xref: /linux/net/ipv4/route.c (revision f4cdf7ca9a1fdcca413157df19753f388a5a224e)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		ROUTE - implementation of the IP router.
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
12  *		Linus Torvalds, <Linus.Torvalds@helsinki.fi>
13  *		Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
14  *
15  * Fixes:
16  *		Alan Cox	:	Verify area fixes.
17  *		Alan Cox	:	cli() protects routing changes
18  *		Rui Oliveira	:	ICMP routing table updates
19  *		(rco@di.uminho.pt)	Routing table insertion and update
20  *		Linus Torvalds	:	Rewrote bits to be sensible
21  *		Alan Cox	:	Added BSD route gw semantics
22  *		Alan Cox	:	Super /proc >4K
23  *		Alan Cox	:	MTU in route table
24  *		Alan Cox	:	MSS actually. Also added the window
25  *					clamper.
26  *		Sam Lantinga	:	Fixed route matching in rt_del()
27  *		Alan Cox	:	Routing cache support.
28  *		Alan Cox	:	Removed compatibility cruft.
29  *		Alan Cox	:	RTF_REJECT support.
30  *		Alan Cox	:	TCP irtt support.
31  *		Jonathan Naylor	:	Added Metric support.
32  *	Miquel van Smoorenburg	:	BSD API fixes.
33  *	Miquel van Smoorenburg	:	Metrics.
34  *		Alan Cox	:	Use __u32 properly
35  *		Alan Cox	:	Aligned routing errors more closely with BSD
36  *					our system is still very different.
37  *		Alan Cox	:	Faster /proc handling
38  *	Alexey Kuznetsov	:	Massive rework to support tree based routing,
39  *					routing caches and better behaviour.
40  *
41  *		Olaf Erb	:	irtt wasn't being copied right.
42  *		Bjorn Ekwall	:	Kerneld route support.
43  *		Alan Cox	:	Multicast fixed (I hope)
44  *		Pavel Krauz	:	Limited broadcast fixed
45  *		Mike McLagan	:	Routing by source
46  *	Alexey Kuznetsov	:	End of old history. Split to fib.c and
47  *					route.c and rewritten from scratch.
48  *		Andi Kleen	:	Load-limit warning messages.
49  *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
50  *	Vitaly E. Lavrov	:	Race condition in ip_route_input_slow.
51  *	Tobias Ringstrom	:	Uninitialized res.type in ip_route_output_slow.
52  *	Vladimir V. Ivanov	:	IP rule info (flowid) is really useful.
53  *		Marc Boucher	:	routing by fwmark
54  *	Robert Olsson		:	Added rt_cache statistics
55  *	Arnaldo C. Melo		:	Convert proc stuff to seq_file
56  *	Eric Dumazet		:	hashed spinlocks and rt_check_expire() fixes.
57  *	Ilia Sotnikov		:	Ignore TOS on PMTUD and Redirect
58  *	Ilia Sotnikov		:	Removed TOS from hash calculations
59  */
60 
61 #define pr_fmt(fmt) "IPv4: " fmt
62 
63 #include <linux/module.h>
64 #include <linux/bitops.h>
65 #include <linux/kernel.h>
66 #include <linux/mm.h>
67 #include <linux/memblock.h>
68 #include <linux/socket.h>
69 #include <linux/errno.h>
70 #include <linux/in.h>
71 #include <linux/inet.h>
72 #include <linux/netdevice.h>
73 #include <linux/proc_fs.h>
74 #include <linux/init.h>
75 #include <linux/skbuff.h>
76 #include <linux/inetdevice.h>
77 #include <linux/igmp.h>
78 #include <linux/pkt_sched.h>
79 #include <linux/mroute.h>
80 #include <linux/netfilter_ipv4.h>
81 #include <linux/random.h>
82 #include <linux/rcupdate.h>
83 #include <linux/slab.h>
84 #include <linux/jhash.h>
85 #include <net/dst.h>
86 #include <net/dst_metadata.h>
87 #include <net/flow.h>
88 #include <net/inet_dscp.h>
89 #include <net/net_namespace.h>
90 #include <net/ip.h>
91 #include <net/route.h>
92 #include <net/inetpeer.h>
93 #include <net/sock.h>
94 #include <net/ip_fib.h>
95 #include <net/nexthop.h>
96 #include <net/tcp.h>
97 #include <net/icmp.h>
98 #include <net/xfrm.h>
99 #include <net/lwtunnel.h>
100 #include <net/netevent.h>
101 #include <net/rtnetlink.h>
102 #ifdef CONFIG_SYSCTL
103 #include <linux/sysctl.h>
104 #endif
105 #include <net/secure_seq.h>
106 #include <net/ip_tunnels.h>
107 
108 #include "fib_lookup.h"
109 
110 #define RT_GC_TIMEOUT (300*HZ)
111 
112 #define DEFAULT_MIN_PMTU (512 + 20 + 20)
113 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
114 #define DEFAULT_MIN_ADVMSS 256
115 static int ip_rt_max_size;
116 static int ip_rt_redirect_number __read_mostly	= 9;
117 static int ip_rt_redirect_load __read_mostly	= HZ / 50;
118 static int ip_rt_redirect_silence __read_mostly	= ((HZ / 50) << (9 + 1));
119 static int ip_rt_error_cost __read_mostly	= HZ;
120 static int ip_rt_error_burst __read_mostly	= 5 * HZ;
121 
122 static int ip_rt_gc_timeout __read_mostly	= RT_GC_TIMEOUT;
123 
124 /*
125  *	Interface to generic destination cache.
126  */
127 
128 INDIRECT_CALLABLE_SCOPE
129 struct dst_entry	*ipv4_dst_check(struct dst_entry *dst, u32 cookie);
130 static unsigned int	 ipv4_default_advmss(const struct dst_entry *dst);
131 INDIRECT_CALLABLE_SCOPE
132 unsigned int		ipv4_mtu(const struct dst_entry *dst);
133 static void		ipv4_negative_advice(struct sock *sk,
134 					     struct dst_entry *dst);
135 static void		 ipv4_link_failure(struct sk_buff *skb);
136 static void		 ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
137 					   struct sk_buff *skb, u32 mtu,
138 					   bool confirm_neigh);
139 static void		 ip_do_redirect(struct dst_entry *dst, struct sock *sk,
140 					struct sk_buff *skb);
141 static void		ipv4_dst_destroy(struct dst_entry *dst);
142 
143 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
144 {
145 	WARN_ON(1);
146 	return NULL;
147 }
148 
149 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
150 					   struct sk_buff *skb,
151 					   const void *daddr);
152 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
153 
154 static struct dst_ops ipv4_dst_ops = {
155 	.family =		AF_INET,
156 	.check =		ipv4_dst_check,
157 	.default_advmss =	ipv4_default_advmss,
158 	.mtu =			ipv4_mtu,
159 	.cow_metrics =		ipv4_cow_metrics,
160 	.destroy =		ipv4_dst_destroy,
161 	.negative_advice =	ipv4_negative_advice,
162 	.link_failure =		ipv4_link_failure,
163 	.update_pmtu =		ip_rt_update_pmtu,
164 	.redirect =		ip_do_redirect,
165 	.local_out =		__ip_local_out,
166 	.neigh_lookup =		ipv4_neigh_lookup,
167 	.confirm_neigh =	ipv4_confirm_neigh,
168 };
169 
170 #define ECN_OR_COST(class)	TC_PRIO_##class
171 
172 const __u8 ip_tos2prio[16] = {
173 	TC_PRIO_BESTEFFORT,
174 	ECN_OR_COST(BESTEFFORT),
175 	TC_PRIO_BESTEFFORT,
176 	ECN_OR_COST(BESTEFFORT),
177 	TC_PRIO_BULK,
178 	ECN_OR_COST(BULK),
179 	TC_PRIO_BULK,
180 	ECN_OR_COST(BULK),
181 	TC_PRIO_INTERACTIVE,
182 	ECN_OR_COST(INTERACTIVE),
183 	TC_PRIO_INTERACTIVE,
184 	ECN_OR_COST(INTERACTIVE),
185 	TC_PRIO_INTERACTIVE_BULK,
186 	ECN_OR_COST(INTERACTIVE_BULK),
187 	TC_PRIO_INTERACTIVE_BULK,
188 	ECN_OR_COST(INTERACTIVE_BULK)
189 };
190 EXPORT_SYMBOL(ip_tos2prio);
191 
192 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
193 #ifndef CONFIG_PREEMPT_RT
194 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
195 #else
196 #define RT_CACHE_STAT_INC(field) this_cpu_inc(rt_cache_stat.field)
197 #endif
198 
199 #ifdef CONFIG_PROC_FS
200 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
201 {
202 	if (*pos)
203 		return NULL;
204 	return SEQ_START_TOKEN;
205 }
206 
207 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
208 {
209 	++*pos;
210 	return NULL;
211 }
212 
213 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
214 {
215 }
216 
217 static int rt_cache_seq_show(struct seq_file *seq, void *v)
218 {
219 	if (v == SEQ_START_TOKEN)
220 		seq_printf(seq, "%-127s\n",
221 			   "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
222 			   "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
223 			   "HHUptod\tSpecDst");
224 	return 0;
225 }
226 
227 static const struct seq_operations rt_cache_seq_ops = {
228 	.start  = rt_cache_seq_start,
229 	.next   = rt_cache_seq_next,
230 	.stop   = rt_cache_seq_stop,
231 	.show   = rt_cache_seq_show,
232 };
233 
234 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
235 {
236 	int cpu;
237 
238 	if (*pos == 0)
239 		return SEQ_START_TOKEN;
240 
241 	for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
242 		if (!cpu_possible(cpu))
243 			continue;
244 		*pos = cpu+1;
245 		return &per_cpu(rt_cache_stat, cpu);
246 	}
247 	return NULL;
248 }
249 
250 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
251 {
252 	int cpu;
253 
254 	for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
255 		if (!cpu_possible(cpu))
256 			continue;
257 		*pos = cpu+1;
258 		return &per_cpu(rt_cache_stat, cpu);
259 	}
260 	(*pos)++;
261 	return NULL;
262 
263 }
264 
265 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
266 {
267 
268 }
269 
270 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
271 {
272 	struct rt_cache_stat *st = v;
273 
274 	if (v == SEQ_START_TOKEN) {
275 		seq_puts(seq, "entries  in_hit   in_slow_tot in_slow_mc in_no_route in_brd   in_martian_dst in_martian_src out_hit  out_slow_tot out_slow_mc gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
276 		return 0;
277 	}
278 
279 	seq_printf(seq, "%08x %08x %08x    %08x   %08x    %08x %08x       "
280 			"%08x       %08x %08x     %08x    %08x %08x   "
281 			"%08x     %08x        %08x        %08x\n",
282 		   dst_entries_get_slow(&ipv4_dst_ops),
283 		   0, /* st->in_hit */
284 		   st->in_slow_tot,
285 		   st->in_slow_mc,
286 		   st->in_no_route,
287 		   st->in_brd,
288 		   st->in_martian_dst,
289 		   st->in_martian_src,
290 
291 		   0, /* st->out_hit */
292 		   st->out_slow_tot,
293 		   st->out_slow_mc,
294 
295 		   0, /* st->gc_total */
296 		   0, /* st->gc_ignored */
297 		   0, /* st->gc_goal_miss */
298 		   0, /* st->gc_dst_overflow */
299 		   0, /* st->in_hlist_search */
300 		   0  /* st->out_hlist_search */
301 		);
302 	return 0;
303 }
304 
305 static const struct seq_operations rt_cpu_seq_ops = {
306 	.start  = rt_cpu_seq_start,
307 	.next   = rt_cpu_seq_next,
308 	.stop   = rt_cpu_seq_stop,
309 	.show   = rt_cpu_seq_show,
310 };
311 
312 #ifdef CONFIG_IP_ROUTE_CLASSID
313 static int rt_acct_proc_show(struct seq_file *m, void *v)
314 {
315 	struct ip_rt_acct *dst, *src;
316 	unsigned int i, j;
317 
318 	dst = kzalloc_objs(struct ip_rt_acct, 256);
319 	if (!dst)
320 		return -ENOMEM;
321 
322 	for_each_possible_cpu(i) {
323 		src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
324 		for (j = 0; j < 256; j++) {
325 			dst[j].o_bytes   += src[j].o_bytes;
326 			dst[j].o_packets += src[j].o_packets;
327 			dst[j].i_bytes   += src[j].i_bytes;
328 			dst[j].i_packets += src[j].i_packets;
329 		}
330 	}
331 
332 	seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
333 	kfree(dst);
334 	return 0;
335 }
336 #endif
337 
338 static int __net_init ip_rt_do_proc_init(struct net *net)
339 {
340 	struct proc_dir_entry *pde;
341 
342 	pde = proc_create_seq("rt_cache", 0444, net->proc_net,
343 			      &rt_cache_seq_ops);
344 	if (!pde)
345 		goto err1;
346 
347 	pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
348 			      &rt_cpu_seq_ops);
349 	if (!pde)
350 		goto err2;
351 
352 #ifdef CONFIG_IP_ROUTE_CLASSID
353 	pde = proc_create_single("rt_acct", 0, net->proc_net,
354 			rt_acct_proc_show);
355 	if (!pde)
356 		goto err3;
357 #endif
358 	return 0;
359 
360 #ifdef CONFIG_IP_ROUTE_CLASSID
361 err3:
362 	remove_proc_entry("rt_cache", net->proc_net_stat);
363 #endif
364 err2:
365 	remove_proc_entry("rt_cache", net->proc_net);
366 err1:
367 	return -ENOMEM;
368 }
369 
370 static void __net_exit ip_rt_do_proc_exit(struct net *net)
371 {
372 	remove_proc_entry("rt_cache", net->proc_net_stat);
373 	remove_proc_entry("rt_cache", net->proc_net);
374 #ifdef CONFIG_IP_ROUTE_CLASSID
375 	remove_proc_entry("rt_acct", net->proc_net);
376 #endif
377 }
378 
379 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
380 	.init = ip_rt_do_proc_init,
381 	.exit = ip_rt_do_proc_exit,
382 };
383 
384 static int __init ip_rt_proc_init(void)
385 {
386 	return register_pernet_subsys(&ip_rt_proc_ops);
387 }
388 
389 #else
390 static inline int ip_rt_proc_init(void)
391 {
392 	return 0;
393 }
394 #endif /* CONFIG_PROC_FS */
395 
396 static inline bool rt_is_expired(const struct rtable *rth)
397 {
398 	bool res;
399 
400 	rcu_read_lock();
401 	res = rth->rt_genid != rt_genid_ipv4(dev_net_rcu(rth->dst.dev));
402 	rcu_read_unlock();
403 
404 	return res;
405 }
406 
407 void rt_cache_flush(struct net *net)
408 {
409 	rt_genid_bump_ipv4(net);
410 }
411 
412 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
413 					   struct sk_buff *skb,
414 					   const void *daddr)
415 {
416 	const struct rtable *rt = container_of(dst, struct rtable, dst);
417 	struct net_device *dev;
418 	struct neighbour *n;
419 
420 	rcu_read_lock();
421 	dev = dst_dev_rcu(dst);
422 	if (likely(rt->rt_gw_family == AF_INET)) {
423 		n = ip_neigh_gw4(dev, rt->rt_gw4);
424 	} else if (rt->rt_gw_family == AF_INET6) {
425 		n = ip_neigh_gw6(dev, &rt->rt_gw6);
426         } else {
427 		__be32 pkey;
428 
429 		pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
430 		n = ip_neigh_gw4(dev, pkey);
431 	}
432 
433 	if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
434 		n = NULL;
435 
436 	rcu_read_unlock();
437 
438 	return n;
439 }
440 
441 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
442 {
443 	const struct rtable *rt = container_of(dst, struct rtable, dst);
444 	struct net_device *dev = dst_dev(dst);
445 	const __be32 *pkey = daddr;
446 
447 	if (rt->rt_gw_family == AF_INET) {
448 		pkey = (const __be32 *)&rt->rt_gw4;
449 	} else if (IS_ENABLED(CONFIG_IPV6) && rt->rt_gw_family == AF_INET6) {
450 		return __ipv6_confirm_neigh(dev, &rt->rt_gw6);
451 	} else if (!daddr ||
452 		 (rt->rt_flags &
453 		  (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
454 		return;
455 	}
456 	__ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
457 }
458 
459 /* Hash tables of size 2048..262144 depending on RAM size.
