xref: /openbmc/linux/net/ipv4/route.c (revision 7d9e5f422150ed00de744e02a80734d74cc9704d)
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/uaccess.h>
65 #include <linux/bitops.h>
66 #include <linux/types.h>
67 #include <linux/kernel.h>
68 #include <linux/mm.h>
69 #include <linux/string.h>
70 #include <linux/socket.h>
71 #include <linux/sockios.h>
72 #include <linux/errno.h>
73 #include <linux/in.h>
74 #include <linux/inet.h>
75 #include <linux/netdevice.h>
76 #include <linux/proc_fs.h>
77 #include <linux/init.h>
78 #include <linux/skbuff.h>
79 #include <linux/inetdevice.h>
80 #include <linux/igmp.h>
81 #include <linux/pkt_sched.h>
82 #include <linux/mroute.h>
83 #include <linux/netfilter_ipv4.h>
84 #include <linux/random.h>
85 #include <linux/rcupdate.h>
86 #include <linux/times.h>
87 #include <linux/slab.h>
88 #include <linux/jhash.h>
89 #include <net/dst.h>
90 #include <net/dst_metadata.h>
91 #include <net/net_namespace.h>
92 #include <net/protocol.h>
93 #include <net/ip.h>
94 #include <net/route.h>
95 #include <net/inetpeer.h>
96 #include <net/sock.h>
97 #include <net/ip_fib.h>
98 #include <net/nexthop.h>
99 #include <net/arp.h>
100 #include <net/tcp.h>
101 #include <net/icmp.h>
102 #include <net/xfrm.h>
103 #include <net/lwtunnel.h>
104 #include <net/netevent.h>
105 #include <net/rtnetlink.h>
106 #ifdef CONFIG_SYSCTL
107 #include <linux/sysctl.h>
108 #endif
109 #include <net/secure_seq.h>
110 #include <net/ip_tunnels.h>
111 #include <net/l3mdev.h>
112 
113 #include "fib_lookup.h"
114 
115 #define RT_FL_TOS(oldflp4) \
116 	((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
117 
118 #define RT_GC_TIMEOUT (300*HZ)
119 
120 static int ip_rt_max_size;
121 static int ip_rt_redirect_number __read_mostly	= 9;
122 static int ip_rt_redirect_load __read_mostly	= HZ / 50;
123 static int ip_rt_redirect_silence __read_mostly	= ((HZ / 50) << (9 + 1));
124 static int ip_rt_error_cost __read_mostly	= HZ;
125 static int ip_rt_error_burst __read_mostly	= 5 * HZ;
126 static int ip_rt_mtu_expires __read_mostly	= 10 * 60 * HZ;
127 static u32 ip_rt_min_pmtu __read_mostly		= 512 + 20 + 20;
128 static int ip_rt_min_advmss __read_mostly	= 256;
129 
130 static int ip_rt_gc_timeout __read_mostly	= RT_GC_TIMEOUT;
131 
132 /*
133  *	Interface to generic destination cache.
134  */
135 
136 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
137 static unsigned int	 ipv4_default_advmss(const struct dst_entry *dst);
138 static unsigned int	 ipv4_mtu(const struct dst_entry *dst);
139 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
140 static void		 ipv4_link_failure(struct sk_buff *skb);
141 static void		 ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
142 					   struct sk_buff *skb, u32 mtu);
143 static void		 ip_do_redirect(struct dst_entry *dst, struct sock *sk,
144 					struct sk_buff *skb);
145 static void		ipv4_dst_destroy(struct dst_entry *dst);
146 
147 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
148 {
149 	WARN_ON(1);
150 	return NULL;
151 }
152 
153 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
154 					   struct sk_buff *skb,
155 					   const void *daddr);
156 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
157 
158 static struct dst_ops ipv4_dst_ops = {
159 	.family =		AF_INET,
160 	.check =		ipv4_dst_check,
161 	.default_advmss =	ipv4_default_advmss,
162 	.mtu =			ipv4_mtu,
163 	.cow_metrics =		ipv4_cow_metrics,
164 	.destroy =		ipv4_dst_destroy,
165 	.negative_advice =	ipv4_negative_advice,
166 	.link_failure =		ipv4_link_failure,
167 	.update_pmtu =		ip_rt_update_pmtu,
168 	.redirect =		ip_do_redirect,
169 	.local_out =		__ip_local_out,
170 	.neigh_lookup =		ipv4_neigh_lookup,
171 	.confirm_neigh =	ipv4_confirm_neigh,
172 };
173 
174 #define ECN_OR_COST(class)	TC_PRIO_##class
175 
176 const __u8 ip_tos2prio[16] = {
177 	TC_PRIO_BESTEFFORT,
178 	ECN_OR_COST(BESTEFFORT),
179 	TC_PRIO_BESTEFFORT,
180 	ECN_OR_COST(BESTEFFORT),
181 	TC_PRIO_BULK,
182 	ECN_OR_COST(BULK),
183 	TC_PRIO_BULK,
184 	ECN_OR_COST(BULK),
185 	TC_PRIO_INTERACTIVE,
186 	ECN_OR_COST(INTERACTIVE),
187 	TC_PRIO_INTERACTIVE,
188 	ECN_OR_COST(INTERACTIVE),
189 	TC_PRIO_INTERACTIVE_BULK,
190 	ECN_OR_COST(INTERACTIVE_BULK),
191 	TC_PRIO_INTERACTIVE_BULK,
192 	ECN_OR_COST(INTERACTIVE_BULK)
193 };
194 EXPORT_SYMBOL(ip_tos2prio);
195 
196 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
197 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
198 
199 #ifdef CONFIG_PROC_FS
200 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
201 {
202 	if (*pos)
203 		return NULL;
204 	return SEQ_START_TOKEN;
205 }
206 
207 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
208 {
209 	++*pos;
210 	return NULL;
211 }
212 
213 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
214 {
215 }
216 
217 static int rt_cache_seq_show(struct seq_file *seq, void *v)
218 {
219 	if (v == SEQ_START_TOKEN)
220 		seq_printf(seq, "%-127s\n",
221 			   "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
222 			   "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
223 			   "HHUptod\tSpecDst");
224 	return 0;
225 }
226 
227 static const struct seq_operations rt_cache_seq_ops = {
228 	.start  = rt_cache_seq_start,
229 	.next   = rt_cache_seq_next,
230 	.stop   = rt_cache_seq_stop,
231 	.show   = rt_cache_seq_show,
232 };
233 
234 static int rt_cache_seq_open(struct inode *inode, struct file *file)
235 {
236 	return seq_open(file, &rt_cache_seq_ops);
237 }
238 
239 static const struct file_operations rt_cache_seq_fops = {
240 	.open	 = rt_cache_seq_open,
241 	.read	 = seq_read,
242 	.llseek	 = seq_lseek,
243 	.release = seq_release,
244 };
245 
246 
247 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
248 {
249 	int cpu;
250 
251 	if (*pos == 0)
252 		return SEQ_START_TOKEN;
253 
254 	for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
255 		if (!cpu_possible(cpu))
256 			continue;
257 		*pos = cpu+1;
258 		return &per_cpu(rt_cache_stat, cpu);
259 	}
260 	return NULL;
261 }
262 
263 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
264 {
265 	int cpu;
266 
267 	for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
268 		if (!cpu_possible(cpu))
269 			continue;
270 		*pos = cpu+1;
271 		return &per_cpu(rt_cache_stat, cpu);
272 	}
273 	return NULL;
274 
275 }
276 
277 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
278 {
279 
280 }
281 
282 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
283 {
284 	struct rt_cache_stat *st = v;
285 
286 	if (v == SEQ_START_TOKEN) {
287 		seq_printf(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");
288 		return 0;
289 	}
290 
291 	seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
292 		   " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
293 		   dst_entries_get_slow(&ipv4_dst_ops),
294 		   0, /* st->in_hit */
295 		   st->in_slow_tot,
296 		   st->in_slow_mc,
297 		   st->in_no_route,
298 		   st->in_brd,
299 		   st->in_martian_dst,
300 		   st->in_martian_src,
301 
302 		   0, /* st->out_hit */
303 		   st->out_slow_tot,
304 		   st->out_slow_mc,
305 
306 		   0, /* st->gc_total */
307 		   0, /* st->gc_ignored */
308 		   0, /* st->gc_goal_miss */
309 		   0, /* st->gc_dst_overflow */
310 		   0, /* st->in_hlist_search */
311 		   0  /* st->out_hlist_search */
312 		);
313 	return 0;
314 }
315 
316 static const struct seq_operations rt_cpu_seq_ops = {
317 	.start  = rt_cpu_seq_start,
318 	.next   = rt_cpu_seq_next,
319 	.stop   = rt_cpu_seq_stop,
320 	.show   = rt_cpu_seq_show,
321 };
322 
323 
324 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
325 {
326 	return seq_open(file, &rt_cpu_seq_ops);
327 }
328 
329 static const struct file_operations rt_cpu_seq_fops = {
330 	.open	 = rt_cpu_seq_open,
331 	.read	 = seq_read,
332 	.llseek	 = seq_lseek,
333 	.release = seq_release,
334 };
335 
336 #ifdef CONFIG_IP_ROUTE_CLASSID
337 static int rt_acct_proc_show(struct seq_file *m, void *v)
338 {
339 	struct ip_rt_acct *dst, *src;
340 	unsigned int i, j;
341 
342 	dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
343 	if (!dst)
344 		return -ENOMEM;
345 
346 	for_each_possible_cpu(i) {
347 		src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
348 		for (j = 0; j < 256; j++) {
349 			dst[j].o_bytes   += src[j].o_bytes;
350 			dst[j].o_packets += src[j].o_packets;
351 			dst[j].i_bytes   += src[j].i_bytes;
352 			dst[j].i_packets += src[j].i_packets;
353 		}
354 	}
355 
356 	seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
357 	kfree(dst);
358 	return 0;
359 }
360 #endif
361 
362 static int __net_init ip_rt_do_proc_init(struct net *net)
363 {
364 	struct proc_dir_entry *pde;
365 
366 	pde = proc_create("rt_cache", 0444, net->proc_net,
367 			  &rt_cache_seq_fops);
368 	if (!pde)
369 		goto err1;
370 
371 	pde = proc_create("rt_cache", 0444,
372 			  net->proc_net_stat, &rt_cpu_seq_fops);
373 	if (!pde)
374 		goto err2;
375 
376 #ifdef CONFIG_IP_ROUTE_CLASSID
377 	pde = proc_create_single("rt_acct", 0, net->proc_net,
378 			rt_acct_proc_show);
379 	if (!pde)
380 		goto err3;
381 #endif
382 	return 0;
383 
384 #ifdef CONFIG_IP_ROUTE_CLASSID
385 err3:
386 	remove_proc_entry("rt_cache", net->proc_net_stat);
387 #endif
388 err2:
389 	remove_proc_entry("rt_cache", net->proc_net);
390 err1:
391 	return -ENOMEM;
392 }
393 
394 static void __net_exit ip_rt_do_proc_exit(struct net *net)
395 {
396 	remove_proc_entry("rt_cache", net->proc_net_stat);
397 	remove_proc_entry("rt_cache", net->proc_net);
398 #ifdef CONFIG_IP_ROUTE_CLASSID
399 	remove_proc_entry("rt_acct", net->proc_net);
400 #endif
401 }
402 
403 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
404 	.init = ip_rt_do_proc_init,
405 	.exit = ip_rt_do_proc_exit,
406 };
407 
408 static int __init ip_rt_proc_init(void)
409 {
410 	return register_pernet_subsys(&ip_rt_proc_ops);
411 }
412 
413 #else
414 static inline int ip_rt_proc_init(void)
415 {
416 	return 0;
417 }
418 #endif /* CONFIG_PROC_FS */
419 
420 static inline bool rt_is_expired(const struct rtable *rth)
421 {
422 	return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
423 }
424 
425 void rt_cache_flush(struct net *net)
426 {
427 	rt_genid_bump_ipv4(net);
428 }
429 
430 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
431 					   struct sk_buff *skb,
432 					   const void *daddr)
433 {
434 	const struct rtable *rt = container_of(dst, struct rtable, dst);
435 	struct net_device *dev = dst->dev;
436 	struct neighbour *n;
437 
438 	rcu_read_lock_bh();
439 
440 	if (likely(rt->rt_gw_family == AF_INET)) {
441 		n = ip_neigh_gw4(dev, rt->rt_gw4);
442 	} else if (rt->rt_gw_family == AF_INET6) {
443 		n = ip_neigh_gw6(dev, &rt->rt_gw6);
444         } else {
445 		__be32 pkey;
446 
447 		pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
448 		n = ip_neigh_gw4(dev, pkey);
449 	}
450 
451 	if (n && !refcount_inc_not_zero(&n->refcnt))
452 		n = NULL;
453 
454 	rcu_read_unlock_bh();
455 
456 	return n;
457 }
458 
459 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
460 {
461 	const struct rtable *rt = container_of(dst, struct rtable, dst);
462 	struct net_device *dev = dst->dev;
463 	const __be32 *pkey = daddr;
464 
465 	if (rt->rt_gw_family == AF_INET) {
466 		pkey = (const __be32 *)&rt->rt_gw4;
467 	} else if (rt->rt_gw_family == AF_INET6) {
468 		return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
469 	} else if (!daddr ||
470 		 (rt->rt_flags &
471 		  (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
472 		return;
473 	}
474 	__ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
475 }
476 
477 #define IP_IDENTS_SZ 2048u
478 
479 static atomic_t *ip_idents __read_mostly;
480 static u32 *ip_tstamps __read_mostly;
481 
482 /* In order to protect privacy, we add a perturbation to identifiers
483  * if one generator is seldom used. This makes hard for an attacker
484  * to infer how many packets were sent between two points in time.
