xref: /linux/net/ipv4/route.c (revision 607975b30db41aad6edc846ed567191aa6b7d893)
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/arp.h>
99 #include <net/tcp.h>
100 #include <net/icmp.h>
101 #include <net/xfrm.h>
102 #include <net/lwtunnel.h>
103 #include <net/netevent.h>
104 #include <net/rtnetlink.h>
105 #ifdef CONFIG_SYSCTL
106 #include <linux/sysctl.h>
107 #endif
108 #include <net/secure_seq.h>
109 #include <net/ip_tunnels.h>
110 #include <net/l3mdev.h>
111 
112 #include "fib_lookup.h"
113 
114 #define RT_FL_TOS(oldflp4) \
115 	((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
116 
117 #define RT_GC_TIMEOUT (300*HZ)
118 
119 static int ip_rt_max_size;
120 static int ip_rt_redirect_number __read_mostly	= 9;
121 static int ip_rt_redirect_load __read_mostly	= HZ / 50;
122 static int ip_rt_redirect_silence __read_mostly	= ((HZ / 50) << (9 + 1));
123 static int ip_rt_error_cost __read_mostly	= HZ;
124 static int ip_rt_error_burst __read_mostly	= 5 * HZ;
125 static int ip_rt_mtu_expires __read_mostly	= 10 * 60 * HZ;
126 static u32 ip_rt_min_pmtu __read_mostly		= 512 + 20 + 20;
127 static int ip_rt_min_advmss __read_mostly	= 256;
128 
129 static int ip_rt_gc_timeout __read_mostly	= RT_GC_TIMEOUT;
130 
131 /*
132  *	Interface to generic destination cache.
133  */
134 
135 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
136 static unsigned int	 ipv4_default_advmss(const struct dst_entry *dst);
137 static unsigned int	 ipv4_mtu(const struct dst_entry *dst);
138 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
139 static void		 ipv4_link_failure(struct sk_buff *skb);
140 static void		 ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
141 					   struct sk_buff *skb, u32 mtu);
142 static void		 ip_do_redirect(struct dst_entry *dst, struct sock *sk,
143 					struct sk_buff *skb);
144 static void		ipv4_dst_destroy(struct dst_entry *dst);
145 
146 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
147 {
148 	WARN_ON(1);
149 	return NULL;
150 }
151 
152 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
153 					   struct sk_buff *skb,
154 					   const void *daddr);
155 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
156 
157 static struct dst_ops ipv4_dst_ops = {
158 	.family =		AF_INET,
159 	.check =		ipv4_dst_check,
160 	.default_advmss =	ipv4_default_advmss,
161 	.mtu =			ipv4_mtu,
162 	.cow_metrics =		ipv4_cow_metrics,
163 	.destroy =		ipv4_dst_destroy,
164 	.negative_advice =	ipv4_negative_advice,
165 	.link_failure =		ipv4_link_failure,
166 	.update_pmtu =		ip_rt_update_pmtu,
167 	.redirect =		ip_do_redirect,
168 	.local_out =		__ip_local_out,
169 	.neigh_lookup =		ipv4_neigh_lookup,
170 	.confirm_neigh =	ipv4_confirm_neigh,
171 };
172 
173 #define ECN_OR_COST(class)	TC_PRIO_##class
174 
175 const __u8 ip_tos2prio[16] = {
176 	TC_PRIO_BESTEFFORT,
177 	ECN_OR_COST(BESTEFFORT),
178 	TC_PRIO_BESTEFFORT,
179 	ECN_OR_COST(BESTEFFORT),
180 	TC_PRIO_BULK,
181 	ECN_OR_COST(BULK),
182 	TC_PRIO_BULK,
183 	ECN_OR_COST(BULK),
184 	TC_PRIO_INTERACTIVE,
185 	ECN_OR_COST(INTERACTIVE),
186 	TC_PRIO_INTERACTIVE,
187 	ECN_OR_COST(INTERACTIVE),
188 	TC_PRIO_INTERACTIVE_BULK,
189 	ECN_OR_COST(INTERACTIVE_BULK),
190 	TC_PRIO_INTERACTIVE_BULK,
191 	ECN_OR_COST(INTERACTIVE_BULK)
192 };
193 EXPORT_SYMBOL(ip_tos2prio);
194 
195 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
196 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
197 
198 #ifdef CONFIG_PROC_FS
199 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
200 {
201 	if (*pos)
202 		return NULL;
203 	return SEQ_START_TOKEN;
204 }
205 
206 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
207 {
208 	++*pos;
209 	return NULL;
210 }
211 
212 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
213 {
214 }
215 
216 static int rt_cache_seq_show(struct seq_file *seq, void *v)
217 {
218 	if (v == SEQ_START_TOKEN)
219 		seq_printf(seq, "%-127s\n",
220 			   "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
221 			   "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
222 			   "HHUptod\tSpecDst");
223 	return 0;
224 }
225 
226 static const struct seq_operations rt_cache_seq_ops = {
227 	.start  = rt_cache_seq_start,
228 	.next   = rt_cache_seq_next,
229 	.stop   = rt_cache_seq_stop,
230 	.show   = rt_cache_seq_show,
231 };
232 
233 static int rt_cache_seq_open(struct inode *inode, struct file *file)
234 {
235 	return seq_open(file, &rt_cache_seq_ops);
236 }
237 
238 static const struct file_operations rt_cache_seq_fops = {
239 	.open	 = rt_cache_seq_open,
240 	.read	 = seq_read,
241 	.llseek	 = seq_lseek,
242 	.release = seq_release,
243 };
244 
245 
246 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
247 {
248 	int cpu;
249 
250 	if (*pos == 0)
251 		return SEQ_START_TOKEN;
252 
253 	for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
254 		if (!cpu_possible(cpu))
255 			continue;
256 		*pos = cpu+1;
257 		return &per_cpu(rt_cache_stat, cpu);
258 	}
259 	return NULL;
260 }
261 
262 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
263 {
264 	int cpu;
265 
266 	for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
267 		if (!cpu_possible(cpu))
268 			continue;
269 		*pos = cpu+1;
270 		return &per_cpu(rt_cache_stat, cpu);
271 	}
272 	return NULL;
273 
274 }
275 
276 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
277 {
278 
279 }
280 
281 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
282 {
283 	struct rt_cache_stat *st = v;
284 
285 	if (v == SEQ_START_TOKEN) {
286 		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");
287 		return 0;
288 	}
289 
290 	seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
291 		   " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
292 		   dst_entries_get_slow(&ipv4_dst_ops),
293 		   0, /* st->in_hit */
294 		   st->in_slow_tot,
295 		   st->in_slow_mc,
296 		   st->in_no_route,
297 		   st->in_brd,
298 		   st->in_martian_dst,
299 		   st->in_martian_src,
300 
301 		   0, /* st->out_hit */
302 		   st->out_slow_tot,
303 		   st->out_slow_mc,
304 
305 		   0, /* st->gc_total */
306 		   0, /* st->gc_ignored */
307 		   0, /* st->gc_goal_miss */
308 		   0, /* st->gc_dst_overflow */
309 		   0, /* st->in_hlist_search */
310 		   0  /* st->out_hlist_search */
311 		);
312 	return 0;
313 }
314 
315 static const struct seq_operations rt_cpu_seq_ops = {
316 	.start  = rt_cpu_seq_start,
317 	.next   = rt_cpu_seq_next,
318 	.stop   = rt_cpu_seq_stop,
319 	.show   = rt_cpu_seq_show,
320 };
321 
322 
323 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
324 {
325 	return seq_open(file, &rt_cpu_seq_ops);
326 }
327 
328 static const struct file_operations rt_cpu_seq_fops = {
329 	.open	 = rt_cpu_seq_open,
330 	.read	 = seq_read,
331 	.llseek	 = seq_lseek,
332 	.release = seq_release,
333 };
334 
335 #ifdef CONFIG_IP_ROUTE_CLASSID
336 static int rt_acct_proc_show(struct seq_file *m, void *v)
337 {
338 	struct ip_rt_acct *dst, *src;
339 	unsigned int i, j;
340 
341 	dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
342 	if (!dst)
343 		return -ENOMEM;
344 
345 	for_each_possible_cpu(i) {
346 		src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
347 		for (j = 0; j < 256; j++) {
348 			dst[j].o_bytes   += src[j].o_bytes;
349 			dst[j].o_packets += src[j].o_packets;
350 			dst[j].i_bytes   += src[j].i_bytes;
351 			dst[j].i_packets += src[j].i_packets;
352 		}
353 	}
354 
355 	seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
356 	kfree(dst);
357 	return 0;
358 }
359 #endif
360 
361 static int __net_init ip_rt_do_proc_init(struct net *net)
362 {
363 	struct proc_dir_entry *pde;
364 
365 	pde = proc_create("rt_cache", 0444, net->proc_net,
366 			  &rt_cache_seq_fops);
367 	if (!pde)
368 		goto err1;
369 
370 	pde = proc_create("rt_cache", 0444,
371 			  net->proc_net_stat, &rt_cpu_seq_fops);
372 	if (!pde)
373 		goto err2;
374 
375 #ifdef CONFIG_IP_ROUTE_CLASSID
376 	pde = proc_create_single("rt_acct", 0, net->proc_net,
377 			rt_acct_proc_show);
378 	if (!pde)
379 		goto err3;
380 #endif
381 	return 0;
382 
383 #ifdef CONFIG_IP_ROUTE_CLASSID
384 err3:
385 	remove_proc_entry("rt_cache", net->proc_net_stat);
386 #endif
387 err2:
388 	remove_proc_entry("rt_cache", net->proc_net);
389 err1:
390 	return -ENOMEM;
391 }
392 
393 static void __net_exit ip_rt_do_proc_exit(struct net *net)
394 {
395 	remove_proc_entry("rt_cache", net->proc_net_stat);
396 	remove_proc_entry("rt_cache", net->proc_net);
397 #ifdef CONFIG_IP_ROUTE_CLASSID
398 	remove_proc_entry("rt_acct", net->proc_net);
399 #endif
400 }
401 
402 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
403 	.init = ip_rt_do_proc_init,
404 	.exit = ip_rt_do_proc_exit,
405 };
406 
407 static int __init ip_rt_proc_init(void)
408 {
409 	return register_pernet_subsys(&ip_rt_proc_ops);
410 }
411 
412 #else
413 static inline int ip_rt_proc_init(void)
414 {
415 	return 0;
416 }
417 #endif /* CONFIG_PROC_FS */
418 
419 static inline bool rt_is_expired(const struct rtable *rth)
420 {
421 	return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
422 }
423 
424 void rt_cache_flush(struct net *net)
425 {
426 	rt_genid_bump_ipv4(net);
427 }
428 
429 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
430 					   struct sk_buff *skb,
431 					   const void *daddr)
432 {
433 	const struct rtable *rt = container_of(dst, struct rtable, dst);
434 	struct net_device *dev = dst->dev;
435 	struct neighbour *n;
436 
437 	rcu_read_lock_bh();
438 
439 	if (likely(rt->rt_gw_family == AF_INET)) {
440 		n = ip_neigh_gw4(dev, rt->rt_gw4);
441 	} else if (rt->rt_gw_family == AF_INET6) {
442 		n = ip_neigh_gw6(dev, &rt->rt_gw6);
443         } else {
444 		__be32 pkey;
445 
446 		pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
447 		n = ip_neigh_gw4(dev, pkey);
448 	}
449 
450 	if (n && !refcount_inc_not_zero(&n->refcnt))
451 		n = NULL;
452 
453 	rcu_read_unlock_bh();
454 
455 	return n;
456 }
457 
458 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
459 {
460 	const struct rtable *rt = container_of(dst, struct rtable, dst);
461 	struct net_device *dev = dst->dev;
462 	const __be32 *pkey = daddr;
463 
464 	if (rt->rt_gw_family == AF_INET) {
465 		pkey = (const __be32 *)&rt->rt_gw4;
466 	} else if (rt->rt_gw_family == AF_INET6) {
467 		return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
468 	} else if (!daddr ||
469 		 (rt->rt_flags &
470 		  (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
471 		return;
472 	}
473 	__ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
474 }
475 
476 #define IP_IDENTS_SZ 2048u
477 
478 static atomic_t *ip_idents __read_mostly;
479 static u32 *ip_tstamps __read_mostly;
480 
481 /* In order to protect privacy, we add a perturbation to identifiers
482  * if one generator is seldom used. This makes hard for an attacker
483  * to infer how many packets were sent between two points in time.
