xref: /linux/net/ipv4/route.c (revision c5fbdf0ba7c1a6ed52dc3650bee73ce00c86cf7f)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		ROUTE - implementation of the IP router.
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
12  *		Linus Torvalds, <Linus.Torvalds@helsinki.fi>
13  *		Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
14  *
15  * Fixes:
16  *		Alan Cox	:	Verify area fixes.
17  *		Alan Cox	:	cli() protects routing changes
18  *		Rui Oliveira	:	ICMP routing table updates
19  *		(rco@di.uminho.pt)	Routing table insertion and update
20  *		Linus Torvalds	:	Rewrote bits to be sensible
21  *		Alan Cox	:	Added BSD route gw semantics
22  *		Alan Cox	:	Super /proc >4K
23  *		Alan Cox	:	MTU in route table
24  *		Alan Cox	:	MSS actually. Also added the window
25  *					clamper.
26  *		Sam Lantinga	:	Fixed route matching in rt_del()
27  *		Alan Cox	:	Routing cache support.
28  *		Alan Cox	:	Removed compatibility cruft.
29  *		Alan Cox	:	RTF_REJECT support.
30  *		Alan Cox	:	TCP irtt support.
31  *		Jonathan Naylor	:	Added Metric support.
32  *	Miquel van Smoorenburg	:	BSD API fixes.
33  *	Miquel van Smoorenburg	:	Metrics.
34  *		Alan Cox	:	Use __u32 properly
35  *		Alan Cox	:	Aligned routing errors more closely with BSD
36  *					our system is still very different.
37  *		Alan Cox	:	Faster /proc handling
38  *	Alexey Kuznetsov	:	Massive rework to support tree based routing,
39  *					routing caches and better behaviour.
40  *
41  *		Olaf Erb	:	irtt wasn't being copied right.
42  *		Bjorn Ekwall	:	Kerneld route support.
43  *		Alan Cox	:	Multicast fixed (I hope)
44  *		Pavel Krauz	:	Limited broadcast fixed
45  *		Mike McLagan	:	Routing by source
46  *	Alexey Kuznetsov	:	End of old history. Split to fib.c and
47  *					route.c and rewritten from scratch.
48  *		Andi Kleen	:	Load-limit warning messages.
49  *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
50  *	Vitaly E. Lavrov	:	Race condition in ip_route_input_slow.
51  *	Tobias Ringstrom	:	Uninitialized res.type in ip_route_output_slow.
52  *	Vladimir V. Ivanov	:	IP rule info (flowid) is really useful.
53  *		Marc Boucher	:	routing by fwmark
54  *	Robert Olsson		:	Added rt_cache statistics
55  *	Arnaldo C. Melo		:	Convert proc stuff to seq_file
56  *	Eric Dumazet		:	hashed spinlocks and rt_check_expire() fixes.
57  *	Ilia Sotnikov		:	Ignore TOS on PMTUD and Redirect
58  *	Ilia Sotnikov		:	Removed TOS from hash calculations
59  */
60 
61 #define pr_fmt(fmt) "IPv4: " fmt
62 
63 #include <linux/module.h>
64 #include <linux/bitops.h>
65 #include <linux/kernel.h>
66 #include <linux/mm.h>
67 #include <linux/memblock.h>
68 #include <linux/socket.h>
69 #include <linux/errno.h>
70 #include <linux/in.h>
71 #include <linux/inet.h>
72 #include <linux/netdevice.h>
73 #include <linux/proc_fs.h>
74 #include <linux/init.h>
75 #include <linux/skbuff.h>
76 #include <linux/inetdevice.h>
77 #include <linux/igmp.h>
78 #include <linux/pkt_sched.h>
79 #include <linux/mroute.h>
80 #include <linux/netfilter_ipv4.h>
81 #include <linux/random.h>
82 #include <linux/rcupdate.h>
83 #include <linux/slab.h>
84 #include <linux/jhash.h>
85 #include <net/dst.h>
86 #include <net/dst_metadata.h>
87 #include <net/inet_dscp.h>
88 #include <net/net_namespace.h>
89 #include <net/ip.h>
90 #include <net/route.h>
91 #include <net/inetpeer.h>
92 #include <net/sock.h>
93 #include <net/ip_fib.h>
94 #include <net/nexthop.h>
95 #include <net/tcp.h>
96 #include <net/icmp.h>
97 #include <net/xfrm.h>
98 #include <net/lwtunnel.h>
99 #include <net/netevent.h>
100 #include <net/rtnetlink.h>
101 #ifdef CONFIG_SYSCTL
102 #include <linux/sysctl.h>
103 #endif
104 #include <net/secure_seq.h>
105 #include <net/ip_tunnels.h>
106 
107 #include "fib_lookup.h"
108 
109 #define RT_GC_TIMEOUT (300*HZ)
110 
111 #define DEFAULT_MIN_PMTU (512 + 20 + 20)
112 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
113 #define DEFAULT_MIN_ADVMSS 256
114 static int ip_rt_max_size;
115 static int ip_rt_redirect_number __read_mostly	= 9;
116 static int ip_rt_redirect_load __read_mostly	= HZ / 50;
117 static int ip_rt_redirect_silence __read_mostly	= ((HZ / 50) << (9 + 1));
118 static int ip_rt_error_cost __read_mostly	= HZ;
119 static int ip_rt_error_burst __read_mostly	= 5 * HZ;
120 
121 static int ip_rt_gc_timeout __read_mostly	= RT_GC_TIMEOUT;
122 
123 /*
124  *	Interface to generic destination cache.
125  */
126 
127 INDIRECT_CALLABLE_SCOPE
128 struct dst_entry	*ipv4_dst_check(struct dst_entry *dst, u32 cookie);
129 static unsigned int	 ipv4_default_advmss(const struct dst_entry *dst);
130 INDIRECT_CALLABLE_SCOPE
131 unsigned int		ipv4_mtu(const struct dst_entry *dst);
132 static void		ipv4_negative_advice(struct sock *sk,
133 					     struct dst_entry *dst);
134 static void		 ipv4_link_failure(struct sk_buff *skb);
135 static void		 ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
136 					   struct sk_buff *skb, u32 mtu,
137 					   bool confirm_neigh);
138 static void		 ip_do_redirect(struct dst_entry *dst, struct sock *sk,
139 					struct sk_buff *skb);
140 static void		ipv4_dst_destroy(struct dst_entry *dst);
141 
142 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
143 {
144 	WARN_ON(1);
145 	return NULL;
146 }
147 
148 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
149 					   struct sk_buff *skb,
150 					   const void *daddr);
151 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
152 
153 static struct dst_ops ipv4_dst_ops = {
154 	.family =		AF_INET,
155 	.check =		ipv4_dst_check,
156 	.default_advmss =	ipv4_default_advmss,
157 	.mtu =			ipv4_mtu,
158 	.cow_metrics =		ipv4_cow_metrics,
159 	.destroy =		ipv4_dst_destroy,
160 	.negative_advice =	ipv4_negative_advice,
161 	.link_failure =		ipv4_link_failure,
162 	.update_pmtu =		ip_rt_update_pmtu,
163 	.redirect =		ip_do_redirect,
164 	.local_out =		__ip_local_out,
165 	.neigh_lookup =		ipv4_neigh_lookup,
166 	.confirm_neigh =	ipv4_confirm_neigh,
167 };
168 
169 #define ECN_OR_COST(class)	TC_PRIO_##class
170 
171 const __u8 ip_tos2prio[16] = {
172 	TC_PRIO_BESTEFFORT,
173 	ECN_OR_COST(BESTEFFORT),
174 	TC_PRIO_BESTEFFORT,
175 	ECN_OR_COST(BESTEFFORT),
176 	TC_PRIO_BULK,
177 	ECN_OR_COST(BULK),
178 	TC_PRIO_BULK,
179 	ECN_OR_COST(BULK),
180 	TC_PRIO_INTERACTIVE,
181 	ECN_OR_COST(INTERACTIVE),
182 	TC_PRIO_INTERACTIVE,
183 	ECN_OR_COST(INTERACTIVE),
184 	TC_PRIO_INTERACTIVE_BULK,
185 	ECN_OR_COST(INTERACTIVE_BULK),
186 	TC_PRIO_INTERACTIVE_BULK,
187 	ECN_OR_COST(INTERACTIVE_BULK)
188 };
189 EXPORT_SYMBOL(ip_tos2prio);
190 
191 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
192 #ifndef CONFIG_PREEMPT_RT
193 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
194 #else
195 #define RT_CACHE_STAT_INC(field) this_cpu_inc(rt_cache_stat.field)
196 #endif
197 
198 #ifdef CONFIG_PROC_FS
199 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
200 {
201 	if (*pos)
202 		return NULL;
203 	return SEQ_START_TOKEN;
204 }
205 
206 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
207 {
208 	++*pos;
209 	return NULL;
210 }
211 
212 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
213 {
214 }
215 
216 static int rt_cache_seq_show(struct seq_file *seq, void *v)
217 {
218 	if (v == SEQ_START_TOKEN)
219 		seq_printf(seq, "%-127s\n",
220 			   "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
221 			   "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
222 			   "HHUptod\tSpecDst");
223 	return 0;
224 }
225 
226 static const struct seq_operations rt_cache_seq_ops = {
227 	.start  = rt_cache_seq_start,
228 	.next   = rt_cache_seq_next,
229 	.stop   = rt_cache_seq_stop,
230 	.show   = rt_cache_seq_show,
231 };
232 
233 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
234 {
235 	int cpu;
236 
237 	if (*pos == 0)
238 		return SEQ_START_TOKEN;
239 
240 	for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
241 		if (!cpu_possible(cpu))
242 			continue;
243 		*pos = cpu+1;
244 		return &per_cpu(rt_cache_stat, cpu);
245 	}
246 	return NULL;
247 }
248 
249 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
250 {
251 	int cpu;
252 
253 	for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
254 		if (!cpu_possible(cpu))
255 			continue;
256 		*pos = cpu+1;
257 		return &per_cpu(rt_cache_stat, cpu);
258 	}
259 	(*pos)++;
260 	return NULL;
261 
262 }
263 
264 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
265 {
266 
267 }
268 
269 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
270 {
271 	struct rt_cache_stat *st = v;
272 
273 	if (v == SEQ_START_TOKEN) {
274 		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");
275 		return 0;
276 	}
277 
278 	seq_printf(seq, "%08x %08x %08x    %08x   %08x    %08x %08x       "
279 			"%08x       %08x %08x     %08x    %08x %08x   "
280 			"%08x     %08x        %08x        %08x\n",
281 		   dst_entries_get_slow(&ipv4_dst_ops),
282 		   0, /* st->in_hit */
283 		   st->in_slow_tot,
284 		   st->in_slow_mc,
285 		   st->in_no_route,
286 		   st->in_brd,
287 		   st->in_martian_dst,
288 		   st->in_martian_src,
289 
290 		   0, /* st->out_hit */
291 		   st->out_slow_tot,
292 		   st->out_slow_mc,
293 
294 		   0, /* st->gc_total */
295 		   0, /* st->gc_ignored */
296 		   0, /* st->gc_goal_miss */
297 		   0, /* st->gc_dst_overflow */
298 		   0, /* st->in_hlist_search */
299 		   0  /* st->out_hlist_search */
300 		);
301 	return 0;
302 }
303 
304 static const struct seq_operations rt_cpu_seq_ops = {
305 	.start  = rt_cpu_seq_start,
306 	.next   = rt_cpu_seq_next,
307 	.stop   = rt_cpu_seq_stop,
308 	.show   = rt_cpu_seq_show,
309 };
310 
311 #ifdef CONFIG_IP_ROUTE_CLASSID
312 static int rt_acct_proc_show(struct seq_file *m, void *v)
313 {
314 	struct ip_rt_acct *dst, *src;
315 	unsigned int i, j;
316 
317 	dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
318 	if (!dst)
319 		return -ENOMEM;
320 
321 	for_each_possible_cpu(i) {
322 		src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
323 		for (j = 0; j < 256; j++) {
324 			dst[j].o_bytes   += src[j].o_bytes;
325 			dst[j].o_packets += src[j].o_packets;
326 			dst[j].i_bytes   += src[j].i_bytes;
327 			dst[j].i_packets += src[j].i_packets;
328 		}
329 	}
330 
331 	seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
332 	kfree(dst);
333 	return 0;
334 }
335 #endif
336 
337 static int __net_init ip_rt_do_proc_init(struct net *net)
338 {
339 	struct proc_dir_entry *pde;
340 
341 	pde = proc_create_seq("rt_cache", 0444, net->proc_net,
342 			      &rt_cache_seq_ops);
343 	if (!pde)
344 		goto err1;
345 
346 	pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
347 			      &rt_cpu_seq_ops);
348 	if (!pde)
349 		goto err2;
350 
351 #ifdef CONFIG_IP_ROUTE_CLASSID
352 	pde = proc_create_single("rt_acct", 0, net->proc_net,
353 			rt_acct_proc_show);
354 	if (!pde)
355 		goto err3;
356 #endif
357 	return 0;
358 
359 #ifdef CONFIG_IP_ROUTE_CLASSID
360 err3:
361 	remove_proc_entry("rt_cache", net->proc_net_stat);
362 #endif
363 err2:
364 	remove_proc_entry("rt_cache", net->proc_net);
365 err1:
366 	return -ENOMEM;
367 }
368 
369 static void __net_exit ip_rt_do_proc_exit(struct net *net)
370 {
371 	remove_proc_entry("rt_cache", net->proc_net_stat);
372 	remove_proc_entry("rt_cache", net->proc_net);
373 #ifdef CONFIG_IP_ROUTE_CLASSID
374 	remove_proc_entry("rt_acct", net->proc_net);
375 #endif
376 }
377 
378 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
379 	.init = ip_rt_do_proc_init,
380 	.exit = ip_rt_do_proc_exit,
381 };
382 
383 static int __init ip_rt_proc_init(void)
384 {
385 	return register_pernet_subsys(&ip_rt_proc_ops);
386 }
387 
388 #else
389 static inline int ip_rt_proc_init(void)
390 {
391 	return 0;
392 }
393 #endif /* CONFIG_PROC_FS */
394 
395 static inline bool rt_is_expired(const struct rtable *rth)
396 {
397 	bool res;
398 
399 	rcu_read_lock();
400 	res = rth->rt_genid != rt_genid_ipv4(dev_net_rcu(rth->dst.dev));
401 	rcu_read_unlock();
402 
403 	return res;
404 }
405 
406 void rt_cache_flush(struct net *net)
407 {
408 	rt_genid_bump_ipv4(net);
409 }
410 
411 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
412 					   struct sk_buff *skb,
413 					   const void *daddr)
414 {
415 	const struct rtable *rt = container_of(dst, struct rtable, dst);
416 	struct net_device *dev = dst->dev;
417 	struct neighbour *n;
418 
419 	rcu_read_lock();
420 
421 	if (likely(rt->rt_gw_family == AF_INET)) {
422 		n = ip_neigh_gw4(dev, rt->rt_gw4);
423 	} else if (rt->rt_gw_family == AF_INET6) {
424 		n = ip_neigh_gw6(dev, &rt->rt_gw6);
425         } else {
426 		__be32 pkey;
427 
428 		pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
429 		n = ip_neigh_gw4(dev, pkey);
430 	}
431 
432 	if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
433 		n = NULL;
434 
435 	rcu_read_unlock();
436 
437 	return n;
438 }
439 
440 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
441 {
442 	const struct rtable *rt = container_of(dst, struct rtable, dst);
443 	struct net_device *dev = dst->dev;
444 	const __be32 *pkey = daddr;
445 
446 	if (rt->rt_gw_family == AF_INET) {
447 		pkey = (const __be32 *)&rt->rt_gw4;
448 	} else if (rt->rt_gw_family == AF_INET6) {
449 		return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
450 	} else if (!daddr ||
451 		 (rt->rt_flags &
452 		  (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
453 		return;
454 	}
455 	__ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
456 }
457 
458 /* Hash tables of size 2048..262144 depending on RAM size.
