xref: /linux/net/ipv4/route.c (revision a5d9265e017f081f0dc133c0e2f45103d027b874)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		ROUTE - implementation of the IP router.
7  *
8  * Authors:	Ross Biro
9  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
11  *		Linus Torvalds, <Linus.Torvalds@helsinki.fi>
12  *		Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
13  *
14  * Fixes:
15  *		Alan Cox	:	Verify area fixes.
16  *		Alan Cox	:	cli() protects routing changes
17  *		Rui Oliveira	:	ICMP routing table updates
18  *		(rco@di.uminho.pt)	Routing table insertion and update
19  *		Linus Torvalds	:	Rewrote bits to be sensible
20  *		Alan Cox	:	Added BSD route gw semantics
21  *		Alan Cox	:	Super /proc >4K
22  *		Alan Cox	:	MTU in route table
23  *		Alan Cox	: 	MSS actually. Also added the window
24  *					clamper.
25  *		Sam Lantinga	:	Fixed route matching in rt_del()
26  *		Alan Cox	:	Routing cache support.
27  *		Alan Cox	:	Removed compatibility cruft.
28  *		Alan Cox	:	RTF_REJECT support.
29  *		Alan Cox	:	TCP irtt support.
30  *		Jonathan Naylor	:	Added Metric support.
31  *	Miquel van Smoorenburg	:	BSD API fixes.
32  *	Miquel van Smoorenburg	:	Metrics.
33  *		Alan Cox	:	Use __u32 properly
34  *		Alan Cox	:	Aligned routing errors more closely with BSD
35  *					our system is still very different.
36  *		Alan Cox	:	Faster /proc handling
37  *	Alexey Kuznetsov	:	Massive rework to support tree based routing,
38  *					routing caches and better behaviour.
39  *
40  *		Olaf Erb	:	irtt wasn't being copied right.
41  *		Bjorn Ekwall	:	Kerneld route support.
42  *		Alan Cox	:	Multicast fixed (I hope)
43  * 		Pavel Krauz	:	Limited broadcast fixed
44  *		Mike McLagan	:	Routing by source
45  *	Alexey Kuznetsov	:	End of old history. Split to fib.c and
46  *					route.c and rewritten from scratch.
47  *		Andi Kleen	:	Load-limit warning messages.
48  *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
49  *	Vitaly E. Lavrov	:	Race condition in ip_route_input_slow.
50  *	Tobias Ringstrom	:	Uninitialized res.type in ip_route_output_slow.
51  *	Vladimir V. Ivanov	:	IP rule info (flowid) is really useful.
52  *		Marc Boucher	:	routing by fwmark
53  *	Robert Olsson		:	Added rt_cache statistics
54  *	Arnaldo C. Melo		:	Convert proc stuff to seq_file
55  *	Eric Dumazet		:	hashed spinlocks and rt_check_expire() fixes.
56  * 	Ilia Sotnikov		:	Ignore TOS on PMTUD and Redirect
57  * 	Ilia Sotnikov		:	Removed TOS from hash calculations
58  *
59  *		This program is free software; you can redistribute it and/or
60  *		modify it under the terms of the GNU General Public License
61  *		as published by the Free Software Foundation; either version
62  *		2 of the License, or (at your option) any later version.
63  */
64 
65 #define pr_fmt(fmt) "IPv4: " fmt
66 
67 #include <linux/module.h>
68 #include <linux/uaccess.h>
69 #include <linux/bitops.h>
70 #include <linux/types.h>
71 #include <linux/kernel.h>
72 #include <linux/mm.h>
73 #include <linux/string.h>
74 #include <linux/socket.h>
75 #include <linux/sockios.h>
76 #include <linux/errno.h>
77 #include <linux/in.h>
78 #include <linux/inet.h>
79 #include <linux/netdevice.h>
80 #include <linux/proc_fs.h>
81 #include <linux/init.h>
82 #include <linux/skbuff.h>
83 #include <linux/inetdevice.h>
84 #include <linux/igmp.h>
85 #include <linux/pkt_sched.h>
86 #include <linux/mroute.h>
87 #include <linux/netfilter_ipv4.h>
88 #include <linux/random.h>
89 #include <linux/rcupdate.h>
90 #include <linux/times.h>
91 #include <linux/slab.h>
92 #include <linux/jhash.h>
93 #include <net/dst.h>
94 #include <net/dst_metadata.h>
95 #include <net/net_namespace.h>
96 #include <net/protocol.h>
97 #include <net/ip.h>
98 #include <net/route.h>
99 #include <net/inetpeer.h>
100 #include <net/sock.h>
101 #include <net/ip_fib.h>
102 #include <net/arp.h>
103 #include <net/tcp.h>
104 #include <net/icmp.h>
105 #include <net/xfrm.h>
106 #include <net/lwtunnel.h>
107 #include <net/netevent.h>
108 #include <net/rtnetlink.h>
109 #ifdef CONFIG_SYSCTL
110 #include <linux/sysctl.h>
111 #endif
112 #include <net/secure_seq.h>
113 #include <net/ip_tunnels.h>
114 #include <net/l3mdev.h>
115 
116 #include "fib_lookup.h"
117 
118 #define RT_FL_TOS(oldflp4) \
119 	((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
120 
121 #define RT_GC_TIMEOUT (300*HZ)
122 
123 static int ip_rt_max_size;
124 static int ip_rt_redirect_number __read_mostly	= 9;
125 static int ip_rt_redirect_load __read_mostly	= HZ / 50;
126 static int ip_rt_redirect_silence __read_mostly	= ((HZ / 50) << (9 + 1));
127 static int ip_rt_error_cost __read_mostly	= HZ;
128 static int ip_rt_error_burst __read_mostly	= 5 * HZ;
129 static int ip_rt_mtu_expires __read_mostly	= 10 * 60 * HZ;
130 static u32 ip_rt_min_pmtu __read_mostly		= 512 + 20 + 20;
131 static int ip_rt_min_advmss __read_mostly	= 256;
132 
133 static int ip_rt_gc_timeout __read_mostly	= RT_GC_TIMEOUT;
134 
135 /*
136  *	Interface to generic destination cache.
137  */
138 
139 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
140 static unsigned int	 ipv4_default_advmss(const struct dst_entry *dst);
141 static unsigned int	 ipv4_mtu(const struct dst_entry *dst);
142 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
143 static void		 ipv4_link_failure(struct sk_buff *skb);
144 static void		 ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
145 					   struct sk_buff *skb, u32 mtu);
146 static void		 ip_do_redirect(struct dst_entry *dst, struct sock *sk,
147 					struct sk_buff *skb);
148 static void		ipv4_dst_destroy(struct dst_entry *dst);
149 
150 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
151 {
152 	WARN_ON(1);
153 	return NULL;
154 }
155 
156 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
157 					   struct sk_buff *skb,
158 					   const void *daddr);
159 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
160 
161 static struct dst_ops ipv4_dst_ops = {
162 	.family =		AF_INET,
163 	.check =		ipv4_dst_check,
164 	.default_advmss =	ipv4_default_advmss,
165 	.mtu =			ipv4_mtu,
166 	.cow_metrics =		ipv4_cow_metrics,
167 	.destroy =		ipv4_dst_destroy,
168 	.negative_advice =	ipv4_negative_advice,
169 	.link_failure =		ipv4_link_failure,
170 	.update_pmtu =		ip_rt_update_pmtu,
171 	.redirect =		ip_do_redirect,
172 	.local_out =		__ip_local_out,
173 	.neigh_lookup =		ipv4_neigh_lookup,
174 	.confirm_neigh =	ipv4_confirm_neigh,
175 };
176 
177 #define ECN_OR_COST(class)	TC_PRIO_##class
178 
179 const __u8 ip_tos2prio[16] = {
180 	TC_PRIO_BESTEFFORT,
181 	ECN_OR_COST(BESTEFFORT),
182 	TC_PRIO_BESTEFFORT,
183 	ECN_OR_COST(BESTEFFORT),
184 	TC_PRIO_BULK,
185 	ECN_OR_COST(BULK),
186 	TC_PRIO_BULK,
187 	ECN_OR_COST(BULK),
188 	TC_PRIO_INTERACTIVE,
189 	ECN_OR_COST(INTERACTIVE),
190 	TC_PRIO_INTERACTIVE,
191 	ECN_OR_COST(INTERACTIVE),
192 	TC_PRIO_INTERACTIVE_BULK,
193 	ECN_OR_COST(INTERACTIVE_BULK),
194 	TC_PRIO_INTERACTIVE_BULK,
195 	ECN_OR_COST(INTERACTIVE_BULK)
196 };
197 EXPORT_SYMBOL(ip_tos2prio);
198 
199 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
200 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
201 
202 #ifdef CONFIG_PROC_FS
203 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
204 {
205 	if (*pos)
206 		return NULL;
207 	return SEQ_START_TOKEN;
208 }
209 
210 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
211 {
212 	++*pos;
213 	return NULL;
214 }
215 
216 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
217 {
218 }
219 
220 static int rt_cache_seq_show(struct seq_file *seq, void *v)
221 {
222 	if (v == SEQ_START_TOKEN)
223 		seq_printf(seq, "%-127s\n",
224 			   "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
225 			   "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
226 			   "HHUptod\tSpecDst");
227 	return 0;
228 }
229 
230 static const struct seq_operations rt_cache_seq_ops = {
231 	.start  = rt_cache_seq_start,
232 	.next   = rt_cache_seq_next,
233 	.stop   = rt_cache_seq_stop,
234 	.show   = rt_cache_seq_show,
235 };
236 
237 static int rt_cache_seq_open(struct inode *inode, struct file *file)
238 {
239 	return seq_open(file, &rt_cache_seq_ops);
240 }
241 
242 static const struct file_operations rt_cache_seq_fops = {
243 	.open	 = rt_cache_seq_open,
244 	.read	 = seq_read,
245 	.llseek	 = seq_lseek,
246 	.release = seq_release,
247 };
248 
249 
250 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
251 {
252 	int cpu;
253 
254 	if (*pos == 0)
255 		return SEQ_START_TOKEN;
256 
257 	for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
258 		if (!cpu_possible(cpu))
259 			continue;
260 		*pos = cpu+1;
261 		return &per_cpu(rt_cache_stat, cpu);
262 	}
263 	return NULL;
264 }
265 
266 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
267 {
268 	int cpu;
269 
270 	for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
271 		if (!cpu_possible(cpu))
272 			continue;
273 		*pos = cpu+1;
274 		return &per_cpu(rt_cache_stat, cpu);
275 	}
276 	return NULL;
277 
278 }
279 
280 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
281 {
282 
283 }
284 
285 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
286 {
287 	struct rt_cache_stat *st = v;
288 
289 	if (v == SEQ_START_TOKEN) {
290 		seq_printf(seq, "entries  in_hit in_slow_tot in_slow_mc in_no_route in_brd in_martian_dst in_martian_src  out_hit out_slow_tot out_slow_mc  gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
291 		return 0;
292 	}
293 
294 	seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
295 		   " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
296 		   dst_entries_get_slow(&ipv4_dst_ops),
297 		   0, /* st->in_hit */
298 		   st->in_slow_tot,
299 		   st->in_slow_mc,
300 		   st->in_no_route,
301 		   st->in_brd,
302 		   st->in_martian_dst,
303 		   st->in_martian_src,
304 
305 		   0, /* st->out_hit */
306 		   st->out_slow_tot,
307 		   st->out_slow_mc,
308 
309 		   0, /* st->gc_total */
310 		   0, /* st->gc_ignored */
311 		   0, /* st->gc_goal_miss */
312 		   0, /* st->gc_dst_overflow */
313 		   0, /* st->in_hlist_search */
314 		   0  /* st->out_hlist_search */
315 		);
316 	return 0;
317 }
318 
319 static const struct seq_operations rt_cpu_seq_ops = {
320 	.start  = rt_cpu_seq_start,
321 	.next   = rt_cpu_seq_next,
322 	.stop   = rt_cpu_seq_stop,
323 	.show   = rt_cpu_seq_show,
324 };
325 
326 
327 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
328 {
329 	return seq_open(file, &rt_cpu_seq_ops);