484  */
485 u32 ip_idents_reserve(u32 hash, int segs)
486 {
487 	u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
488 	atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
489 	u32 old = READ_ONCE(*p_tstamp);
490 	u32 now = (u32)jiffies;
491 	u32 new, delta = 0;
492 
493 	if (old != now && cmpxchg(p_tstamp, old, now) == old)
494 		delta = prandom_u32_max(now - old);
495 
496 	/* Do not use atomic_add_return() as it makes UBSAN unhappy */
497 	do {
498 		old = (u32)atomic_read(p_id);
499 		new = old + delta + segs;
500 	} while (atomic_cmpxchg(p_id, old, new) != old);
501 
502 	return new - segs;
503 }
504 EXPORT_SYMBOL(ip_idents_reserve);
505 
506 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
507 {
508 	u32 hash, id;
509 
510 	/* Note the following code is not safe, but this is okay. */
511 	if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
512 		get_random_bytes(&net->ipv4.ip_id_key,
513 				 sizeof(net->ipv4.ip_id_key));
514 
515 	hash = siphash_3u32((__force u32)iph->daddr,
516 			    (__force u32)iph->saddr,
517 			    iph->protocol,
518 			    &net->ipv4.ip_id_key);
519 	id = ip_idents_reserve(hash, segs);
520 	iph->id = htons(id);
521 }
522 EXPORT_SYMBOL(__ip_select_ident);
523 
524 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
525 			     const struct sock *sk,
526 			     const struct iphdr *iph,
527 			     int oif, u8 tos,
528 			     u8 prot, u32 mark, int flow_flags)
529 {
530 	if (sk) {
531 		const struct inet_sock *inet = inet_sk(sk);
532 
533 		oif = sk->sk_bound_dev_if;
534 		mark = sk->sk_mark;
535 		tos = RT_CONN_FLAGS(sk);
536 		prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
537 	}
538 	flowi4_init_output(fl4, oif, mark, tos,
539 			   RT_SCOPE_UNIVERSE, prot,
540 			   flow_flags,
541 			   iph->daddr, iph->saddr, 0, 0,
542 			   sock_net_uid(net, sk));
543 }
544 
545 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
546 			       const struct sock *sk)
547 {
548 	const struct net *net = dev_net(skb->dev);
549 	const struct iphdr *iph = ip_hdr(skb);
550 	int oif = skb->dev->ifindex;
551 	u8 tos = RT_TOS(iph->tos);
552 	u8 prot = iph->protocol;
553 	u32 mark = skb->mark;
554 
555 	__build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
556 }
557 
558 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
559 {
560 	const struct inet_sock *inet = inet_sk(sk);
561 	const struct ip_options_rcu *inet_opt;
562 	__be32 daddr = inet->inet_daddr;
563 
564 	rcu_read_lock();
565 	inet_opt = rcu_dereference(inet->inet_opt);
566 	if (inet_opt && inet_opt->opt.srr)
567 		daddr = inet_opt->opt.faddr;
568 	flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
569 			   RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
570 			   inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
571 			   inet_sk_flowi_flags(sk),
572 			   daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
573 	rcu_read_unlock();
574 }
575 
576 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
577 				 const struct sk_buff *skb)
578 {
579 	if (skb)
580 		build_skb_flow_key(fl4, skb, sk);
581 	else
582 		build_sk_flow_key(fl4, sk);
583 }
584 
585 static DEFINE_SPINLOCK(fnhe_lock);
586 
587 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
588 {
589 	struct rtable *rt;
590 
591 	rt = rcu_dereference(fnhe->fnhe_rth_input);
592 	if (rt) {
593 		RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
594 		dst_dev_put(&rt->dst);
595 		dst_release(&rt->dst);
596 	}
597 	rt = rcu_dereference(fnhe->fnhe_rth_output);
598 	if (rt) {
599 		RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
600 		dst_dev_put(&rt->dst);
601 		dst_release(&rt->dst);
602 	}
603 }
604 
605 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
606 {
607 	struct fib_nh_exception *fnhe, *oldest;
608 
609 	oldest = rcu_dereference(hash->chain);
610 	for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
611 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
612 		if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
613 			oldest = fnhe;
614 	}
615 	fnhe_flush_routes(oldest);
616 	return oldest;
617 }
618 
619 static inline u32 fnhe_hashfun(__be32 daddr)
620 {
621 	static u32 fnhe_hashrnd __read_mostly;
622 	u32 hval;
623 
624 	net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
625 	hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
626 	return hash_32(hval, FNHE_HASH_SHIFT);
627 }
628 
629 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
630 {
631 	rt->rt_pmtu = fnhe->fnhe_pmtu;
632 	rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
633 	rt->dst.expires = fnhe->fnhe_expires;
634 
635 	if (fnhe->fnhe_gw) {
636 		rt->rt_flags |= RTCF_REDIRECTED;
637 		rt->rt_gw_family = AF_INET;
638 		rt->rt_gw4 = fnhe->fnhe_gw;
639 	}
640 }
641 
642 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
643 				  __be32 gw, u32 pmtu, bool lock,
644 				  unsigned long expires)
645 {
646 	struct fnhe_hash_bucket *hash;
647 	struct fib_nh_exception *fnhe;
648 	struct rtable *rt;
649 	u32 genid, hval;
650 	unsigned int i;
651 	int depth;
652 
653 	genid = fnhe_genid(dev_net(nhc->nhc_dev));
654 	hval = fnhe_hashfun(daddr);
655 
656 	spin_lock_bh(&fnhe_lock);
657 
658 	hash = rcu_dereference(nhc->nhc_exceptions);
659 	if (!hash) {
660 		hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
661 		if (!hash)
662 			goto out_unlock;
663 		rcu_assign_pointer(nhc->nhc_exceptions, hash);
664 	}
665 
666 	hash += hval;
667 
668 	depth = 0;
669 	for (fnhe = rcu_dereference(hash->chain); fnhe;
670 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
671 		if (fnhe->fnhe_daddr == daddr)
672 			break;
673 		depth++;
674 	}
675 
676 	if (fnhe) {
677 		if (fnhe->fnhe_genid != genid)
678 			fnhe->fnhe_genid = genid;
679 		if (gw)
680 			fnhe->fnhe_gw = gw;
681 		if (pmtu) {
682 			fnhe->fnhe_pmtu = pmtu;
683 			fnhe->fnhe_mtu_locked = lock;
684 		}
685 		fnhe->fnhe_expires = max(1UL, expires);
686 		/* Update all cached dsts too */
687 		rt = rcu_dereference(fnhe->fnhe_rth_input);
688 		if (rt)
689 			fill_route_from_fnhe(rt, fnhe);
690 		rt = rcu_dereference(fnhe->fnhe_rth_output);
691 		if (rt)
692 			fill_route_from_fnhe(rt, fnhe);
693 	} else {
694 		if (depth > FNHE_RECLAIM_DEPTH)
695 			fnhe = fnhe_oldest(hash);
696 		else {
697 			fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
698 			if (!fnhe)
699 				goto out_unlock;
700 
701 			fnhe->fnhe_next = hash->chain;
702 			rcu_assign_pointer(hash->chain, fnhe);
703 		}
704 		fnhe->fnhe_genid = genid;
705 		fnhe->fnhe_daddr = daddr;
706 		fnhe->fnhe_gw = gw;
707 		fnhe->fnhe_pmtu = pmtu;
708 		fnhe->fnhe_mtu_locked = lock;
709 		fnhe->fnhe_expires = max(1UL, expires);
710 
711 		/* Exception created; mark the cached routes for the nexthop
712 		 * stale, so anyone caching it rechecks if this exception
713 		 * applies to them.
714 		 */
715 		rt = rcu_dereference(nhc->nhc_rth_input);
716 		if (rt)
717 			rt->dst.obsolete = DST_OBSOLETE_KILL;
718 
719 		for_each_possible_cpu(i) {
720 			struct rtable __rcu **prt;
721 			prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
722 			rt = rcu_dereference(*prt);
723 			if (rt)
724 				rt->dst.obsolete = DST_OBSOLETE_KILL;
725 		}
726 	}
727 
728 	fnhe->fnhe_stamp = jiffies;
729 
730 out_unlock:
731 	spin_unlock_bh(&fnhe_lock);
732 }
733 
734 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
735 			     bool kill_route)
736 {
737 	__be32 new_gw = icmp_hdr(skb)->un.gateway;
738 	__be32 old_gw = ip_hdr(skb)->saddr;
739 	struct net_device *dev = skb->dev;
740 	struct in_device *in_dev;
741 	struct fib_result res;
742 	struct neighbour *n;
743 	struct net *net;
744 
745 	switch (icmp_hdr(skb)->code & 7) {
746 	case ICMP_REDIR_NET:
747 	case ICMP_REDIR_NETTOS:
748 	case ICMP_REDIR_HOST:
749 	case ICMP_REDIR_HOSTTOS:
750 		break;
751 
752 	default:
753 		return;
754 	}
755 
756 	if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
757 		return;
758 
759 	in_dev = __in_dev_get_rcu(dev);
760 	if (!in_dev)
761 		return;
762 
763 	net = dev_net(dev);
764 	if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
765 	    ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
766 	    ipv4_is_zeronet(new_gw))
767 		goto reject_redirect;
768 
769 	if (!IN_DEV_SHARED_MEDIA(in_dev)) {
770 		if (!inet_addr_onlink(in_dev, new_gw, old_gw))
771 			goto reject_redirect;
772 		if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
773 			goto reject_redirect;
774 	} else {
775 		if (inet_addr_type(net, new_gw) != RTN_UNICAST)
776 			goto reject_redirect;
777 	}
778 
779 	n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
780 	if (!n)
781 		n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
782 	if (!IS_ERR(n)) {
783 		if (!(n->nud_state & NUD_VALID)) {
784 			neigh_event_send(n, NULL);
785 		} else {
786 			if (fib_lookup(net, fl4, &res, 0) == 0) {
787 				struct fib_nh_common *nhc = FIB_RES_NHC(res);
788 
789 				update_or_create_fnhe(nhc, fl4->daddr, new_gw,
790 						0, false,
791 						jiffies + ip_rt_gc_timeout);
792 			}
793 			if (kill_route)
794 				rt->dst.obsolete = DST_OBSOLETE_KILL;
795 			call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
796 		}
797 		neigh_release(n);
798 	}
799 	return;
800 
801 reject_redirect:
802 #ifdef CONFIG_IP_ROUTE_VERBOSE
803 	if (IN_DEV_LOG_MARTIANS(in_dev)) {
804 		const struct iphdr *iph = (const struct iphdr *) skb->data;
805 		__be32 daddr = iph->daddr;
806 		__be32 saddr = iph->saddr;
807 
808 		net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
809 				     "  Advised path = %pI4 -> %pI4\n",
810 				     &old_gw, dev->name, &new_gw,
811 				     &saddr, &daddr);
812 	}
813 #endif
814 	;
815 }
816 
817 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
818 {
819 	struct rtable *rt;
820 	struct flowi4 fl4;
821 	const struct iphdr *iph = (const struct iphdr *) skb->data;
822 	struct net *net = dev_net(skb->dev);
823 	int oif = skb->dev->ifindex;
824 	u8 tos = RT_TOS(iph->tos);
825 	u8 prot = iph->protocol;
826 	u32 mark = skb->mark;
827 
828 	rt = (struct rtable *) dst;
829 
830 	__build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
831 	__ip_do_redirect(rt, skb, &fl4, true);
832 }
833 
834 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
835 {
836 	struct rtable *rt = (struct rtable *)dst;
837 	struct dst_entry *ret = dst;
838 
839 	if (rt) {
840 		if (dst->obsolete > 0) {
841 			ip_rt_put(rt);
842 			ret = NULL;
843 		} else if ((rt->rt_flags & RTCF_REDIRECTED) ||
844 			   rt->dst.expires) {
845 			ip_rt_put(rt);
846 			ret = NULL;
847 		}
848 	}
849 	return ret;
850 }
851 
852 /*
853  * Algorithm:
854  *	1. The first ip_rt_redirect_number redirects are sent
855  *	   with exponential backoff, then we stop sending them at all,
856  *	   assuming that the host ignores our redirects.
857  *	2. If we did not see packets requiring redirects
858  *	   during ip_rt_redirect_silence, we assume that the host
859  *	   forgot redirected route and start to send redirects again.
860  *
861  * This algorithm is much cheaper and more intelligent than dumb load limiting
862  * in icmp.c.
863  *
864  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
865  * and "frag. need" (breaks PMTU discovery) in icmp.c.
866  */
867 
868 void ip_rt_send_redirect(struct sk_buff *skb)
869 {
870 	struct rtable *rt = skb_rtable(skb);
871 	struct in_device *in_dev;
872 	struct inet_peer *peer;
873 	struct net *net;
874 	int log_martians;
875 	int vif;
876 
877 	rcu_read_lock();
878 	in_dev = __in_dev_get_rcu(rt->dst.dev);
879 	if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
880 		rcu_read_unlock();
881 		return;
882 	}
883 	log_martians = IN_DEV_LOG_MARTIANS(in_dev);
884 	vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
885 	rcu_read_unlock();
886 
887 	net = dev_net(rt->dst.dev);
888 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
889 	if (!peer) {
890 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
891 			  rt_nexthop(rt, ip_hdr(skb)->daddr));
892 		return;
893 	}
894 
895 	/* No redirected packets during ip_rt_redirect_silence;
896 	 * reset the algorithm.
897 	 */
898 	if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
899 		peer->rate_tokens = 0;
900 		peer->n_redirects = 0;
901 	}
902 
903 	/* Too many ignored redirects; do not send anything
904 	 * set dst.rate_last to the last seen redirected packet.
905 	 */
906 	if (peer->n_redirects >= ip_rt_redirect_number) {
907 		peer->rate_last = jiffies;
908 		goto out_put_peer;
909 	}
910 
911 	/* Check for load limit; set rate_last to the latest sent
912 	 * redirect.