459  * Each bucket uses 8 bytes.
460  */
461 static u32 ip_idents_mask __read_mostly;
462 static atomic_t *ip_idents __read_mostly;
463 static u32 *ip_tstamps __read_mostly;
464 
465 /* In order to protect privacy, we add a perturbation to identifiers
466  * if one generator is seldom used. This makes hard for an attacker
467  * to infer how many packets were sent between two points in time.
468  */
469 static u32 ip_idents_reserve(u32 hash, int segs)
470 {
471 	u32 bucket, old, now = (u32)jiffies;
472 	atomic_t *p_id;
473 	u32 *p_tstamp;
474 	u32 delta = 0;
475 
476 	bucket = hash & ip_idents_mask;
477 	p_tstamp = ip_tstamps + bucket;
478 	p_id = ip_idents + bucket;
479 	old = READ_ONCE(*p_tstamp);
480 
481 	if (old != now && cmpxchg(p_tstamp, old, now) == old)
482 		delta = get_random_u32_below(now - old);
483 
484 	/* If UBSAN reports an error there, please make sure your compiler
485 	 * supports -fno-strict-overflow before reporting it that was a bug
486 	 * in UBSAN, and it has been fixed in GCC-8.
487 	 */
488 	return atomic_add_return(segs + delta, p_id) - segs;
489 }
490 
491 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
492 {
493 	u32 hash, id;
494 
495 	/* Note the following code is not safe, but this is okay. */
496 	if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
497 		get_random_bytes(&net->ipv4.ip_id_key,
498 				 sizeof(net->ipv4.ip_id_key));
499 
500 	hash = siphash_3u32((__force u32)iph->daddr,
501 			    (__force u32)iph->saddr,
502 			    iph->protocol,
503 			    &net->ipv4.ip_id_key);
504 	id = ip_idents_reserve(hash, segs);
505 	iph->id = htons(id);
506 }
507 EXPORT_SYMBOL(__ip_select_ident);
508 
509 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
510 			     const struct sock *sk, const struct iphdr *iph,
511 			     int oif, __u8 tos, u8 prot, u32 mark,
512 			     int flow_flags)
513 {
514 	__u8 scope = RT_SCOPE_UNIVERSE;
515 
516 	if (sk) {
517 		oif = sk->sk_bound_dev_if;
518 		mark = READ_ONCE(sk->sk_mark);
519 		tos = ip_sock_rt_tos(sk);
520 		scope = ip_sock_rt_scope(sk);
521 		prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW :
522 						    sk->sk_protocol;
523 	}
524 
525 	flowi4_init_output(fl4, oif, mark, tos & INET_DSCP_MASK, scope,
526 			   prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
527 			   sock_net_uid(net, sk));
528 }
529 
530 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
531 			       const struct sock *sk)
532 {
533 	const struct net *net = dev_net(skb->dev);
534 	const struct iphdr *iph = ip_hdr(skb);
535 	int oif = skb->dev->ifindex;
536 	u8 prot = iph->protocol;
537 	u32 mark = skb->mark;
538 	__u8 tos = iph->tos;
539 
540 	__build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
541 }
542 
543 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
544 {
545 	const struct inet_sock *inet = inet_sk(sk);
546 	const struct ip_options_rcu *inet_opt;
547 	__be32 daddr = inet->inet_daddr;
548 
549 	rcu_read_lock();
550 	inet_opt = rcu_dereference(inet->inet_opt);
551 	if (inet_opt && inet_opt->opt.srr)
552 		daddr = inet_opt->opt.faddr;
553 	flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark),
554 			   ip_sock_rt_tos(sk),
555 			   ip_sock_rt_scope(sk),
556 			   inet_test_bit(HDRINCL, sk) ?
557 				IPPROTO_RAW : sk->sk_protocol,
558 			   inet_sk_flowi_flags(sk),
559 			   daddr, inet->inet_saddr, 0, 0,
560 			   sk_uid(sk));
561 	rcu_read_unlock();
562 }
563 
564 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
565 				 const struct sk_buff *skb)
566 {
567 	if (skb)
568 		build_skb_flow_key(fl4, skb, sk);
569 	else
570 		build_sk_flow_key(fl4, sk);
571 }
572 
573 static DEFINE_SPINLOCK(fnhe_lock);
574 
575 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
576 {
577 	struct rtable *rt;
578 
579 	rt = rcu_dereference(fnhe->fnhe_rth_input);
580 	if (rt) {
581 		RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
582 		dst_dev_put(&rt->dst);
583 		dst_release(&rt->dst);
584 	}
585 	rt = rcu_dereference(fnhe->fnhe_rth_output);
586 	if (rt) {
587 		RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
588 		dst_dev_put(&rt->dst);
589 		dst_release(&rt->dst);
590 	}
591 }
592 
593 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
594 {
595 	struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
596 	struct fib_nh_exception *fnhe, *oldest = NULL;
597 
598 	for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
599 		fnhe = rcu_dereference_protected(*fnhe_p,
600 						 lockdep_is_held(&fnhe_lock));
601 		if (!fnhe)
602 			break;
603 		if (!oldest ||
604 		    time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
605 			oldest = fnhe;
606 			oldest_p = fnhe_p;
607 		}
608 	}
609 	fnhe_flush_routes(oldest);
610 	*oldest_p = oldest->fnhe_next;
611 	kfree_rcu(oldest, rcu);
612 }
613 
614 static u32 fnhe_hashfun(__be32 daddr)
615 {
616 	static siphash_aligned_key_t fnhe_hash_key;
617 	u64 hval;
618 
619 	net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
620 	hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
621 	return hash_64(hval, FNHE_HASH_SHIFT);
622 }
623 
624 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
625 {
626 	rt->rt_pmtu = fnhe->fnhe_pmtu;
627 	rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
628 	rt->dst.expires = fnhe->fnhe_expires;
629 
630 	if (fnhe->fnhe_gw) {
631 		rt->rt_flags |= RTCF_REDIRECTED;
632 		rt->rt_uses_gateway = 1;
633 		rt->rt_gw_family = AF_INET;
634 		rt->rt_gw4 = fnhe->fnhe_gw;
635 	}
636 }
637 
638 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
639 				  __be32 gw, u32 pmtu, bool lock,
640 				  unsigned long expires)
641 {
642 	struct fnhe_hash_bucket *hash;
643 	struct fib_nh_exception *fnhe;
644 	struct rtable *rt;
645 	u32 genid, hval;
646 	unsigned int i;
647 	int depth;
648 
649 	genid = fnhe_genid(dev_net(nhc->nhc_dev));
650 	hval = fnhe_hashfun(daddr);
651 
652 	spin_lock_bh(&fnhe_lock);
653 
654 	hash = rcu_dereference(nhc->nhc_exceptions);
655 	if (!hash) {
656 		hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
657 		if (!hash)
658 			goto out_unlock;
659 		rcu_assign_pointer(nhc->nhc_exceptions, hash);
660 	}
661 
662 	hash += hval;
663 
664 	depth = 0;
665 	for (fnhe = rcu_dereference(hash->chain); fnhe;
666 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
667 		if (fnhe->fnhe_daddr == daddr)
668 			break;
669 		depth++;
670 	}
671 
672 	if (fnhe) {
673 		if (fnhe->fnhe_genid != genid)
674 			fnhe->fnhe_genid = genid;
675 		if (gw)
676 			fnhe->fnhe_gw = gw;
677 		if (pmtu) {
678 			fnhe->fnhe_pmtu = pmtu;
679 			fnhe->fnhe_mtu_locked = lock;
680 		}
681 		fnhe->fnhe_expires = max(1UL, expires);
682 		/* Update all cached dsts too */
683 		rt = rcu_dereference(fnhe->fnhe_rth_input);
684 		if (rt)
685 			fill_route_from_fnhe(rt, fnhe);
686 		rt = rcu_dereference(fnhe->fnhe_rth_output);
687 		if (rt)
688 			fill_route_from_fnhe(rt, fnhe);
689 	} else {
690 		/* Randomize max depth to avoid some side channels attacks. */
691 		int max_depth = FNHE_RECLAIM_DEPTH +
692 				get_random_u32_below(FNHE_RECLAIM_DEPTH);
693 
694 		while (depth > max_depth) {
695 			fnhe_remove_oldest(hash);
696 			depth--;
697 		}
698 
699 		fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
700 		if (!fnhe)
701 			goto out_unlock;
702 
703 		fnhe->fnhe_next = hash->chain;
704 
705 		fnhe->fnhe_genid = genid;
706 		fnhe->fnhe_daddr = daddr;
707 		fnhe->fnhe_gw = gw;
708 		fnhe->fnhe_pmtu = pmtu;
709 		fnhe->fnhe_mtu_locked = lock;
710 		fnhe->fnhe_expires = max(1UL, expires);
711 
712 		rcu_assign_pointer(hash->chain, fnhe);
713 
714 		/* Exception created; mark the cached routes for the nexthop
715 		 * stale, so anyone caching it rechecks if this exception
716 		 * applies to them.
717 		 */
718 		rt = rcu_dereference(nhc->nhc_rth_input);
719 		if (rt)
720 			rt->dst.obsolete = DST_OBSOLETE_KILL;
721 
722 		for_each_possible_cpu(i) {
723 			struct rtable __rcu **prt;
724 
725 			prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
726 			rt = rcu_dereference(*prt);
727 			if (rt)
728 				rt->dst.obsolete = DST_OBSOLETE_KILL;
729 		}
730 	}
731 
732 	fnhe->fnhe_stamp = jiffies;
733 
734 out_unlock:
735 	spin_unlock_bh(&fnhe_lock);
736 }
737 
738 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
739 			     bool kill_route)
740 {
741 	__be32 new_gw = icmp_hdr(skb)->un.gateway;
742 	__be32 old_gw = ip_hdr(skb)->saddr;
743 	struct net_device *dev = skb->dev;
744 	struct in_device *in_dev;
745 	struct fib_result res;
746 	struct neighbour *n;
747 	struct net *net;
748 
749 	switch (icmp_hdr(skb)->code & 7) {
750 	case ICMP_REDIR_NET:
751 	case ICMP_REDIR_NETTOS:
752 	case ICMP_REDIR_HOST:
753 	case ICMP_REDIR_HOSTTOS:
754 		break;
755 
756 	default:
757 		return;
758 	}
759 
760 	if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
761 		return;
762 
763 	in_dev = __in_dev_get_rcu(dev);
764 	if (!in_dev)
765 		return;
766 
767 	net = dev_net(dev);
768 	if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
769 	    ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
770 	    ipv4_is_zeronet(new_gw))
771 		goto reject_redirect;
772 
773 	if (!IN_DEV_SHARED_MEDIA(in_dev)) {
774 		if (!inet_addr_onlink(in_dev, new_gw, old_gw))
775 			goto reject_redirect;
776 		if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
777 			goto reject_redirect;
778 	} else {
779 		if (inet_addr_type(net, new_gw) != RTN_UNICAST)
780 			goto reject_redirect;
781 	}
782 
783 	n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw);
784 	if (!n)
785 		n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
786 	if (!IS_ERR(n)) {
787 		if (!(READ_ONCE(n->nud_state) & NUD_VALID)) {
788 			neigh_event_send(n, NULL);
789 		} else {
790 			if (fib_lookup(net, fl4, &res, 0) == 0) {
791 				struct fib_nh_common *nhc;
792 
793 				fib_select_path(net, &res, fl4, skb);
794 				nhc = FIB_RES_NHC(res);
795 				update_or_create_fnhe(nhc, fl4->daddr, new_gw,
796 						0, false,
797 						jiffies + ip_rt_gc_timeout);
798 			}
799 			if (kill_route)
800 				rt->dst.obsolete = DST_OBSOLETE_KILL;
801 			call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
802 		}
803 		neigh_release(n);
804 	}
805 	return;
806 
807 reject_redirect:
808 #ifdef CONFIG_IP_ROUTE_VERBOSE
809 	if (IN_DEV_LOG_MARTIANS(in_dev)) {
810 		const struct iphdr *iph = (const struct iphdr *) skb->data;
811 		__be32 daddr = iph->daddr;
812 		__be32 saddr = iph->saddr;
813 
814 		net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
815 				     "  Advised path = %pI4 -> %pI4\n",
816 				     &old_gw, dev->name, &new_gw,
817 				     &saddr, &daddr);
818 	}
819 #endif
820 	;
821 }
822 
823 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
824 {
825 	struct rtable *rt;
826 	struct flowi4 fl4;
827 	const struct iphdr *iph = (const struct iphdr *) skb->data;
828 	struct net *net = dev_net(skb->dev);
829 	int oif = skb->dev->ifindex;
830 	u8 prot = iph->protocol;
831 	u32 mark = skb->mark;
832 	__u8 tos = iph->tos;
833 
834 	rt = dst_rtable(dst);
835 
836 	__build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
837 	__ip_do_redirect(rt, skb, &fl4, true);
838 }
839 
840 static void ipv4_negative_advice(struct sock *sk,
841 				 struct dst_entry *dst)
842 {
843 	struct rtable *rt = dst_rtable(dst);
844 
845 	if ((dst->obsolete > 0) ||
846 	    (rt->rt_flags & RTCF_REDIRECTED) ||
847 	    rt->dst.expires)
848 		sk_dst_reset(sk);
849 }
850 
851 /*
852  * Algorithm:
853  *	1. The first ip_rt_redirect_number redirects are sent
854  *	   with exponential backoff, then we stop sending them at all,
855  *	   assuming that the host ignores our redirects.
856  *	2. If we did not see packets requiring redirects
857  *	   during ip_rt_redirect_silence, we assume that the host
858  *	   forgot redirected route and start to send redirects again.
859  *
860  * This algorithm is much cheaper and more intelligent than dumb load limiting
861  * in icmp.c.
862  *
863  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
864  * and "frag. need" (breaks PMTU discovery) in icmp.c.
865  */
866 
867 void ip_rt_send_redirect(struct sk_buff *skb)
868 {
869 	struct rtable *rt = skb_rtable(skb);
870 	struct in_device *in_dev;
871 	struct inet_peer *peer;
872 	struct net *net;
873 	int log_martians;
874 	int vif;
875 
876 	rcu_read_lock();
877 	in_dev = __in_dev_get_rcu(rt->dst.dev);
878 	if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
879 		rcu_read_unlock();
880 		return;
881 	}
882 	log_martians = IN_DEV_LOG_MARTIANS(in_dev);
883 	vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
884 
885 	net = dev_net(rt->dst.dev);
886 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif);
887 	if (!peer) {
888 		rcu_read_unlock();
889 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
890 			  rt_nexthop(rt, ip_hdr(skb)->daddr));
891 		return;
892 	}
893 
894 	/* No redirected packets during ip_rt_redirect_silence;
895 	 * reset the algorithm.
896 	 */
897 	if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
898 		peer->rate_tokens = 0;
899 		peer->n_redirects = 0;
900 	}
901 
902 	/* Too many ignored redirects; do not send anything
903 	 * set dst.rate_last to the last seen redirected packet.
904 	 */
905 	if (peer->n_redirects >= ip_rt_redirect_number) {
906 		peer->rate_last = jiffies;
907 		goto out_unlock;
908 	}
909 
910 	/* Check for load limit; set rate_last to the latest sent
911 	 * redirect.