460  * Each bucket uses 8 bytes.
461  */
462 static u32 ip_idents_mask __read_mostly;
463 static atomic_t *ip_idents __read_mostly;
464 static u32 *ip_tstamps __read_mostly;
465 
466 /* In order to protect privacy, we add a perturbation to identifiers
467  * if one generator is seldom used. This makes hard for an attacker
468  * to infer how many packets were sent between two points in time.
469  */
470 static u32 ip_idents_reserve(u32 hash, int segs)
471 {
472 	u32 bucket, old, now = (u32)jiffies;
473 	atomic_t *p_id;
474 	u32 *p_tstamp;
475 	u32 delta = 0;
476 
477 	bucket = hash & ip_idents_mask;
478 	p_tstamp = ip_tstamps + bucket;
479 	p_id = ip_idents + bucket;
480 	old = READ_ONCE(*p_tstamp);
481 
482 	if (old != now && cmpxchg(p_tstamp, old, now) == old)
483 		delta = get_random_u32_below(now - old);
484 
485 	/* If UBSAN reports an error there, please make sure your compiler
486 	 * supports -fno-strict-overflow before reporting it that was a bug
487 	 * in UBSAN, and it has been fixed in GCC-8.
488 	 */
489 	return atomic_add_return(segs + delta, p_id) - segs;
490 }
491 
492 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
493 {
494 	u32 hash, id;
495 
496 	/* Note the following code is not safe, but this is okay. */
497 	if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
498 		get_random_bytes(&net->ipv4.ip_id_key,
499 				 sizeof(net->ipv4.ip_id_key));
500 
501 	hash = siphash_3u32((__force u32)iph->daddr,
502 			    (__force u32)iph->saddr,
503 			    iph->protocol,
504 			    &net->ipv4.ip_id_key);
505 	id = ip_idents_reserve(hash, segs);
506 	iph->id = htons(id);
507 }
508 EXPORT_SYMBOL(__ip_select_ident);
509 
510 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
511 			     const struct sock *sk, const struct iphdr *iph,
512 			     int oif, __u8 tos, u8 prot, u32 mark,
513 			     int flow_flags)
514 {
515 	__u8 scope = RT_SCOPE_UNIVERSE;
516 
517 	if (sk) {
518 		oif = sk->sk_bound_dev_if;
519 		mark = READ_ONCE(sk->sk_mark);
520 		tos = ip_sock_rt_tos(sk);
521 		scope = ip_sock_rt_scope(sk);
522 		prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW :
523 						    sk->sk_protocol;
524 	}
525 
526 	flowi4_init_output(fl4, oif, mark, tos & INET_DSCP_MASK, scope,
527 			   prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
528 			   sock_net_uid(net, sk));
529 }
530 
531 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
532 			       const struct sock *sk)
533 {
534 	const struct net *net = dev_net(skb->dev);
535 	const struct iphdr *iph = ip_hdr(skb);
536 	int oif = skb->dev->ifindex;
537 	u8 prot = iph->protocol;
538 	u32 mark = skb->mark;
539 	__u8 tos = iph->tos;
540 
541 	__build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
542 }
543 
544 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
545 {
546 	const struct inet_sock *inet = inet_sk(sk);
547 	const struct ip_options_rcu *inet_opt;
548 	__be32 daddr = inet->inet_daddr;
549 
550 	rcu_read_lock();
551 	inet_opt = rcu_dereference(inet->inet_opt);
552 	if (inet_opt && inet_opt->opt.srr)
553 		daddr = inet_opt->opt.faddr;
554 	flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark),
555 			   ip_sock_rt_tos(sk),
556 			   ip_sock_rt_scope(sk),
557 			   inet_test_bit(HDRINCL, sk) ?
558 				IPPROTO_RAW : sk->sk_protocol,
559 			   inet_sk_flowi_flags(sk),
560 			   daddr, inet->inet_saddr, 0, 0,
561 			   sk_uid(sk));
562 	rcu_read_unlock();
563 }
564 
565 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
566 				 const struct sk_buff *skb)
567 {
568 	if (skb)
569 		build_skb_flow_key(fl4, skb, sk);
570 	else
571 		build_sk_flow_key(fl4, sk);
572 }
573 
574 static DEFINE_SPINLOCK(fnhe_lock);
575 
576 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
577 {
578 	struct rtable *rt;
579 
580 	rt = rcu_dereference(fnhe->fnhe_rth_input);
581 	if (rt) {
582 		RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
583 		dst_dev_put(&rt->dst);
584 		dst_release(&rt->dst);
585 	}
586 	rt = rcu_dereference(fnhe->fnhe_rth_output);
587 	if (rt) {
588 		RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
589 		dst_dev_put(&rt->dst);
590 		dst_release(&rt->dst);
591 	}
592 }
593 
594 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
595 {
596 	struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
597 	struct fib_nh_exception *fnhe, *oldest = NULL;
598 
599 	for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
600 		fnhe = rcu_dereference_protected(*fnhe_p,
601 						 lockdep_is_held(&fnhe_lock));
602 		if (!fnhe)
603 			break;
604 		if (!oldest ||
605 		    time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
606 			oldest = fnhe;
607 			oldest_p = fnhe_p;
608 		}
609 	}
610 
611 	/* Clear oldest->fnhe_daddr to prevent this fnhe from being
612 	 * rebound with new dsts in rt_bind_exception().
613 	 */
614 	oldest->fnhe_daddr = 0;
615 	fnhe_flush_routes(oldest);
616 	*oldest_p = oldest->fnhe_next;
617 	kfree_rcu(oldest, rcu);
618 }
619 
620 static u32 fnhe_hashfun(__be32 daddr)
621 {
622 	static siphash_aligned_key_t fnhe_hash_key;
623 	u64 hval;
624 
625 	net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
626 	hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
627 	return hash_64(hval, FNHE_HASH_SHIFT);
628 }
629 
630 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
631 {
632 	rt->rt_pmtu = fnhe->fnhe_pmtu;
633 	rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
634 	rt->dst.expires = fnhe->fnhe_expires;
635 
636 	if (fnhe->fnhe_gw) {
637 		rt->rt_flags |= RTCF_REDIRECTED;
638 		rt->rt_uses_gateway = 1;
639 		rt->rt_gw_family = AF_INET;
640 		rt->rt_gw4 = fnhe->fnhe_gw;
641 	}
642 }
643 
644 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
645 				  __be32 gw, u32 pmtu, bool lock,
646 				  unsigned long expires)
647 {
648 	struct fnhe_hash_bucket *hash;
649 	struct fib_nh_exception *fnhe;
650 	struct rtable *rt;
651 	u32 genid, hval;
652 	unsigned int i;
653 	int depth;
654 
655 	genid = fnhe_genid(dev_net(nhc->nhc_dev));
656 	hval = fnhe_hashfun(daddr);
657 
658 	spin_lock_bh(&fnhe_lock);
659 
660 	hash = rcu_dereference(nhc->nhc_exceptions);
661 	if (!hash) {
662 		hash = kzalloc_objs(*hash, FNHE_HASH_SIZE, GFP_ATOMIC);
663 		if (!hash)
664 			goto out_unlock;
665 		rcu_assign_pointer(nhc->nhc_exceptions, hash);
666 	}
667 
668 	hash += hval;
669 
670 	depth = 0;
671 	for (fnhe = rcu_dereference(hash->chain); fnhe;
672 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
673 		if (fnhe->fnhe_daddr == daddr)
674 			break;
675 		depth++;
676 	}
677 
678 	if (fnhe) {
679 		if (fnhe->fnhe_genid != genid)
680 			fnhe->fnhe_genid = genid;
681 		if (gw)
682 			fnhe->fnhe_gw = gw;
683 		if (pmtu) {
684 			fnhe->fnhe_pmtu = pmtu;
685 			fnhe->fnhe_mtu_locked = lock;
686 		}
687 		fnhe->fnhe_expires = max(1UL, expires);
688 		/* Update all cached dsts too */
689 		rt = rcu_dereference(fnhe->fnhe_rth_input);
690 		if (rt)
691 			fill_route_from_fnhe(rt, fnhe);
692 		rt = rcu_dereference(fnhe->fnhe_rth_output);
693 		if (rt)
694 			fill_route_from_fnhe(rt, fnhe);
695 	} else {
696 		/* Randomize max depth to avoid some side channels attacks. */
697 		int max_depth = FNHE_RECLAIM_DEPTH +
698 				get_random_u32_below(FNHE_RECLAIM_DEPTH);
699 
700 		while (depth > max_depth) {
701 			fnhe_remove_oldest(hash);
702 			depth--;
703 		}
704 
705 		fnhe = kzalloc_obj(*fnhe, GFP_ATOMIC);
706 		if (!fnhe)
707 			goto out_unlock;
708 
709 		fnhe->fnhe_next = hash->chain;
710 
711 		fnhe->fnhe_genid = genid;
712 		fnhe->fnhe_daddr = daddr;
713 		fnhe->fnhe_gw = gw;
714 		fnhe->fnhe_pmtu = pmtu;
715 		fnhe->fnhe_mtu_locked = lock;
716 		fnhe->fnhe_expires = max(1UL, expires);
717 
718 		rcu_assign_pointer(hash->chain, fnhe);
719 
720 		/* Exception created; mark the cached routes for the nexthop
721 		 * stale, so anyone caching it rechecks if this exception
722 		 * applies to them.
723 		 */
724 		rt = rcu_dereference(nhc->nhc_rth_input);
725 		if (rt)
726 			WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL);
727 
728 		for_each_possible_cpu(i) {
729 			struct rtable __rcu **prt;
730 
731 			prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
732 			rt = rcu_dereference(*prt);
733 			if (rt)
734 				WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL);
735 		}
736 	}
737 
738 	fnhe->fnhe_stamp = jiffies;
739 
740 out_unlock:
741 	spin_unlock_bh(&fnhe_lock);
742 }
743 
744 /* Update the PMTU of an exception when:
745  * - the new MTU of the first hop becomes smaller than the PMTU
746  * - the old MTU was the same as the PMTU, and it limited discovery of
747  *   larger MTUs on the path. With that limit raised, we can now
748  *   discover larger MTUs
749  * A special case is locked exceptions, for which the PMTU is smaller
750  * than the minimal accepted PMTU:
751  * - if the new MTU is greater than the PMTU, don't make any change
752  * - otherwise, unlock and set PMTU
753  *
754  * fnhe_lock keeps fnhe_pmtu and fnhe_mtu_locked consistent against
755  * update_or_create_fnhe(), which sets both under the same lock.
756  */
757 void fnhe_update_pmtu(struct fib_nh_exception *fnhe, u32 new, u32 orig)
758 {
759 	spin_lock_bh(&fnhe_lock);
760 
761 	if (fnhe->fnhe_mtu_locked) {
762 		if (new <= fnhe->fnhe_pmtu) {
763 			fnhe->fnhe_pmtu = new;
764 			fnhe->fnhe_mtu_locked = false;
765 		}
766 	} else if (new < fnhe->fnhe_pmtu || orig == fnhe->fnhe_pmtu) {
767 		fnhe->fnhe_pmtu = new;
768 	}
769 
770 	spin_unlock_bh(&fnhe_lock);
771 }
772 
773 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
774 			     bool kill_route)
775 {
776 	__be32 new_gw = icmp_hdr(skb)->un.gateway;
777 	__be32 old_gw = ip_hdr(skb)->saddr;
778 	struct net_device *dev = skb->dev;
779 	struct in_device *in_dev;
780 	struct fib_result res;
781 	struct neighbour *n;
782 	struct net *net;
783 
784 	switch (icmp_hdr(skb)->code & 7) {
785 	case ICMP_REDIR_NET:
786 	case ICMP_REDIR_NETTOS:
787 	case ICMP_REDIR_HOST:
788 	case ICMP_REDIR_HOSTTOS:
789 		break;
790 
791 	default:
792 		return;
793 	}
794 
795 	if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
796 		return;
797 
798 	in_dev = __in_dev_get_rcu(dev);
799 	if (!in_dev)
800 		return;
801 
802 	net = dev_net(dev);
803 	if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
804 	    ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
805 	    ipv4_is_zeronet(new_gw))
806 		goto reject_redirect;
807 
808 	if (!IN_DEV_SHARED_MEDIA(in_dev)) {
809 		if (!inet_addr_onlink(in_dev, new_gw, old_gw))
810 			goto reject_redirect;
811 		if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
812 			goto reject_redirect;
813 	} else {
814 		if (inet_addr_type(net, new_gw) != RTN_UNICAST)
815 			goto reject_redirect;
816 	}
817 
818 	n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw);
819 	if (!n)
820 		n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
821 	if (!IS_ERR(n)) {
822 		if (!(READ_ONCE(n->nud_state) & NUD_VALID)) {
823 			neigh_event_send(n, NULL);
824 		} else {
825 			if (fib_lookup(net, fl4, &res, 0) == 0) {
826 				struct fib_nh_common *nhc;
827 
828 				fib_select_path(net, &res, fl4, skb);
829 				nhc = FIB_RES_NHC(res);
830 				update_or_create_fnhe(nhc, fl4->daddr, new_gw,
831 						0, false,
832 						jiffies + ip_rt_gc_timeout);
833 			}
834 			if (kill_route)
835 				WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL);
836 			call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
837 		}
838 		neigh_release(n);
839 	}
840 	return;
841 
842 reject_redirect:
843 #ifdef CONFIG_IP_ROUTE_VERBOSE
844 	if (IN_DEV_LOG_MARTIANS(in_dev)) {
845 		const struct iphdr *iph = (const struct iphdr *) skb->data;
846 		__be32 daddr = iph->daddr;
847 		__be32 saddr = iph->saddr;
848 
849 		net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
850 				     "  Advised path = %pI4 -> %pI4\n",
851 				     &old_gw, dev->name, &new_gw,
852 				     &saddr, &daddr);
853 	}
854 #endif
855 	;
856 }
857 
858 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
859 {
860 	struct rtable *rt;
861 	struct flowi4 fl4;
862 	const struct iphdr *iph = (const struct iphdr *) skb->data;
863 	struct net *net = dev_net(skb->dev);
864 	int oif = skb->dev->ifindex;
865 	u8 prot = iph->protocol;
866 	u32 mark = skb->mark;
867 	__u8 tos = iph->tos;
868 
869 	rt = dst_rtable(dst);
870 
871 	__build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
872 	__ip_do_redirect(rt, skb, &fl4, true);
873 }
874 
875 static void ipv4_negative_advice(struct sock *sk,
876 				 struct dst_entry *dst)
877 {
878 	struct rtable *rt = dst_rtable(dst);
879 
880 	if ((READ_ONCE(dst->obsolete) > 0) ||
881 	    (rt->rt_flags & RTCF_REDIRECTED) ||
882 	    READ_ONCE(rt->dst.expires))
883 		sk_dst_reset(sk);
884 }
885 
886 /*
887  * Algorithm:
888  *	1. The first ip_rt_redirect_number redirects are sent
889  *	   with exponential backoff, then we stop sending them at all,
890  *	   assuming that the host ignores our redirects.
891  *	2. If we did not see packets requiring redirects
892  *	   during ip_rt_redirect_silence, we assume that the host
893  *	   forgot redirected route and start to send redirects again.
894  *
895  * This algorithm is much cheaper and more intelligent than dumb load limiting
896  * in icmp.c.
897  *
898  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
899  * and "frag. need" (breaks PMTU discovery) in icmp.c.
900  */
901 
902 void ip_rt_send_redirect(struct sk_buff *skb)
903 {
904 	struct rtable *rt = skb_rtable(skb);
905 	struct in_device *in_dev;
906 	struct inet_peer *peer;
907 	struct net *net;
908 	int log_martians;
909 	int vif;
910 
911 	rcu_read_lock();
912 	in_dev = __in_dev_get_rcu(rt->dst.dev);
913 	if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
914 		rcu_read_unlock();
915 		return;
916 	}
917 	log_martians = IN_DEV_LOG_MARTIANS(in_dev);
918 	vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
919 
920 	net = dev_net(rt->dst.dev);
921 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif);
922 	if (!peer) {
923 		rcu_read_unlock();
924 		return;
925 	}
926 
927 	/* No redirected packets during ip_rt_redirect_silence;
928 	 * reset the algorithm.