913 	 */
914 	if (peer->rate_tokens == 0 ||
915 	    time_after(jiffies,
916 		       (peer->rate_last +
917 			(ip_rt_redirect_load << peer->rate_tokens)))) {
918 		__be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
919 
920 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
921 		peer->rate_last = jiffies;
922 		++peer->rate_tokens;
923 		++peer->n_redirects;
924 #ifdef CONFIG_IP_ROUTE_VERBOSE
925 		if (log_martians &&
926 		    peer->rate_tokens == ip_rt_redirect_number)
927 			net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
928 					     &ip_hdr(skb)->saddr, inet_iif(skb),
929 					     &ip_hdr(skb)->daddr, &gw);
930 #endif
931 	}
932 out_put_peer:
933 	inet_putpeer(peer);
934 }
935 
936 static int ip_error(struct sk_buff *skb)
937 {
938 	struct rtable *rt = skb_rtable(skb);
939 	struct net_device *dev = skb->dev;
940 	struct in_device *in_dev;
941 	struct inet_peer *peer;
942 	unsigned long now;
943 	struct net *net;
944 	bool send;
945 	int code;
946 
947 	if (netif_is_l3_master(skb->dev)) {
948 		dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
949 		if (!dev)
950 			goto out;
951 	}
952 
953 	in_dev = __in_dev_get_rcu(dev);
954 
955 	/* IP on this device is disabled. */
956 	if (!in_dev)
957 		goto out;
958 
959 	net = dev_net(rt->dst.dev);
960 	if (!IN_DEV_FORWARD(in_dev)) {
961 		switch (rt->dst.error) {
962 		case EHOSTUNREACH:
963 			__IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
964 			break;
965 
966 		case ENETUNREACH:
967 			__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
968 			break;
969 		}
970 		goto out;
971 	}
972 
973 	switch (rt->dst.error) {
974 	case EINVAL:
975 	default:
976 		goto out;
977 	case EHOSTUNREACH:
978 		code = ICMP_HOST_UNREACH;
979 		break;
980 	case ENETUNREACH:
981 		code = ICMP_NET_UNREACH;
982 		__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
983 		break;
984 	case EACCES:
985 		code = ICMP_PKT_FILTERED;
986 		break;
987 	}
988 
989 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
990 			       l3mdev_master_ifindex(skb->dev), 1);
991 
992 	send = true;
993 	if (peer) {
994 		now = jiffies;
995 		peer->rate_tokens += now - peer->rate_last;
996 		if (peer->rate_tokens > ip_rt_error_burst)
997 			peer->rate_tokens = ip_rt_error_burst;
998 		peer->rate_last = now;
999 		if (peer->rate_tokens >= ip_rt_error_cost)
1000 			peer->rate_tokens -= ip_rt_error_cost;
1001 		else
1002 			send = false;
1003 		inet_putpeer(peer);
1004 	}
1005 	if (send)
1006 		icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1007 
1008 out:	kfree_skb(skb);
1009 	return 0;
1010 }
1011 
1012 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1013 {
1014 	struct dst_entry *dst = &rt->dst;
1015 	u32 old_mtu = ipv4_mtu(dst);
1016 	struct fib_result res;
1017 	bool lock = false;
1018 
1019 	if (ip_mtu_locked(dst))
1020 		return;
1021 
1022 	if (old_mtu < mtu)
1023 		return;
1024 
1025 	if (mtu < ip_rt_min_pmtu) {
1026 		lock = true;
1027 		mtu = min(old_mtu, ip_rt_min_pmtu);
1028 	}
1029 
1030 	if (rt->rt_pmtu == mtu && !lock &&
1031 	    time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1032 		return;
1033 
1034 	rcu_read_lock();
1035 	if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
1036 		struct fib_nh_common *nhc = FIB_RES_NHC(res);
1037 
1038 		update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1039 				      jiffies + ip_rt_mtu_expires);
1040 	}
1041 	rcu_read_unlock();
1042 }
1043 
1044 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1045 			      struct sk_buff *skb, u32 mtu)
1046 {
1047 	struct rtable *rt = (struct rtable *) dst;
1048 	struct flowi4 fl4;
1049 
1050 	ip_rt_build_flow_key(&fl4, sk, skb);
1051 	__ip_rt_update_pmtu(rt, &fl4, mtu);
1052 }
1053 
1054 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1055 		      int oif, u8 protocol)
1056 {
1057 	const struct iphdr *iph = (const struct iphdr *) skb->data;
1058 	struct flowi4 fl4;
1059 	struct rtable *rt;
1060 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
1061 
1062 	__build_flow_key(net, &fl4, NULL, iph, oif,
1063 			 RT_TOS(iph->tos), protocol, mark, 0);
1064 	rt = __ip_route_output_key(net, &fl4);
1065 	if (!IS_ERR(rt)) {
1066 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1067 		ip_rt_put(rt);
1068 	}
1069 }
1070 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1071 
1072 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1073 {
1074 	const struct iphdr *iph = (const struct iphdr *) skb->data;
1075 	struct flowi4 fl4;
1076 	struct rtable *rt;
1077 
1078 	__build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1079 
1080 	if (!fl4.flowi4_mark)
1081 		fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1082 
1083 	rt = __ip_route_output_key(sock_net(sk), &fl4);
1084 	if (!IS_ERR(rt)) {
1085 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1086 		ip_rt_put(rt);
1087 	}
1088 }
1089 
1090 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1091 {
1092 	const struct iphdr *iph = (const struct iphdr *) skb->data;
1093 	struct flowi4 fl4;
1094 	struct rtable *rt;
1095 	struct dst_entry *odst = NULL;
1096 	bool new = false;
1097 	struct net *net = sock_net(sk);
1098 
1099 	bh_lock_sock(sk);
1100 
1101 	if (!ip_sk_accept_pmtu(sk))
1102 		goto out;
1103 
1104 	odst = sk_dst_get(sk);
1105 
1106 	if (sock_owned_by_user(sk) || !odst) {
1107 		__ipv4_sk_update_pmtu(skb, sk, mtu);
1108 		goto out;
1109 	}
1110 
1111 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1112 
1113 	rt = (struct rtable *)odst;
1114 	if (odst->obsolete && !odst->ops->check(odst, 0)) {
1115 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1116 		if (IS_ERR(rt))
1117 			goto out;
1118 
1119 		new = true;
1120 	}
1121 
1122 	__ip_rt_update_pmtu((struct rtable *) xfrm_dst_path(&rt->dst), &fl4, mtu);
1123 
1124 	if (!dst_check(&rt->dst, 0)) {
1125 		if (new)
1126 			dst_release(&rt->dst);
1127 
1128 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1129 		if (IS_ERR(rt))
1130 			goto out;
1131 
1132 		new = true;
1133 	}
1134 
1135 	if (new)
1136 		sk_dst_set(sk, &rt->dst);
1137 
1138 out:
1139 	bh_unlock_sock(sk);
1140 	dst_release(odst);
1141 }
1142 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1143 
1144 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1145 		   int oif, u8 protocol)
1146 {
1147 	const struct iphdr *iph = (const struct iphdr *) skb->data;
1148 	struct flowi4 fl4;
1149 	struct rtable *rt;
1150 
1151 	__build_flow_key(net, &fl4, NULL, iph, oif,
1152 			 RT_TOS(iph->tos), protocol, 0, 0);
1153 	rt = __ip_route_output_key(net, &fl4);
1154 	if (!IS_ERR(rt)) {
1155 		__ip_do_redirect(rt, skb, &fl4, false);
1156 		ip_rt_put(rt);
1157 	}
1158 }
1159 EXPORT_SYMBOL_GPL(ipv4_redirect);
1160 
1161 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1162 {
1163 	const struct iphdr *iph = (const struct iphdr *) skb->data;
1164 	struct flowi4 fl4;
1165 	struct rtable *rt;
1166 	struct net *net = sock_net(sk);
1167 
1168 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1169 	rt = __ip_route_output_key(net, &fl4);
1170 	if (!IS_ERR(rt)) {
1171 		__ip_do_redirect(rt, skb, &fl4, false);
1172 		ip_rt_put(rt);
1173 	}
1174 }
1175 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1176 
1177 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1178 {
1179 	struct rtable *rt = (struct rtable *) dst;
1180 
1181 	/* All IPV4 dsts are created with ->obsolete set to the value
1182 	 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1183 	 * into this function always.
1184 	 *
1185 	 * When a PMTU/redirect information update invalidates a route,
1186 	 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1187 	 * DST_OBSOLETE_DEAD.
1188 	 */
1189 	if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1190 		return NULL;
1191 	return dst;
1192 }
1193 
1194 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1195 {
1196 	struct ip_options opt;
1197 	int res;
1198 
1199 	/* Recompile ip options since IPCB may not be valid anymore.
1200 	 * Also check we have a reasonable ipv4 header.
1201 	 */
1202 	if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1203 	    ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1204 		return;
1205 
1206 	memset(&opt, 0, sizeof(opt));
1207 	if (ip_hdr(skb)->ihl > 5) {
1208 		if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1209 			return;
1210 		opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1211 
1212 		rcu_read_lock();
1213 		res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1214 		rcu_read_unlock();
1215 
1216 		if (res)
1217 			return;
1218 	}
1219 	__icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1220 }
1221 
1222 static void ipv4_link_failure(struct sk_buff *skb)
1223 {
1224 	struct rtable *rt;
1225 
1226 	ipv4_send_dest_unreach(skb);
1227 
1228 	rt = skb_rtable(skb);
1229 	if (rt)
1230 		dst_set_expires(&rt->dst, 0);
1231 }
1232 
1233 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1234 {
1235 	pr_debug("%s: %pI4 -> %pI4, %s\n",
1236 		 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1237 		 skb->dev ? skb->dev->name : "?");
1238 	kfree_skb(skb);
1239 	WARN_ON(1);
1240 	return 0;
1241 }
1242 
1243 /*
1244    We do not cache source address of outgoing interface,
1245    because it is used only by IP RR, TS and SRR options,
1246    so that it out of fast path.
1247 
1248    BTW remember: "addr" is allowed to be not aligned
1249    in IP options!
1250  */
1251 
1252 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1253 {
1254 	__be32 src;
1255 
1256 	if (rt_is_output_route(rt))
1257 		src = ip_hdr(skb)->saddr;
1258 	else {
1259 		struct fib_result res;
1260 		struct iphdr *iph = ip_hdr(skb);
1261 		struct flowi4 fl4 = {
1262 			.daddr = iph->daddr,
1263 			.saddr = iph->saddr,
1264 			.flowi4_tos = RT_TOS(iph->tos),
1265 			.flowi4_oif = rt->dst.dev->ifindex,
1266 			.flowi4_iif = skb->dev->ifindex,
1267 			.flowi4_mark = skb->mark,
1268 		};
1269 
1270 		rcu_read_lock();
1271 		if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1272 			src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1273 		else
1274 			src = inet_select_addr(rt->dst.dev,
1275 					       rt_nexthop(rt, iph->daddr),
1276 					       RT_SCOPE_UNIVERSE);
1277 		rcu_read_unlock();
1278 	}
1279 	memcpy(addr, &src, 4);
1280 }
1281 
1282 #ifdef CONFIG_IP_ROUTE_CLASSID
1283 static void set_class_tag(struct rtable *rt, u32 tag)
1284 {
1285 	if (!(rt->dst.tclassid & 0xFFFF))
1286 		rt->dst.tclassid |= tag & 0xFFFF;
1287 	if (!(rt->dst.tclassid & 0xFFFF0000))
1288 		rt->dst.tclassid |= tag & 0xFFFF0000;
1289 }
1290 #endif
1291 
1292 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1293 {
1294 	unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1295 	unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1296 				    ip_rt_min_advmss);
1297 
1298 	return min(advmss, IPV4_MAX_PMTU - header_size);
1299 }
1300 
1301 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1302 {
1303 	const struct rtable *rt = (const struct rtable *) dst;
1304 	unsigned int mtu = rt->rt_pmtu;
1305 
1306 	if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1307 		mtu = dst_metric_raw(dst, RTAX_MTU);
1308 
1309 	if (mtu)
1310 		return mtu;
1311 
1312 	mtu = READ_ONCE(dst->dev->mtu);
1313 
1314 	if (unlikely(ip_mtu_locked(dst))) {
1315 		if (rt->rt_gw_family && mtu > 576)
1316 			mtu = 576;
1317 	}
1318 
1319 	mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1320 
1321 	return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1322 }
1323 
1324 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1325 {
1326 	struct fnhe_hash_bucket *hash;
1327 	struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1328 	u32 hval = fnhe_hashfun(daddr);
1329 
1330 	spin_lock_bh(&fnhe_lock);
1331 
1332 	hash = rcu_dereference_protected(nhc->nhc_exceptions,
1333 					 lockdep_is_held(&fnhe_lock));
1334 	hash += hval;
1335 
1336 	fnhe_p = &hash->chain;
1337 	fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1338 	while (fnhe) {
1339 		if (fnhe->fnhe_daddr == daddr) {
1340 			rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1341 				fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1342 			/* set fnhe_daddr to 0 to ensure it won't bind with
1343 			 * new dsts in rt_bind_exception().