912 	 */
913 	if (peer->n_redirects == 0 ||
914 	    time_after(jiffies,
915 		       (peer->rate_last +
916 			(ip_rt_redirect_load << peer->n_redirects)))) {
917 		__be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
918 
919 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
920 		peer->rate_last = jiffies;
921 		++peer->n_redirects;
922 		if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians &&
923 		    peer->n_redirects == ip_rt_redirect_number)
924 			net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
925 					     &ip_hdr(skb)->saddr, inet_iif(skb),
926 					     &ip_hdr(skb)->daddr, &gw);
927 	}
928 out_unlock:
929 	rcu_read_unlock();
930 }
931 
932 static int ip_error(struct sk_buff *skb)
933 {
934 	struct rtable *rt = skb_rtable(skb);
935 	struct net_device *dev = skb->dev;
936 	struct in_device *in_dev;
937 	struct inet_peer *peer;
938 	unsigned long now;
939 	struct net *net;
940 	SKB_DR(reason);
941 	bool send;
942 	int code;
943 
944 	if (netif_is_l3_master(skb->dev)) {
945 		dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
946 		if (!dev)
947 			goto out;
948 	}
949 
950 	in_dev = __in_dev_get_rcu(dev);
951 
952 	/* IP on this device is disabled. */
953 	if (!in_dev)
954 		goto out;
955 
956 	net = dev_net(rt->dst.dev);
957 	if (!IN_DEV_FORWARD(in_dev)) {
958 		switch (rt->dst.error) {
959 		case EHOSTUNREACH:
960 			SKB_DR_SET(reason, IP_INADDRERRORS);
961 			__IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
962 			break;
963 
964 		case ENETUNREACH:
965 			SKB_DR_SET(reason, IP_INNOROUTES);
966 			__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
967 			break;
968 		}
969 		goto out;
970 	}
971 
972 	switch (rt->dst.error) {
973 	case EINVAL:
974 	default:
975 		goto out;
976 	case EHOSTUNREACH:
977 		code = ICMP_HOST_UNREACH;
978 		break;
979 	case ENETUNREACH:
980 		code = ICMP_NET_UNREACH;
981 		SKB_DR_SET(reason, IP_INNOROUTES);
982 		__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
983 		break;
984 	case EACCES:
985 		code = ICMP_PKT_FILTERED;
986 		break;
987 	}
988 
989 	rcu_read_lock();
990 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
991 			       l3mdev_master_ifindex_rcu(skb->dev));
992 	send = true;
993 	if (peer) {
994 		now = jiffies;
995 		peer->rate_tokens += now - peer->rate_last;
996 		if (peer->rate_tokens > ip_rt_error_burst)
997 			peer->rate_tokens = ip_rt_error_burst;
998 		peer->rate_last = now;
999 		if (peer->rate_tokens >= ip_rt_error_cost)
1000 			peer->rate_tokens -= ip_rt_error_cost;
1001 		else
1002 			send = false;
1003 	}
1004 	rcu_read_unlock();
1005 
1006 	if (send)
1007 		icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1008 
1009 out:	kfree_skb_reason(skb, reason);
1010 	return 0;
1011 }
1012 
1013 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1014 {
1015 	struct dst_entry *dst = &rt->dst;
1016 	struct fib_result res;
1017 	bool lock = false;
1018 	struct net *net;
1019 	u32 old_mtu;
1020 
1021 	if (ip_mtu_locked(dst))
1022 		return;
1023 
1024 	old_mtu = ipv4_mtu(dst);
1025 	if (old_mtu < mtu)
1026 		return;
1027 
1028 	rcu_read_lock();
1029 	net = dev_net_rcu(dst->dev);
1030 	if (mtu < net->ipv4.ip_rt_min_pmtu) {
1031 		lock = true;
1032 		mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1033 	}
1034 
1035 	if (rt->rt_pmtu == mtu && !lock &&
1036 	    time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2))
1037 		goto out;
1038 
1039 	if (fib_lookup(net, fl4, &res, 0) == 0) {
1040 		struct fib_nh_common *nhc;
1041 
1042 		fib_select_path(net, &res, fl4, NULL);
1043 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1044 		if (fib_info_num_path(res.fi) > 1) {
1045 			int nhsel;
1046 
1047 			for (nhsel = 0; nhsel < fib_info_num_path(res.fi); nhsel++) {
1048 				nhc = fib_info_nhc(res.fi, nhsel);
1049 				update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1050 						      jiffies + net->ipv4.ip_rt_mtu_expires);
1051 			}
1052 			goto out;
1053 		}
1054 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1055 		nhc = FIB_RES_NHC(res);
1056 		update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1057 				      jiffies + net->ipv4.ip_rt_mtu_expires);
1058 	}
1059 out:
1060 	rcu_read_unlock();
1061 }
1062 
1063 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1064 			      struct sk_buff *skb, u32 mtu,
1065 			      bool confirm_neigh)
1066 {
1067 	struct rtable *rt = dst_rtable(dst);
1068 	struct flowi4 fl4;
1069 
1070 	ip_rt_build_flow_key(&fl4, sk, skb);
1071 
1072 	/* Don't make lookup fail for bridged encapsulations */
1073 	if (skb && netif_is_any_bridge_port(skb->dev))
1074 		fl4.flowi4_oif = 0;
1075 
1076 	__ip_rt_update_pmtu(rt, &fl4, mtu);
1077 }
1078 
1079 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1080 		      int oif, u8 protocol)
1081 {
1082 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1083 	struct flowi4 fl4;
1084 	struct rtable *rt;
1085 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
1086 
1087 	__build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1088 			 0);
1089 	rt = __ip_route_output_key(net, &fl4);
1090 	if (!IS_ERR(rt)) {
1091 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1092 		ip_rt_put(rt);
1093 	}
1094 }
1095 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1096 
1097 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1098 {
1099 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1100 	struct flowi4 fl4;
1101 	struct rtable *rt;
1102 
1103 	__build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1104 
1105 	if (!fl4.flowi4_mark)
1106 		fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1107 
1108 	rt = __ip_route_output_key(sock_net(sk), &fl4);
1109 	if (!IS_ERR(rt)) {
1110 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1111 		ip_rt_put(rt);
1112 	}
1113 }
1114 
1115 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1116 {
1117 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1118 	struct flowi4 fl4;
1119 	struct rtable *rt;
1120 	struct dst_entry *odst = NULL;
1121 	bool new = false;
1122 	struct net *net = sock_net(sk);
1123 
1124 	bh_lock_sock(sk);
1125 
1126 	if (!ip_sk_accept_pmtu(sk))
1127 		goto out;
1128 
1129 	odst = sk_dst_get(sk);
1130 
1131 	if (sock_owned_by_user(sk) || !odst) {
1132 		__ipv4_sk_update_pmtu(skb, sk, mtu);
1133 		goto out;
1134 	}
1135 
1136 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1137 
1138 	rt = dst_rtable(odst);
1139 	if (odst->obsolete && !odst->ops->check(odst, 0)) {
1140 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1141 		if (IS_ERR(rt))
1142 			goto out;
1143 
1144 		new = true;
1145 	}
1146 
1147 	__ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu);
1148 
1149 	if (!dst_check(&rt->dst, 0)) {
1150 		if (new)
1151 			dst_release(&rt->dst);
1152 
1153 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1154 		if (IS_ERR(rt))
1155 			goto out;
1156 
1157 		new = true;
1158 	}
1159 
1160 	if (new)
1161 		sk_dst_set(sk, &rt->dst);
1162 
1163 out:
1164 	bh_unlock_sock(sk);
1165 	dst_release(odst);
1166 }
1167 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1168 
1169 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1170 		   int oif, u8 protocol)
1171 {
1172 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1173 	struct flowi4 fl4;
1174 	struct rtable *rt;
1175 
1176 	__build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1177 	rt = __ip_route_output_key(net, &fl4);
1178 	if (!IS_ERR(rt)) {
1179 		__ip_do_redirect(rt, skb, &fl4, false);
1180 		ip_rt_put(rt);
1181 	}
1182 }
1183 EXPORT_SYMBOL_GPL(ipv4_redirect);
1184 
1185 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1186 {
1187 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1188 	struct flowi4 fl4;
1189 	struct rtable *rt;
1190 	struct net *net = sock_net(sk);
1191 
1192 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1193 	rt = __ip_route_output_key(net, &fl4);
1194 	if (!IS_ERR(rt)) {
1195 		__ip_do_redirect(rt, skb, &fl4, false);
1196 		ip_rt_put(rt);
1197 	}
1198 }
1199 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1200 
1201 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1202 							 u32 cookie)
1203 {
1204 	struct rtable *rt = dst_rtable(dst);
1205 
1206 	/* All IPV4 dsts are created with ->obsolete set to the value
1207 	 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1208 	 * into this function always.
1209 	 *
1210 	 * When a PMTU/redirect information update invalidates a route,
1211 	 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1212 	 * DST_OBSOLETE_DEAD.
1213 	 */
1214 	if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1215 		return NULL;
1216 	return dst;
1217 }
1218 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1219 
1220 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1221 {
1222 	struct net_device *dev;
1223 	struct ip_options opt;
1224 	int res;
1225 
1226 	/* Recompile ip options since IPCB may not be valid anymore.
1227 	 * Also check we have a reasonable ipv4 header.
1228 	 */
1229 	if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1230 	    ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1231 		return;
1232 
1233 	memset(&opt, 0, sizeof(opt));
1234 	if (ip_hdr(skb)->ihl > 5) {
1235 		if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1236 			return;
1237 		opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1238 
1239 		rcu_read_lock();
1240 		dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1241 		res = __ip_options_compile(dev_net(dev), &opt, skb, NULL);
1242 		rcu_read_unlock();
1243 
1244 		if (res)
1245 			return;
1246 	}
1247 	__icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1248 }
1249 
1250 static void ipv4_link_failure(struct sk_buff *skb)
1251 {
1252 	struct rtable *rt;
1253 
1254 	ipv4_send_dest_unreach(skb);
1255 
1256 	rt = skb_rtable(skb);
1257 	if (rt)
1258 		dst_set_expires(&rt->dst, 0);
1259 }
1260 
1261 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1262 {
1263 	pr_debug("%s: %pI4 -> %pI4, %s\n",
1264 		 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1265 		 skb->dev ? skb->dev->name : "?");
1266 	kfree_skb(skb);
1267 	WARN_ON(1);
1268 	return 0;
1269 }
1270 
1271 /*
1272  * We do not cache source address of outgoing interface,
1273  * because it is used only by IP RR, TS and SRR options,
1274  * so that it out of fast path.
1275  *
1276  * BTW remember: "addr" is allowed to be not aligned
1277  * in IP options!
1278  */
1279 
1280 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1281 {
1282 	__be32 src;
1283 
1284 	if (rt_is_output_route(rt))
1285 		src = ip_hdr(skb)->saddr;
1286 	else {
1287 		struct fib_result res;
1288 		struct iphdr *iph = ip_hdr(skb);
1289 		struct flowi4 fl4 = {
1290 			.daddr = iph->daddr,
1291 			.saddr = iph->saddr,
1292 			.flowi4_tos = inet_dscp_to_dsfield(ip4h_dscp(iph)),
1293 			.flowi4_oif = rt->dst.dev->ifindex,
1294 			.flowi4_iif = skb->dev->ifindex,
1295 			.flowi4_mark = skb->mark,
1296 		};
1297 
1298 		rcu_read_lock();
1299 		if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1300 			src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1301 		else
1302 			src = inet_select_addr(rt->dst.dev,
1303 					       rt_nexthop(rt, iph->daddr),
1304 					       RT_SCOPE_UNIVERSE);
1305 		rcu_read_unlock();
1306 	}
1307 	memcpy(addr, &src, 4);
1308 }
1309 
1310 #ifdef CONFIG_IP_ROUTE_CLASSID
1311 static void set_class_tag(struct rtable *rt, u32 tag)
1312 {
1313 	if (!(rt->dst.tclassid & 0xFFFF))
1314 		rt->dst.tclassid |= tag & 0xFFFF;
1315 	if (!(rt->dst.tclassid & 0xFFFF0000))
1316 		rt->dst.tclassid |= tag & 0xFFFF0000;
1317 }
1318 #endif
1319 
1320 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1321 {
1322 	unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1323 	unsigned int advmss;
1324 	struct net *net;
1325 
1326 	rcu_read_lock();
1327 	net = dev_net_rcu(dst->dev);
1328 	advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1329 				   net->ipv4.ip_rt_min_advmss);
1330 	rcu_read_unlock();
1331 
1332 	return min(advmss, IPV4_MAX_PMTU - header_size);
1333 }
1334 
1335 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1336 {
1337 	return ip_dst_mtu_maybe_forward(dst, false);
1338 }
1339 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1340 
1341 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1342 {
1343 	struct fnhe_hash_bucket *hash;
1344 	struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1345 	u32 hval = fnhe_hashfun(daddr);
1346 
1347 	spin_lock_bh(&fnhe_lock);
1348 
1349 	hash = rcu_dereference_protected(nhc->nhc_exceptions,
1350 					 lockdep_is_held(&fnhe_lock));
1351 	hash += hval;
1352 
1353 	fnhe_p = &hash->chain;
1354 	fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1355 	while (fnhe) {
1356 		if (fnhe->fnhe_daddr == daddr) {
1357 			rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1358 				fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1359 			/* set fnhe_daddr to 0 to ensure it won't bind with
1360 			 * new dsts in rt_bind_exception().