485  */
486 u32 ip_idents_reserve(u32 hash, int segs)
487 {
488 	u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
489 	atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
490 	u32 old = READ_ONCE(*p_tstamp);
491 	u32 now = (u32)jiffies;
492 	u32 new, delta = 0;
493 
494 	if (old != now && cmpxchg(p_tstamp, old, now) == old)
495 		delta = prandom_u32_max(now - old);
496 
497 	/* Do not use atomic_add_return() as it makes UBSAN unhappy */
498 	do {
499 		old = (u32)atomic_read(p_id);
500 		new = old + delta + segs;
501 	} while (atomic_cmpxchg(p_id, old, new) != old);
502 
503 	return new - segs;
504 }
505 EXPORT_SYMBOL(ip_idents_reserve);
506 
507 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
508 {
509 	u32 hash, id;
510 
511 	/* Note the following code is not safe, but this is okay. */
512 	if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
513 		get_random_bytes(&net->ipv4.ip_id_key,
514 				 sizeof(net->ipv4.ip_id_key));
515 
516 	hash = siphash_3u32((__force u32)iph->daddr,
517 			    (__force u32)iph->saddr,
518 			    iph->protocol,
519 			    &net->ipv4.ip_id_key);
520 	id = ip_idents_reserve(hash, segs);
521 	iph->id = htons(id);
522 }
523 EXPORT_SYMBOL(__ip_select_ident);
524 
525 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
526 			     const struct sock *sk,
527 			     const struct iphdr *iph,
528 			     int oif, u8 tos,
529 			     u8 prot, u32 mark, int flow_flags)
530 {
531 	if (sk) {
532 		const struct inet_sock *inet = inet_sk(sk);
533 
534 		oif = sk->sk_bound_dev_if;
535 		mark = sk->sk_mark;
536 		tos = RT_CONN_FLAGS(sk);
537 		prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
538 	}
539 	flowi4_init_output(fl4, oif, mark, tos,
540 			   RT_SCOPE_UNIVERSE, prot,
541 			   flow_flags,
542 			   iph->daddr, iph->saddr, 0, 0,
543 			   sock_net_uid(net, sk));
544 }
545 
546 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
547 			       const struct sock *sk)
548 {
549 	const struct net *net = dev_net(skb->dev);
550 	const struct iphdr *iph = ip_hdr(skb);
551 	int oif = skb->dev->ifindex;
552 	u8 tos = RT_TOS(iph->tos);
553 	u8 prot = iph->protocol;
554 	u32 mark = skb->mark;
555 
556 	__build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
557 }
558 
559 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
560 {
561 	const struct inet_sock *inet = inet_sk(sk);
562 	const struct ip_options_rcu *inet_opt;
563 	__be32 daddr = inet->inet_daddr;
564 
565 	rcu_read_lock();
566 	inet_opt = rcu_dereference(inet->inet_opt);
567 	if (inet_opt && inet_opt->opt.srr)
568 		daddr = inet_opt->opt.faddr;
569 	flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
570 			   RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
571 			   inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
572 			   inet_sk_flowi_flags(sk),
573 			   daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
574 	rcu_read_unlock();
575 }
576 
577 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
578 				 const struct sk_buff *skb)
579 {
580 	if (skb)
581 		build_skb_flow_key(fl4, skb, sk);
582 	else
583 		build_sk_flow_key(fl4, sk);
584 }
585 
586 static DEFINE_SPINLOCK(fnhe_lock);
587 
588 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
589 {
590 	struct rtable *rt;
591 
592 	rt = rcu_dereference(fnhe->fnhe_rth_input);
593 	if (rt) {
594 		RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
595 		dst_dev_put(&rt->dst);
596 		dst_release(&rt->dst);
597 	}
598 	rt = rcu_dereference(fnhe->fnhe_rth_output);
599 	if (rt) {
600 		RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
601 		dst_dev_put(&rt->dst);
602 		dst_release(&rt->dst);
603 	}
604 }
605 
606 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
607 {
608 	struct fib_nh_exception *fnhe, *oldest;
609 
610 	oldest = rcu_dereference(hash->chain);
611 	for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
612 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
613 		if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
614 			oldest = fnhe;
615 	}
616 	fnhe_flush_routes(oldest);
617 	return oldest;
618 }
619 
620 static inline u32 fnhe_hashfun(__be32 daddr)
621 {
622 	static u32 fnhe_hashrnd __read_mostly;
623 	u32 hval;
624 
625 	net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
626 	hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
627 	return hash_32(hval, FNHE_HASH_SHIFT);
628 }
629 
630 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
631 {
632 	rt->rt_pmtu = fnhe->fnhe_pmtu;
633 	rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
634 	rt->dst.expires = fnhe->fnhe_expires;
635 
636 	if (fnhe->fnhe_gw) {
637 		rt->rt_flags |= RTCF_REDIRECTED;
638 		rt->rt_gw_family = AF_INET;
639 		rt->rt_gw4 = fnhe->fnhe_gw;
640 	}
641 }
642 
643 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
644 				  __be32 gw, u32 pmtu, bool lock,
645 				  unsigned long expires)
646 {
647 	struct fnhe_hash_bucket *hash;
648 	struct fib_nh_exception *fnhe;
649 	struct rtable *rt;
650 	u32 genid, hval;
651 	unsigned int i;
652 	int depth;
653 
654 	genid = fnhe_genid(dev_net(nhc->nhc_dev));
655 	hval = fnhe_hashfun(daddr);
656 
657 	spin_lock_bh(&fnhe_lock);
658 
659 	hash = rcu_dereference(nhc->nhc_exceptions);
660 	if (!hash) {
661 		hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
662 		if (!hash)
663 			goto out_unlock;
664 		rcu_assign_pointer(nhc->nhc_exceptions, hash);
665 	}
666 
667 	hash += hval;
668 
669 	depth = 0;
670 	for (fnhe = rcu_dereference(hash->chain); fnhe;
671 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
672 		if (fnhe->fnhe_daddr == daddr)
673 			break;
674 		depth++;
675 	}
676 
677 	if (fnhe) {
678 		if (fnhe->fnhe_genid != genid)
679 			fnhe->fnhe_genid = genid;
680 		if (gw)
681 			fnhe->fnhe_gw = gw;
682 		if (pmtu) {
683 			fnhe->fnhe_pmtu = pmtu;
684 			fnhe->fnhe_mtu_locked = lock;
685 		}
686 		fnhe->fnhe_expires = max(1UL, expires);
687 		/* Update all cached dsts too */
688 		rt = rcu_dereference(fnhe->fnhe_rth_input);
689 		if (rt)
690 			fill_route_from_fnhe(rt, fnhe);
691 		rt = rcu_dereference(fnhe->fnhe_rth_output);
692 		if (rt)
693 			fill_route_from_fnhe(rt, fnhe);
694 	} else {
695 		if (depth > FNHE_RECLAIM_DEPTH)
696 			fnhe = fnhe_oldest(hash);
697 		else {
698 			fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
699 			if (!fnhe)
700 				goto out_unlock;
701 
702 			fnhe->fnhe_next = hash->chain;
703 			rcu_assign_pointer(hash->chain, fnhe);
704 		}
705 		fnhe->fnhe_genid = genid;
706 		fnhe->fnhe_daddr = daddr;
707 		fnhe->fnhe_gw = gw;
708 		fnhe->fnhe_pmtu = pmtu;
709 		fnhe->fnhe_mtu_locked = lock;
710 		fnhe->fnhe_expires = max(1UL, expires);
711 
712 		/* Exception created; mark the cached routes for the nexthop
713 		 * stale, so anyone caching it rechecks if this exception
714 		 * applies to them.
715 		 */
716 		rt = rcu_dereference(nhc->nhc_rth_input);
717 		if (rt)
718 			rt->dst.obsolete = DST_OBSOLETE_KILL;
719 
720 		for_each_possible_cpu(i) {
721 			struct rtable __rcu **prt;
722 			prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
723 			rt = rcu_dereference(*prt);
724 			if (rt)
725 				rt->dst.obsolete = DST_OBSOLETE_KILL;
726 		}
727 	}
728 
729 	fnhe->fnhe_stamp = jiffies;
730 
731 out_unlock:
732 	spin_unlock_bh(&fnhe_lock);
733 }
734 
735 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
736 			     bool kill_route)
737 {
738 	__be32 new_gw = icmp_hdr(skb)->un.gateway;
739 	__be32 old_gw = ip_hdr(skb)->saddr;
740 	struct net_device *dev = skb->dev;
741 	struct in_device *in_dev;
742 	struct fib_result res;
743 	struct neighbour *n;
744 	struct net *net;
745 
746 	switch (icmp_hdr(skb)->code & 7) {
747 	case ICMP_REDIR_NET:
748 	case ICMP_REDIR_NETTOS:
749 	case ICMP_REDIR_HOST:
750 	case ICMP_REDIR_HOSTTOS:
751 		break;
752 
753 	default:
754 		return;
755 	}
756 
757 	if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
758 		return;
759 
760 	in_dev = __in_dev_get_rcu(dev);
761 	if (!in_dev)
762 		return;
763 
764 	net = dev_net(dev);
765 	if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
766 	    ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
767 	    ipv4_is_zeronet(new_gw))
768 		goto reject_redirect;
769 
770 	if (!IN_DEV_SHARED_MEDIA(in_dev)) {
771 		if (!inet_addr_onlink(in_dev, new_gw, old_gw))
772 			goto reject_redirect;
773 		if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
774 			goto reject_redirect;
775 	} else {
776 		if (inet_addr_type(net, new_gw) != RTN_UNICAST)
777 			goto reject_redirect;
778 	}
779 
780 	n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
781 	if (!n)
782 		n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
783 	if (!IS_ERR(n)) {
784 		if (!(n->nud_state & NUD_VALID)) {
785 			neigh_event_send(n, NULL);
786 		} else {
787 			if (fib_lookup(net, fl4, &res, 0) == 0) {
788 				struct fib_nh_common *nhc = FIB_RES_NHC(res);
789 
790 				update_or_create_fnhe(nhc, fl4->daddr, new_gw,
791 						0, false,
792 						jiffies + ip_rt_gc_timeout);
793 			}
794 			if (kill_route)
795 				rt->dst.obsolete = DST_OBSOLETE_KILL;
796 			call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
797 		}
798 		neigh_release(n);
799 	}
800 	return;
801 
802 reject_redirect:
803 #ifdef CONFIG_IP_ROUTE_VERBOSE
804 	if (IN_DEV_LOG_MARTIANS(in_dev)) {
805 		const struct iphdr *iph = (const struct iphdr *) skb->data;
806 		__be32 daddr = iph->daddr;
807 		__be32 saddr = iph->saddr;
808 
809 		net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
810 				     "  Advised path = %pI4 -> %pI4\n",
811 				     &old_gw, dev->name, &new_gw,
812 				     &saddr, &daddr);
813 	}
814 #endif
815 	;
816 }
817 
818 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
819 {
820 	struct rtable *rt;
821 	struct flowi4 fl4;
822 	const struct iphdr *iph = (const struct iphdr *) skb->data;
823 	struct net *net = dev_net(skb->dev);
824 	int oif = skb->dev->ifindex;
825 	u8 tos = RT_TOS(iph->tos);
826 	u8 prot = iph->protocol;
827 	u32 mark = skb->mark;
828 
829 	rt = (struct rtable *) dst;
830 
831 	__build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
832 	__ip_do_redirect(rt, skb, &fl4, true);
833 }
834 
835 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
836 {
837 	struct rtable *rt = (struct rtable *)dst;
838 	struct dst_entry *ret = dst;
839 
840 	if (rt) {
841 		if (dst->obsolete > 0) {
842 			ip_rt_put(rt);
843 			ret = NULL;
844 		} else if ((rt->rt_flags & RTCF_REDIRECTED) ||
845 			   rt->dst.expires) {
846 			ip_rt_put(rt);
847 			ret = NULL;
848 		}
849 	}
850 	return ret;
851 }
852 
853 /*
854  * Algorithm:
855  *	1. The first ip_rt_redirect_number redirects are sent
856  *	   with exponential backoff, then we stop sending them at all,
857  *	   assuming that the host ignores our redirects.
858  *	2. If we did not see packets requiring redirects
859  *	   during ip_rt_redirect_silence, we assume that the host
860  *	   forgot redirected route and start to send redirects again.
861  *
862  * This algorithm is much cheaper and more intelligent than dumb load limiting
863  * in icmp.c.
864  *
865  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
866  * and "frag. need" (breaks PMTU discovery) in icmp.c.
867  */
868 
869 void ip_rt_send_redirect(struct sk_buff *skb)
870 {
871 	struct rtable *rt = skb_rtable(skb);
872 	struct in_device *in_dev;
873 	struct inet_peer *peer;
874 	struct net *net;
875 	int log_martians;
876 	int vif;
877 
878 	rcu_read_lock();
879 	in_dev = __in_dev_get_rcu(rt->dst.dev);
880 	if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
881 		rcu_read_unlock();
882 		return;
883 	}
884 	log_martians = IN_DEV_LOG_MARTIANS(in_dev);
885 	vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
886 	rcu_read_unlock();
887 
888 	net = dev_net(rt->dst.dev);
889 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
890 	if (!peer) {
891 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
892 			  rt_nexthop(rt, ip_hdr(skb)->daddr));
893 		return;
894 	}
895 
896 	/* No redirected packets during ip_rt_redirect_silence;
897 	 * reset the algorithm.