330 }
331 
332 static const struct file_operations rt_cpu_seq_fops = {
333 	.open	 = rt_cpu_seq_open,
334 	.read	 = seq_read,
335 	.llseek	 = seq_lseek,
336 	.release = seq_release,
337 };
338 
339 #ifdef CONFIG_IP_ROUTE_CLASSID
340 static int rt_acct_proc_show(struct seq_file *m, void *v)
341 {
342 	struct ip_rt_acct *dst, *src;
343 	unsigned int i, j;
344 
345 	dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
346 	if (!dst)
347 		return -ENOMEM;
348 
349 	for_each_possible_cpu(i) {
350 		src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
351 		for (j = 0; j < 256; j++) {
352 			dst[j].o_bytes   += src[j].o_bytes;
353 			dst[j].o_packets += src[j].o_packets;
354 			dst[j].i_bytes   += src[j].i_bytes;
355 			dst[j].i_packets += src[j].i_packets;
356 		}
357 	}
358 
359 	seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
360 	kfree(dst);
361 	return 0;
362 }
363 #endif
364 
365 static int __net_init ip_rt_do_proc_init(struct net *net)
366 {
367 	struct proc_dir_entry *pde;
368 
369 	pde = proc_create("rt_cache", 0444, net->proc_net,
370 			  &rt_cache_seq_fops);
371 	if (!pde)
372 		goto err1;
373 
374 	pde = proc_create("rt_cache", 0444,
375 			  net->proc_net_stat, &rt_cpu_seq_fops);
376 	if (!pde)
377 		goto err2;
378 
379 #ifdef CONFIG_IP_ROUTE_CLASSID
380 	pde = proc_create_single("rt_acct", 0, net->proc_net,
381 			rt_acct_proc_show);
382 	if (!pde)
383 		goto err3;
384 #endif
385 	return 0;
386 
387 #ifdef CONFIG_IP_ROUTE_CLASSID
388 err3:
389 	remove_proc_entry("rt_cache", net->proc_net_stat);
390 #endif
391 err2:
392 	remove_proc_entry("rt_cache", net->proc_net);
393 err1:
394 	return -ENOMEM;
395 }
396 
397 static void __net_exit ip_rt_do_proc_exit(struct net *net)
398 {
399 	remove_proc_entry("rt_cache", net->proc_net_stat);
400 	remove_proc_entry("rt_cache", net->proc_net);
401 #ifdef CONFIG_IP_ROUTE_CLASSID
402 	remove_proc_entry("rt_acct", net->proc_net);
403 #endif
404 }
405 
406 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
407 	.init = ip_rt_do_proc_init,
408 	.exit = ip_rt_do_proc_exit,
409 };
410 
411 static int __init ip_rt_proc_init(void)
412 {
413 	return register_pernet_subsys(&ip_rt_proc_ops);
414 }
415 
416 #else
417 static inline int ip_rt_proc_init(void)
418 {
419 	return 0;
420 }
421 #endif /* CONFIG_PROC_FS */
422 
423 static inline bool rt_is_expired(const struct rtable *rth)
424 {
425 	return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
426 }
427 
428 void rt_cache_flush(struct net *net)
429 {
430 	rt_genid_bump_ipv4(net);
431 }
432 
433 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
434 					   struct sk_buff *skb,
435 					   const void *daddr)
436 {
437 	struct net_device *dev = dst->dev;
438 	const __be32 *pkey = daddr;
439 	const struct rtable *rt;
440 	struct neighbour *n;
441 
442 	rt = (const struct rtable *) dst;
443 	if (rt->rt_gateway)
444 		pkey = (const __be32 *) &rt->rt_gateway;
445 	else if (skb)
446 		pkey = &ip_hdr(skb)->daddr;
447 
448 	n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
449 	if (n)
450 		return n;
451 	return neigh_create(&arp_tbl, pkey, dev);
452 }
453 
454 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
455 {
456 	struct net_device *dev = dst->dev;
457 	const __be32 *pkey = daddr;
458 	const struct rtable *rt;
459 
460 	rt = (const struct rtable *)dst;
461 	if (rt->rt_gateway)
462 		pkey = (const __be32 *)&rt->rt_gateway;
463 	else if (!daddr ||
464 		 (rt->rt_flags &
465 		  (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL)))
466 		return;
467 
468 	__ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
469 }
470 
471 #define IP_IDENTS_SZ 2048u
472 
473 static atomic_t *ip_idents __read_mostly;
474 static u32 *ip_tstamps __read_mostly;
475 
476 /* In order to protect privacy, we add a perturbation to identifiers
477  * if one generator is seldom used. This makes hard for an attacker
478  * to infer how many packets were sent between two points in time.
479  */
480 u32 ip_idents_reserve(u32 hash, int segs)
481 {
482 	u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
483 	atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
484 	u32 old = READ_ONCE(*p_tstamp);
485 	u32 now = (u32)jiffies;
486 	u32 new, delta = 0;
487 
488 	if (old != now && cmpxchg(p_tstamp, old, now) == old)
489 		delta = prandom_u32_max(now - old);
490 
491 	/* Do not use atomic_add_return() as it makes UBSAN unhappy */
492 	do {
493 		old = (u32)atomic_read(p_id);
494 		new = old + delta + segs;
495 	} while (atomic_cmpxchg(p_id, old, new) != old);
496 
497 	return new - segs;
498 }
499 EXPORT_SYMBOL(ip_idents_reserve);
500 
501 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
502 {
503 	static u32 ip_idents_hashrnd __read_mostly;
504 	u32 hash, id;
505 
506 	net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
507 
508 	hash = jhash_3words((__force u32)iph->daddr,
509 			    (__force u32)iph->saddr,
510 			    iph->protocol ^ net_hash_mix(net),
511 			    ip_idents_hashrnd);
512 	id = ip_idents_reserve(hash, segs);
513 	iph->id = htons(id);
514 }
515 EXPORT_SYMBOL(__ip_select_ident);
516 
517 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
518 			     const struct sock *sk,
519 			     const struct iphdr *iph,
520 			     int oif, u8 tos,
521 			     u8 prot, u32 mark, int flow_flags)
522 {
523 	if (sk) {
524 		const struct inet_sock *inet = inet_sk(sk);
525 
526 		oif = sk->sk_bound_dev_if;
527 		mark = sk->sk_mark;
528 		tos = RT_CONN_FLAGS(sk);
529 		prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
530 	}
531 	flowi4_init_output(fl4, oif, mark, tos,
532 			   RT_SCOPE_UNIVERSE, prot,
533 			   flow_flags,
534 			   iph->daddr, iph->saddr, 0, 0,
535 			   sock_net_uid(net, sk));
536 }
537 
538 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
539 			       const struct sock *sk)
540 {
541 	const struct net *net = dev_net(skb->dev);
542 	const struct iphdr *iph = ip_hdr(skb);
543 	int oif = skb->dev->ifindex;
544 	u8 tos = RT_TOS(iph->tos);
545 	u8 prot = iph->protocol;
546 	u32 mark = skb->mark;
547 
548 	__build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
549 }
550 
551 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
552 {
553 	const struct inet_sock *inet = inet_sk(sk);
554 	const struct ip_options_rcu *inet_opt;
555 	__be32 daddr = inet->inet_daddr;
556 
557 	rcu_read_lock();
558 	inet_opt = rcu_dereference(inet->inet_opt);
559 	if (inet_opt && inet_opt->opt.srr)
560 		daddr = inet_opt->opt.faddr;
561 	flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
562 			   RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
563 			   inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
564 			   inet_sk_flowi_flags(sk),
565 			   daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
566 	rcu_read_unlock();
567 }
568 
569 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
570 				 const struct sk_buff *skb)
571 {
572 	if (skb)
573 		build_skb_flow_key(fl4, skb, sk);
574 	else
575 		build_sk_flow_key(fl4, sk);
576 }
577 
578 static DEFINE_SPINLOCK(fnhe_lock);
579 
580 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
581 {
582 	struct rtable *rt;
583 
584 	rt = rcu_dereference(fnhe->fnhe_rth_input);
585 	if (rt) {
586 		RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
587 		dst_dev_put(&rt->dst);
588 		dst_release(&rt->dst);
589 	}
590 	rt = rcu_dereference(fnhe->fnhe_rth_output);
591 	if (rt) {
592 		RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
593 		dst_dev_put(&rt->dst);
594 		dst_release(&rt->dst);
595 	}
596 }
597 
598 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
599 {
600 	struct fib_nh_exception *fnhe, *oldest;
601 
602 	oldest = rcu_dereference(hash->chain);
603 	for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
604 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
605 		if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
606 			oldest = fnhe;
607 	}
608 	fnhe_flush_routes(oldest);
609 	return oldest;
610 }
611 
612 static inline u32 fnhe_hashfun(__be32 daddr)
613 {
614 	static u32 fnhe_hashrnd __read_mostly;
615 	u32 hval;
616 
617 	net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
618 	hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
619 	return hash_32(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_gateway = fnhe->fnhe_gw;
631 		rt->rt_uses_gateway = 1;
632 	}
633 }
634 
635 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
636 				  u32 pmtu, bool lock, unsigned long expires)
637 {
638 	struct fnhe_hash_bucket *hash;
639 	struct fib_nh_exception *fnhe;
640 	struct rtable *rt;
641 	u32 genid, hval;
642 	unsigned int i;
643 	int depth;
644 
645 	genid = fnhe_genid(dev_net(nh->nh_dev));
646 	hval = fnhe_hashfun(daddr);
647 
648 	spin_lock_bh(&fnhe_lock);
649 
650 	hash = rcu_dereference(nh->nh_exceptions);
651 	if (!hash) {
652 		hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
653 		if (!hash)
654 			goto out_unlock;
655 		rcu_assign_pointer(nh->nh_exceptions, hash);
656 	}
657 
658 	hash += hval;
659 
660 	depth = 0;
661 	for (fnhe = rcu_dereference(hash->chain); fnhe;
662 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
663 		if (fnhe->fnhe_daddr == daddr)
664 			break;
665 		depth++;
666 	}
667 
668 	if (fnhe) {
669 		if (fnhe->fnhe_genid != genid)
670 			fnhe->fnhe_genid = genid;
671 		if (gw)
672 			fnhe->fnhe_gw = gw;
673 		if (pmtu) {
674 			fnhe->fnhe_pmtu = pmtu;
675 			fnhe->fnhe_mtu_locked = lock;
676 		}
677 		fnhe->fnhe_expires = max(1UL, expires);
678 		/* Update all cached dsts too */
679 		rt = rcu_dereference(fnhe->fnhe_rth_input);
680 		if (rt)
681 			fill_route_from_fnhe(rt, fnhe);
682 		rt = rcu_dereference(fnhe->fnhe_rth_output);
683 		if (rt)
684 			fill_route_from_fnhe(rt, fnhe);
685 	} else {
686 		if (depth > FNHE_RECLAIM_DEPTH)
687 			fnhe = fnhe_oldest(hash);
688 		else {
689 			fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
690 			if (!fnhe)
691 				goto out_unlock;
692 
693 			fnhe->fnhe_next = hash->chain;
694 			rcu_assign_pointer(hash->chain, fnhe);
695 		}
696 		fnhe->fnhe_genid = genid;
697 		fnhe->fnhe_daddr = daddr;
698 		fnhe->fnhe_gw = gw;
699 		fnhe->fnhe_pmtu = pmtu;
700 		fnhe->fnhe_mtu_locked = lock;
701 		fnhe->fnhe_expires = max(1UL, expires);
702 
703 		/* Exception created; mark the cached routes for the nexthop
704 		 * stale, so anyone caching it rechecks if this exception
705 		 * applies to them.