1361 			 */
1362 			fnhe->fnhe_daddr = 0;
1363 			fnhe_flush_routes(fnhe);
1364 			kfree_rcu(fnhe, rcu);
1365 			break;
1366 		}
1367 		fnhe_p = &fnhe->fnhe_next;
1368 		fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1369 						 lockdep_is_held(&fnhe_lock));
1370 	}
1371 
1372 	spin_unlock_bh(&fnhe_lock);
1373 }
1374 
1375 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1376 					       __be32 daddr)
1377 {
1378 	struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1379 	struct fib_nh_exception *fnhe;
1380 	u32 hval;
1381 
1382 	if (!hash)
1383 		return NULL;
1384 
1385 	hval = fnhe_hashfun(daddr);
1386 
1387 	for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1388 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
1389 		if (fnhe->fnhe_daddr == daddr) {
1390 			if (fnhe->fnhe_expires &&
1391 			    time_after(jiffies, fnhe->fnhe_expires)) {
1392 				ip_del_fnhe(nhc, daddr);
1393 				break;
1394 			}
1395 			return fnhe;
1396 		}
1397 	}
1398 	return NULL;
1399 }
1400 
1401 /* MTU selection:
1402  * 1. mtu on route is locked - use it
1403  * 2. mtu from nexthop exception
1404  * 3. mtu from egress device
1405  */
1406 
1407 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1408 {
1409 	struct fib_nh_common *nhc = res->nhc;
1410 	struct net_device *dev = nhc->nhc_dev;
1411 	struct fib_info *fi = res->fi;
1412 	u32 mtu = 0;
1413 
1414 	if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1415 	    fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1416 		mtu = fi->fib_mtu;
1417 
1418 	if (likely(!mtu)) {
1419 		struct fib_nh_exception *fnhe;
1420 
1421 		fnhe = find_exception(nhc, daddr);
1422 		if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1423 			mtu = fnhe->fnhe_pmtu;
1424 	}
1425 
1426 	if (likely(!mtu))
1427 		mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1428 
1429 	return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1430 }
1431 
1432 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1433 			      __be32 daddr, const bool do_cache)
1434 {
1435 	bool ret = false;
1436 
1437 	spin_lock_bh(&fnhe_lock);
1438 
1439 	if (daddr == fnhe->fnhe_daddr) {
1440 		struct rtable __rcu **porig;
1441 		struct rtable *orig;
1442 		int genid = fnhe_genid(dev_net(rt->dst.dev));
1443 
1444 		if (rt_is_input_route(rt))
1445 			porig = &fnhe->fnhe_rth_input;
1446 		else
1447 			porig = &fnhe->fnhe_rth_output;
1448 		orig = rcu_dereference(*porig);
1449 
1450 		if (fnhe->fnhe_genid != genid) {
1451 			fnhe->fnhe_genid = genid;
1452 			fnhe->fnhe_gw = 0;
1453 			fnhe->fnhe_pmtu = 0;
1454 			fnhe->fnhe_expires = 0;
1455 			fnhe->fnhe_mtu_locked = false;
1456 			fnhe_flush_routes(fnhe);
1457 			orig = NULL;
1458 		}
1459 		fill_route_from_fnhe(rt, fnhe);
1460 		if (!rt->rt_gw4) {
1461 			rt->rt_gw4 = daddr;
1462 			rt->rt_gw_family = AF_INET;
1463 		}
1464 
1465 		if (do_cache) {
1466 			dst_hold(&rt->dst);
1467 			rcu_assign_pointer(*porig, rt);
1468 			if (orig) {
1469 				dst_dev_put(&orig->dst);
1470 				dst_release(&orig->dst);
1471 			}
1472 			ret = true;
1473 		}
1474 
1475 		fnhe->fnhe_stamp = jiffies;
1476 	}
1477 	spin_unlock_bh(&fnhe_lock);
1478 
1479 	return ret;
1480 }
1481 
1482 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1483 {
1484 	struct rtable *orig, *prev, **p;
1485 	bool ret = true;
1486 
1487 	if (rt_is_input_route(rt)) {
1488 		p = (struct rtable **)&nhc->nhc_rth_input;
1489 	} else {
1490 		p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1491 	}
1492 	orig = *p;
1493 
1494 	/* hold dst before doing cmpxchg() to avoid race condition
1495 	 * on this dst
1496 	 */
1497 	dst_hold(&rt->dst);
1498 	prev = cmpxchg(p, orig, rt);
1499 	if (prev == orig) {
1500 		if (orig) {
1501 			rt_add_uncached_list(orig);
1502 			dst_release(&orig->dst);
1503 		}
1504 	} else {
1505 		dst_release(&rt->dst);
1506 		ret = false;
1507 	}
1508 
1509 	return ret;
1510 }
1511 
1512 struct uncached_list {
1513 	spinlock_t		lock;
1514 	struct list_head	head;
1515 };
1516 
1517 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1518 
1519 void rt_add_uncached_list(struct rtable *rt)
1520 {
1521 	struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1522 
1523 	rt->dst.rt_uncached_list = ul;
1524 
1525 	spin_lock_bh(&ul->lock);
1526 	list_add_tail(&rt->dst.rt_uncached, &ul->head);
1527 	spin_unlock_bh(&ul->lock);
1528 }
1529 
1530 void rt_del_uncached_list(struct rtable *rt)
1531 {
1532 	if (!list_empty(&rt->dst.rt_uncached)) {
1533 		struct uncached_list *ul = rt->dst.rt_uncached_list;
1534 
1535 		spin_lock_bh(&ul->lock);
1536 		list_del_init(&rt->dst.rt_uncached);
1537 		spin_unlock_bh(&ul->lock);
1538 	}
1539 }
1540 
1541 static void ipv4_dst_destroy(struct dst_entry *dst)
1542 {
1543 	ip_dst_metrics_put(dst);
1544 	rt_del_uncached_list(dst_rtable(dst));
1545 }
1546 
1547 void rt_flush_dev(struct net_device *dev)
1548 {
1549 	struct rtable *rt, *safe;
1550 	int cpu;
1551 
1552 	for_each_possible_cpu(cpu) {
1553 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1554 
1555 		if (list_empty(&ul->head))
1556 			continue;
1557 
1558 		spin_lock_bh(&ul->lock);
1559 		list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1560 			if (rt->dst.dev != dev)
1561 				continue;
1562 			rt->dst.dev = blackhole_netdev;
1563 			netdev_ref_replace(dev, blackhole_netdev,
1564 					   &rt->dst.dev_tracker, GFP_ATOMIC);
1565 			list_del_init(&rt->dst.rt_uncached);
1566 		}
1567 		spin_unlock_bh(&ul->lock);
1568 	}
1569 }
1570 
1571 static bool rt_cache_valid(const struct rtable *rt)
1572 {
1573 	return	rt &&
1574 		rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1575 		!rt_is_expired(rt);
1576 }
1577 
1578 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1579 			   const struct fib_result *res,
1580 			   struct fib_nh_exception *fnhe,
1581 			   struct fib_info *fi, u16 type, u32 itag,
1582 			   const bool do_cache)
1583 {
1584 	bool cached = false;
1585 
1586 	if (fi) {
1587 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1588 
1589 		if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1590 			rt->rt_uses_gateway = 1;
1591 			rt->rt_gw_family = nhc->nhc_gw_family;
1592 			/* only INET and INET6 are supported */
1593 			if (likely(nhc->nhc_gw_family == AF_INET))
1594 				rt->rt_gw4 = nhc->nhc_gw.ipv4;
1595 			else
1596 				rt->rt_gw6 = nhc->nhc_gw.ipv6;
1597 		}
1598 
1599 		ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1600 
1601 #ifdef CONFIG_IP_ROUTE_CLASSID
1602 		if (nhc->nhc_family == AF_INET) {
1603 			struct fib_nh *nh;
1604 
1605 			nh = container_of(nhc, struct fib_nh, nh_common);
1606 			rt->dst.tclassid = nh->nh_tclassid;
1607 		}
1608 #endif
1609 		rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1610 		if (unlikely(fnhe))
1611 			cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1612 		else if (do_cache)
1613 			cached = rt_cache_route(nhc, rt);
1614 		if (unlikely(!cached)) {
1615 			/* Routes we intend to cache in nexthop exception or
1616 			 * FIB nexthop have the DST_NOCACHE bit clear.
1617 			 * However, if we are unsuccessful at storing this
1618 			 * route into the cache we really need to set it.
1619 			 */
1620 			if (!rt->rt_gw4) {
1621 				rt->rt_gw_family = AF_INET;
1622 				rt->rt_gw4 = daddr;
1623 			}
1624 			rt_add_uncached_list(rt);
1625 		}
1626 	} else
1627 		rt_add_uncached_list(rt);
1628 
1629 #ifdef CONFIG_IP_ROUTE_CLASSID
1630 #ifdef CONFIG_IP_MULTIPLE_TABLES
1631 	set_class_tag(rt, res->tclassid);
1632 #endif
1633 	set_class_tag(rt, itag);
1634 #endif
1635 }
1636 
1637 struct rtable *rt_dst_alloc(struct net_device *dev,
1638 			    unsigned int flags, u16 type,
1639 			    bool noxfrm)
1640 {
1641 	struct rtable *rt;
1642 
1643 	rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1644 		       (noxfrm ? DST_NOXFRM : 0));
1645 
1646 	if (rt) {
1647 		rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1648 		rt->rt_flags = flags;
1649 		rt->rt_type = type;
1650 		rt->rt_is_input = 0;
1651 		rt->rt_iif = 0;
1652 		rt->rt_pmtu = 0;
1653 		rt->rt_mtu_locked = 0;
1654 		rt->rt_uses_gateway = 0;
1655 		rt->rt_gw_family = 0;
1656 		rt->rt_gw4 = 0;
1657 
1658 		rt->dst.output = ip_output;
1659 		if (flags & RTCF_LOCAL)
1660 			rt->dst.input = ip_local_deliver;
1661 	}
1662 
1663 	return rt;
1664 }
1665 EXPORT_SYMBOL(rt_dst_alloc);
1666 
1667 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1668 {
1669 	struct rtable *new_rt;
1670 
1671 	new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1672 			   rt->dst.flags);
1673 
1674 	if (new_rt) {
1675 		new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1676 		new_rt->rt_flags = rt->rt_flags;
1677 		new_rt->rt_type = rt->rt_type;
1678 		new_rt->rt_is_input = rt->rt_is_input;
1679 		new_rt->rt_iif = rt->rt_iif;
1680 		new_rt->rt_pmtu = rt->rt_pmtu;
1681 		new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1682 		new_rt->rt_gw_family = rt->rt_gw_family;
1683 		if (rt->rt_gw_family == AF_INET)
1684 			new_rt->rt_gw4 = rt->rt_gw4;
1685 		else if (rt->rt_gw_family == AF_INET6)
1686 			new_rt->rt_gw6 = rt->rt_gw6;
1687 
1688 		new_rt->dst.input = rt->dst.input;
1689 		new_rt->dst.output = rt->dst.output;
1690 		new_rt->dst.error = rt->dst.error;
1691 		new_rt->dst.lastuse = jiffies;
1692 		new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1693 	}
1694 	return new_rt;
1695 }
1696 EXPORT_SYMBOL(rt_dst_clone);
1697 
1698 /* called in rcu_read_lock() section */
1699 enum skb_drop_reason
1700 ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1701 		      dscp_t dscp, struct net_device *dev,
1702 		      struct in_device *in_dev, u32 *itag)
1703 {
1704 	enum skb_drop_reason reason;
1705 
1706 	/* Primary sanity checks. */
1707 	if (!in_dev)
1708 		return SKB_DROP_REASON_NOT_SPECIFIED;
1709 
1710 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1711 		return SKB_DROP_REASON_IP_INVALID_SOURCE;
1712 
1713 	if (skb->protocol != htons(ETH_P_IP))
1714 		return SKB_DROP_REASON_INVALID_PROTO;
1715 
1716 	if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1717 		return SKB_DROP_REASON_IP_LOCALNET;
1718 
1719 	if (ipv4_is_zeronet(saddr)) {
1720 		if (!ipv4_is_local_multicast(daddr) &&
1721 		    ip_hdr(skb)->protocol != IPPROTO_IGMP)
1722 			return SKB_DROP_REASON_IP_INVALID_SOURCE;
1723 	} else {
1724 		reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
1725 						    dev, in_dev, itag);
1726 		if (reason)
1727 			return reason;
1728 	}
1729 	return SKB_NOT_DROPPED_YET;
1730 }
1731 
1732 /* called in rcu_read_lock() section */
1733 static enum skb_drop_reason
1734 ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1735 		  dscp_t dscp, struct net_device *dev, int our)
1736 {
1737 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1738 	unsigned int flags = RTCF_MULTICAST;
1739 	enum skb_drop_reason reason;
1740 	struct rtable *rth;
1741 	u32 itag = 0;
1742 
1743 	reason = ip_mc_validate_source(skb, daddr, saddr, dscp, dev, in_dev,
1744 				       &itag);
1745 	if (reason)
1746 		return reason;
1747 
1748 	if (our)
1749 		flags |= RTCF_LOCAL;
1750 
1751 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1752 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
1753 
1754 	rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1755 			   false);
1756 	if (!rth)
1757 		return SKB_DROP_REASON_NOMEM;
1758 
1759 #ifdef CONFIG_IP_ROUTE_CLASSID
1760 	rth->dst.tclassid = itag;
1761 #endif
1762 	rth->dst.output = ip_rt_bug;
1763 	rth->rt_is_input= 1;
1764 
1765 #ifdef CONFIG_IP_MROUTE
1766 	if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1767 		rth->dst.input = ip_mr_input;
1768 #endif
1769 	RT_CACHE_STAT_INC(in_slow_mc);
1770 
1771 	skb_dst_drop(skb);
1772 	skb_dst_set(skb, &rth->dst);
1773 	return SKB_NOT_DROPPED_YET;
1774 }
1775 
1776 
1777 static void ip_handle_martian_source(struct net_device *dev,
1778 				     struct in_device *in_dev,
1779 				     struct sk_buff *skb,
1780 				     __be32 daddr,
1781 				     __be32 saddr)
1782 {
1783 	RT_CACHE_STAT_INC(in_martian_src);
1784 #ifdef CONFIG_IP_ROUTE_VERBOSE
1785 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1786 		/*
1787 		 *	RFC1812 recommendation, if source is martian,
1788 		 *	the only hint is MAC header.
1789 		 */
1790 		pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1791 			&daddr, &saddr, dev->name);
1792 		if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1793 			print_hex_dump(KERN_WARNING, "ll header: ",
1794 				       DUMP_PREFIX_OFFSET, 16, 1,
1795 				       skb_mac_header(skb),
1796 				       dev->hard_header_len, false);
1797 		}
1798 	}
1799 #endif
1800 }
1801 
1802 /* called in rcu_read_lock() section */
1803 static enum skb_drop_reason
1804 __mkroute_input(struct sk_buff *skb, const struct fib_result *res,
1805 		struct in_device *in_dev, __be32 daddr,
1806 		__be32 saddr, dscp_t dscp)
1807 {
1808 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1809 	struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1810 	struct net_device *dev = nhc->nhc_dev;
1811 	struct fib_nh_exception *fnhe;
1812 	struct rtable *rth;
1813 	int err;
1814 	struct in_device *out_dev;
1815 	bool do_cache;
1816 	u32 itag = 0;
1817 
1818 	/* get a working reference to the output device */
1819 	out_dev = __in_dev_get_rcu(dev);
1820 	if (!out_dev) {
1821 		net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1822 		return reason;
1823 	}
1824 
1825 	err = fib_validate_source(skb, saddr, daddr, dscp, FIB_RES_OIF(*res),
1826 				  in_dev->dev, in_dev, &itag);
1827 	if (err < 0) {
1828 		reason = -err;
1829 		ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1830 					 saddr);
1831 
1832 		goto cleanup;
1833 	}
1834 
1835 	do_cache = res->fi && !itag;
1836 	if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1837 	    skb->protocol == htons(ETH_P_IP)) {
1838 		__be32 gw;
1839 
1840 		gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1841 		if (IN_DEV_SHARED_MEDIA(out_dev) ||
1842 		    inet_addr_onlink(out_dev, saddr, gw))
1843 			IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1844 	}
1845 
1846 	if (skb->protocol != htons(ETH_P_IP)) {
1847 		/* Not IP (i.e. ARP). Do not create route, if it is
1848 		 * invalid for proxy arp. DNAT routes are always valid.
1849 		 *
1850 		 * Proxy arp feature have been extended to allow, ARP
1851 		 * replies back to the same interface, to support
1852 		 * Private VLAN switch technologies. See arp.c.
1853 		 */
1854 		if (out_dev == in_dev &&
1855 		    IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1856 			reason = SKB_DROP_REASON_ARP_PVLAN_DISABLE;
1857 			goto cleanup;
1858 		}
1859 	}
1860 
1861 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1862 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
1863 
1864 	fnhe = find_exception(nhc, daddr);
1865 	if (do_cache) {
1866 		if (fnhe)
1867 			rth = rcu_dereference(fnhe->fnhe_rth_input);
1868 		else
1869 			rth = rcu_dereference(nhc->nhc_rth_input);
1870 		if (rt_cache_valid(rth)) {
1871 			skb_dst_set_noref(skb, &rth->dst);
1872 			goto out;
1873 		}
1874 	}
1875 
1876 	rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1877 			   IN_DEV_ORCONF(out_dev, NOXFRM));
1878 	if (!rth) {
1879 		reason = SKB_DROP_REASON_NOMEM;
1880 		goto cleanup;
1881 	}
1882 
1883 	rth->rt_is_input = 1;
1884 	RT_CACHE_STAT_INC(in_slow_tot);
1885 
1886 	rth->dst.input = ip_forward;
1887 
1888 	rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1889 		       do_cache);
1890 	lwtunnel_set_redirect(&rth->dst);
1891 	skb_dst_set(skb, &rth->dst);
1892 out:
1893 	reason = SKB_NOT_DROPPED_YET;
1894 cleanup:
1895 	return reason;
1896 }
1897 
1898 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1899 /* To make ICMP packets follow the right flow, the multipath hash is
1900  * calculated from the inner IP addresses.