898 	 */
899 	if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
900 		peer->rate_tokens = 0;
901 		peer->n_redirects = 0;
902 	}
903 
904 	/* Too many ignored redirects; do not send anything
905 	 * set dst.rate_last to the last seen redirected packet.
906 	 */
907 	if (peer->n_redirects >= ip_rt_redirect_number) {
908 		peer->rate_last = jiffies;
909 		goto out_put_peer;
910 	}
911 
912 	/* Check for load limit; set rate_last to the latest sent
913 	 * redirect.
914 	 */
915 	if (peer->rate_tokens == 0 ||
916 	    time_after(jiffies,
917 		       (peer->rate_last +
918 			(ip_rt_redirect_load << peer->rate_tokens)))) {
919 		__be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
920 
921 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
922 		peer->rate_last = jiffies;
923 		++peer->rate_tokens;
924 		++peer->n_redirects;
925 #ifdef CONFIG_IP_ROUTE_VERBOSE
926 		if (log_martians &&
927 		    peer->rate_tokens == ip_rt_redirect_number)
928 			net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
929 					     &ip_hdr(skb)->saddr, inet_iif(skb),
930 					     &ip_hdr(skb)->daddr, &gw);
931 #endif
932 	}
933 out_put_peer:
934 	inet_putpeer(peer);
935 }
936 
937 static int ip_error(struct sk_buff *skb)
938 {
939 	struct rtable *rt = skb_rtable(skb);
940 	struct net_device *dev = skb->dev;
941 	struct in_device *in_dev;
942 	struct inet_peer *peer;
943 	unsigned long now;
944 	struct net *net;
945 	bool send;
946 	int code;
947 
948 	if (netif_is_l3_master(skb->dev)) {
949 		dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
950 		if (!dev)
951 			goto out;
952 	}
953 
954 	in_dev = __in_dev_get_rcu(dev);
955 
956 	/* IP on this device is disabled. */
957 	if (!in_dev)
958 		goto out;
959 
960 	net = dev_net(rt->dst.dev);
961 	if (!IN_DEV_FORWARD(in_dev)) {
962 		switch (rt->dst.error) {
963 		case EHOSTUNREACH:
964 			__IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
965 			break;
966 
967 		case ENETUNREACH:
968 			__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
969 			break;
970 		}
971 		goto out;
972 	}
973 
974 	switch (rt->dst.error) {
975 	case EINVAL:
976 	default:
977 		goto out;
978 	case EHOSTUNREACH:
979 		code = ICMP_HOST_UNREACH;
980 		break;
981 	case ENETUNREACH:
982 		code = ICMP_NET_UNREACH;
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(skb);
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 	u32 old_mtu = ipv4_mtu(dst);
1017 	struct fib_result res;
1018 	bool lock = false;
1019 
1020 	if (ip_mtu_locked(dst))
1021 		return;
1022 
1023 	if (old_mtu < mtu)
1024 		return;
1025 
1026 	if (mtu < ip_rt_min_pmtu) {
1027 		lock = true;
1028 		mtu = min(old_mtu, ip_rt_min_pmtu);
1029 	}
1030 
1031 	if (rt->rt_pmtu == mtu && !lock &&
1032 	    time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1033 		return;
1034 
1035 	rcu_read_lock();
1036 	if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
1037 		struct fib_nh_common *nhc = FIB_RES_NHC(res);
1038 
1039 		update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1040 				      jiffies + ip_rt_mtu_expires);
1041 	}
1042 	rcu_read_unlock();
1043 }
1044 
1045 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1046 			      struct sk_buff *skb, u32 mtu)
1047 {
1048 	struct rtable *rt = (struct rtable *) dst;
1049 	struct flowi4 fl4;
1050 
1051 	ip_rt_build_flow_key(&fl4, sk, skb);
1052 	__ip_rt_update_pmtu(rt, &fl4, mtu);
1053 }
1054 
1055 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1056 		      int oif, u8 protocol)
1057 {
1058 	const struct iphdr *iph = (const struct iphdr *) skb->data;
1059 	struct flowi4 fl4;
1060 	struct rtable *rt;
1061 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
1062 
1063 	__build_flow_key(net, &fl4, NULL, iph, oif,
1064 			 RT_TOS(iph->tos), protocol, mark, 0);
1065 	rt = __ip_route_output_key(net, &fl4);
1066 	if (!IS_ERR(rt)) {
1067 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1068 		ip_rt_put(rt);
1069 	}
1070 }
1071 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1072 
1073 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1074 {
1075 	const struct iphdr *iph = (const struct iphdr *) skb->data;
1076 	struct flowi4 fl4;
1077 	struct rtable *rt;
1078 
1079 	__build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1080 
1081 	if (!fl4.flowi4_mark)
1082 		fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1083 
1084 	rt = __ip_route_output_key(sock_net(sk), &fl4);
1085 	if (!IS_ERR(rt)) {
1086 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1087 		ip_rt_put(rt);
1088 	}
1089 }
1090 
1091 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1092 {
1093 	const struct iphdr *iph = (const struct iphdr *) skb->data;
1094 	struct flowi4 fl4;
1095 	struct rtable *rt;
1096 	struct dst_entry *odst = NULL;
1097 	bool new = false;
1098 	struct net *net = sock_net(sk);
1099 
1100 	bh_lock_sock(sk);
1101 
1102 	if (!ip_sk_accept_pmtu(sk))
1103 		goto out;
1104 
1105 	odst = sk_dst_get(sk);
1106 
1107 	if (sock_owned_by_user(sk) || !odst) {
1108 		__ipv4_sk_update_pmtu(skb, sk, mtu);
1109 		goto out;
1110 	}
1111 
1112 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1113 
1114 	rt = (struct rtable *)odst;
1115 	if (odst->obsolete && !odst->ops->check(odst, 0)) {
1116 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1117 		if (IS_ERR(rt))
1118 			goto out;
1119 
1120 		new = true;
1121 	}
1122 
1123 	__ip_rt_update_pmtu((struct rtable *) xfrm_dst_path(&rt->dst), &fl4, mtu);
1124 
1125 	if (!dst_check(&rt->dst, 0)) {
1126 		if (new)
1127 			dst_release(&rt->dst);
1128 
1129 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1130 		if (IS_ERR(rt))
1131 			goto out;
1132 
1133 		new = true;
1134 	}
1135 
1136 	if (new)
1137 		sk_dst_set(sk, &rt->dst);
1138 
1139 out:
1140 	bh_unlock_sock(sk);
1141 	dst_release(odst);
1142 }
1143 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1144 
1145 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1146 		   int oif, u8 protocol)
1147 {
1148 	const struct iphdr *iph = (const struct iphdr *) skb->data;
1149 	struct flowi4 fl4;
1150 	struct rtable *rt;
1151 
1152 	__build_flow_key(net, &fl4, NULL, iph, oif,
1153 			 RT_TOS(iph->tos), protocol, 0, 0);
1154 	rt = __ip_route_output_key(net, &fl4);
1155 	if (!IS_ERR(rt)) {
1156 		__ip_do_redirect(rt, skb, &fl4, false);
1157 		ip_rt_put(rt);
1158 	}
1159 }
1160 EXPORT_SYMBOL_GPL(ipv4_redirect);
1161 
1162 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1163 {
1164 	const struct iphdr *iph = (const struct iphdr *) skb->data;
1165 	struct flowi4 fl4;
1166 	struct rtable *rt;
1167 	struct net *net = sock_net(sk);
1168 
1169 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1170 	rt = __ip_route_output_key(net, &fl4);
1171 	if (!IS_ERR(rt)) {
1172 		__ip_do_redirect(rt, skb, &fl4, false);
1173 		ip_rt_put(rt);
1174 	}
1175 }
1176 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1177 
1178 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1179 {
1180 	struct rtable *rt = (struct rtable *) dst;
1181 
1182 	/* All IPV4 dsts are created with ->obsolete set to the value
1183 	 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1184 	 * into this function always.
1185 	 *
1186 	 * When a PMTU/redirect information update invalidates a route,
1187 	 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1188 	 * DST_OBSOLETE_DEAD.
1189 	 */
1190 	if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1191 		return NULL;
1192 	return dst;
1193 }
1194 
1195 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1196 {
1197 	struct ip_options opt;
1198 	int res;
1199 
1200 	/* Recompile ip options since IPCB may not be valid anymore.
1201 	 * Also check we have a reasonable ipv4 header.
1202 	 */
1203 	if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1204 	    ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1205 		return;
1206 
1207 	memset(&opt, 0, sizeof(opt));
1208 	if (ip_hdr(skb)->ihl > 5) {
1209 		if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1210 			return;
1211 		opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1212 
1213 		rcu_read_lock();
1214 		res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1215 		rcu_read_unlock();
1216 
1217 		if (res)
1218 			return;
1219 	}
1220 	__icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1221 }
1222 
1223 static void ipv4_link_failure(struct sk_buff *skb)
1224 {
1225 	struct rtable *rt;
1226 
1227 	ipv4_send_dest_unreach(skb);
1228 
1229 	rt = skb_rtable(skb);
1230 	if (rt)
1231 		dst_set_expires(&rt->dst, 0);
1232 }
1233 
1234 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1235 {
1236 	pr_debug("%s: %pI4 -> %pI4, %s\n",
1237 		 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1238 		 skb->dev ? skb->dev->name : "?");
1239 	kfree_skb(skb);
1240 	WARN_ON(1);
1241 	return 0;
1242 }
1243 
1244 /*
1245    We do not cache source address of outgoing interface,
1246    because it is used only by IP RR, TS and SRR options,
1247    so that it out of fast path.
1248 
1249    BTW remember: "addr" is allowed to be not aligned
1250    in IP options!
1251  */
1252 
1253 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1254 {
1255 	__be32 src;
1256 
1257 	if (rt_is_output_route(rt))
1258 		src = ip_hdr(skb)->saddr;
1259 	else {
1260 		struct fib_result res;
1261 		struct iphdr *iph = ip_hdr(skb);
1262 		struct flowi4 fl4 = {
1263 			.daddr = iph->daddr,
1264 			.saddr = iph->saddr,
1265 			.flowi4_tos = RT_TOS(iph->tos),
1266 			.flowi4_oif = rt->dst.dev->ifindex,
1267 			.flowi4_iif = skb->dev->ifindex,
1268 			.flowi4_mark = skb->mark,
1269 		};
1270 
1271 		rcu_read_lock();
1272 		if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1273 			src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1274 		else
1275 			src = inet_select_addr(rt->dst.dev,
1276 					       rt_nexthop(rt, iph->daddr),
1277 					       RT_SCOPE_UNIVERSE);
1278 		rcu_read_unlock();
1279 	}
1280 	memcpy(addr, &src, 4);
1281 }
1282 
1283 #ifdef CONFIG_IP_ROUTE_CLASSID
1284 static void set_class_tag(struct rtable *rt, u32 tag)
1285 {
1286 	if (!(rt->dst.tclassid & 0xFFFF))
1287 		rt->dst.tclassid |= tag & 0xFFFF;
1288 	if (!(rt->dst.tclassid & 0xFFFF0000))
1289 		rt->dst.tclassid |= tag & 0xFFFF0000;
1290 }
1291 #endif
1292 
1293 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1294 {
1295 	unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1296 	unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1297 				    ip_rt_min_advmss);
1298 
1299 	return min(advmss, IPV4_MAX_PMTU - header_size);
1300 }
1301 
1302 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1303 {
1304 	const struct rtable *rt = (const struct rtable *) dst;
1305 	unsigned int mtu = rt->rt_pmtu;
1306 
1307 	if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1308 		mtu = dst_metric_raw(dst, RTAX_MTU);
1309 
1310 	if (mtu)
1311 		return mtu;
1312 
1313 	mtu = READ_ONCE(dst->dev->mtu);
1314 
1315 	if (unlikely(ip_mtu_locked(dst))) {
1316 		if (rt->rt_gw_family && mtu > 576)
1317 			mtu = 576;
1318 	}
1319 
1320 	mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1321 
1322 	return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1323 }
1324 
1325 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1326 {
1327 	struct fnhe_hash_bucket *hash;
1328 	struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1329 	u32 hval = fnhe_hashfun(daddr);
1330 
1331 	spin_lock_bh(&fnhe_lock);
1332 
1333 	hash = rcu_dereference_protected(nhc->nhc_exceptions,
1334 					 lockdep_is_held(&fnhe_lock));
1335 	hash += hval;
1336 
1337 	fnhe_p = &hash->chain;
1338 	fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1339 	while (fnhe) {
1340 		if (fnhe->fnhe_daddr == daddr) {
1341 			rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1342 				fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1343 			/* set fnhe_daddr to 0 to ensure it won't bind with
1344 			 * new dsts in rt_bind_exception().
