xref: /openbmc/linux/net/ipv4/route.c (revision fbb6b31a)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		ROUTE - implementation of the IP router.
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
12  *		Linus Torvalds, <Linus.Torvalds@helsinki.fi>
13  *		Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
14  *
15  * Fixes:
16  *		Alan Cox	:	Verify area fixes.
17  *		Alan Cox	:	cli() protects routing changes
18  *		Rui Oliveira	:	ICMP routing table updates
19  *		(rco@di.uminho.pt)	Routing table insertion and update
20  *		Linus Torvalds	:	Rewrote bits to be sensible
21  *		Alan Cox	:	Added BSD route gw semantics
22  *		Alan Cox	:	Super /proc >4K
23  *		Alan Cox	:	MTU in route table
24  *		Alan Cox	:	MSS actually. Also added the window
25  *					clamper.
26  *		Sam Lantinga	:	Fixed route matching in rt_del()
27  *		Alan Cox	:	Routing cache support.
28  *		Alan Cox	:	Removed compatibility cruft.
29  *		Alan Cox	:	RTF_REJECT support.
30  *		Alan Cox	:	TCP irtt support.
31  *		Jonathan Naylor	:	Added Metric support.
32  *	Miquel van Smoorenburg	:	BSD API fixes.
33  *	Miquel van Smoorenburg	:	Metrics.
34  *		Alan Cox	:	Use __u32 properly
35  *		Alan Cox	:	Aligned routing errors more closely with BSD
36  *					our system is still very different.
37  *		Alan Cox	:	Faster /proc handling
38  *	Alexey Kuznetsov	:	Massive rework to support tree based routing,
39  *					routing caches and better behaviour.
40  *
41  *		Olaf Erb	:	irtt wasn't being copied right.
42  *		Bjorn Ekwall	:	Kerneld route support.
43  *		Alan Cox	:	Multicast fixed (I hope)
44  *		Pavel Krauz	:	Limited broadcast fixed
45  *		Mike McLagan	:	Routing by source
46  *	Alexey Kuznetsov	:	End of old history. Split to fib.c and
47  *					route.c and rewritten from scratch.
48  *		Andi Kleen	:	Load-limit warning messages.
49  *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
50  *	Vitaly E. Lavrov	:	Race condition in ip_route_input_slow.
51  *	Tobias Ringstrom	:	Uninitialized res.type in ip_route_output_slow.
52  *	Vladimir V. Ivanov	:	IP rule info (flowid) is really useful.
53  *		Marc Boucher	:	routing by fwmark
54  *	Robert Olsson		:	Added rt_cache statistics
55  *	Arnaldo C. Melo		:	Convert proc stuff to seq_file
56  *	Eric Dumazet		:	hashed spinlocks and rt_check_expire() fixes.
57  *	Ilia Sotnikov		:	Ignore TOS on PMTUD and Redirect
58  *	Ilia Sotnikov		:	Removed TOS from hash calculations
59  */
60 
61 #define pr_fmt(fmt) "IPv4: " fmt
62 
63 #include <linux/module.h>
64 #include <linux/bitops.h>
65 #include <linux/kernel.h>
66 #include <linux/mm.h>
67 #include <linux/memblock.h>
68 #include <linux/socket.h>
69 #include <linux/errno.h>
70 #include <linux/in.h>
71 #include <linux/inet.h>
72 #include <linux/netdevice.h>
73 #include <linux/proc_fs.h>
74 #include <linux/init.h>
75 #include <linux/skbuff.h>
76 #include <linux/inetdevice.h>
77 #include <linux/igmp.h>
78 #include <linux/pkt_sched.h>
79 #include <linux/mroute.h>
80 #include <linux/netfilter_ipv4.h>
81 #include <linux/random.h>
82 #include <linux/rcupdate.h>
83 #include <linux/slab.h>
84 #include <linux/jhash.h>
85 #include <net/dst.h>
86 #include <net/dst_metadata.h>
87 #include <net/inet_dscp.h>
88 #include <net/net_namespace.h>
89 #include <net/ip.h>
90 #include <net/route.h>
91 #include <net/inetpeer.h>
92 #include <net/sock.h>
93 #include <net/ip_fib.h>
94 #include <net/nexthop.h>
95 #include <net/tcp.h>
96 #include <net/icmp.h>
97 #include <net/xfrm.h>
98 #include <net/lwtunnel.h>
99 #include <net/netevent.h>
100 #include <net/rtnetlink.h>
101 #ifdef CONFIG_SYSCTL
102 #include <linux/sysctl.h>
103 #endif
104 #include <net/secure_seq.h>
105 #include <net/ip_tunnels.h>
106 
107 #include "fib_lookup.h"
108 
109 #define RT_FL_TOS(oldflp4) \
110 	((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
111 
112 #define RT_GC_TIMEOUT (300*HZ)
113 
114 #define DEFAULT_MIN_PMTU (512 + 20 + 20)
115 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
116 #define DEFAULT_MIN_ADVMSS 256
117 static int ip_rt_max_size;
118 static int ip_rt_redirect_number __read_mostly	= 9;
119 static int ip_rt_redirect_load __read_mostly	= HZ / 50;
120 static int ip_rt_redirect_silence __read_mostly	= ((HZ / 50) << (9 + 1));
121 static int ip_rt_error_cost __read_mostly	= HZ;
122 static int ip_rt_error_burst __read_mostly	= 5 * HZ;
123 
124 static int ip_rt_gc_timeout __read_mostly	= RT_GC_TIMEOUT;
125 
126 /*
127  *	Interface to generic destination cache.
128  */
129 
130 INDIRECT_CALLABLE_SCOPE
131 struct dst_entry	*ipv4_dst_check(struct dst_entry *dst, u32 cookie);
132 static unsigned int	 ipv4_default_advmss(const struct dst_entry *dst);
133 INDIRECT_CALLABLE_SCOPE
134 unsigned int		ipv4_mtu(const struct dst_entry *dst);
135 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
136 static void		 ipv4_link_failure(struct sk_buff *skb);
137 static void		 ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
138 					   struct sk_buff *skb, u32 mtu,
139 					   bool confirm_neigh);
140 static void		 ip_do_redirect(struct dst_entry *dst, struct sock *sk,
141 					struct sk_buff *skb);
142 static void		ipv4_dst_destroy(struct dst_entry *dst);
143 
144 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
145 {
146 	WARN_ON(1);
147 	return NULL;
148 }
149 
150 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
151 					   struct sk_buff *skb,
152 					   const void *daddr);
153 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
154 
155 static struct dst_ops ipv4_dst_ops = {
156 	.family =		AF_INET,
157 	.check =		ipv4_dst_check,
158 	.default_advmss =	ipv4_default_advmss,
159 	.mtu =			ipv4_mtu,
160 	.cow_metrics =		ipv4_cow_metrics,
161 	.destroy =		ipv4_dst_destroy,
162 	.negative_advice =	ipv4_negative_advice,
163 	.link_failure =		ipv4_link_failure,
164 	.update_pmtu =		ip_rt_update_pmtu,
165 	.redirect =		ip_do_redirect,
166 	.local_out =		__ip_local_out,
167 	.neigh_lookup =		ipv4_neigh_lookup,
168 	.confirm_neigh =	ipv4_confirm_neigh,
169 };
170 
171 #define ECN_OR_COST(class)	TC_PRIO_##class
172 
173 const __u8 ip_tos2prio[16] = {
174 	TC_PRIO_BESTEFFORT,
175 	ECN_OR_COST(BESTEFFORT),
176 	TC_PRIO_BESTEFFORT,
177 	ECN_OR_COST(BESTEFFORT),
178 	TC_PRIO_BULK,
179 	ECN_OR_COST(BULK),
180 	TC_PRIO_BULK,
181 	ECN_OR_COST(BULK),
182 	TC_PRIO_INTERACTIVE,
183 	ECN_OR_COST(INTERACTIVE),
184 	TC_PRIO_INTERACTIVE,
185 	ECN_OR_COST(INTERACTIVE),
186 	TC_PRIO_INTERACTIVE_BULK,
187 	ECN_OR_COST(INTERACTIVE_BULK),
188 	TC_PRIO_INTERACTIVE_BULK,
189 	ECN_OR_COST(INTERACTIVE_BULK)
190 };
191 EXPORT_SYMBOL(ip_tos2prio);
192 
193 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
194 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
195 
196 #ifdef CONFIG_PROC_FS
197 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
198 {
199 	if (*pos)
200 		return NULL;
201 	return SEQ_START_TOKEN;
202 }
203 
204 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
205 {
206 	++*pos;
207 	return NULL;
208 }
209 
210 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
211 {
212 }
213 
214 static int rt_cache_seq_show(struct seq_file *seq, void *v)
215 {
216 	if (v == SEQ_START_TOKEN)
217 		seq_printf(seq, "%-127s\n",
218 			   "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
219 			   "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
220 			   "HHUptod\tSpecDst");
221 	return 0;
222 }
223 
224 static const struct seq_operations rt_cache_seq_ops = {
225 	.start  = rt_cache_seq_start,
226 	.next   = rt_cache_seq_next,
227 	.stop   = rt_cache_seq_stop,
228 	.show   = rt_cache_seq_show,
229 };
230 
231 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
232 {
233 	int cpu;
234 
235 	if (*pos == 0)
236 		return SEQ_START_TOKEN;
237 
238 	for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
239 		if (!cpu_possible(cpu))
240 			continue;
241 		*pos = cpu+1;
242 		return &per_cpu(rt_cache_stat, cpu);
243 	}
244 	return NULL;
245 }
246 
247 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
248 {
249 	int cpu;
250 
251 	for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
252 		if (!cpu_possible(cpu))
253 			continue;
254 		*pos = cpu+1;
255 		return &per_cpu(rt_cache_stat, cpu);
256 	}
257 	(*pos)++;
258 	return NULL;
259 
260 }
261 
262 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
263 {
264 
265 }
266 
267 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
268 {
269 	struct rt_cache_stat *st = v;
270 
271 	if (v == SEQ_START_TOKEN) {
272 		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");
273 		return 0;
274 	}
275 
276 	seq_printf(seq, "%08x %08x %08x    %08x   %08x    %08x %08x       "
277 			"%08x       %08x %08x     %08x    %08x %08x   "
278 			"%08x     %08x        %08x        %08x\n",
279 		   dst_entries_get_slow(&ipv4_dst_ops),
280 		   0, /* st->in_hit */
281 		   st->in_slow_tot,
282 		   st->in_slow_mc,
283 		   st->in_no_route,
284 		   st->in_brd,
285 		   st->in_martian_dst,
286 		   st->in_martian_src,
287 
288 		   0, /* st->out_hit */
289 		   st->out_slow_tot,
290 		   st->out_slow_mc,
291 
292 		   0, /* st->gc_total */
293 		   0, /* st->gc_ignored */
294 		   0, /* st->gc_goal_miss */
295 		   0, /* st->gc_dst_overflow */
296 		   0, /* st->in_hlist_search */
297 		   0  /* st->out_hlist_search */
298 		);
299 	return 0;
300 }
301 
302 static const struct seq_operations rt_cpu_seq_ops = {
303 	.start  = rt_cpu_seq_start,
304 	.next   = rt_cpu_seq_next,
305 	.stop   = rt_cpu_seq_stop,
306 	.show   = rt_cpu_seq_show,
307 };
308 
309 #ifdef CONFIG_IP_ROUTE_CLASSID
310 static int rt_acct_proc_show(struct seq_file *m, void *v)
311 {
312 	struct ip_rt_acct *dst, *src;
313 	unsigned int i, j;
314 
315 	dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
316 	if (!dst)
317 		return -ENOMEM;
318 
319 	for_each_possible_cpu(i) {
320 		src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
321 		for (j = 0; j < 256; j++) {
322 			dst[j].o_bytes   += src[j].o_bytes;
323 			dst[j].o_packets += src[j].o_packets;
324 			dst[j].i_bytes   += src[j].i_bytes;
325 			dst[j].i_packets += src[j].i_packets;
326 		}
327 	}
328 
329 	seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
330 	kfree(dst);
331 	return 0;
332 }
333 #endif
334 
335 static int __net_init ip_rt_do_proc_init(struct net *net)
336 {
337 	struct proc_dir_entry *pde;
338 
339 	pde = proc_create_seq("rt_cache", 0444, net->proc_net,
340 			      &rt_cache_seq_ops);
341 	if (!pde)
342 		goto err1;
343 
344 	pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
345 			      &rt_cpu_seq_ops);
346 	if (!pde)
347 		goto err2;
348 
349 #ifdef CONFIG_IP_ROUTE_CLASSID
350 	pde = proc_create_single("rt_acct", 0, net->proc_net,
351 			rt_acct_proc_show);
352 	if (!pde)
353 		goto err3;
354 #endif
355 	return 0;
356 
357 #ifdef CONFIG_IP_ROUTE_CLASSID
358 err3:
359 	remove_proc_entry("rt_cache", net->proc_net_stat);
360 #endif
361 err2:
362 	remove_proc_entry("rt_cache", net->proc_net);
363 err1:
364 	return -ENOMEM;
365 }
366 
367 static void __net_exit ip_rt_do_proc_exit(struct net *net)
368 {
369 	remove_proc_entry("rt_cache", net->proc_net_stat);
370 	remove_proc_entry("rt_cache", net->proc_net);
371 #ifdef CONFIG_IP_ROUTE_CLASSID
372 	remove_proc_entry("rt_acct", net->proc_net);
373 #endif
374 }
375 
376 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
377 	.init = ip_rt_do_proc_init,
378 	.exit = ip_rt_do_proc_exit,
379 };
380 
381 static int __init ip_rt_proc_init(void)
382 {
383 	return register_pernet_subsys(&ip_rt_proc_ops);
384 }
385 
386 #else
387 static inline int ip_rt_proc_init(void)
388 {
389 	return 0;
390 }
391 #endif /* CONFIG_PROC_FS */
392 
393 static inline bool rt_is_expired(const struct rtable *rth)
394 {
395 	return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
396 }
397 
398 void rt_cache_flush(struct net *net)
399 {
400 	rt_genid_bump_ipv4(net);
401 }
402 
403 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
404 					   struct sk_buff *skb,
405 					   const void *daddr)
406 {
407 	const struct rtable *rt = container_of(dst, struct rtable, dst);
408 	struct net_device *dev = dst->dev;
409 	struct neighbour *n;
410 
411 	rcu_read_lock_bh();
412 
413 	if (likely(rt->rt_gw_family == AF_INET)) {
414 		n = ip_neigh_gw4(dev, rt->rt_gw4);
415 	} else if (rt->rt_gw_family == AF_INET6) {
416 		n = ip_neigh_gw6(dev, &rt->rt_gw6);
417         } else {
418 		__be32 pkey;
419 
420 		pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
421 		n = ip_neigh_gw4(dev, pkey);
422 	}
423 
424 	if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
425 		n = NULL;
426 
427 	rcu_read_unlock_bh();
428 
429 	return n;
430 }
431 
432 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
433 {
434 	const struct rtable *rt = container_of(dst, struct rtable, dst);
435 	struct net_device *dev = dst->dev;
436 	const __be32 *pkey = daddr;
437 
438 	if (rt->rt_gw_family == AF_INET) {
439 		pkey = (const __be32 *)&rt->rt_gw4;
440 	} else if (rt->rt_gw_family == AF_INET6) {
441 		return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
442 	} else if (!daddr ||
443 		 (rt->rt_flags &
444 		  (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
445 		return;
446 	}
447 	__ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
448 }
449 
450 /* Hash tables of size 2048..262144 depending on RAM size.
451  * Each bucket uses 8 bytes.