1901  */
1902 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1903 				 struct flow_keys *hash_keys)
1904 {
1905 	const struct iphdr *outer_iph = ip_hdr(skb);
1906 	const struct iphdr *key_iph = outer_iph;
1907 	const struct iphdr *inner_iph;
1908 	const struct icmphdr *icmph;
1909 	struct iphdr _inner_iph;
1910 	struct icmphdr _icmph;
1911 
1912 	if (likely(outer_iph->protocol != IPPROTO_ICMP))
1913 		goto out;
1914 
1915 	if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1916 		goto out;
1917 
1918 	icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1919 				   &_icmph);
1920 	if (!icmph)
1921 		goto out;
1922 
1923 	if (!icmp_is_err(icmph->type))
1924 		goto out;
1925 
1926 	inner_iph = skb_header_pointer(skb,
1927 				       outer_iph->ihl * 4 + sizeof(_icmph),
1928 				       sizeof(_inner_iph), &_inner_iph);
1929 	if (!inner_iph)
1930 		goto out;
1931 
1932 	key_iph = inner_iph;
1933 out:
1934 	hash_keys->addrs.v4addrs.src = key_iph->saddr;
1935 	hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1936 }
1937 
1938 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1939 					   const struct sk_buff *skb,
1940 					   bool *p_has_inner)
1941 {
1942 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1943 	struct flow_keys keys, hash_keys;
1944 
1945 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1946 		return 0;
1947 
1948 	memset(&hash_keys, 0, sizeof(hash_keys));
1949 	skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1950 
1951 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1952 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1953 		hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1954 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1955 		hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1956 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1957 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
1958 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1959 		hash_keys.ports.src = keys.ports.src;
1960 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1961 		hash_keys.ports.dst = keys.ports.dst;
1962 
1963 	*p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1964 	return fib_multipath_hash_from_keys(net, &hash_keys);
1965 }
1966 
1967 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1968 					   const struct sk_buff *skb,
1969 					   bool has_inner)
1970 {
1971 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1972 	struct flow_keys keys, hash_keys;
1973 
1974 	/* We assume the packet carries an encapsulation, but if none was
1975 	 * encountered during dissection of the outer flow, then there is no
1976 	 * point in calling the flow dissector again.
1977 	 */
1978 	if (!has_inner)
1979 		return 0;
1980 
1981 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1982 		return 0;
1983 
1984 	memset(&hash_keys, 0, sizeof(hash_keys));
1985 	skb_flow_dissect_flow_keys(skb, &keys, 0);
1986 
1987 	if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1988 		return 0;
1989 
1990 	if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1991 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1992 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1993 			hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1994 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1995 			hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1996 	} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1997 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1998 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1999 			hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2000 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2001 			hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2002 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2003 			hash_keys.tags.flow_label = keys.tags.flow_label;
2004 	}
2005 
2006 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2007 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
2008 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2009 		hash_keys.ports.src = keys.ports.src;
2010 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2011 		hash_keys.ports.dst = keys.ports.dst;
2012 
2013 	return fib_multipath_hash_from_keys(net, &hash_keys);
2014 }
2015 
2016 static u32 fib_multipath_custom_hash_skb(const struct net *net,
2017 					 const struct sk_buff *skb)
2018 {
2019 	u32 mhash, mhash_inner;
2020 	bool has_inner = true;
2021 
2022 	mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2023 	mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2024 
2025 	return jhash_2words(mhash, mhash_inner, 0);
2026 }
2027 
2028 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2029 					 const struct flowi4 *fl4)
2030 {
2031 	u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2032 	struct flow_keys hash_keys;
2033 
2034 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2035 		return 0;
2036 
2037 	memset(&hash_keys, 0, sizeof(hash_keys));
2038 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2039 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2040 		hash_keys.addrs.v4addrs.src = fl4->saddr;
2041 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2042 		hash_keys.addrs.v4addrs.dst = fl4->daddr;
2043 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2044 		hash_keys.basic.ip_proto = fl4->flowi4_proto;
2045 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) {
2046 		if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT)
2047 			hash_keys.ports.src = (__force __be16)get_random_u16();
2048 		else
2049 			hash_keys.ports.src = fl4->fl4_sport;
2050 	}
2051 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2052 		hash_keys.ports.dst = fl4->fl4_dport;
2053 
2054 	return fib_multipath_hash_from_keys(net, &hash_keys);
2055 }
2056 
2057 /* if skb is set it will be used and fl4 can be NULL */
2058 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2059 		       const struct sk_buff *skb, struct flow_keys *flkeys)
2060 {
2061 	u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2062 	struct flow_keys hash_keys;
2063 	u32 mhash = 0;
2064 
2065 	switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2066 	case 0:
2067 		memset(&hash_keys, 0, sizeof(hash_keys));
2068 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2069 		if (skb) {
2070 			ip_multipath_l3_keys(skb, &hash_keys);
2071 		} else {
2072 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2073 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2074 		}
2075 		mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2076 		break;
2077 	case 1:
2078 		/* skb is currently provided only when forwarding */
2079 		if (skb) {
2080 			unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2081 			struct flow_keys keys;
2082 
2083 			/* short-circuit if we already have L4 hash present */
2084 			if (skb->l4_hash)
2085 				return skb_get_hash_raw(skb) >> 1;
2086 
2087 			memset(&hash_keys, 0, sizeof(hash_keys));
2088 
2089 			if (!flkeys) {
2090 				skb_flow_dissect_flow_keys(skb, &keys, flag);
2091 				flkeys = &keys;
2092 			}
2093 
2094 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2095 			hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2096 			hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2097 			hash_keys.ports.src = flkeys->ports.src;
2098 			hash_keys.ports.dst = flkeys->ports.dst;
2099 			hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2100 		} else {
2101 			memset(&hash_keys, 0, sizeof(hash_keys));
2102 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2103 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2104 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2105 			if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT)
2106 				hash_keys.ports.src = (__force __be16)get_random_u16();
2107 			else
2108 				hash_keys.ports.src = fl4->fl4_sport;
2109 			hash_keys.ports.dst = fl4->fl4_dport;
2110 			hash_keys.basic.ip_proto = fl4->flowi4_proto;
2111 		}
2112 		mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2113 		break;
2114 	case 2:
2115 		memset(&hash_keys, 0, sizeof(hash_keys));
2116 		/* skb is currently provided only when forwarding */
2117 		if (skb) {
2118 			struct flow_keys keys;
2119 
2120 			skb_flow_dissect_flow_keys(skb, &keys, 0);
2121 			/* Inner can be v4 or v6 */
2122 			if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2123 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2124 				hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2125 				hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2126 			} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2127 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2128 				hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2129 				hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2130 				hash_keys.tags.flow_label = keys.tags.flow_label;
2131 				hash_keys.basic.ip_proto = keys.basic.ip_proto;
2132 			} else {
2133 				/* Same as case 0 */
2134 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2135 				ip_multipath_l3_keys(skb, &hash_keys);
2136 			}
2137 		} else {
2138 			/* Same as case 0 */
2139 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2140 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2141 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2142 		}
2143 		mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2144 		break;
2145 	case 3:
2146 		if (skb)
2147 			mhash = fib_multipath_custom_hash_skb(net, skb);
2148 		else
2149 			mhash = fib_multipath_custom_hash_fl4(net, fl4);
2150 		break;
2151 	}
2152 
2153 	if (multipath_hash)
2154 		mhash = jhash_2words(mhash, multipath_hash, 0);
2155 
2156 	return mhash >> 1;
2157 }
2158 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2159 
2160 static enum skb_drop_reason
2161 ip_mkroute_input(struct sk_buff *skb, struct fib_result *res,
2162 		 struct in_device *in_dev, __be32 daddr,
2163 		 __be32 saddr, dscp_t dscp, struct flow_keys *hkeys)
2164 {
2165 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2166 	if (res->fi && fib_info_num_path(res->fi) > 1) {
2167 		int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2168 
2169 		fib_select_multipath(res, h, NULL);
2170 		IPCB(skb)->flags |= IPSKB_MULTIPATH;
2171 	}
2172 #endif
2173 
2174 	/* create a routing cache entry */
2175 	return __mkroute_input(skb, res, in_dev, daddr, saddr, dscp);
2176 }
2177 
2178 /* Implements all the saddr-related checks as ip_route_input_slow(),
2179  * assuming daddr is valid and the destination is not a local broadcast one.
2180  * Uses the provided hint instead of performing a route lookup.
2181  */
2182 enum skb_drop_reason
2183 ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2184 		  dscp_t dscp, struct net_device *dev,
2185 		  const struct sk_buff *hint)
2186 {
2187 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2188 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2189 	struct rtable *rt = skb_rtable(hint);
2190 	struct net *net = dev_net(dev);
2191 	u32 tag = 0;
2192 
2193 	if (!in_dev)
2194 		return reason;
2195 
2196 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2197 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2198 		goto martian_source;
2199 	}
2200 
2201 	if (ipv4_is_zeronet(saddr)) {
2202 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2203 		goto martian_source;
2204 	}
2205 
2206 	if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2207 		reason = SKB_DROP_REASON_IP_LOCALNET;
2208 		goto martian_source;
2209 	}
2210 
2211 	if (rt->rt_type != RTN_LOCAL)
2212 		goto skip_validate_source;
2213 
2214 	reason = fib_validate_source_reason(skb, saddr, daddr, dscp, 0, dev,
2215 					    in_dev, &tag);
2216 	if (reason)
2217 		goto martian_source;
2218 
2219 skip_validate_source:
2220 	skb_dst_copy(skb, hint);
2221 	return SKB_NOT_DROPPED_YET;
2222 
2223 martian_source:
2224 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2225 	return reason;
2226 }
2227 
2228 /* get device for dst_alloc with local routes */
2229 static struct net_device *ip_rt_get_dev(struct net *net,
2230 					const struct fib_result *res)
2231 {
2232 	struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2233 	struct net_device *dev = NULL;
2234 
2235 	if (nhc)
2236 		dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2237 
2238 	return dev ? : net->loopback_dev;
2239 }
2240 
2241 /*
2242  *	NOTE. We drop all the packets that has local source
2243  *	addresses, because every properly looped back packet
2244  *	must have correct destination already attached by output routine.
2245  *	Changes in the enforced policies must be applied also to
2246  *	ip_route_use_hint().
2247  *
2248  *	Such approach solves two big problems:
2249  *	1. Not simplex devices are handled properly.
2250  *	2. IP spoofing attempts are filtered with 100% of guarantee.
2251  *	called with rcu_read_lock()
2252  */
2253 
2254 static enum skb_drop_reason
2255 ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2256 		    dscp_t dscp, struct net_device *dev,
2257 		    struct fib_result *res)
2258 {
2259 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2260 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2261 	struct flow_keys *flkeys = NULL, _flkeys;
2262 	struct net    *net = dev_net(dev);
2263 	struct ip_tunnel_info *tun_info;
2264 	int		err = -EINVAL;
2265 	unsigned int	flags = 0;
2266 	u32		itag = 0;
2267 	struct rtable	*rth;
2268 	struct flowi4	fl4;
2269 	bool do_cache = true;
2270 
2271 	/* IP on this device is disabled. */
2272 
2273 	if (!in_dev)
2274 		goto out;
2275 
2276 	/* Check for the most weird martians, which can be not detected
2277 	 * by fib_lookup.
2278 	 */
2279 
2280 	tun_info = skb_tunnel_info(skb);
2281 	if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2282 		fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2283 	else
2284 		fl4.flowi4_tun_key.tun_id = 0;
2285 	skb_dst_drop(skb);
2286 
2287 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2288 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2289 		goto martian_source;
2290 	}
2291 
2292 	res->fi = NULL;
2293 	res->table = NULL;
2294 	if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2295 		goto brd_input;
2296 
2297 	/* Accept zero addresses only to limited broadcast;
2298 	 * I even do not know to fix it or not. Waiting for complains :-)
2299 	 */
2300 	if (ipv4_is_zeronet(saddr)) {
2301 		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2302 		goto martian_source;
2303 	}
2304 
2305 	if (ipv4_is_zeronet(daddr)) {
2306 		reason = SKB_DROP_REASON_IP_INVALID_DEST;
2307 		goto martian_destination;
2308 	}
2309 
2310 	/* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2311 	 * and call it once if daddr or/and saddr are loopback addresses
2312 	 */
2313 	if (ipv4_is_loopback(daddr)) {
2314 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2315 			reason = SKB_DROP_REASON_IP_LOCALNET;
2316 			goto martian_destination;
2317 		}
2318 	} else if (ipv4_is_loopback(saddr)) {
2319 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2320 			reason = SKB_DROP_REASON_IP_LOCALNET;
2321 			goto martian_source;
2322 		}
2323 	}
2324 
2325 	/*
2326 	 *	Now we are ready to route packet.
2327 	 */
2328 	fl4.flowi4_l3mdev = 0;
2329 	fl4.flowi4_oif = 0;
2330 	fl4.flowi4_iif = dev->ifindex;
2331 	fl4.flowi4_mark = skb->mark;
2332 	fl4.flowi4_tos = inet_dscp_to_dsfield(dscp);
2333 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2334 	fl4.flowi4_flags = 0;
2335 	fl4.daddr = daddr;
2336 	fl4.saddr = saddr;
2337 	fl4.flowi4_uid = sock_net_uid(net, NULL);
2338 	fl4.flowi4_multipath_hash = 0;
2339 
2340 	if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2341 		flkeys = &_flkeys;
2342 	} else {
2343 		fl4.flowi4_proto = 0;
2344 		fl4.fl4_sport = 0;
2345 		fl4.fl4_dport = 0;
2346 	}
2347 
2348 	err = fib_lookup(net, &fl4, res, 0);
2349 	if (err != 0) {
2350 		if (!IN_DEV_FORWARD(in_dev))
2351 			err = -EHOSTUNREACH;
2352 		goto no_route;
2353 	}
2354 
2355 	if (res->type == RTN_BROADCAST) {
2356 		if (IN_DEV_BFORWARD(in_dev))
2357 			goto make_route;
2358 		/* not do cache if bc_forwarding is enabled */
2359 		if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING))
2360 			do_cache = false;
2361 		goto brd_input;
2362 	}
2363 
2364 	err = -EINVAL;
2365 	if (res->type == RTN_LOCAL) {
2366 		reason = fib_validate_source_reason(skb, saddr, daddr, dscp,
2367 						    0, dev, in_dev, &itag);
2368 		if (reason)
2369 			goto martian_source;
2370 		goto local_input;
2371 	}
2372 
2373 	if (!IN_DEV_FORWARD(in_dev)) {
2374 		err = -EHOSTUNREACH;
2375 		goto no_route;
2376 	}
2377 	if (res->type != RTN_UNICAST) {
2378 		reason = SKB_DROP_REASON_IP_INVALID_DEST;
2379 		goto martian_destination;
2380 	}
2381 
2382 make_route:
2383 	reason = ip_mkroute_input(skb, res, in_dev, daddr, saddr, dscp,
2384 				  flkeys);
2385 
2386 out:
2387 	return reason;
2388 
2389 brd_input:
2390 	if (skb->protocol != htons(ETH_P_IP)) {
2391 		reason = SKB_DROP_REASON_INVALID_PROTO;
2392 		goto out;
2393 	}
2394 
2395 	if (!ipv4_is_zeronet(saddr)) {
2396 		reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
2397 						    dev, in_dev, &itag);
2398 		if (reason)
2399 			goto martian_source;
2400 	}
2401 	flags |= RTCF_BROADCAST;
2402 	res->type = RTN_BROADCAST;
2403 	RT_CACHE_STAT_INC(in_brd);
2404 
2405 local_input:
2406 	if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2407 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
2408 
2409 	do_cache &= res->fi && !itag;
2410 	if (do_cache) {
2411 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2412 
2413 		rth = rcu_dereference(nhc->nhc_rth_input);
2414 		if (rt_cache_valid(rth)) {
2415 			skb_dst_set_noref(skb, &rth->dst);
2416 			reason = SKB_NOT_DROPPED_YET;
2417 			goto out;
2418 		}
2419 	}
2420 
2421 	rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2422 			   flags | RTCF_LOCAL, res->type, false);
2423 	if (!rth)
2424 		goto e_nobufs;
2425 
2426 	rth->dst.output= ip_rt_bug;
2427 #ifdef CONFIG_IP_ROUTE_CLASSID
2428 	rth->dst.tclassid = itag;
2429 #endif
2430 	rth->rt_is_input = 1;
2431 
2432 	RT_CACHE_STAT_INC(in_slow_tot);
2433 	if (res->type == RTN_UNREACHABLE) {
2434 		rth->dst.input= ip_error;
2435 		rth->dst.error= -err;
2436 		rth->rt_flags	&= ~RTCF_LOCAL;
2437 	}
2438 
2439 	if (do_cache) {
2440 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2441 
2442 		rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2443 		if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2444 			WARN_ON(rth->dst.input == lwtunnel_input);
2445 			rth->dst.lwtstate->orig_input = rth->dst.input;
2446 			rth->dst.input = lwtunnel_input;
2447 		}
2448 
2449 		if (unlikely(!rt_cache_route(nhc, rth)))
2450 			rt_add_uncached_list(rth);
2451 	}
2452 	skb_dst_set(skb, &rth->dst);
2453 	reason = SKB_NOT_DROPPED_YET;
2454 	goto out;
2455 
2456 no_route:
2457 	RT_CACHE_STAT_INC(in_no_route);
2458 	res->type = RTN_UNREACHABLE;
2459 	res->fi = NULL;
2460 	res->table = NULL;
2461 	goto local_input;
2462 
2463 	/*
2464 	 *	Do not cache martian addresses: they should be logged (RFC1812)
2465 	 */
2466 martian_destination:
2467 	RT_CACHE_STAT_INC(in_martian_dst);
2468 #ifdef CONFIG_IP_ROUTE_VERBOSE
2469 	if (IN_DEV_LOG_MARTIANS(in_dev))
2470 		net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2471 				     &daddr, &saddr, dev->name);
2472 #endif
2473 	goto out;
2474 
2475 e_nobufs:
2476 	reason = SKB_DROP_REASON_NOMEM;
2477 	goto out;
2478 
2479 martian_source:
2480 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2481 	goto out;
2482 }
2483 
2484 /* called with rcu_read_lock held */
2485 static enum skb_drop_reason
2486 ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2487 		   dscp_t dscp, struct net_device *dev,
2488 		   struct fib_result *res)
2489 {
2490 	/* Multicast recognition logic is moved from route cache to here.