1345 			 */
1346 			fnhe->fnhe_daddr = 0;
1347 			fnhe_flush_routes(fnhe);
1348 			kfree_rcu(fnhe, rcu);
1349 			break;
1350 		}
1351 		fnhe_p = &fnhe->fnhe_next;
1352 		fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1353 						 lockdep_is_held(&fnhe_lock));
1354 	}
1355 
1356 	spin_unlock_bh(&fnhe_lock);
1357 }
1358 
1359 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1360 					       __be32 daddr)
1361 {
1362 	struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1363 	struct fib_nh_exception *fnhe;
1364 	u32 hval;
1365 
1366 	if (!hash)
1367 		return NULL;
1368 
1369 	hval = fnhe_hashfun(daddr);
1370 
1371 	for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1372 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
1373 		if (fnhe->fnhe_daddr == daddr) {
1374 			if (fnhe->fnhe_expires &&
1375 			    time_after(jiffies, fnhe->fnhe_expires)) {
1376 				ip_del_fnhe(nhc, daddr);
1377 				break;
1378 			}
1379 			return fnhe;
1380 		}
1381 	}
1382 	return NULL;
1383 }
1384 
1385 /* MTU selection:
1386  * 1. mtu on route is locked - use it
1387  * 2. mtu from nexthop exception
1388  * 3. mtu from egress device
1389  */
1390 
1391 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1392 {
1393 	struct fib_nh_common *nhc = res->nhc;
1394 	struct net_device *dev = nhc->nhc_dev;
1395 	struct fib_info *fi = res->fi;
1396 	u32 mtu = 0;
1397 
1398 	if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu ||
1399 	    fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1400 		mtu = fi->fib_mtu;
1401 
1402 	if (likely(!mtu)) {
1403 		struct fib_nh_exception *fnhe;
1404 
1405 		fnhe = find_exception(nhc, daddr);
1406 		if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1407 			mtu = fnhe->fnhe_pmtu;
1408 	}
1409 
1410 	if (likely(!mtu))
1411 		mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1412 
1413 	return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1414 }
1415 
1416 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1417 			      __be32 daddr, const bool do_cache)
1418 {
1419 	bool ret = false;
1420 
1421 	spin_lock_bh(&fnhe_lock);
1422 
1423 	if (daddr == fnhe->fnhe_daddr) {
1424 		struct rtable __rcu **porig;
1425 		struct rtable *orig;
1426 		int genid = fnhe_genid(dev_net(rt->dst.dev));
1427 
1428 		if (rt_is_input_route(rt))
1429 			porig = &fnhe->fnhe_rth_input;
1430 		else
1431 			porig = &fnhe->fnhe_rth_output;
1432 		orig = rcu_dereference(*porig);
1433 
1434 		if (fnhe->fnhe_genid != genid) {
1435 			fnhe->fnhe_genid = genid;
1436 			fnhe->fnhe_gw = 0;
1437 			fnhe->fnhe_pmtu = 0;
1438 			fnhe->fnhe_expires = 0;
1439 			fnhe->fnhe_mtu_locked = false;
1440 			fnhe_flush_routes(fnhe);
1441 			orig = NULL;
1442 		}
1443 		fill_route_from_fnhe(rt, fnhe);
1444 		if (!rt->rt_gw4) {
1445 			rt->rt_gw4 = daddr;
1446 			rt->rt_gw_family = AF_INET;
1447 		}
1448 
1449 		if (do_cache) {
1450 			dst_hold(&rt->dst);
1451 			rcu_assign_pointer(*porig, rt);
1452 			if (orig) {
1453 				dst_dev_put(&orig->dst);
1454 				dst_release(&orig->dst);
1455 			}
1456 			ret = true;
1457 		}
1458 
1459 		fnhe->fnhe_stamp = jiffies;
1460 	}
1461 	spin_unlock_bh(&fnhe_lock);
1462 
1463 	return ret;
1464 }
1465 
1466 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1467 {
1468 	struct rtable *orig, *prev, **p;
1469 	bool ret = true;
1470 
1471 	if (rt_is_input_route(rt)) {
1472 		p = (struct rtable **)&nhc->nhc_rth_input;
1473 	} else {
1474 		p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1475 	}
1476 	orig = *p;
1477 
1478 	/* hold dst before doing cmpxchg() to avoid race condition
1479 	 * on this dst
1480 	 */
1481 	dst_hold(&rt->dst);
1482 	prev = cmpxchg(p, orig, rt);
1483 	if (prev == orig) {
1484 		if (orig) {
1485 			dst_dev_put(&orig->dst);
1486 			dst_release(&orig->dst);
1487 		}
1488 	} else {
1489 		dst_release(&rt->dst);
1490 		ret = false;
1491 	}
1492 
1493 	return ret;
1494 }
1495 
1496 struct uncached_list {
1497 	spinlock_t		lock;
1498 	struct list_head	head;
1499 };
1500 
1501 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1502 
1503 void rt_add_uncached_list(struct rtable *rt)
1504 {
1505 	struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1506 
1507 	rt->rt_uncached_list = ul;
1508 
1509 	spin_lock_bh(&ul->lock);
1510 	list_add_tail(&rt->rt_uncached, &ul->head);
1511 	spin_unlock_bh(&ul->lock);
1512 }
1513 
1514 void rt_del_uncached_list(struct rtable *rt)
1515 {
1516 	if (!list_empty(&rt->rt_uncached)) {
1517 		struct uncached_list *ul = rt->rt_uncached_list;
1518 
1519 		spin_lock_bh(&ul->lock);
1520 		list_del(&rt->rt_uncached);
1521 		spin_unlock_bh(&ul->lock);
1522 	}
1523 }
1524 
1525 static void ipv4_dst_destroy(struct dst_entry *dst)
1526 {
1527 	struct rtable *rt = (struct rtable *)dst;
1528 
1529 	ip_dst_metrics_put(dst);
1530 	rt_del_uncached_list(rt);
1531 }
1532 
1533 void rt_flush_dev(struct net_device *dev)
1534 {
1535 	struct net *net = dev_net(dev);
1536 	struct rtable *rt;
1537 	int cpu;
1538 
1539 	for_each_possible_cpu(cpu) {
1540 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1541 
1542 		spin_lock_bh(&ul->lock);
1543 		list_for_each_entry(rt, &ul->head, rt_uncached) {
1544 			if (rt->dst.dev != dev)
1545 				continue;
1546 			rt->dst.dev = net->loopback_dev;
1547 			dev_hold(rt->dst.dev);
1548 			dev_put(dev);
1549 		}
1550 		spin_unlock_bh(&ul->lock);
1551 	}
1552 }
1553 
1554 static bool rt_cache_valid(const struct rtable *rt)
1555 {
1556 	return	rt &&
1557 		rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1558 		!rt_is_expired(rt);
1559 }
1560 
1561 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1562 			   const struct fib_result *res,
1563 			   struct fib_nh_exception *fnhe,
1564 			   struct fib_info *fi, u16 type, u32 itag,
1565 			   const bool do_cache)
1566 {
1567 	bool cached = false;
1568 
1569 	if (fi) {
1570 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1571 
1572 		if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1573 			rt->rt_gw_family = nhc->nhc_gw_family;
1574 			/* only INET and INET6 are supported */
1575 			if (likely(nhc->nhc_gw_family == AF_INET))
1576 				rt->rt_gw4 = nhc->nhc_gw.ipv4;
1577 			else
1578 				rt->rt_gw6 = nhc->nhc_gw.ipv6;
1579 		}
1580 
1581 		ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1582 
1583 #ifdef CONFIG_IP_ROUTE_CLASSID
1584 		if (nhc->nhc_family == AF_INET) {
1585 			struct fib_nh *nh;
1586 
1587 			nh = container_of(nhc, struct fib_nh, nh_common);
1588 			rt->dst.tclassid = nh->nh_tclassid;
1589 		}
1590 #endif
1591 		rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1592 		if (unlikely(fnhe))
1593 			cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1594 		else if (do_cache)
1595 			cached = rt_cache_route(nhc, rt);
1596 		if (unlikely(!cached)) {
1597 			/* Routes we intend to cache in nexthop exception or
1598 			 * FIB nexthop have the DST_NOCACHE bit clear.
1599 			 * However, if we are unsuccessful at storing this
1600 			 * route into the cache we really need to set it.
1601 			 */
1602 			if (!rt->rt_gw4) {
1603 				rt->rt_gw_family = AF_INET;
1604 				rt->rt_gw4 = daddr;
1605 			}
1606 			rt_add_uncached_list(rt);
1607 		}
1608 	} else
1609 		rt_add_uncached_list(rt);
1610 
1611 #ifdef CONFIG_IP_ROUTE_CLASSID
1612 #ifdef CONFIG_IP_MULTIPLE_TABLES
1613 	set_class_tag(rt, res->tclassid);
1614 #endif
1615 	set_class_tag(rt, itag);
1616 #endif
1617 }
1618 
1619 struct rtable *rt_dst_alloc(struct net_device *dev,
1620 			    unsigned int flags, u16 type,
1621 			    bool nopolicy, bool noxfrm, bool will_cache)
1622 {
1623 	struct rtable *rt;
1624 
1625 	rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1626 		       (will_cache ? 0 : DST_HOST) |
1627 		       (nopolicy ? DST_NOPOLICY : 0) |
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_gw_family = 0;
1639 		rt->rt_gw4 = 0;
1640 		INIT_LIST_HEAD(&rt->rt_uncached);
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 /* called in rcu_read_lock() section */
1652 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1653 			  u8 tos, struct net_device *dev,
1654 			  struct in_device *in_dev, u32 *itag)
1655 {
1656 	int err;
1657 
1658 	/* Primary sanity checks. */
1659 	if (!in_dev)
1660 		return -EINVAL;
1661 
1662 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1663 	    skb->protocol != htons(ETH_P_IP))
1664 		return -EINVAL;
1665 
1666 	if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1667 		return -EINVAL;
1668 
1669 	if (ipv4_is_zeronet(saddr)) {
1670 		if (!ipv4_is_local_multicast(daddr) &&
1671 		    ip_hdr(skb)->protocol != IPPROTO_IGMP)
1672 			return -EINVAL;
1673 	} else {
1674 		err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1675 					  in_dev, itag);
1676 		if (err < 0)
1677 			return err;
1678 	}
1679 	return 0;
1680 }
1681 
1682 /* called in rcu_read_lock() section */
1683 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1684 			     u8 tos, struct net_device *dev, int our)
1685 {
1686 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1687 	unsigned int flags = RTCF_MULTICAST;
1688 	struct rtable *rth;
1689 	u32 itag = 0;
1690 	int err;
1691 
1692 	err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1693 	if (err)
1694 		return err;
1695 
1696 	if (our)
1697 		flags |= RTCF_LOCAL;
1698 
1699 	rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1700 			   IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1701 	if (!rth)
1702 		return -ENOBUFS;
1703 
1704 #ifdef CONFIG_IP_ROUTE_CLASSID
1705 	rth->dst.tclassid = itag;
1706 #endif
1707 	rth->dst.output = ip_rt_bug;
1708 	rth->rt_is_input= 1;
1709 
1710 #ifdef CONFIG_IP_MROUTE
1711 	if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1712 		rth->dst.input = ip_mr_input;
1713 #endif
1714 	RT_CACHE_STAT_INC(in_slow_mc);
1715 
1716 	skb_dst_set(skb, &rth->dst);
1717 	return 0;
1718 }
1719 
1720 
1721 static void ip_handle_martian_source(struct net_device *dev,
1722 				     struct in_device *in_dev,
1723 				     struct sk_buff *skb,
1724 				     __be32 daddr,
1725 				     __be32 saddr)
1726 {
1727 	RT_CACHE_STAT_INC(in_martian_src);
1728 #ifdef CONFIG_IP_ROUTE_VERBOSE
1729 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1730 		/*
1731 		 *	RFC1812 recommendation, if source is martian,
1732 		 *	the only hint is MAC header.
1733 		 */
1734 		pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1735 			&daddr, &saddr, dev->name);
1736 		if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1737 			print_hex_dump(KERN_WARNING, "ll header: ",
1738 				       DUMP_PREFIX_OFFSET, 16, 1,
1739 				       skb_mac_header(skb),
1740 				       dev->hard_header_len, false);
1741 		}
1742 	}
1743 #endif
1744 }
1745 
1746 /* called in rcu_read_lock() section */
1747 static int __mkroute_input(struct sk_buff *skb,
1748 			   const struct fib_result *res,
1749 			   struct in_device *in_dev,
1750 			   __be32 daddr, __be32 saddr, u32 tos)
1751 {
1752 	struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1753 	struct net_device *dev = nhc->nhc_dev;
1754 	struct fib_nh_exception *fnhe;
1755 	struct rtable *rth;
1756 	int err;
1757 	struct in_device *out_dev;
1758 	bool do_cache;
1759 	u32 itag = 0;
1760 
1761 	/* get a working reference to the output device */
1762 	out_dev = __in_dev_get_rcu(dev);
1763 	if (!out_dev) {
1764 		net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1765 		return -EINVAL;
1766 	}
1767 
1768 	err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1769 				  in_dev->dev, in_dev, &itag);
1770 	if (err < 0) {
1771 		ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1772 					 saddr);
1773 
1774 		goto cleanup;
1775 	}
1776 
1777 	do_cache = res->fi && !itag;
1778 	if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1779 	    skb->protocol == htons(ETH_P_IP)) {
1780 		__be32 gw;
1781 
1782 		gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1783 		if (IN_DEV_SHARED_MEDIA(out_dev) ||
1784 		    inet_addr_onlink(out_dev, saddr, gw))
1785 			IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1786 	}
1787 
1788 	if (skb->protocol != htons(ETH_P_IP)) {
1789 		/* Not IP (i.e. ARP). Do not create route, if it is
1790 		 * invalid for proxy arp. DNAT routes are always valid.
1791 		 *
1792 		 * Proxy arp feature have been extended to allow, ARP
1793 		 * replies back to the same interface, to support
1794 		 * Private VLAN switch technologies. See arp.c.