706 		 */
707 		rt = rcu_dereference(nh->nh_rth_input);
708 		if (rt)
709 			rt->dst.obsolete = DST_OBSOLETE_KILL;
710 
711 		for_each_possible_cpu(i) {
712 			struct rtable __rcu **prt;
713 			prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
714 			rt = rcu_dereference(*prt);
715 			if (rt)
716 				rt->dst.obsolete = DST_OBSOLETE_KILL;
717 		}
718 	}
719 
720 	fnhe->fnhe_stamp = jiffies;
721 
722 out_unlock:
723 	spin_unlock_bh(&fnhe_lock);
724 }
725 
726 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
727 			     bool kill_route)
728 {
729 	__be32 new_gw = icmp_hdr(skb)->un.gateway;
730 	__be32 old_gw = ip_hdr(skb)->saddr;
731 	struct net_device *dev = skb->dev;
732 	struct in_device *in_dev;
733 	struct fib_result res;
734 	struct neighbour *n;
735 	struct net *net;
736 
737 	switch (icmp_hdr(skb)->code & 7) {
738 	case ICMP_REDIR_NET:
739 	case ICMP_REDIR_NETTOS:
740 	case ICMP_REDIR_HOST:
741 	case ICMP_REDIR_HOSTTOS:
742 		break;
743 
744 	default:
745 		return;
746 	}
747 
748 	if (rt->rt_gateway != old_gw)
749 		return;
750 
751 	in_dev = __in_dev_get_rcu(dev);
752 	if (!in_dev)
753 		return;
754 
755 	net = dev_net(dev);
756 	if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
757 	    ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
758 	    ipv4_is_zeronet(new_gw))
759 		goto reject_redirect;
760 
761 	if (!IN_DEV_SHARED_MEDIA(in_dev)) {
762 		if (!inet_addr_onlink(in_dev, new_gw, old_gw))
763 			goto reject_redirect;
764 		if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
765 			goto reject_redirect;
766 	} else {
767 		if (inet_addr_type(net, new_gw) != RTN_UNICAST)
768 			goto reject_redirect;
769 	}
770 
771 	n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
772 	if (!n)
773 		n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
774 	if (!IS_ERR(n)) {
775 		if (!(n->nud_state & NUD_VALID)) {
776 			neigh_event_send(n, NULL);
777 		} else {
778 			if (fib_lookup(net, fl4, &res, 0) == 0) {
779 				struct fib_nh *nh = &FIB_RES_NH(res);
780 
781 				update_or_create_fnhe(nh, fl4->daddr, new_gw,
782 						0, false,
783 						jiffies + ip_rt_gc_timeout);
784 			}
785 			if (kill_route)
786 				rt->dst.obsolete = DST_OBSOLETE_KILL;
787 			call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
788 		}
789 		neigh_release(n);
790 	}
791 	return;
792 
793 reject_redirect:
794 #ifdef CONFIG_IP_ROUTE_VERBOSE
795 	if (IN_DEV_LOG_MARTIANS(in_dev)) {
796 		const struct iphdr *iph = (const struct iphdr *) skb->data;
797 		__be32 daddr = iph->daddr;
798 		__be32 saddr = iph->saddr;
799 
800 		net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
801 				     "  Advised path = %pI4 -> %pI4\n",
802 				     &old_gw, dev->name, &new_gw,
803 				     &saddr, &daddr);
804 	}
805 #endif
806 	;
807 }
808 
809 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
810 {
811 	struct rtable *rt;
812 	struct flowi4 fl4;
813 	const struct iphdr *iph = (const struct iphdr *) skb->data;
814 	struct net *net = dev_net(skb->dev);
815 	int oif = skb->dev->ifindex;
816 	u8 tos = RT_TOS(iph->tos);
817 	u8 prot = iph->protocol;
818 	u32 mark = skb->mark;
819 
820 	rt = (struct rtable *) dst;
821 
822 	__build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
823 	__ip_do_redirect(rt, skb, &fl4, true);
824 }
825 
826 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
827 {
828 	struct rtable *rt = (struct rtable *)dst;
829 	struct dst_entry *ret = dst;
830 
831 	if (rt) {
832 		if (dst->obsolete > 0) {
833 			ip_rt_put(rt);
834 			ret = NULL;
835 		} else if ((rt->rt_flags & RTCF_REDIRECTED) ||
836 			   rt->dst.expires) {
837 			ip_rt_put(rt);
838 			ret = NULL;
839 		}
840 	}
841 	return ret;
842 }
843 
844 /*
845  * Algorithm:
846  *	1. The first ip_rt_redirect_number redirects are sent
847  *	   with exponential backoff, then we stop sending them at all,
848  *	   assuming that the host ignores our redirects.
849  *	2. If we did not see packets requiring redirects
850  *	   during ip_rt_redirect_silence, we assume that the host
851  *	   forgot redirected route and start to send redirects again.
852  *
853  * This algorithm is much cheaper and more intelligent than dumb load limiting
854  * in icmp.c.
855  *
856  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
857  * and "frag. need" (breaks PMTU discovery) in icmp.c.
858  */
859 
860 void ip_rt_send_redirect(struct sk_buff *skb)
861 {
862 	struct rtable *rt = skb_rtable(skb);
863 	struct in_device *in_dev;
864 	struct inet_peer *peer;
865 	struct net *net;
866 	int log_martians;
867 	int vif;
868 
869 	rcu_read_lock();
870 	in_dev = __in_dev_get_rcu(rt->dst.dev);
871 	if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
872 		rcu_read_unlock();
873 		return;
874 	}
875 	log_martians = IN_DEV_LOG_MARTIANS(in_dev);
876 	vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
877 	rcu_read_unlock();
878 
879 	net = dev_net(rt->dst.dev);
880 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
881 	if (!peer) {
882 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
883 			  rt_nexthop(rt, ip_hdr(skb)->daddr));
884 		return;
885 	}
886 
887 	/* No redirected packets during ip_rt_redirect_silence;
888 	 * reset the algorithm.
889 	 */
890 	if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
891 		peer->rate_tokens = 0;
892 		peer->n_redirects = 0;
893 	}
894 
895 	/* Too many ignored redirects; do not send anything
896 	 * set dst.rate_last to the last seen redirected packet.
897 	 */
898 	if (peer->n_redirects >= ip_rt_redirect_number) {
899 		peer->rate_last = jiffies;
900 		goto out_put_peer;
901 	}
902 
903 	/* Check for load limit; set rate_last to the latest sent
904 	 * redirect.