2491 	 * The problem was that too many Ethernet cards have broken/missing
2492 	 * hardware multicast filters :-( As result the host on multicasting
2493 	 * network acquires a lot of useless route cache entries, sort of
2494 	 * SDR messages from all the world. Now we try to get rid of them.
2495 	 * Really, provided software IP multicast filter is organized
2496 	 * reasonably (at least, hashed), it does not result in a slowdown
2497 	 * comparing with route cache reject entries.
2498 	 * Note, that multicast routers are not affected, because
2499 	 * route cache entry is created eventually.
2500 	 */
2501 	if (ipv4_is_multicast(daddr)) {
2502 		enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2503 		struct in_device *in_dev = __in_dev_get_rcu(dev);
2504 		int our = 0;
2505 
2506 		if (!in_dev)
2507 			return reason;
2508 
2509 		our = ip_check_mc_rcu(in_dev, daddr, saddr,
2510 				      ip_hdr(skb)->protocol);
2511 
2512 		/* check l3 master if no match yet */
2513 		if (!our && netif_is_l3_slave(dev)) {
2514 			struct in_device *l3_in_dev;
2515 
2516 			l3_in_dev = __in_dev_get_rcu(skb->dev);
2517 			if (l3_in_dev)
2518 				our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2519 						      ip_hdr(skb)->protocol);
2520 		}
2521 
2522 		if (our
2523 #ifdef CONFIG_IP_MROUTE
2524 			||
2525 		    (!ipv4_is_local_multicast(daddr) &&
2526 		     IN_DEV_MFORWARD(in_dev))
2527 #endif
2528 		   ) {
2529 			reason = ip_route_input_mc(skb, daddr, saddr, dscp,
2530 						   dev, our);
2531 		}
2532 		return reason;
2533 	}
2534 
2535 	return ip_route_input_slow(skb, daddr, saddr, dscp, dev, res);
2536 }
2537 
2538 enum skb_drop_reason ip_route_input_noref(struct sk_buff *skb, __be32 daddr,
2539 					  __be32 saddr, dscp_t dscp,
2540 					  struct net_device *dev)
2541 {
2542 	enum skb_drop_reason reason;
2543 	struct fib_result res;
2544 
2545 	rcu_read_lock();
2546 	reason = ip_route_input_rcu(skb, daddr, saddr, dscp, dev, &res);
2547 	rcu_read_unlock();
2548 
2549 	return reason;
2550 }
2551 EXPORT_SYMBOL(ip_route_input_noref);
2552 
2553 /* called with rcu_read_lock() */
2554 static struct rtable *__mkroute_output(const struct fib_result *res,
2555 				       const struct flowi4 *fl4, int orig_oif,
2556 				       struct net_device *dev_out,
2557 				       unsigned int flags)
2558 {
2559 	struct fib_info *fi = res->fi;
2560 	struct fib_nh_exception *fnhe;
2561 	struct in_device *in_dev;
2562 	u16 type = res->type;
2563 	struct rtable *rth;
2564 	bool do_cache;
2565 
2566 	in_dev = __in_dev_get_rcu(dev_out);
2567 	if (!in_dev)
2568 		return ERR_PTR(-EINVAL);
2569 
2570 	if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2571 		if (ipv4_is_loopback(fl4->saddr) &&
2572 		    !(dev_out->flags & IFF_LOOPBACK) &&
2573 		    !netif_is_l3_master(dev_out))
2574 			return ERR_PTR(-EINVAL);
2575 
2576 	if (ipv4_is_lbcast(fl4->daddr))
2577 		type = RTN_BROADCAST;
2578 	else if (ipv4_is_multicast(fl4->daddr))
2579 		type = RTN_MULTICAST;
2580 	else if (ipv4_is_zeronet(fl4->daddr))
2581 		return ERR_PTR(-EINVAL);
2582 
2583 	if (dev_out->flags & IFF_LOOPBACK)
2584 		flags |= RTCF_LOCAL;
2585 
2586 	do_cache = true;
2587 	if (type == RTN_BROADCAST) {
2588 		flags |= RTCF_BROADCAST | RTCF_LOCAL;
2589 		fi = NULL;
2590 	} else if (type == RTN_MULTICAST) {
2591 		flags |= RTCF_MULTICAST | RTCF_LOCAL;
2592 		if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2593 				     fl4->flowi4_proto))
2594 			flags &= ~RTCF_LOCAL;
2595 		else
2596 			do_cache = false;
2597 		/* If multicast route do not exist use
2598 		 * default one, but do not gateway in this case.
2599 		 * Yes, it is hack.
2600 		 */
2601 		if (fi && res->prefixlen < 4)
2602 			fi = NULL;
2603 	} else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2604 		   (orig_oif != dev_out->ifindex)) {
2605 		/* For local routes that require a particular output interface
2606 		 * we do not want to cache the result.  Caching the result
2607 		 * causes incorrect behaviour when there are multiple source
2608 		 * addresses on the interface, the end result being that if the
2609 		 * intended recipient is waiting on that interface for the
2610 		 * packet he won't receive it because it will be delivered on
2611 		 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2612 		 * be set to the loopback interface as well.
2613 		 */
2614 		do_cache = false;
2615 	}
2616 
2617 	fnhe = NULL;
2618 	do_cache &= fi != NULL;
2619 	if (fi) {
2620 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2621 		struct rtable __rcu **prth;
2622 
2623 		fnhe = find_exception(nhc, fl4->daddr);
2624 		if (!do_cache)
2625 			goto add;
2626 		if (fnhe) {
2627 			prth = &fnhe->fnhe_rth_output;
2628 		} else {
2629 			if (unlikely(fl4->flowi4_flags &
2630 				     FLOWI_FLAG_KNOWN_NH &&
2631 				     !(nhc->nhc_gw_family &&
2632 				       nhc->nhc_scope == RT_SCOPE_LINK))) {
2633 				do_cache = false;
2634 				goto add;
2635 			}
2636 			prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2637 		}
2638 		rth = rcu_dereference(*prth);
2639 		if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2640 			return rth;
2641 	}
2642 
2643 add:
2644 	rth = rt_dst_alloc(dev_out, flags, type,
2645 			   IN_DEV_ORCONF(in_dev, NOXFRM));
2646 	if (!rth)
2647 		return ERR_PTR(-ENOBUFS);
2648 
2649 	rth->rt_iif = orig_oif;
2650 
2651 	RT_CACHE_STAT_INC(out_slow_tot);
2652 
2653 	if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2654 		if (flags & RTCF_LOCAL &&
2655 		    !(dev_out->flags & IFF_LOOPBACK)) {
2656 			rth->dst.output = ip_mc_output;
2657 			RT_CACHE_STAT_INC(out_slow_mc);
2658 		}
2659 #ifdef CONFIG_IP_MROUTE
2660 		if (type == RTN_MULTICAST) {
2661 			if (IN_DEV_MFORWARD(in_dev) &&
2662 			    !ipv4_is_local_multicast(fl4->daddr)) {
2663 				rth->dst.input = ip_mr_input;
2664 				rth->dst.output = ip_mr_output;
2665 			}
2666 		}
2667 #endif
2668 	}
2669 
2670 	rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2671 	lwtunnel_set_redirect(&rth->dst);
2672 
2673 	return rth;
2674 }
2675 
2676 /*
2677  * Major route resolver routine.
2678  */
2679 
2680 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2681 					const struct sk_buff *skb)
2682 {
2683 	struct fib_result res = {
2684 		.type		= RTN_UNSPEC,
2685 		.fi		= NULL,
2686 		.table		= NULL,
2687 		.tclassid	= 0,
2688 	};
2689 	struct rtable *rth;
2690 
2691 	fl4->flowi4_iif = LOOPBACK_IFINDEX;
2692 	fl4->flowi4_tos &= INET_DSCP_MASK;
2693 
2694 	rcu_read_lock();
2695 	rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2696 	rcu_read_unlock();
2697 
2698 	return rth;
2699 }
2700 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2701 
2702 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2703 					    struct fib_result *res,
2704 					    const struct sk_buff *skb)
2705 {
2706 	struct net_device *dev_out = NULL;
2707 	int orig_oif = fl4->flowi4_oif;
2708 	unsigned int flags = 0;
2709 	struct rtable *rth;
2710 	int err;
2711 
2712 	if (fl4->saddr) {
2713 		if (ipv4_is_multicast(fl4->saddr) ||
2714 		    ipv4_is_lbcast(fl4->saddr)) {
2715 			rth = ERR_PTR(-EINVAL);
2716 			goto out;
2717 		}
2718 
2719 		rth = ERR_PTR(-ENETUNREACH);
2720 
2721 		/* I removed check for oif == dev_out->oif here.
2722 		 * It was wrong for two reasons:
2723 		 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2724 		 *    is assigned to multiple interfaces.
2725 		 * 2. Moreover, we are allowed to send packets with saddr
2726 		 *    of another iface. --ANK
2727 		 */
2728 
2729 		if (fl4->flowi4_oif == 0 &&
2730 		    (ipv4_is_multicast(fl4->daddr) ||
2731 		     ipv4_is_lbcast(fl4->daddr))) {
2732 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2733 			dev_out = __ip_dev_find(net, fl4->saddr, false);
2734 			if (!dev_out)
2735 				goto out;
2736 
2737 			/* Special hack: user can direct multicasts
2738 			 * and limited broadcast via necessary interface
2739 			 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2740 			 * This hack is not just for fun, it allows
2741 			 * vic,vat and friends to work.
2742 			 * They bind socket to loopback, set ttl to zero
2743 			 * and expect that it will work.
2744 			 * From the viewpoint of routing cache they are broken,
2745 			 * because we are not allowed to build multicast path
2746 			 * with loopback source addr (look, routing cache
2747 			 * cannot know, that ttl is zero, so that packet
2748 			 * will not leave this host and route is valid).
2749 			 * Luckily, this hack is good workaround.
2750 			 */
2751 
2752 			fl4->flowi4_oif = dev_out->ifindex;
2753 			goto make_route;
2754 		}
2755 
2756 		if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2757 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2758 			if (!__ip_dev_find(net, fl4->saddr, false))
2759 				goto out;
2760 		}
2761 	}
2762 
2763 
2764 	if (fl4->flowi4_oif) {
2765 		dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2766 		rth = ERR_PTR(-ENODEV);
2767 		if (!dev_out)
2768 			goto out;
2769 
2770 		/* RACE: Check return value of inet_select_addr instead. */
2771 		if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2772 			rth = ERR_PTR(-ENETUNREACH);
2773 			goto out;
2774 		}
2775 		if (ipv4_is_local_multicast(fl4->daddr) ||
2776 		    ipv4_is_lbcast(fl4->daddr) ||
2777 		    fl4->flowi4_proto == IPPROTO_IGMP) {
2778 			if (!fl4->saddr)
2779 				fl4->saddr = inet_select_addr(dev_out, 0,
2780 							      RT_SCOPE_LINK);
2781 			goto make_route;
2782 		}
2783 		if (!fl4->saddr) {
2784 			if (ipv4_is_multicast(fl4->daddr))
2785 				fl4->saddr = inet_select_addr(dev_out, 0,
2786 							      fl4->flowi4_scope);
2787 			else if (!fl4->daddr)
2788 				fl4->saddr = inet_select_addr(dev_out, 0,
2789 							      RT_SCOPE_HOST);
2790 		}
2791 	}
2792 
2793 	if (!fl4->daddr) {
2794 		fl4->daddr = fl4->saddr;
2795 		if (!fl4->daddr)
2796 			fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2797 		dev_out = net->loopback_dev;
2798 		fl4->flowi4_oif = LOOPBACK_IFINDEX;
2799 		res->type = RTN_LOCAL;
2800 		flags |= RTCF_LOCAL;
2801 		goto make_route;
2802 	}
2803 
2804 	err = fib_lookup(net, fl4, res, 0);
2805 	if (err) {
2806 		res->fi = NULL;
2807 		res->table = NULL;
2808 		if (fl4->flowi4_oif &&
2809 		    (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2810 			/* Apparently, routing tables are wrong. Assume,
2811 			 * that the destination is on link.
2812 			 *
2813 			 * WHY? DW.
2814 			 * Because we are allowed to send to iface
2815 			 * even if it has NO routes and NO assigned
2816 			 * addresses. When oif is specified, routing
2817 			 * tables are looked up with only one purpose:
2818 			 * to catch if destination is gatewayed, rather than
2819 			 * direct. Moreover, if MSG_DONTROUTE is set,
2820 			 * we send packet, ignoring both routing tables
2821 			 * and ifaddr state. --ANK
2822 			 *
2823 			 *
2824 			 * We could make it even if oif is unknown,
2825 			 * likely IPv6, but we do not.