1795 		 */
1796 		if (out_dev == in_dev &&
1797 		    IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1798 			err = -EINVAL;
1799 			goto cleanup;
1800 		}
1801 	}
1802 
1803 	fnhe = find_exception(nhc, daddr);
1804 	if (do_cache) {
1805 		if (fnhe)
1806 			rth = rcu_dereference(fnhe->fnhe_rth_input);
1807 		else
1808 			rth = rcu_dereference(nhc->nhc_rth_input);
1809 		if (rt_cache_valid(rth)) {
1810 			skb_dst_set_noref(skb, &rth->dst);
1811 			goto out;
1812 		}
1813 	}
1814 
1815 	rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1816 			   IN_DEV_CONF_GET(in_dev, NOPOLICY),
1817 			   IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1818 	if (!rth) {
1819 		err = -ENOBUFS;
1820 		goto cleanup;
1821 	}
1822 
1823 	rth->rt_is_input = 1;
1824 	RT_CACHE_STAT_INC(in_slow_tot);
1825 
1826 	rth->dst.input = ip_forward;
1827 
1828 	rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1829 		       do_cache);
1830 	lwtunnel_set_redirect(&rth->dst);
1831 	skb_dst_set(skb, &rth->dst);
1832 out:
1833 	err = 0;
1834  cleanup:
1835 	return err;
1836 }
1837 
1838 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1839 /* To make ICMP packets follow the right flow, the multipath hash is
1840  * calculated from the inner IP addresses.
1841  */
1842 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1843 				 struct flow_keys *hash_keys)
1844 {
1845 	const struct iphdr *outer_iph = ip_hdr(skb);
1846 	const struct iphdr *key_iph = outer_iph;
1847 	const struct iphdr *inner_iph;
1848 	const struct icmphdr *icmph;
1849 	struct iphdr _inner_iph;
1850 	struct icmphdr _icmph;
1851 
1852 	if (likely(outer_iph->protocol != IPPROTO_ICMP))
1853 		goto out;
1854 
1855 	if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1856 		goto out;
1857 
1858 	icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1859 				   &_icmph);
1860 	if (!icmph)
1861 		goto out;
1862 
1863 	if (icmph->type != ICMP_DEST_UNREACH &&
1864 	    icmph->type != ICMP_REDIRECT &&
1865 	    icmph->type != ICMP_TIME_EXCEEDED &&
1866 	    icmph->type != ICMP_PARAMETERPROB)
1867 		goto out;
1868 
1869 	inner_iph = skb_header_pointer(skb,
1870 				       outer_iph->ihl * 4 + sizeof(_icmph),
1871 				       sizeof(_inner_iph), &_inner_iph);
1872 	if (!inner_iph)
1873 		goto out;
1874 
1875 	key_iph = inner_iph;
1876 out:
1877 	hash_keys->addrs.v4addrs.src = key_iph->saddr;
1878 	hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1879 }
1880 
1881 /* if skb is set it will be used and fl4 can be NULL */
1882 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
1883 		       const struct sk_buff *skb, struct flow_keys *flkeys)
1884 {
1885 	u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
1886 	struct flow_keys hash_keys;
1887 	u32 mhash;
1888 
1889 	switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1890 	case 0:
1891 		memset(&hash_keys, 0, sizeof(hash_keys));
1892 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1893 		if (skb) {
1894 			ip_multipath_l3_keys(skb, &hash_keys);
1895 		} else {
1896 			hash_keys.addrs.v4addrs.src = fl4->saddr;
1897 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
1898 		}
1899 		break;
1900 	case 1:
1901 		/* skb is currently provided only when forwarding */
1902 		if (skb) {
1903 			unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1904 			struct flow_keys keys;
1905 
1906 			/* short-circuit if we already have L4 hash present */
1907 			if (skb->l4_hash)
1908 				return skb_get_hash_raw(skb) >> 1;
1909 
1910 			memset(&hash_keys, 0, sizeof(hash_keys));
1911 
1912 			if (!flkeys) {
1913 				skb_flow_dissect_flow_keys(skb, &keys, flag);
1914 				flkeys = &keys;
1915 			}
1916 
1917 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1918 			hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
1919 			hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
1920 			hash_keys.ports.src = flkeys->ports.src;
1921 			hash_keys.ports.dst = flkeys->ports.dst;
1922 			hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
1923 		} else {
1924 			memset(&hash_keys, 0, sizeof(hash_keys));
1925 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1926 			hash_keys.addrs.v4addrs.src = fl4->saddr;
1927 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
1928 			hash_keys.ports.src = fl4->fl4_sport;
1929 			hash_keys.ports.dst = fl4->fl4_dport;
1930 			hash_keys.basic.ip_proto = fl4->flowi4_proto;
1931 		}
1932 		break;
1933 	case 2:
1934 		memset(&hash_keys, 0, sizeof(hash_keys));
1935 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1936 		/* skb is currently provided only when forwarding */
1937 		if (skb) {
1938 			struct flow_keys keys;
1939 
1940 			skb_flow_dissect_flow_keys(skb, &keys, 0);
1941 
1942 			hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1943 			hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1944 		} else {
1945 			/* Same as case 0 */
1946 			hash_keys.addrs.v4addrs.src = fl4->saddr;
1947 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
1948 		}
1949 		break;
1950 	}
1951 	mhash = flow_hash_from_keys(&hash_keys);
1952 
1953 	if (multipath_hash)
1954 		mhash = jhash_2words(mhash, multipath_hash, 0);
1955 
1956 	return mhash >> 1;
1957 }
1958 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1959 
1960 static int ip_mkroute_input(struct sk_buff *skb,
1961 			    struct fib_result *res,
1962 			    struct in_device *in_dev,
1963 			    __be32 daddr, __be32 saddr, u32 tos,
1964 			    struct flow_keys *hkeys)
1965 {
1966 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1967 	if (res->fi && fib_info_num_path(res->fi) > 1) {
1968 		int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
1969 
1970 		fib_select_multipath(res, h);
1971 	}
1972 #endif
1973 
1974 	/* create a routing cache entry */
1975 	return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1976 }
1977 
1978 /*
1979  *	NOTE. We drop all the packets that has local source
1980  *	addresses, because every properly looped back packet
1981  *	must have correct destination already attached by output routine.
1982  *
1983  *	Such approach solves two big problems:
1984  *	1. Not simplex devices are handled properly.
1985  *	2. IP spoofing attempts are filtered with 100% of guarantee.
1986  *	called with rcu_read_lock()
1987  */
1988 
1989 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1990 			       u8 tos, struct net_device *dev,
1991 			       struct fib_result *res)
1992 {
1993 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1994 	struct flow_keys *flkeys = NULL, _flkeys;
1995 	struct net    *net = dev_net(dev);
1996 	struct ip_tunnel_info *tun_info;
1997 	int		err = -EINVAL;
1998 	unsigned int	flags = 0;
1999 	u32		itag = 0;
2000 	struct rtable	*rth;
2001 	struct flowi4	fl4;
2002 	bool do_cache = true;
2003 
2004 	/* IP on this device is disabled. */
2005 
2006 	if (!in_dev)
2007 		goto out;
2008 
2009 	/* Check for the most weird martians, which can be not detected
2010 	   by fib_lookup.
2011 	 */
2012 
2013 	tun_info = skb_tunnel_info(skb);
2014 	if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2015 		fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2016 	else
2017 		fl4.flowi4_tun_key.tun_id = 0;
2018 	skb_dst_drop(skb);
2019 
2020 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2021 		goto martian_source;
2022 
2023 	res->fi = NULL;
2024 	res->table = NULL;
2025 	if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2026 		goto brd_input;
2027 
2028 	/* Accept zero addresses only to limited broadcast;
2029 	 * I even do not know to fix it or not. Waiting for complains :-)
2030 	 */
2031 	if (ipv4_is_zeronet(saddr))
2032 		goto martian_source;
2033 
2034 	if (ipv4_is_zeronet(daddr))
2035 		goto martian_destination;
2036 
2037 	/* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2038 	 * and call it once if daddr or/and saddr are loopback addresses
2039 	 */
2040 	if (ipv4_is_loopback(daddr)) {
2041 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2042 			goto martian_destination;
2043 	} else if (ipv4_is_loopback(saddr)) {
2044 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2045 			goto martian_source;
2046 	}
2047 
2048 	/*
2049 	 *	Now we are ready to route packet.
2050 	 */
2051 	fl4.flowi4_oif = 0;
2052 	fl4.flowi4_iif = dev->ifindex;
2053 	fl4.flowi4_mark = skb->mark;
2054 	fl4.flowi4_tos = tos;
2055 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2056 	fl4.flowi4_flags = 0;
2057 	fl4.daddr = daddr;
2058 	fl4.saddr = saddr;
2059 	fl4.flowi4_uid = sock_net_uid(net, NULL);
2060 
2061 	if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2062 		flkeys = &_flkeys;
2063 	} else {
2064 		fl4.flowi4_proto = 0;
2065 		fl4.fl4_sport = 0;
2066 		fl4.fl4_dport = 0;
2067 	}
2068 
2069 	err = fib_lookup(net, &fl4, res, 0);
2070 	if (err != 0) {
2071 		if (!IN_DEV_FORWARD(in_dev))
2072 			err = -EHOSTUNREACH;
2073 		goto no_route;
2074 	}
2075 
2076 	if (res->type == RTN_BROADCAST) {
2077 		if (IN_DEV_BFORWARD(in_dev))
2078 			goto make_route;
2079 		/* not do cache if bc_forwarding is enabled */
2080 		if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2081 			do_cache = false;
2082 		goto brd_input;
2083 	}
2084 
2085 	if (res->type == RTN_LOCAL) {
2086 		err = fib_validate_source(skb, saddr, daddr, tos,
2087 					  0, dev, in_dev, &itag);
2088 		if (err < 0)
2089 			goto martian_source;
2090 		goto local_input;
2091 	}
2092 
2093 	if (!IN_DEV_FORWARD(in_dev)) {
2094 		err = -EHOSTUNREACH;
2095 		goto no_route;
2096 	}
2097 	if (res->type != RTN_UNICAST)
2098 		goto martian_destination;
2099 
2100 make_route:
2101 	err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2102 out:	return err;
2103 
2104 brd_input:
2105 	if (skb->protocol != htons(ETH_P_IP))
2106 		goto e_inval;
2107 
2108 	if (!ipv4_is_zeronet(saddr)) {
2109 		err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2110 					  in_dev, &itag);
2111 		if (err < 0)
2112 			goto martian_source;
2113 	}
2114 	flags |= RTCF_BROADCAST;
2115 	res->type = RTN_BROADCAST;
2116 	RT_CACHE_STAT_INC(in_brd);
2117 
2118 local_input:
2119 	do_cache &= res->fi && !itag;
2120 	if (do_cache) {
2121 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2122 
2123 		rth = rcu_dereference(nhc->nhc_rth_input);
2124 		if (rt_cache_valid(rth)) {
2125 			skb_dst_set_noref(skb, &rth->dst);
2126 			err = 0;
2127 			goto out;
2128 		}
2129 	}
2130 
2131 	rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2132 			   flags | RTCF_LOCAL, res->type,
2133 			   IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
2134 	if (!rth)
2135 		goto e_nobufs;
2136 
2137 	rth->dst.output= ip_rt_bug;
2138 #ifdef CONFIG_IP_ROUTE_CLASSID
2139 	rth->dst.tclassid = itag;
2140 #endif
2141 	rth->rt_is_input = 1;
2142 
2143 	RT_CACHE_STAT_INC(in_slow_tot);
2144 	if (res->type == RTN_UNREACHABLE) {
2145 		rth->dst.input= ip_error;
2146 		rth->dst.error= -err;
2147 		rth->rt_flags 	&= ~RTCF_LOCAL;
2148 	}
2149 
2150 	if (do_cache) {
2151 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2152 
2153 		rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2154 		if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2155 			WARN_ON(rth->dst.input == lwtunnel_input);
2156 			rth->dst.lwtstate->orig_input = rth->dst.input;
2157 			rth->dst.input = lwtunnel_input;
2158 		}
2159 
2160 		if (unlikely(!rt_cache_route(nhc, rth)))
2161 			rt_add_uncached_list(rth);
2162 	}
2163 	skb_dst_set(skb, &rth->dst);
2164 	err = 0;
2165 	goto out;
2166 
2167 no_route:
2168 	RT_CACHE_STAT_INC(in_no_route);
2169 	res->type = RTN_UNREACHABLE;
2170 	res->fi = NULL;
2171 	res->table = NULL;
2172 	goto local_input;
2173 
2174 	/*
2175 	 *	Do not cache martian addresses: they should be logged (RFC1812)
2176 	 */
2177 martian_destination:
2178 	RT_CACHE_STAT_INC(in_martian_dst);
2179 #ifdef CONFIG_IP_ROUTE_VERBOSE
2180 	if (IN_DEV_LOG_MARTIANS(in_dev))
2181 		net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2182 				     &daddr, &saddr, dev->name);
2183 #endif
2184 
2185 e_inval:
2186 	err = -EINVAL;
2187 	goto out;
2188 
2189 e_nobufs:
2190 	err = -ENOBUFS;
2191 	goto out;
2192 
2193 martian_source:
2194 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2195 	goto out;
2196 }
2197 
2198 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2199 			 u8 tos, struct net_device *dev)
2200 {
2201 	struct fib_result res;
2202 	int err;
2203 
2204 	tos &= IPTOS_RT_MASK;
2205 	rcu_read_lock();
2206 	err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2207 	rcu_read_unlock();
2208 
2209 	return err;
2210 }
2211 EXPORT_SYMBOL(ip_route_input_noref);
2212 
2213 /* called with rcu_read_lock held */
2214 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2215 		       u8 tos, struct net_device *dev, struct fib_result *res)
2216 {
2217 	/* Multicast recognition logic is moved from route cache to here.