929 	 */
930 	if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
931 		peer->rate_tokens = 0;
932 		peer->n_redirects = 0;
933 	}
934 
935 	/* Too many ignored redirects; do not send anything
936 	 * set dst.rate_last to the last seen redirected packet.
937 	 */
938 	if (peer->n_redirects >= ip_rt_redirect_number) {
939 		peer->rate_last = jiffies;
940 		goto out_unlock;
941 	}
942 
943 	/* Check for load limit; set rate_last to the latest sent
944 	 * redirect.
945 	 */
946 	if (peer->n_redirects == 0 ||
947 	    time_after(jiffies,
948 		       (peer->rate_last +
949 			(ip_rt_redirect_load << peer->n_redirects)))) {
950 		__be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
951 
952 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
953 		peer->rate_last = jiffies;
954 		++peer->n_redirects;
955 		if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians &&
956 		    peer->n_redirects == ip_rt_redirect_number)
957 			net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
958 					     &ip_hdr(skb)->saddr, inet_iif(skb),
959 					     &ip_hdr(skb)->daddr, &gw);
960 	}
961 out_unlock:
962 	rcu_read_unlock();
963 }
964 
965 static int ip_error(struct sk_buff *skb)
966 {
967 	struct rtable *rt = skb_rtable(skb);
968 	struct net_device *dev = skb->dev;
969 	struct in_device *in_dev;
970 	struct inet_peer *peer;
971 	unsigned long now;
972 	struct net *net;
973 	SKB_DR(reason);
974 	bool send;
975 	int code;
976 
977 	if (netif_is_l3_master(skb->dev)) {
978 		dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
979 		if (!dev)
980 			goto out;
981 	}
982 
983 	in_dev = __in_dev_get_rcu(dev);
984 
985 	/* IP on this device is disabled. */
986 	if (!in_dev)
987 		goto out;
988 
989 	net = dev_net(rt->dst.dev);
990 	if (!IN_DEV_FORWARD(in_dev)) {
991 		switch (rt->dst.error) {
992 		case EHOSTUNREACH:
993 			SKB_DR_SET(reason, IP_INADDRERRORS);
994 			__IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
995 			break;
996 
997 		case ENETUNREACH:
998 			SKB_DR_SET(reason, IP_INNOROUTES);
999 			__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
1000 			break;
1001 		}
1002 		goto out;
1003 	}
1004 
1005 	switch (rt->dst.error) {
1006 	case EINVAL:
1007 	default:
1008 		goto out;
1009 	case EHOSTUNREACH:
1010 		code = ICMP_HOST_UNREACH;
1011 		break;
1012 	case ENETUNREACH:
1013 		code = ICMP_NET_UNREACH;
1014 		SKB_DR_SET(reason, IP_INNOROUTES);
1015 		__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
1016 		break;
1017 	case EACCES:
1018 		code = ICMP_PKT_FILTERED;
1019 		break;
1020 	}
1021 
1022 	rcu_read_lock();
1023 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
1024 			       l3mdev_master_ifindex_rcu(skb->dev));
1025 	send = true;
1026 	if (peer) {
1027 		now = jiffies;
1028 		peer->rate_tokens += now - peer->rate_last;
1029 		if (peer->rate_tokens > ip_rt_error_burst)
1030 			peer->rate_tokens = ip_rt_error_burst;
1031 		peer->rate_last = now;
1032 		if (peer->rate_tokens >= ip_rt_error_cost)
1033 			peer->rate_tokens -= ip_rt_error_cost;
1034 		else
1035 			send = false;
1036 	}
1037 	rcu_read_unlock();
1038 
1039 	if (send)
1040 		icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1041 
1042 out:	kfree_skb_reason(skb, reason);
1043 	return 0;
1044 }
1045 
1046 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1047 {
1048 	struct dst_entry *dst = &rt->dst;
1049 	struct fib_result res;
1050 	bool lock = false;
1051 	struct net *net;
1052 	u32 old_mtu;
1053 
1054 	if (ip_mtu_locked(dst))
1055 		return;
1056 
1057 	old_mtu = ipv4_mtu(dst);
1058 	if (old_mtu < mtu)
1059 		return;
1060 
1061 	rcu_read_lock();
1062 	net = dst_dev_net_rcu(dst);
1063 	if (mtu < net->ipv4.ip_rt_min_pmtu) {
1064 		lock = true;
1065 		mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1066 	}
1067 
1068 	if (rt->rt_pmtu == mtu && !lock &&
1069 	    time_before(jiffies, READ_ONCE(dst->expires) -
1070 				 net->ipv4.ip_rt_mtu_expires / 2))
1071 		goto out;
1072 
1073 	if (fib_lookup(net, fl4, &res, 0) == 0) {
1074 		struct fib_nh_common *nhc;
1075 
1076 		fib_select_path(net, &res, fl4, NULL);
1077 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1078 		if (fib_info_num_path(res.fi) > 1) {
1079 			int nhsel;
1080 
1081 			for (nhsel = 0; nhsel < fib_info_num_path(res.fi); nhsel++) {
1082 				nhc = fib_info_nhc(res.fi, nhsel);
1083 				update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1084 						      jiffies + net->ipv4.ip_rt_mtu_expires);
1085 			}
1086 			goto out;
1087 		}
1088 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1089 		nhc = FIB_RES_NHC(res);
1090 		update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1091 				      jiffies + net->ipv4.ip_rt_mtu_expires);
1092 	}
1093 out:
1094 	rcu_read_unlock();
1095 }
1096 
1097 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1098 			      struct sk_buff *skb, u32 mtu,
1099 			      bool confirm_neigh)
1100 {
1101 	struct rtable *rt = dst_rtable(dst);
1102 	struct flowi4 fl4;
1103 
1104 	ip_rt_build_flow_key(&fl4, sk, skb);
1105 
1106 	/* Don't make lookup fail for bridged encapsulations */
1107 	if (skb && netif_is_any_bridge_port(skb->dev))
1108 		fl4.flowi4_oif = 0;
1109 
1110 	__ip_rt_update_pmtu(rt, &fl4, mtu);
1111 }
1112 
1113 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1114 		      int oif, u8 protocol)
1115 {
1116 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1117 	struct flowi4 fl4;
1118 	struct rtable *rt;
1119 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
1120 
1121 	__build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1122 			 0);
1123 	rt = __ip_route_output_key(net, &fl4);
1124 	if (!IS_ERR(rt)) {
1125 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1126 		ip_rt_put(rt);
1127 	}
1128 }
1129 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1130 
1131 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1132 {
1133 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1134 	struct flowi4 fl4;
1135 	struct rtable *rt;
1136 
1137 	__build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1138 
1139 	if (!fl4.flowi4_mark)
1140 		fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1141 
1142 	rt = __ip_route_output_key(sock_net(sk), &fl4);
1143 	if (!IS_ERR(rt)) {
1144 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1145 		ip_rt_put(rt);
1146 	}
1147 }
1148 
1149 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1150 {
1151 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1152 	struct flowi4 fl4;
1153 	struct rtable *rt;
1154 	struct dst_entry *odst = NULL;
1155 	bool new = false;
1156 	struct net *net = sock_net(sk);
1157 
1158 	bh_lock_sock(sk);
1159 
1160 	if (!ip_sk_accept_pmtu(sk))
1161 		goto out;
1162 
1163 	odst = sk_dst_get(sk);
1164 
1165 	if (sock_owned_by_user(sk) || !odst) {
1166 		__ipv4_sk_update_pmtu(skb, sk, mtu);
1167 		goto out;
1168 	}
1169 
1170 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1171 
1172 	rt = dst_rtable(odst);
1173 	if (READ_ONCE(odst->obsolete) && !odst->ops->check(odst, 0)) {
1174 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1175 		if (IS_ERR(rt))
1176 			goto out;
1177 
1178 		new = true;
1179 	}
1180 
1181 	__ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu);
1182 
1183 	if (!dst_check(&rt->dst, 0)) {
1184 		if (new)
1185 			dst_release(&rt->dst);
1186 
1187 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1188 		if (IS_ERR(rt))
1189 			goto out;
1190 
1191 		new = true;
1192 	}
1193 
1194 	if (new)
1195 		sk_dst_set(sk, &rt->dst);
1196 
1197 out:
1198 	bh_unlock_sock(sk);
1199 	dst_release(odst);
1200 }
1201 
1202 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1203 		   int oif, u8 protocol)
1204 {
1205 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1206 	struct flowi4 fl4;
1207 	struct rtable *rt;
1208 
1209 	__build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1210 	rt = __ip_route_output_key(net, &fl4);
1211 	if (!IS_ERR(rt)) {
1212 		__ip_do_redirect(rt, skb, &fl4, false);
1213 		ip_rt_put(rt);
1214 	}
1215 }
1216 EXPORT_SYMBOL_GPL(ipv4_redirect);
1217 
1218 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1219 {
1220 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1221 	struct flowi4 fl4;
1222 	struct rtable *rt;
1223 	struct net *net = sock_net(sk);
1224 
1225 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1226 	rt = __ip_route_output_key(net, &fl4);
1227 	if (!IS_ERR(rt)) {
1228 		__ip_do_redirect(rt, skb, &fl4, false);
1229 		ip_rt_put(rt);
1230 	}
1231 }
1232 
1233 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1234 							 u32 cookie)
1235 {
1236 	struct rtable *rt = dst_rtable(dst);
1237 
1238 	/* All IPV4 dsts are created with ->obsolete set to the value
1239 	 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1240 	 * into this function always.
1241 	 *
1242 	 * When a PMTU/redirect information update invalidates a route,
1243 	 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1244 	 * DST_OBSOLETE_DEAD.
1245 	 */
1246 	if (READ_ONCE(dst->obsolete) != DST_OBSOLETE_FORCE_CHK ||
1247 	    rt_is_expired(rt))
1248 		return NULL;
1249 	return dst;
1250 }
1251 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1252 
1253 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1254 {
1255 	struct inet_skb_parm parm;
1256 	struct net_device *dev;
1257 	int res;
1258 
1259 	/* Recompile ip options since IPCB may not be valid anymore.
1260 	 * Also check we have a reasonable ipv4 header.
1261 	 */
1262 	if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1263 	    ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1264 		return;
1265 
1266 	memset(&parm, 0, sizeof(parm));
1267 	if (ip_hdr(skb)->ihl > 5) {
1268 		if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1269 			return;
1270 		parm.opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1271 
1272 		rcu_read_lock();
1273 		dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1274 		res = __ip_options_compile(dev_net(dev), &parm.opt, skb, NULL);
1275 		rcu_read_unlock();
1276 
1277 		if (res)
1278 			return;
1279 	}
1280 	__icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &parm);
1281 }
1282 
1283 static void ipv4_link_failure(struct sk_buff *skb)
1284 {
1285 	struct rtable *rt;
1286 
1287 	ipv4_send_dest_unreach(skb);
1288 
1289 	rt = skb_rtable(skb);
1290 	if (rt)
1291 		dst_set_expires(&rt->dst, 0);
1292 }
1293 
1294 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1295 {
1296 	pr_debug("%s: %pI4 -> %pI4, %s\n",
1297 		 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1298 		 skb->dev ? skb->dev->name : "?");
1299 	kfree_skb(skb);
1300 	WARN_ON_ONCE(1);
1301 	return 0;
1302 }
1303 
1304 /*
1305  * We do not cache source address of outgoing interface,
1306  * because it is used only by IP RR, TS and SRR options,
1307  * so that it out of fast path.
1308  *
1309  * BTW remember: "addr" is allowed to be not aligned
1310  * in IP options!
1311  */
1312 
1313 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1314 {
1315 	__be32 src;
1316 
1317 	if (rt_is_output_route(rt))
1318 		src = ip_hdr(skb)->saddr;
1319 	else {
1320 		struct fib_result res;
1321 		struct iphdr *iph = ip_hdr(skb);
1322 		struct flowi4 fl4 = {
1323 			.daddr = iph->daddr,
1324 			.saddr = iph->saddr,
1325 			.flowi4_dscp = ip4h_dscp(iph),
1326 			.flowi4_oif = rt->dst.dev->ifindex,
1327 			.flowi4_iif = skb->dev->ifindex,
1328 			.flowi4_mark = skb->mark,
1329 		};
1330 
1331 		rcu_read_lock();
1332 		if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1333 			src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1334 		else
1335 			src = inet_select_addr(rt->dst.dev,
1336 					       rt_nexthop(rt, iph->daddr),
1337 					       RT_SCOPE_UNIVERSE);
1338 		rcu_read_unlock();
1339 	}
1340 	memcpy(addr, &src, 4);
1341 }
1342 
1343 #ifdef CONFIG_IP_ROUTE_CLASSID
1344 static void set_class_tag(struct rtable *rt, u32 tag)
1345 {
1346 	if (!(rt->dst.tclassid & 0xFFFF))
1347 		rt->dst.tclassid |= tag & 0xFFFF;
1348 	if (!(rt->dst.tclassid & 0xFFFF0000))
1349 		rt->dst.tclassid |= tag & 0xFFFF0000;
1350 }
1351 #endif
1352 
1353 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1354 {
1355 	unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1356 	unsigned int advmss;
1357 	struct net *net;
1358 
1359 	rcu_read_lock();
1360 	net = dst_dev_net_rcu(dst);
1361 	advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1362 				   net->ipv4.ip_rt_min_advmss);
1363 	rcu_read_unlock();
1364 
1365 	return min(advmss, IPV4_MAX_PMTU - header_size);
1366 }
1367 
1368 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1369 {
1370 	return ip_dst_mtu_maybe_forward(dst, false);
1371 }
1372 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1373 
1374 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1375 {
1376 	struct fnhe_hash_bucket *hash;
1377 	struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1378 	u32 hval = fnhe_hashfun(daddr);
1379 
1380 	spin_lock_bh(&fnhe_lock);
1381 
1382 	hash = rcu_dereference_protected(nhc->nhc_exceptions,
1383 					 lockdep_is_held(&fnhe_lock));
1384 	hash += hval;
1385 
1386 	fnhe_p = &hash->chain;
1387 	fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1388 	while (fnhe) {
1389 		if (fnhe->fnhe_daddr == daddr) {
1390 			rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1391 				fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1392 			/* set fnhe_daddr to 0 to ensure it won't bind with
1393 			 * new dsts in rt_bind_exception().