452  */
453 static u32 ip_idents_mask __read_mostly;
454 static atomic_t *ip_idents __read_mostly;
455 static u32 *ip_tstamps __read_mostly;
456 
457 /* In order to protect privacy, we add a perturbation to identifiers
458  * if one generator is seldom used. This makes hard for an attacker
459  * to infer how many packets were sent between two points in time.
460  */
461 static u32 ip_idents_reserve(u32 hash, int segs)
462 {
463 	u32 bucket, old, now = (u32)jiffies;
464 	atomic_t *p_id;
465 	u32 *p_tstamp;
466 	u32 delta = 0;
467 
468 	bucket = hash & ip_idents_mask;
469 	p_tstamp = ip_tstamps + bucket;
470 	p_id = ip_idents + bucket;
471 	old = READ_ONCE(*p_tstamp);
472 
473 	if (old != now && cmpxchg(p_tstamp, old, now) == old)
474 		delta = prandom_u32_max(now - old);
475 
476 	/* If UBSAN reports an error there, please make sure your compiler
477 	 * supports -fno-strict-overflow before reporting it that was a bug
478 	 * in UBSAN, and it has been fixed in GCC-8.
479 	 */
480 	return atomic_add_return(segs + delta, p_id) - segs;
481 }
482 
483 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
484 {
485 	u32 hash, id;
486 
487 	/* Note the following code is not safe, but this is okay. */
488 	if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
489 		get_random_bytes(&net->ipv4.ip_id_key,
490 				 sizeof(net->ipv4.ip_id_key));
491 
492 	hash = siphash_3u32((__force u32)iph->daddr,
493 			    (__force u32)iph->saddr,
494 			    iph->protocol,
495 			    &net->ipv4.ip_id_key);
496 	id = ip_idents_reserve(hash, segs);
497 	iph->id = htons(id);
498 }
499 EXPORT_SYMBOL(__ip_select_ident);
500 
501 static void ip_rt_fix_tos(struct flowi4 *fl4)
502 {
503 	__u8 tos = RT_FL_TOS(fl4);
504 
505 	fl4->flowi4_tos = tos & IPTOS_RT_MASK;
506 	fl4->flowi4_scope = tos & RTO_ONLINK ?
507 			    RT_SCOPE_LINK : RT_SCOPE_UNIVERSE;
508 }
509 
510 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
511 			     const struct sock *sk,
512 			     const struct iphdr *iph,
513 			     int oif, u8 tos,
514 			     u8 prot, u32 mark, int flow_flags)
515 {
516 	if (sk) {
517 		const struct inet_sock *inet = inet_sk(sk);
518 
519 		oif = sk->sk_bound_dev_if;
520 		mark = sk->sk_mark;
521 		tos = RT_CONN_FLAGS(sk);
522 		prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
523 	}
524 	flowi4_init_output(fl4, oif, mark, tos,
525 			   RT_SCOPE_UNIVERSE, prot,
526 			   flow_flags,
527 			   iph->daddr, iph->saddr, 0, 0,
528 			   sock_net_uid(net, sk));
529 }
530 
531 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
532 			       const struct sock *sk)
533 {
534 	const struct net *net = dev_net(skb->dev);
535 	const struct iphdr *iph = ip_hdr(skb);
536 	int oif = skb->dev->ifindex;
537 	u8 tos = RT_TOS(iph->tos);
538 	u8 prot = iph->protocol;
539 	u32 mark = skb->mark;
540 
541 	__build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
542 }
543 
544 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
545 {
546 	const struct inet_sock *inet = inet_sk(sk);
547 	const struct ip_options_rcu *inet_opt;
548 	__be32 daddr = inet->inet_daddr;
549 
550 	rcu_read_lock();
551 	inet_opt = rcu_dereference(inet->inet_opt);
552 	if (inet_opt && inet_opt->opt.srr)
553 		daddr = inet_opt->opt.faddr;
554 	flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
555 			   RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
556 			   inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
557 			   inet_sk_flowi_flags(sk),
558 			   daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
559 	rcu_read_unlock();
560 }
561 
562 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
563 				 const struct sk_buff *skb)
564 {
565 	if (skb)
566 		build_skb_flow_key(fl4, skb, sk);
567 	else
568 		build_sk_flow_key(fl4, sk);
569 }
570 
571 static DEFINE_SPINLOCK(fnhe_lock);
572 
573 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
574 {
575 	struct rtable *rt;
576 
577 	rt = rcu_dereference(fnhe->fnhe_rth_input);
578 	if (rt) {
579 		RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
580 		dst_dev_put(&rt->dst);
581 		dst_release(&rt->dst);
582 	}
583 	rt = rcu_dereference(fnhe->fnhe_rth_output);
584 	if (rt) {
585 		RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
586 		dst_dev_put(&rt->dst);
587 		dst_release(&rt->dst);
588 	}
589 }
590 
591 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
592 {
593 	struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
594 	struct fib_nh_exception *fnhe, *oldest = NULL;
595 
596 	for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
597 		fnhe = rcu_dereference_protected(*fnhe_p,
598 						 lockdep_is_held(&fnhe_lock));
599 		if (!fnhe)
600 			break;
601 		if (!oldest ||
602 		    time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
603 			oldest = fnhe;
604 			oldest_p = fnhe_p;
605 		}
606 	}
607 	fnhe_flush_routes(oldest);
608 	*oldest_p = oldest->fnhe_next;
609 	kfree_rcu(oldest, rcu);
610 }
611 
612 static u32 fnhe_hashfun(__be32 daddr)
613 {
614 	static siphash_aligned_key_t fnhe_hash_key;
615 	u64 hval;
616 
617 	net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
618 	hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
619 	return hash_64(hval, FNHE_HASH_SHIFT);
620 }
621 
622 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
623 {
624 	rt->rt_pmtu = fnhe->fnhe_pmtu;
625 	rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
626 	rt->dst.expires = fnhe->fnhe_expires;
627 
628 	if (fnhe->fnhe_gw) {
629 		rt->rt_flags |= RTCF_REDIRECTED;
630 		rt->rt_uses_gateway = 1;
631 		rt->rt_gw_family = AF_INET;
632 		rt->rt_gw4 = fnhe->fnhe_gw;
633 	}
634 }
635 
636 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
637 				  __be32 gw, u32 pmtu, bool lock,
638 				  unsigned long expires)
639 {
640 	struct fnhe_hash_bucket *hash;
641 	struct fib_nh_exception *fnhe;
642 	struct rtable *rt;
643 	u32 genid, hval;
644 	unsigned int i;
645 	int depth;
646 
647 	genid = fnhe_genid(dev_net(nhc->nhc_dev));
648 	hval = fnhe_hashfun(daddr);
649 
650 	spin_lock_bh(&fnhe_lock);
651 
652 	hash = rcu_dereference(nhc->nhc_exceptions);
653 	if (!hash) {
654 		hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
655 		if (!hash)
656 			goto out_unlock;
657 		rcu_assign_pointer(nhc->nhc_exceptions, hash);
658 	}
659 
660 	hash += hval;
661 
662 	depth = 0;
663 	for (fnhe = rcu_dereference(hash->chain); fnhe;
664 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
665 		if (fnhe->fnhe_daddr == daddr)
666 			break;
667 		depth++;
668 	}
669 
670 	if (fnhe) {
671 		if (fnhe->fnhe_genid != genid)
672 			fnhe->fnhe_genid = genid;
673 		if (gw)
674 			fnhe->fnhe_gw = gw;
675 		if (pmtu) {
676 			fnhe->fnhe_pmtu = pmtu;
677 			fnhe->fnhe_mtu_locked = lock;
678 		}
679 		fnhe->fnhe_expires = max(1UL, expires);
680 		/* Update all cached dsts too */
681 		rt = rcu_dereference(fnhe->fnhe_rth_input);
682 		if (rt)
683 			fill_route_from_fnhe(rt, fnhe);
684 		rt = rcu_dereference(fnhe->fnhe_rth_output);
685 		if (rt)
686 			fill_route_from_fnhe(rt, fnhe);
687 	} else {
688 		/* Randomize max depth to avoid some side channels attacks. */
689 		int max_depth = FNHE_RECLAIM_DEPTH +
690 				prandom_u32_max(FNHE_RECLAIM_DEPTH);
691 
692 		while (depth > max_depth) {
693 			fnhe_remove_oldest(hash);
694 			depth--;
695 		}
696 
697 		fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
698 		if (!fnhe)
699 			goto out_unlock;
700 
701 		fnhe->fnhe_next = hash->chain;
702 
703 		fnhe->fnhe_genid = genid;
704 		fnhe->fnhe_daddr = daddr;
705 		fnhe->fnhe_gw = gw;
706 		fnhe->fnhe_pmtu = pmtu;
707 		fnhe->fnhe_mtu_locked = lock;
708 		fnhe->fnhe_expires = max(1UL, expires);
709 
710 		rcu_assign_pointer(hash->chain, fnhe);
711 
712 		/* Exception created; mark the cached routes for the nexthop
713 		 * stale, so anyone caching it rechecks if this exception
714 		 * applies to them.
715 		 */
716 		rt = rcu_dereference(nhc->nhc_rth_input);
717 		if (rt)
718 			rt->dst.obsolete = DST_OBSOLETE_KILL;
719 
720 		for_each_possible_cpu(i) {
721 			struct rtable __rcu **prt;
722 
723 			prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
724 			rt = rcu_dereference(*prt);
725 			if (rt)
726 				rt->dst.obsolete = DST_OBSOLETE_KILL;
727 		}
728 	}
729 
730 	fnhe->fnhe_stamp = jiffies;
731 
732 out_unlock:
733 	spin_unlock_bh(&fnhe_lock);
734 }
735 
736 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
737 			     bool kill_route)
738 {
739 	__be32 new_gw = icmp_hdr(skb)->un.gateway;
740 	__be32 old_gw = ip_hdr(skb)->saddr;
741 	struct net_device *dev = skb->dev;
742 	struct in_device *in_dev;
743 	struct fib_result res;
744 	struct neighbour *n;
745 	struct net *net;
746 
747 	switch (icmp_hdr(skb)->code & 7) {
748 	case ICMP_REDIR_NET:
749 	case ICMP_REDIR_NETTOS:
750 	case ICMP_REDIR_HOST:
751 	case ICMP_REDIR_HOSTTOS:
752 		break;
753 
754 	default:
755 		return;
756 	}
757 
758 	if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
759 		return;
760 
761 	in_dev = __in_dev_get_rcu(dev);
762 	if (!in_dev)
763 		return;
764 
765 	net = dev_net(dev);
766 	if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
767 	    ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
768 	    ipv4_is_zeronet(new_gw))
769 		goto reject_redirect;
770 
771 	if (!IN_DEV_SHARED_MEDIA(in_dev)) {
772 		if (!inet_addr_onlink(in_dev, new_gw, old_gw))
773 			goto reject_redirect;
774 		if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
775 			goto reject_redirect;
776 	} else {
777 		if (inet_addr_type(net, new_gw) != RTN_UNICAST)
778 			goto reject_redirect;
779 	}
780 
781 	n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
782 	if (!n)
783 		n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
784 	if (!IS_ERR(n)) {
785 		if (!(n->nud_state & NUD_VALID)) {
786 			neigh_event_send(n, NULL);
787 		} else {
788 			if (fib_lookup(net, fl4, &res, 0) == 0) {
789 				struct fib_nh_common *nhc;
790 
791 				fib_select_path(net, &res, fl4, skb);
792 				nhc = FIB_RES_NHC(res);
793 				update_or_create_fnhe(nhc, fl4->daddr, new_gw,
794 						0, false,
795 						jiffies + ip_rt_gc_timeout);
796 			}
797 			if (kill_route)
798 				rt->dst.obsolete = DST_OBSOLETE_KILL;
799 			call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
800 		}
801 		neigh_release(n);
802 	}
803 	return;
804 
805 reject_redirect:
806 #ifdef CONFIG_IP_ROUTE_VERBOSE
807 	if (IN_DEV_LOG_MARTIANS(in_dev)) {
808 		const struct iphdr *iph = (const struct iphdr *) skb->data;
809 		__be32 daddr = iph->daddr;
810 		__be32 saddr = iph->saddr;
811 
812 		net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
813 				     "  Advised path = %pI4 -> %pI4\n",
814 				     &old_gw, dev->name, &new_gw,
815 				     &saddr, &daddr);
816 	}
817 #endif
818 	;
819 }
820 
821 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
822 {
823 	struct rtable *rt;
824 	struct flowi4 fl4;
825 	const struct iphdr *iph = (const struct iphdr *) skb->data;
826 	struct net *net = dev_net(skb->dev);
827 	int oif = skb->dev->ifindex;
828 	u8 tos = RT_TOS(iph->tos);
829 	u8 prot = iph->protocol;
830 	u32 mark = skb->mark;
831 
832 	rt = (struct rtable *) dst;
833 
834 	__build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
835 	ip_rt_fix_tos(&fl4);
836 	__ip_do_redirect(rt, skb, &fl4, true);
837 }
838 
839 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
840 {
841 	struct rtable *rt = (struct rtable *)dst;
842 	struct dst_entry *ret = dst;
843 
844 	if (rt) {
845 		if (dst->obsolete > 0) {
846 			ip_rt_put(rt);
847 			ret = NULL;
848 		} else if ((rt->rt_flags & RTCF_REDIRECTED) ||
849 			   rt->dst.expires) {
850 			ip_rt_put(rt);
851 			ret = NULL;
852 		}
853 	}
854 	return ret;
855 }
856 
857 /*
858  * Algorithm:
859  *	1. The first ip_rt_redirect_number redirects are sent
860  *	   with exponential backoff, then we stop sending them at all,
861  *	   assuming that the host ignores our redirects.
862  *	2. If we did not see packets requiring redirects
863  *	   during ip_rt_redirect_silence, we assume that the host
864  *	   forgot redirected route and start to send redirects again.
865  *
866  * This algorithm is much cheaper and more intelligent than dumb load limiting
867  * in icmp.c.
868  *
869  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
870  * and "frag. need" (breaks PMTU discovery) in icmp.c.
871  */
872 
873 void ip_rt_send_redirect(struct sk_buff *skb)
874 {
875 	struct rtable *rt = skb_rtable(skb);
876 	struct in_device *in_dev;
877 	struct inet_peer *peer;
878 	struct net *net;
879 	int log_martians;
880 	int vif;
881 
882 	rcu_read_lock();
883 	in_dev = __in_dev_get_rcu(rt->dst.dev);
884 	if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
885 		rcu_read_unlock();
886 		return;
887 	}
888 	log_martians = IN_DEV_LOG_MARTIANS(in_dev);
889 	vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
890 	rcu_read_unlock();
891 
892 	net = dev_net(rt->dst.dev);
893 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
894 	if (!peer) {
895 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
896 			  rt_nexthop(rt, ip_hdr(skb)->daddr));
897 		return;
898 	}
899 
900 	/* No redirected packets during ip_rt_redirect_silence;
901 	 * reset the algorithm.
902 	 */
903 	if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
904 		peer->rate_tokens = 0;
905 		peer->n_redirects = 0;
906 	}
907 
908 	/* Too many ignored redirects; do not send anything
909 	 * set dst.rate_last to the last seen redirected packet.
910 	 */
911 	if (peer->n_redirects >= ip_rt_redirect_number) {
912 		peer->rate_last = jiffies;
913 		goto out_put_peer;
914 	}
915 
916 	/* Check for load limit; set rate_last to the latest sent
917 	 * redirect.