2218 	   The problem was that too many Ethernet cards have broken/missing
2219 	   hardware multicast filters :-( As result the host on multicasting
2220 	   network acquires a lot of useless route cache entries, sort of
2221 	   SDR messages from all the world. Now we try to get rid of them.
2222 	   Really, provided software IP multicast filter is organized
2223 	   reasonably (at least, hashed), it does not result in a slowdown
2224 	   comparing with route cache reject entries.
2225 	   Note, that multicast routers are not affected, because
2226 	   route cache entry is created eventually.
2227 	 */
2228 	if (ipv4_is_multicast(daddr)) {
2229 		struct in_device *in_dev = __in_dev_get_rcu(dev);
2230 		int our = 0;
2231 		int err = -EINVAL;
2232 
2233 		if (!in_dev)
2234 			return err;
2235 		our = ip_check_mc_rcu(in_dev, daddr, saddr,
2236 				      ip_hdr(skb)->protocol);
2237 
2238 		/* check l3 master if no match yet */
2239 		if (!our && netif_is_l3_slave(dev)) {
2240 			struct in_device *l3_in_dev;
2241 
2242 			l3_in_dev = __in_dev_get_rcu(skb->dev);
2243 			if (l3_in_dev)
2244 				our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2245 						      ip_hdr(skb)->protocol);
2246 		}
2247 
2248 		if (our
2249 #ifdef CONFIG_IP_MROUTE
2250 			||
2251 		    (!ipv4_is_local_multicast(daddr) &&
2252 		     IN_DEV_MFORWARD(in_dev))
2253 #endif
2254 		   ) {
2255 			err = ip_route_input_mc(skb, daddr, saddr,
2256 						tos, dev, our);
2257 		}
2258 		return err;
2259 	}
2260 
2261 	return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2262 }
2263 
2264 /* called with rcu_read_lock() */
2265 static struct rtable *__mkroute_output(const struct fib_result *res,
2266 				       const struct flowi4 *fl4, int orig_oif,
2267 				       struct net_device *dev_out,
2268 				       unsigned int flags)
2269 {
2270 	struct fib_info *fi = res->fi;
2271 	struct fib_nh_exception *fnhe;
2272 	struct in_device *in_dev;
2273 	u16 type = res->type;
2274 	struct rtable *rth;
2275 	bool do_cache;
2276 
2277 	in_dev = __in_dev_get_rcu(dev_out);
2278 	if (!in_dev)
2279 		return ERR_PTR(-EINVAL);
2280 
2281 	if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2282 		if (ipv4_is_loopback(fl4->saddr) &&
2283 		    !(dev_out->flags & IFF_LOOPBACK) &&
2284 		    !netif_is_l3_master(dev_out))
2285 			return ERR_PTR(-EINVAL);
2286 
2287 	if (ipv4_is_lbcast(fl4->daddr))
2288 		type = RTN_BROADCAST;
2289 	else if (ipv4_is_multicast(fl4->daddr))
2290 		type = RTN_MULTICAST;
2291 	else if (ipv4_is_zeronet(fl4->daddr))
2292 		return ERR_PTR(-EINVAL);
2293 
2294 	if (dev_out->flags & IFF_LOOPBACK)
2295 		flags |= RTCF_LOCAL;
2296 
2297 	do_cache = true;
2298 	if (type == RTN_BROADCAST) {
2299 		flags |= RTCF_BROADCAST | RTCF_LOCAL;
2300 		fi = NULL;
2301 	} else if (type == RTN_MULTICAST) {
2302 		flags |= RTCF_MULTICAST | RTCF_LOCAL;
2303 		if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2304 				     fl4->flowi4_proto))
2305 			flags &= ~RTCF_LOCAL;
2306 		else
2307 			do_cache = false;
2308 		/* If multicast route do not exist use
2309 		 * default one, but do not gateway in this case.
2310 		 * Yes, it is hack.
2311 		 */
2312 		if (fi && res->prefixlen < 4)
2313 			fi = NULL;
2314 	} else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2315 		   (orig_oif != dev_out->ifindex)) {
2316 		/* For local routes that require a particular output interface
2317 		 * we do not want to cache the result.  Caching the result
2318 		 * causes incorrect behaviour when there are multiple source
2319 		 * addresses on the interface, the end result being that if the
2320 		 * intended recipient is waiting on that interface for the
2321 		 * packet he won't receive it because it will be delivered on
2322 		 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2323 		 * be set to the loopback interface as well.
2324 		 */
2325 		do_cache = false;
2326 	}
2327 
2328 	fnhe = NULL;
2329 	do_cache &= fi != NULL;
2330 	if (fi) {
2331 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2332 		struct rtable __rcu **prth;
2333 
2334 		fnhe = find_exception(nhc, fl4->daddr);
2335 		if (!do_cache)
2336 			goto add;
2337 		if (fnhe) {
2338 			prth = &fnhe->fnhe_rth_output;
2339 		} else {
2340 			if (unlikely(fl4->flowi4_flags &
2341 				     FLOWI_FLAG_KNOWN_NH &&
2342 				     !(nhc->nhc_gw_family &&
2343 				       nhc->nhc_scope == RT_SCOPE_LINK))) {
2344 				do_cache = false;
2345 				goto add;
2346 			}
2347 			prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2348 		}
2349 		rth = rcu_dereference(*prth);
2350 		if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2351 			return rth;
2352 	}
2353 
2354 add:
2355 	rth = rt_dst_alloc(dev_out, flags, type,
2356 			   IN_DEV_CONF_GET(in_dev, NOPOLICY),
2357 			   IN_DEV_CONF_GET(in_dev, NOXFRM),
2358 			   do_cache);
2359 	if (!rth)
2360 		return ERR_PTR(-ENOBUFS);
2361 
2362 	rth->rt_iif = orig_oif;
2363 
2364 	RT_CACHE_STAT_INC(out_slow_tot);
2365 
2366 	if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2367 		if (flags & RTCF_LOCAL &&
2368 		    !(dev_out->flags & IFF_LOOPBACK)) {
2369 			rth->dst.output = ip_mc_output;
2370 			RT_CACHE_STAT_INC(out_slow_mc);
2371 		}
2372 #ifdef CONFIG_IP_MROUTE
2373 		if (type == RTN_MULTICAST) {
2374 			if (IN_DEV_MFORWARD(in_dev) &&
2375 			    !ipv4_is_local_multicast(fl4->daddr)) {
2376 				rth->dst.input = ip_mr_input;
2377 				rth->dst.output = ip_mc_output;
2378 			}
2379 		}
2380 #endif
2381 	}
2382 
2383 	rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2384 	lwtunnel_set_redirect(&rth->dst);
2385 
2386 	return rth;
2387 }
2388 
2389 /*
2390  * Major route resolver routine.
2391  */
2392 
2393 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2394 					const struct sk_buff *skb)
2395 {
2396 	__u8 tos = RT_FL_TOS(fl4);
2397 	struct fib_result res = {
2398 		.type		= RTN_UNSPEC,
2399 		.fi		= NULL,
2400 		.table		= NULL,
2401 		.tclassid	= 0,
2402 	};
2403 	struct rtable *rth;
2404 
2405 	fl4->flowi4_iif = LOOPBACK_IFINDEX;
2406 	fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2407 	fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2408 			 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2409 
2410 	rcu_read_lock();
2411 	rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2412 	rcu_read_unlock();
2413 
2414 	return rth;
2415 }
2416 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2417 
2418 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2419 					    struct fib_result *res,
2420 					    const struct sk_buff *skb)
2421 {
2422 	struct net_device *dev_out = NULL;
2423 	int orig_oif = fl4->flowi4_oif;
2424 	unsigned int flags = 0;
2425 	struct rtable *rth;
2426 	int err = -ENETUNREACH;
2427 
2428 	if (fl4->saddr) {
2429 		rth = ERR_PTR(-EINVAL);
2430 		if (ipv4_is_multicast(fl4->saddr) ||
2431 		    ipv4_is_lbcast(fl4->saddr) ||
2432 		    ipv4_is_zeronet(fl4->saddr))
2433 			goto out;
2434 
2435 		/* I removed check for oif == dev_out->oif here.
2436 		   It was wrong for two reasons:
2437 		   1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2438 		      is assigned to multiple interfaces.
2439 		   2. Moreover, we are allowed to send packets with saddr
2440 		      of another iface. --ANK
2441 		 */
2442 
2443 		if (fl4->flowi4_oif == 0 &&
2444 		    (ipv4_is_multicast(fl4->daddr) ||
2445 		     ipv4_is_lbcast(fl4->daddr))) {
2446 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2447 			dev_out = __ip_dev_find(net, fl4->saddr, false);
2448 			if (!dev_out)
2449 				goto out;
2450 
2451 			/* Special hack: user can direct multicasts
2452 			   and limited broadcast via necessary interface
2453 			   without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2454 			   This hack is not just for fun, it allows
2455 			   vic,vat and friends to work.
2456 			   They bind socket to loopback, set ttl to zero
2457 			   and expect that it will work.
2458 			   From the viewpoint of routing cache they are broken,
2459 			   because we are not allowed to build multicast path
2460 			   with loopback source addr (look, routing cache
2461 			   cannot know, that ttl is zero, so that packet
2462 			   will not leave this host and route is valid).
2463 			   Luckily, this hack is good workaround.
2464 			 */
2465 
2466 			fl4->flowi4_oif = dev_out->ifindex;
2467 			goto make_route;
2468 		}
2469 
2470 		if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2471 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2472 			if (!__ip_dev_find(net, fl4->saddr, false))
2473 				goto out;
2474 		}
2475 	}
2476 
2477 
2478 	if (fl4->flowi4_oif) {
2479 		dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2480 		rth = ERR_PTR(-ENODEV);
2481 		if (!dev_out)
2482 			goto out;
2483 
2484 		/* RACE: Check return value of inet_select_addr instead. */
2485 		if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2486 			rth = ERR_PTR(-ENETUNREACH);
2487 			goto out;
2488 		}
2489 		if (ipv4_is_local_multicast(fl4->daddr) ||
2490 		    ipv4_is_lbcast(fl4->daddr) ||
2491 		    fl4->flowi4_proto == IPPROTO_IGMP) {
2492 			if (!fl4->saddr)
2493 				fl4->saddr = inet_select_addr(dev_out, 0,
2494 							      RT_SCOPE_LINK);
2495 			goto make_route;
2496 		}
2497 		if (!fl4->saddr) {
2498 			if (ipv4_is_multicast(fl4->daddr))
2499 				fl4->saddr = inet_select_addr(dev_out, 0,
2500 							      fl4->flowi4_scope);
2501 			else if (!fl4->daddr)
2502 				fl4->saddr = inet_select_addr(dev_out, 0,
2503 							      RT_SCOPE_HOST);
2504 		}
2505 	}
2506 
2507 	if (!fl4->daddr) {
2508 		fl4->daddr = fl4->saddr;
2509 		if (!fl4->daddr)
2510 			fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2511 		dev_out = net->loopback_dev;
2512 		fl4->flowi4_oif = LOOPBACK_IFINDEX;
2513 		res->type = RTN_LOCAL;
2514 		flags |= RTCF_LOCAL;
2515 		goto make_route;
2516 	}
2517 
2518 	err = fib_lookup(net, fl4, res, 0);
2519 	if (err) {
2520 		res->fi = NULL;
2521 		res->table = NULL;
2522 		if (fl4->flowi4_oif &&
2523 		    (ipv4_is_multicast(fl4->daddr) ||
2524 		    !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2525 			/* Apparently, routing tables are wrong. Assume,
2526 			   that the destination is on link.
2527 
2528 			   WHY? DW.
2529 			   Because we are allowed to send to iface
2530 			   even if it has NO routes and NO assigned
2531 			   addresses. When oif is specified, routing
2532 			   tables are looked up with only one purpose:
2533 			   to catch if destination is gatewayed, rather than
2534 			   direct. Moreover, if MSG_DONTROUTE is set,
2535 			   we send packet, ignoring both routing tables
2536 			   and ifaddr state. --ANK
2537 
2538 
2539 			   We could make it even if oif is unknown,
2540 			   likely IPv6, but we do not.