1344 			 */
1345 			fnhe->fnhe_daddr = 0;
1346 			fnhe_flush_routes(fnhe);
1347 			kfree_rcu(fnhe, rcu);
1348 			break;
1349 		}
1350 		fnhe_p = &fnhe->fnhe_next;
1351 		fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1352 						 lockdep_is_held(&fnhe_lock));
1353 	}
1354 
1355 	spin_unlock_bh(&fnhe_lock);
1356 }
1357 
1358 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1359 					       __be32 daddr)
1360 {
1361 	struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1362 	struct fib_nh_exception *fnhe;
1363 	u32 hval;
1364 
1365 	if (!hash)
1366 		return NULL;
1367 
1368 	hval = fnhe_hashfun(daddr);
1369 
1370 	for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1371 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
1372 		if (fnhe->fnhe_daddr == daddr) {
1373 			if (fnhe->fnhe_expires &&
1374 			    time_after(jiffies, fnhe->fnhe_expires)) {
1375 				ip_del_fnhe(nhc, daddr);
1376 				break;
1377 			}
1378 			return fnhe;
1379 		}
1380 	}
1381 	return NULL;
1382 }
1383 
1384 /* MTU selection:
1385  * 1. mtu on route is locked - use it
1386  * 2. mtu from nexthop exception
1387  * 3. mtu from egress device
1388  */
1389 
1390 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1391 {
1392 	struct fib_nh_common *nhc = res->nhc;
1393 	struct net_device *dev = nhc->nhc_dev;
1394 	struct fib_info *fi = res->fi;
1395 	u32 mtu = 0;
1396 
1397 	if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu ||
1398 	    fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1399 		mtu = fi->fib_mtu;
1400 
1401 	if (likely(!mtu)) {
1402 		struct fib_nh_exception *fnhe;
1403 
1404 		fnhe = find_exception(nhc, daddr);
1405 		if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1406 			mtu = fnhe->fnhe_pmtu;
1407 	}
1408 
1409 	if (likely(!mtu))
1410 		mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1411 
1412 	return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1413 }
1414 
1415 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1416 			      __be32 daddr, const bool do_cache)
1417 {
1418 	bool ret = false;
1419 
1420 	spin_lock_bh(&fnhe_lock);
1421 
1422 	if (daddr == fnhe->fnhe_daddr) {
1423 		struct rtable __rcu **porig;
1424 		struct rtable *orig;
1425 		int genid = fnhe_genid(dev_net(rt->dst.dev));
1426 
1427 		if (rt_is_input_route(rt))
1428 			porig = &fnhe->fnhe_rth_input;
1429 		else
1430 			porig = &fnhe->fnhe_rth_output;
1431 		orig = rcu_dereference(*porig);
1432 
1433 		if (fnhe->fnhe_genid != genid) {
1434 			fnhe->fnhe_genid = genid;
1435 			fnhe->fnhe_gw = 0;
1436 			fnhe->fnhe_pmtu = 0;
1437 			fnhe->fnhe_expires = 0;
1438 			fnhe->fnhe_mtu_locked = false;
1439 			fnhe_flush_routes(fnhe);
1440 			orig = NULL;
1441 		}
1442 		fill_route_from_fnhe(rt, fnhe);
1443 		if (!rt->rt_gw4) {
1444 			rt->rt_gw4 = daddr;
1445 			rt->rt_gw_family = AF_INET;
1446 		}
1447 
1448 		if (do_cache) {
1449 			dst_hold(&rt->dst);
1450 			rcu_assign_pointer(*porig, rt);
1451 			if (orig) {
1452 				dst_dev_put(&orig->dst);
1453 				dst_release(&orig->dst);
1454 			}
1455 			ret = true;
1456 		}
1457 
1458 		fnhe->fnhe_stamp = jiffies;
1459 	}
1460 	spin_unlock_bh(&fnhe_lock);
1461 
1462 	return ret;
1463 }
1464 
1465 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1466 {
1467 	struct rtable *orig, *prev, **p;
1468 	bool ret = true;
1469 
1470 	if (rt_is_input_route(rt)) {
1471 		p = (struct rtable **)&nhc->nhc_rth_input;
1472 	} else {
1473 		p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1474 	}
1475 	orig = *p;
1476 
1477 	/* hold dst before doing cmpxchg() to avoid race condition
1478 	 * on this dst
1479 	 */
1480 	dst_hold(&rt->dst);
1481 	prev = cmpxchg(p, orig, rt);
1482 	if (prev == orig) {
1483 		if (orig) {
1484 			dst_dev_put(&orig->dst);
1485 			dst_release(&orig->dst);
1486 		}
1487 	} else {
1488 		dst_release(&rt->dst);
1489 		ret = false;
1490 	}
1491 
1492 	return ret;
1493 }
1494 
1495 struct uncached_list {
1496 	spinlock_t		lock;
1497 	struct list_head	head;
1498 };
1499 
1500 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1501 
1502 void rt_add_uncached_list(struct rtable *rt)
1503 {
1504 	struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1505 
1506 	rt->rt_uncached_list = ul;
1507 
1508 	spin_lock_bh(&ul->lock);
1509 	list_add_tail(&rt->rt_uncached, &ul->head);
1510 	spin_unlock_bh(&ul->lock);
1511 }
1512 
1513 void rt_del_uncached_list(struct rtable *rt)
1514 {
1515 	if (!list_empty(&rt->rt_uncached)) {
1516 		struct uncached_list *ul = rt->rt_uncached_list;
1517 
1518 		spin_lock_bh(&ul->lock);
1519 		list_del(&rt->rt_uncached);
1520 		spin_unlock_bh(&ul->lock);
1521 	}
1522 }
1523 
1524 static void ipv4_dst_destroy(struct dst_entry *dst)
1525 {
1526 	struct rtable *rt = (struct rtable *)dst;
1527 
1528 	ip_dst_metrics_put(dst);
1529 	rt_del_uncached_list(rt);
1530 }
1531 
1532 void rt_flush_dev(struct net_device *dev)
1533 {
1534 	struct net *net = dev_net(dev);
1535 	struct rtable *rt;
1536 	int cpu;
1537 
1538 	for_each_possible_cpu(cpu) {
1539 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1540 
1541 		spin_lock_bh(&ul->lock);
1542 		list_for_each_entry(rt, &ul->head, rt_uncached) {
1543 			if (rt->dst.dev != dev)
1544 				continue;
1545 			rt->dst.dev = net->loopback_dev;
1546 			dev_hold(rt->dst.dev);
1547 			dev_put(dev);
1548 		}
1549 		spin_unlock_bh(&ul->lock);
1550 	}
1551 }
1552 
1553 static bool rt_cache_valid(const struct rtable *rt)
1554 {
1555 	return	rt &&
1556 		rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1557 		!rt_is_expired(rt);
1558 }
1559 
1560 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1561 			   const struct fib_result *res,
1562 			   struct fib_nh_exception *fnhe,
1563 			   struct fib_info *fi, u16 type, u32 itag,
1564 			   const bool do_cache)
1565 {
1566 	bool cached = false;
1567 
1568 	if (fi) {
1569 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1570 
1571 		if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1572 			rt->rt_gw_family = nhc->nhc_gw_family;
1573 			/* only INET and INET6 are supported */
1574 			if (likely(nhc->nhc_gw_family == AF_INET))
1575 				rt->rt_gw4 = nhc->nhc_gw.ipv4;
1576 			else
1577 				rt->rt_gw6 = nhc->nhc_gw.ipv6;
1578 		}
1579 
1580 		ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1581 
1582 #ifdef CONFIG_IP_ROUTE_CLASSID
1583 		{
1584 			struct fib_nh *nh;
1585 
1586 			nh = container_of(nhc, struct fib_nh, nh_common);
1587 			rt->dst.tclassid = nh->nh_tclassid;
1588 		}
1589 #endif
1590 		rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1591 		if (unlikely(fnhe))
1592 			cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1593 		else if (do_cache)
1594 			cached = rt_cache_route(nhc, rt);
1595 		if (unlikely(!cached)) {
1596 			/* Routes we intend to cache in nexthop exception or
1597 			 * FIB nexthop have the DST_NOCACHE bit clear.
1598 			 * However, if we are unsuccessful at storing this
1599 			 * route into the cache we really need to set it.
1600 			 */
1601 			if (!rt->rt_gw4) {
1602 				rt->rt_gw_family = AF_INET;
1603 				rt->rt_gw4 = daddr;
1604 			}
1605 			rt_add_uncached_list(rt);
1606 		}
1607 	} else
1608 		rt_add_uncached_list(rt);
1609 
1610 #ifdef CONFIG_IP_ROUTE_CLASSID
1611 #ifdef CONFIG_IP_MULTIPLE_TABLES
1612 	set_class_tag(rt, res->tclassid);
1613 #endif
1614 	set_class_tag(rt, itag);
1615 #endif
1616 }
1617 
1618 struct rtable *rt_dst_alloc(struct net_device *dev,
1619 			    unsigned int flags, u16 type,
1620 			    bool nopolicy, bool noxfrm, bool will_cache)
1621 {
1622 	struct rtable *rt;
1623 
1624 	rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1625 		       (will_cache ? 0 : DST_HOST) |
1626 		       (nopolicy ? DST_NOPOLICY : 0) |
1627 		       (noxfrm ? DST_NOXFRM : 0));
1628 
1629 	if (rt) {
1630 		rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1631 		rt->rt_flags = flags;
1632 		rt->rt_type = type;
1633 		rt->rt_is_input = 0;
1634 		rt->rt_iif = 0;
1635 		rt->rt_pmtu = 0;
1636 		rt->rt_mtu_locked = 0;
1637 		rt->rt_gw_family = 0;
1638 		rt->rt_gw4 = 0;
1639 		INIT_LIST_HEAD(&rt->rt_uncached);
1640 
1641 		rt->dst.output = ip_output;
1642 		if (flags & RTCF_LOCAL)
1643 			rt->dst.input = ip_local_deliver;
1644 	}
1645 
1646 	return rt;
1647 }
1648 EXPORT_SYMBOL(rt_dst_alloc);
1649 
1650 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1651 {
1652 	struct rtable *new_rt;
1653 
1654 	new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1655 			   rt->dst.flags);
1656 
1657 	if (new_rt) {
1658 		new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1659 		new_rt->rt_flags = rt->rt_flags;
1660 		new_rt->rt_type = rt->rt_type;
1661 		new_rt->rt_is_input = rt->rt_is_input;
1662 		new_rt->rt_iif = rt->rt_iif;
1663 		new_rt->rt_pmtu = rt->rt_pmtu;
1664 		new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1665 		new_rt->rt_gw_family = rt->rt_gw_family;
1666 		if (rt->rt_gw_family == AF_INET)
1667 			new_rt->rt_gw4 = rt->rt_gw4;
1668 		else if (rt->rt_gw_family == AF_INET6)
1669 			new_rt->rt_gw6 = rt->rt_gw6;
1670 		INIT_LIST_HEAD(&new_rt->rt_uncached);
1671 
1672 		new_rt->dst.flags |= DST_HOST;
1673 		new_rt->dst.input = rt->dst.input;
1674 		new_rt->dst.output = rt->dst.output;
1675 		new_rt->dst.error = rt->dst.error;
1676 		new_rt->dst.lastuse = jiffies;
1677 		new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1678 	}
1679 	return new_rt;
1680 }
1681 EXPORT_SYMBOL(rt_dst_clone);
1682 
1683 /* called in rcu_read_lock() section */
1684 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1685 			  u8 tos, struct net_device *dev,
1686 			  struct in_device *in_dev, u32 *itag)
1687 {
1688 	int err;
1689 
1690 	/* Primary sanity checks. */
1691 	if (!in_dev)
1692 		return -EINVAL;
1693 
1694 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1695 	    skb->protocol != htons(ETH_P_IP))
1696 		return -EINVAL;
1697 
1698 	if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1699 		return -EINVAL;
1700 
1701 	if (ipv4_is_zeronet(saddr)) {
1702 		if (!ipv4_is_local_multicast(daddr) &&
1703 		    ip_hdr(skb)->protocol != IPPROTO_IGMP)
1704 			return -EINVAL;
1705 	} else {
1706 		err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1707 					  in_dev, itag);
1708 		if (err < 0)
1709 			return err;
1710 	}
1711 	return 0;
1712 }
1713 
1714 /* called in rcu_read_lock() section */
1715 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1716 			     u8 tos, struct net_device *dev, int our)
1717 {
1718 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1719 	unsigned int flags = RTCF_MULTICAST;
1720 	struct rtable *rth;
1721 	u32 itag = 0;
1722 	int err;
1723 
1724 	err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1725 	if (err)
1726 		return err;
1727 
1728 	if (our)
1729 		flags |= RTCF_LOCAL;
1730 
1731 	rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1732 			   IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1733 	if (!rth)
1734 		return -ENOBUFS;
1735 
1736 #ifdef CONFIG_IP_ROUTE_CLASSID
1737 	rth->dst.tclassid = itag;
1738 #endif
1739 	rth->dst.output = ip_rt_bug;
1740 	rth->rt_is_input= 1;
1741 
1742 #ifdef CONFIG_IP_MROUTE
1743 	if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1744 		rth->dst.input = ip_mr_input;
1745 #endif
1746 	RT_CACHE_STAT_INC(in_slow_mc);
1747 
1748 	skb_dst_set(skb, &rth->dst);
1749 	return 0;
1750 }
1751 
1752 
1753 static void ip_handle_martian_source(struct net_device *dev,
1754 				     struct in_device *in_dev,
1755 				     struct sk_buff *skb,
1756 				     __be32 daddr,
1757 				     __be32 saddr)
1758 {
1759 	RT_CACHE_STAT_INC(in_martian_src);
1760 #ifdef CONFIG_IP_ROUTE_VERBOSE
1761 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1762 		/*
1763 		 *	RFC1812 recommendation, if source is martian,
1764 		 *	the only hint is MAC header.
1765 		 */
1766 		pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1767 			&daddr, &saddr, dev->name);
1768 		if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1769 			print_hex_dump(KERN_WARNING, "ll header: ",
1770 				       DUMP_PREFIX_OFFSET, 16, 1,
1771 				       skb_mac_header(skb),
1772 				       dev->hard_header_len, false);
1773 		}
1774 	}
1775 #endif
1776 }
1777 
1778 /* called in rcu_read_lock() section */
1779 static int __mkroute_input(struct sk_buff *skb,
1780 			   const struct fib_result *res,
1781 			   struct in_device *in_dev,
1782 			   __be32 daddr, __be32 saddr, u32 tos)
1783 {
1784 	struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1785 	struct net_device *dev = nhc->nhc_dev;
1786 	struct fib_nh_exception *fnhe;
1787 	struct rtable *rth;
1788 	int err;
1789 	struct in_device *out_dev;
1790 	bool do_cache;
1791 	u32 itag = 0;
1792 
1793 	/* get a working reference to the output device */
1794 	out_dev = __in_dev_get_rcu(dev);
1795 	if (!out_dev) {
1796 		net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1797 		return -EINVAL;
1798 	}
1799 
1800 	err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1801 				  in_dev->dev, in_dev, &itag);
1802 	if (err < 0) {
1803 		ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1804 					 saddr);
1805 
1806 		goto cleanup;
1807 	}
1808 
1809 	do_cache = res->fi && !itag;
1810 	if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1811 	    skb->protocol == htons(ETH_P_IP)) {
1812 		__be32 gw;
1813 
1814 		gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1815 		if (IN_DEV_SHARED_MEDIA(out_dev) ||
1816 		    inet_addr_onlink(out_dev, saddr, gw))
1817 			IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1818 	}
1819 
1820 	if (skb->protocol != htons(ETH_P_IP)) {
1821 		/* Not IP (i.e. ARP). Do not create route, if it is
1822 		 * invalid for proxy arp. DNAT routes are always valid.