918 	 */
919 	if (peer->n_redirects == 0 ||
920 	    time_after(jiffies,
921 		       (peer->rate_last +
922 			(ip_rt_redirect_load << peer->n_redirects)))) {
923 		__be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
924 
925 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
926 		peer->rate_last = jiffies;
927 		++peer->n_redirects;
928 #ifdef CONFIG_IP_ROUTE_VERBOSE
929 		if (log_martians &&
930 		    peer->n_redirects == ip_rt_redirect_number)
931 			net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
932 					     &ip_hdr(skb)->saddr, inet_iif(skb),
933 					     &ip_hdr(skb)->daddr, &gw);
934 #endif
935 	}
936 out_put_peer:
937 	inet_putpeer(peer);
938 }
939 
940 static int ip_error(struct sk_buff *skb)
941 {
942 	struct rtable *rt = skb_rtable(skb);
943 	struct net_device *dev = skb->dev;
944 	struct in_device *in_dev;
945 	struct inet_peer *peer;
946 	unsigned long now;
947 	struct net *net;
948 	bool send;
949 	int code;
950 
951 	if (netif_is_l3_master(skb->dev)) {
952 		dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
953 		if (!dev)
954 			goto out;
955 	}
956 
957 	in_dev = __in_dev_get_rcu(dev);
958 
959 	/* IP on this device is disabled. */
960 	if (!in_dev)
961 		goto out;
962 
963 	net = dev_net(rt->dst.dev);
964 	if (!IN_DEV_FORWARD(in_dev)) {
965 		switch (rt->dst.error) {
966 		case EHOSTUNREACH:
967 			__IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
968 			break;
969 
970 		case ENETUNREACH:
971 			__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
972 			break;
973 		}
974 		goto out;
975 	}
976 
977 	switch (rt->dst.error) {
978 	case EINVAL:
979 	default:
980 		goto out;
981 	case EHOSTUNREACH:
982 		code = ICMP_HOST_UNREACH;
983 		break;
984 	case ENETUNREACH:
985 		code = ICMP_NET_UNREACH;
986 		__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
987 		break;
988 	case EACCES:
989 		code = ICMP_PKT_FILTERED;
990 		break;
991 	}
992 
993 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
994 			       l3mdev_master_ifindex(skb->dev), 1);
995 
996 	send = true;
997 	if (peer) {
998 		now = jiffies;
999 		peer->rate_tokens += now - peer->rate_last;
1000 		if (peer->rate_tokens > ip_rt_error_burst)
1001 			peer->rate_tokens = ip_rt_error_burst;
1002 		peer->rate_last = now;
1003 		if (peer->rate_tokens >= ip_rt_error_cost)
1004 			peer->rate_tokens -= ip_rt_error_cost;
1005 		else
1006 			send = false;
1007 		inet_putpeer(peer);
1008 	}
1009 	if (send)
1010 		icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1011 
1012 out:	kfree_skb(skb);
1013 	return 0;
1014 }
1015 
1016 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1017 {
1018 	struct dst_entry *dst = &rt->dst;
1019 	struct net *net = dev_net(dst->dev);
1020 	struct fib_result res;
1021 	bool lock = false;
1022 	u32 old_mtu;
1023 
1024 	if (ip_mtu_locked(dst))
1025 		return;
1026 
1027 	old_mtu = ipv4_mtu(dst);
1028 	if (old_mtu < mtu)
1029 		return;
1030 
1031 	if (mtu < net->ipv4.ip_rt_min_pmtu) {
1032 		lock = true;
1033 		mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1034 	}
1035 
1036 	if (rt->rt_pmtu == mtu && !lock &&
1037 	    time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2))
1038 		return;
1039 
1040 	rcu_read_lock();
1041 	if (fib_lookup(net, fl4, &res, 0) == 0) {
1042 		struct fib_nh_common *nhc;
1043 
1044 		fib_select_path(net, &res, fl4, NULL);
1045 		nhc = FIB_RES_NHC(res);
1046 		update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1047 				      jiffies + net->ipv4.ip_rt_mtu_expires);
1048 	}
1049 	rcu_read_unlock();
1050 }
1051 
1052 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1053 			      struct sk_buff *skb, u32 mtu,
1054 			      bool confirm_neigh)
1055 {
1056 	struct rtable *rt = (struct rtable *) dst;
1057 	struct flowi4 fl4;
1058 
1059 	ip_rt_build_flow_key(&fl4, sk, skb);
1060 	ip_rt_fix_tos(&fl4);
1061 
1062 	/* Don't make lookup fail for bridged encapsulations */
1063 	if (skb && netif_is_any_bridge_port(skb->dev))
1064 		fl4.flowi4_oif = 0;
1065 
1066 	__ip_rt_update_pmtu(rt, &fl4, mtu);
1067 }
1068 
1069 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1070 		      int oif, u8 protocol)
1071 {
1072 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1073 	struct flowi4 fl4;
1074 	struct rtable *rt;
1075 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
1076 
1077 	__build_flow_key(net, &fl4, NULL, iph, oif,
1078 			 RT_TOS(iph->tos), protocol, mark, 0);
1079 	rt = __ip_route_output_key(net, &fl4);
1080 	if (!IS_ERR(rt)) {
1081 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1082 		ip_rt_put(rt);
1083 	}
1084 }
1085 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1086 
1087 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1088 {
1089 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1090 	struct flowi4 fl4;
1091 	struct rtable *rt;
1092 
1093 	__build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1094 
1095 	if (!fl4.flowi4_mark)
1096 		fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1097 
1098 	rt = __ip_route_output_key(sock_net(sk), &fl4);
1099 	if (!IS_ERR(rt)) {
1100 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1101 		ip_rt_put(rt);
1102 	}
1103 }
1104 
1105 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1106 {
1107 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1108 	struct flowi4 fl4;
1109 	struct rtable *rt;
1110 	struct dst_entry *odst = NULL;
1111 	bool new = false;
1112 	struct net *net = sock_net(sk);
1113 
1114 	bh_lock_sock(sk);
1115 
1116 	if (!ip_sk_accept_pmtu(sk))
1117 		goto out;
1118 
1119 	odst = sk_dst_get(sk);
1120 
1121 	if (sock_owned_by_user(sk) || !odst) {
1122 		__ipv4_sk_update_pmtu(skb, sk, mtu);
1123 		goto out;
1124 	}
1125 
1126 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1127 
1128 	rt = (struct rtable *)odst;
1129 	if (odst->obsolete && !odst->ops->check(odst, 0)) {
1130 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1131 		if (IS_ERR(rt))
1132 			goto out;
1133 
1134 		new = true;
1135 	} else {
1136 		ip_rt_fix_tos(&fl4);
1137 	}
1138 
1139 	__ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu);
1140 
1141 	if (!dst_check(&rt->dst, 0)) {
1142 		if (new)
1143 			dst_release(&rt->dst);
1144 
1145 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1146 		if (IS_ERR(rt))
1147 			goto out;
1148 
1149 		new = true;
1150 	}
1151 
1152 	if (new)
1153 		sk_dst_set(sk, &rt->dst);
1154 
1155 out:
1156 	bh_unlock_sock(sk);
1157 	dst_release(odst);
1158 }
1159 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1160 
1161 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1162 		   int oif, u8 protocol)
1163 {
1164 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1165 	struct flowi4 fl4;
1166 	struct rtable *rt;
1167 
1168 	__build_flow_key(net, &fl4, NULL, iph, oif,
1169 			 RT_TOS(iph->tos), protocol, 0, 0);
1170 	rt = __ip_route_output_key(net, &fl4);
1171 	if (!IS_ERR(rt)) {
1172 		__ip_do_redirect(rt, skb, &fl4, false);
1173 		ip_rt_put(rt);
1174 	}
1175 }
1176 EXPORT_SYMBOL_GPL(ipv4_redirect);
1177 
1178 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1179 {
1180 	const struct iphdr *iph = (const struct iphdr *)skb->data;
1181 	struct flowi4 fl4;
1182 	struct rtable *rt;
1183 	struct net *net = sock_net(sk);
1184 
1185 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1186 	rt = __ip_route_output_key(net, &fl4);
1187 	if (!IS_ERR(rt)) {
1188 		__ip_do_redirect(rt, skb, &fl4, false);
1189 		ip_rt_put(rt);
1190 	}
1191 }
1192 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1193 
1194 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1195 							 u32 cookie)
1196 {
1197 	struct rtable *rt = (struct rtable *) dst;
1198 
1199 	/* All IPV4 dsts are created with ->obsolete set to the value
1200 	 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1201 	 * into this function always.
1202 	 *
1203 	 * When a PMTU/redirect information update invalidates a route,
1204 	 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1205 	 * DST_OBSOLETE_DEAD.
1206 	 */
1207 	if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1208 		return NULL;
1209 	return dst;
1210 }
1211 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1212 
1213 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1214 {
1215 	struct ip_options opt;
1216 	int res;
1217 
1218 	/* Recompile ip options since IPCB may not be valid anymore.
1219 	 * Also check we have a reasonable ipv4 header.
1220 	 */
1221 	if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1222 	    ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1223 		return;
1224 
1225 	memset(&opt, 0, sizeof(opt));
1226 	if (ip_hdr(skb)->ihl > 5) {
1227 		if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1228 			return;
1229 		opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1230 
1231 		rcu_read_lock();
1232 		res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1233 		rcu_read_unlock();
1234 
1235 		if (res)
1236 			return;
1237 	}
1238 	__icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1239 }
1240 
1241 static void ipv4_link_failure(struct sk_buff *skb)
1242 {
1243 	struct rtable *rt;
1244 
1245 	ipv4_send_dest_unreach(skb);
1246 
1247 	rt = skb_rtable(skb);
1248 	if (rt)
1249 		dst_set_expires(&rt->dst, 0);
1250 }
1251 
1252 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1253 {
1254 	pr_debug("%s: %pI4 -> %pI4, %s\n",
1255 		 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1256 		 skb->dev ? skb->dev->name : "?");
1257 	kfree_skb(skb);
1258 	WARN_ON(1);
1259 	return 0;
1260 }
1261 
1262 /*
1263  * We do not cache source address of outgoing interface,
1264  * because it is used only by IP RR, TS and SRR options,
1265  * so that it out of fast path.
1266  *
1267  * BTW remember: "addr" is allowed to be not aligned
1268  * in IP options!
1269  */
1270 
1271 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1272 {
1273 	__be32 src;
1274 
1275 	if (rt_is_output_route(rt))
1276 		src = ip_hdr(skb)->saddr;
1277 	else {
1278 		struct fib_result res;
1279 		struct iphdr *iph = ip_hdr(skb);
1280 		struct flowi4 fl4 = {
1281 			.daddr = iph->daddr,
1282 			.saddr = iph->saddr,
1283 			.flowi4_tos = RT_TOS(iph->tos),
1284 			.flowi4_oif = rt->dst.dev->ifindex,
1285 			.flowi4_iif = skb->dev->ifindex,
1286 			.flowi4_mark = skb->mark,
1287 		};
1288 
1289 		rcu_read_lock();
1290 		if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1291 			src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1292 		else
1293 			src = inet_select_addr(rt->dst.dev,
1294 					       rt_nexthop(rt, iph->daddr),
1295 					       RT_SCOPE_UNIVERSE);
1296 		rcu_read_unlock();
1297 	}
1298 	memcpy(addr, &src, 4);
1299 }
1300 
1301 #ifdef CONFIG_IP_ROUTE_CLASSID
1302 static void set_class_tag(struct rtable *rt, u32 tag)
1303 {
1304 	if (!(rt->dst.tclassid & 0xFFFF))
1305 		rt->dst.tclassid |= tag & 0xFFFF;
1306 	if (!(rt->dst.tclassid & 0xFFFF0000))
1307 		rt->dst.tclassid |= tag & 0xFFFF0000;
1308 }
1309 #endif
1310 
1311 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1312 {
1313 	struct net *net = dev_net(dst->dev);
1314 	unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1315 	unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1316 				    net->ipv4.ip_rt_min_advmss);
1317 
1318 	return min(advmss, IPV4_MAX_PMTU - header_size);
1319 }
1320 
1321 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1322 {
1323 	return ip_dst_mtu_maybe_forward(dst, false);
1324 }
1325 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1326 
1327 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1328 {
1329 	struct fnhe_hash_bucket *hash;
1330 	struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1331 	u32 hval = fnhe_hashfun(daddr);
1332 
1333 	spin_lock_bh(&fnhe_lock);
1334 
1335 	hash = rcu_dereference_protected(nhc->nhc_exceptions,
1336 					 lockdep_is_held(&fnhe_lock));
1337 	hash += hval;
1338 
1339 	fnhe_p = &hash->chain;
1340 	fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1341 	while (fnhe) {
1342 		if (fnhe->fnhe_daddr == daddr) {
1343 			rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1344 				fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1345 			/* set fnhe_daddr to 0 to ensure it won't bind with
1346 			 * new dsts in rt_bind_exception().