1394 			 */
1395 			fnhe->fnhe_daddr = 0;
1396 			fnhe_flush_routes(fnhe);
1397 			kfree_rcu(fnhe, rcu);
1398 			break;
1399 		}
1400 		fnhe_p = &fnhe->fnhe_next;
1401 		fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1402 						 lockdep_is_held(&fnhe_lock));
1403 	}
1404 
1405 	spin_unlock_bh(&fnhe_lock);
1406 }
1407 
1408 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1409 					       __be32 daddr)
1410 {
1411 	struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1412 	struct fib_nh_exception *fnhe;
1413 	u32 hval;
1414 
1415 	if (!hash)
1416 		return NULL;
1417 
1418 	hval = fnhe_hashfun(daddr);
1419 
1420 	for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1421 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
1422 		if (fnhe->fnhe_daddr == daddr) {
1423 			if (fnhe->fnhe_expires &&
1424 			    time_after(jiffies, fnhe->fnhe_expires)) {
1425 				ip_del_fnhe(nhc, daddr);
1426 				break;
1427 			}
1428 			return fnhe;
1429 		}
1430 	}
1431 	return NULL;
1432 }
1433 
1434 /* MTU selection:
1435  * 1. mtu on route is locked - use it
1436  * 2. mtu from nexthop exception
1437  * 3. mtu from egress device
1438  */
1439 
1440 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1441 {
1442 	struct fib_nh_common *nhc = res->nhc;
1443 	struct net_device *dev = nhc->nhc_dev;
1444 	struct fib_info *fi = res->fi;
1445 	u32 mtu = 0;
1446 
1447 	if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1448 	    fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1449 		mtu = fi->fib_mtu;
1450 
1451 	if (likely(!mtu)) {
1452 		struct fib_nh_exception *fnhe;
1453 
1454 		fnhe = find_exception(nhc, daddr);
1455 		if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1456 			mtu = fnhe->fnhe_pmtu;
1457 	}
1458 
1459 	if (likely(!mtu))
1460 		mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1461 
1462 	return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1463 }
1464 
1465 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1466 			      __be32 daddr, const bool do_cache)
1467 {
1468 	bool ret = false;
1469 
1470 	spin_lock_bh(&fnhe_lock);
1471 
1472 	if (daddr == fnhe->fnhe_daddr) {
1473 		struct rtable __rcu **porig;
1474 		struct rtable *orig;
1475 		int genid = fnhe_genid(dev_net(rt->dst.dev));
1476 
1477 		if (rt_is_input_route(rt))
1478 			porig = &fnhe->fnhe_rth_input;
1479 		else
1480 			porig = &fnhe->fnhe_rth_output;
1481 		orig = rcu_dereference(*porig);
1482 
1483 		if (fnhe->fnhe_genid != genid) {
1484 			fnhe->fnhe_genid = genid;
1485 			fnhe->fnhe_gw = 0;
1486 			fnhe->fnhe_pmtu = 0;
1487 			fnhe->fnhe_expires = 0;
1488 			fnhe->fnhe_mtu_locked = false;
1489 			fnhe_flush_routes(fnhe);
1490 			orig = NULL;
1491 		}
1492 		fill_route_from_fnhe(rt, fnhe);
1493 		if (!rt->rt_gw4) {
1494 			rt->rt_gw4 = daddr;
1495 			rt->rt_gw_family = AF_INET;
1496 		}
1497 
1498 		if (do_cache) {
1499 			dst_hold(&rt->dst);
1500 			rcu_assign_pointer(*porig, rt);
1501 			if (orig) {
1502 				dst_dev_put(&orig->dst);
1503 				dst_release(&orig->dst);
1504 			}
1505 			ret = true;
1506 		}
1507 
1508 		fnhe->fnhe_stamp = jiffies;
1509 	}
1510 	spin_unlock_bh(&fnhe_lock);
1511 
1512 	return ret;
1513 }
1514 
1515 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1516 {
1517 	struct rtable *orig, *prev, **p;
1518 	bool ret = true;
1519 
1520 	if (rt_is_input_route(rt)) {
1521 		p = (struct rtable **)&nhc->nhc_rth_input;
1522 	} else {
1523 		p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1524 	}
1525 	orig = *p;
1526 
1527 	/* hold dst before doing cmpxchg() to avoid race condition
1528 	 * on this dst
1529 	 */
1530 	dst_hold(&rt->dst);
1531 	prev = cmpxchg(p, orig, rt);
1532 	if (prev == orig) {
1533 		if (orig) {
1534 			rt_add_uncached_list(orig);
1535 			dst_release(&orig->dst);
1536 		}
1537 	} else {
1538 		dst_release(&rt->dst);
1539 		ret = false;
1540 	}
1541 
1542 	return ret;
1543 }
1544 
1545 struct uncached_list {
1546 	spinlock_t		lock;
1547 	struct list_head	head;
1548 };
1549 
1550 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1551 
1552 void rt_add_uncached_list(struct rtable *rt)
1553 {
1554 	struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1555 
1556 	rt->dst.rt_uncached_list = ul;
1557 
1558 	spin_lock_bh(&ul->lock);
1559 	list_add_tail(&rt->dst.rt_uncached, &ul->head);
1560 	spin_unlock_bh(&ul->lock);
1561 }
1562 
1563 void rt_del_uncached_list(struct rtable *rt)
1564 {
1565 	struct uncached_list *ul = rt->dst.rt_uncached_list;
1566 
1567 	if (ul) {
1568 		spin_lock_bh(&ul->lock);
1569 		list_del_init(&rt->dst.rt_uncached);
1570 		spin_unlock_bh(&ul->lock);
1571 	}
1572 }
1573 
1574 static void ipv4_dst_destroy(struct dst_entry *dst)
1575 {
1576 	ip_dst_metrics_put(dst);
1577 	rt_del_uncached_list(dst_rtable(dst));
1578 }
1579 
1580 void rt_flush_dev(struct net_device *dev)
1581 {
1582 	struct rtable *rt, *safe;
1583 	int cpu;
1584 
1585 	for_each_possible_cpu(cpu) {
1586 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1587 
1588 		if (list_empty(&ul->head))
1589 			continue;
1590 
1591 		spin_lock_bh(&ul->lock);
1592 		list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1593 			if (rt->dst.dev != dev)
1594 				continue;
1595 			rt->dst.dev = blackhole_netdev;
1596 			netdev_ref_replace(dev, blackhole_netdev,
1597 					   &rt->dst.dev_tracker, GFP_ATOMIC);
1598 			list_del_init(&rt->dst.rt_uncached);
1599 		}
1600 		spin_unlock_bh(&ul->lock);
1601 	}
1602 }
1603 
1604 static bool rt_cache_valid(const struct rtable *rt)
1605 {
1606 	return	rt &&
1607 		READ_ONCE(rt->dst.obsolete) == DST_OBSOLETE_FORCE_CHK &&
1608 		!rt_is_expired(rt);
1609 }
1610 
1611 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1612 			   const struct fib_result *res,
1613 			   struct fib_nh_exception *fnhe,
1614 			   struct fib_info *fi, u16 type, u32 itag,
1615 			   const bool do_cache)
1616 {
1617 	bool cached = false;
1618 
1619 	if (fi) {
1620 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1621 
1622 		if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1623 			rt->rt_uses_gateway = 1;
1624 			rt->rt_gw_family = nhc->nhc_gw_family;
1625 			/* only INET and INET6 are supported */
1626 			if (likely(nhc->nhc_gw_family == AF_INET))
1627 				rt->rt_gw4 = nhc->nhc_gw.ipv4;
1628 			else
1629 				rt->rt_gw6 = nhc->nhc_gw.ipv6;
1630 		}
1631 
1632 		ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1633 
1634 #ifdef CONFIG_IP_ROUTE_CLASSID
1635 		if (nhc->nhc_family == AF_INET) {
1636 			struct fib_nh *nh;
1637 
1638 			nh = container_of(nhc, struct fib_nh, nh_common);
1639 			rt->dst.tclassid = nh->nh_tclassid;
1640 		}
1641 #endif
1642 		rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1643 		if (unlikely(fnhe))
1644 			cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1645 		else if (do_cache)
1646 			cached = rt_cache_route(nhc, rt);
1647 		if (unlikely(!cached)) {
1648 			/* Routes we intend to cache in nexthop exception or
1649 			 * FIB nexthop have the DST_NOCACHE bit clear.
1650 			 * However, if we are unsuccessful at storing this
1651 			 * route into the cache we really need to set it.
1652 			 */
1653 			if (!rt->rt_gw4) {
1654 				rt->rt_gw_family = AF_INET;
1655 				rt->rt_gw4 = daddr;
1656 			}
1657 			rt_add_uncached_list(rt);
1658 		}
1659 	} else
1660 		rt_add_uncached_list(rt);
1661 
1662 #ifdef CONFIG_IP_ROUTE_CLASSID
1663 #ifdef CONFIG_IP_MULTIPLE_TABLES
1664 	set_class_tag(rt, res->tclassid);
1665 #endif
1666 	set_class_tag(rt, itag);
1667 #endif
1668 }
1669 
1670 struct rtable *rt_dst_alloc(struct net_device *dev,
1671 			    unsigned int flags, u16 type,
1672 			    bool noxfrm)
1673 {
1674 	struct rtable *rt;
1675 
1676 	rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1677 		       (noxfrm ? DST_NOXFRM : 0));
1678 
1679 	if (rt) {
1680 		rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1681 		rt->rt_flags = flags;
1682 		rt->rt_type = type;
1683 		rt->rt_is_input = 0;
1684 		rt->rt_iif = 0;
1685 		rt->rt_pmtu = 0;
1686 		rt->rt_mtu_locked = 0;
1687 		rt->rt_uses_gateway = 0;
1688 		rt->rt_gw_family = 0;
1689 		rt->rt_gw4 = 0;
1690 
1691 		rt->dst.output = ip_output;
1692 		if (flags & RTCF_LOCAL)
1693 			rt->dst.input = ip_local_deliver;
1694 	}
1695 
1696 	return rt;
1697 }
1698 EXPORT_SYMBOL(rt_dst_alloc);
1699 
1700 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1701 {
1702 	struct rtable *new_rt;
1703 
1704 	new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1705 			   rt->dst.flags);
1706 
1707 	if (new_rt) {
1708 		new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1709 		new_rt->rt_flags = rt->rt_flags;
1710 		new_rt->rt_type = rt->rt_type;
1711 		new_rt->rt_is_input = rt->rt_is_input;
1712 		new_rt->rt_iif = rt->rt_iif;
1713 		new_rt->rt_pmtu = rt->rt_pmtu;
1714 		new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1715 		new_rt->rt_gw_family = rt->rt_gw_family;
1716 		if (rt->rt_gw_family == AF_INET)
1717 			new_rt->rt_gw4 = rt->rt_gw4;
1718 		else if (rt->rt_gw_family == AF_INET6)
1719 			new_rt->rt_gw6 = rt->rt_gw6;
1720 
1721 		new_rt->dst.input = READ_ONCE(rt->dst.input);
1722 		new_rt->dst.output = READ_ONCE(rt->dst.output);
1723 		new_rt->dst.error = rt->dst.error;
1724 		new_rt->dst.lastuse = jiffies;
1725 		new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1726 	}
1727 	return new_rt;
1728 }
1729 
1730 /* called in rcu_read_lock() section */
1731 enum skb_drop_reason
1732 ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1733 		      dscp_t dscp, struct net_device *dev,
1734 		      struct in_device *in_dev, u32 *itag)
1735 {
1736 	enum skb_drop_reason reason;
1737 
1738 	/* Primary sanity checks. */
1739 	if (!in_dev)
1740 		return SKB_DROP_REASON_NOT_SPECIFIED;
1741 
1742 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1743 		return SKB_DROP_REASON_IP_INVALID_SOURCE;
1744 
1745 	if (skb->protocol != htons(ETH_P_IP))
1746 		return SKB_DROP_REASON_INVALID_PROTO;
1747 
1748 	if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1749 		return SKB_DROP_REASON_IP_LOCALNET;
1750 
1751 	if (ipv4_is_zeronet(saddr)) {
1752 		if (!ipv4_is_local_multicast(daddr) &&
1753 		    ip_hdr(skb)->protocol != IPPROTO_IGMP)
1754 			return SKB_DROP_REASON_IP_INVALID_SOURCE;
1755 	} else {
1756 		reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
1757 						    dev, in_dev, itag);
1758 		if (reason)
1759 			return reason;
1760 	}
1761 	return SKB_NOT_DROPPED_YET;
1762 }
1763 
1764 /* called in rcu_read_lock() section */
1765 static enum skb_drop_reason
1766 ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1767 		  dscp_t dscp, struct net_device *dev, int our)
1768 {
1769 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1770 	unsigned int flags = RTCF_MULTICAST;
1771 	enum skb_drop_reason reason;
1772 	struct rtable *rth;
1773 	u32 itag = 0;
1774 
1775 	reason = ip_mc_validate_source(skb, daddr, saddr, dscp, dev, in_dev,
1776 				       &itag);
1777 	if (reason)
1778 		return reason;
1779 
1780 	if (our)
1781 		flags |= RTCF_LOCAL;
1782 
1783 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1784 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
1785 
1786 	rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1787 			   false);
1788 	if (!rth)
1789 		return SKB_DROP_REASON_NOMEM;
1790 
1791 #ifdef CONFIG_IP_ROUTE_CLASSID
1792 	rth->dst.tclassid = itag;
1793 #endif
1794 	rth->dst.output = ip_rt_bug;
1795 	rth->rt_is_input= 1;
1796 
1797 #ifdef CONFIG_IP_MROUTE
1798 	if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1799 		rth->dst.input = ip_mr_input;
1800 #endif
1801 	RT_CACHE_STAT_INC(in_slow_mc);
1802 
1803 	skb_dst_drop(skb);
1804 	skb_dst_set(skb, &rth->dst);
1805 	return SKB_NOT_DROPPED_YET;
1806 }
1807 
1808 
1809 static void ip_handle_martian_source(struct net_device *dev,
1810 				     struct in_device *in_dev,
1811 				     struct sk_buff *skb,
1812 				     __be32 daddr,
1813 				     __be32 saddr)
1814 {
1815 	RT_CACHE_STAT_INC(in_martian_src);
1816 #ifdef CONFIG_IP_ROUTE_VERBOSE
1817 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1818 		/*
1819 		 *	RFC1812 recommendation, if source is martian,
1820 		 *	the only hint is MAC header.
1821 		 */
1822 		pr_warn("martian source (src=%pI4, dst=%pI4, dev=%s)\n",
1823 			&saddr, &daddr, dev->name);
1824 		if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1825 			print_hex_dump(KERN_WARNING, "ll header: ",
1826 				       DUMP_PREFIX_OFFSET, 16, 1,
1827 				       skb_mac_header(skb),
1828 				       dev->hard_header_len, false);
1829 		}
1830 	}
1831 #endif
1832 }
1833 
1834 /* called in rcu_read_lock() section */
1835 static enum skb_drop_reason
1836 __mkroute_input(struct sk_buff *skb, const struct fib_result *res,
1837 		struct in_device *in_dev, __be32 daddr,
1838 		__be32 saddr, dscp_t dscp)
1839 {
1840 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1841 	struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1842 	struct net_device *dev = nhc->nhc_dev;
1843 	struct fib_nh_exception *fnhe;
1844 	struct rtable *rth;
1845 	int err;
1846 	struct in_device *out_dev;
1847 	bool do_cache;
1848 	u32 itag = 0;
1849 
1850 	/* get a working reference to the output device */
1851 	out_dev = __in_dev_get_rcu(dev);
1852 	if (!out_dev) {
1853 		net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1854 		return reason;
1855 	}
1856 
1857 	err = fib_validate_source(skb, saddr, daddr, dscp, FIB_RES_OIF(*res),
1858 				  in_dev->dev, in_dev, &itag);
1859 	if (err < 0) {
1860 		reason = -err;
1861 		ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1862 					 saddr);
1863 
1864 		goto cleanup;
1865 	}
1866 
1867 	do_cache = res->fi && !itag;
1868 	if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1869 	    skb->protocol == htons(ETH_P_IP)) {
1870 		__be32 gw;
1871 
1872 		gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1873 		if (IN_DEV_SHARED_MEDIA(out_dev) ||
1874 		    inet_addr_onlink(out_dev, saddr, gw))
1875 			IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1876 	}
1877 
1878 	if (skb->protocol != htons(ETH_P_IP)) {
1879 		/* Not IP (i.e. ARP). Do not create route, if it is
1880 		 * invalid for proxy arp. DNAT routes are always valid.
1881 		 *
1882 		 * Proxy arp feature have been extended to allow, ARP
1883 		 * replies back to the same interface, to support
1884 		 * Private VLAN switch technologies. See arp.c.
1885 		 */
1886 		if (out_dev == in_dev &&
1887 		    IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1888 			reason = SKB_DROP_REASON_ARP_PVLAN_DISABLE;
1889 			goto cleanup;
1890 		}
1891 	}
1892 
1893 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1894 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
1895 
1896 	fnhe = find_exception(nhc, daddr);
1897 	if (do_cache) {
1898 		if (fnhe)
1899 			rth = rcu_dereference(fnhe->fnhe_rth_input);
1900 		else
1901 			rth = rcu_dereference(nhc->nhc_rth_input);
1902 		if (rt_cache_valid(rth)) {
1903 			skb_dst_set_noref(skb, &rth->dst);
1904 			goto out;
1905 		}
1906 	}
1907 
1908 	rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1909 			   IN_DEV_ORCONF(out_dev, NOXFRM));
1910 	if (!rth) {
1911 		reason = SKB_DROP_REASON_NOMEM;
1912 		goto cleanup;
1913 	}
1914 
1915 	rth->rt_is_input = 1;
1916 	RT_CACHE_STAT_INC(in_slow_tot);
1917 
1918 	rth->dst.input = ip_forward;
1919 
1920 	rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1921 		       do_cache);
1922 	lwtunnel_set_redirect(&rth->dst);
1923 	skb_dst_set(skb, &rth->dst);
1924 out:
1925 	reason = SKB_NOT_DROPPED_YET;
1926 cleanup:
1927 	return reason;
1928 }
1929 
1930 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1931 /* To make ICMP packets follow the right flow, the multipath hash is
1932  * calculated from the inner IP addresses.