1823 		 *
1824 		 * Proxy arp feature have been extended to allow, ARP
1825 		 * replies back to the same interface, to support
1826 		 * Private VLAN switch technologies. See arp.c.
1827 		 */
1828 		if (out_dev == in_dev &&
1829 		    IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1830 			err = -EINVAL;
1831 			goto cleanup;
1832 		}
1833 	}
1834 
1835 	fnhe = find_exception(nhc, daddr);
1836 	if (do_cache) {
1837 		if (fnhe)
1838 			rth = rcu_dereference(fnhe->fnhe_rth_input);
1839 		else
1840 			rth = rcu_dereference(nhc->nhc_rth_input);
1841 		if (rt_cache_valid(rth)) {
1842 			skb_dst_set_noref(skb, &rth->dst);
1843 			goto out;
1844 		}
1845 	}
1846 
1847 	rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1848 			   IN_DEV_CONF_GET(in_dev, NOPOLICY),
1849 			   IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1850 	if (!rth) {
1851 		err = -ENOBUFS;
1852 		goto cleanup;
1853 	}
1854 
1855 	rth->rt_is_input = 1;
1856 	RT_CACHE_STAT_INC(in_slow_tot);
1857 
1858 	rth->dst.input = ip_forward;
1859 
1860 	rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1861 		       do_cache);
1862 	lwtunnel_set_redirect(&rth->dst);
1863 	skb_dst_set(skb, &rth->dst);
1864 out:
1865 	err = 0;
1866  cleanup:
1867 	return err;
1868 }
1869 
1870 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1871 /* To make ICMP packets follow the right flow, the multipath hash is
1872  * calculated from the inner IP addresses.
1873  */
1874 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1875 				 struct flow_keys *hash_keys)
1876 {
1877 	const struct iphdr *outer_iph = ip_hdr(skb);
1878 	const struct iphdr *key_iph = outer_iph;
1879 	const struct iphdr *inner_iph;
1880 	const struct icmphdr *icmph;
1881 	struct iphdr _inner_iph;
1882 	struct icmphdr _icmph;
1883 
1884 	if (likely(outer_iph->protocol != IPPROTO_ICMP))
1885 		goto out;
1886 
1887 	if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1888 		goto out;
1889 
1890 	icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1891 				   &_icmph);
1892 	if (!icmph)
1893 		goto out;
1894 
1895 	if (icmph->type != ICMP_DEST_UNREACH &&
1896 	    icmph->type != ICMP_REDIRECT &&
1897 	    icmph->type != ICMP_TIME_EXCEEDED &&
1898 	    icmph->type != ICMP_PARAMETERPROB)
1899 		goto out;
1900 
1901 	inner_iph = skb_header_pointer(skb,
1902 				       outer_iph->ihl * 4 + sizeof(_icmph),
1903 				       sizeof(_inner_iph), &_inner_iph);
1904 	if (!inner_iph)
1905 		goto out;
1906 
1907 	key_iph = inner_iph;
1908 out:
1909 	hash_keys->addrs.v4addrs.src = key_iph->saddr;
1910 	hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1911 }
1912 
1913 /* if skb is set it will be used and fl4 can be NULL */
1914 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
1915 		       const struct sk_buff *skb, struct flow_keys *flkeys)
1916 {
1917 	u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
1918 	struct flow_keys hash_keys;
1919 	u32 mhash;
1920 
1921 	switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1922 	case 0:
1923 		memset(&hash_keys, 0, sizeof(hash_keys));
1924 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1925 		if (skb) {
1926 			ip_multipath_l3_keys(skb, &hash_keys);
1927 		} else {
1928 			hash_keys.addrs.v4addrs.src = fl4->saddr;
1929 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
1930 		}
1931 		break;
1932 	case 1:
1933 		/* skb is currently provided only when forwarding */
1934 		if (skb) {
1935 			unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1936 			struct flow_keys keys;
1937 
1938 			/* short-circuit if we already have L4 hash present */
1939 			if (skb->l4_hash)
1940 				return skb_get_hash_raw(skb) >> 1;
1941 
1942 			memset(&hash_keys, 0, sizeof(hash_keys));
1943 
1944 			if (!flkeys) {
1945 				skb_flow_dissect_flow_keys(skb, &keys, flag);
1946 				flkeys = &keys;
1947 			}
1948 
1949 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1950 			hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
1951 			hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
1952 			hash_keys.ports.src = flkeys->ports.src;
1953 			hash_keys.ports.dst = flkeys->ports.dst;
1954 			hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
1955 		} else {
1956 			memset(&hash_keys, 0, sizeof(hash_keys));
1957 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1958 			hash_keys.addrs.v4addrs.src = fl4->saddr;
1959 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
1960 			hash_keys.ports.src = fl4->fl4_sport;
1961 			hash_keys.ports.dst = fl4->fl4_dport;
1962 			hash_keys.basic.ip_proto = fl4->flowi4_proto;
1963 		}
1964 		break;
1965 	}
1966 	mhash = flow_hash_from_keys(&hash_keys);
1967 
1968 	if (multipath_hash)
1969 		mhash = jhash_2words(mhash, multipath_hash, 0);
1970 
1971 	return mhash >> 1;
1972 }
1973 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1974 
1975 static int ip_mkroute_input(struct sk_buff *skb,
1976 			    struct fib_result *res,
1977 			    struct in_device *in_dev,
1978 			    __be32 daddr, __be32 saddr, u32 tos,
1979 			    struct flow_keys *hkeys)
1980 {
1981 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1982 	if (res->fi && res->fi->fib_nhs > 1) {
1983 		int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
1984 
1985 		fib_select_multipath(res, h);
1986 	}
1987 #endif
1988 
1989 	/* create a routing cache entry */
1990 	return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1991 }
1992 
1993 /*
1994  *	NOTE. We drop all the packets that has local source
1995  *	addresses, because every properly looped back packet
1996  *	must have correct destination already attached by output routine.
1997  *
1998  *	Such approach solves two big problems:
1999  *	1. Not simplex devices are handled properly.
2000  *	2. IP spoofing attempts are filtered with 100% of guarantee.
2001  *	called with rcu_read_lock()
2002  */
2003 
2004 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2005 			       u8 tos, struct net_device *dev,
2006 			       struct fib_result *res)
2007 {
2008 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2009 	struct flow_keys *flkeys = NULL, _flkeys;
2010 	struct net    *net = dev_net(dev);
2011 	struct ip_tunnel_info *tun_info;
2012 	int		err = -EINVAL;
2013 	unsigned int	flags = 0;
2014 	u32		itag = 0;
2015 	struct rtable	*rth;
2016 	struct flowi4	fl4;
2017 	bool do_cache = true;
2018 
2019 	/* IP on this device is disabled. */
2020 
2021 	if (!in_dev)
2022 		goto out;
2023 
2024 	/* Check for the most weird martians, which can be not detected
2025 	   by fib_lookup.
2026 	 */
2027 
2028 	tun_info = skb_tunnel_info(skb);
2029 	if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2030 		fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2031 	else
2032 		fl4.flowi4_tun_key.tun_id = 0;
2033 	skb_dst_drop(skb);
2034 
2035 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2036 		goto martian_source;
2037 
2038 	res->fi = NULL;
2039 	res->table = NULL;
2040 	if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2041 		goto brd_input;
2042 
2043 	/* Accept zero addresses only to limited broadcast;
2044 	 * I even do not know to fix it or not. Waiting for complains :-)
2045 	 */
2046 	if (ipv4_is_zeronet(saddr))
2047 		goto martian_source;
2048 
2049 	if (ipv4_is_zeronet(daddr))
2050 		goto martian_destination;
2051 
2052 	/* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2053 	 * and call it once if daddr or/and saddr are loopback addresses
2054 	 */
2055 	if (ipv4_is_loopback(daddr)) {
2056 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2057 			goto martian_destination;
2058 	} else if (ipv4_is_loopback(saddr)) {
2059 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2060 			goto martian_source;
2061 	}
2062 
2063 	/*
2064 	 *	Now we are ready to route packet.
2065 	 */
2066 	fl4.flowi4_oif = 0;
2067 	fl4.flowi4_iif = dev->ifindex;
2068 	fl4.flowi4_mark = skb->mark;
2069 	fl4.flowi4_tos = tos;
2070 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2071 	fl4.flowi4_flags = 0;
2072 	fl4.daddr = daddr;
2073 	fl4.saddr = saddr;
2074 	fl4.flowi4_uid = sock_net_uid(net, NULL);
2075 
2076 	if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2077 		flkeys = &_flkeys;
2078 	} else {
2079 		fl4.flowi4_proto = 0;
2080 		fl4.fl4_sport = 0;
2081 		fl4.fl4_dport = 0;
2082 	}
2083 
2084 	err = fib_lookup(net, &fl4, res, 0);
2085 	if (err != 0) {
2086 		if (!IN_DEV_FORWARD(in_dev))
2087 			err = -EHOSTUNREACH;
2088 		goto no_route;
2089 	}
2090 
2091 	if (res->type == RTN_BROADCAST) {
2092 		if (IN_DEV_BFORWARD(in_dev))
2093 			goto make_route;
2094 		/* not do cache if bc_forwarding is enabled */
2095 		if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2096 			do_cache = false;
2097 		goto brd_input;
2098 	}
2099 
2100 	if (res->type == RTN_LOCAL) {
2101 		err = fib_validate_source(skb, saddr, daddr, tos,
2102 					  0, dev, in_dev, &itag);
2103 		if (err < 0)
2104 			goto martian_source;
2105 		goto local_input;
2106 	}
2107 
2108 	if (!IN_DEV_FORWARD(in_dev)) {
2109 		err = -EHOSTUNREACH;
2110 		goto no_route;
2111 	}
2112 	if (res->type != RTN_UNICAST)
2113 		goto martian_destination;
2114 
2115 make_route:
2116 	err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2117 out:	return err;
2118 
2119 brd_input:
2120 	if (skb->protocol != htons(ETH_P_IP))
2121 		goto e_inval;
2122 
2123 	if (!ipv4_is_zeronet(saddr)) {
2124 		err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2125 					  in_dev, &itag);
2126 		if (err < 0)
2127 			goto martian_source;
2128 	}
2129 	flags |= RTCF_BROADCAST;
2130 	res->type = RTN_BROADCAST;
2131 	RT_CACHE_STAT_INC(in_brd);
2132 
2133 local_input:
2134 	do_cache &= res->fi && !itag;
2135 	if (do_cache) {
2136 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2137 
2138 		rth = rcu_dereference(nhc->nhc_rth_input);
2139 		if (rt_cache_valid(rth)) {
2140 			skb_dst_set_noref(skb, &rth->dst);
2141 			err = 0;
2142 			goto out;
2143 		}
2144 	}
2145 
2146 	rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2147 			   flags | RTCF_LOCAL, res->type,
2148 			   IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
2149 	if (!rth)
2150 		goto e_nobufs;
2151 
2152 	rth->dst.output= ip_rt_bug;
2153 #ifdef CONFIG_IP_ROUTE_CLASSID
2154 	rth->dst.tclassid = itag;
2155 #endif
2156 	rth->rt_is_input = 1;
2157 
2158 	RT_CACHE_STAT_INC(in_slow_tot);
2159 	if (res->type == RTN_UNREACHABLE) {
2160 		rth->dst.input= ip_error;
2161 		rth->dst.error= -err;
2162 		rth->rt_flags 	&= ~RTCF_LOCAL;
2163 	}
2164 
2165 	if (do_cache) {
2166 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2167 
2168 		rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2169 		if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2170 			WARN_ON(rth->dst.input == lwtunnel_input);
2171 			rth->dst.lwtstate->orig_input = rth->dst.input;
2172 			rth->dst.input = lwtunnel_input;
2173 		}
2174 
2175 		if (unlikely(!rt_cache_route(nhc, rth)))
2176 			rt_add_uncached_list(rth);
2177 	}
2178 	skb_dst_set(skb, &rth->dst);
2179 	err = 0;
2180 	goto out;
2181 
2182 no_route:
2183 	RT_CACHE_STAT_INC(in_no_route);
2184 	res->type = RTN_UNREACHABLE;
2185 	res->fi = NULL;
2186 	res->table = NULL;
2187 	goto local_input;
2188 
2189 	/*
2190 	 *	Do not cache martian addresses: they should be logged (RFC1812)
2191 	 */
2192 martian_destination:
2193 	RT_CACHE_STAT_INC(in_martian_dst);
2194 #ifdef CONFIG_IP_ROUTE_VERBOSE
2195 	if (IN_DEV_LOG_MARTIANS(in_dev))
2196 		net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2197 				     &daddr, &saddr, dev->name);
2198 #endif
2199 
2200 e_inval:
2201 	err = -EINVAL;
2202 	goto out;
2203 
2204 e_nobufs:
2205 	err = -ENOBUFS;
2206 	goto out;
2207 
2208 martian_source:
2209 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2210 	goto out;
2211 }
2212 
2213 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2214 			 u8 tos, struct net_device *dev)
2215 {
2216 	struct fib_result res;
2217 	int err;
2218 
2219 	tos &= IPTOS_RT_MASK;
2220 	rcu_read_lock();
2221 	err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2222 	rcu_read_unlock();
2223 
2224 	return err;
2225 }
2226 EXPORT_SYMBOL(ip_route_input_noref);
2227 
2228 /* called with rcu_read_lock held */
2229 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2230 		       u8 tos, struct net_device *dev, struct fib_result *res)
2231 {
2232 	/* Multicast recognition logic is moved from route cache to here.