1347 			 */
1348 			fnhe->fnhe_daddr = 0;
1349 			fnhe_flush_routes(fnhe);
1350 			kfree_rcu(fnhe, rcu);
1351 			break;
1352 		}
1353 		fnhe_p = &fnhe->fnhe_next;
1354 		fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1355 						 lockdep_is_held(&fnhe_lock));
1356 	}
1357 
1358 	spin_unlock_bh(&fnhe_lock);
1359 }
1360 
1361 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1362 					       __be32 daddr)
1363 {
1364 	struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1365 	struct fib_nh_exception *fnhe;
1366 	u32 hval;
1367 
1368 	if (!hash)
1369 		return NULL;
1370 
1371 	hval = fnhe_hashfun(daddr);
1372 
1373 	for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1374 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
1375 		if (fnhe->fnhe_daddr == daddr) {
1376 			if (fnhe->fnhe_expires &&
1377 			    time_after(jiffies, fnhe->fnhe_expires)) {
1378 				ip_del_fnhe(nhc, daddr);
1379 				break;
1380 			}
1381 			return fnhe;
1382 		}
1383 	}
1384 	return NULL;
1385 }
1386 
1387 /* MTU selection:
1388  * 1. mtu on route is locked - use it
1389  * 2. mtu from nexthop exception
1390  * 3. mtu from egress device
1391  */
1392 
1393 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1394 {
1395 	struct fib_nh_common *nhc = res->nhc;
1396 	struct net_device *dev = nhc->nhc_dev;
1397 	struct fib_info *fi = res->fi;
1398 	u32 mtu = 0;
1399 
1400 	if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu ||
1401 	    fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1402 		mtu = fi->fib_mtu;
1403 
1404 	if (likely(!mtu)) {
1405 		struct fib_nh_exception *fnhe;
1406 
1407 		fnhe = find_exception(nhc, daddr);
1408 		if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1409 			mtu = fnhe->fnhe_pmtu;
1410 	}
1411 
1412 	if (likely(!mtu))
1413 		mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1414 
1415 	return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1416 }
1417 
1418 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1419 			      __be32 daddr, const bool do_cache)
1420 {
1421 	bool ret = false;
1422 
1423 	spin_lock_bh(&fnhe_lock);
1424 
1425 	if (daddr == fnhe->fnhe_daddr) {
1426 		struct rtable __rcu **porig;
1427 		struct rtable *orig;
1428 		int genid = fnhe_genid(dev_net(rt->dst.dev));
1429 
1430 		if (rt_is_input_route(rt))
1431 			porig = &fnhe->fnhe_rth_input;
1432 		else
1433 			porig = &fnhe->fnhe_rth_output;
1434 		orig = rcu_dereference(*porig);
1435 
1436 		if (fnhe->fnhe_genid != genid) {
1437 			fnhe->fnhe_genid = genid;
1438 			fnhe->fnhe_gw = 0;
1439 			fnhe->fnhe_pmtu = 0;
1440 			fnhe->fnhe_expires = 0;
1441 			fnhe->fnhe_mtu_locked = false;
1442 			fnhe_flush_routes(fnhe);
1443 			orig = NULL;
1444 		}
1445 		fill_route_from_fnhe(rt, fnhe);
1446 		if (!rt->rt_gw4) {
1447 			rt->rt_gw4 = daddr;
1448 			rt->rt_gw_family = AF_INET;
1449 		}
1450 
1451 		if (do_cache) {
1452 			dst_hold(&rt->dst);
1453 			rcu_assign_pointer(*porig, rt);
1454 			if (orig) {
1455 				dst_dev_put(&orig->dst);
1456 				dst_release(&orig->dst);
1457 			}
1458 			ret = true;
1459 		}
1460 
1461 		fnhe->fnhe_stamp = jiffies;
1462 	}
1463 	spin_unlock_bh(&fnhe_lock);
1464 
1465 	return ret;
1466 }
1467 
1468 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1469 {
1470 	struct rtable *orig, *prev, **p;
1471 	bool ret = true;
1472 
1473 	if (rt_is_input_route(rt)) {
1474 		p = (struct rtable **)&nhc->nhc_rth_input;
1475 	} else {
1476 		p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1477 	}
1478 	orig = *p;
1479 
1480 	/* hold dst before doing cmpxchg() to avoid race condition
1481 	 * on this dst
1482 	 */
1483 	dst_hold(&rt->dst);
1484 	prev = cmpxchg(p, orig, rt);
1485 	if (prev == orig) {
1486 		if (orig) {
1487 			rt_add_uncached_list(orig);
1488 			dst_release(&orig->dst);
1489 		}
1490 	} else {
1491 		dst_release(&rt->dst);
1492 		ret = false;
1493 	}
1494 
1495 	return ret;
1496 }
1497 
1498 struct uncached_list {
1499 	spinlock_t		lock;
1500 	struct list_head	head;
1501 	struct list_head	quarantine;
1502 };
1503 
1504 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1505 
1506 void rt_add_uncached_list(struct rtable *rt)
1507 {
1508 	struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1509 
1510 	rt->rt_uncached_list = ul;
1511 
1512 	spin_lock_bh(&ul->lock);
1513 	list_add_tail(&rt->rt_uncached, &ul->head);
1514 	spin_unlock_bh(&ul->lock);
1515 }
1516 
1517 void rt_del_uncached_list(struct rtable *rt)
1518 {
1519 	if (!list_empty(&rt->rt_uncached)) {
1520 		struct uncached_list *ul = rt->rt_uncached_list;
1521 
1522 		spin_lock_bh(&ul->lock);
1523 		list_del_init(&rt->rt_uncached);
1524 		spin_unlock_bh(&ul->lock);
1525 	}
1526 }
1527 
1528 static void ipv4_dst_destroy(struct dst_entry *dst)
1529 {
1530 	struct rtable *rt = (struct rtable *)dst;
1531 
1532 	ip_dst_metrics_put(dst);
1533 	rt_del_uncached_list(rt);
1534 }
1535 
1536 void rt_flush_dev(struct net_device *dev)
1537 {
1538 	struct rtable *rt, *safe;
1539 	int cpu;
1540 
1541 	for_each_possible_cpu(cpu) {
1542 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1543 
1544 		if (list_empty(&ul->head))
1545 			continue;
1546 
1547 		spin_lock_bh(&ul->lock);
1548 		list_for_each_entry_safe(rt, safe, &ul->head, rt_uncached) {
1549 			if (rt->dst.dev != dev)
1550 				continue;
1551 			rt->dst.dev = blackhole_netdev;
1552 			dev_replace_track(dev, blackhole_netdev,
1553 					  &rt->dst.dev_tracker,
1554 					  GFP_ATOMIC);
1555 			list_move(&rt->rt_uncached, &ul->quarantine);
1556 		}
1557 		spin_unlock_bh(&ul->lock);
1558 	}
1559 }
1560 
1561 static bool rt_cache_valid(const struct rtable *rt)
1562 {
1563 	return	rt &&
1564 		rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1565 		!rt_is_expired(rt);
1566 }
1567 
1568 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1569 			   const struct fib_result *res,
1570 			   struct fib_nh_exception *fnhe,
1571 			   struct fib_info *fi, u16 type, u32 itag,
1572 			   const bool do_cache)
1573 {
1574 	bool cached = false;
1575 
1576 	if (fi) {
1577 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1578 
1579 		if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1580 			rt->rt_uses_gateway = 1;
1581 			rt->rt_gw_family = nhc->nhc_gw_family;
1582 			/* only INET and INET6 are supported */
1583 			if (likely(nhc->nhc_gw_family == AF_INET))
1584 				rt->rt_gw4 = nhc->nhc_gw.ipv4;
1585 			else
1586 				rt->rt_gw6 = nhc->nhc_gw.ipv6;
1587 		}
1588 
1589 		ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1590 
1591 #ifdef CONFIG_IP_ROUTE_CLASSID
1592 		if (nhc->nhc_family == AF_INET) {
1593 			struct fib_nh *nh;
1594 
1595 			nh = container_of(nhc, struct fib_nh, nh_common);
1596 			rt->dst.tclassid = nh->nh_tclassid;
1597 		}
1598 #endif
1599 		rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1600 		if (unlikely(fnhe))
1601 			cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1602 		else if (do_cache)
1603 			cached = rt_cache_route(nhc, rt);
1604 		if (unlikely(!cached)) {
1605 			/* Routes we intend to cache in nexthop exception or
1606 			 * FIB nexthop have the DST_NOCACHE bit clear.
1607 			 * However, if we are unsuccessful at storing this
1608 			 * route into the cache we really need to set it.
1609 			 */
1610 			if (!rt->rt_gw4) {
1611 				rt->rt_gw_family = AF_INET;
1612 				rt->rt_gw4 = daddr;
1613 			}
1614 			rt_add_uncached_list(rt);
1615 		}
1616 	} else
1617 		rt_add_uncached_list(rt);
1618 
1619 #ifdef CONFIG_IP_ROUTE_CLASSID
1620 #ifdef CONFIG_IP_MULTIPLE_TABLES
1621 	set_class_tag(rt, res->tclassid);
1622 #endif
1623 	set_class_tag(rt, itag);
1624 #endif
1625 }
1626 
1627 struct rtable *rt_dst_alloc(struct net_device *dev,
1628 			    unsigned int flags, u16 type,
1629 			    bool nopolicy, bool noxfrm)
1630 {
1631 	struct rtable *rt;
1632 
1633 	rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1634 		       (nopolicy ? DST_NOPOLICY : 0) |
1635 		       (noxfrm ? DST_NOXFRM : 0));
1636 
1637 	if (rt) {
1638 		rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1639 		rt->rt_flags = flags;
1640 		rt->rt_type = type;
1641 		rt->rt_is_input = 0;
1642 		rt->rt_iif = 0;
1643 		rt->rt_pmtu = 0;
1644 		rt->rt_mtu_locked = 0;
1645 		rt->rt_uses_gateway = 0;
1646 		rt->rt_gw_family = 0;
1647 		rt->rt_gw4 = 0;
1648 		INIT_LIST_HEAD(&rt->rt_uncached);
1649 
1650 		rt->dst.output = ip_output;
1651 		if (flags & RTCF_LOCAL)
1652 			rt->dst.input = ip_local_deliver;
1653 	}
1654 
1655 	return rt;
1656 }
1657 EXPORT_SYMBOL(rt_dst_alloc);
1658 
1659 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1660 {
1661 	struct rtable *new_rt;
1662 
1663 	new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1664 			   rt->dst.flags);
1665 
1666 	if (new_rt) {
1667 		new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1668 		new_rt->rt_flags = rt->rt_flags;
1669 		new_rt->rt_type = rt->rt_type;
1670 		new_rt->rt_is_input = rt->rt_is_input;
1671 		new_rt->rt_iif = rt->rt_iif;
1672 		new_rt->rt_pmtu = rt->rt_pmtu;
1673 		new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1674 		new_rt->rt_gw_family = rt->rt_gw_family;
1675 		if (rt->rt_gw_family == AF_INET)
1676 			new_rt->rt_gw4 = rt->rt_gw4;
1677 		else if (rt->rt_gw_family == AF_INET6)
1678 			new_rt->rt_gw6 = rt->rt_gw6;
1679 		INIT_LIST_HEAD(&new_rt->rt_uncached);
1680 
1681 		new_rt->dst.input = rt->dst.input;
1682 		new_rt->dst.output = rt->dst.output;
1683 		new_rt->dst.error = rt->dst.error;
1684 		new_rt->dst.lastuse = jiffies;
1685 		new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1686 	}
1687 	return new_rt;
1688 }
1689 EXPORT_SYMBOL(rt_dst_clone);
1690 
1691 /* called in rcu_read_lock() section */
1692 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1693 			  u8 tos, struct net_device *dev,
1694 			  struct in_device *in_dev, u32 *itag)
1695 {
1696 	int err;
1697 
1698 	/* Primary sanity checks. */
1699 	if (!in_dev)
1700 		return -EINVAL;
1701 
1702 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1703 	    skb->protocol != htons(ETH_P_IP))
1704 		return -EINVAL;
1705 
1706 	if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1707 		return -EINVAL;
1708 
1709 	if (ipv4_is_zeronet(saddr)) {
1710 		if (!ipv4_is_local_multicast(daddr) &&
1711 		    ip_hdr(skb)->protocol != IPPROTO_IGMP)
1712 			return -EINVAL;
1713 	} else {
1714 		err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1715 					  in_dev, itag);
1716 		if (err < 0)
1717 			return err;
1718 	}
1719 	return 0;
1720 }
1721 
1722 /* called in rcu_read_lock() section */
1723 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1724 			     u8 tos, struct net_device *dev, int our)
1725 {
1726 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1727 	unsigned int flags = RTCF_MULTICAST;
1728 	struct rtable *rth;
1729 	bool no_policy;
1730 	u32 itag = 0;
1731 	int err;
1732 
1733 	err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1734 	if (err)
1735 		return err;
1736 
1737 	if (our)
1738 		flags |= RTCF_LOCAL;
1739 
1740 	no_policy = IN_DEV_ORCONF(in_dev, NOPOLICY);
1741 	if (no_policy)
1742 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
1743 
1744 	rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1745 			   no_policy, false);
1746 	if (!rth)
1747 		return -ENOBUFS;
1748 
1749 #ifdef CONFIG_IP_ROUTE_CLASSID
1750 	rth->dst.tclassid = itag;
1751 #endif
1752 	rth->dst.output = ip_rt_bug;
1753 	rth->rt_is_input= 1;
1754 
1755 #ifdef CONFIG_IP_MROUTE
1756 	if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1757 		rth->dst.input = ip_mr_input;
1758 #endif
1759 	RT_CACHE_STAT_INC(in_slow_mc);
1760 
1761 	skb_dst_drop(skb);
1762 	skb_dst_set(skb, &rth->dst);
1763 	return 0;
1764 }
1765 
1766 
1767 static void ip_handle_martian_source(struct net_device *dev,
1768 				     struct in_device *in_dev,
1769 				     struct sk_buff *skb,
1770 				     __be32 daddr,
1771 				     __be32 saddr)
1772 {
1773 	RT_CACHE_STAT_INC(in_martian_src);
1774 #ifdef CONFIG_IP_ROUTE_VERBOSE
1775 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1776 		/*
1777 		 *	RFC1812 recommendation, if source is martian,
1778 		 *	the only hint is MAC header.
1779 		 */
1780 		pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1781 			&daddr, &saddr, dev->name);
1782 		if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1783 			print_hex_dump(KERN_WARNING, "ll header: ",
1784 				       DUMP_PREFIX_OFFSET, 16, 1,
1785 				       skb_mac_header(skb),
1786 				       dev->hard_header_len, false);
1787 		}
1788 	}
1789 #endif
1790 }
1791 
1792 /* called in rcu_read_lock() section */
1793 static int __mkroute_input(struct sk_buff *skb,
1794 			   const struct fib_result *res,
1795 			   struct in_device *in_dev,
1796 			   __be32 daddr, __be32 saddr, u32 tos)
1797 {
1798 	struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1799 	struct net_device *dev = nhc->nhc_dev;
1800 	struct fib_nh_exception *fnhe;
1801 	struct rtable *rth;
1802 	int err;
1803 	struct in_device *out_dev;
1804 	bool do_cache, no_policy;
1805 	u32 itag = 0;
1806 
1807 	/* get a working reference to the output device */
1808 	out_dev = __in_dev_get_rcu(dev);
1809 	if (!out_dev) {
1810 		net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1811 		return -EINVAL;
1812 	}
1813 
1814 	err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1815 				  in_dev->dev, in_dev, &itag);
1816 	if (err < 0) {
1817 		ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1818 					 saddr);
1819 
1820 		goto cleanup;
1821 	}
1822 
1823 	do_cache = res->fi && !itag;
1824 	if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1825 	    skb->protocol == htons(ETH_P_IP)) {
1826 		__be32 gw;
1827 
1828 		gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1829 		if (IN_DEV_SHARED_MEDIA(out_dev) ||
1830 		    inet_addr_onlink(out_dev, saddr, gw))
1831 			IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1832 	}
1833 
1834 	if (skb->protocol != htons(ETH_P_IP)) {
1835 		/* Not IP (i.e. ARP). Do not create route, if it is
1836 		 * invalid for proxy arp. DNAT routes are always valid.
1837 		 *
1838 		 * Proxy arp feature have been extended to allow, ARP
1839 		 * replies back to the same interface, to support
1840 		 * Private VLAN switch technologies. See arp.c.
1841 		 */
1842 		if (out_dev == in_dev &&
1843 		    IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1844 			err = -EINVAL;
1845 			goto cleanup;
1846 		}
1847 	}
1848 
1849 	no_policy = IN_DEV_ORCONF(in_dev, NOPOLICY);
1850 	if (no_policy)
1851 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
1852 
1853 	fnhe = find_exception(nhc, daddr);
1854 	if (do_cache) {
1855 		if (fnhe)
1856 			rth = rcu_dereference(fnhe->fnhe_rth_input);
1857 		else
1858 			rth = rcu_dereference(nhc->nhc_rth_input);
1859 		if (rt_cache_valid(rth)) {
1860 			skb_dst_set_noref(skb, &rth->dst);
1861 			goto out;
1862 		}
1863 	}
1864 
1865 	rth = rt_dst_alloc(out_dev->dev, 0, res->type, no_policy,
1866 			   IN_DEV_ORCONF(out_dev, NOXFRM));
1867 	if (!rth) {
1868 		err = -ENOBUFS;
1869 		goto cleanup;
1870 	}
1871 
1872 	rth->rt_is_input = 1;
1873 	RT_CACHE_STAT_INC(in_slow_tot);
1874 
1875 	rth->dst.input = ip_forward;
1876 
1877 	rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1878 		       do_cache);
1879 	lwtunnel_set_redirect(&rth->dst);
1880 	skb_dst_set(skb, &rth->dst);
1881 out:
1882 	err = 0;
1883  cleanup:
1884 	return err;
1885 }
1886 
1887 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1888 /* To make ICMP packets follow the right flow, the multipath hash is
1889  * calculated from the inner IP addresses.