2826 			 */
2827 
2828 			if (fl4->saddr == 0)
2829 				fl4->saddr = inet_select_addr(dev_out, 0,
2830 							      RT_SCOPE_LINK);
2831 			res->type = RTN_UNICAST;
2832 			goto make_route;
2833 		}
2834 		rth = ERR_PTR(err);
2835 		goto out;
2836 	}
2837 
2838 	if (res->type == RTN_LOCAL) {
2839 		if (!fl4->saddr) {
2840 			if (res->fi->fib_prefsrc)
2841 				fl4->saddr = res->fi->fib_prefsrc;
2842 			else
2843 				fl4->saddr = fl4->daddr;
2844 		}
2845 
2846 		/* L3 master device is the loopback for that domain */
2847 		dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2848 			net->loopback_dev;
2849 
2850 		/* make sure orig_oif points to fib result device even
2851 		 * though packet rx/tx happens over loopback or l3mdev
2852 		 */
2853 		orig_oif = FIB_RES_OIF(*res);
2854 
2855 		fl4->flowi4_oif = dev_out->ifindex;
2856 		flags |= RTCF_LOCAL;
2857 		goto make_route;
2858 	}
2859 
2860 	fib_select_path(net, res, fl4, skb);
2861 
2862 	dev_out = FIB_RES_DEV(*res);
2863 
2864 make_route:
2865 	rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2866 
2867 out:
2868 	return rth;
2869 }
2870 
2871 static struct dst_ops ipv4_dst_blackhole_ops = {
2872 	.family			= AF_INET,
2873 	.default_advmss		= ipv4_default_advmss,
2874 	.neigh_lookup		= ipv4_neigh_lookup,
2875 	.check			= dst_blackhole_check,
2876 	.cow_metrics		= dst_blackhole_cow_metrics,
2877 	.update_pmtu		= dst_blackhole_update_pmtu,
2878 	.redirect		= dst_blackhole_redirect,
2879 	.mtu			= dst_blackhole_mtu,
2880 };
2881 
2882 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2883 {
2884 	struct rtable *ort = dst_rtable(dst_orig);
2885 	struct rtable *rt;
2886 
2887 	rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0);
2888 	if (rt) {
2889 		struct dst_entry *new = &rt->dst;
2890 
2891 		new->__use = 1;
2892 		new->input = dst_discard;
2893 		new->output = dst_discard_out;
2894 
2895 		new->dev = net->loopback_dev;
2896 		netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2897 
2898 		rt->rt_is_input = ort->rt_is_input;
2899 		rt->rt_iif = ort->rt_iif;
2900 		rt->rt_pmtu = ort->rt_pmtu;
2901 		rt->rt_mtu_locked = ort->rt_mtu_locked;
2902 
2903 		rt->rt_genid = rt_genid_ipv4(net);
2904 		rt->rt_flags = ort->rt_flags;
2905 		rt->rt_type = ort->rt_type;
2906 		rt->rt_uses_gateway = ort->rt_uses_gateway;
2907 		rt->rt_gw_family = ort->rt_gw_family;
2908 		if (rt->rt_gw_family == AF_INET)
2909 			rt->rt_gw4 = ort->rt_gw4;
2910 		else if (rt->rt_gw_family == AF_INET6)
2911 			rt->rt_gw6 = ort->rt_gw6;
2912 	}
2913 
2914 	dst_release(dst_orig);
2915 
2916 	return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2917 }
2918 
2919 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2920 				    const struct sock *sk)
2921 {
2922 	struct rtable *rt = __ip_route_output_key(net, flp4);
2923 
2924 	if (IS_ERR(rt))
2925 		return rt;
2926 
2927 	if (flp4->flowi4_proto) {
2928 		flp4->flowi4_oif = rt->dst.dev->ifindex;
2929 		rt = dst_rtable(xfrm_lookup_route(net, &rt->dst,
2930 						  flowi4_to_flowi(flp4),
2931 						  sk, 0));
2932 	}
2933 
2934 	return rt;
2935 }
2936 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2937 
2938 /* called with rcu_read_lock held */
2939 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2940 			struct rtable *rt, u32 table_id, dscp_t dscp,
2941 			struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2942 			u32 seq, unsigned int flags)
2943 {
2944 	struct rtmsg *r;
2945 	struct nlmsghdr *nlh;
2946 	unsigned long expires = 0;
2947 	u32 error;
2948 	u32 metrics[RTAX_MAX];
2949 
2950 	nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2951 	if (!nlh)
2952 		return -EMSGSIZE;
2953 
2954 	r = nlmsg_data(nlh);
2955 	r->rtm_family	 = AF_INET;
2956 	r->rtm_dst_len	= 32;
2957 	r->rtm_src_len	= 0;
2958 	r->rtm_tos	= inet_dscp_to_dsfield(dscp);
2959 	r->rtm_table	= table_id < 256 ? table_id : RT_TABLE_COMPAT;
2960 	if (nla_put_u32(skb, RTA_TABLE, table_id))
2961 		goto nla_put_failure;
2962 	r->rtm_type	= rt->rt_type;
2963 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2964 	r->rtm_protocol = RTPROT_UNSPEC;
2965 	r->rtm_flags	= (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2966 	if (rt->rt_flags & RTCF_NOTIFY)
2967 		r->rtm_flags |= RTM_F_NOTIFY;
2968 	if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2969 		r->rtm_flags |= RTCF_DOREDIRECT;
2970 
2971 	if (nla_put_in_addr(skb, RTA_DST, dst))
2972 		goto nla_put_failure;
2973 	if (src) {
2974 		r->rtm_src_len = 32;
2975 		if (nla_put_in_addr(skb, RTA_SRC, src))
2976 			goto nla_put_failure;
2977 	}
2978 	if (rt->dst.dev &&
2979 	    nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2980 		goto nla_put_failure;
2981 	if (lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2982 		goto nla_put_failure;
2983 #ifdef CONFIG_IP_ROUTE_CLASSID
2984 	if (rt->dst.tclassid &&
2985 	    nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2986 		goto nla_put_failure;
2987 #endif
2988 	if (fl4 && !rt_is_input_route(rt) &&
2989 	    fl4->saddr != src) {
2990 		if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2991 			goto nla_put_failure;
2992 	}
2993 	if (rt->rt_uses_gateway) {
2994 		if (rt->rt_gw_family == AF_INET &&
2995 		    nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2996 			goto nla_put_failure;
2997 		} else if (rt->rt_gw_family == AF_INET6) {
2998 			int alen = sizeof(struct in6_addr);
2999 			struct nlattr *nla;
3000 			struct rtvia *via;
3001 
3002 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
3003 			if (!nla)
3004 				goto nla_put_failure;
3005 
3006 			via = nla_data(nla);
3007 			via->rtvia_family = AF_INET6;
3008 			memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
3009 		}
3010 	}
3011 
3012 	expires = rt->dst.expires;
3013 	if (expires) {
3014 		unsigned long now = jiffies;
3015 
3016 		if (time_before(now, expires))
3017 			expires -= now;
3018 		else
3019 			expires = 0;
3020 	}
3021 
3022 	memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3023 	if (rt->rt_pmtu && expires)
3024 		metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3025 	if (rt->rt_mtu_locked && expires)
3026 		metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3027 	if (rtnetlink_put_metrics(skb, metrics) < 0)
3028 		goto nla_put_failure;
3029 
3030 	if (fl4) {
3031 		if (fl4->flowi4_mark &&
3032 		    nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3033 			goto nla_put_failure;
3034 
3035 		if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3036 		    nla_put_u32(skb, RTA_UID,
3037 				from_kuid_munged(current_user_ns(),
3038 						 fl4->flowi4_uid)))
3039 			goto nla_put_failure;
3040 
3041 		if (rt_is_input_route(rt)) {
3042 #ifdef CONFIG_IP_MROUTE
3043 			if (ipv4_is_multicast(dst) &&
3044 			    !ipv4_is_local_multicast(dst) &&
3045 			    IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) {
3046 				int err = ipmr_get_route(net, skb,
3047 							 fl4->saddr, fl4->daddr,
3048 							 r, portid);
3049 
3050 				if (err <= 0) {
3051 					if (err == 0)
3052 						return 0;
3053 					goto nla_put_failure;
3054 				}
3055 			} else
3056 #endif
3057 				if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3058 					goto nla_put_failure;
3059 		}
3060 	}
3061 
3062 	error = rt->dst.error;
3063 
3064 	if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3065 		goto nla_put_failure;
3066 
3067 	nlmsg_end(skb, nlh);
3068 	return 0;
3069 
3070 nla_put_failure:
3071 	nlmsg_cancel(skb, nlh);
3072 	return -EMSGSIZE;
3073 }
3074 
3075 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3076 			    struct netlink_callback *cb, u32 table_id,
3077 			    struct fnhe_hash_bucket *bucket, int genid,
3078 			    int *fa_index, int fa_start, unsigned int flags)
3079 {
3080 	int i;
3081 
3082 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
3083 		struct fib_nh_exception *fnhe;
3084 
3085 		for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3086 		     fnhe = rcu_dereference(fnhe->fnhe_next)) {
3087 			struct rtable *rt;
3088 			int err;
3089 
3090 			if (*fa_index < fa_start)
3091 				goto next;
3092 
3093 			if (fnhe->fnhe_genid != genid)
3094 				goto next;
3095 
3096 			if (fnhe->fnhe_expires &&
3097 			    time_after(jiffies, fnhe->fnhe_expires))
3098 				goto next;
3099 
3100 			rt = rcu_dereference(fnhe->fnhe_rth_input);
3101 			if (!rt)
3102 				rt = rcu_dereference(fnhe->fnhe_rth_output);
3103 			if (!rt)
3104 				goto next;
3105 
3106 			err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3107 					   table_id, 0, NULL, skb,
3108 					   NETLINK_CB(cb->skb).portid,
3109 					   cb->nlh->nlmsg_seq, flags);
3110 			if (err)
3111 				return err;
3112 next:
3113 			(*fa_index)++;
3114 		}
3115 	}
3116 
3117 	return 0;
3118 }
3119 
3120 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3121 		       u32 table_id, struct fib_info *fi,
3122 		       int *fa_index, int fa_start, unsigned int flags)
3123 {
3124 	struct net *net = sock_net(cb->skb->sk);
3125 	int nhsel, genid = fnhe_genid(net);
3126 
3127 	for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3128 		struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3129 		struct fnhe_hash_bucket *bucket;
3130 		int err;
3131 
3132 		if (nhc->nhc_flags & RTNH_F_DEAD)
3133 			continue;
3134 
3135 		rcu_read_lock();
3136 		bucket = rcu_dereference(nhc->nhc_exceptions);
3137 		err = 0;
3138 		if (bucket)
3139 			err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3140 					       genid, fa_index, fa_start,
3141 					       flags);
3142 		rcu_read_unlock();
3143 		if (err)
3144 			return err;
3145 	}
3146 
3147 	return 0;
3148 }
3149 
3150 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3151 						   u8 ip_proto, __be16 sport,
3152 						   __be16 dport)
3153 {
3154 	struct sk_buff *skb;
3155 	struct iphdr *iph;
3156 
3157 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3158 	if (!skb)
3159 		return NULL;
3160 
3161 	/* Reserve room for dummy headers, this skb can pass
3162 	 * through good chunk of routing engine.
3163 	 */
3164 	skb_reset_mac_header(skb);
3165 	skb_reset_network_header(skb);
3166 	skb->protocol = htons(ETH_P_IP);
3167 	iph = skb_put(skb, sizeof(struct iphdr));
3168 	iph->protocol = ip_proto;
3169 	iph->saddr = src;
3170 	iph->daddr = dst;
3171 	iph->version = 0x4;
3172 	iph->frag_off = 0;
3173 	iph->ihl = 0x5;
3174 	skb_set_transport_header(skb, skb->len);
3175 
3176 	switch (iph->protocol) {
3177 	case IPPROTO_UDP: {
3178 		struct udphdr *udph;
3179 
3180 		udph = skb_put_zero(skb, sizeof(struct udphdr));
3181 		udph->source = sport;
3182 		udph->dest = dport;
3183 		udph->len = htons(sizeof(struct udphdr));
3184 		udph->check = 0;
3185 		break;
3186 	}
3187 	case IPPROTO_TCP: {
3188 		struct tcphdr *tcph;
3189 
3190 		tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3191 		tcph->source	= sport;
3192 		tcph->dest	= dport;
3193 		tcph->doff	= sizeof(struct tcphdr) / 4;
3194 		tcph->rst = 1;
3195 		tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3196 					    src, dst, 0);
3197 		break;
3198 	}
3199 	case IPPROTO_ICMP: {
3200 		struct icmphdr *icmph;
3201 
3202 		icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3203 		icmph->type = ICMP_ECHO;
3204 		icmph->code = 0;
3205 	}
3206 	}
3207 
3208 	return skb;
3209 }
3210 
3211 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3212 				       const struct nlmsghdr *nlh,
3213 				       struct nlattr **tb,
3214 				       struct netlink_ext_ack *extack)
3215 {
3216 	struct rtmsg *rtm;
3217 	int i, err;
3218 
3219 	rtm = nlmsg_payload(nlh, sizeof(*rtm));
3220 	if (!rtm) {
3221 		NL_SET_ERR_MSG(extack,
3222 			       "ipv4: Invalid header for route get request");
3223 		return -EINVAL;
3224 	}
3225 
3226 	if (!netlink_strict_get_check(skb))
3227 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3228 					      rtm_ipv4_policy, extack);
3229 
3230 	if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3231 	    (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3232 	    rtm->rtm_table || rtm->rtm_protocol ||
3233 	    rtm->rtm_scope || rtm->rtm_type) {
3234 		NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3235 		return -EINVAL;
3236 	}
3237 
3238 	if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3239 			       RTM_F_LOOKUP_TABLE |
3240 			       RTM_F_FIB_MATCH)) {
3241 		NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3242 		return -EINVAL;
3243 	}
3244 
3245 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3246 					    rtm_ipv4_policy, extack);
3247 	if (err)
3248 		return err;
3249 
3250 	if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3251 	    (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3252 		NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3253 		return -EINVAL;
3254 	}
3255 
3256 	for (i = 0; i <= RTA_MAX; i++) {
3257 		if (!tb[i])
3258 			continue;
3259 
3260 		switch (i) {
3261 		case RTA_IIF:
3262 		case RTA_OIF:
3263 		case RTA_SRC:
3264 		case RTA_DST:
3265 		case RTA_IP_PROTO:
3266 		case RTA_SPORT:
3267 		case RTA_DPORT:
3268 		case RTA_MARK:
3269 		case RTA_UID:
3270 			break;
3271 		default:
3272 			NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3273 			return -EINVAL;
3274 		}
3275 	}
3276 
3277 	return 0;
3278 }
3279 
3280 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3281 			     struct netlink_ext_ack *extack)
3282 {
3283 	struct net *net = sock_net(in_skb->sk);
3284 	struct nlattr *tb[RTA_MAX+1];
3285 	u32 table_id = RT_TABLE_MAIN;
3286 	__be16 sport = 0, dport = 0;
3287 	struct fib_result res = {};
3288 	u8 ip_proto = IPPROTO_UDP;
3289 	struct rtable *rt = NULL;
3290 	struct sk_buff *skb;
3291 	struct rtmsg *rtm;
3292 	struct flowi4 fl4 = {};
3293 	__be32 dst = 0;
3294 	__be32 src = 0;
3295 	dscp_t dscp;
3296 	kuid_t uid;
3297 	u32 iif;
3298 	int err;
3299 	int mark;
3300 
3301 	err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3302 	if (err < 0)
3303 		return err;