2541 			 */
2542 
2543 			if (fl4->saddr == 0)
2544 				fl4->saddr = inet_select_addr(dev_out, 0,
2545 							      RT_SCOPE_LINK);
2546 			res->type = RTN_UNICAST;
2547 			goto make_route;
2548 		}
2549 		rth = ERR_PTR(err);
2550 		goto out;
2551 	}
2552 
2553 	if (res->type == RTN_LOCAL) {
2554 		if (!fl4->saddr) {
2555 			if (res->fi->fib_prefsrc)
2556 				fl4->saddr = res->fi->fib_prefsrc;
2557 			else
2558 				fl4->saddr = fl4->daddr;
2559 		}
2560 
2561 		/* L3 master device is the loopback for that domain */
2562 		dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2563 			net->loopback_dev;
2564 
2565 		/* make sure orig_oif points to fib result device even
2566 		 * though packet rx/tx happens over loopback or l3mdev
2567 		 */
2568 		orig_oif = FIB_RES_OIF(*res);
2569 
2570 		fl4->flowi4_oif = dev_out->ifindex;
2571 		flags |= RTCF_LOCAL;
2572 		goto make_route;
2573 	}
2574 
2575 	fib_select_path(net, res, fl4, skb);
2576 
2577 	dev_out = FIB_RES_DEV(*res);
2578 	fl4->flowi4_oif = dev_out->ifindex;
2579 
2580 
2581 make_route:
2582 	rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2583 
2584 out:
2585 	return rth;
2586 }
2587 
2588 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2589 {
2590 	return NULL;
2591 }
2592 
2593 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2594 {
2595 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2596 
2597 	return mtu ? : dst->dev->mtu;
2598 }
2599 
2600 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2601 					  struct sk_buff *skb, u32 mtu)
2602 {
2603 }
2604 
2605 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2606 				       struct sk_buff *skb)
2607 {
2608 }
2609 
2610 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2611 					  unsigned long old)
2612 {
2613 	return NULL;
2614 }
2615 
2616 static struct dst_ops ipv4_dst_blackhole_ops = {
2617 	.family			=	AF_INET,
2618 	.check			=	ipv4_blackhole_dst_check,
2619 	.mtu			=	ipv4_blackhole_mtu,
2620 	.default_advmss		=	ipv4_default_advmss,
2621 	.update_pmtu		=	ipv4_rt_blackhole_update_pmtu,
2622 	.redirect		=	ipv4_rt_blackhole_redirect,
2623 	.cow_metrics		=	ipv4_rt_blackhole_cow_metrics,
2624 	.neigh_lookup		=	ipv4_neigh_lookup,
2625 };
2626 
2627 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2628 {
2629 	struct rtable *ort = (struct rtable *) dst_orig;
2630 	struct rtable *rt;
2631 
2632 	rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2633 	if (rt) {
2634 		struct dst_entry *new = &rt->dst;
2635 
2636 		new->__use = 1;
2637 		new->input = dst_discard;
2638 		new->output = dst_discard_out;
2639 
2640 		new->dev = net->loopback_dev;
2641 		if (new->dev)
2642 			dev_hold(new->dev);
2643 
2644 		rt->rt_is_input = ort->rt_is_input;
2645 		rt->rt_iif = ort->rt_iif;
2646 		rt->rt_pmtu = ort->rt_pmtu;
2647 		rt->rt_mtu_locked = ort->rt_mtu_locked;
2648 
2649 		rt->rt_genid = rt_genid_ipv4(net);
2650 		rt->rt_flags = ort->rt_flags;
2651 		rt->rt_type = ort->rt_type;
2652 		rt->rt_gw_family = ort->rt_gw_family;
2653 		if (rt->rt_gw_family == AF_INET)
2654 			rt->rt_gw4 = ort->rt_gw4;
2655 		else if (rt->rt_gw_family == AF_INET6)
2656 			rt->rt_gw6 = ort->rt_gw6;
2657 
2658 		INIT_LIST_HEAD(&rt->rt_uncached);
2659 	}
2660 
2661 	dst_release(dst_orig);
2662 
2663 	return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2664 }
2665 
2666 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2667 				    const struct sock *sk)
2668 {
2669 	struct rtable *rt = __ip_route_output_key(net, flp4);
2670 
2671 	if (IS_ERR(rt))
2672 		return rt;
2673 
2674 	if (flp4->flowi4_proto)
2675 		rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2676 							flowi4_to_flowi(flp4),
2677 							sk, 0);
2678 
2679 	return rt;
2680 }
2681 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2682 
2683 /* called with rcu_read_lock held */
2684 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2685 			struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2686 			struct sk_buff *skb, u32 portid, u32 seq)
2687 {
2688 	struct rtmsg *r;
2689 	struct nlmsghdr *nlh;
2690 	unsigned long expires = 0;
2691 	u32 error;
2692 	u32 metrics[RTAX_MAX];
2693 
2694 	nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0);
2695 	if (!nlh)
2696 		return -EMSGSIZE;
2697 
2698 	r = nlmsg_data(nlh);
2699 	r->rtm_family	 = AF_INET;
2700 	r->rtm_dst_len	= 32;
2701 	r->rtm_src_len	= 0;
2702 	r->rtm_tos	= fl4->flowi4_tos;
2703 	r->rtm_table	= table_id < 256 ? table_id : RT_TABLE_COMPAT;
2704 	if (nla_put_u32(skb, RTA_TABLE, table_id))
2705 		goto nla_put_failure;
2706 	r->rtm_type	= rt->rt_type;
2707 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2708 	r->rtm_protocol = RTPROT_UNSPEC;
2709 	r->rtm_flags	= (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2710 	if (rt->rt_flags & RTCF_NOTIFY)
2711 		r->rtm_flags |= RTM_F_NOTIFY;
2712 	if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2713 		r->rtm_flags |= RTCF_DOREDIRECT;
2714 
2715 	if (nla_put_in_addr(skb, RTA_DST, dst))
2716 		goto nla_put_failure;
2717 	if (src) {
2718 		r->rtm_src_len = 32;
2719 		if (nla_put_in_addr(skb, RTA_SRC, src))
2720 			goto nla_put_failure;
2721 	}
2722 	if (rt->dst.dev &&
2723 	    nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2724 		goto nla_put_failure;
2725 #ifdef CONFIG_IP_ROUTE_CLASSID
2726 	if (rt->dst.tclassid &&
2727 	    nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2728 		goto nla_put_failure;
2729 #endif
2730 	if (!rt_is_input_route(rt) &&
2731 	    fl4->saddr != src) {
2732 		if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2733 			goto nla_put_failure;
2734 	}
2735 	if (rt->rt_gw_family == AF_INET &&
2736 	    nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2737 		goto nla_put_failure;
2738 	} else if (rt->rt_gw_family == AF_INET6) {
2739 		int alen = sizeof(struct in6_addr);
2740 		struct nlattr *nla;
2741 		struct rtvia *via;
2742 
2743 		nla = nla_reserve(skb, RTA_VIA, alen + 2);
2744 		if (!nla)
2745 			goto nla_put_failure;
2746 
2747 		via = nla_data(nla);
2748 		via->rtvia_family = AF_INET6;
2749 		memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2750 	}
2751 
2752 	expires = rt->dst.expires;
2753 	if (expires) {
2754 		unsigned long now = jiffies;
2755 
2756 		if (time_before(now, expires))
2757 			expires -= now;
2758 		else
2759 			expires = 0;
2760 	}
2761 
2762 	memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2763 	if (rt->rt_pmtu && expires)
2764 		metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2765 	if (rt->rt_mtu_locked && expires)
2766 		metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2767 	if (rtnetlink_put_metrics(skb, metrics) < 0)
2768 		goto nla_put_failure;
2769 
2770 	if (fl4->flowi4_mark &&
2771 	    nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2772 		goto nla_put_failure;
2773 
2774 	if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2775 	    nla_put_u32(skb, RTA_UID,
2776 			from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2777 		goto nla_put_failure;
2778 
2779 	error = rt->dst.error;
2780 
2781 	if (rt_is_input_route(rt)) {
2782 #ifdef CONFIG_IP_MROUTE
2783 		if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2784 		    IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2785 			int err = ipmr_get_route(net, skb,
2786 						 fl4->saddr, fl4->daddr,
2787 						 r, portid);
2788 
2789 			if (err <= 0) {
2790 				if (err == 0)
2791 					return 0;
2792 				goto nla_put_failure;
2793 			}
2794 		} else
2795 #endif
2796 			if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
2797 				goto nla_put_failure;
2798 	}
2799 
2800 	if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2801 		goto nla_put_failure;
2802 
2803 	nlmsg_end(skb, nlh);
2804 	return 0;
2805 
2806 nla_put_failure:
2807 	nlmsg_cancel(skb, nlh);
2808 	return -EMSGSIZE;
2809 }
2810 
2811 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
2812 						   u8 ip_proto, __be16 sport,
2813 						   __be16 dport)
2814 {
2815 	struct sk_buff *skb;
2816 	struct iphdr *iph;
2817 
2818 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2819 	if (!skb)
2820 		return NULL;
2821 
2822 	/* Reserve room for dummy headers, this skb can pass
2823 	 * through good chunk of routing engine.
2824 	 */
2825 	skb_reset_mac_header(skb);
2826 	skb_reset_network_header(skb);
2827 	skb->protocol = htons(ETH_P_IP);
2828 	iph = skb_put(skb, sizeof(struct iphdr));
2829 	iph->protocol = ip_proto;
2830 	iph->saddr = src;
2831 	iph->daddr = dst;
2832 	iph->version = 0x4;
2833 	iph->frag_off = 0;
2834 	iph->ihl = 0x5;
2835 	skb_set_transport_header(skb, skb->len);
2836 
2837 	switch (iph->protocol) {
2838 	case IPPROTO_UDP: {
2839 		struct udphdr *udph;
2840 
2841 		udph = skb_put_zero(skb, sizeof(struct udphdr));
2842 		udph->source = sport;
2843 		udph->dest = dport;
2844 		udph->len = sizeof(struct udphdr);
2845 		udph->check = 0;
2846 		break;
2847 	}
2848 	case IPPROTO_TCP: {
2849 		struct tcphdr *tcph;
2850 
2851 		tcph = skb_put_zero(skb, sizeof(struct tcphdr));
2852 		tcph->source	= sport;
2853 		tcph->dest	= dport;
2854 		tcph->doff	= sizeof(struct tcphdr) / 4;
2855 		tcph->rst = 1;
2856 		tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
2857 					    src, dst, 0);
2858 		break;
2859 	}
2860 	case IPPROTO_ICMP: {
2861 		struct icmphdr *icmph;
2862 
2863 		icmph = skb_put_zero(skb, sizeof(struct icmphdr));
2864 		icmph->type = ICMP_ECHO;
2865 		icmph->code = 0;
2866 	}
2867 	}
2868 
2869 	return skb;
2870 }
2871 
2872 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
2873 				       const struct nlmsghdr *nlh,
2874 				       struct nlattr **tb,
2875 				       struct netlink_ext_ack *extack)
2876 {
2877 	struct rtmsg *rtm;
2878 	int i, err;
2879 
2880 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2881 		NL_SET_ERR_MSG(extack,
2882 			       "ipv4: Invalid header for route get request");
2883 		return -EINVAL;
2884 	}
2885 
2886 	if (!netlink_strict_get_check(skb))
2887 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2888 					      rtm_ipv4_policy, extack);
2889 
2890 	rtm = nlmsg_data(nlh);
2891 	if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
2892 	    (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
2893 	    rtm->rtm_table || rtm->rtm_protocol ||
2894 	    rtm->rtm_scope || rtm->rtm_type) {
2895 		NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
2896 		return -EINVAL;
2897 	}
2898 
2899 	if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
2900 			       RTM_F_LOOKUP_TABLE |
2901 			       RTM_F_FIB_MATCH)) {
2902 		NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
2903 		return -EINVAL;
2904 	}
2905 
2906 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2907 					    rtm_ipv4_policy, extack);
2908 	if (err)
2909 		return err;
2910 
2911 	if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
2912 	    (tb[RTA_DST] && !rtm->rtm_dst_len)) {
2913 		NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
2914 		return -EINVAL;
2915 	}
2916 
2917 	for (i = 0; i <= RTA_MAX; i++) {
2918 		if (!tb[i])
2919 			continue;
2920 
2921 		switch (i) {
2922 		case RTA_IIF:
2923 		case RTA_OIF:
2924 		case RTA_SRC:
2925 		case RTA_DST:
2926 		case RTA_IP_PROTO:
2927 		case RTA_SPORT:
2928 		case RTA_DPORT:
2929 		case RTA_MARK:
2930 		case RTA_UID:
2931 			break;
2932 		default:
2933 			NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
2934 			return -EINVAL;
2935 		}
2936 	}
2937 
2938 	return 0;
2939 }
2940 
2941 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2942 			     struct netlink_ext_ack *extack)
2943 {
2944 	struct net *net = sock_net(in_skb->sk);
2945 	struct nlattr *tb[RTA_MAX+1];
2946 	u32 table_id = RT_TABLE_MAIN;
2947 	__be16 sport = 0, dport = 0;
2948 	struct fib_result res = {};
2949 	u8 ip_proto = IPPROTO_UDP;
2950 	struct rtable *rt = NULL;
2951 	struct sk_buff *skb;
2952 	struct rtmsg *rtm;