1933  */
1934 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1935 				 struct flow_keys *hash_keys)
1936 {
1937 	const struct iphdr *outer_iph = ip_hdr(skb);
1938 	const struct iphdr *key_iph = outer_iph;
1939 	const struct iphdr *inner_iph;
1940 	const struct icmphdr *icmph;
1941 	struct iphdr _inner_iph;
1942 	struct icmphdr _icmph;
1943 
1944 	if (likely(outer_iph->protocol != IPPROTO_ICMP))
1945 		goto out;
1946 
1947 	if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1948 		goto out;
1949 
1950 	icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1951 				   &_icmph);
1952 	if (!icmph)
1953 		goto out;
1954 
1955 	if (!icmp_is_err(icmph->type))
1956 		goto out;
1957 
1958 	inner_iph = skb_header_pointer(skb,
1959 				       outer_iph->ihl * 4 + sizeof(_icmph),
1960 				       sizeof(_inner_iph), &_inner_iph);
1961 	if (!inner_iph)
1962 		goto out;
1963 
1964 	key_iph = inner_iph;
1965 out:
1966 	hash_keys->addrs.v4addrs.src = key_iph->saddr;
1967 	hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1968 }
1969 
1970 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1971 					   const struct sk_buff *skb,
1972 					   bool *p_has_inner)
1973 {
1974 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1975 	struct flow_keys keys, hash_keys;
1976 
1977 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1978 		return 0;
1979 
1980 	memset(&hash_keys, 0, sizeof(hash_keys));
1981 	skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1982 
1983 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1984 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1985 		hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1986 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1987 		hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1988 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1989 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
1990 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1991 		hash_keys.ports.src = keys.ports.src;
1992 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1993 		hash_keys.ports.dst = keys.ports.dst;
1994 
1995 	*p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1996 	return fib_multipath_hash_from_keys(net, &hash_keys);
1997 }
1998 
1999 static u32 fib_multipath_custom_hash_inner(const struct net *net,
2000 					   const struct sk_buff *skb,
2001 					   bool has_inner)
2002 {
2003 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2004 	struct flow_keys keys, hash_keys;
2005 
2006 	/* We assume the packet carries an encapsulation, but if none was
2007 	 * encountered during dissection of the outer flow, then there is no
2008 	 * point in calling the flow dissector again.
2009 	 */
2010 	if (!has_inner)
2011 		return 0;
2012 
2013 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
2014 		return 0;
2015 
2016 	memset(&hash_keys, 0, sizeof(hash_keys));
2017 	skb_flow_dissect_flow_keys(skb, &keys, 0);
2018 
2019 	if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
2020 		return 0;
2021 
2022 	if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2023 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2024 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2025 			hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2026 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2027 			hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2028 	} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2029 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2030 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2031 			hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2032 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2033 			hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2034 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2035 			hash_keys.tags.flow_label = keys.tags.flow_label;
2036 	}
2037 
2038 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2039 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
2040 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2041 		hash_keys.ports.src = keys.ports.src;
2042 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2043 		hash_keys.ports.dst = keys.ports.dst;
2044 
2045 	return fib_multipath_hash_from_keys(net, &hash_keys);
2046 }
2047 
2048 static u32 fib_multipath_custom_hash_skb(const struct net *net,
2049 					 const struct sk_buff *skb)
2050 {
2051 	u32 mhash, mhash_inner;
2052 	bool has_inner = true;
2053 
2054 	mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2055 	mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2056 
2057 	return jhash_2words(mhash, mhash_inner, 0);
2058 }
2059 
2060 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2061 					 const struct flowi4 *fl4)
2062 {
2063 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2064 	struct flow_keys hash_keys;
2065 
2066 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2067 		return 0;
2068 
2069 	memset(&hash_keys, 0, sizeof(hash_keys));
2070 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2071 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2072 		hash_keys.addrs.v4addrs.src = fl4->saddr;
2073 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2074 		hash_keys.addrs.v4addrs.dst = fl4->daddr;
2075 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2076 		hash_keys.basic.ip_proto = fl4->flowi4_proto;
2077 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) {
2078 		if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT)
2079 			hash_keys.ports.src = (__force __be16)get_random_u16();
2080 		else
2081 			hash_keys.ports.src = fl4->fl4_sport;
2082 	}
2083 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2084 		hash_keys.ports.dst = fl4->fl4_dport;
2085 
2086 	return fib_multipath_hash_from_keys(net, &hash_keys);
2087 }
2088 
2089 /* if skb is set it will be used and fl4 can be NULL */
2090 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2091 		       const struct sk_buff *skb, struct flow_keys *flkeys)
2092 {
2093 	u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2094 	struct flow_keys hash_keys;
2095 	u32 mhash = 0;
2096 
2097 	switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2098 	case 0:
2099 		memset(&hash_keys, 0, sizeof(hash_keys));
2100 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2101 		if (skb) {
2102 			ip_multipath_l3_keys(skb, &hash_keys);
2103 		} else {
2104 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2105 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2106 		}
2107 		mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2108 		break;
2109 	case 1:
2110 		/* skb is currently provided only when forwarding */
2111 		if (skb) {
2112 			unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2113 			struct flow_keys keys;
2114 
2115 			/* short-circuit if we already have L4 hash present */
2116 			if (skb->l4_hash)
2117 				return skb_get_hash_raw(skb) >> 1;
2118 
2119 			memset(&hash_keys, 0, sizeof(hash_keys));
2120 
2121 			if (!flkeys) {
2122 				skb_flow_dissect_flow_keys(skb, &keys, flag);
2123 				flkeys = &keys;
2124 			}
2125 
2126 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2127 			hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2128 			hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2129 			hash_keys.ports.src = flkeys->ports.src;
2130 			hash_keys.ports.dst = flkeys->ports.dst;
2131 			hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2132 		} else {
2133 			memset(&hash_keys, 0, sizeof(hash_keys));
2134 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2135 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2136 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2137 			if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT)
2138 				hash_keys.ports.src = (__force __be16)get_random_u16();
2139 			else
2140 				hash_keys.ports.src = fl4->fl4_sport;
2141 			hash_keys.ports.dst = fl4->fl4_dport;
2142 			hash_keys.basic.ip_proto = fl4->flowi4_proto;
2143 		}
2144 		mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2145 		break;
2146 	case 2:
2147 		memset(&hash_keys, 0, sizeof(hash_keys));
2148 		/* skb is currently provided only when forwarding */
2149 		if (skb) {
2150 			struct flow_keys keys;
2151 
2152 			skb_flow_dissect_flow_keys(skb, &keys, 0);
2153 			/* Inner can be v4 or v6 */
2154 			if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2155 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2156 				hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2157 				hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2158 			} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2159 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2160 				hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2161 				hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2162 				hash_keys.tags.flow_label = keys.tags.flow_label;
2163 				hash_keys.basic.ip_proto = keys.basic.ip_proto;
2164 			} else {
2165 				/* Same as case 0 */
2166 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2167 				ip_multipath_l3_keys(skb, &hash_keys);
2168 			}
2169 		} else {
2170 			/* Same as case 0 */
2171 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2172 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2173 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2174 		}
2175 		mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2176 		break;
2177 	case 3:
2178 		if (skb)
2179 			mhash = fib_multipath_custom_hash_skb(net, skb);
2180 		else
2181 			mhash = fib_multipath_custom_hash_fl4(net, fl4);
2182 		break;
2183 	}
2184 
2185 	if (multipath_hash)
2186 		mhash = jhash_2words(mhash, multipath_hash, 0);
2187 
2188 	return mhash >> 1;
2189 }
2190 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2191 
2192 static enum skb_drop_reason
2193 ip_mkroute_input(struct sk_buff *skb, struct fib_result *res,
2194 		 struct in_device *in_dev, __be32 daddr,
2195 		 __be32 saddr, dscp_t dscp, struct flow_keys *hkeys)
2196 {
2197 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2198 	if (res->fi && fib_info_num_path(res->fi) > 1) {
2199 		int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2200 
2201 		fib_select_multipath(res, h, NULL);
2202 		IPCB(skb)->flags |= IPSKB_MULTIPATH;
2203 	}
2204 #endif
2205 
2206 	/* create a routing cache entry */
2207 	return __mkroute_input(skb, res, in_dev, daddr, saddr, dscp);
2208 }
2209 
2210 /* Implements all the saddr-related checks as ip_route_input_slow(),
2211  * assuming daddr is valid and the destination is not a local broadcast one.
2212  * Uses the provided hint instead of performing a route lookup.
2213  */
2214 enum skb_drop_reason
2215 ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2216 		  dscp_t dscp, struct net_device *dev,
2217 		  const struct sk_buff *hint)
2218 {
2219 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2220 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2221 	struct rtable *rt = skb_rtable(hint);
2222 	struct net *net = dev_net(dev);
2223 	u32 tag = 0;
2224 
2225 	if (!in_dev)
2226 		return reason;
2227 
2228 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2229 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2230 		goto martian_source;
2231 	}
2232 
2233 	if (ipv4_is_zeronet(saddr)) {
2234 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2235 		goto martian_source;
2236 	}
2237 
2238 	if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2239 		reason = SKB_DROP_REASON_IP_LOCALNET;
2240 		goto martian_source;
2241 	}
2242 
2243 	if (!(rt->rt_flags & RTCF_LOCAL))
2244 		goto skip_validate_source;
2245 
2246 	reason = fib_validate_source_reason(skb, saddr, daddr, dscp, 0, dev,
2247 					    in_dev, &tag);
2248 	if (reason)
2249 		goto martian_source;
2250 
2251 skip_validate_source:
2252 	skb_dst_copy(skb, hint);
2253 	return SKB_NOT_DROPPED_YET;
2254 
2255 martian_source:
2256 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2257 	return reason;
2258 }
2259 
2260 /* get device for dst_alloc with local routes */
2261 static struct net_device *ip_rt_get_dev(struct net *net,
2262 					const struct fib_result *res)
2263 {
2264 	struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2265 	struct net_device *dev = NULL;
2266 
2267 	if (nhc)
2268 		dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2269 
2270 	return dev ? : net->loopback_dev;
2271 }
2272 
2273 /*
2274  *	NOTE. We drop all the packets that has local source
2275  *	addresses, because every properly looped back packet
2276  *	must have correct destination already attached by output routine.
2277  *	Changes in the enforced policies must be applied also to
2278  *	ip_route_use_hint().
2279  *
2280  *	Such approach solves two big problems:
2281  *	1. Not simplex devices are handled properly.
2282  *	2. IP spoofing attempts are filtered with 100% of guarantee.
2283  *	called with rcu_read_lock()
2284  */
2285 
2286 static enum skb_drop_reason
2287 ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2288 		    dscp_t dscp, struct net_device *dev,
2289 		    struct fib_result *res)
2290 {
2291 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2292 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2293 	struct flow_keys *flkeys = NULL, _flkeys;
2294 	struct net    *net = dev_net(dev);
2295 	struct ip_tunnel_info *tun_info;
2296 	int		err = -EINVAL;
2297 	unsigned int	flags = 0;
2298 	u32		itag = 0;
2299 	struct rtable	*rth;
2300 	struct flowi4	fl4;
2301 	bool do_cache = true;
2302 
2303 	/* IP on this device is disabled. */
2304 
2305 	if (!in_dev)
2306 		goto out;
2307 
2308 	/* Check for the most weird martians, which can be not detected
2309 	 * by fib_lookup.
2310 	 */
2311 
2312 	tun_info = skb_tunnel_info(skb);
2313 	if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2314 		fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2315 	else
2316 		fl4.flowi4_tun_key.tun_id = 0;
2317 	skb_dst_drop(skb);
2318 
2319 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2320 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2321 		goto martian_source;
2322 	}
2323 
2324 	res->fi = NULL;
2325 	res->table = NULL;
2326 	if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2327 		goto brd_input;
2328 
2329 	/* Accept zero addresses only to limited broadcast;
2330 	 * I even do not know to fix it or not. Waiting for complains :-)
2331 	 */
2332 	if (ipv4_is_zeronet(saddr)) {
2333 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2334 		goto martian_source;
2335 	}
2336 
2337 	if (ipv4_is_zeronet(daddr)) {
2338 		reason = SKB_DROP_REASON_IP_INVALID_DEST;
2339 		goto martian_destination;
2340 	}
2341 
2342 	/* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2343 	 * and call it once if daddr or/and saddr are loopback addresses
2344 	 */
2345 	if (ipv4_is_loopback(daddr)) {
2346 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2347 			reason = SKB_DROP_REASON_IP_LOCALNET;
2348 			goto martian_destination;
2349 		}
2350 	} else if (ipv4_is_loopback(saddr)) {
2351 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2352 			reason = SKB_DROP_REASON_IP_LOCALNET;
2353 			goto martian_source;
2354 		}
2355 	}
2356 
2357 	/*
2358 	 *	Now we are ready to route packet.