1890  */
1891 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1892 				 struct flow_keys *hash_keys)
1893 {
1894 	const struct iphdr *outer_iph = ip_hdr(skb);
1895 	const struct iphdr *key_iph = outer_iph;
1896 	const struct iphdr *inner_iph;
1897 	const struct icmphdr *icmph;
1898 	struct iphdr _inner_iph;
1899 	struct icmphdr _icmph;
1900 
1901 	if (likely(outer_iph->protocol != IPPROTO_ICMP))
1902 		goto out;
1903 
1904 	if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1905 		goto out;
1906 
1907 	icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1908 				   &_icmph);
1909 	if (!icmph)
1910 		goto out;
1911 
1912 	if (!icmp_is_err(icmph->type))
1913 		goto out;
1914 
1915 	inner_iph = skb_header_pointer(skb,
1916 				       outer_iph->ihl * 4 + sizeof(_icmph),
1917 				       sizeof(_inner_iph), &_inner_iph);
1918 	if (!inner_iph)
1919 		goto out;
1920 
1921 	key_iph = inner_iph;
1922 out:
1923 	hash_keys->addrs.v4addrs.src = key_iph->saddr;
1924 	hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1925 }
1926 
1927 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1928 					   const struct sk_buff *skb,
1929 					   bool *p_has_inner)
1930 {
1931 	u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
1932 	struct flow_keys keys, hash_keys;
1933 
1934 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1935 		return 0;
1936 
1937 	memset(&hash_keys, 0, sizeof(hash_keys));
1938 	skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1939 
1940 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1941 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1942 		hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1943 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1944 		hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1945 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1946 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
1947 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1948 		hash_keys.ports.src = keys.ports.src;
1949 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1950 		hash_keys.ports.dst = keys.ports.dst;
1951 
1952 	*p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1953 	return flow_hash_from_keys(&hash_keys);
1954 }
1955 
1956 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1957 					   const struct sk_buff *skb,
1958 					   bool has_inner)
1959 {
1960 	u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
1961 	struct flow_keys keys, hash_keys;
1962 
1963 	/* We assume the packet carries an encapsulation, but if none was
1964 	 * encountered during dissection of the outer flow, then there is no
1965 	 * point in calling the flow dissector again.
1966 	 */
1967 	if (!has_inner)
1968 		return 0;
1969 
1970 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1971 		return 0;
1972 
1973 	memset(&hash_keys, 0, sizeof(hash_keys));
1974 	skb_flow_dissect_flow_keys(skb, &keys, 0);
1975 
1976 	if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1977 		return 0;
1978 
1979 	if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1980 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1981 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1982 			hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1983 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1984 			hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1985 	} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1986 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1987 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1988 			hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1989 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1990 			hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1991 		if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1992 			hash_keys.tags.flow_label = keys.tags.flow_label;
1993 	}
1994 
1995 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1996 		hash_keys.basic.ip_proto = keys.basic.ip_proto;
1997 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1998 		hash_keys.ports.src = keys.ports.src;
1999 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2000 		hash_keys.ports.dst = keys.ports.dst;
2001 
2002 	return flow_hash_from_keys(&hash_keys);
2003 }
2004 
2005 static u32 fib_multipath_custom_hash_skb(const struct net *net,
2006 					 const struct sk_buff *skb)
2007 {
2008 	u32 mhash, mhash_inner;
2009 	bool has_inner = true;
2010 
2011 	mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2012 	mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2013 
2014 	return jhash_2words(mhash, mhash_inner, 0);
2015 }
2016 
2017 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2018 					 const struct flowi4 *fl4)
2019 {
2020 	u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
2021 	struct flow_keys hash_keys;
2022 
2023 	if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2024 		return 0;
2025 
2026 	memset(&hash_keys, 0, sizeof(hash_keys));
2027 	hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2028 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2029 		hash_keys.addrs.v4addrs.src = fl4->saddr;
2030 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2031 		hash_keys.addrs.v4addrs.dst = fl4->daddr;
2032 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2033 		hash_keys.basic.ip_proto = fl4->flowi4_proto;
2034 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2035 		hash_keys.ports.src = fl4->fl4_sport;
2036 	if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2037 		hash_keys.ports.dst = fl4->fl4_dport;
2038 
2039 	return flow_hash_from_keys(&hash_keys);
2040 }
2041 
2042 /* if skb is set it will be used and fl4 can be NULL */
2043 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2044 		       const struct sk_buff *skb, struct flow_keys *flkeys)
2045 {
2046 	u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2047 	struct flow_keys hash_keys;
2048 	u32 mhash = 0;
2049 
2050 	switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
2051 	case 0:
2052 		memset(&hash_keys, 0, sizeof(hash_keys));
2053 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2054 		if (skb) {
2055 			ip_multipath_l3_keys(skb, &hash_keys);
2056 		} else {
2057 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2058 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2059 		}
2060 		mhash = flow_hash_from_keys(&hash_keys);
2061 		break;
2062 	case 1:
2063 		/* skb is currently provided only when forwarding */
2064 		if (skb) {
2065 			unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2066 			struct flow_keys keys;
2067 
2068 			/* short-circuit if we already have L4 hash present */
2069 			if (skb->l4_hash)
2070 				return skb_get_hash_raw(skb) >> 1;
2071 
2072 			memset(&hash_keys, 0, sizeof(hash_keys));
2073 
2074 			if (!flkeys) {
2075 				skb_flow_dissect_flow_keys(skb, &keys, flag);
2076 				flkeys = &keys;
2077 			}
2078 
2079 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2080 			hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2081 			hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2082 			hash_keys.ports.src = flkeys->ports.src;
2083 			hash_keys.ports.dst = flkeys->ports.dst;
2084 			hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2085 		} else {
2086 			memset(&hash_keys, 0, sizeof(hash_keys));
2087 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2088 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2089 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2090 			hash_keys.ports.src = fl4->fl4_sport;
2091 			hash_keys.ports.dst = fl4->fl4_dport;
2092 			hash_keys.basic.ip_proto = fl4->flowi4_proto;
2093 		}
2094 		mhash = flow_hash_from_keys(&hash_keys);
2095 		break;
2096 	case 2:
2097 		memset(&hash_keys, 0, sizeof(hash_keys));
2098 		/* skb is currently provided only when forwarding */
2099 		if (skb) {
2100 			struct flow_keys keys;
2101 
2102 			skb_flow_dissect_flow_keys(skb, &keys, 0);
2103 			/* Inner can be v4 or v6 */
2104 			if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2105 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2106 				hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2107 				hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2108 			} else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2109 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2110 				hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2111 				hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2112 				hash_keys.tags.flow_label = keys.tags.flow_label;
2113 				hash_keys.basic.ip_proto = keys.basic.ip_proto;
2114 			} else {
2115 				/* Same as case 0 */
2116 				hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2117 				ip_multipath_l3_keys(skb, &hash_keys);
2118 			}
2119 		} else {
2120 			/* Same as case 0 */
2121 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2122 			hash_keys.addrs.v4addrs.src = fl4->saddr;
2123 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
2124 		}
2125 		mhash = flow_hash_from_keys(&hash_keys);
2126 		break;
2127 	case 3:
2128 		if (skb)
2129 			mhash = fib_multipath_custom_hash_skb(net, skb);
2130 		else
2131 			mhash = fib_multipath_custom_hash_fl4(net, fl4);
2132 		break;
2133 	}
2134 
2135 	if (multipath_hash)
2136 		mhash = jhash_2words(mhash, multipath_hash, 0);
2137 
2138 	return mhash >> 1;
2139 }
2140 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2141 
2142 static int ip_mkroute_input(struct sk_buff *skb,
2143 			    struct fib_result *res,
2144 			    struct in_device *in_dev,
2145 			    __be32 daddr, __be32 saddr, u32 tos,
2146 			    struct flow_keys *hkeys)
2147 {
2148 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2149 	if (res->fi && fib_info_num_path(res->fi) > 1) {
2150 		int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2151 
2152 		fib_select_multipath(res, h);
2153 	}
2154 #endif
2155 
2156 	/* create a routing cache entry */
2157 	return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2158 }
2159 
2160 /* Implements all the saddr-related checks as ip_route_input_slow(),
2161  * assuming daddr is valid and the destination is not a local broadcast one.
2162  * Uses the provided hint instead of performing a route lookup.
2163  */
2164 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2165 		      u8 tos, struct net_device *dev,
2166 		      const struct sk_buff *hint)
2167 {
2168 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2169 	struct rtable *rt = skb_rtable(hint);
2170 	struct net *net = dev_net(dev);
2171 	int err = -EINVAL;
2172 	u32 tag = 0;
2173 
2174 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2175 		goto martian_source;
2176 
2177 	if (ipv4_is_zeronet(saddr))
2178 		goto martian_source;
2179 
2180 	if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2181 		goto martian_source;
2182 
2183 	if (rt->rt_type != RTN_LOCAL)
2184 		goto skip_validate_source;
2185 
2186 	tos &= IPTOS_RT_MASK;
2187 	err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2188 	if (err < 0)
2189 		goto martian_source;
2190 
2191 skip_validate_source:
2192 	skb_dst_copy(skb, hint);
2193 	return 0;
2194 
2195 martian_source:
2196 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2197 	return err;
2198 }
2199 
2200 /* get device for dst_alloc with local routes */
2201 static struct net_device *ip_rt_get_dev(struct net *net,
2202 					const struct fib_result *res)
2203 {
2204 	struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2205 	struct net_device *dev = NULL;
2206 
2207 	if (nhc)
2208 		dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2209 
2210 	return dev ? : net->loopback_dev;
2211 }
2212 
2213 /*
2214  *	NOTE. We drop all the packets that has local source
2215  *	addresses, because every properly looped back packet
2216  *	must have correct destination already attached by output routine.
2217  *	Changes in the enforced policies must be applied also to
2218  *	ip_route_use_hint().
2219  *
2220  *	Such approach solves two big problems:
2221  *	1. Not simplex devices are handled properly.
2222  *	2. IP spoofing attempts are filtered with 100% of guarantee.
2223  *	called with rcu_read_lock()
2224  */
2225 
2226 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2227 			       u8 tos, struct net_device *dev,
2228 			       struct fib_result *res)
2229 {
2230 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2231 	struct flow_keys *flkeys = NULL, _flkeys;
2232 	struct net    *net = dev_net(dev);
2233 	struct ip_tunnel_info *tun_info;
2234 	int		err = -EINVAL;
2235 	unsigned int	flags = 0;
2236 	u32		itag = 0;
2237 	struct rtable	*rth;
2238 	struct flowi4	fl4;
2239 	bool do_cache = true;
2240 	bool no_policy;
2241 
2242 	/* IP on this device is disabled. */
2243 
2244 	if (!in_dev)
2245 		goto out;
2246 
2247 	/* Check for the most weird martians, which can be not detected
2248 	 * by fib_lookup.
2249 	 */
2250 
2251 	tun_info = skb_tunnel_info(skb);
2252 	if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2253 		fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2254 	else
2255 		fl4.flowi4_tun_key.tun_id = 0;
2256 	skb_dst_drop(skb);
2257 
2258 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2259 		goto martian_source;
2260 
2261 	res->fi = NULL;
2262 	res->table = NULL;
2263 	if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2264 		goto brd_input;
2265 
2266 	/* Accept zero addresses only to limited broadcast;
2267 	 * I even do not know to fix it or not. Waiting for complains :-)
2268 	 */
2269 	if (ipv4_is_zeronet(saddr))
2270 		goto martian_source;
2271 
2272 	if (ipv4_is_zeronet(daddr))
2273 		goto martian_destination;
2274 
2275 	/* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2276 	 * and call it once if daddr or/and saddr are loopback addresses
2277 	 */
2278 	if (ipv4_is_loopback(daddr)) {
2279 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2280 			goto martian_destination;
2281 	} else if (ipv4_is_loopback(saddr)) {
2282 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2283 			goto martian_source;
2284 	}
2285 
2286 	/*
2287 	 *	Now we are ready to route packet.
2288 	 */
2289 	fl4.flowi4_l3mdev = 0;
2290 	fl4.flowi4_oif = 0;
2291 	fl4.flowi4_iif = dev->ifindex;
2292 	fl4.flowi4_mark = skb->mark;
2293 	fl4.flowi4_tos = tos;
2294 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2295 	fl4.flowi4_flags = 0;
2296 	fl4.daddr = daddr;
2297 	fl4.saddr = saddr;
2298 	fl4.flowi4_uid = sock_net_uid(net, NULL);
2299 	fl4.flowi4_multipath_hash = 0;
2300 
2301 	if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2302 		flkeys = &_flkeys;
2303 	} else {
2304 		fl4.flowi4_proto = 0;
2305 		fl4.fl4_sport = 0;
2306 		fl4.fl4_dport = 0;
2307 	}
2308 
2309 	err = fib_lookup(net, &fl4, res, 0);
2310 	if (err != 0) {
2311 		if (!IN_DEV_FORWARD(in_dev))
2312 			err = -EHOSTUNREACH;
2313 		goto no_route;
2314 	}
2315 
2316 	if (res->type == RTN_BROADCAST) {
2317 		if (IN_DEV_BFORWARD(in_dev))
2318 			goto make_route;
2319 		/* not do cache if bc_forwarding is enabled */
2320 		if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2321 			do_cache = false;
2322 		goto brd_input;
2323 	}
2324 
2325 	if (res->type == RTN_LOCAL) {
2326 		err = fib_validate_source(skb, saddr, daddr, tos,
2327 					  0, dev, in_dev, &itag);
2328 		if (err < 0)
2329 			goto martian_source;
2330 		goto local_input;
2331 	}
2332 
2333 	if (!IN_DEV_FORWARD(in_dev)) {
2334 		err = -EHOSTUNREACH;
2335 		goto no_route;
2336 	}
2337 	if (res->type != RTN_UNICAST)
2338 		goto martian_destination;
2339 
2340 make_route:
2341 	err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2342 out:	return err;
2343 
2344 brd_input:
2345 	if (skb->protocol != htons(ETH_P_IP))
2346 		goto e_inval;
2347 
2348 	if (!ipv4_is_zeronet(saddr)) {
2349 		err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2350 					  in_dev, &itag);
2351 		if (err < 0)
2352 			goto martian_source;
2353 	}
2354 	flags |= RTCF_BROADCAST;
2355 	res->type = RTN_BROADCAST;
2356 	RT_CACHE_STAT_INC(in_brd);
2357 
2358 local_input:
2359 	no_policy = IN_DEV_ORCONF(in_dev, NOPOLICY);
2360 	if (no_policy)
2361 		IPCB(skb)->flags |= IPSKB_NOPOLICY;
2362 
2363 	do_cache &= res->fi && !itag;
2364 	if (do_cache) {
2365 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2366 
2367 		rth = rcu_dereference(nhc->nhc_rth_input);
2368 		if (rt_cache_valid(rth)) {
2369 			skb_dst_set_noref(skb, &rth->dst);
2370 			err = 0;
2371 			goto out;
2372 		}
2373 	}
2374 
2375 	rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2376 			   flags | RTCF_LOCAL, res->type,
2377 			   no_policy, false);
2378 	if (!rth)
2379 		goto e_nobufs;
2380 
2381 	rth->dst.output= ip_rt_bug;
2382 #ifdef CONFIG_IP_ROUTE_CLASSID
2383 	rth->dst.tclassid = itag;
2384 #endif
2385 	rth->rt_is_input = 1;
2386 
2387 	RT_CACHE_STAT_INC(in_slow_tot);
2388 	if (res->type == RTN_UNREACHABLE) {
2389 		rth->dst.input= ip_error;
2390 		rth->dst.error= -err;
2391 		rth->rt_flags	&= ~RTCF_LOCAL;
2392 	}
2393 
2394 	if (do_cache) {
2395 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2396 
2397 		rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2398 		if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2399 			WARN_ON(rth->dst.input == lwtunnel_input);
2400 			rth->dst.lwtstate->orig_input = rth->dst.input;
2401 			rth->dst.input = lwtunnel_input;
2402 		}
2403 
2404 		if (unlikely(!rt_cache_route(nhc, rth)))
2405 			rt_add_uncached_list(rth);
2406 	}
2407 	skb_dst_set(skb, &rth->dst);
2408 	err = 0;
2409 	goto out;
2410 
2411 no_route:
2412 	RT_CACHE_STAT_INC(in_no_route);
2413 	res->type = RTN_UNREACHABLE;
2414 	res->fi = NULL;
2415 	res->table = NULL;
2416 	goto local_input;
2417 
2418 	/*
2419 	 *	Do not cache martian addresses: they should be logged (RFC1812)
2420 	 */
2421 martian_destination:
2422 	RT_CACHE_STAT_INC(in_martian_dst);
2423 #ifdef CONFIG_IP_ROUTE_VERBOSE
2424 	if (IN_DEV_LOG_MARTIANS(in_dev))
2425 		net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2426 				     &daddr, &saddr, dev->name);
2427 #endif
2428 
2429 e_inval:
2430 	err = -EINVAL;
2431 	goto out;
2432 
2433 e_nobufs:
2434 	err = -ENOBUFS;
2435 	goto out;
2436 
2437 martian_source:
2438 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2439 	goto out;
2440 }
2441 
2442 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2443 			 u8 tos, struct net_device *dev)
2444 {
2445 	struct fib_result res;
2446 	int err;
2447 
2448 	tos &= IPTOS_RT_MASK;
2449 	rcu_read_lock();
2450 	err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2451 	rcu_read_unlock();
2452 
2453 	return err;
2454 }
2455 EXPORT_SYMBOL(ip_route_input_noref);
2456 
2457 /* called with rcu_read_lock held */
2458 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2459 		       u8 tos, struct net_device *dev, struct fib_result *res)
2460 {
2461 	/* Multicast recognition logic is moved from route cache to here.