905 	 */
906 	if (peer->rate_tokens == 0 ||
907 	    time_after(jiffies,
908 		       (peer->rate_last +
909 			(ip_rt_redirect_load << peer->rate_tokens)))) {
910 		__be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
911 
912 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
913 		peer->rate_last = jiffies;
914 		++peer->rate_tokens;
915 		++peer->n_redirects;
916 #ifdef CONFIG_IP_ROUTE_VERBOSE
917 		if (log_martians &&
918 		    peer->rate_tokens == ip_rt_redirect_number)
919 			net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
920 					     &ip_hdr(skb)->saddr, inet_iif(skb),
921 					     &ip_hdr(skb)->daddr, &gw);
922 #endif
923 	}
924 out_put_peer:
925 	inet_putpeer(peer);
926 }
927 
928 static int ip_error(struct sk_buff *skb)
929 {
930 	struct rtable *rt = skb_rtable(skb);
931 	struct net_device *dev = skb->dev;
932 	struct in_device *in_dev;
933 	struct inet_peer *peer;
934 	unsigned long now;
935 	struct net *net;
936 	bool send;
937 	int code;
938 
939 	if (netif_is_l3_master(skb->dev)) {
940 		dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
941 		if (!dev)
942 			goto out;
943 	}
944 
945 	in_dev = __in_dev_get_rcu(dev);
946 
947 	/* IP on this device is disabled. */
948 	if (!in_dev)
949 		goto out;
950 
951 	net = dev_net(rt->dst.dev);
952 	if (!IN_DEV_FORWARD(in_dev)) {
953 		switch (rt->dst.error) {
954 		case EHOSTUNREACH:
955 			__IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
956 			break;
957 
958 		case ENETUNREACH:
959 			__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
960 			break;
961 		}
962 		goto out;
963 	}
964 
965 	switch (rt->dst.error) {
966 	case EINVAL:
967 	default:
968 		goto out;
969 	case EHOSTUNREACH:
970 		code = ICMP_HOST_UNREACH;
971 		break;
972 	case ENETUNREACH:
973 		code = ICMP_NET_UNREACH;
974 		__IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
975 		break;
976 	case EACCES:
977 		code = ICMP_PKT_FILTERED;
978 		break;
979 	}
980 
981 	peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
982 			       l3mdev_master_ifindex(skb->dev), 1);
983 
984 	send = true;
985 	if (peer) {
986 		now = jiffies;
987 		peer->rate_tokens += now - peer->rate_last;
988 		if (peer->rate_tokens > ip_rt_error_burst)
989 			peer->rate_tokens = ip_rt_error_burst;
990 		peer->rate_last = now;
991 		if (peer->rate_tokens >= ip_rt_error_cost)
992 			peer->rate_tokens -= ip_rt_error_cost;
993 		else
994 			send = false;
995 		inet_putpeer(peer);
996 	}
997 	if (send)
998 		icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
999 
1000 out:	kfree_skb(skb);
1001 	return 0;
1002 }
1003 
1004 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1005 {
1006 	struct dst_entry *dst = &rt->dst;
1007 	u32 old_mtu = ipv4_mtu(dst);
1008 	struct fib_result res;
1009 	bool lock = false;
1010 
1011 	if (ip_mtu_locked(dst))
1012 		return;
1013 
1014 	if (old_mtu < mtu)
1015 		return;
1016 
1017 	if (mtu < ip_rt_min_pmtu) {
1018 		lock = true;
1019 		mtu = min(old_mtu, ip_rt_min_pmtu);
1020 	}
1021 
1022 	if (rt->rt_pmtu == mtu && !lock &&
1023 	    time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1024 		return;
1025 
1026 	rcu_read_lock();
1027 	if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
1028 		struct fib_nh *nh = &FIB_RES_NH(res);
1029 
1030 		update_or_create_fnhe(nh, fl4->daddr, 0, mtu, lock,
1031 				      jiffies + ip_rt_mtu_expires);
1032 	}
1033 	rcu_read_unlock();
1034 }
1035 
1036 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1037 			      struct sk_buff *skb, u32 mtu)
1038 {
1039 	struct rtable *rt = (struct rtable *) dst;
1040 	struct flowi4 fl4;
1041 
1042 	ip_rt_build_flow_key(&fl4, sk, skb);
1043 	__ip_rt_update_pmtu(rt, &fl4, mtu);
1044 }
1045 
1046 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1047 		      int oif, u8 protocol)
1048 {
1049 	const struct iphdr *iph = (const struct iphdr *) skb->data;
1050 	struct flowi4 fl4;
1051 	struct rtable *rt;
1052 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
1053 
1054 	__build_flow_key(net, &fl4, NULL, iph, oif,
1055 			 RT_TOS(iph->tos), protocol, mark, 0);
1056 	rt = __ip_route_output_key(net, &fl4);
1057 	if (!IS_ERR(rt)) {
1058 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1059 		ip_rt_put(rt);
1060 	}
1061 }
1062 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1063 
1064 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1065 {
1066 	const struct iphdr *iph = (const struct iphdr *) skb->data;
1067 	struct flowi4 fl4;
1068 	struct rtable *rt;
1069 
1070 	__build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1071 
1072 	if (!fl4.flowi4_mark)
1073 		fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1074 
1075 	rt = __ip_route_output_key(sock_net(sk), &fl4);
1076 	if (!IS_ERR(rt)) {
1077 		__ip_rt_update_pmtu(rt, &fl4, mtu);
1078 		ip_rt_put(rt);
1079 	}
1080 }
1081 
1082 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1083 {
1084 	const struct iphdr *iph = (const struct iphdr *) skb->data;
1085 	struct flowi4 fl4;
1086 	struct rtable *rt;
1087 	struct dst_entry *odst = NULL;
1088 	bool new = false;
1089 	struct net *net = sock_net(sk);
1090 
1091 	bh_lock_sock(sk);
1092 
1093 	if (!ip_sk_accept_pmtu(sk))
1094 		goto out;
1095 
1096 	odst = sk_dst_get(sk);
1097 
1098 	if (sock_owned_by_user(sk) || !odst) {
1099 		__ipv4_sk_update_pmtu(skb, sk, mtu);
1100 		goto out;
1101 	}
1102 
1103 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1104 
1105 	rt = (struct rtable *)odst;
1106 	if (odst->obsolete && !odst->ops->check(odst, 0)) {
1107 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1108 		if (IS_ERR(rt))
1109 			goto out;
1110 
1111 		new = true;
1112 	}
1113 
1114 	__ip_rt_update_pmtu((struct rtable *) xfrm_dst_path(&rt->dst), &fl4, mtu);
1115 
1116 	if (!dst_check(&rt->dst, 0)) {
1117 		if (new)
1118 			dst_release(&rt->dst);
1119 
1120 		rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1121 		if (IS_ERR(rt))
1122 			goto out;
1123 
1124 		new = true;
1125 	}
1126 
1127 	if (new)
1128 		sk_dst_set(sk, &rt->dst);
1129 
1130 out:
1131 	bh_unlock_sock(sk);
1132 	dst_release(odst);
1133 }
1134 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1135 
1136 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1137 		   int oif, u8 protocol)
1138 {
1139 	const struct iphdr *iph = (const struct iphdr *) skb->data;
1140 	struct flowi4 fl4;
1141 	struct rtable *rt;
1142 
1143 	__build_flow_key(net, &fl4, NULL, iph, oif,
1144 			 RT_TOS(iph->tos), protocol, 0, 0);
1145 	rt = __ip_route_output_key(net, &fl4);
1146 	if (!IS_ERR(rt)) {
1147 		__ip_do_redirect(rt, skb, &fl4, false);
1148 		ip_rt_put(rt);
1149 	}
1150 }
1151 EXPORT_SYMBOL_GPL(ipv4_redirect);
1152 
1153 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1154 {
1155 	const struct iphdr *iph = (const struct iphdr *) skb->data;
1156 	struct flowi4 fl4;
1157 	struct rtable *rt;
1158 	struct net *net = sock_net(sk);
1159 
1160 	__build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1161 	rt = __ip_route_output_key(net, &fl4);
1162 	if (!IS_ERR(rt)) {
1163 		__ip_do_redirect(rt, skb, &fl4, false);
1164 		ip_rt_put(rt);
1165 	}
1166 }
1167 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1168 
1169 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1170 {
1171 	struct rtable *rt = (struct rtable *) dst;
1172 
1173 	/* All IPV4 dsts are created with ->obsolete set to the value
1174 	 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1175 	 * into this function always.
1176 	 *
1177 	 * When a PMTU/redirect information update invalidates a route,
1178 	 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1179 	 * DST_OBSOLETE_DEAD by dst_free().
1180 	 */
1181 	if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1182 		return NULL;
1183 	return dst;
1184 }
1185 
1186 static void ipv4_link_failure(struct sk_buff *skb)
1187 {
1188 	struct rtable *rt;
1189 
1190 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1191 
1192 	rt = skb_rtable(skb);
1193 	if (rt)
1194 		dst_set_expires(&rt->dst, 0);
1195 }
1196 
1197 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1198 {
1199 	pr_debug("%s: %pI4 -> %pI4, %s\n",
1200 		 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1201 		 skb->dev ? skb->dev->name : "?");
1202 	kfree_skb(skb);
1203 	WARN_ON(1);
1204 	return 0;
1205 }
1206 
1207 /*
1208    We do not cache source address of outgoing interface,
1209    because it is used only by IP RR, TS and SRR options,
1210    so that it out of fast path.
1211 
1212    BTW remember: "addr" is allowed to be not aligned
1213    in IP options!
1214  */
1215 
1216 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1217 {
1218 	__be32 src;
1219 
1220 	if (rt_is_output_route(rt))
1221 		src = ip_hdr(skb)->saddr;
1222 	else {
1223 		struct fib_result res;
1224 		struct iphdr *iph = ip_hdr(skb);
1225 		struct flowi4 fl4 = {
1226 			.daddr = iph->daddr,
1227 			.saddr = iph->saddr,
1228 			.flowi4_tos = RT_TOS(iph->tos),
1229 			.flowi4_oif = rt->dst.dev->ifindex,
1230 			.flowi4_iif = skb->dev->ifindex,
1231 			.flowi4_mark = skb->mark,
1232 		};
1233 
1234 		rcu_read_lock();
1235 		if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1236 			src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1237 		else
1238 			src = inet_select_addr(rt->dst.dev,
1239 					       rt_nexthop(rt, iph->daddr),
1240 					       RT_SCOPE_UNIVERSE);
1241 		rcu_read_unlock();
1242 	}
1243 	memcpy(addr, &src, 4);
1244 }
1245 
1246 #ifdef CONFIG_IP_ROUTE_CLASSID
1247 static void set_class_tag(struct rtable *rt, u32 tag)
1248 {
1249 	if (!(rt->dst.tclassid & 0xFFFF))
1250 		rt->dst.tclassid |= tag & 0xFFFF;
1251 	if (!(rt->dst.tclassid & 0xFFFF0000))
1252 		rt->dst.tclassid |= tag & 0xFFFF0000;
1253 }
1254 #endif
1255 
1256 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1257 {
1258 	unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1259 	unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1260 				    ip_rt_min_advmss);
1261 
1262 	return min(advmss, IPV4_MAX_PMTU - header_size);
1263 }
1264 
1265 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1266 {
1267 	const struct rtable *rt = (const struct rtable *) dst;
1268 	unsigned int mtu = rt->rt_pmtu;
1269 
1270 	if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1271 		mtu = dst_metric_raw(dst, RTAX_MTU);
1272 
1273 	if (mtu)
1274 		return mtu;
1275 
1276 	mtu = READ_ONCE(dst->dev->mtu);
1277 
1278 	if (unlikely(ip_mtu_locked(dst))) {
1279 		if (rt->rt_uses_gateway && mtu > 576)
1280 			mtu = 576;
1281 	}
1282 
1283 	mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1284 
1285 	return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1286 }
1287 
1288 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1289 {