2359 	 */
2360 	fl4.flowi4_l3mdev = 0;
2361 	fl4.flowi4_oif = 0;
2362 	fl4.flowi4_iif = dev->ifindex;
2363 	fl4.flowi4_mark = skb->mark;
2364 	fl4.flowi4_dscp = dscp;
2365 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2366 	fl4.flowi4_flags = 0;
2367 	fl4.daddr = daddr;
2368 	fl4.saddr = saddr;
2369 	fl4.flowi4_uid = sock_net_uid(net, NULL);
2370 	fl4.flowi4_multipath_hash = 0;
2371 
2372 	if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2373 		flkeys = &_flkeys;
2374 	} else {
2375 		fl4.flowi4_proto = 0;
2376 		fl4.fl4_sport = 0;
2377 		fl4.fl4_dport = 0;
2378 	}
2379 
2380 	err = fib_lookup(net, &fl4, res, 0);
2381 	if (err != 0) {
2382 		if (!IN_DEV_FORWARD(in_dev))
2383 			err = -EHOSTUNREACH;
2384 		goto no_route;
2385 	}
2386 
2387 	if (res->type == RTN_BROADCAST) {
2388 		if (IN_DEV_BFORWARD(in_dev))
2389 			goto make_route;
2390 		/* not do cache if bc_forwarding is enabled */
2391 		if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING))
2392 			do_cache = false;
2393 		goto brd_input;
2394 	}
2395 
2396 	err = -EINVAL;
2397 	if (res->type == RTN_LOCAL) {
2398 		reason = fib_validate_source_reason(skb, saddr, daddr, dscp,
2399 						    0, dev, in_dev, &itag);
2400 		if (reason)
2401 			goto martian_source;
2402 		goto local_input;
2403 	}
2404 
2405 	if (!IN_DEV_FORWARD(in_dev)) {
2406 		err = -EHOSTUNREACH;
2407 		goto no_route;
2408 	}
2409 	if (res->type != RTN_UNICAST) {
2410 		reason = SKB_DROP_REASON_IP_INVALID_DEST;
2411 		goto martian_destination;
2412 	}
2413 
2414 make_route:
2415 	reason = ip_mkroute_input(skb, res, in_dev, daddr, saddr, dscp,
2416 				  flkeys);
2417 
2418 out:
2419 	return reason;
2420 
2421 brd_input:
2422 	if (skb->protocol != htons(ETH_P_IP)) {
2423 		reason = SKB_DROP_REASON_INVALID_PROTO;
2424 		goto out;
2425 	}
2426 
2427 	if (!ipv4_is_zeronet(saddr)) {
2428 		reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
2429 						    dev, in_dev, &itag);
2430 		if (reason)
2431 			goto martian_source;
2432 	}
2433 	flags |= RTCF_BROADCAST;
2434 	res->type = RTN_BROADCAST;
2435 	RT_CACHE_STAT_INC(in_brd);
2436 
2437 local_input:
2438 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2439 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
2440 
2441 	do_cache &= res->fi && !itag;
2442 	if (do_cache) {
2443 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2444 
2445 		rth = rcu_dereference(nhc->nhc_rth_input);
2446 		if (rt_cache_valid(rth)) {
2447 			skb_dst_set_noref(skb, &rth->dst);
2448 			reason = SKB_NOT_DROPPED_YET;
2449 			goto out;
2450 		}
2451 	}
2452 
2453 	rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2454 			   flags | RTCF_LOCAL, res->type, false);
2455 	if (!rth)
2456 		goto e_nobufs;
2457 
2458 	rth->dst.output= ip_rt_bug;
2459 #ifdef CONFIG_IP_ROUTE_CLASSID
2460 	rth->dst.tclassid = itag;
2461 #endif
2462 	rth->rt_is_input = 1;
2463 
2464 	RT_CACHE_STAT_INC(in_slow_tot);
2465 	if (res->type == RTN_UNREACHABLE) {
2466 		rth->dst.input= ip_error;
2467 		rth->dst.error= -err;
2468 		rth->rt_flags	&= ~RTCF_LOCAL;
2469 	}
2470 
2471 	if (do_cache) {
2472 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2473 
2474 		rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2475 		if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2476 			WARN_ON(rth->dst.input == lwtunnel_input);
2477 			rth->dst.lwtstate->orig_input = rth->dst.input;
2478 			rth->dst.input = lwtunnel_input;
2479 		}
2480 
2481 		if (unlikely(!rt_cache_route(nhc, rth)))
2482 			rt_add_uncached_list(rth);
2483 	}
2484 	skb_dst_set(skb, &rth->dst);
2485 	reason = SKB_NOT_DROPPED_YET;
2486 	goto out;
2487 
2488 no_route:
2489 	RT_CACHE_STAT_INC(in_no_route);
2490 	res->type = RTN_UNREACHABLE;
2491 	res->fi = NULL;
2492 	res->table = NULL;
2493 	goto local_input;
2494 
2495 	/*
2496 	 *	Do not cache martian addresses: they should be logged (RFC1812)
2497 	 */
2498 martian_destination:
2499 	RT_CACHE_STAT_INC(in_martian_dst);
2500 #ifdef CONFIG_IP_ROUTE_VERBOSE
2501 	if (IN_DEV_LOG_MARTIANS(in_dev))
2502 		net_warn_ratelimited("martian destination (src=%pI4, dst=%pI4, dev=%s)\n",
2503 				     &saddr, &daddr, dev->name);
2504 #endif
2505 	goto out;
2506 
2507 e_nobufs:
2508 	reason = SKB_DROP_REASON_NOMEM;
2509 	goto out;
2510 
2511 martian_source:
2512 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2513 	goto out;
2514 }
2515 
2516 /* called with rcu_read_lock held */
2517 static enum skb_drop_reason
2518 ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2519 		   dscp_t dscp, struct net_device *dev,
2520 		   struct fib_result *res)
2521 {
2522 	/* Multicast recognition logic is moved from route cache to here.
2523 	 * The problem was that too many Ethernet cards have broken/missing
2524 	 * hardware multicast filters :-( As result the host on multicasting
2525 	 * network acquires a lot of useless route cache entries, sort of
2526 	 * SDR messages from all the world. Now we try to get rid of them.
2527 	 * Really, provided software IP multicast filter is organized
2528 	 * reasonably (at least, hashed), it does not result in a slowdown
2529 	 * comparing with route cache reject entries.
2530 	 * Note, that multicast routers are not affected, because
2531 	 * route cache entry is created eventually.
2532 	 */
2533 	if (ipv4_is_multicast(daddr)) {
2534 		enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2535 		struct in_device *in_dev = __in_dev_get_rcu(dev);
2536 		int our = 0;
2537 
2538 		if (!in_dev)
2539 			return reason;
2540 
2541 		our = ip_check_mc_rcu(in_dev, daddr, saddr,
2542 				      ip_hdr(skb)->protocol);
2543 
2544 		/* check l3 master if no match yet */
2545 		if (!our && netif_is_l3_slave(dev)) {
2546 			struct in_device *l3_in_dev;
2547 
2548 			l3_in_dev = __in_dev_get_rcu(skb->dev);
2549 			if (l3_in_dev)
2550 				our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2551 						      ip_hdr(skb)->protocol);
2552 		}
2553 
2554 		if (our
2555 #ifdef CONFIG_IP_MROUTE
2556 			||
2557 		    (!ipv4_is_local_multicast(daddr) &&
2558 		     IN_DEV_MFORWARD(in_dev))
2559 #endif
2560 		   ) {
2561 			reason = ip_route_input_mc(skb, daddr, saddr, dscp,
2562 						   dev, our);
2563 		}
2564 		return reason;
2565 	}
2566 
2567 	return ip_route_input_slow(skb, daddr, saddr, dscp, dev, res);
2568 }
2569 
2570 enum skb_drop_reason ip_route_input_noref(struct sk_buff *skb, __be32 daddr,
2571 					  __be32 saddr, dscp_t dscp,
2572 					  struct net_device *dev)
2573 {
2574 	enum skb_drop_reason reason;
2575 	struct fib_result res;
2576 
2577 	rcu_read_lock();
2578 	reason = ip_route_input_rcu(skb, daddr, saddr, dscp, dev, &res);
2579 	rcu_read_unlock();
2580 
2581 	return reason;
2582 }
2583 EXPORT_SYMBOL(ip_route_input_noref);
2584 
2585 /* called with rcu_read_lock() */
2586 static struct rtable *__mkroute_output(const struct fib_result *res,
2587 				       const struct flowi4 *fl4, int orig_oif,
2588 				       struct net_device *dev_out,
2589 				       unsigned int flags)
2590 {
2591 	struct fib_info *fi = res->fi;
2592 	struct fib_nh_exception *fnhe;
2593 	struct in_device *in_dev;
2594 	u16 type = res->type;
2595 	struct rtable *rth;
2596 	bool do_cache;
2597 
2598 	in_dev = __in_dev_get_rcu(dev_out);
2599 	if (!in_dev)
2600 		return ERR_PTR(-EINVAL);
2601 
2602 	if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2603 		if (ipv4_is_loopback(fl4->saddr) &&
2604 		    !(dev_out->flags & IFF_LOOPBACK) &&
2605 		    !netif_is_l3_master(dev_out))
2606 			return ERR_PTR(-EINVAL);
2607 
2608 	if (ipv4_is_lbcast(fl4->daddr)) {
2609 		type = RTN_BROADCAST;
2610 
2611 		/* reset fi to prevent gateway resolution */
2612 		fi = NULL;
2613 	} else if (ipv4_is_multicast(fl4->daddr)) {
2614 		type = RTN_MULTICAST;
2615 	} else if (ipv4_is_zeronet(fl4->daddr)) {
2616 		return ERR_PTR(-EINVAL);
2617 	}
2618 
2619 	if (dev_out->flags & IFF_LOOPBACK)
2620 		flags |= RTCF_LOCAL;
2621 
2622 	do_cache = true;
2623 	if (type == RTN_BROADCAST) {
2624 		flags |= RTCF_BROADCAST | RTCF_LOCAL;
2625 	} else if (type == RTN_MULTICAST) {
2626 		flags |= RTCF_MULTICAST | RTCF_LOCAL;
2627 		if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2628 				     fl4->flowi4_proto))
2629 			flags &= ~RTCF_LOCAL;
2630 		else
2631 			do_cache = false;
2632 		/* If multicast route do not exist use
2633 		 * default one, but do not gateway in this case.
2634 		 * Yes, it is hack.
2635 		 */
2636 		if (fi && res->prefixlen < 4)
2637 			fi = NULL;
2638 	} else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2639 		   (orig_oif != dev_out->ifindex)) {
2640 		/* For local routes that require a particular output interface
2641 		 * we do not want to cache the result.  Caching the result
2642 		 * causes incorrect behaviour when there are multiple source
2643 		 * addresses on the interface, the end result being that if the
2644 		 * intended recipient is waiting on that interface for the
2645 		 * packet he won't receive it because it will be delivered on
2646 		 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2647 		 * be set to the loopback interface as well.
2648 		 */
2649 		do_cache = false;
2650 	}
2651 
2652 	fnhe = NULL;
2653 	do_cache &= fi != NULL;
2654 	if (fi) {
2655 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2656 		struct rtable __rcu **prth;
2657 
2658 		fnhe = find_exception(nhc, fl4->daddr);
2659 		if (!do_cache)
2660 			goto add;
2661 		if (fnhe) {
2662 			prth = &fnhe->fnhe_rth_output;
2663 		} else {
2664 			if (unlikely(fl4->flowi4_flags &
2665 				     FLOWI_FLAG_KNOWN_NH &&
2666 				     !(nhc->nhc_gw_family &&
2667 				       nhc->nhc_scope == RT_SCOPE_LINK))) {
2668 				do_cache = false;
2669 				goto add;
2670 			}
2671 			prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2672 		}
2673 		rth = rcu_dereference(*prth);
2674 		if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2675 			return rth;
2676 	}
2677 
2678 add:
2679 	rth = rt_dst_alloc(dev_out, flags, type,
2680 			   IN_DEV_ORCONF(in_dev, NOXFRM));
2681 	if (!rth)
2682 		return ERR_PTR(-ENOBUFS);
2683 
2684 	rth->rt_iif = orig_oif;
2685 
2686 	RT_CACHE_STAT_INC(out_slow_tot);
2687 
2688 	if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2689 		if (flags & RTCF_LOCAL &&
2690 		    !(dev_out->flags & IFF_LOOPBACK)) {
2691 			rth->dst.output = ip_mc_output;
2692 			RT_CACHE_STAT_INC(out_slow_mc);
2693 		}
2694 #ifdef CONFIG_IP_MROUTE
2695 		if (type == RTN_MULTICAST) {
2696 			if (IN_DEV_MFORWARD(in_dev) &&
2697 			    !ipv4_is_local_multicast(fl4->daddr)) {
2698 				rth->dst.input = ip_mr_input;
2699 				rth->dst.output = ip_mr_output;
2700 			}
2701 		}
2702 #endif
2703 	}
2704 
2705 	rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2706 	lwtunnel_set_redirect(&rth->dst);
2707 
2708 	return rth;
2709 }
2710 
2711 /*
2712  * Major route resolver routine.
2713  */
2714 
2715 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2716 					const struct sk_buff *skb)
2717 {
2718 	struct fib_result res = {
2719 		.type		= RTN_UNSPEC,
2720 		.fi		= NULL,
2721 		.table		= NULL,
2722 		.tclassid	= 0,
2723 	};
2724 	struct rtable *rth;
2725 
2726 	fl4->flowi4_iif = LOOPBACK_IFINDEX;
2727 
2728 	rcu_read_lock();
2729 	rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2730 	rcu_read_unlock();
2731 
2732 	return rth;
2733 }
2734 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2735 
2736 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2737 					    struct fib_result *res,
2738 					    const struct sk_buff *skb)
2739 {
2740 	struct net_device *dev_out = NULL;
2741 	int orig_oif = fl4->flowi4_oif;
2742 	unsigned int flags = 0;
2743 	struct rtable *rth;
2744 	int err;
2745 
2746 	if (fl4->saddr) {
2747 		if (ipv4_is_multicast(fl4->saddr) ||
2748 		    ipv4_is_lbcast(fl4->saddr)) {
2749 			rth = ERR_PTR(-EINVAL);
2750 			goto out;
2751 		}
2752 
2753 		rth = ERR_PTR(-ENETUNREACH);
2754 
2755 		/* I removed check for oif == dev_out->oif here.
2756 		 * It was wrong for two reasons:
2757 		 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2758 		 *    is assigned to multiple interfaces.
2759 		 * 2. Moreover, we are allowed to send packets with saddr
2760 		 *    of another iface. --ANK
2761 		 */
2762 
2763 		if (fl4->flowi4_oif == 0 &&
2764 		    (ipv4_is_multicast(fl4->daddr) ||
2765 		     ipv4_is_lbcast(fl4->daddr))) {
2766 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2767 			dev_out = __ip_dev_find(net, fl4->saddr, false);
2768 			if (!dev_out)
2769 				goto out;
2770 
2771 			/* Special hack: user can direct multicasts
2772 			 * and limited broadcast via necessary interface
2773 			 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2774 			 * This hack is not just for fun, it allows
2775 			 * vic,vat and friends to work.
2776 			 * They bind socket to loopback, set ttl to zero
2777 			 * and expect that it will work.
2778 			 * From the viewpoint of routing cache they are broken,
2779 			 * because we are not allowed to build multicast path
2780 			 * with loopback source addr (look, routing cache
2781 			 * cannot know, that ttl is zero, so that packet
2782 			 * will not leave this host and route is valid).
2783 			 * Luckily, this hack is good workaround.
2784 			 */
2785 
2786 			fl4->flowi4_oif = dev_out->ifindex;
2787 			goto make_route;
2788 		}
2789 
2790 		if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2791 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2792 			if (!__ip_dev_find(net, fl4->saddr, false))
2793 				goto out;
2794 		}
2795 	}
2796 
2797 
2798 	if (fl4->flowi4_oif) {
2799 		dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2800 		rth = ERR_PTR(-ENODEV);
2801 		if (!dev_out)
2802 			goto out;
2803 
2804 		/* RACE: Check return value of inet_select_addr instead. */
2805 		if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2806 			rth = ERR_PTR(-ENETUNREACH);
2807 			goto out;
2808 		}
2809 		if (ipv4_is_local_multicast(fl4->daddr) ||
2810 		    ipv4_is_lbcast(fl4->daddr) ||
2811 		    fl4->flowi4_proto == IPPROTO_IGMP) {
2812 			if (!fl4->saddr)
2813 				fl4->saddr = inet_select_addr(dev_out, 0,
2814 							      RT_SCOPE_LINK);
2815 			goto make_route;
2816 		}
2817 		if (!fl4->saddr) {
2818 			if (ipv4_is_multicast(fl4->daddr))
2819 				fl4->saddr = inet_select_addr(dev_out, 0,
2820 							      fl4->flowi4_scope);
2821 			else if (!fl4->daddr)
2822 				fl4->saddr = inet_select_addr(dev_out, 0,
2823 							      RT_SCOPE_HOST);
2824 		}
2825 	}
2826 
2827 	if (!fl4->daddr) {
2828 		fl4->daddr = fl4->saddr;
2829 		if (!fl4->daddr)
2830 			fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2831 		dev_out = net->loopback_dev;
2832 		fl4->flowi4_oif = LOOPBACK_IFINDEX;
2833 		res->type = RTN_LOCAL;
2834 		flags |= RTCF_LOCAL;
2835 		goto make_route;
2836 	}
2837 
2838 	err = fib_lookup(net, fl4, res, 0);
2839 	if (err) {
2840 		res->fi = NULL;
2841 		res->table = NULL;
2842 		if (fl4->flowi4_oif &&
2843 		    (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2844 			/* Apparently, routing tables are wrong. Assume,
2845 			 * that the destination is on link.
2846 			 *
2847 			 * WHY? DW.
2848 			 * Because we are allowed to send to iface
2849 			 * even if it has NO routes and NO assigned
2850 			 * addresses. When oif is specified, routing
2851 			 * tables are looked up with only one purpose:
2852 			 * to catch if destination is gatewayed, rather than
2853 			 * direct. Moreover, if MSG_DONTROUTE is set,
2854 			 * we send packet, ignoring both routing tables
2855 			 * and ifaddr state. --ANK
2856 			 *
2857 			 *
2858 			 * We could make it even if oif is unknown,
2859 			 * likely IPv6, but we do not.