2462 	 * The problem was that too many Ethernet cards have broken/missing
2463 	 * hardware multicast filters :-( As result the host on multicasting
2464 	 * network acquires a lot of useless route cache entries, sort of
2465 	 * SDR messages from all the world. Now we try to get rid of them.
2466 	 * Really, provided software IP multicast filter is organized
2467 	 * reasonably (at least, hashed), it does not result in a slowdown
2468 	 * comparing with route cache reject entries.
2469 	 * Note, that multicast routers are not affected, because
2470 	 * route cache entry is created eventually.
2471 	 */
2472 	if (ipv4_is_multicast(daddr)) {
2473 		struct in_device *in_dev = __in_dev_get_rcu(dev);
2474 		int our = 0;
2475 		int err = -EINVAL;
2476 
2477 		if (!in_dev)
2478 			return err;
2479 		our = ip_check_mc_rcu(in_dev, daddr, saddr,
2480 				      ip_hdr(skb)->protocol);
2481 
2482 		/* check l3 master if no match yet */
2483 		if (!our && netif_is_l3_slave(dev)) {
2484 			struct in_device *l3_in_dev;
2485 
2486 			l3_in_dev = __in_dev_get_rcu(skb->dev);
2487 			if (l3_in_dev)
2488 				our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2489 						      ip_hdr(skb)->protocol);
2490 		}
2491 
2492 		if (our
2493 #ifdef CONFIG_IP_MROUTE
2494 			||
2495 		    (!ipv4_is_local_multicast(daddr) &&
2496 		     IN_DEV_MFORWARD(in_dev))
2497 #endif
2498 		   ) {
2499 			err = ip_route_input_mc(skb, daddr, saddr,
2500 						tos, dev, our);
2501 		}
2502 		return err;
2503 	}
2504 
2505 	return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2506 }
2507 
2508 /* called with rcu_read_lock() */
2509 static struct rtable *__mkroute_output(const struct fib_result *res,
2510 				       const struct flowi4 *fl4, int orig_oif,
2511 				       struct net_device *dev_out,
2512 				       unsigned int flags)
2513 {
2514 	struct fib_info *fi = res->fi;
2515 	struct fib_nh_exception *fnhe;
2516 	struct in_device *in_dev;
2517 	u16 type = res->type;
2518 	struct rtable *rth;
2519 	bool do_cache;
2520 
2521 	in_dev = __in_dev_get_rcu(dev_out);
2522 	if (!in_dev)
2523 		return ERR_PTR(-EINVAL);
2524 
2525 	if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2526 		if (ipv4_is_loopback(fl4->saddr) &&
2527 		    !(dev_out->flags & IFF_LOOPBACK) &&
2528 		    !netif_is_l3_master(dev_out))
2529 			return ERR_PTR(-EINVAL);
2530 
2531 	if (ipv4_is_lbcast(fl4->daddr))
2532 		type = RTN_BROADCAST;
2533 	else if (ipv4_is_multicast(fl4->daddr))
2534 		type = RTN_MULTICAST;
2535 	else if (ipv4_is_zeronet(fl4->daddr))
2536 		return ERR_PTR(-EINVAL);
2537 
2538 	if (dev_out->flags & IFF_LOOPBACK)
2539 		flags |= RTCF_LOCAL;
2540 
2541 	do_cache = true;
2542 	if (type == RTN_BROADCAST) {
2543 		flags |= RTCF_BROADCAST | RTCF_LOCAL;
2544 		fi = NULL;
2545 	} else if (type == RTN_MULTICAST) {
2546 		flags |= RTCF_MULTICAST | RTCF_LOCAL;
2547 		if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2548 				     fl4->flowi4_proto))
2549 			flags &= ~RTCF_LOCAL;
2550 		else
2551 			do_cache = false;
2552 		/* If multicast route do not exist use
2553 		 * default one, but do not gateway in this case.
2554 		 * Yes, it is hack.
2555 		 */
2556 		if (fi && res->prefixlen < 4)
2557 			fi = NULL;
2558 	} else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2559 		   (orig_oif != dev_out->ifindex)) {
2560 		/* For local routes that require a particular output interface
2561 		 * we do not want to cache the result.  Caching the result
2562 		 * causes incorrect behaviour when there are multiple source
2563 		 * addresses on the interface, the end result being that if the
2564 		 * intended recipient is waiting on that interface for the
2565 		 * packet he won't receive it because it will be delivered on
2566 		 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2567 		 * be set to the loopback interface as well.
2568 		 */
2569 		do_cache = false;
2570 	}
2571 
2572 	fnhe = NULL;
2573 	do_cache &= fi != NULL;
2574 	if (fi) {
2575 		struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2576 		struct rtable __rcu **prth;
2577 
2578 		fnhe = find_exception(nhc, fl4->daddr);
2579 		if (!do_cache)
2580 			goto add;
2581 		if (fnhe) {
2582 			prth = &fnhe->fnhe_rth_output;
2583 		} else {
2584 			if (unlikely(fl4->flowi4_flags &
2585 				     FLOWI_FLAG_KNOWN_NH &&
2586 				     !(nhc->nhc_gw_family &&
2587 				       nhc->nhc_scope == RT_SCOPE_LINK))) {
2588 				do_cache = false;
2589 				goto add;
2590 			}
2591 			prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2592 		}
2593 		rth = rcu_dereference(*prth);
2594 		if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2595 			return rth;
2596 	}
2597 
2598 add:
2599 	rth = rt_dst_alloc(dev_out, flags, type,
2600 			   IN_DEV_ORCONF(in_dev, NOPOLICY),
2601 			   IN_DEV_ORCONF(in_dev, NOXFRM));
2602 	if (!rth)
2603 		return ERR_PTR(-ENOBUFS);
2604 
2605 	rth->rt_iif = orig_oif;
2606 
2607 	RT_CACHE_STAT_INC(out_slow_tot);
2608 
2609 	if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2610 		if (flags & RTCF_LOCAL &&
2611 		    !(dev_out->flags & IFF_LOOPBACK)) {
2612 			rth->dst.output = ip_mc_output;
2613 			RT_CACHE_STAT_INC(out_slow_mc);
2614 		}
2615 #ifdef CONFIG_IP_MROUTE
2616 		if (type == RTN_MULTICAST) {
2617 			if (IN_DEV_MFORWARD(in_dev) &&
2618 			    !ipv4_is_local_multicast(fl4->daddr)) {
2619 				rth->dst.input = ip_mr_input;
2620 				rth->dst.output = ip_mc_output;
2621 			}
2622 		}
2623 #endif
2624 	}
2625 
2626 	rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2627 	lwtunnel_set_redirect(&rth->dst);
2628 
2629 	return rth;
2630 }
2631 
2632 /*
2633  * Major route resolver routine.
2634  */
2635 
2636 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2637 					const struct sk_buff *skb)
2638 {
2639 	struct fib_result res = {
2640 		.type		= RTN_UNSPEC,
2641 		.fi		= NULL,
2642 		.table		= NULL,
2643 		.tclassid	= 0,
2644 	};
2645 	struct rtable *rth;
2646 
2647 	fl4->flowi4_iif = LOOPBACK_IFINDEX;
2648 	ip_rt_fix_tos(fl4);
2649 
2650 	rcu_read_lock();
2651 	rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2652 	rcu_read_unlock();
2653 
2654 	return rth;
2655 }
2656 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2657 
2658 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2659 					    struct fib_result *res,
2660 					    const struct sk_buff *skb)
2661 {
2662 	struct net_device *dev_out = NULL;
2663 	int orig_oif = fl4->flowi4_oif;
2664 	unsigned int flags = 0;
2665 	struct rtable *rth;
2666 	int err;
2667 
2668 	if (fl4->saddr) {
2669 		if (ipv4_is_multicast(fl4->saddr) ||
2670 		    ipv4_is_lbcast(fl4->saddr) ||
2671 		    ipv4_is_zeronet(fl4->saddr)) {
2672 			rth = ERR_PTR(-EINVAL);
2673 			goto out;
2674 		}
2675 
2676 		rth = ERR_PTR(-ENETUNREACH);
2677 
2678 		/* I removed check for oif == dev_out->oif here.
2679 		 * It was wrong for two reasons:
2680 		 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2681 		 *    is assigned to multiple interfaces.
2682 		 * 2. Moreover, we are allowed to send packets with saddr
2683 		 *    of another iface. --ANK
2684 		 */
2685 
2686 		if (fl4->flowi4_oif == 0 &&
2687 		    (ipv4_is_multicast(fl4->daddr) ||
2688 		     ipv4_is_lbcast(fl4->daddr))) {
2689 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2690 			dev_out = __ip_dev_find(net, fl4->saddr, false);
2691 			if (!dev_out)
2692 				goto out;
2693 
2694 			/* Special hack: user can direct multicasts
2695 			 * and limited broadcast via necessary interface
2696 			 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2697 			 * This hack is not just for fun, it allows
2698 			 * vic,vat and friends to work.
2699 			 * They bind socket to loopback, set ttl to zero
2700 			 * and expect that it will work.
2701 			 * From the viewpoint of routing cache they are broken,
2702 			 * because we are not allowed to build multicast path
2703 			 * with loopback source addr (look, routing cache
2704 			 * cannot know, that ttl is zero, so that packet
2705 			 * will not leave this host and route is valid).
2706 			 * Luckily, this hack is good workaround.
2707 			 */
2708 
2709 			fl4->flowi4_oif = dev_out->ifindex;
2710 			goto make_route;
2711 		}
2712 
2713 		if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2714 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2715 			if (!__ip_dev_find(net, fl4->saddr, false))
2716 				goto out;
2717 		}
2718 	}
2719 
2720 
2721 	if (fl4->flowi4_oif) {
2722 		dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2723 		rth = ERR_PTR(-ENODEV);
2724 		if (!dev_out)
2725 			goto out;
2726 
2727 		/* RACE: Check return value of inet_select_addr instead. */
2728 		if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2729 			rth = ERR_PTR(-ENETUNREACH);
2730 			goto out;
2731 		}
2732 		if (ipv4_is_local_multicast(fl4->daddr) ||
2733 		    ipv4_is_lbcast(fl4->daddr) ||
2734 		    fl4->flowi4_proto == IPPROTO_IGMP) {
2735 			if (!fl4->saddr)
2736 				fl4->saddr = inet_select_addr(dev_out, 0,
2737 							      RT_SCOPE_LINK);
2738 			goto make_route;
2739 		}
2740 		if (!fl4->saddr) {
2741 			if (ipv4_is_multicast(fl4->daddr))
2742 				fl4->saddr = inet_select_addr(dev_out, 0,
2743 							      fl4->flowi4_scope);
2744 			else if (!fl4->daddr)
2745 				fl4->saddr = inet_select_addr(dev_out, 0,
2746 							      RT_SCOPE_HOST);
2747 		}
2748 	}
2749 
2750 	if (!fl4->daddr) {
2751 		fl4->daddr = fl4->saddr;
2752 		if (!fl4->daddr)
2753 			fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2754 		dev_out = net->loopback_dev;
2755 		fl4->flowi4_oif = LOOPBACK_IFINDEX;
2756 		res->type = RTN_LOCAL;
2757 		flags |= RTCF_LOCAL;
2758 		goto make_route;
2759 	}
2760 
2761 	err = fib_lookup(net, fl4, res, 0);
2762 	if (err) {
2763 		res->fi = NULL;
2764 		res->table = NULL;
2765 		if (fl4->flowi4_oif &&
2766 		    (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2767 			/* Apparently, routing tables are wrong. Assume,
2768 			 * that the destination is on link.
2769 			 *
2770 			 * WHY? DW.
2771 			 * Because we are allowed to send to iface
2772 			 * even if it has NO routes and NO assigned
2773 			 * addresses. When oif is specified, routing
2774 			 * tables are looked up with only one purpose:
2775 			 * to catch if destination is gatewayed, rather than
2776 			 * direct. Moreover, if MSG_DONTROUTE is set,
2777 			 * we send packet, ignoring both routing tables
2778 			 * and ifaddr state. --ANK
2779 			 *
2780 			 *
2781 			 * We could make it even if oif is unknown,
2782 			 * likely IPv6, but we do not.