1290 	struct fnhe_hash_bucket *hash;
1291 	struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1292 	u32 hval = fnhe_hashfun(daddr);
1293 
1294 	spin_lock_bh(&fnhe_lock);
1295 
1296 	hash = rcu_dereference_protected(nh->nh_exceptions,
1297 					 lockdep_is_held(&fnhe_lock));
1298 	hash += hval;
1299 
1300 	fnhe_p = &hash->chain;
1301 	fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1302 	while (fnhe) {
1303 		if (fnhe->fnhe_daddr == daddr) {
1304 			rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1305 				fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1306 			fnhe_flush_routes(fnhe);
1307 			kfree_rcu(fnhe, rcu);
1308 			break;
1309 		}
1310 		fnhe_p = &fnhe->fnhe_next;
1311 		fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1312 						 lockdep_is_held(&fnhe_lock));
1313 	}
1314 
1315 	spin_unlock_bh(&fnhe_lock);
1316 }
1317 
1318 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1319 {
1320 	struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1321 	struct fib_nh_exception *fnhe;
1322 	u32 hval;
1323 
1324 	if (!hash)
1325 		return NULL;
1326 
1327 	hval = fnhe_hashfun(daddr);
1328 
1329 	for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1330 	     fnhe = rcu_dereference(fnhe->fnhe_next)) {
1331 		if (fnhe->fnhe_daddr == daddr) {
1332 			if (fnhe->fnhe_expires &&
1333 			    time_after(jiffies, fnhe->fnhe_expires)) {
1334 				ip_del_fnhe(nh, daddr);
1335 				break;
1336 			}
1337 			return fnhe;
1338 		}
1339 	}
1340 	return NULL;
1341 }
1342 
1343 /* MTU selection:
1344  * 1. mtu on route is locked - use it
1345  * 2. mtu from nexthop exception
1346  * 3. mtu from egress device
1347  */
1348 
1349 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1350 {
1351 	struct fib_info *fi = res->fi;
1352 	struct fib_nh *nh = &fi->fib_nh[res->nh_sel];
1353 	struct net_device *dev = nh->nh_dev;
1354 	u32 mtu = 0;
1355 
1356 	if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu ||
1357 	    fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1358 		mtu = fi->fib_mtu;
1359 
1360 	if (likely(!mtu)) {
1361 		struct fib_nh_exception *fnhe;
1362 
1363 		fnhe = find_exception(nh, daddr);
1364 		if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1365 			mtu = fnhe->fnhe_pmtu;
1366 	}
1367 
1368 	if (likely(!mtu))
1369 		mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1370 
1371 	return mtu - lwtunnel_headroom(nh->nh_lwtstate, mtu);
1372 }
1373 
1374 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1375 			      __be32 daddr, const bool do_cache)
1376 {
1377 	bool ret = false;
1378 
1379 	spin_lock_bh(&fnhe_lock);
1380 
1381 	if (daddr == fnhe->fnhe_daddr) {
1382 		struct rtable __rcu **porig;
1383 		struct rtable *orig;
1384 		int genid = fnhe_genid(dev_net(rt->dst.dev));
1385 
1386 		if (rt_is_input_route(rt))
1387 			porig = &fnhe->fnhe_rth_input;
1388 		else
1389 			porig = &fnhe->fnhe_rth_output;
1390 		orig = rcu_dereference(*porig);
1391 
1392 		if (fnhe->fnhe_genid != genid) {
1393 			fnhe->fnhe_genid = genid;
1394 			fnhe->fnhe_gw = 0;
1395 			fnhe->fnhe_pmtu = 0;
1396 			fnhe->fnhe_expires = 0;
1397 			fnhe->fnhe_mtu_locked = false;
1398 			fnhe_flush_routes(fnhe);
1399 			orig = NULL;
1400 		}
1401 		fill_route_from_fnhe(rt, fnhe);
1402 		if (!rt->rt_gateway)
1403 			rt->rt_gateway = daddr;
1404 
1405 		if (do_cache) {
1406 			dst_hold(&rt->dst);
1407 			rcu_assign_pointer(*porig, rt);
1408 			if (orig) {
1409 				dst_dev_put(&orig->dst);
1410 				dst_release(&orig->dst);
1411 			}
1412 			ret = true;
1413 		}
1414 
1415 		fnhe->fnhe_stamp = jiffies;
1416 	}
1417 	spin_unlock_bh(&fnhe_lock);
1418 
1419 	return ret;
1420 }
1421 
1422 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1423 {
1424 	struct rtable *orig, *prev, **p;
1425 	bool ret = true;
1426 
1427 	if (rt_is_input_route(rt)) {
1428 		p = (struct rtable **)&nh->nh_rth_input;
1429 	} else {
1430 		p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1431 	}
1432 	orig = *p;
1433 
1434 	/* hold dst before doing cmpxchg() to avoid race condition
1435 	 * on this dst
1436 	 */
1437 	dst_hold(&rt->dst);
1438 	prev = cmpxchg(p, orig, rt);
1439 	if (prev == orig) {
1440 		if (orig) {
1441 			dst_dev_put(&orig->dst);
1442 			dst_release(&orig->dst);
1443 		}
1444 	} else {
1445 		dst_release(&rt->dst);
1446 		ret = false;
1447 	}
1448 
1449 	return ret;
1450 }
1451 
1452 struct uncached_list {
1453 	spinlock_t		lock;
1454 	struct list_head	head;
1455 };
1456 
1457 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1458 
1459 void rt_add_uncached_list(struct rtable *rt)
1460 {
1461 	struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1462 
1463 	rt->rt_uncached_list = ul;
1464 
1465 	spin_lock_bh(&ul->lock);
1466 	list_add_tail(&rt->rt_uncached, &ul->head);
1467 	spin_unlock_bh(&ul->lock);
1468 }
1469 
1470 void rt_del_uncached_list(struct rtable *rt)
1471 {
1472 	if (!list_empty(&rt->rt_uncached)) {
1473 		struct uncached_list *ul = rt->rt_uncached_list;
1474 
1475 		spin_lock_bh(&ul->lock);
1476 		list_del(&rt->rt_uncached);
1477 		spin_unlock_bh(&ul->lock);
1478 	}
1479 }
1480 
1481 static void ipv4_dst_destroy(struct dst_entry *dst)
1482 {
1483 	struct rtable *rt = (struct rtable *)dst;
1484 
1485 	ip_dst_metrics_put(dst);
1486 	rt_del_uncached_list(rt);
1487 }
1488 
1489 void rt_flush_dev(struct net_device *dev)
1490 {
1491 	struct net *net = dev_net(dev);
1492 	struct rtable *rt;
1493 	int cpu;
1494 
1495 	for_each_possible_cpu(cpu) {
1496 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1497 
1498 		spin_lock_bh(&ul->lock);
1499 		list_for_each_entry(rt, &ul->head, rt_uncached) {
1500 			if (rt->dst.dev != dev)
1501 				continue;
1502 			rt->dst.dev = net->loopback_dev;
1503 			dev_hold(rt->dst.dev);
1504 			dev_put(dev);
1505 		}
1506 		spin_unlock_bh(&ul->lock);
1507 	}
1508 }
1509 
1510 static bool rt_cache_valid(const struct rtable *rt)
1511 {
1512 	return	rt &&
1513 		rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1514 		!rt_is_expired(rt);
1515 }
1516 
1517 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1518 			   const struct fib_result *res,
1519 			   struct fib_nh_exception *fnhe,
1520 			   struct fib_info *fi, u16 type, u32 itag,
1521 			   const bool do_cache)
1522 {
1523 	bool cached = false;
1524 
1525 	if (fi) {
1526 		struct fib_nh *nh = &FIB_RES_NH(*res);
1527 
1528 		if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1529 			rt->rt_gateway = nh->nh_gw;
1530 			rt->rt_uses_gateway = 1;
1531 		}
1532 		ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1533 
1534 #ifdef CONFIG_IP_ROUTE_CLASSID
1535 		rt->dst.tclassid = nh->nh_tclassid;
1536 #endif
1537 		rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1538 		if (unlikely(fnhe))
1539 			cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1540 		else if (do_cache)
1541 			cached = rt_cache_route(nh, rt);
1542 		if (unlikely(!cached)) {
1543 			/* Routes we intend to cache in nexthop exception or
1544 			 * FIB nexthop have the DST_NOCACHE bit clear.
1545 			 * However, if we are unsuccessful at storing this
1546 			 * route into the cache we really need to set it.
1547 			 */
1548 			if (!rt->rt_gateway)
1549 				rt->rt_gateway = daddr;
1550 			rt_add_uncached_list(rt);
1551 		}
1552 	} else
1553 		rt_add_uncached_list(rt);
1554 
1555 #ifdef CONFIG_IP_ROUTE_CLASSID
1556 #ifdef CONFIG_IP_MULTIPLE_TABLES
1557 	set_class_tag(rt, res->tclassid);
1558 #endif
1559 	set_class_tag(rt, itag);
1560 #endif
1561 }
1562 
1563 struct rtable *rt_dst_alloc(struct net_device *dev,
1564 			    unsigned int flags, u16 type,
1565 			    bool nopolicy, bool noxfrm, bool will_cache)
1566 {
1567 	struct rtable *rt;
1568 
1569 	rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1570 		       (will_cache ? 0 : DST_HOST) |
1571 		       (nopolicy ? DST_NOPOLICY : 0) |
1572 		       (noxfrm ? DST_NOXFRM : 0));
1573 
1574 	if (rt) {
1575 		rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1576 		rt->rt_flags = flags;
1577 		rt->rt_type = type;
1578 		rt->rt_is_input = 0;
1579 		rt->rt_iif = 0;
1580 		rt->rt_pmtu = 0;
1581 		rt->rt_mtu_locked = 0;
1582 		rt->rt_gateway = 0;
1583 		rt->rt_uses_gateway = 0;
1584 		INIT_LIST_HEAD(&rt->rt_uncached);
1585 
1586 		rt->dst.output = ip_output;
1587 		if (flags & RTCF_LOCAL)
1588 			rt->dst.input = ip_local_deliver;
1589 	}
1590 
1591 	return rt;
1592 }
1593 EXPORT_SYMBOL(rt_dst_alloc);
1594 
1595 /* called in rcu_read_lock() section */
1596 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1597 			  u8 tos, struct net_device *dev,
1598 			  struct in_device *in_dev, u32 *itag)
1599 {
1600 	int err;
1601 
1602 	/* Primary sanity checks. */
1603 	if (!in_dev)
1604 		return -EINVAL;
1605 
1606 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1607 	    skb->protocol != htons(ETH_P_IP))
1608 		return -EINVAL;
1609 
1610 	if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1611 		return -EINVAL;
1612 
1613 	if (ipv4_is_zeronet(saddr)) {
1614 		if (!ipv4_is_local_multicast(daddr) &&
1615 		    ip_hdr(skb)->protocol != IPPROTO_IGMP)
1616 			return -EINVAL;
1617 	} else {
1618 		err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1619 					  in_dev, itag);
1620 		if (err < 0)
1621 			return err;
1622 	}
1623 	return 0;
1624 }
1625 
1626 /* called in rcu_read_lock() section */
1627 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1628 			     u8 tos, struct net_device *dev, int our)
1629 {
1630 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1631 	unsigned int flags = RTCF_MULTICAST;
1632 	struct rtable *rth;
1633 	u32 itag = 0;
1634 	int err;
1635 
1636 	err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1637 	if (err)
1638 		return err;
1639 
1640 	if (our)
1641 		flags |= RTCF_LOCAL;
1642 
1643 	rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1644 			   IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1645 	if (!rth)
1646 		return -ENOBUFS;
1647 
1648 #ifdef CONFIG_IP_ROUTE_CLASSID
1649 	rth->dst.tclassid = itag;
1650 #endif
1651 	rth->dst.output = ip_rt_bug;
1652 	rth->rt_is_input= 1;
1653 
1654 #ifdef CONFIG_IP_MROUTE
1655 	if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1656 		rth->dst.input = ip_mr_input;
1657 #endif
1658 	RT_CACHE_STAT_INC(in_slow_mc);
1659 
1660 	skb_dst_set(skb, &rth->dst);
1661 	return 0;
1662 }
1663 
1664 
1665 static void ip_handle_martian_source(struct net_device *dev,
1666 				     struct in_device *in_dev,
1667 				     struct sk_buff *skb,
1668 				     __be32 daddr,
1669 				     __be32 saddr)
1670 {
1671 	RT_CACHE_STAT_INC(in_martian_src);
1672 #ifdef CONFIG_IP_ROUTE_VERBOSE
1673 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1674 		/*
1675 		 *	RFC1812 recommendation, if source is martian,
1676 		 *	the only hint is MAC header.