2233 	   The problem was that too many Ethernet cards have broken/missing
2234 	   hardware multicast filters :-( As result the host on multicasting
2235 	   network acquires a lot of useless route cache entries, sort of
2236 	   SDR messages from all the world. Now we try to get rid of them.
2237 	   Really, provided software IP multicast filter is organized
2238 	   reasonably (at least, hashed), it does not result in a slowdown
2239 	   comparing with route cache reject entries.
2240 	   Note, that multicast routers are not affected, because
2241 	   route cache entry is created eventually.
2242 	 */
2243 	if (ipv4_is_multicast(daddr)) {
2244 		struct in_device *in_dev = __in_dev_get_rcu(dev);
2245 		int our = 0;
2246 		int err = -EINVAL;
2247 
2248 		if (!in_dev)
2249 			return err;
2250 		our = ip_check_mc_rcu(in_dev, daddr, saddr,
2251 				      ip_hdr(skb)->protocol);
2252 
2253 		/* check l3 master if no match yet */
2254 		if (!our && netif_is_l3_slave(dev)) {
2255 			struct in_device *l3_in_dev;
2256 
2257 			l3_in_dev = __in_dev_get_rcu(skb->dev);
2258 			if (l3_in_dev)
2259 				our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2260 						      ip_hdr(skb)->protocol);
2261 		}
2262 
2263 		if (our
2264 #ifdef CONFIG_IP_MROUTE
2265 			||
2266 		    (!ipv4_is_local_multicast(daddr) &&
2267 		     IN_DEV_MFORWARD(in_dev))
2268 #endif
2269 		   ) {
2270 			err = ip_route_input_mc(skb, daddr, saddr,
2271 						tos, dev, our);
2272 		}
2273 		return err;
2274 	}
2275 
2276 	return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2277 }
2278 
2279 /* called with rcu_read_lock() */
2280 static struct rtable *__mkroute_output(const struct fib_result *res,
2281 				       const struct flowi4 *fl4, int orig_oif,
2282 				       struct net_device *dev_out,
2283 				       unsigned int flags)
2284 {
2285 	struct fib_info *fi = res->fi;
2286 	struct fib_nh_exception *fnhe;
2287 	struct in_device *in_dev;
2288 	u16 type = res->type;
2289 	struct rtable *rth;
2290 	bool do_cache;
2291 
2292 	in_dev = __in_dev_get_rcu(dev_out);
2293 	if (!in_dev)
2294 		return ERR_PTR(-EINVAL);
2295 
2296 	if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2297 		if (ipv4_is_loopback(fl4->saddr) &&
2298 		    !(dev_out->flags & IFF_LOOPBACK) &&
2299 		    !netif_is_l3_master(dev_out))
2300 			return ERR_PTR(-EINVAL);
2301 
2302 	if (ipv4_is_lbcast(fl4->daddr))
2303 		type = RTN_BROADCAST;
2304 	else if (ipv4_is_multicast(fl4->daddr))
2305 		type = RTN_MULTICAST;
2306 	else if (ipv4_is_zeronet(fl4->daddr))
2307 		return ERR_PTR(-EINVAL);
2308 
2309 	if (dev_out->flags & IFF_LOOPBACK)
2310 		flags |= RTCF_LOCAL;
2311 
2312 	do_cache = true;
2313 	if (type == RTN_BROADCAST) {
2314 		flags |= RTCF_BROADCAST | RTCF_LOCAL;
2315 		fi = NULL;
2316 	} else if (type == RTN_MULTICAST) {
2317 		flags |= RTCF_MULTICAST | RTCF_LOCAL;
2318 		if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2319 				     fl4->flowi4_proto))
2320 			flags &= ~RTCF_LOCAL;
2321 		else
2322 			do_cache = false;
2323 		/* If multicast route do not exist use
2324 		 * default one, but do not gateway in this case.
2325 		 * Yes, it is hack.
2326 		 */
2327 		if (fi && res->prefixlen < 4)
2328 			fi = NULL;
2329 	} else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2330 		   (orig_oif != dev_out->ifindex)) {
2331 		/* For local routes that require a particular output interface
2332 		 * we do not want to cache the result.  Caching the result
2333 		 * causes incorrect behaviour when there are multiple source
2334 		 * addresses on the interface, the end result being that if the
2335 		 * intended recipient is waiting on that interface for the
2336 		 * packet he won't receive it because it will be delivered on
2337 		 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2338 		 * be set to the loopback interface as well.
2339 		 */
2340 		do_cache = false;
2341 	}
2342 
2343 	fnhe = NULL;
2344 	do_cache &= fi != NULL;
2345 	if (fi) {
2346 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2347 		struct rtable __rcu **prth;
2348 
2349 		fnhe = find_exception(nhc, fl4->daddr);
2350 		if (!do_cache)
2351 			goto add;
2352 		if (fnhe) {
2353 			prth = &fnhe->fnhe_rth_output;
2354 		} else {
2355 			if (unlikely(fl4->flowi4_flags &
2356 				     FLOWI_FLAG_KNOWN_NH &&
2357 				     !(nhc->nhc_gw_family &&
2358 				       nhc->nhc_scope == RT_SCOPE_LINK))) {
2359 				do_cache = false;
2360 				goto add;
2361 			}
2362 			prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2363 		}
2364 		rth = rcu_dereference(*prth);
2365 		if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2366 			return rth;
2367 	}
2368 
2369 add:
2370 	rth = rt_dst_alloc(dev_out, flags, type,
2371 			   IN_DEV_CONF_GET(in_dev, NOPOLICY),
2372 			   IN_DEV_CONF_GET(in_dev, NOXFRM),
2373 			   do_cache);
2374 	if (!rth)
2375 		return ERR_PTR(-ENOBUFS);
2376 
2377 	rth->rt_iif = orig_oif;
2378 
2379 	RT_CACHE_STAT_INC(out_slow_tot);
2380 
2381 	if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2382 		if (flags & RTCF_LOCAL &&
2383 		    !(dev_out->flags & IFF_LOOPBACK)) {
2384 			rth->dst.output = ip_mc_output;
2385 			RT_CACHE_STAT_INC(out_slow_mc);
2386 		}
2387 #ifdef CONFIG_IP_MROUTE
2388 		if (type == RTN_MULTICAST) {
2389 			if (IN_DEV_MFORWARD(in_dev) &&
2390 			    !ipv4_is_local_multicast(fl4->daddr)) {
2391 				rth->dst.input = ip_mr_input;
2392 				rth->dst.output = ip_mc_output;
2393 			}
2394 		}
2395 #endif
2396 	}
2397 
2398 	rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2399 	lwtunnel_set_redirect(&rth->dst);
2400 
2401 	return rth;
2402 }
2403 
2404 /*
2405  * Major route resolver routine.
2406  */
2407 
2408 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2409 					const struct sk_buff *skb)
2410 {
2411 	__u8 tos = RT_FL_TOS(fl4);
2412 	struct fib_result res = {
2413 		.type		= RTN_UNSPEC,
2414 		.fi		= NULL,
2415 		.table		= NULL,
2416 		.tclassid	= 0,
2417 	};
2418 	struct rtable *rth;
2419 
2420 	fl4->flowi4_iif = LOOPBACK_IFINDEX;
2421 	fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2422 	fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2423 			 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2424 
2425 	rcu_read_lock();
2426 	rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2427 	rcu_read_unlock();
2428 
2429 	return rth;
2430 }
2431 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2432 
2433 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2434 					    struct fib_result *res,
2435 					    const struct sk_buff *skb)
2436 {
2437 	struct net_device *dev_out = NULL;
2438 	int orig_oif = fl4->flowi4_oif;
2439 	unsigned int flags = 0;
2440 	struct rtable *rth;
2441 	int err = -ENETUNREACH;
2442 
2443 	if (fl4->saddr) {
2444 		rth = ERR_PTR(-EINVAL);
2445 		if (ipv4_is_multicast(fl4->saddr) ||
2446 		    ipv4_is_lbcast(fl4->saddr) ||
2447 		    ipv4_is_zeronet(fl4->saddr))
2448 			goto out;
2449 
2450 		/* I removed check for oif == dev_out->oif here.
2451 		   It was wrong for two reasons:
2452 		   1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2453 		      is assigned to multiple interfaces.
2454 		   2. Moreover, we are allowed to send packets with saddr
2455 		      of another iface. --ANK
2456 		 */
2457 
2458 		if (fl4->flowi4_oif == 0 &&
2459 		    (ipv4_is_multicast(fl4->daddr) ||
2460 		     ipv4_is_lbcast(fl4->daddr))) {
2461 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2462 			dev_out = __ip_dev_find(net, fl4->saddr, false);
2463 			if (!dev_out)
2464 				goto out;
2465 
2466 			/* Special hack: user can direct multicasts
2467 			   and limited broadcast via necessary interface
2468 			   without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2469 			   This hack is not just for fun, it allows
2470 			   vic,vat and friends to work.
2471 			   They bind socket to loopback, set ttl to zero
2472 			   and expect that it will work.
2473 			   From the viewpoint of routing cache they are broken,
2474 			   because we are not allowed to build multicast path
2475 			   with loopback source addr (look, routing cache
2476 			   cannot know, that ttl is zero, so that packet
2477 			   will not leave this host and route is valid).
2478 			   Luckily, this hack is good workaround.
2479 			 */
2480 
2481 			fl4->flowi4_oif = dev_out->ifindex;
2482 			goto make_route;
2483 		}
2484 
2485 		if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2486 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2487 			if (!__ip_dev_find(net, fl4->saddr, false))
2488 				goto out;
2489 		}
2490 	}
2491 
2492 
2493 	if (fl4->flowi4_oif) {
2494 		dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2495 		rth = ERR_PTR(-ENODEV);
2496 		if (!dev_out)
2497 			goto out;
2498 
2499 		/* RACE: Check return value of inet_select_addr instead. */
2500 		if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2501 			rth = ERR_PTR(-ENETUNREACH);
2502 			goto out;
2503 		}
2504 		if (ipv4_is_local_multicast(fl4->daddr) ||
2505 		    ipv4_is_lbcast(fl4->daddr) ||
2506 		    fl4->flowi4_proto == IPPROTO_IGMP) {
2507 			if (!fl4->saddr)
2508 				fl4->saddr = inet_select_addr(dev_out, 0,
2509 							      RT_SCOPE_LINK);
2510 			goto make_route;
2511 		}
2512 		if (!fl4->saddr) {
2513 			if (ipv4_is_multicast(fl4->daddr))
2514 				fl4->saddr = inet_select_addr(dev_out, 0,
2515 							      fl4->flowi4_scope);
2516 			else if (!fl4->daddr)
2517 				fl4->saddr = inet_select_addr(dev_out, 0,
2518 							      RT_SCOPE_HOST);
2519 		}
2520 	}
2521 
2522 	if (!fl4->daddr) {
2523 		fl4->daddr = fl4->saddr;
2524 		if (!fl4->daddr)
2525 			fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2526 		dev_out = net->loopback_dev;
2527 		fl4->flowi4_oif = LOOPBACK_IFINDEX;
2528 		res->type = RTN_LOCAL;
2529 		flags |= RTCF_LOCAL;
2530 		goto make_route;
2531 	}
2532 
2533 	err = fib_lookup(net, fl4, res, 0);
2534 	if (err) {
2535 		res->fi = NULL;
2536 		res->table = NULL;
2537 		if (fl4->flowi4_oif &&
2538 		    (ipv4_is_multicast(fl4->daddr) ||
2539 		    !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2540 			/* Apparently, routing tables are wrong. Assume,
2541 			   that the destination is on link.
2542 
2543 			   WHY? DW.
2544 			   Because we are allowed to send to iface
2545 			   even if it has NO routes and NO assigned
2546 			   addresses. When oif is specified, routing
2547 			   tables are looked up with only one purpose:
2548 			   to catch if destination is gatewayed, rather than
2549 			   direct. Moreover, if MSG_DONTROUTE is set,
2550 			   we send packet, ignoring both routing tables
2551 			   and ifaddr state. --ANK
2552 
2553 
2554 			   We could make it even if oif is unknown,
2555 			   likely IPv6, but we do not.