2783 			 */
2784 
2785 			if (fl4->saddr == 0)
2786 				fl4->saddr = inet_select_addr(dev_out, 0,
2787 							      RT_SCOPE_LINK);
2788 			res->type = RTN_UNICAST;
2789 			goto make_route;
2790 		}
2791 		rth = ERR_PTR(err);
2792 		goto out;
2793 	}
2794 
2795 	if (res->type == RTN_LOCAL) {
2796 		if (!fl4->saddr) {
2797 			if (res->fi->fib_prefsrc)
2798 				fl4->saddr = res->fi->fib_prefsrc;
2799 			else
2800 				fl4->saddr = fl4->daddr;
2801 		}
2802 
2803 		/* L3 master device is the loopback for that domain */
2804 		dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2805 			net->loopback_dev;
2806 
2807 		/* make sure orig_oif points to fib result device even
2808 		 * though packet rx/tx happens over loopback or l3mdev
2809 		 */
2810 		orig_oif = FIB_RES_OIF(*res);
2811 
2812 		fl4->flowi4_oif = dev_out->ifindex;
2813 		flags |= RTCF_LOCAL;
2814 		goto make_route;
2815 	}
2816 
2817 	fib_select_path(net, res, fl4, skb);
2818 
2819 	dev_out = FIB_RES_DEV(*res);
2820 
2821 make_route:
2822 	rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2823 
2824 out:
2825 	return rth;
2826 }
2827 
2828 static struct dst_ops ipv4_dst_blackhole_ops = {
2829 	.family			= AF_INET,
2830 	.default_advmss		= ipv4_default_advmss,
2831 	.neigh_lookup		= ipv4_neigh_lookup,
2832 	.check			= dst_blackhole_check,
2833 	.cow_metrics		= dst_blackhole_cow_metrics,
2834 	.update_pmtu		= dst_blackhole_update_pmtu,
2835 	.redirect		= dst_blackhole_redirect,
2836 	.mtu			= dst_blackhole_mtu,
2837 };
2838 
2839 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2840 {
2841 	struct rtable *ort = (struct rtable *) dst_orig;
2842 	struct rtable *rt;
2843 
2844 	rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2845 	if (rt) {
2846 		struct dst_entry *new = &rt->dst;
2847 
2848 		new->__use = 1;
2849 		new->input = dst_discard;
2850 		new->output = dst_discard_out;
2851 
2852 		new->dev = net->loopback_dev;
2853 		dev_hold_track(new->dev, &new->dev_tracker, GFP_ATOMIC);
2854 
2855 		rt->rt_is_input = ort->rt_is_input;
2856 		rt->rt_iif = ort->rt_iif;
2857 		rt->rt_pmtu = ort->rt_pmtu;
2858 		rt->rt_mtu_locked = ort->rt_mtu_locked;
2859 
2860 		rt->rt_genid = rt_genid_ipv4(net);
2861 		rt->rt_flags = ort->rt_flags;
2862 		rt->rt_type = ort->rt_type;
2863 		rt->rt_uses_gateway = ort->rt_uses_gateway;
2864 		rt->rt_gw_family = ort->rt_gw_family;
2865 		if (rt->rt_gw_family == AF_INET)
2866 			rt->rt_gw4 = ort->rt_gw4;
2867 		else if (rt->rt_gw_family == AF_INET6)
2868 			rt->rt_gw6 = ort->rt_gw6;
2869 
2870 		INIT_LIST_HEAD(&rt->rt_uncached);
2871 	}
2872 
2873 	dst_release(dst_orig);
2874 
2875 	return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2876 }
2877 
2878 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2879 				    const struct sock *sk)
2880 {
2881 	struct rtable *rt = __ip_route_output_key(net, flp4);
2882 
2883 	if (IS_ERR(rt))
2884 		return rt;
2885 
2886 	if (flp4->flowi4_proto) {
2887 		flp4->flowi4_oif = rt->dst.dev->ifindex;
2888 		rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2889 							flowi4_to_flowi(flp4),
2890 							sk, 0);
2891 	}
2892 
2893 	return rt;
2894 }
2895 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2896 
2897 struct rtable *ip_route_output_tunnel(struct sk_buff *skb,
2898 				      struct net_device *dev,
2899 				      struct net *net, __be32 *saddr,
2900 				      const struct ip_tunnel_info *info,
2901 				      u8 protocol, bool use_cache)
2902 {
2903 #ifdef CONFIG_DST_CACHE
2904 	struct dst_cache *dst_cache;
2905 #endif
2906 	struct rtable *rt = NULL;
2907 	struct flowi4 fl4;
2908 	__u8 tos;
2909 
2910 #ifdef CONFIG_DST_CACHE
2911 	dst_cache = (struct dst_cache *)&info->dst_cache;
2912 	if (use_cache) {
2913 		rt = dst_cache_get_ip4(dst_cache, saddr);
2914 		if (rt)
2915 			return rt;
2916 	}
2917 #endif
2918 	memset(&fl4, 0, sizeof(fl4));
2919 	fl4.flowi4_mark = skb->mark;
2920 	fl4.flowi4_proto = protocol;
2921 	fl4.daddr = info->key.u.ipv4.dst;
2922 	fl4.saddr = info->key.u.ipv4.src;
2923 	tos = info->key.tos;
2924 	fl4.flowi4_tos = RT_TOS(tos);
2925 
2926 	rt = ip_route_output_key(net, &fl4);
2927 	if (IS_ERR(rt)) {
2928 		netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
2929 		return ERR_PTR(-ENETUNREACH);
2930 	}
2931 	if (rt->dst.dev == dev) { /* is this necessary? */
2932 		netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr);
2933 		ip_rt_put(rt);
2934 		return ERR_PTR(-ELOOP);
2935 	}
2936 #ifdef CONFIG_DST_CACHE
2937 	if (use_cache)
2938 		dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2939 #endif
2940 	*saddr = fl4.saddr;
2941 	return rt;
2942 }
2943 EXPORT_SYMBOL_GPL(ip_route_output_tunnel);
2944 
2945 /* called with rcu_read_lock held */
2946 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2947 			struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2948 			struct sk_buff *skb, u32 portid, u32 seq,
2949 			unsigned int flags)
2950 {
2951 	struct rtmsg *r;
2952 	struct nlmsghdr *nlh;
2953 	unsigned long expires = 0;
2954 	u32 error;
2955 	u32 metrics[RTAX_MAX];
2956 
2957 	nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2958 	if (!nlh)
2959 		return -EMSGSIZE;
2960 
2961 	r = nlmsg_data(nlh);
2962 	r->rtm_family	 = AF_INET;
2963 	r->rtm_dst_len	= 32;
2964 	r->rtm_src_len	= 0;
2965 	r->rtm_tos	= fl4 ? fl4->flowi4_tos : 0;
2966 	r->rtm_table	= table_id < 256 ? table_id : RT_TABLE_COMPAT;
2967 	if (nla_put_u32(skb, RTA_TABLE, table_id))
2968 		goto nla_put_failure;
2969 	r->rtm_type	= rt->rt_type;
2970 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2971 	r->rtm_protocol = RTPROT_UNSPEC;
2972 	r->rtm_flags	= (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2973 	if (rt->rt_flags & RTCF_NOTIFY)
2974 		r->rtm_flags |= RTM_F_NOTIFY;
2975 	if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2976 		r->rtm_flags |= RTCF_DOREDIRECT;
2977 
2978 	if (nla_put_in_addr(skb, RTA_DST, dst))
2979 		goto nla_put_failure;
2980 	if (src) {
2981 		r->rtm_src_len = 32;
2982 		if (nla_put_in_addr(skb, RTA_SRC, src))
2983 			goto nla_put_failure;
2984 	}
2985 	if (rt->dst.dev &&
2986 	    nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2987 		goto nla_put_failure;
2988 	if (rt->dst.lwtstate &&
2989 	    lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2990 		goto nla_put_failure;
2991 #ifdef CONFIG_IP_ROUTE_CLASSID
2992 	if (rt->dst.tclassid &&
2993 	    nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2994 		goto nla_put_failure;
2995 #endif
2996 	if (fl4 && !rt_is_input_route(rt) &&
2997 	    fl4->saddr != src) {
2998 		if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2999 			goto nla_put_failure;
3000 	}
3001 	if (rt->rt_uses_gateway) {
3002 		if (rt->rt_gw_family == AF_INET &&
3003 		    nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
3004 			goto nla_put_failure;
3005 		} else if (rt->rt_gw_family == AF_INET6) {
3006 			int alen = sizeof(struct in6_addr);
3007 			struct nlattr *nla;
3008 			struct rtvia *via;
3009 
3010 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
3011 			if (!nla)
3012 				goto nla_put_failure;
3013 
3014 			via = nla_data(nla);
3015 			via->rtvia_family = AF_INET6;
3016 			memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
3017 		}
3018 	}
3019 
3020 	expires = rt->dst.expires;
3021 	if (expires) {
3022 		unsigned long now = jiffies;
3023 
3024 		if (time_before(now, expires))
3025 			expires -= now;
3026 		else
3027 			expires = 0;
3028 	}
3029 
3030 	memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3031 	if (rt->rt_pmtu && expires)
3032 		metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3033 	if (rt->rt_mtu_locked && expires)
3034 		metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3035 	if (rtnetlink_put_metrics(skb, metrics) < 0)
3036 		goto nla_put_failure;
3037 
3038 	if (fl4) {
3039 		if (fl4->flowi4_mark &&
3040 		    nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3041 			goto nla_put_failure;
3042 
3043 		if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3044 		    nla_put_u32(skb, RTA_UID,
3045 				from_kuid_munged(current_user_ns(),
3046 						 fl4->flowi4_uid)))
3047 			goto nla_put_failure;
3048 
3049 		if (rt_is_input_route(rt)) {
3050 #ifdef CONFIG_IP_MROUTE
3051 			if (ipv4_is_multicast(dst) &&
3052 			    !ipv4_is_local_multicast(dst) &&
3053 			    IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
3054 				int err = ipmr_get_route(net, skb,
3055 							 fl4->saddr, fl4->daddr,
3056 							 r, portid);
3057 
3058 				if (err <= 0) {
3059 					if (err == 0)
3060 						return 0;
3061 					goto nla_put_failure;
3062 				}
3063 			} else
3064 #endif
3065 				if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3066 					goto nla_put_failure;
3067 		}
3068 	}
3069 
3070 	error = rt->dst.error;
3071 
3072 	if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3073 		goto nla_put_failure;
3074 
3075 	nlmsg_end(skb, nlh);
3076 	return 0;
3077 
3078 nla_put_failure:
3079 	nlmsg_cancel(skb, nlh);
3080 	return -EMSGSIZE;
3081 }
3082 
3083 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3084 			    struct netlink_callback *cb, u32 table_id,
3085 			    struct fnhe_hash_bucket *bucket, int genid,
3086 			    int *fa_index, int fa_start, unsigned int flags)
3087 {
3088 	int i;
3089 
3090 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
3091 		struct fib_nh_exception *fnhe;
3092 
3093 		for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3094 		     fnhe = rcu_dereference(fnhe->fnhe_next)) {
3095 			struct rtable *rt;
3096 			int err;
3097 
3098 			if (*fa_index < fa_start)
3099 				goto next;
3100 
3101 			if (fnhe->fnhe_genid != genid)
3102 				goto next;
3103 
3104 			if (fnhe->fnhe_expires &&
3105 			    time_after(jiffies, fnhe->fnhe_expires))
3106 				goto next;
3107 
3108 			rt = rcu_dereference(fnhe->fnhe_rth_input);
3109 			if (!rt)
3110 				rt = rcu_dereference(fnhe->fnhe_rth_output);
3111 			if (!rt)
3112 				goto next;
3113 
3114 			err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3115 					   table_id, NULL, skb,
3116 					   NETLINK_CB(cb->skb).portid,
3117 					   cb->nlh->nlmsg_seq, flags);
3118 			if (err)
3119 				return err;
3120 next:
3121 			(*fa_index)++;
3122 		}
3123 	}
3124 
3125 	return 0;
3126 }
3127 
3128 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3129 		       u32 table_id, struct fib_info *fi,
3130 		       int *fa_index, int fa_start, unsigned int flags)
3131 {
3132 	struct net *net = sock_net(cb->skb->sk);
3133 	int nhsel, genid = fnhe_genid(net);
3134 
3135 	for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3136 		struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3137 		struct fnhe_hash_bucket *bucket;
3138 		int err;
3139 
3140 		if (nhc->nhc_flags & RTNH_F_DEAD)
3141 			continue;
3142 
3143 		rcu_read_lock();
3144 		bucket = rcu_dereference(nhc->nhc_exceptions);
3145 		err = 0;
3146 		if (bucket)
3147 			err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3148 					       genid, fa_index, fa_start,
3149 					       flags);
3150 		rcu_read_unlock();
3151 		if (err)
3152 			return err;
3153 	}
3154 
3155 	return 0;
3156 }
3157 
3158 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3159 						   u8 ip_proto, __be16 sport,
3160 						   __be16 dport)
3161 {
3162 	struct sk_buff *skb;
3163 	struct iphdr *iph;
3164 
3165 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3166 	if (!skb)
3167 		return NULL;
3168 
3169 	/* Reserve room for dummy headers, this skb can pass
3170 	 * through good chunk of routing engine.