2860 			 */
2861 
2862 			if (fl4->saddr == 0)
2863 				fl4->saddr = inet_select_addr(dev_out, 0,
2864 							      RT_SCOPE_LINK);
2865 			res->type = RTN_UNICAST;
2866 			goto make_route;
2867 		}
2868 		rth = ERR_PTR(err);
2869 		goto out;
2870 	}
2871 
2872 	if (res->type == RTN_LOCAL) {
2873 		if (!fl4->saddr) {
2874 			if (res->fi->fib_prefsrc)
2875 				fl4->saddr = res->fi->fib_prefsrc;
2876 			else
2877 				fl4->saddr = fl4->daddr;
2878 		}
2879 
2880 		/* L3 master device is the loopback for that domain */
2881 		dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2882 			net->loopback_dev;
2883 
2884 		/* make sure orig_oif points to fib result device even
2885 		 * though packet rx/tx happens over loopback or l3mdev
2886 		 */
2887 		orig_oif = FIB_RES_OIF(*res);
2888 
2889 		fl4->flowi4_oif = dev_out->ifindex;
2890 		flags |= RTCF_LOCAL;
2891 		goto make_route;
2892 	}
2893 
2894 	fib_select_path(net, res, fl4, skb);
2895 
2896 	dev_out = FIB_RES_DEV(*res);
2897 
2898 make_route:
2899 	rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2900 
2901 out:
2902 	return rth;
2903 }
2904 
2905 static struct dst_ops ipv4_dst_blackhole_ops = {
2906 	.family			= AF_INET,
2907 	.default_advmss		= ipv4_default_advmss,
2908 	.neigh_lookup		= ipv4_neigh_lookup,
2909 	.check			= dst_blackhole_check,
2910 	.cow_metrics		= dst_blackhole_cow_metrics,
2911 	.update_pmtu		= dst_blackhole_update_pmtu,
2912 	.redirect		= dst_blackhole_redirect,
2913 	.mtu			= dst_blackhole_mtu,
2914 };
2915 
2916 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2917 {
2918 	struct rtable *ort = dst_rtable(dst_orig);
2919 	struct rtable *rt;
2920 
2921 	rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0);
2922 	if (rt) {
2923 		struct dst_entry *new = &rt->dst;
2924 
2925 		new->__use = 1;
2926 		new->input = dst_discard;
2927 		new->output = dst_discard_out;
2928 
2929 		new->dev = net->loopback_dev;
2930 		netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2931 
2932 		rt->rt_is_input = ort->rt_is_input;
2933 		rt->rt_iif = ort->rt_iif;
2934 		rt->rt_pmtu = ort->rt_pmtu;
2935 		rt->rt_mtu_locked = ort->rt_mtu_locked;
2936 
2937 		rt->rt_genid = rt_genid_ipv4(net);
2938 		rt->rt_flags = ort->rt_flags;
2939 		rt->rt_type = ort->rt_type;
2940 		rt->rt_uses_gateway = ort->rt_uses_gateway;
2941 		rt->rt_gw_family = ort->rt_gw_family;
2942 		if (rt->rt_gw_family == AF_INET)
2943 			rt->rt_gw4 = ort->rt_gw4;
2944 		else if (rt->rt_gw_family == AF_INET6)
2945 			rt->rt_gw6 = ort->rt_gw6;
2946 	}
2947 
2948 	dst_release(dst_orig);
2949 
2950 	return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2951 }
2952 
2953 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2954 				    const struct sock *sk)
2955 {
2956 	struct rtable *rt = __ip_route_output_key(net, flp4);
2957 
2958 	if (IS_ERR(rt))
2959 		return rt;
2960 
2961 	if (flp4->flowi4_proto) {
2962 		flp4->flowi4_oif = rt->dst.dev->ifindex;
2963 		rt = dst_rtable(xfrm_lookup_route(net, &rt->dst,
2964 						  flowi4_to_flowi(flp4),
2965 						  sk, 0));
2966 	}
2967 
2968 	return rt;
2969 }
2970 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2971 
2972 /* called with rcu_read_lock held */
2973 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2974 			struct rtable *rt, u32 table_id, dscp_t dscp,
2975 			struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2976 			u32 seq, unsigned int flags)
2977 {
2978 	struct rtmsg *r;
2979 	struct nlmsghdr *nlh;
2980 	unsigned long expires = 0;
2981 	u32 error;
2982 	u32 metrics[RTAX_MAX];
2983 
2984 	nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2985 	if (!nlh)
2986 		return -EMSGSIZE;
2987 
2988 	r = nlmsg_data(nlh);
2989 	r->rtm_family	 = AF_INET;
2990 	r->rtm_dst_len	= 32;
2991 	r->rtm_src_len	= 0;
2992 	r->rtm_tos	= inet_dscp_to_dsfield(dscp);
2993 	r->rtm_table	= table_id < 256 ? table_id : RT_TABLE_COMPAT;
2994 	if (nla_put_u32(skb, RTA_TABLE, table_id))
2995 		goto nla_put_failure;
2996 	r->rtm_type	= rt->rt_type;
2997 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2998 	r->rtm_protocol = RTPROT_UNSPEC;
2999 	r->rtm_flags	= (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
3000 	if (rt->rt_flags & RTCF_NOTIFY)
3001 		r->rtm_flags |= RTM_F_NOTIFY;
3002 	if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
3003 		r->rtm_flags |= RTCF_DOREDIRECT;
3004 
3005 	if (nla_put_in_addr(skb, RTA_DST, dst))
3006 		goto nla_put_failure;
3007 	if (src) {
3008 		r->rtm_src_len = 32;
3009 		if (nla_put_in_addr(skb, RTA_SRC, src))
3010 			goto nla_put_failure;
3011 	}
3012 	if (rt->dst.dev &&
3013 	    nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
3014 		goto nla_put_failure;
3015 	if (lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
3016 		goto nla_put_failure;
3017 #ifdef CONFIG_IP_ROUTE_CLASSID
3018 	if (rt->dst.tclassid &&
3019 	    nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
3020 		goto nla_put_failure;
3021 #endif
3022 	if (fl4 && !rt_is_input_route(rt) &&
3023 	    fl4->saddr != src) {
3024 		if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
3025 			goto nla_put_failure;
3026 	}
3027 	if (rt->rt_uses_gateway) {
3028 		if (rt->rt_gw_family == AF_INET &&
3029 		    nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
3030 			goto nla_put_failure;
3031 		} else if (rt->rt_gw_family == AF_INET6) {
3032 			int alen = sizeof(struct in6_addr);
3033 			struct nlattr *nla;
3034 			struct rtvia *via;
3035 
3036 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
3037 			if (!nla)
3038 				goto nla_put_failure;
3039 
3040 			via = nla_data(nla);
3041 			via->rtvia_family = AF_INET6;
3042 			memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
3043 		}
3044 	}
3045 
3046 	expires = READ_ONCE(rt->dst.expires);
3047 	if (expires) {
3048 		unsigned long now = jiffies;
3049 
3050 		if (time_before(now, expires))
3051 			expires -= now;
3052 		else
3053 			expires = 0;
3054 	}
3055 
3056 	memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3057 	if (rt->rt_pmtu && expires)
3058 		metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3059 	if (rt->rt_mtu_locked && expires)
3060 		metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3061 	if (rtnetlink_put_metrics(skb, metrics) < 0)
3062 		goto nla_put_failure;
3063 
3064 	if (fl4) {
3065 		if (fl4->flowi4_mark &&
3066 		    nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3067 			goto nla_put_failure;
3068 
3069 		if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3070 		    nla_put_u32(skb, RTA_UID,
3071 				from_kuid_munged(current_user_ns(),
3072 						 fl4->flowi4_uid)))
3073 			goto nla_put_failure;
3074 
3075 		if (rt_is_input_route(rt)) {
3076 #ifdef CONFIG_IP_MROUTE
3077 			if (ipv4_is_multicast(dst) &&
3078 			    !ipv4_is_local_multicast(dst) &&
3079 			    IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) {
3080 				int err = ipmr_get_route(net, skb,
3081 							 fl4->saddr, fl4->daddr,
3082 							 r, portid);
3083 
3084 				if (err <= 0) {
3085 					if (err == 0)
3086 						return 0;
3087 					goto nla_put_failure;
3088 				}
3089 			} else
3090 #endif
3091 				if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3092 					goto nla_put_failure;
3093 		}
3094 	}
3095 
3096 	error = rt->dst.error;
3097 
3098 	if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3099 		goto nla_put_failure;
3100 
3101 	nlmsg_end(skb, nlh);
3102 	return 0;
3103 
3104 nla_put_failure:
3105 	nlmsg_cancel(skb, nlh);
3106 	return -EMSGSIZE;
3107 }
3108 
3109 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3110 			    struct netlink_callback *cb, u32 table_id,
3111 			    struct fnhe_hash_bucket *bucket, int genid,
3112 			    int *fa_index, int fa_start, unsigned int flags)
3113 {
3114 	int i;
3115 
3116 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
3117 		struct fib_nh_exception *fnhe;
3118 
3119 		for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3120 		     fnhe = rcu_dereference(fnhe->fnhe_next)) {
3121 			struct rtable *rt;
3122 			int err;
3123 
3124 			if (*fa_index < fa_start)
3125 				goto next;
3126 
3127 			if (fnhe->fnhe_genid != genid)
3128 				goto next;
3129 
3130 			if (fnhe->fnhe_expires &&
3131 			    time_after(jiffies, fnhe->fnhe_expires))
3132 				goto next;
3133 
3134 			rt = rcu_dereference(fnhe->fnhe_rth_input);
3135 			if (!rt)
3136 				rt = rcu_dereference(fnhe->fnhe_rth_output);
3137 			if (!rt)
3138 				goto next;
3139 
3140 			err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3141 					   table_id, 0, NULL, skb,
3142 					   NETLINK_CB(cb->skb).portid,
3143 					   cb->nlh->nlmsg_seq, flags);
3144 			if (err)
3145 				return err;
3146 next:
3147 			(*fa_index)++;
3148 		}
3149 	}
3150 
3151 	return 0;
3152 }
3153 
3154 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3155 		       u32 table_id, struct fib_info *fi,
3156 		       int *fa_index, int fa_start, unsigned int flags)
3157 {
3158 	struct net *net = sock_net(cb->skb->sk);
3159 	int nhsel, genid = fnhe_genid(net);
3160 
3161 	for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3162 		struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3163 		struct fnhe_hash_bucket *bucket;
3164 		int err;
3165 
3166 		if (nhc->nhc_flags & RTNH_F_DEAD)
3167 			continue;
3168 
3169 		rcu_read_lock();
3170 		bucket = rcu_dereference(nhc->nhc_exceptions);
3171 		err = 0;
3172 		if (bucket)
3173 			err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3174 					       genid, fa_index, fa_start,
3175 					       flags);
3176 		rcu_read_unlock();
3177 		if (err)
3178 			return err;
3179 	}
3180 
3181 	return 0;
3182 }
3183 
3184 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3185 						   u8 ip_proto, __be16 sport,
3186 						   __be16 dport)
3187 {
3188 	struct sk_buff *skb;
3189 	struct iphdr *iph;
3190 
3191 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3192 	if (!skb)
3193 		return NULL;
3194 
3195 	/* Reserve room for dummy headers, this skb can pass
3196 	 * through good chunk of routing engine.
3197 	 */
3198 	skb_reset_mac_header(skb);
3199 	skb_reset_network_header(skb);
3200 	skb->protocol = htons(ETH_P_IP);
3201 	iph = skb_put(skb, sizeof(struct iphdr));
3202 	iph->protocol = ip_proto;
3203 	iph->saddr = src;
3204 	iph->daddr = dst;
3205 	iph->version = 0x4;
3206 	iph->frag_off = 0;
3207 	iph->ihl = 0x5;
3208 	skb_set_transport_header(skb, skb->len);
3209 
3210 	switch (iph->protocol) {
3211 	case IPPROTO_UDP: {
3212 		struct udphdr *udph;
3213 
3214 		udph = skb_put_zero(skb, sizeof(struct udphdr));
3215 		udph->source = sport;
3216 		udph->dest = dport;
3217 		udph->len = htons(sizeof(struct udphdr));
3218 		udph->check = 0;
3219 		break;
3220 	}
3221 	case IPPROTO_TCP: {
3222 		struct tcphdr *tcph;
3223 
3224 		tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3225 		tcph->source	= sport;
3226 		tcph->dest	= dport;
3227 		tcph->doff	= sizeof(struct tcphdr) / 4;
3228 		tcph->rst = 1;
3229 		tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3230 					    src, dst, 0);
3231 		break;
3232 	}
3233 	case IPPROTO_ICMP: {
3234 		struct icmphdr *icmph;
3235 
3236 		icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3237 		icmph->type = ICMP_ECHO;
3238 		icmph->code = 0;
3239 	}
3240 	}
3241 
3242 	return skb;
3243 }
3244 
3245 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3246 				       const struct nlmsghdr *nlh,
3247 				       struct nlattr **tb,
3248 				       struct netlink_ext_ack *extack)
3249 {
3250 	struct rtmsg *rtm;
3251 	int i, err;
3252 
3253 	rtm = nlmsg_payload(nlh, sizeof(*rtm));
3254 	if (!rtm) {
3255 		NL_SET_ERR_MSG(extack,
3256 			       "ipv4: Invalid header for route get request");
3257 		return -EINVAL;
3258 	}
3259 
3260 	if (!netlink_strict_get_check(skb))
3261 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3262 					      rtm_ipv4_policy, extack);
3263 
3264 	if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3265 	    (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3266 	    rtm->rtm_table || rtm->rtm_protocol ||
3267 	    rtm->rtm_scope || rtm->rtm_type) {
3268 		NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3269 		return -EINVAL;
3270 	}
3271 
3272 	if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3273 			       RTM_F_LOOKUP_TABLE |
3274 			       RTM_F_FIB_MATCH)) {
3275 		NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3276 		return -EINVAL;
3277 	}
3278 
3279 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3280 					    rtm_ipv4_policy, extack);
3281 	if (err)
3282 		return err;
3283 
3284 	if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3285 	    (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3286 		NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3287 		return -EINVAL;
3288 	}
3289 
3290 	for (i = 0; i <= RTA_MAX; i++) {
3291 		if (!tb[i])
3292 			continue;
3293 
3294 		switch (i) {
3295 		case RTA_IIF:
3296 		case RTA_OIF:
3297 		case RTA_SRC:
3298 		case RTA_DST:
3299 		case RTA_IP_PROTO:
3300 		case RTA_SPORT:
3301 		case RTA_DPORT:
3302 		case RTA_MARK:
3303 		case RTA_UID:
3304 			break;
3305 		default:
3306 			NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3307 			return -EINVAL;
3308 		}
3309 	}
3310 
3311 	return 0;
3312 }
3313 
3314 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3315 			     struct netlink_ext_ack *extack)
3316 {
3317 	struct net *net = sock_net(in_skb->sk);
3318 	struct nlattr *tb[RTA_MAX+1];
3319 	u32 table_id = RT_TABLE_MAIN;
3320 	__be16 sport = 0, dport = 0;
3321 	struct fib_result res = {};
3322 	u8 ip_proto = IPPROTO_UDP;
3323 	struct rtable *rt = NULL;
3324 	struct sk_buff *skb;
3325 	struct rtmsg *rtm;
3326 	struct flowi4 fl4 = {};
3327 	__be32 dst = 0;
3328 	__be32 src = 0;
3329 	dscp_t dscp;
3330 	kuid_t uid;
3331 	u32 iif;
3332 	int err;
3333 	int mark;
3334 
3335 	err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3336 	if (err < 0)
