route_ops = {
3645 	.init = sysctl_route_net_init,
3646 	.exit = sysctl_route_net_exit,
3647 };
3648 #endif
3649 
3650 static __net_init int netns_ip_rt_init(struct net *net)
3651 {
3652 	/* Set default value for namespaceified sysctls */
3653 	net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3654 	net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3655 	net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3656 	return 0;
3657 }
3658 
3659 static struct pernet_operations __net_initdata ip_rt_ops = {
3660 	.init = netns_ip_rt_init,
3661 };
3662 
3663 static __net_init int rt_genid_init(struct net *net)
3664 {
3665 	atomic_set(&net->ipv4.rt_genid, 0);
3666 	atomic_set(&net->fnhe_genid, 0);
3667 	atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3668 	return 0;
3669 }
3670 
3671 static __net_initdata struct pernet_operations rt_genid_ops = {
3672 	.init = rt_genid_init,
3673 };
3674 
3675 static int __net_init ipv4_inetpeer_init(struct net *net)
3676 {
3677 	struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3678 
3679 	if (!bp)
3680 		return -ENOMEM;
3681 	inet_peer_base_init(bp);
3682 	net->ipv4.peers = bp;
3683 	return 0;
3684 }
3685 
3686 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3687 {
3688 	struct inet_peer_base *bp = net->ipv4.peers;
3689 
3690 	net->ipv4.peers = NULL;
3691 	inetpeer_invalidate_tree(bp);
3692 	kfree(bp);
3693 }
3694 
3695 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3696 	.init	=	ipv4_inetpeer_init,
3697 	.exit	=	ipv4_inetpeer_exit,
3698 };
3699 
3700 #ifdef CONFIG_IP_ROUTE_CLASSID
3701 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3702 #endif /* CONFIG_IP_ROUTE_CLASSID */
3703 
3704 int __init ip_rt_init(void)
3705 {
3706 	void *idents_hash;
3707 	int cpu;
3708 
3709 	/* For modern hosts, this will use 2 MB of memory */
3710 	idents_hash = alloc_large_system_hash("IP idents",
3711 					      sizeof(*ip_idents) + sizeof(*ip_tstamps),
3712 					      0,
3713 					      16, /* one bucket per 64 KB */
3714 					      HASH_ZERO,
3715 					      NULL,
3716 					      &ip_idents_mask,
3717 					      2048,
3718 					      256*1024);
3719 
3720 	ip_idents = idents_hash;
3721 
3722 	get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3723 
3724 	ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3725 
3726 	for_each_possible_cpu(cpu) {
3727 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3728 
3729 		INIT_LIST_HEAD(&ul->head);
3730 		INIT_LIST_HEAD(&ul->quarantine);
3731 		spin_lock_init(&ul->lock);
3732 	}
3733 #ifdef CONFIG_IP_ROUTE_CLASSID
3734 	ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3735 	if (!ip_rt_acct)
3736 		panic("IP: failed to allocate ip_rt_acct\n");
3737 #endif
3738 
3739 	ipv4_dst_ops.kmem_cachep =
3740 		kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3741 				  SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3742 
3743 	ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3744 
3745 	if (dst_entries_init(&ipv4_dst_ops) < 0)
3746 		panic("IP: failed to allocate ipv4_dst_ops counter\n");
3747 
3748 	if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3749 		panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3750 
3751 	ipv4_dst_ops.gc_thresh = ~0;
3752 	ip_rt_max_size = INT_MAX;
3753 
3754 	devinet_init();
3755 	ip_fib_init();
3756 
3757 	if (ip_rt_proc_init())
3758 		pr_err("Unable to create route proc files\n");
3759 #ifdef CONFIG_XFRM
3760 	xfrm_init();
3761 	xfrm4_init();
3762 #endif
3763 	rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3764 		      RTNL_FLAG_DOIT_UNLOCKED);
3765 
3766 #ifdef CONFIG_SYSCTL
3767 	register_pernet_subsys(&sysctl_route_ops);
3768 #endif
3769 	register_pernet_subsys(&ip_rt_ops);
3770 	register_pernet_subsys(&rt_genid_ops);
3771 	register_pernet_subsys(&ipv4_inetpeer_ops);
3772 	return 0;
3773 }
3774 
3775 #ifdef CONFIG_SYSCTL
3776 /*
3777  * We really need to sanitize the damn ipv4 init order, then all
3778  * this nonsense will go away.
3779  */
3780 void __init ip_static_sysctl_init(void)
3781 {
3782 	register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3783 }
3784 #endif
3785