2556 			 */
2557 
2558 			if (fl4->saddr == 0)
2559 				fl4->saddr = inet_select_addr(dev_out, 0,
2560 							      RT_SCOPE_LINK);
2561 			res->type = RTN_UNICAST;
2562 			goto make_route;
2563 		}
2564 		rth = ERR_PTR(err);
2565 		goto out;
2566 	}
2567 
2568 	if (res->type == RTN_LOCAL) {
2569 		if (!fl4->saddr) {
2570 			if (res->fi->fib_prefsrc)
2571 				fl4->saddr = res->fi->fib_prefsrc;
2572 			else
2573 				fl4->saddr = fl4->daddr;
2574 		}
2575 
2576 		/* L3 master device is the loopback for that domain */
2577 		dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2578 			net->loopback_dev;
2579 
2580 		/* make sure orig_oif points to fib result device even
2581 		 * though packet rx/tx happens over loopback or l3mdev
2582 		 */
2583 		orig_oif = FIB_RES_OIF(*res);
2584 
2585 		fl4->flowi4_oif = dev_out->ifindex;
2586 		flags |= RTCF_LOCAL;
2587 		goto make_route;
2588 	}
2589 
2590 	fib_select_path(net, res, fl4, skb);
2591 
2592 	dev_out = FIB_RES_DEV(*res);
2593 	fl4->flowi4_oif = dev_out->ifindex;
2594 
2595 
2596 make_route:
2597 	rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2598 
2599 out:
2600 	return rth;
2601 }
2602 
2603 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2604 {
2605 	return NULL;
2606 }
2607 
2608 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2609 {
2610 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2611 
2612 	return mtu ? : dst->dev->mtu;
2613 }
2614 
2615 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2616 					  struct sk_buff *skb, u32 mtu)
2617 {
2618 }
2619 
2620 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2621 				       struct sk_buff *skb)
2622 {
2623 }
2624 
2625 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2626 					  unsigned long old)
2627 {
2628 	return NULL;
2629 }
2630 
2631 static struct dst_ops ipv4_dst_blackhole_ops = {
2632 	.family			=	AF_INET,
2633 	.check			=	ipv4_blackhole_dst_check,
2634 	.mtu			=	ipv4_blackhole_mtu,
2635 	.default_advmss		=	ipv4_default_advmss,
2636 	.update_pmtu		=	ipv4_rt_blackhole_update_pmtu,
2637 	.redirect		=	ipv4_rt_blackhole_redirect,
2638 	.cow_metrics		=	ipv4_rt_blackhole_cow_metrics,
2639 	.neigh_lookup		=	ipv4_neigh_lookup,
2640 };
2641 
2642 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2643 {
2644 	struct rtable *ort = (struct rtable *) dst_orig;
2645 	struct rtable *rt;
2646 
2647 	rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2648 	if (rt) {
2649 		struct dst_entry *new = &rt->dst;
2650 
2651 		new->__use = 1;
2652 		new->input = dst_discard;
2653 		new->output = dst_discard_out;
2654 
2655 		new->dev = net->loopback_dev;
2656 		if (new->dev)
2657 			dev_hold(new->dev);
2658 
2659 		rt->rt_is_input = ort->rt_is_input;
2660 		rt->rt_iif = ort->rt_iif;
2661 		rt->rt_pmtu = ort->rt_pmtu;
2662 		rt->rt_mtu_locked = ort->rt_mtu_locked;
2663 
2664 		rt->rt_genid = rt_genid_ipv4(net);
2665 		rt->rt_flags = ort->rt_flags;
2666 		rt->rt_type = ort->rt_type;
2667 		rt->rt_gw_family = ort->rt_gw_family;
2668 		if (rt->rt_gw_family == AF_INET)
2669 			rt->rt_gw4 = ort->rt_gw4;
2670 		else if (rt->rt_gw_family == AF_INET6)
2671 			rt->rt_gw6 = ort->rt_gw6;
2672 
2673 		INIT_LIST_HEAD(&rt->rt_uncached);
2674 	}
2675 
2676 	dst_release(dst_orig);
2677 
2678 	return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2679 }
2680 
2681 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2682 				    const struct sock *sk)
2683 {
2684 	struct rtable *rt = __ip_route_output_key(net, flp4);
2685 
2686 	if (IS_ERR(rt))
2687 		return rt;
2688 
2689 	if (flp4->flowi4_proto)
2690 		rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2691 							flowi4_to_flowi(flp4),
2692 							sk, 0);
2693 
2694 	return rt;
2695 }
2696 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2697 
2698 /* called with rcu_read_lock held */
2699 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2700 			struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2701 			struct sk_buff *skb, u32 portid, u32 seq)
2702 {
2703 	struct rtmsg *r;
2704 	struct nlmsghdr *nlh;
2705 	unsigned long expires = 0;
2706 	u32 error;
2707 	u32 metrics[RTAX_MAX];
2708 
2709 	nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0);
2710 	if (!nlh)
2711 		return -EMSGSIZE;
2712 
2713 	r = nlmsg_data(nlh);
2714 	r->rtm_family	 = AF_INET;
2715 	r->rtm_dst_len	= 32;
2716 	r->rtm_src_len	= 0;
2717 	r->rtm_tos	= fl4->flowi4_tos;
2718 	r->rtm_table	= table_id < 256 ? table_id : RT_TABLE_COMPAT;
2719 	if (nla_put_u32(skb, RTA_TABLE, table_id))
2720 		goto nla_put_failure;
2721 	r->rtm_type	= rt->rt_type;
2722 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2723 	r->rtm_protocol = RTPROT_UNSPEC;
2724 	r->rtm_flags	= (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2725 	if (rt->rt_flags & RTCF_NOTIFY)
2726 		r->rtm_flags |= RTM_F_NOTIFY;
2727 	if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2728 		r->rtm_flags |= RTCF_DOREDIRECT;
2729 
2730 	if (nla_put_in_addr(skb, RTA_DST, dst))
2731 		goto nla_put_failure;
2732 	if (src) {
2733 		r->rtm_src_len = 32;
2734 		if (nla_put_in_addr(skb, RTA_SRC, src))
2735 			goto nla_put_failure;
2736 	}
2737 	if (rt->dst.dev &&
2738 	    nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2739 		goto nla_put_failure;
2740 #ifdef CONFIG_IP_ROUTE_CLASSID
2741 	if (rt->dst.tclassid &&
2742 	    nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2743 		goto nla_put_failure;
2744 #endif
2745 	if (!rt_is_input_route(rt) &&
2746 	    fl4->saddr != src) {
2747 		if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2748 			goto nla_put_failure;
2749 	}
2750 	if (rt->rt_gw_family == AF_INET &&
2751 	    nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2752 		goto nla_put_failure;
2753 	} else if (rt->rt_gw_family == AF_INET6) {
2754 		int alen = sizeof(struct in6_addr);
2755 		struct nlattr *nla;
2756 		struct rtvia *via;
2757 
2758 		nla = nla_reserve(skb, RTA_VIA, alen + 2);
2759 		if (!nla)
2760 			goto nla_put_failure;
2761 
2762 		via = nla_data(nla);
2763 		via->rtvia_family = AF_INET6;
2764 		memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2765 	}
2766 
2767 	expires = rt->dst.expires;
2768 	if (expires) {
2769 		unsigned long now = jiffies;
2770 
2771 		if (time_before(now, expires))
2772 			expires -= now;
2773 		else
2774 			expires = 0;
2775 	}
2776 
2777 	memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2778 	if (rt->rt_pmtu && expires)
2779 		metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2780 	if (rt->rt_mtu_locked && expires)
2781 		metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2782 	if (rtnetlink_put_metrics(skb, metrics) < 0)
2783 		goto nla_put_failure;
2784 
2785 	if (fl4->flowi4_mark &&
2786 	    nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2787 		goto nla_put_failure;
2788 
2789 	if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2790 	    nla_put_u32(skb, RTA_UID,
2791 			from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2792 		goto nla_put_failure;
2793 
2794 	error = rt->dst.error;
2795 
2796 	if (rt_is_input_route(rt)) {
2797 #ifdef CONFIG_IP_MROUTE
2798 		if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2799 		    IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2800 			int err = ipmr_get_route(net, skb,
2801 						 fl4->saddr, fl4->daddr,
2802 						 r, portid);
2803 
2804 			if (err <= 0) {
2805 				if (err == 0)
2806 					return 0;
2807 				goto nla_put_failure;
2808 			}
2809 		} else
2810 #endif
2811 			if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
2812 				goto nla_put_failure;
2813 	}
2814 
2815 	if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2816 		goto nla_put_failure;
2817 
2818 	nlmsg_end(skb, nlh);
2819 	return 0;
2820 
2821 nla_put_failure:
2822 	nlmsg_cancel(skb, nlh);
2823 	return -EMSGSIZE;
2824 }
2825 
2826 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
2827 						   u8 ip_proto, __be16 sport,
2828 						   __be16 dport)
2829 {
2830 	struct sk_buff *skb;
2831 	struct iphdr *iph;
2832 
2833 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2834 	if (!skb)
2835 		return NULL;
2836 
2837 	/* Reserve room for dummy headers, this skb can pass
2838 	 * through good chunk of routing engine.
2839 	 */
2840 	skb_reset_mac_header(skb);
2841 	skb_reset_network_header(skb);
2842 	skb->protocol = htons(ETH_P_IP);
2843 	iph = skb_put(skb, sizeof(struct iphdr));
2844 	iph->protocol = ip_proto;
2845 	iph->saddr = src;
2846 	iph->daddr = dst;
2847 	iph->version = 0x4;
2848 	iph->frag_off = 0;
2849 	iph->ihl = 0x5;
2850 	skb_set_transport_header(skb, skb->len);
2851 
2852 	switch (iph->protocol) {
2853 	case IPPROTO_UDP: {
2854 		struct udphdr *udph;
2855 
2856 		udph = skb_put_zero(skb, sizeof(struct udphdr));
2857 		udph->source = sport;
2858 		udph->dest = dport;
2859 		udph->len = sizeof(struct udphdr);
2860 		udph->check = 0;
2861 		break;
2862 	}
2863 	case IPPROTO_TCP: {
2864 		struct tcphdr *tcph;
2865 
2866 		tcph = skb_put_zero(skb, sizeof(struct tcphdr));
2867 		tcph->source	= sport;
2868 		tcph->dest	= dport;
2869 		tcph->doff	= sizeof(struct tcphdr) / 4;
2870 		tcph->rst = 1;
2871 		tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
2872 					    src, dst, 0);
2873 		break;
2874 	}
2875 	case IPPROTO_ICMP: {
2876 		struct icmphdr *icmph;
2877 
2878 		icmph = skb_put_zero(skb, sizeof(struct icmphdr));
2879 		icmph->type = ICMP_ECHO;
2880 		icmph->code = 0;
2881 	}
2882 	}
2883 
2884 	return skb;
2885 }
2886 
2887 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
2888 				       const struct nlmsghdr *nlh,
2889 				       struct nlattr **tb,
2890 				       struct netlink_ext_ack *extack)
2891 {
2892 	struct rtmsg *rtm;
2893 	int i, err;
2894 
2895 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2896 		NL_SET_ERR_MSG(extack,
2897 			       "ipv4: Invalid header for route get request");
2898 		return -EINVAL;
2899 	}
2900 
2901 	if (!netlink_strict_get_check(skb))
2902 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2903 					      rtm_ipv4_policy, extack);
2904 
2905 	rtm = nlmsg_data(nlh);
2906 	if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
2907 	    (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
2908 	    rtm->rtm_table || rtm->rtm_protocol ||
2909 	    rtm->rtm_scope || rtm->rtm_type) {
2910 		NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
2911 		return -EINVAL;
2912 	}
2913 
2914 	if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
2915 			       RTM_F_LOOKUP_TABLE |
2916 			       RTM_F_FIB_MATCH)) {
2917 		NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
2918 		return -EINVAL;
2919 	}
2920 
2921 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2922 					    rtm_ipv4_policy, extack);
2923 	if (err)
2924 		return err;
2925 
2926 	if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
2927 	    (tb[RTA_DST] && !rtm->rtm_dst_len)) {
2928 		NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
2929 		return -EINVAL;
2930 	}
2931 
2932 	for (i = 0; i <= RTA_MAX; i++) {
2933 		if (!tb[i])
2934 			continue;
2935 
2936 		switch (i) {
2937 		case RTA_IIF:
2938 		case RTA_OIF:
2939 		case RTA_SRC:
2940 		case RTA_DST:
2941 		case RTA_IP_PROTO:
2942 		case RTA_SPORT:
2943 		case RTA_DPORT:
2944 		case RTA_MARK:
2945 		case RTA_UID:
2946 			break;
2947 		default:
2948 			NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
2949 			return -EINVAL;
2950 		}
2951 	}
2952 
2953 	return 0;
2954 }
2955 
2956 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2957 			     struct netlink_ext_ack *extack)
2958 {
2959 	struct net *net = sock_net(in_skb->sk);
2960 	struct nlattr *tb[RTA_MAX+1];
2961 	u32 table_id = RT_TABLE_MAIN;
2962 	__be16 sport = 0, dport = 0;
2963 	struct fib_result res = {};
2964 	u8 ip_proto = IPPROTO_UDP;
2965 	struct rtable *rt = NULL;
2966 	struct sk_buff *skb;
2967 	struct rtmsg *rtm;