3337 		return err;
3338 
3339 	rtm = nlmsg_data(nlh);
3340 	src = nla_get_in_addr_default(tb[RTA_SRC], 0);
3341 	dst = nla_get_in_addr_default(tb[RTA_DST], 0);
3342 	iif = nla_get_u32_default(tb[RTA_IIF], 0);
3343 	mark = nla_get_u32_default(tb[RTA_MARK], 0);
3344 	dscp = inet_dsfield_to_dscp(rtm->rtm_tos);
3345 	if (tb[RTA_UID])
3346 		uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3347 	else
3348 		uid = (iif ? INVALID_UID : current_uid());
3349 
3350 	if (tb[RTA_IP_PROTO]) {
3351 		err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3352 						  &ip_proto, AF_INET, extack);
3353 		if (err)
3354 			return err;
3355 	}
3356 
3357 	if (tb[RTA_SPORT])
3358 		sport = nla_get_be16(tb[RTA_SPORT]);
3359 
3360 	if (tb[RTA_DPORT])
3361 		dport = nla_get_be16(tb[RTA_DPORT]);
3362 
3363 	skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3364 	if (!skb)
3365 		return -ENOBUFS;
3366 
3367 	fl4.daddr = dst;
3368 	fl4.saddr = src;
3369 	fl4.flowi4_dscp = dscp;
3370 	fl4.flowi4_oif = nla_get_u32_default(tb[RTA_OIF], 0);
3371 	fl4.flowi4_mark = mark;
3372 	fl4.flowi4_uid = uid;
3373 	if (sport)
3374 		fl4.fl4_sport = sport;
3375 	if (dport)
3376 		fl4.fl4_dport = dport;
3377 	fl4.flowi4_proto = ip_proto;
3378 
3379 	rcu_read_lock();
3380 
3381 	if (iif) {
3382 		struct net_device *dev;
3383 
3384 		dev = dev_get_by_index_rcu(net, iif);
3385 		if (!dev) {
3386 			err = -ENODEV;
3387 			goto errout_rcu;
3388 		}
3389 
3390 		fl4.flowi4_iif = iif; /* for rt_fill_info */
3391 		skb->dev	= dev;
3392 		skb->mark	= mark;
3393 		err = ip_route_input_rcu(skb, dst, src, dscp, dev,
3394 					 &res) ? -EINVAL : 0;
3395 
3396 		rt = skb_rtable(skb);
3397 		if (err == 0 && rt->dst.error)
3398 			err = -rt->dst.error;
3399 	} else {
3400 		fl4.flowi4_iif = LOOPBACK_IFINDEX;
3401 		skb->dev = net->loopback_dev;
3402 		rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3403 		err = 0;
3404 		if (IS_ERR(rt))
3405 			err = PTR_ERR(rt);
3406 		else
3407 			skb_dst_set(skb, &rt->dst);
3408 	}
3409 
3410 	if (err)
3411 		goto errout_rcu;
3412 
3413 	if (rtm->rtm_flags & RTM_F_NOTIFY)
3414 		rt->rt_flags |= RTCF_NOTIFY;
3415 
3416 	if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3417 		table_id = res.table ? res.table->tb_id : 0;
3418 
3419 	/* reset skb for netlink reply msg */
3420 	skb_trim(skb, 0);
3421 	skb_reset_network_header(skb);
3422 	skb_reset_transport_header(skb);
3423 	skb_reset_mac_header(skb);
3424 
3425 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3426 		struct fib_rt_info fri;
3427 
3428 		if (!res.fi) {
3429 			err = fib_props[res.type].error;
3430 			if (!err)
3431 				err = -EHOSTUNREACH;
3432 			goto errout_rcu;
3433 		}
3434 		fri.fi = res.fi;
3435 		fri.tb_id = table_id;
3436 		fri.dst = res.prefix;
3437 		fri.dst_len = res.prefixlen;
3438 		fri.dscp = res.dscp;
3439 		fri.type = rt->rt_type;
3440 		fri.offload = 0;
3441 		fri.trap = 0;
3442 		fri.offload_failed = 0;
3443 		if (res.fa_head) {
3444 			struct fib_alias *fa;
3445 
3446 			hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3447 				u8 slen = 32 - fri.dst_len;
3448 
3449 				if (fa->fa_slen == slen &&
3450 				    fa->tb_id == fri.tb_id &&
3451 				    fa->fa_dscp == fri.dscp &&
3452 				    fa->fa_info == res.fi &&
3453 				    fa->fa_type == fri.type) {
3454 					fri.offload = READ_ONCE(fa->offload);
3455 					fri.trap = READ_ONCE(fa->trap);
3456 					fri.offload_failed =
3457 						READ_ONCE(fa->offload_failed);
3458 					break;
3459 				}
3460 			}
3461 		}
3462 		err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3463 				    nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3464 	} else {
3465 		err = rt_fill_info(net, dst, src, rt, table_id, res.dscp, &fl4,
3466 				   skb, NETLINK_CB(in_skb).portid,
3467 				   nlh->nlmsg_seq, 0);
3468 	}
3469 	if (err < 0)
3470 		goto errout_rcu;
3471 
3472 	rcu_read_unlock();
3473 
3474 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3475 
3476 errout_free:
3477 	return err;
3478 errout_rcu:
3479 	rcu_read_unlock();
3480 	kfree_skb(skb);
3481 	goto errout_free;
3482 }
3483 
3484 void ip_rt_multicast_event(struct in_device *in_dev)
3485 {
3486 	rt_cache_flush(dev_net(in_dev->dev));
3487 }
3488 
3489 #ifdef CONFIG_SYSCTL
3490 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3491 static int ip_rt_gc_min_interval __read_mostly	= HZ / 2;
3492 static int ip_rt_gc_elasticity __read_mostly	= 8;
3493 static int ip_min_valid_pmtu __read_mostly	= IPV4_MIN_MTU;
3494 
3495 static int ipv4_sysctl_rtcache_flush(const struct ctl_table *__ctl, int write,
3496 		void *buffer, size_t *lenp, loff_t *ppos)
3497 {
3498 	struct net *net = (struct net *)__ctl->extra1;
3499 
3500 	if (write) {
3501 		rt_cache_flush(net);
3502 		fnhe_genid_bump(net);
3503 		return 0;
3504 	}
3505 
3506 	return -EINVAL;
3507 }
3508 
3509 static struct ctl_table ipv4_route_table[] = {
3510 	{
3511 		.procname	= "gc_thresh",
3512 		.data		= &ipv4_dst_ops.gc_thresh,
3513 		.maxlen		= sizeof(int),
3514 		.mode		= 0644,
3515 		.proc_handler	= proc_dointvec,
3516 	},
3517 	{
3518 		.procname	= "max_size",
3519 		.data		= &ip_rt_max_size,
3520 		.maxlen		= sizeof(int),
3521 		.mode		= 0644,
3522 		.proc_handler	= proc_dointvec,
3523 	},
3524 	{
3525 		/*  Deprecated. Use gc_min_interval_ms */
3526 
3527 		.procname	= "gc_min_interval",
3528 		.data		= &ip_rt_gc_min_interval,
3529 		.maxlen		= sizeof(int),
3530 		.mode		= 0644,
3531 		.proc_handler	= proc_dointvec_jiffies,
3532 	},
3533 	{
3534 		.procname	= "gc_min_interval_ms",
3535 		.data		= &ip_rt_gc_min_interval,
3536 		.maxlen		= sizeof(int),
3537 		.mode		= 0644,
3538 		.proc_handler	= proc_dointvec_ms_jiffies,
3539 	},
3540 	{
3541 		.procname	= "gc_timeout",
3542 		.data		= &ip_rt_gc_timeout,
3543 		.maxlen		= sizeof(int),
3544 		.mode		= 0644,
3545 		.proc_handler	= proc_dointvec_jiffies,
3546 	},
3547 	{
3548 		.procname	= "gc_interval",
3549 		.data		= &ip_rt_gc_interval,
3550 		.maxlen		= sizeof(int),
3551 		.mode		= 0644,
3552 		.proc_handler	= proc_dointvec_jiffies,
3553 	},
3554 	{
3555 		.procname	= "redirect_load",
3556 		.data		= &ip_rt_redirect_load,
3557 		.maxlen		= sizeof(int),
3558 		.mode		= 0644,
3559 		.proc_handler	= proc_dointvec,
3560 	},
3561 	{
3562 		.procname	= "redirect_number",
3563 		.data		= &ip_rt_redirect_number,
3564 		.maxlen		= sizeof(int),
3565 		.mode		= 0644,
3566 		.proc_handler	= proc_dointvec,
3567 	},
3568 	{
3569 		.procname	= "redirect_silence",
3570 		.data		= &ip_rt_redirect_silence,
3571 		.maxlen		= sizeof(int),
3572 		.mode		= 0644,
3573 		.proc_handler	= proc_dointvec,
3574 	},
3575 	{
3576 		.procname	= "error_cost",
3577 		.data		= &ip_rt_error_cost,
3578 		.maxlen		= sizeof(int),
3579 		.mode		= 0644,
3580 		.proc_handler	= proc_dointvec,
3581 	},
3582 	{
3583 		.procname	= "error_burst",
3584 		.data		= &ip_rt_error_burst,
3585 		.maxlen		= sizeof(int),
3586 		.mode		= 0644,
3587 		.proc_handler	= proc_dointvec,
3588 	},
3589 	{
3590 		.procname	= "gc_elasticity",
3591 		.data		= &ip_rt_gc_elasticity,
3592 		.maxlen		= sizeof(int),
3593 		.mode		= 0644,
3594 		.proc_handler	= proc_dointvec,
3595 	},
3596 };
3597 
3598 static const char ipv4_route_flush_procname[] = "flush";
3599 
3600 static struct ctl_table ipv4_route_netns_table[] = {
3601 	{
3602 		.procname	= ipv4_route_flush_procname,
3603 		.maxlen		= sizeof(int),
3604 		.mode		= 0200,
3605 		.proc_handler	= ipv4_sysctl_rtcache_flush,
3606 	},
3607 	{
3608 		.procname       = "min_pmtu",
3609 		.data           = &init_net.ipv4.ip_rt_min_pmtu,
3610 		.maxlen         = sizeof(int),
3611 		.mode           = 0644,
3612 		.proc_handler   = proc_dointvec_minmax,
3613 		.extra1         = &ip_min_valid_pmtu,
3614 	},
3615 	{
3616 		.procname       = "mtu_expires",
3617 		.data           = &init_net.ipv4.ip_rt_mtu_expires,
3618 		.maxlen         = sizeof(int),
3619 		.mode           = 0644,
3620 		.proc_handler   = proc_dointvec_jiffies,
3621 	},
3622 	{
3623 		.procname   = "min_adv_mss",
3624 		.data       = &init_net.ipv4.ip_rt_min_advmss,
3625 		.maxlen     = sizeof(int),
3626 		.mode       = 0644,
3627 		.proc_handler   = proc_dointvec,
3628 	},
3629 };
3630 
3631 static __net_init int sysctl_route_net_init(struct net *net)
3632 {
3633 	struct ctl_table *tbl;
3634 	size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3635 
3636 	tbl = ipv4_route_netns_table;
3637 	if (!net_eq(net, &init_net)) {
3638 		int i;
3639 
3640 		tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3641 		if (!tbl)
3642 			goto err_dup;
3643 
3644 		/* Don't export non-whitelisted sysctls to unprivileged users */
3645 		if (net->user_ns != &init_user_ns) {
3646 			if (tbl[0].procname != ipv4_route_flush_procname)
3647 				table_size = 0;
3648 		}
3649 
3650 		/* Update the variables to point into the current struct net
3651 		 * except for the first element flush
3652 		 */
3653 		for (i = 1; i < table_size; i++)
3654 			tbl[i].data += (void *)net - (void *)&init_net;
3655 	}
3656 	tbl[0].extra1 = net;
3657 
3658 	net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3659 						     tbl, table_size);
3660 	if (!net->ipv4.route_hdr)
3661 		goto err_reg;
3662 	return 0;
3663 
3664 err_reg:
3665 	if (tbl != ipv4_route_netns_table)
3666 		kfree(tbl);
3667 err_dup:
3668 	return -ENOMEM;
3669 }
3670 
3671 static __net_exit void sysctl_route_net_exit(struct net *net)
3672 {
3673 	const struct ctl_table *tbl;
3674 
3675 	tbl = net->ipv4.route_hdr->ctl_table_arg;
3676 	unregister_net_sysctl_table(net->ipv4.route_hdr);
3677 	BUG_ON(tbl == ipv4_route_netns_table);
3678 	kfree(tbl);
3679 }
3680 
3681 static __net_initdata struct pernet_operations sysctl_route_ops = {
3682 	.init = sysctl_route_net_init,
3683 	.exit = sysctl_route_net_exit,
3684 };
3685 #endif
3686 
3687 static __net_init int netns_ip_rt_init(struct net *net)
3688 {
3689 	/* Set default value for namespaceified sysctls */
3690 	net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3691 	net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3692 	net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3693 	return 0;
3694 }
3695 
3696 static struct pernet_operations __net_initdata ip_rt_ops = {
3697 	.init = netns_ip_rt_init,
3698 };
3699 
3700 static __net_init int rt_genid_init(struct net *net)
3701 {
3702 	atomic_set(&net->ipv4.rt_genid, 0);
3703 	atomic_set(&net->fnhe_genid, 0);
3704 	atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3705 	return 0;
3706 }
3707 
3708 static __net_initdata struct pernet_operations rt_genid_ops = {
3709 	.init = rt_genid_init,
3710 };
3711 
3712 static int __net_init ipv4_inetpeer_init(struct net *net)
3713 {
3714 	struct inet_peer_base *bp = kmalloc_obj(*bp);
3715 
3716 	if (!bp)
3717 		return -ENOMEM;
3718 	inet_peer_base_init(bp);
3719 	net->ipv4.peers = bp;
3720 	return 0;
3721 }
3722 
3723 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3724 {
3725 	struct inet_peer_base *bp = net->ipv4.peers;
3726 
3727 	net->ipv4.peers = NULL;
3728 	inetpeer_invalidate_tree(bp);
3729 	kfree(bp);
3730 }
3731 
3732 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3733 	.init	=	ipv4_inetpeer_init,
3734 	.exit	=	ipv4_inetpeer_exit,
3735 };
3736 
3737 #ifdef CONFIG_IP_ROUTE_CLASSID
3738 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3739 #endif /* CONFIG_IP_ROUTE_CLASSID */
3740 
3741 static const struct rtnl_msg_handler ip_rt_rtnl_msg_handlers[] __initconst = {
3742 	{.protocol = PF_INET, .msgtype = RTM_GETROUTE,
3743 	 .doit = inet_rtm_getroute, .flags = RTNL_FLAG_DOIT_UNLOCKED},
3744 };
3745 
3746 int __init ip_rt_init(void)
3747 {
3748 	void *idents_hash;
3749 	int cpu;
3750 
3751 	/* For modern hosts, this will use 2 MB of memory */
3752 	idents_hash = alloc_large_system_hash("IP idents",
3753 					      sizeof(*ip_idents) + sizeof(*ip_tstamps),
3754 					      0,
3755 					      16, /* one bucket per 64 KB */
3756 					      HASH_ZERO,
3757 					      NULL,
3758 					      &ip_idents_mask,
3759 					      2048,
3760 					      256*1024);
3761 
3762 	ip_idents = idents_hash;
3763 
3764 	get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3765 
3766 	ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3767 
3768 	for_each_possible_cpu(cpu) {
3769 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3770 
3771 		INIT_LIST_HEAD(&ul->head);
3772 		spin_lock_init(&ul->lock);
3773 	}
3774 #ifdef CONFIG_IP_ROUTE_CLASSID
3775 	ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3776 	if (!ip_rt_acct)
3777 		panic("IP: failed to allocate ip_rt_acct\n");
3778 #endif
3779 
3780 	ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable,
3781 					      SLAB_HWCACHE_ALIGN | SLAB_PANIC);
3782 
3783 	ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3784 
3785 	if (dst_entries_init(&ipv4_dst_ops) < 0)
3786 		panic("IP: failed to allocate ipv4_dst_ops counter\n");
3787 
3788 	if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3789 		panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3790 
3791 	ipv4_dst_ops.gc_thresh = ~0;
3792 	ip_rt_max_size = INT_MAX;
3793 
3794 	devinet_init();
3795 	ip_fib_init();
3796 
3797 	if (ip_rt_proc_init())
3798 		pr_err("Unable to create route proc files\n");
3799 #ifdef CONFIG_XFRM
3800 	xfrm_init();
3801 	xfrm4_init();
3802 #endif
3803 	rtnl_register_many(ip_rt_rtnl_msg_handlers);
3804 
3805 #ifdef CONFIG_SYSCTL
3806 	register_pernet_subsys(&sysctl_route_ops);
3807 #endif
3808 	register_pernet_subsys(&ip_rt_ops);
3809 	register_pernet_subsys(&rt_genid_ops);
3810 	register_pernet_subsys(&ipv4_inetpeer_ops);
3811 	return 0;
3812 }
3813 
3814 #ifdef CONFIG_SYSCTL
3815 /*
3816  * We really need to sanitize the damn ipv4 init order, then all
3817  * this nonsense will go away.
3818  */
3819 void __init ip_static_sysctl_init(void)
3820 {
3821 	register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3822 }
3823 #endif
3824