1677 		 */
1678 		pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1679 			&daddr, &saddr, dev->name);
1680 		if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1681 			print_hex_dump(KERN_WARNING, "ll header: ",
1682 				       DUMP_PREFIX_OFFSET, 16, 1,
1683 				       skb_mac_header(skb),
1684 				       dev->hard_header_len, false);
1685 		}
1686 	}
1687 #endif
1688 }
1689 
1690 /* called in rcu_read_lock() section */
1691 static int __mkroute_input(struct sk_buff *skb,
1692 			   const struct fib_result *res,
1693 			   struct in_device *in_dev,
1694 			   __be32 daddr, __be32 saddr, u32 tos)
1695 {
1696 	struct fib_nh_exception *fnhe;
1697 	struct rtable *rth;
1698 	int err;
1699 	struct in_device *out_dev;
1700 	bool do_cache;
1701 	u32 itag = 0;
1702 
1703 	/* get a working reference to the output device */
1704 	out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1705 	if (!out_dev) {
1706 		net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1707 		return -EINVAL;
1708 	}
1709 
1710 	err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1711 				  in_dev->dev, in_dev, &itag);
1712 	if (err < 0) {
1713 		ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1714 					 saddr);
1715 
1716 		goto cleanup;
1717 	}
1718 
1719 	do_cache = res->fi && !itag;
1720 	if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1721 	    skb->protocol == htons(ETH_P_IP) &&
1722 	    (IN_DEV_SHARED_MEDIA(out_dev) ||
1723 	     inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1724 		IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1725 
1726 	if (skb->protocol != htons(ETH_P_IP)) {
1727 		/* Not IP (i.e. ARP). Do not create route, if it is
1728 		 * invalid for proxy arp. DNAT routes are always valid.
1729 		 *
1730 		 * Proxy arp feature have been extended to allow, ARP
1731 		 * replies back to the same interface, to support
1732 		 * Private VLAN switch technologies. See arp.c.
1733 		 */
1734 		if (out_dev == in_dev &&
1735 		    IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1736 			err = -EINVAL;
1737 			goto cleanup;
1738 		}
1739 	}
1740 
1741 	fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1742 	if (do_cache) {
1743 		if (fnhe)
1744 			rth = rcu_dereference(fnhe->fnhe_rth_input);
1745 		else
1746 			rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1747 		if (rt_cache_valid(rth)) {
1748 			skb_dst_set_noref(skb, &rth->dst);
1749 			goto out;
1750 		}
1751 	}
1752 
1753 	rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1754 			   IN_DEV_CONF_GET(in_dev, NOPOLICY),
1755 			   IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1756 	if (!rth) {
1757 		err = -ENOBUFS;
1758 		goto cleanup;
1759 	}
1760 
1761 	rth->rt_is_input = 1;
1762 	RT_CACHE_STAT_INC(in_slow_tot);
1763 
1764 	rth->dst.input = ip_forward;
1765 
1766 	rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1767 		       do_cache);
1768 	lwtunnel_set_redirect(&rth->dst);
1769 	skb_dst_set(skb, &rth->dst);
1770 out:
1771 	err = 0;
1772  cleanup:
1773 	return err;
1774 }
1775 
1776 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1777 /* To make ICMP packets follow the right flow, the multipath hash is
1778  * calculated from the inner IP addresses.
1779  */
1780 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1781 				 struct flow_keys *hash_keys)
1782 {
1783 	const struct iphdr *outer_iph = ip_hdr(skb);
1784 	const struct iphdr *key_iph = outer_iph;
1785 	const struct iphdr *inner_iph;
1786 	const struct icmphdr *icmph;
1787 	struct iphdr _inner_iph;
1788 	struct icmphdr _icmph;
1789 
1790 	if (likely(outer_iph->protocol != IPPROTO_ICMP))
1791 		goto out;
1792 
1793 	if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1794 		goto out;
1795 
1796 	icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1797 				   &_icmph);
1798 	if (!icmph)
1799 		goto out;
1800 
1801 	if (icmph->type != ICMP_DEST_UNREACH &&
1802 	    icmph->type != ICMP_REDIRECT &&
1803 	    icmph->type != ICMP_TIME_EXCEEDED &&
1804 	    icmph->type != ICMP_PARAMETERPROB)
1805 		goto out;
1806 
1807 	inner_iph = skb_header_pointer(skb,
1808 				       outer_iph->ihl * 4 + sizeof(_icmph),
1809 				       sizeof(_inner_iph), &_inner_iph);
1810 	if (!inner_iph)
1811 		goto out;
1812 
1813 	key_iph = inner_iph;
1814 out:
1815 	hash_keys->addrs.v4addrs.src = key_iph->saddr;
1816 	hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1817 }
1818 
1819 /* if skb is set it will be used and fl4 can be NULL */
1820 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
1821 		       const struct sk_buff *skb, struct flow_keys *flkeys)
1822 {
1823 	struct flow_keys hash_keys;
1824 	u32 mhash;
1825 
1826 	switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1827 	case 0:
1828 		memset(&hash_keys, 0, sizeof(hash_keys));
1829 		hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1830 		if (skb) {
1831 			ip_multipath_l3_keys(skb, &hash_keys);
1832 		} else {
1833 			hash_keys.addrs.v4addrs.src = fl4->saddr;
1834 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
1835 		}
1836 		break;
1837 	case 1:
1838 		/* skb is currently provided only when forwarding */
1839 		if (skb) {
1840 			unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1841 			struct flow_keys keys;
1842 
1843 			/* short-circuit if we already have L4 hash present */
1844 			if (skb->l4_hash)
1845 				return skb_get_hash_raw(skb) >> 1;
1846 
1847 			memset(&hash_keys, 0, sizeof(hash_keys));
1848 
1849 			if (!flkeys) {
1850 				skb_flow_dissect_flow_keys(skb, &keys, flag);
1851 				flkeys = &keys;
1852 			}
1853 
1854 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1855 			hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
1856 			hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
1857 			hash_keys.ports.src = flkeys->ports.src;
1858 			hash_keys.ports.dst = flkeys->ports.dst;
1859 			hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
1860 		} else {
1861 			memset(&hash_keys, 0, sizeof(hash_keys));
1862 			hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1863 			hash_keys.addrs.v4addrs.src = fl4->saddr;
1864 			hash_keys.addrs.v4addrs.dst = fl4->daddr;
1865 			hash_keys.ports.src = fl4->fl4_sport;
1866 			hash_keys.ports.dst = fl4->fl4_dport;
1867 			hash_keys.basic.ip_proto = fl4->flowi4_proto;
1868 		}
1869 		break;
1870 	}
1871 	mhash = flow_hash_from_keys(&hash_keys);
1872 
1873 	return mhash >> 1;
1874 }
1875 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1876 
1877 static int ip_mkroute_input(struct sk_buff *skb,
1878 			    struct fib_result *res,
1879 			    struct in_device *in_dev,
1880 			    __be32 daddr, __be32 saddr, u32 tos,
1881 			    struct flow_keys *hkeys)
1882 {
1883 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1884 	if (res->fi && res->fi->fib_nhs > 1) {
1885 		int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
1886 
1887 		fib_select_multipath(res, h);
1888 	}
1889 #endif
1890 
1891 	/* create a routing cache entry */
1892 	return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1893 }
1894 
1895 /*
1896  *	NOTE. We drop all the packets that has local source
1897  *	addresses, because every properly looped back packet
1898  *	must have correct destination already attached by output routine.
1899  *
1900  *	Such approach solves two big problems:
1901  *	1. Not simplex devices are handled properly.
1902  *	2. IP spoofing attempts are filtered with 100% of guarantee.
1903  *	called with rcu_read_lock()
1904  */
1905 
1906 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1907 			       u8 tos, struct net_device *dev,
1908 			       struct fib_result *res)
1909 {
1910 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1911 	struct flow_keys *flkeys = NULL, _flkeys;
1912 	struct net    *net = dev_net(dev);
1913 	struct ip_tunnel_info *tun_info;
1914 	int		err = -EINVAL;
1915 	unsigned int	flags = 0;
1916 	u32		itag = 0;
1917 	struct rtable	*rth;
1918 	struct flowi4	fl4;
1919 	bool do_cache;
1920 
1921 	/* IP on this device is disabled. */
1922 
1923 	if (!in_dev)
1924 		goto out;
1925 
1926 	/* Check for the most weird martians, which can be not detected
1927 	   by fib_lookup.
1928 	 */
1929 
1930 	tun_info = skb_tunnel_info(skb);
1931 	if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1932 		fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1933 	else
1934 		fl4.flowi4_tun_key.tun_id = 0;
1935 	skb_dst_drop(skb);
1936 
1937 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1938 		goto martian_source;
1939 
1940 	res->fi = NULL;
1941 	res->table = NULL;
1942 	if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1943 		goto brd_input;
1944 
1945 	/* Accept zero addresses only to limited broadcast;
1946 	 * I even do not know to fix it or not. Waiting for complains :-)
1947 	 */
1948 	if (ipv4_is_zeronet(saddr))
1949 		goto martian_source;
1950 
1951 	if (ipv4_is_zeronet(daddr))
1952 		goto martian_destination;
1953 
1954 	/* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1955 	 * and call it once if daddr or/and saddr are loopback addresses
1956 	 */
1957 	if (ipv4_is_loopback(daddr)) {
1958 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1959 			goto martian_destination;
1960 	} else if (ipv4_is_loopback(saddr)) {
1961 		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1962 			goto martian_source;
1963 	}
1964 
1965 	/*
1966 	 *	Now we are ready to route packet.