3171 	 */
3172 	skb_reset_mac_header(skb);
3173 	skb_reset_network_header(skb);
3174 	skb->protocol = htons(ETH_P_IP);
3175 	iph = skb_put(skb, sizeof(struct iphdr));
3176 	iph->protocol = ip_proto;
3177 	iph->saddr = src;
3178 	iph->daddr = dst;
3179 	iph->version = 0x4;
3180 	iph->frag_off = 0;
3181 	iph->ihl = 0x5;
3182 	skb_set_transport_header(skb, skb->len);
3183 
3184 	switch (iph->protocol) {
3185 	case IPPROTO_UDP: {
3186 		struct udphdr *udph;
3187 
3188 		udph = skb_put_zero(skb, sizeof(struct udphdr));
3189 		udph->source = sport;
3190 		udph->dest = dport;
3191 		udph->len = htons(sizeof(struct udphdr));
3192 		udph->check = 0;
3193 		break;
3194 	}
3195 	case IPPROTO_TCP: {
3196 		struct tcphdr *tcph;
3197 
3198 		tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3199 		tcph->source	= sport;
3200 		tcph->dest	= dport;
3201 		tcph->doff	= sizeof(struct tcphdr) / 4;
3202 		tcph->rst = 1;
3203 		tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3204 					    src, dst, 0);
3205 		break;
3206 	}
3207 	case IPPROTO_ICMP: {
3208 		struct icmphdr *icmph;
3209 
3210 		icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3211 		icmph->type = ICMP_ECHO;
3212 		icmph->code = 0;
3213 	}
3214 	}
3215 
3216 	return skb;
3217 }
3218 
3219 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3220 				       const struct nlmsghdr *nlh,
3221 				       struct nlattr **tb,
3222 				       struct netlink_ext_ack *extack)
3223 {
3224 	struct rtmsg *rtm;
3225 	int i, err;
3226 
3227 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3228 		NL_SET_ERR_MSG(extack,
3229 			       "ipv4: Invalid header for route get request");
3230 		return -EINVAL;
3231 	}
3232 
3233 	if (!netlink_strict_get_check(skb))
3234 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3235 					      rtm_ipv4_policy, extack);
3236 
3237 	rtm = nlmsg_data(nlh);
3238 	if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3239 	    (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3240 	    rtm->rtm_table || rtm->rtm_protocol ||
3241 	    rtm->rtm_scope || rtm->rtm_type) {
3242 		NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3243 		return -EINVAL;
3244 	}
3245 
3246 	if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3247 			       RTM_F_LOOKUP_TABLE |
3248 			       RTM_F_FIB_MATCH)) {
3249 		NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3250 		return -EINVAL;
3251 	}
3252 
3253 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3254 					    rtm_ipv4_policy, extack);
3255 	if (err)
3256 		return err;
3257 
3258 	if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3259 	    (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3260 		NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3261 		return -EINVAL;
3262 	}
3263 
3264 	for (i = 0; i <= RTA_MAX; i++) {
3265 		if (!tb[i])
3266 			continue;
3267 
3268 		switch (i) {
3269 		case RTA_IIF:
3270 		case RTA_OIF:
3271 		case RTA_SRC:
3272 		case RTA_DST:
3273 		case RTA_IP_PROTO:
3274 		case RTA_SPORT:
3275 		case RTA_DPORT:
3276 		case RTA_MARK:
3277 		case RTA_UID:
3278 			break;
3279 		default:
3280 			NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3281 			return -EINVAL;
3282 		}
3283 	}
3284 
3285 	return 0;
3286 }
3287 
3288 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3289 			     struct netlink_ext_ack *extack)
3290 {
3291 	struct net *net = sock_net(in_skb->sk);
3292 	struct nlattr *tb[RTA_MAX+1];
3293 	u32 table_id = RT_TABLE_MAIN;
3294 	__be16 sport = 0, dport = 0;
3295 	struct fib_result res = {};
3296 	u8 ip_proto = IPPROTO_UDP;
3297 	struct rtable *rt = NULL;
3298 	struct sk_buff *skb;
3299 	struct rtmsg *rtm;
3300 	struct flowi4 fl4 = {};
3301 	__be32 dst = 0;
3302 	__be32 src = 0;
3303 	kuid_t uid;
3304 	u32 iif;
3305 	int err;
3306 	int mark;
3307 
3308 	err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3309 	if (err < 0)
3310 		return err;
3311 
3312 	rtm = nlmsg_data(nlh);
3313 	src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3314 	dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3315 	iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3316 	mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3317 	if (tb[RTA_UID])
3318 		uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3319 	else
3320 		uid = (iif ? INVALID_UID : current_uid());
3321 
3322 	if (tb[RTA_IP_PROTO]) {
3323 		err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3324 						  &ip_proto, AF_INET, extack);
3325 		if (err)
3326 			return err;
3327 	}
3328 
3329 	if (tb[RTA_SPORT])
3330 		sport = nla_get_be16(tb[RTA_SPORT]);
3331 
3332 	if (tb[RTA_DPORT])
3333 		dport = nla_get_be16(tb[RTA_DPORT]);
3334 
3335 	skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3336 	if (!skb)
3337 		return -ENOBUFS;
3338 
3339 	fl4.daddr = dst;
3340 	fl4.saddr = src;
3341 	fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3342 	fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3343 	fl4.flowi4_mark = mark;
3344 	fl4.flowi4_uid = uid;
3345 	if (sport)
3346 		fl4.fl4_sport = sport;
3347 	if (dport)
3348 		fl4.fl4_dport = dport;
3349 	fl4.flowi4_proto = ip_proto;
3350 
3351 	rcu_read_lock();
3352 
3353 	if (iif) {
3354 		struct net_device *dev;
3355 
3356 		dev = dev_get_by_index_rcu(net, iif);
3357 		if (!dev) {
3358 			err = -ENODEV;
3359 			goto errout_rcu;
3360 		}
3361 
3362 		fl4.flowi4_iif = iif; /* for rt_fill_info */
3363 		skb->dev	= dev;
3364 		skb->mark	= mark;
3365 		err = ip_route_input_rcu(skb, dst, src,
3366 					 rtm->rtm_tos & IPTOS_RT_MASK, dev,
3367 					 &res);
3368 
3369 		rt = skb_rtable(skb);
3370 		if (err == 0 && rt->dst.error)
3371 			err = -rt->dst.error;
3372 	} else {
3373 		fl4.flowi4_iif = LOOPBACK_IFINDEX;
3374 		skb->dev = net->loopback_dev;
3375 		rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3376 		err = 0;
3377 		if (IS_ERR(rt))
3378 			err = PTR_ERR(rt);
3379 		else
3380 			skb_dst_set(skb, &rt->dst);
3381 	}
3382 
3383 	if (err)
3384 		goto errout_rcu;
3385 
3386 	if (rtm->rtm_flags & RTM_F_NOTIFY)
3387 		rt->rt_flags |= RTCF_NOTIFY;
3388 
3389 	if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3390 		table_id = res.table ? res.table->tb_id : 0;
3391 
3392 	/* reset skb for netlink reply msg */
3393 	skb_trim(skb, 0);
3394 	skb_reset_network_header(skb);
3395 	skb_reset_transport_header(skb);
3396 	skb_reset_mac_header(skb);
3397 
3398 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3399 		struct fib_rt_info fri;
3400 
3401 		if (!res.fi) {
3402 			err = fib_props[res.type].error;
3403 			if (!err)
3404 				err = -EHOSTUNREACH;
3405 			goto errout_rcu;
3406 		}
3407 		fri.fi = res.fi;
3408 		fri.tb_id = table_id;
3409 		fri.dst = res.prefix;
3410 		fri.dst_len = res.prefixlen;
3411 		fri.tos = fl4.flowi4_tos;
3412 		fri.type = rt->rt_type;
3413 		fri.offload = 0;
3414 		fri.trap = 0;
3415 		fri.offload_failed = 0;
3416 		if (res.fa_head) {
3417 			struct fib_alias *fa;
3418 
3419 			hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3420 				u8 slen = 32 - fri.dst_len;
3421 
3422 				if (fa->fa_slen == slen &&
3423 				    fa->tb_id == fri.tb_id &&
3424 				    fa->fa_dscp == inet_dsfield_to_dscp(fri.tos) &&
3425 				    fa->fa_info == res.fi &&
3426 				    fa->fa_type == fri.type) {
3427 					fri.offload = READ_ONCE(fa->offload);
3428 					fri.trap = READ_ONCE(fa->trap);
3429 					break;
3430 				}
3431 			}
3432 		}
3433 		err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3434 				    nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3435 	} else {
3436 		err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3437 				   NETLINK_CB(in_skb).portid,
3438 				   nlh->nlmsg_seq, 0);
3439 	}
3440 	if (err < 0)
3441 		goto errout_rcu;
3442 
3443 	rcu_read_unlock();
3444 
3445 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3446 
3447 errout_free:
3448 	return err;
3449 errout_rcu:
3450 	rcu_read_unlock();
3451 	kfree_skb(skb);
3452 	goto errout_free;
3453 }
3454 
3455 void ip_rt_multicast_event(struct in_device *in_dev)
3456 {
3457 	rt_cache_flush(dev_net(in_dev->dev));
3458 }
3459 
3460 #ifdef CONFIG_SYSCTL
3461 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3462 static int ip_rt_gc_min_interval __read_mostly	= HZ / 2;
3463 static int ip_rt_gc_elasticity __read_mostly	= 8;
3464 static int ip_min_valid_pmtu __read_mostly	= IPV4_MIN_MTU;
3465 
3466 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3467 		void *buffer, size_t *lenp, loff_t *ppos)
3468 {
3469 	struct net *net = (struct net *)__ctl->extra1;
3470 
3471 	if (write) {
3472 		rt_cache_flush(net);
3473 		fnhe_genid_bump(net);
3474 		return 0;
3475 	}
3476 
3477 	return -EINVAL;
3478 }
3479 
3480 static struct ctl_table ipv4_route_table[] = {
3481 	{
3482 		.procname	= "gc_thresh",
3483 		.data		= &ipv4_dst_ops.gc_thresh,
3484 		.maxlen		= sizeof(int),
3485 		.mode		= 0644,
3486 		.proc_handler	= proc_dointvec,
3487 	},
3488 	{
3489 		.procname	= "max_size",
3490 		.data		= &ip_rt_max_size,
3491 		.maxlen		= sizeof(int),
3492 		.mode		= 0644,
3493 		.proc_handler	= proc_dointvec,
3494 	},
3495 	{
3496 		/*  Deprecated. Use gc_min_interval_ms */
3497 
3498 		.procname	= "gc_min_interval",
3499 		.data		= &ip_rt_gc_min_interval,
3500 		.maxlen		= sizeof(int),
3501 		.mode		= 0644,
3502 		.proc_handler	= proc_dointvec_jiffies,
3503 	},
3504 	{
3505 		.procname	= "gc_min_interval_ms",
3506 		.data		= &ip_rt_gc_min_interval,
3507 		.maxlen		= sizeof(int),
3508 		.mode		= 0644,
3509 		.proc_handler	= proc_dointvec_ms_jiffies,
3510 	},
3511 	{
3512 		.procname	= "gc_timeout",
3513 		.data		= &ip_rt_gc_timeout,
3514 		.maxlen		= sizeof(int),
3515 		.mode		= 0644,
3516 		.proc_handler	= proc_dointvec_jiffies,
3517 	},
3518 	{
3519 		.procname	= "gc_interval",
3520 		.data		= &ip_rt_gc_interval,
3521 		.maxlen		= sizeof(int),
3522 		.mode		= 0644,
3523 		.proc_handler	= proc_dointvec_jiffies,
3524 	},
3525 	{
3526 		.procname	= "redirect_load",
3527 		.data		= &ip_rt_redirect_load,
3528 		.maxlen		= sizeof(int),
3529 		.mode		= 0644,
3530 		.proc_handler	= proc_dointvec,
3531 	},
3532 	{
3533 		.procname	= "redirect_number",
3534 		.data		= &ip_rt_redirect_number,
3535 		.maxlen		= sizeof(int),
3536 		.mode		= 0644,
3537 		.proc_handler	= proc_dointvec,
3538 	},
3539 	{
3540 		.procname	= "redirect_silence",
3541 		.data		= &ip_rt_redirect_silence,
3542 		.maxlen		= sizeof(int),
3543 		.mode		= 0644,
3544 		.proc_handler	= proc_dointvec,
3545 	},
3546 	{
3547 		.procname	= "error_cost",
3548 		.data		= &ip_rt_error_cost,
3549 		.maxlen		= sizeof(int),
3550 		.mode		= 0644,
3551 		.proc_handler	= proc_dointvec,
3552 	},
3553 	{
3554 		.procname	= "error_burst",
3555 		.data		= &ip_rt_error_burst,
3556 		.maxlen		= sizeof(int),
3557 		.mode		= 0644,
3558 		.proc_handler	= proc_dointvec,
3559 	},
3560 	{
3561 		.procname	= "gc_elasticity",
3562 		.data		= &ip_rt_gc_elasticity,
3563 		.maxlen		= sizeof(int),
3564 		.mode		= 0644,
3565 		.proc_handler	= proc_dointvec,
3566 	},
3567 	{ }
3568 };
3569 
3570 static const char ipv4_route_flush_procname[] = "flush";
3571 
3572 static struct ctl_table ipv4_route_netns_table[] = {
3573 	{
3574 		.procname	= ipv4_route_flush_procname,
3575 		.maxlen		= sizeof(int),
3576 		.mode		= 0200,
3577 		.proc_handler	= ipv4_sysctl_rtcache_flush,
3578 	},
3579 	{
3580 		.procname       = "min_pmtu",
3581 		.data           = &init_net.ipv4.ip_rt_min_pmtu,
3582 		.maxlen         = sizeof(int),
3583 		.mode           = 0644,
3584 		.proc_handler   = proc_dointvec_minmax,
3585 		.extra1         = &ip_min_valid_pmtu,
3586 	},
3587 	{
3588 		.procname       = "mtu_expires",
3589 		.data           = &init_net.ipv4.ip_rt_mtu_expires,
3590 		.maxlen         = sizeof(int),
3591 		.mode           = 0644,
3592 		.proc_handler   = proc_dointvec_jiffies,
3593 	},
3594 	{
3595 		.procname   = "min_adv_mss",
3596 		.data       = &init_net.ipv4.ip_rt_min_advmss,
3597 		.maxlen     = sizeof(int),
3598 		.mode       = 0644,
3599 		.proc_handler   = proc_dointvec,
3600 	},
3601 	{ },
3602 };
3603 
3604 static __net_init int sysctl_route_net_init(struct net *net)
3605 {
3606 	struct ctl_table *tbl;
3607 
3608 	tbl = ipv4_route_netns_table;
3609 	if (!net_eq(net, &init_net)) {
3610 		int i;
3611 
3612 		tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3613 		if (!tbl)
3614 			goto err_dup;
3615 
3616 		/* Don't export non-whitelisted sysctls to unprivileged users */
3617 		if (net->user_ns != &init_user_ns) {
3618 			if (tbl[0].procname != ipv4_route_flush_procname)
3619 				tbl[0].procname = NULL;
3620 		}
3621 
3622 		/* Update the variables to point into the current struct net
3623 		 * except for the first element flush
3624 		 */
3625 		for (i = 1; i < ARRAY_SIZE(ipv4_route_netns_table) - 1; i++)
3626 			tbl[i].data += (void *)net - (void *)&init_net;
3627 	}
3628 	tbl[0].extra1 = net;
3629 
3630 	net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3631 	if (!net->ipv4.route_hdr)
3632 		goto err_reg;
3633 	return 0;
3634 
3635 err_reg:
3636 	if (tbl != ipv4_route_netns_table)
3637 		kfree(tbl);
3638 err_dup:
3639 	return -ENOMEM;
3640 }
3641 
3642 static __net_exit void sysctl_route_net_exit(struct net *net)
3643 {
3644 	struct ctl_table *tbl;
3645 
3646 	tbl = net->ipv4.route_hdr->ctl_table_arg;
3647 	unregister_net_sysctl_table(net->ipv4.route_hdr);
3648 	BUG_ON(tbl == ipv4_route_netns_table);
3649 	kfree(tbl);
3650 }
3651 
3652 static __net_initdata struct pernet_operations sysctl_route_ops = {
3653 	.init = sysctl_route_net_init,
3654 	.exit = sysctl_route_net_exit,
3655 };
3656 #endif
3657 
3658 static __net_init int netns_ip_rt_init(struct net *net)
3659 {
3660 	/* Set default value for namespaceified sysctls */
3661 	net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3662 	net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3663 	net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3664 	return 0;
3665 }
3666 
3667 static struct pernet_operations __net_initdata ip_rt_ops = {
3668 	.init = netns_ip_rt_init,
3669 };
3670 
3671 static __net_init int rt_genid_init(struct net *net)
3672 {
3673 	atomic_set(&net->ipv4.rt_genid, 0);
3674 	atomic_set(&net->fnhe_genid, 0);
3675 	atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3676 	return 0;
3677 }
3678 
3679 static __net_initdata struct pernet_operations rt_genid_ops = {
3680 	.init = rt_genid_init,
3681 };
3682 
3683 static int __net_init ipv4_inetpeer_init(struct net *net)
3684 {
3685 	struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3686 
3687 	if (!bp)
3688 		return -ENOMEM;
3689 	inet_peer_base_init(bp);
3690 	net->ipv4.peers = bp;
3691 	return 0;
3692 }
3693 
3694 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3695 {
3696 	struct inet_peer_base *bp = net->ipv4.peers;
3697 
3698 	net->ipv4.peers = NULL;
3699 	inetpeer_invalidate_tree(bp);
3700 	kfree(bp);
3701 }
3702 
3703 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3704 	.init	=	ipv4_inetpeer_init,
3705 	.exit	=	ipv4_inetpeer_exit,
3706 };
3707 
3708 #ifdef CONFIG_IP_ROUTE_CLASSID
3709 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3710 #endif /* CONFIG_IP_ROUTE_CLASSID */
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 	prandom_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 		INIT_LIST_HEAD(&ul->quarantine);
3739 		spin_lock_init(&ul->lock);
3740 	}
3741 #ifdef CONFIG_IP_ROUTE_CLASSID
3742 	ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3743 	if (!ip_rt_acct)
3744 		panic("IP: failed to allocate ip_rt_acct\n");
3745 #endif
3746 
3747 	ipv4_dst_ops.kmem_cachep =
3748 		kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3749 				  SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3750 
3751 	ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3752 
3753 	if (dst_entries_init(&ipv4_dst_ops) < 0)
3754 		panic("IP: failed to allocate ipv4_dst_ops counter\n");
3755 
3756 	if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3757 		panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3758 
3759 	ipv4_dst_ops.gc_thresh = ~0;
3760 	ip_rt_max_size = INT_MAX;
3761 
3762 	devinet_init();
3763 	ip_fib_init();
3764 
3765 	if (ip_rt_proc_init())
3766 		pr_err("Unable to create route proc files\n");
3767 #ifdef CONFIG_XFRM
3768 	xfrm_init();
3769 	xfrm4_init();
3770 #endif
3771 	rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3772 		      RTNL_FLAG_DOIT_UNLOCKED);
3773 
3774 #ifdef CONFIG_SYSCTL
3775 	register_pernet_subsys(&sysctl_route_ops);
3776 #endif
3777 	register_pernet_subsys(&ip_rt_ops);
3778 	register_pernet_subsys(&rt_genid_ops);
3779 	register_pernet_subsys(&ipv4_inetpeer_ops);
3780 	return 0;
3781 }
3782 
3783 #ifdef CONFIG_SYSCTL
3784 /*
3785  * We really need to sanitize the damn ipv4 init order, then all
3786  * this nonsense will go away.
3787  */
3788 void __init ip_static_sysctl_init(void)
3789 {
3790 	register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3791 }
3792 #endif
3793