2968 	struct flowi4 fl4 = {};
2969 	__be32 dst = 0;
2970 	__be32 src = 0;
2971 	kuid_t uid;
2972 	u32 iif;
2973 	int err;
2974 	int mark;
2975 
2976 	err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
2977 	if (err < 0)
2978 		return err;
2979 
2980 	rtm = nlmsg_data(nlh);
2981 	src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2982 	dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2983 	iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2984 	mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2985 	if (tb[RTA_UID])
2986 		uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2987 	else
2988 		uid = (iif ? INVALID_UID : current_uid());
2989 
2990 	if (tb[RTA_IP_PROTO]) {
2991 		err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
2992 						  &ip_proto, AF_INET, extack);
2993 		if (err)
2994 			return err;
2995 	}
2996 
2997 	if (tb[RTA_SPORT])
2998 		sport = nla_get_be16(tb[RTA_SPORT]);
2999 
3000 	if (tb[RTA_DPORT])
3001 		dport = nla_get_be16(tb[RTA_DPORT]);
3002 
3003 	skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3004 	if (!skb)
3005 		return -ENOBUFS;
3006 
3007 	fl4.daddr = dst;
3008 	fl4.saddr = src;
3009 	fl4.flowi4_tos = rtm->rtm_tos;
3010 	fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3011 	fl4.flowi4_mark = mark;
3012 	fl4.flowi4_uid = uid;
3013 	if (sport)
3014 		fl4.fl4_sport = sport;
3015 	if (dport)
3016 		fl4.fl4_dport = dport;
3017 	fl4.flowi4_proto = ip_proto;
3018 
3019 	rcu_read_lock();
3020 
3021 	if (iif) {
3022 		struct net_device *dev;
3023 
3024 		dev = dev_get_by_index_rcu(net, iif);
3025 		if (!dev) {
3026 			err = -ENODEV;
3027 			goto errout_rcu;
3028 		}
3029 
3030 		fl4.flowi4_iif = iif; /* for rt_fill_info */
3031 		skb->dev	= dev;
3032 		skb->mark	= mark;
3033 		err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos,
3034 					 dev, &res);
3035 
3036 		rt = skb_rtable(skb);
3037 		if (err == 0 && rt->dst.error)
3038 			err = -rt->dst.error;
3039 	} else {
3040 		fl4.flowi4_iif = LOOPBACK_IFINDEX;
3041 		skb->dev = net->loopback_dev;
3042 		rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3043 		err = 0;
3044 		if (IS_ERR(rt))
3045 			err = PTR_ERR(rt);
3046 		else
3047 			skb_dst_set(skb, &rt->dst);
3048 	}
3049 
3050 	if (err)
3051 		goto errout_rcu;
3052 
3053 	if (rtm->rtm_flags & RTM_F_NOTIFY)
3054 		rt->rt_flags |= RTCF_NOTIFY;
3055 
3056 	if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3057 		table_id = res.table ? res.table->tb_id : 0;
3058 
3059 	/* reset skb for netlink reply msg */
3060 	skb_trim(skb, 0);
3061 	skb_reset_network_header(skb);
3062 	skb_reset_transport_header(skb);
3063 	skb_reset_mac_header(skb);
3064 
3065 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3066 		if (!res.fi) {
3067 			err = fib_props[res.type].error;
3068 			if (!err)
3069 				err = -EHOSTUNREACH;
3070 			goto errout_rcu;
3071 		}
3072 		err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3073 				    nlh->nlmsg_seq, RTM_NEWROUTE, table_id,
3074 				    rt->rt_type, res.prefix, res.prefixlen,
3075 				    fl4.flowi4_tos, res.fi, 0);
3076 	} else {
3077 		err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3078 				   NETLINK_CB(in_skb).portid, nlh->nlmsg_seq);
3079 	}
3080 	if (err < 0)
3081 		goto errout_rcu;
3082 
3083 	rcu_read_unlock();
3084 
3085 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3086 
3087 errout_free:
3088 	return err;
3089 errout_rcu:
3090 	rcu_read_unlock();
3091 	kfree_skb(skb);
3092 	goto errout_free;
3093 }
3094 
3095 void ip_rt_multicast_event(struct in_device *in_dev)
3096 {
3097 	rt_cache_flush(dev_net(in_dev->dev));
3098 }
3099 
3100 #ifdef CONFIG_SYSCTL
3101 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3102 static int ip_rt_gc_min_interval __read_mostly	= HZ / 2;
3103 static int ip_rt_gc_elasticity __read_mostly	= 8;
3104 static int ip_min_valid_pmtu __read_mostly	= IPV4_MIN_MTU;
3105 
3106 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3107 					void __user *buffer,
3108 					size_t *lenp, loff_t *ppos)
3109 {
3110 	struct net *net = (struct net *)__ctl->extra1;
3111 
3112 	if (write) {
3113 		rt_cache_flush(net);
3114 		fnhe_genid_bump(net);
3115 		return 0;
3116 	}
3117 
3118 	return -EINVAL;
3119 }
3120 
3121 static struct ctl_table ipv4_route_table[] = {
3122 	{
3123 		.procname	= "gc_thresh",
3124 		.data		= &ipv4_dst_ops.gc_thresh,
3125 		.maxlen		= sizeof(int),
3126 		.mode		= 0644,
3127 		.proc_handler	= proc_dointvec,
3128 	},
3129 	{
3130 		.procname	= "max_size",
3131 		.data		= &ip_rt_max_size,
3132 		.maxlen		= sizeof(int),
3133 		.mode		= 0644,
3134 		.proc_handler	= proc_dointvec,
3135 	},
3136 	{
3137 		/*  Deprecated. Use gc_min_interval_ms */
3138 
3139 		.procname	= "gc_min_interval",
3140 		.data		= &ip_rt_gc_min_interval,
3141 		.maxlen		= sizeof(int),
3142 		.mode		= 0644,
3143 		.proc_handler	= proc_dointvec_jiffies,
3144 	},
3145 	{
3146 		.procname	= "gc_min_interval_ms",
3147 		.data		= &ip_rt_gc_min_interval,
3148 		.maxlen		= sizeof(int),
3149 		.mode		= 0644,
3150 		.proc_handler	= proc_dointvec_ms_jiffies,
3151 	},
3152 	{
3153 		.procname	= "gc_timeout",
3154 		.data		= &ip_rt_gc_timeout,
3155 		.maxlen		= sizeof(int),
3156 		.mode		= 0644,
3157 		.proc_handler	= proc_dointvec_jiffies,
3158 	},
3159 	{
3160 		.procname	= "gc_interval",
3161 		.data		= &ip_rt_gc_interval,
3162 		.maxlen		= sizeof(int),
3163 		.mode		= 0644,
3164 		.proc_handler	= proc_dointvec_jiffies,
3165 	},
3166 	{
3167 		.procname	= "redirect_load",
3168 		.data		= &ip_rt_redirect_load,
3169 		.maxlen		= sizeof(int),
3170 		.mode		= 0644,
3171 		.proc_handler	= proc_dointvec,
3172 	},
3173 	{
3174 		.procname	= "redirect_number",
3175 		.data		= &ip_rt_redirect_number,
3176 		.maxlen		= sizeof(int),
3177 		.mode		= 0644,
3178 		.proc_handler	= proc_dointvec,
3179 	},
3180 	{
3181 		.procname	= "redirect_silence",
3182 		.data		= &ip_rt_redirect_silence,
3183 		.maxlen		= sizeof(int),
3184 		.mode		= 0644,
3185 		.proc_handler	= proc_dointvec,
3186 	},
3187 	{
3188 		.procname	= "error_cost",
3189 		.data		= &ip_rt_error_cost,
3190 		.maxlen		= sizeof(int),
3191 		.mode		= 0644,
3192 		.proc_handler	= proc_dointvec,
3193 	},
3194 	{
3195 		.procname	= "error_burst",
3196 		.data		= &ip_rt_error_burst,
3197 		.maxlen		= sizeof(int),
3198 		.mode		= 0644,
3199 		.proc_handler	= proc_dointvec,
3200 	},
3201 	{
3202 		.procname	= "gc_elasticity",
3203 		.data		= &ip_rt_gc_elasticity,
3204 		.maxlen		= sizeof(int),
3205 		.mode		= 0644,
3206 		.proc_handler	= proc_dointvec,
3207 	},
3208 	{
3209 		.procname	= "mtu_expires",
3210 		.data		= &ip_rt_mtu_expires,
3211 		.maxlen		= sizeof(int),
3212 		.mode		= 0644,
3213 		.proc_handler	= proc_dointvec_jiffies,
3214 	},
3215 	{
3216 		.procname	= "min_pmtu",
3217 		.data		= &ip_rt_min_pmtu,
3218 		.maxlen		= sizeof(int),
3219 		.mode		= 0644,
3220 		.proc_handler	= proc_dointvec_minmax,
3221 		.extra1		= &ip_min_valid_pmtu,
3222 	},
3223 	{
3224 		.procname	= "min_adv_mss",
3225 		.data		= &ip_rt_min_advmss,
3226 		.maxlen		= sizeof(int),
3227 		.mode		= 0644,
3228 		.proc_handler	= proc_dointvec,
3229 	},
3230 	{ }
3231 };
3232 
3233 static struct ctl_table ipv4_route_flush_table[] = {
3234 	{
3235 		.procname	= "flush",
3236 		.maxlen		= sizeof(int),
3237 		.mode		= 0200,
3238 		.proc_handler	= ipv4_sysctl_rtcache_flush,
3239 	},
3240 	{ },
3241 };
3242 
3243 static __net_init int sysctl_route_net_init(struct net *net)
3244 {
3245 	struct ctl_table *tbl;
3246 
3247 	tbl = ipv4_route_flush_table;
3248 	if (!net_eq(net, &init_net)) {
3249 		tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3250 		if (!tbl)
3251 			goto err_dup;
3252 
3253 		/* Don't export sysctls to unprivileged users */
3254 		if (net->user_ns != &init_user_ns)
3255 			tbl[0].procname = NULL;
3256 	}
3257 	tbl[0].extra1 = net;
3258 
3259 	net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3260 	if (!net->ipv4.route_hdr)
3261 		goto err_reg;
3262 	return 0;
3263 
3264 err_reg:
3265 	if (tbl != ipv4_route_flush_table)
3266 		kfree(tbl);
3267 err_dup:
3268 	return -ENOMEM;
3269 }
3270 
3271 static __net_exit void sysctl_route_net_exit(struct net *net)
3272 {
3273 	struct ctl_table *tbl;
3274 
3275 	tbl = net->ipv4.route_hdr->ctl_table_arg;
3276 	unregister_net_sysctl_table(net->ipv4.route_hdr);
3277 	BUG_ON(tbl == ipv4_route_flush_table);
3278 	kfree(tbl);
3279 }
3280 
3281 static __net_initdata struct pernet_operations sysctl_route_ops = {
3282 	.init = sysctl_route_net_init,
3283 	.exit = sysctl_route_net_exit,
3284 };
3285 #endif
3286 
3287 static __net_init int rt_genid_init(struct net *net)
3288 {
3289 	atomic_set(&net->ipv4.rt_genid, 0);
3290 	atomic_set(&net->fnhe_genid, 0);
3291 	atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3292 	return 0;
3293 }
3294 
3295 static __net_initdata struct pernet_operations rt_genid_ops = {
3296 	.init = rt_genid_init,
3297 };
3298 
3299 static int __net_init ipv4_inetpeer_init(struct net *net)
3300 {
3301 	struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3302 
3303 	if (!bp)
3304 		return -ENOMEM;
3305 	inet_peer_base_init(bp);
3306 	net->ipv4.peers = bp;
3307 	return 0;
3308 }
3309 
3310 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3311 {
3312 	struct inet_peer_base *bp = net->ipv4.peers;
3313 
3314 	net->ipv4.peers = NULL;
3315 	inetpeer_invalidate_tree(bp);
3316 	kfree(bp);
3317 }
3318 
3319 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3320 	.init	=	ipv4_inetpeer_init,
3321 	.exit	=	ipv4_inetpeer_exit,
3322 };
3323 
3324 #ifdef CONFIG_IP_ROUTE_CLASSID
3325 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3326 #endif /* CONFIG_IP_ROUTE_CLASSID */
3327 
3328 int __init ip_rt_init(void)
3329 {
3330 	int cpu;
3331 
3332 	ip_idents = kmalloc_array(IP_IDENTS_SZ, sizeof(*ip_idents),
3333 				  GFP_KERNEL);
3334 	if (!ip_idents)
3335 		panic("IP: failed to allocate ip_idents\n");
3336 
3337 	prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
3338 
3339 	ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
3340 	if (!ip_tstamps)
3341 		panic("IP: failed to allocate ip_tstamps\n");
3342 
3343 	for_each_possible_cpu(cpu) {
3344 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3345 
3346 		INIT_LIST_HEAD(&ul->head);
3347 		spin_lock_init(&ul->lock);
3348 	}
3349 #ifdef CONFIG_IP_ROUTE_CLASSID
3350 	ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3351 	if (!ip_rt_acct)
3352 		panic("IP: failed to allocate ip_rt_acct\n");
3353 #endif
3354 
3355 	ipv4_dst_ops.kmem_cachep =
3356 		kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3357 				  SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3358 
3359 	ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3360 
3361 	if (dst_entries_init(&ipv4_dst_ops) < 0)
3362 		panic("IP: failed to allocate ipv4_dst_ops counter\n");
3363 
3364 	if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3365 		panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3366 
3367 	ipv4_dst_ops.gc_thresh = ~0;
3368 	ip_rt_max_size = INT_MAX;
3369 
3370 	devinet_init();
3371 	ip_fib_init();
3372 
3373 	if (ip_rt_proc_init())
3374 		pr_err("Unable to create route proc files\n");
3375 #ifdef CONFIG_XFRM
3376 	xfrm_init();
3377 	xfrm4_init();
3378 #endif
3379 	rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3380 		      RTNL_FLAG_DOIT_UNLOCKED);
3381 
3382 #ifdef CONFIG_SYSCTL
3383 	register_pernet_subsys(&sysctl_route_ops);
3384 #endif
3385 	register_pernet_subsys(&rt_genid_ops);
3386 	register_pernet_subsys(&ipv4_inetpeer_ops);
3387 	return 0;
3388 }
3389 
3390 #ifdef CONFIG_SYSCTL
3391 /*
3392  * We really need to sanitize the damn ipv4 init order, then all
3393  * this nonsense will go away.
3394  */
3395 void __init ip_static_sysctl_init(void)
3396 {
3397 	register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3398 }
3399 #endif
3400