3304 
3305 	rtm = nlmsg_data(nlh);
3306 	src = nla_get_in_addr_default(tb[RTA_SRC], 0);
3307 	dst = nla_get_in_addr_default(tb[RTA_DST], 0);
3308 	iif = nla_get_u32_default(tb[RTA_IIF], 0);
3309 	mark = nla_get_u32_default(tb[RTA_MARK], 0);
3310 	dscp = inet_dsfield_to_dscp(rtm->rtm_tos);
3311 	if (tb[RTA_UID])
3312 		uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3313 	else
3314 		uid = (iif ? INVALID_UID : current_uid());
3315 
3316 	if (tb[RTA_IP_PROTO]) {
3317 		err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3318 						  &ip_proto, AF_INET, extack);
3319 		if (err)
3320 			return err;
3321 	}
3322 
3323 	if (tb[RTA_SPORT])
3324 		sport = nla_get_be16(tb[RTA_SPORT]);
3325 
3326 	if (tb[RTA_DPORT])
3327 		dport = nla_get_be16(tb[RTA_DPORT]);
3328 
3329 	skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3330 	if (!skb)
3331 		return -ENOBUFS;
3332 
3333 	fl4.daddr = dst;
3334 	fl4.saddr = src;
3335 	fl4.flowi4_tos = inet_dscp_to_dsfield(dscp);
3336 	fl4.flowi4_oif = nla_get_u32_default(tb[RTA_OIF], 0);
3337 	fl4.flowi4_mark = mark;
3338 	fl4.flowi4_uid = uid;
3339 	if (sport)
3340 		fl4.fl4_sport = sport;
3341 	if (dport)
3342 		fl4.fl4_dport = dport;
3343 	fl4.flowi4_proto = ip_proto;
3344 
3345 	rcu_read_lock();
3346 
3347 	if (iif) {
3348 		struct net_device *dev;
3349 
3350 		dev = dev_get_by_index_rcu(net, iif);
3351 		if (!dev) {
3352 			err = -ENODEV;
3353 			goto errout_rcu;
3354 		}
3355 
3356 		fl4.flowi4_iif = iif; /* for rt_fill_info */
3357 		skb->dev	= dev;
3358 		skb->mark	= mark;
3359 		err = ip_route_input_rcu(skb, dst, src, dscp, dev,
3360 					 &res) ? -EINVAL : 0;
3361 
3362 		rt = skb_rtable(skb);
3363 		if (err == 0 && rt->dst.error)
3364 			err = -rt->dst.error;
3365 	} else {
3366 		fl4.flowi4_iif = LOOPBACK_IFINDEX;
3367 		skb->dev = net->loopback_dev;
3368 		rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3369 		err = 0;
3370 		if (IS_ERR(rt))
3371 			err = PTR_ERR(rt);
3372 		else
3373 			skb_dst_set(skb, &rt->dst);
3374 	}
3375 
3376 	if (err)
3377 		goto errout_rcu;
3378 
3379 	if (rtm->rtm_flags & RTM_F_NOTIFY)
3380 		rt->rt_flags |= RTCF_NOTIFY;
3381 
3382 	if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3383 		table_id = res.table ? res.table->tb_id : 0;
3384 
3385 	/* reset skb for netlink reply msg */
3386 	skb_trim(skb, 0);
3387 	skb_reset_network_header(skb);
3388 	skb_reset_transport_header(skb);
3389 	skb_reset_mac_header(skb);
3390 
3391 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3392 		struct fib_rt_info fri;
3393 
3394 		if (!res.fi) {
3395 			err = fib_props[res.type].error;
3396 			if (!err)
3397 				err = -EHOSTUNREACH;
3398 			goto errout_rcu;
3399 		}
3400 		fri.fi = res.fi;
3401 		fri.tb_id = table_id;
3402 		fri.dst = res.prefix;
3403 		fri.dst_len = res.prefixlen;
3404 		fri.dscp = res.dscp;
3405 		fri.type = rt->rt_type;
3406 		fri.offload = 0;
3407 		fri.trap = 0;
3408 		fri.offload_failed = 0;
3409 		if (res.fa_head) {
3410 			struct fib_alias *fa;
3411 
3412 			hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3413 				u8 slen = 32 - fri.dst_len;
3414 
3415 				if (fa->fa_slen == slen &&
3416 				    fa->tb_id == fri.tb_id &&
3417 				    fa->fa_dscp == fri.dscp &&
3418 				    fa->fa_info == res.fi &&
3419 				    fa->fa_type == fri.type) {
3420 					fri.offload = READ_ONCE(fa->offload);
3421 					fri.trap = READ_ONCE(fa->trap);
3422 					fri.offload_failed =
3423 						READ_ONCE(fa->offload_failed);
3424 					break;
3425 				}
3426 			}
3427 		}
3428 		err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3429 				    nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3430 	} else {
3431 		err = rt_fill_info(net, dst, src, rt, table_id, res.dscp, &fl4,
3432 				   skb, NETLINK_CB(in_skb).portid,
3433 				   nlh->nlmsg_seq, 0);
3434 	}
3435 	if (err < 0)
3436 		goto errout_rcu;
3437 
3438 	rcu_read_unlock();
3439 
3440 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3441 
3442 errout_free:
3443 	return err;
3444 errout_rcu:
3445 	rcu_read_unlock();
3446 	kfree_skb(skb);
3447 	goto errout_free;
3448 }
3449 
3450 void ip_rt_multicast_event(struct in_device *in_dev)
3451 {
3452 	rt_cache_flush(dev_net(in_dev->dev));
3453 }
3454 
3455 #ifdef CONFIG_SYSCTL
3456 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3457 static int ip_rt_gc_min_interval __read_mostly	= HZ / 2;
3458 static int ip_rt_gc_elasticity __read_mostly	= 8;
3459 static int ip_min_valid_pmtu __read_mostly	= IPV4_MIN_MTU;
3460 
3461 static int ipv4_sysctl_rtcache_flush(const struct ctl_table *__ctl, int write,
3462 		void *buffer, size_t *lenp, loff_t *ppos)
3463 {
3464 	struct net *net = (struct net *)__ctl->extra1;
3465 
3466 	if (write) {
3467 		rt_cache_flush(net);
3468 		fnhe_genid_bump(net);
3469 		return 0;
3470 	}
3471 
3472 	return -EINVAL;
3473 }
3474 
3475 static struct ctl_table ipv4_route_table[] = {
3476 	{
3477 		.procname	= "gc_thresh",
3478 		.data		= &ipv4_dst_ops.gc_thresh,
3479 		.maxlen		= sizeof(int),
3480 		.mode		= 0644,
3481 		.proc_handler	= proc_dointvec,
3482 	},
3483 	{
3484 		.procname	= "max_size",
3485 		.data		= &ip_rt_max_size,
3486 		.maxlen		= sizeof(int),
3487 		.mode		= 0644,
3488 		.proc_handler	= proc_dointvec,
3489 	},
3490 	{
3491 		/*  Deprecated. Use gc_min_interval_ms */
3492 
3493 		.procname	= "gc_min_interval",
3494 		.data		= &ip_rt_gc_min_interval,
3495 		.maxlen		= sizeof(int),
3496 		.mode		= 0644,
3497 		.proc_handler	= proc_dointvec_jiffies,
3498 	},
3499 	{
3500 		.procname	= "gc_min_interval_ms",
3501 		.data		= &ip_rt_gc_min_interval,
3502 		.maxlen		= sizeof(int),
3503 		.mode		= 0644,
3504 		.proc_handler	= proc_dointvec_ms_jiffies,
3505 	},
3506 	{
3507 		.procname	= "gc_timeout",
3508 		.data		= &ip_rt_gc_timeout,
3509 		.maxlen		= sizeof(int),
3510 		.mode		= 0644,
3511 		.proc_handler	= proc_dointvec_jiffies,
3512 	},
3513 	{
3514 		.procname	= "gc_interval",
3515 		.data		= &ip_rt_gc_interval,
3516 		.maxlen		= sizeof(int),
3517 		.mode		= 0644,
3518 		.proc_handler	= proc_dointvec_jiffies,
3519 	},
3520 	{
3521 		.procname	= "redirect_load",
3522 		.data		= &ip_rt_redirect_load,
3523 		.maxlen		= sizeof(int),
3524 		.mode		= 0644,
3525 		.proc_handler	= proc_dointvec,
3526 	},
3527 	{
3528 		.procname	= "redirect_number",
3529 		.data		= &ip_rt_redirect_number,
3530 		.maxlen		= sizeof(int),
3531 		.mode		= 0644,
3532 		.proc_handler	= proc_dointvec,
3533 	},
3534 	{
3535 		.procname	= "redirect_silence",
3536 		.data		= &ip_rt_redirect_silence,
3537 		.maxlen		= sizeof(int),
3538 		.mode		= 0644,
3539 		.proc_handler	= proc_dointvec,
3540 	},
3541 	{
3542 		.procname	= "error_cost",
3543 		.data		= &ip_rt_error_cost,
3544 		.maxlen		= sizeof(int),
3545 		.mode		= 0644,
3546 		.proc_handler	= proc_dointvec,
3547 	},
3548 	{
3549 		.procname	= "error_burst",
3550 		.data		= &ip_rt_error_burst,
3551 		.maxlen		= sizeof(int),
3552 		.mode		= 0644,
3553 		.proc_handler	= proc_dointvec,
3554 	},
3555 	{
3556 		.procname	= "gc_elasticity",
3557 		.data		= &ip_rt_gc_elasticity,
3558 		.maxlen		= sizeof(int),
3559 		.mode		= 0644,
3560 		.proc_handler	= proc_dointvec,
3561 	},
3562 };
3563 
3564 static const char ipv4_route_flush_procname[] = "flush";
3565 
3566 static struct ctl_table ipv4_route_netns_table[] = {
3567 	{
3568 		.procname	= ipv4_route_flush_procname,
3569 		.maxlen		= sizeof(int),
3570 		.mode		= 0200,
3571 		.proc_handler	= ipv4_sysctl_rtcache_flush,
3572 	},
3573 	{
3574 		.procname       = "min_pmtu",
3575 		.data           = &init_net.ipv4.ip_rt_min_pmtu,
3576 		.maxlen         = sizeof(int),
3577 		.mode           = 0644,
3578 		.proc_handler   = proc_dointvec_minmax,
3579 		.extra1         = &ip_min_valid_pmtu,
3580 	},
3581 	{
3582 		.procname       = "mtu_expires",
3583 		.data           = &init_net.ipv4.ip_rt_mtu_expires,
3584 		.maxlen         = sizeof(int),
3585 		.mode           = 0644,
3586 		.proc_handler   = proc_dointvec_jiffies,
3587 	},
3588 	{
3589 		.procname   = "min_adv_mss",
3590 		.data       = &init_net.ipv4.ip_rt_min_advmss,
3591 		.maxlen     = sizeof(int),
3592 		.mode       = 0644,
3593 		.proc_handler   = proc_dointvec,
3594 	},
3595 };
3596 
3597 static __net_init int sysctl_route_net_init(struct net *net)
3598 {
3599 	struct ctl_table *tbl;
3600 	size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3601 
3602 	tbl = ipv4_route_netns_table;
3603 	if (!net_eq(net, &init_net)) {
3604 		int i;
3605 
3606 		tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3607 		if (!tbl)
3608 			goto err_dup;
3609 
3610 		/* Don't export non-whitelisted sysctls to unprivileged users */
3611 		if (net->user_ns != &init_user_ns) {
3612 			if (tbl[0].procname != ipv4_route_flush_procname)
3613 				table_size = 0;
3614 		}
3615 
3616 		/* Update the variables to point into the current struct net
3617 		 * except for the first element flush
3618 		 */
3619 		for (i = 1; i < table_size; i++)
3620 			tbl[i].data += (void *)net - (void *)&init_net;
3621 	}
3622 	tbl[0].extra1 = net;
3623 
3624 	net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3625 						     tbl, table_size);
3626 	if (!net->ipv4.route_hdr)
3627 		goto err_reg;
3628 	return 0;
3629 
3630 err_reg:
3631 	if (tbl != ipv4_route_netns_table)
3632 		kfree(tbl);
3633 err_dup:
3634 	return -ENOMEM;
3635 }
3636 
3637 static __net_exit void sysctl_route_net_exit(struct net *net)
3638 {
3639 	const struct ctl_table *tbl;
3640 
3641 	tbl = net->ipv4.route_hdr->ctl_table_arg;
3642 	unregister_net_sysctl_table(net->ipv4.route_hdr);
3643 	BUG_ON(tbl == ipv4_route_netns_table);
3644 	kfree(tbl);
3645 }
3646 
3647 static __net_initdata struct pernet_operations sysctl_route_ops = {
3648 	.init = sysctl_route_net_init,
3649 	.exit = sysctl_route_net_exit,
3650 };
3651 #endif
3652 
3653 static __net_init int netns_ip_rt_init(struct net *net)
3654 {
3655 	/* Set default value for namespaceified sysctls */
3656 	net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3657 	net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3658 	net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3659 	return 0;
3660 }
3661 
3662 static struct pernet_operations __net_initdata ip_rt_ops = {
3663 	.init = netns_ip_rt_init,
3664 };
3665 
3666 static __net_init int rt_genid_init(struct net *net)
3667 {
3668 	atomic_set(&net->ipv4.rt_genid, 0);
3669 	atomic_set(&net->fnhe_genid, 0);
3670 	atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3671 	return 0;
3672 }
3673 
3674 static __net_initdata struct pernet_operations rt_genid_ops = {
3675 	.init = rt_genid_init,
3676 };
3677 
3678 static int __net_init ipv4_inetpeer_init(struct net *net)
3679 {
3680 	struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3681 
3682 	if (!bp)
3683 		return -ENOMEM;
3684 	inet_peer_base_init(bp);
3685 	net->ipv4.peers = bp;
3686 	return 0;
3687 }
3688 
3689 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3690 {
3691 	struct inet_peer_base *bp = net->ipv4.peers;
3692 
3693 	net->ipv4.peers = NULL;
3694 	inetpeer_invalidate_tree(bp);
3695 	kfree(bp);
3696 }
3697 
3698 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3699 	.init	=	ipv4_inetpeer_init,
3700 	.exit	=	ipv4_inetpeer_exit,
3701 };
3702 
3703 #ifdef CONFIG_IP_ROUTE_CLASSID
3704 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3705 #endif /* CONFIG_IP_ROUTE_CLASSID */
3706 
3707 static const struct rtnl_msg_handler ip_rt_rtnl_msg_handlers[] __initconst = {
3708 	{.protocol = PF_INET, .msgtype = RTM_GETROUTE,
3709 	 .doit = inet_rtm_getroute, .flags = RTNL_FLAG_DOIT_UNLOCKED},
3710 };
3711 
3712 int __init ip_rt_init(void)
3713 {
3714 	void *idents_hash;
3715 	int cpu;
3716 
3717 	/* For modern hosts, this will use 2 MB of memory */
3718 	idents_hash = alloc_large_system_hash("IP idents",
3719 					      sizeof(*ip_idents) + sizeof(*ip_tstamps),
3720 					      0,
3721 					      16, /* one bucket per 64 KB */
3722 					      HASH_ZERO,
3723 					      NULL,
3724 					      &ip_idents_mask,
3725 					      2048,
3726 					      256*1024);
3727 
3728 	ip_idents = idents_hash;
3729 
3730 	get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3731 
3732 	ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3733 
3734 	for_each_possible_cpu(cpu) {
3735 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3736 
3737 		INIT_LIST_HEAD(&ul->head);
3738 		spin_lock_init(&ul->lock);
3739 	}
3740 #ifdef CONFIG_IP_ROUTE_CLASSID
3741 	ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3742 	if (!ip_rt_acct)
3743 		panic("IP: failed to allocate ip_rt_acct\n");
3744 #endif
3745 
3746 	ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable,
3747 					      SLAB_HWCACHE_ALIGN | SLAB_PANIC);
3748 
3749 	ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3750 
3751 	if (dst_entries_init(&ipv4_dst_ops) < 0)
3752 		panic("IP: failed to allocate ipv4_dst_ops counter\n");
3753 
3754 	if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3755 		panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3756 
3757 	ipv4_dst_ops.gc_thresh = ~0;
3758 	ip_rt_max_size = INT_MAX;
3759 
3760 	devinet_init();
3761 	ip_fib_init();
3762 
3763 	if (ip_rt_proc_init())
3764 		pr_err("Unable to create route proc files\n");
3765 #ifdef CONFIG_XFRM
3766 	xfrm_init();
3767 	xfrm4_init();
3768 #endif
3769 	rtnl_register_many(ip_rt_rtnl_msg_handlers);
3770 
3771 #ifdef CONFIG_SYSCTL
3772 	register_pernet_subsys(&sysctl_route_ops);
3773 #endif
3774 	register_pernet_subsys(&ip_rt_ops);
3775 	register_pernet_subsys(&rt_genid_ops);
3776 	register_pernet_subsys(&ipv4_inetpeer_ops);
3777 	return 0;
3778 }
3779 
3780 #ifdef CONFIG_SYSCTL
3781 /*
3782  * We really need to sanitize the damn ipv4 init order, then all
3783  * this nonsense will go away.
3784  */
3785 void __init ip_static_sysctl_init(void)
3786 {
3787 	register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3788 }
3789 #endif
3790