2953 	struct flowi4 fl4 = {};
2954 	__be32 dst = 0;
2955 	__be32 src = 0;
2956 	kuid_t uid;
2957 	u32 iif;
2958 	int err;
2959 	int mark;
2960 
2961 	err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
2962 	if (err < 0)
2963 		return err;
2964 
2965 	rtm = nlmsg_data(nlh);
2966 	src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2967 	dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2968 	iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2969 	mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2970 	if (tb[RTA_UID])
2971 		uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2972 	else
2973 		uid = (iif ? INVALID_UID : current_uid());
2974 
2975 	if (tb[RTA_IP_PROTO]) {
2976 		err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
2977 						  &ip_proto, AF_INET, extack);
2978 		if (err)
2979 			return err;
2980 	}
2981 
2982 	if (tb[RTA_SPORT])
2983 		sport = nla_get_be16(tb[RTA_SPORT]);
2984 
2985 	if (tb[RTA_DPORT])
2986 		dport = nla_get_be16(tb[RTA_DPORT]);
2987 
2988 	skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
2989 	if (!skb)
2990 		return -ENOBUFS;
2991 
2992 	fl4.daddr = dst;
2993 	fl4.saddr = src;
2994 	fl4.flowi4_tos = rtm->rtm_tos;
2995 	fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2996 	fl4.flowi4_mark = mark;
2997 	fl4.flowi4_uid = uid;
2998 	if (sport)
2999 		fl4.fl4_sport = sport;
3000 	if (dport)
3001 		fl4.fl4_dport = dport;
3002 	fl4.flowi4_proto = ip_proto;
3003 
3004 	rcu_read_lock();
3005 
3006 	if (iif) {
3007 		struct net_device *dev;
3008 
3009 		dev = dev_get_by_index_rcu(net, iif);
3010 		if (!dev) {
3011 			err = -ENODEV;
3012 			goto errout_rcu;
3013 		}
3014 
3015 		fl4.flowi4_iif = iif; /* for rt_fill_info */
3016 		skb->dev	= dev;
3017 		skb->mark	= mark;
3018 		err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos,
3019 					 dev, &res);
3020 
3021 		rt = skb_rtable(skb);
3022 		if (err == 0 && rt->dst.error)
3023 			err = -rt->dst.error;
3024 	} else {
3025 		fl4.flowi4_iif = LOOPBACK_IFINDEX;
3026 		skb->dev = net->loopback_dev;
3027 		rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3028 		err = 0;
3029 		if (IS_ERR(rt))
3030 			err = PTR_ERR(rt);
3031 		else
3032 			skb_dst_set(skb, &rt->dst);
3033 	}
3034 
3035 	if (err)
3036 		goto errout_rcu;
3037 
3038 	if (rtm->rtm_flags & RTM_F_NOTIFY)
3039 		rt->rt_flags |= RTCF_NOTIFY;
3040 
3041 	if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3042 		table_id = res.table ? res.table->tb_id : 0;
3043 
3044 	/* reset skb for netlink reply msg */
3045 	skb_trim(skb, 0);
3046 	skb_reset_network_header(skb);
3047 	skb_reset_transport_header(skb);
3048 	skb_reset_mac_header(skb);
3049 
3050 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3051 		if (!res.fi) {
3052 			err = fib_props[res.type].error;
3053 			if (!err)
3054 				err = -EHOSTUNREACH;
3055 			goto errout_rcu;
3056 		}
3057 		err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3058 				    nlh->nlmsg_seq, RTM_NEWROUTE, table_id,
3059 				    rt->rt_type, res.prefix, res.prefixlen,
3060 				    fl4.flowi4_tos, res.fi, 0);
3061 	} else {
3062 		err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3063 				   NETLINK_CB(in_skb).portid, nlh->nlmsg_seq);
3064 	}
3065 	if (err < 0)
3066 		goto errout_rcu;
3067 
3068 	rcu_read_unlock();
3069 
3070 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3071 
3072 errout_free:
3073 	return err;
3074 errout_rcu:
3075 	rcu_read_unlock();
3076 	kfree_skb(skb);
3077 	goto errout_free;
3078 }
3079 
3080 void ip_rt_multicast_event(struct in_device *in_dev)
3081 {
3082 	rt_cache_flush(dev_net(in_dev->dev));
3083 }
3084 
3085 #ifdef CONFIG_SYSCTL
3086 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3087 static int ip_rt_gc_min_interval __read_mostly	= HZ / 2;
3088 static int ip_rt_gc_elasticity __read_mostly	= 8;
3089 static int ip_min_valid_pmtu __read_mostly	= IPV4_MIN_MTU;
3090 
3091 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3092 					void __user *buffer,
3093 					size_t *lenp, loff_t *ppos)
3094 {
3095 	struct net *net = (struct net *)__ctl->extra1;
3096 
3097 	if (write) {
3098 		rt_cache_flush(net);
3099 		fnhe_genid_bump(net);
3100 		return 0;
3101 	}
3102 
3103 	return -EINVAL;
3104 }
3105 
3106 static struct ctl_table ipv4_route_table[] = {
3107 	{
3108 		.procname	= "gc_thresh",
3109 		.data		= &ipv4_dst_ops.gc_thresh,
3110 		.maxlen		= sizeof(int),
3111 		.mode		= 0644,
3112 		.proc_handler	= proc_dointvec,
3113 	},
3114 	{
3115 		.procname	= "max_size",
3116 		.data		= &ip_rt_max_size,
3117 		.maxlen		= sizeof(int),
3118 		.mode		= 0644,
3119 		.proc_handler	= proc_dointvec,
3120 	},
3121 	{
3122 		/*  Deprecated. Use gc_min_interval_ms */
3123 
3124 		.procname	= "gc_min_interval",
3125 		.data		= &ip_rt_gc_min_interval,
3126 		.maxlen		= sizeof(int),
3127 		.mode		= 0644,
3128 		.proc_handler	= proc_dointvec_jiffies,
3129 	},
3130 	{
3131 		.procname	= "gc_min_interval_ms",
3132 		.data		= &ip_rt_gc_min_interval,
3133 		.maxlen		= sizeof(int),
3134 		.mode		= 0644,
3135 		.proc_handler	= proc_dointvec_ms_jiffies,
3136 	},
3137 	{
3138 		.procname	= "gc_timeout",
3139 		.data		= &ip_rt_gc_timeout,
3140 		.maxlen		= sizeof(int),
3141 		.mode		= 0644,
3142 		.proc_handler	= proc_dointvec_jiffies,
3143 	},
3144 	{
3145 		.procname	= "gc_interval",
3146 		.data		= &ip_rt_gc_interval,
3147 		.maxlen		= sizeof(int),
3148 		.mode		= 0644,
3149 		.proc_handler	= proc_dointvec_jiffies,
3150 	},
3151 	{
3152 		.procname	= "redirect_load",
3153 		.data		= &ip_rt_redirect_load,
3154 		.maxlen		= sizeof(int),
3155 		.mode		= 0644,
3156 		.proc_handler	= proc_dointvec,
3157 	},
3158 	{
3159 		.procname	= "redirect_number",
3160 		.data		= &ip_rt_redirect_number,
3161 		.maxlen		= sizeof(int),
3162 		.mode		= 0644,
3163 		.proc_handler	= proc_dointvec,
3164 	},
3165 	{
3166 		.procname	= "redirect_silence",
3167 		.data		= &ip_rt_redirect_silence,
3168 		.maxlen		= sizeof(int),
3169 		.mode		= 0644,
3170 		.proc_handler	= proc_dointvec,
3171 	},
3172 	{
3173 		.procname	= "error_cost",
3174 		.data		= &ip_rt_error_cost,
3175 		.maxlen		= sizeof(int),
3176 		.mode		= 0644,
3177 		.proc_handler	= proc_dointvec,
3178 	},
3179 	{
3180 		.procname	= "error_burst",
3181 		.data		= &ip_rt_error_burst,
3182 		.maxlen		= sizeof(int),
3183 		.mode		= 0644,
3184 		.proc_handler	= proc_dointvec,
3185 	},
3186 	{
3187 		.procname	= "gc_elasticity",
3188 		.data		= &ip_rt_gc_elasticity,
3189 		.maxlen		= sizeof(int),
3190 		.mode		= 0644,
3191 		.proc_handler	= proc_dointvec,
3192 	},
3193 	{
3194 		.procname	= "mtu_expires",
3195 		.data		= &ip_rt_mtu_expires,
3196 		.maxlen		= sizeof(int),
3197 		.mode		= 0644,
3198 		.proc_handler	= proc_dointvec_jiffies,
3199 	},
3200 	{
3201 		.procname	= "min_pmtu",
3202 		.data		= &ip_rt_min_pmtu,
3203 		.maxlen		= sizeof(int),
3204 		.mode		= 0644,
3205 		.proc_handler	= proc_dointvec_minmax,
3206 		.extra1		= &ip_min_valid_pmtu,
3207 	},
3208 	{
3209 		.procname	= "min_adv_mss",
3210 		.data		= &ip_rt_min_advmss,
3211 		.maxlen		= sizeof(int),
3212 		.mode		= 0644,
3213 		.proc_handler	= proc_dointvec,
3214 	},
3215 	{ }
3216 };
3217 
3218 static struct ctl_table ipv4_route_flush_table[] = {
3219 	{
3220 		.procname	= "flush",
3221 		.maxlen		= sizeof(int),
3222 		.mode		= 0200,
3223 		.proc_handler	= ipv4_sysctl_rtcache_flush,
3224 	},
3225 	{ },
3226 };
3227 
3228 static __net_init int sysctl_route_net_init(struct net *net)
3229 {
3230 	struct ctl_table *tbl;
3231 
3232 	tbl = ipv4_route_flush_table;
3233 	if (!net_eq(net, &init_net)) {
3234 		tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3235 		if (!tbl)
3236 			goto err_dup;
3237 
3238 		/* Don't export sysctls to unprivileged users */
3239 		if (net->user_ns != &init_user_ns)
3240 			tbl[0].procname = NULL;
3241 	}
3242 	tbl[0].extra1 = net;
3243 
3244 	net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3245 	if (!net->ipv4.route_hdr)
3246 		goto err_reg;
3247 	return 0;
3248 
3249 err_reg:
3250 	if (tbl != ipv4_route_flush_table)
3251 		kfree(tbl);
3252 err_dup:
3253 	return -ENOMEM;
3254 }
3255 
3256 static __net_exit void sysctl_route_net_exit(struct net *net)
3257 {
3258 	struct ctl_table *tbl;
3259 
3260 	tbl = net->ipv4.route_hdr->ctl_table_arg;
3261 	unregister_net_sysctl_table(net->ipv4.route_hdr);
3262 	BUG_ON(tbl == ipv4_route_flush_table);
3263 	kfree(tbl);
3264 }
3265 
3266 static __net_initdata struct pernet_operations sysctl_route_ops = {
3267 	.init = sysctl_route_net_init,
3268 	.exit = sysctl_route_net_exit,
3269 };
3270 #endif
3271 
3272 static __net_init int rt_genid_init(struct net *net)
3273 {
3274 	atomic_set(&net->ipv4.rt_genid, 0);
3275 	atomic_set(&net->fnhe_genid, 0);
3276 	atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3277 	return 0;
3278 }
3279 
3280 static __net_initdata struct pernet_operations rt_genid_ops = {
3281 	.init = rt_genid_init,
3282 };
3283 
3284 static int __net_init ipv4_inetpeer_init(struct net *net)
3285 {
3286 	struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3287 
3288 	if (!bp)
3289 		return -ENOMEM;
3290 	inet_peer_base_init(bp);
3291 	net->ipv4.peers = bp;
3292 	return 0;
3293 }
3294 
3295 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3296 {
3297 	struct inet_peer_base *bp = net->ipv4.peers;
3298 
3299 	net->ipv4.peers = NULL;
3300 	inetpeer_invalidate_tree(bp);
3301 	kfree(bp);
3302 }
3303 
3304 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3305 	.init	=	ipv4_inetpeer_init,
3306 	.exit	=	ipv4_inetpeer_exit,
3307 };
3308 
3309 #ifdef CONFIG_IP_ROUTE_CLASSID
3310 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3311 #endif /* CONFIG_IP_ROUTE_CLASSID */
3312 
3313 int __init ip_rt_init(void)
3314 {
3315 	int cpu;
3316 
3317 	ip_idents = kmalloc_array(IP_IDENTS_SZ, sizeof(*ip_idents),
3318 				  GFP_KERNEL);
3319 	if (!ip_idents)
3320 		panic("IP: failed to allocate ip_idents\n");
3321 
3322 	prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
3323 
3324 	ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
3325 	if (!ip_tstamps)
3326 		panic("IP: failed to allocate ip_tstamps\n");
3327 
3328 	for_each_possible_cpu(cpu) {
3329 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3330 
3331 		INIT_LIST_HEAD(&ul->head);
3332 		spin_lock_init(&ul->lock);
3333 	}
3334 #ifdef CONFIG_IP_ROUTE_CLASSID
3335 	ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3336 	if (!ip_rt_acct)
3337 		panic("IP: failed to allocate ip_rt_acct\n");
3338 #endif
3339 
3340 	ipv4_dst_ops.kmem_cachep =
3341 		kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3342 				  SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3343 
3344 	ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3345 
3346 	if (dst_entries_init(&ipv4_dst_ops) < 0)
3347 		panic("IP: failed to allocate ipv4_dst_ops counter\n");
3348 
3349 	if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3350 		panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3351 
3352 	ipv4_dst_ops.gc_thresh = ~0;
3353 	ip_rt_max_size = INT_MAX;
3354 
3355 	devinet_init();
3356 	ip_fib_init();
3357 
3358 	if (ip_rt_proc_init())
3359 		pr_err("Unable to create route proc files\n");
3360 #ifdef CONFIG_XFRM
3361 	xfrm_init();
3362 	xfrm4_init();
3363 #endif
3364 	rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3365 		      RTNL_FLAG_DOIT_UNLOCKED);
3366 
3367 #ifdef CONFIG_SYSCTL
3368 	register_pernet_subsys(&sysctl_route_ops);
3369 #endif
3370 	register_pernet_subsys(&rt_genid_ops);
3371 	register_pernet_subsys(&ipv4_inetpeer_ops);
3372 	return 0;
3373 }
3374 
3375 #ifdef CONFIG_SYSCTL
3376 /*
3377  * We really need to sanitize the damn ipv4 init order, then all
3378  * this nonsense will go away.
3379  */
3380 void __init ip_static_sysctl_init(void)
3381 {
3382 	register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3383 }
3384 #endif
3385