1967 	 */
1968 	fl4.flowi4_oif = 0;
1969 	fl4.flowi4_iif = dev->ifindex;
1970 	fl4.flowi4_mark = skb->mark;
1971 	fl4.flowi4_tos = tos;
1972 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1973 	fl4.flowi4_flags = 0;
1974 	fl4.daddr = daddr;
1975 	fl4.saddr = saddr;
1976 	fl4.flowi4_uid = sock_net_uid(net, NULL);
1977 
1978 	if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
1979 		flkeys = &_flkeys;
1980 	} else {
1981 		fl4.flowi4_proto = 0;
1982 		fl4.fl4_sport = 0;
1983 		fl4.fl4_dport = 0;
1984 	}
1985 
1986 	err = fib_lookup(net, &fl4, res, 0);
1987 	if (err != 0) {
1988 		if (!IN_DEV_FORWARD(in_dev))
1989 			err = -EHOSTUNREACH;
1990 		goto no_route;
1991 	}
1992 
1993 	if (res->type == RTN_BROADCAST) {
1994 		if (IN_DEV_BFORWARD(in_dev))
1995 			goto make_route;
1996 		goto brd_input;
1997 	}
1998 
1999 	if (res->type == RTN_LOCAL) {
2000 		err = fib_validate_source(skb, saddr, daddr, tos,
2001 					  0, dev, in_dev, &itag);
2002 		if (err < 0)
2003 			goto martian_source;
2004 		goto local_input;
2005 	}
2006 
2007 	if (!IN_DEV_FORWARD(in_dev)) {
2008 		err = -EHOSTUNREACH;
2009 		goto no_route;
2010 	}
2011 	if (res->type != RTN_UNICAST)
2012 		goto martian_destination;
2013 
2014 make_route:
2015 	err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2016 out:	return err;
2017 
2018 brd_input:
2019 	if (skb->protocol != htons(ETH_P_IP))
2020 		goto e_inval;
2021 
2022 	if (!ipv4_is_zeronet(saddr)) {
2023 		err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2024 					  in_dev, &itag);
2025 		if (err < 0)
2026 			goto martian_source;
2027 	}
2028 	flags |= RTCF_BROADCAST;
2029 	res->type = RTN_BROADCAST;
2030 	RT_CACHE_STAT_INC(in_brd);
2031 
2032 local_input:
2033 	do_cache = false;
2034 	if (res->fi) {
2035 		if (!itag) {
2036 			rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
2037 			if (rt_cache_valid(rth)) {
2038 				skb_dst_set_noref(skb, &rth->dst);
2039 				err = 0;
2040 				goto out;
2041 			}
2042 			do_cache = true;
2043 		}
2044 	}
2045 
2046 	rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2047 			   flags | RTCF_LOCAL, res->type,
2048 			   IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
2049 	if (!rth)
2050 		goto e_nobufs;
2051 
2052 	rth->dst.output= ip_rt_bug;
2053 #ifdef CONFIG_IP_ROUTE_CLASSID
2054 	rth->dst.tclassid = itag;
2055 #endif
2056 	rth->rt_is_input = 1;
2057 
2058 	RT_CACHE_STAT_INC(in_slow_tot);
2059 	if (res->type == RTN_UNREACHABLE) {
2060 		rth->dst.input= ip_error;
2061 		rth->dst.error= -err;
2062 		rth->rt_flags 	&= ~RTCF_LOCAL;
2063 	}
2064 
2065 	if (do_cache) {
2066 		struct fib_nh *nh = &FIB_RES_NH(*res);
2067 
2068 		rth->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
2069 		if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2070 			WARN_ON(rth->dst.input == lwtunnel_input);
2071 			rth->dst.lwtstate->orig_input = rth->dst.input;
2072 			rth->dst.input = lwtunnel_input;
2073 		}
2074 
2075 		if (unlikely(!rt_cache_route(nh, rth)))
2076 			rt_add_uncached_list(rth);
2077 	}
2078 	skb_dst_set(skb, &rth->dst);
2079 	err = 0;
2080 	goto out;
2081 
2082 no_route:
2083 	RT_CACHE_STAT_INC(in_no_route);
2084 	res->type = RTN_UNREACHABLE;
2085 	res->fi = NULL;
2086 	res->table = NULL;
2087 	goto local_input;
2088 
2089 	/*
2090 	 *	Do not cache martian addresses: they should be logged (RFC1812)
2091 	 */
2092 martian_destination:
2093 	RT_CACHE_STAT_INC(in_martian_dst);
2094 #ifdef CONFIG_IP_ROUTE_VERBOSE
2095 	if (IN_DEV_LOG_MARTIANS(in_dev))
2096 		net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2097 				     &daddr, &saddr, dev->name);
2098 #endif
2099 
2100 e_inval:
2101 	err = -EINVAL;
2102 	goto out;
2103 
2104 e_nobufs:
2105 	err = -ENOBUFS;
2106 	goto out;
2107 
2108 martian_source:
2109 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2110 	goto out;
2111 }
2112 
2113 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2114 			 u8 tos, struct net_device *dev)
2115 {
2116 	struct fib_result res;
2117 	int err;
2118 
2119 	tos &= IPTOS_RT_MASK;
2120 	rcu_read_lock();
2121 	err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2122 	rcu_read_unlock();
2123 
2124 	return err;
2125 }
2126 EXPORT_SYMBOL(ip_route_input_noref);
2127 
2128 /* called with rcu_read_lock held */
2129 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2130 		       u8 tos, struct net_device *dev, struct fib_result *res)
2131 {
2132 	/* Multicast recognition logic is moved from route cache to here.
2133 	   The problem was that too many Ethernet cards have broken/missing
2134 	   hardware multicast filters :-( As result the host on multicasting
2135 	   network acquires a lot of useless route cache entries, sort of
2136 	   SDR messages from all the world. Now we try to get rid of them.
2137 	   Really, provided software IP multicast filter is organized
2138 	   reasonably (at least, hashed), it does not result in a slowdown
2139 	   comparing with route cache reject entries.
2140 	   Note, that multicast routers are not affected, because
2141 	   route cache entry is created eventually.
2142 	 */
2143 	if (ipv4_is_multicast(daddr)) {
2144 		struct in_device *in_dev = __in_dev_get_rcu(dev);
2145 		int our = 0;
2146 		int err = -EINVAL;
2147 
2148 		if (in_dev)
2149 			our = ip_check_mc_rcu(in_dev, daddr, saddr,
2150 					      ip_hdr(skb)->protocol);
2151 
2152 		/* check l3 master if no match yet */
2153 		if ((!in_dev || !our) && netif_is_l3_slave(dev)) {
2154 			struct in_device *l3_in_dev;
2155 
2156 			l3_in_dev = __in_dev_get_rcu(skb->dev);
2157 			if (l3_in_dev)
2158 				our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2159 						      ip_hdr(skb)->protocol);
2160 		}
2161 
2162 		if (our
2163 #ifdef CONFIG_IP_MROUTE
2164 			||
2165 		    (!ipv4_is_local_multicast(daddr) &&
2166 		     IN_DEV_MFORWARD(in_dev))
2167 #endif
2168 		   ) {
2169 			err = ip_route_input_mc(skb, daddr, saddr,
2170 						tos, dev, our);
2171 		}
2172 		return err;
2173 	}
2174 
2175 	return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2176 }
2177 
2178 /* called with rcu_read_lock() */
2179 static struct rtable *__mkroute_output(const struct fib_result *res,
2180 				       const struct flowi4 *fl4, int orig_oif,
2181 				       struct net_device *dev_out,
2182 				       unsigned int flags)
2183 {
2184 	struct fib_info *fi = res->fi;
2185 	struct fib_nh_exception *fnhe;
2186 	struct in_device *in_dev;
2187 	u16 type = res->type;
2188 	struct rtable *rth;
2189 	bool do_cache;
2190 
2191 	in_dev = __in_dev_get_rcu(dev_out);
2192 	if (!in_dev)
2193 		return ERR_PTR(-EINVAL);
2194 
2195 	if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2196 		if (ipv4_is_loopback(fl4->saddr) &&
2197 		    !(dev_out->flags & IFF_LOOPBACK) &&
2198 		    !netif_is_l3_master(dev_out))
2199 			return ERR_PTR(-EINVAL);
2200 
2201 	if (ipv4_is_lbcast(fl4->daddr))
2202 		type = RTN_BROADCAST;
2203 	else if (ipv4_is_multicast(fl4->daddr))
2204 		type = RTN_MULTICAST;
2205 	else if (ipv4_is_zeronet(fl4->daddr))
2206 		return ERR_PTR(-EINVAL);
2207 
2208 	if (dev_out->flags & IFF_LOOPBACK)
2209 		flags |= RTCF_LOCAL;
2210 
2211 	do_cache = true;
2212 	if (type == RTN_BROADCAST) {
2213 		flags |= RTCF_BROADCAST | RTCF_LOCAL;
2214 		fi = NULL;
2215 	} else if (type == RTN_MULTICAST) {
2216 		flags |= RTCF_MULTICAST | RTCF_LOCAL;
2217 		if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2218 				     fl4->flowi4_proto))
2219 			flags &= ~RTCF_LOCAL;
2220 		else
2221 			do_cache = false;
2222 		/* If multicast route do not exist use
2223 		 * default one, but do not gateway in this case.
2224 		 * Yes, it is hack.
2225 		 */
2226 		if (fi && res->prefixlen < 4)
2227 			fi = NULL;
2228 	} else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2229 		   (orig_oif != dev_out->ifindex)) {
2230 		/* For local routes that require a particular output interface
2231 		 * we do not want to cache the result.  Caching the result
2232 		 * causes incorrect behaviour when there are multiple source
2233 		 * addresses on the interface, the end result being that if the
2234 		 * intended recipient is waiting on that interface for the
2235 		 * packet he won't receive it because it will be delivered on
2236 		 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2237 		 * be set to the loopback interface as well.
2238 		 */
2239 		do_cache = false;
2240 	}
2241 
2242 	fnhe = NULL;
2243 	do_cache &= fi != NULL;
2244 	if (fi) {
2245 		struct rtable __rcu **prth;
2246 		struct fib_nh *nh = &FIB_RES_NH(*res);
2247 
2248 		fnhe = find_exception(nh, fl4->daddr);
2249 		if (!do_cache)
2250 			goto add;
2251 		if (fnhe) {
2252 			prth = &fnhe->fnhe_rth_output;
2253 		} else {
2254 			if (unlikely(fl4->flowi4_flags &
2255 				     FLOWI_FLAG_KNOWN_NH &&
2256 				     !(nh->nh_gw &&
2257 				       nh->nh_scope == RT_SCOPE_LINK))) {
2258 				do_cache = false;
2259 				goto add;
2260 			}
2261 			prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2262 		}
2263 		rth = rcu_dereference(*prth);
2264 		if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2265 			return rth;
2266 	}
2267 
2268 add:
2269 	rth = rt_dst_alloc(dev_out, flags, type,
2270 			   IN_DEV_CONF_GET(in_dev, NOPOLICY),
2271 			   IN_DEV_CONF_GET(in_dev, NOXFRM),
2272 			   do_cache);
2273 	if (!rth)
2274 		return ERR_PTR(-ENOBUFS);
2275 
2276 	rth->rt_iif = orig_oif;
2277 
2278 	RT_CACHE_STAT_INC(out_slow_tot);
2279 
2280 	if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2281 		if (flags & RTCF_LOCAL &&
2282 		    !(dev_out->flags & IFF_LOOPBACK)) {
2283 			rth->dst.output = ip_mc_output;
2284 			RT_CACHE_STAT_INC(out_slow_mc);
2285 		}
2286 #ifdef CONFIG_IP_MROUTE
2287 		if (type == RTN_MULTICAST) {
2288 			if (IN_DEV_MFORWARD(in_dev) &&
2289 			    !ipv4_is_local_multicast(fl4->daddr)) {
2290 				rth->dst.input = ip_mr_input;
2291 				rth->dst.output = ip_mc_output;
2292 			}
2293 		}
2294 #endif
2295 	}
2296 
2297 	rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2298 	lwtunnel_set_redirect(&rth->dst);
2299 
2300 	return rth;
2301 }
2302 
2303 /*
2304  * Major route resolver routine.
2305  */
2306 
2307 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2308 					const struct sk_buff *skb)
2309 {
2310 	__u8 tos = RT_FL_TOS(fl4);
2311 	struct fib_result res = {
2312 		.type		= RTN_UNSPEC,
2313 		.fi		= NULL,
2314 		.table		= NULL,
2315 		.tclassid	= 0,
2316 	};
2317 	struct rtable *rth;
2318 
2319 	fl4->flowi4_iif = LOOPBACK_IFINDEX;
2320 	fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2321 	fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2322 			 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2323 
2324 	rcu_read_lock();
2325 	rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2326 	rcu_read_unlock();
2327 
2328 	return rth;
2329 }
2330 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2331 
2332 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2333 					    struct fib_result *res,
2334 					    const struct sk_buff *skb)
2335 {
2336 	struct net_device *dev_out = NULL;
2337 	int orig_oif = fl4->flowi4_oif;
2338 	unsigned int flags = 0;
2339 	struct rtable *rth;
2340 	int err = -ENETUNREACH;
2341 
2342 	if (fl4->saddr) {
2343 		rth = ERR_PTR(-EINVAL);
2344 		if (ipv4_is_multicast(fl4->saddr) ||
2345 		    ipv4_is_lbcast(fl4->saddr) ||
2346 		    ipv4_is_zeronet(fl4->saddr))
2347 			goto out;
2348 
2349 		/* I removed check for oif == dev_out->oif here.
2350 		   It was wrong for two reasons:
2351 		   1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2352 		      is assigned to multiple interfaces.
2353 		   2. Moreover, we are allowed to send packets with saddr
2354 		      of another iface. --ANK
2355 		 */
2356 
2357 		if (fl4->flowi4_oif == 0 &&
2358 		    (ipv4_is_multicast(fl4->daddr) ||
2359 		     ipv4_is_lbcast(fl4->daddr))) {
2360 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2361 			dev_out = __ip_dev_find(net, fl4->saddr, false);
2362 			if (!dev_out)
2363 				goto out;
2364 
2365 			/* Special hack: user can direct multicasts
2366 			   and limited broadcast via necessary interface
2367 			   without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2368 			   This hack is not just for fun, it allows
2369 			   vic,vat and friends to work.
2370 			   They bind socket to loopback, set ttl to zero
2371 			   and expect that it will work.
2372 			   From the viewpoint of routing cache they are broken,
2373 			   because we are not allowed to build multicast path
2374 			   with loopback source addr (look, routing cache
2375 			   cannot know, that ttl is zero, so that packet
2376 			   will not leave this host and route is valid).
2377 			   Luckily, this hack is good workaround.
2378 			 */
2379 
2380 			fl4->flowi4_oif = dev_out->ifindex;
2381 			goto make_route;
2382 		}
2383 
2384 		if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2385 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2386 			if (!__ip_dev_find(net, fl4->saddr, false))
2387 				goto out;
2388 		}
2389 	}
2390 
2391 
2392 	if (fl4->flowi4_oif) {
2393 		dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2394 		rth = ERR_PTR(-ENODEV);
2395 		if (!dev_out)
2396 			goto out;
2397 
2398 		/* RACE: Check return value of inet_select_addr instead. */
2399 		if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2400 			rth = ERR_PTR(-ENETUNREACH);
2401 			goto out;
2402 		}
2403 		if (ipv4_is_local_multicast(fl4->daddr) ||
2404 		    ipv4_is_lbcast(fl4->daddr) ||
2405 		    fl4->flowi4_proto == IPPROTO_IGMP) {
2406 			if (!fl4->saddr)
2407 				fl4->saddr = inet_select_addr(dev_out, 0,
2408 							      RT_SCOPE_LINK);
2409 			goto make_route;
2410 		}
2411 		if (!fl4->saddr) {
2412 			if (ipv4_is_multicast(fl4->daddr))
2413 				fl4->saddr = inet_select_addr(dev_out, 0,
2414 							      fl4->flowi4_scope);
2415 			else if (!fl4->daddr)
2416 				fl4->saddr = inet_select_addr(dev_out, 0,
2417 							      RT_SCOPE_HOST);
2418 		}
2419 	}
2420 
2421 	if (!fl4->daddr) {
2422 		fl4->daddr = fl4->saddr;
2423 		if (!fl4->daddr)
2424 			fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2425 		dev_out = net->loopback_dev;
2426 		fl4->flowi4_oif = LOOPBACK_IFINDEX;
2427 		res->type = RTN_LOCAL;
2428 		flags |= RTCF_LOCAL;
2429 		goto make_route;
2430 	}
2431 
2432 	err = fib_lookup(net, fl4, res, 0);
2433 	if (err) {
2434 		res->fi = NULL;
2435 		res->table = NULL;
2436 		if (fl4->flowi4_oif &&
2437 		    (ipv4_is_multicast(fl4->daddr) ||
2438 		    !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2439 			/* Apparently, routing tables are wrong. Assume,
2440 			   that the destination is on link.
2441 
2442 			   WHY? DW.
2443 			   Because we are allowed to send to iface
2444 			   even if it has NO routes and NO assigned
2445 			   addresses. When oif is specified, routing
2446 			   tables are looked up with only one purpose:
2447 			   to catch if destination is gatewayed, rather than
2448 			   direct. Moreover, if MSG_DONTROUTE is set,
2449 			   we send packet, ignoring both routing tables
2450 			   and ifaddr state. --ANK
2451 
2452 
2453 			   We could make it even if oif is unknown,
2454 			   likely IPv6, but we do not.
2455 			 */
2456 
2457 			if (fl4->saddr == 0)
2458 				fl4->saddr = inet_select_addr(dev_out, 0,
2459 							      RT_SCOPE_LINK);
2460 			res->type = RTN_UNICAST;
2461 			goto make_route;
2462 		}
2463 		rth = ERR_PTR(err);
2464 		goto out;
2465 	}
2466 
2467 	if (res->type == RTN_LOCAL) {
2468 		if (!fl4->saddr) {
2469 			if (res->fi->fib_prefsrc)
2470 				fl4->saddr = res->fi->fib_prefsrc;
2471 			else
2472 				fl4->saddr = fl4->daddr;
2473 		}
2474 
2475 		/* L3 master device is the loopback for that domain */
2476 		dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2477 			net->loopback_dev;
2478 
2479 		/* make sure orig_oif points to fib result device even
2480 		 * though packet rx/tx happens over loopback or l3mdev
2481 		 */
2482 		orig_oif = FIB_RES_OIF(*res);
2483 
2484 		fl4->flowi4_oif = dev_out->ifindex;
2485 		flags |= RTCF_LOCAL;
2486 		goto make_route;
2487 	}
2488 
2489 	fib_select_path(net, res, fl4, skb);
2490 
2491 	dev_out = FIB_RES_DEV(*res);
2492 	fl4->flowi4_oif = dev_out->ifindex;
2493 
2494 
2495 make_route:
2496 	rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2497 
2498 out:
2499 	return rth;
2500 }
2501 
2502 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2503 {
2504 	return NULL;
2505 }
2506 
2507 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2508 {
2509 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2510 
2511 	return mtu ? : dst->dev->mtu;
2512 }
2513 
2514 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2515 					  struct sk_buff *skb, u32 mtu)
2516 {
2517 }
2518 
2519 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2520 				       struct sk_buff *skb)
2521 {
2522 }
2523 
2524 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2525 					  unsigned long old)
2526 {
2527 	return NULL;
2528 }
2529 
2530 static struct dst_ops ipv4_dst_blackhole_ops = {
2531 	.family			=	AF_INET,
2532 	.check			=	ipv4_blackhole_dst_check,
2533 	.mtu			=	ipv4_blackhole_mtu,
2534 	.default_advmss		=	ipv4_default_advmss,
2535 	.update_pmtu		=	ipv4_rt_blackhole_update_pmtu,
2536 	.redirect		=	ipv4_rt_blackhole_redirect,
2537 	.cow_metrics		=	ipv4_rt_blackhole_cow_metrics,
2538 	.neigh_lookup		=	ipv4_neigh_lookup,
2539 };
2540 
2541 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2542 {
2543 	struct rtable *ort = (struct rtable *) dst_orig;
2544 	struct rtable *rt;
2545 
2546 	rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2547 	if (rt) {
2548 		struct dst_entry *new = &rt->dst;
2549 
2550 		new->__use = 1;
2551 		new->input = dst_discard;
2552 		new->output = dst_discard_out;
2553 
2554 		new->dev = net->loopback_dev;
2555 		if (new->dev)
2556 			dev_hold(new->dev);
2557 
2558 		rt->rt_is_input = ort->rt_is_input;
2559 		rt->rt_iif = ort->rt_iif;
2560 		rt->rt_pmtu = ort->rt_pmtu;
2561 		rt->rt_mtu_locked = ort->rt_mtu_locked;
2562 
2563 		rt->rt_genid = rt_genid_ipv4(net);
2564 		rt->rt_flags = ort->rt_flags;
2565 		rt->rt_type = ort->rt_type;
2566 		rt->rt_gateway = ort->rt_gateway;
2567 		rt->rt_uses_gateway = ort->rt_uses_gateway;
2568 
2569 		INIT_LIST_HEAD(&rt->rt_uncached);
2570 	}
2571 
2572 	dst_release(dst_orig);
2573 
2574 	return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2575 }
2576 
2577 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2578 				    const struct sock *sk)
2579 {
2580 	struct rtable *rt = __ip_route_output_key(net, flp4);
2581 
2582 	if (IS_ERR(rt))
2583 		return rt;
2584 
2585 	if (flp4->flowi4_proto)
2586 		rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2587 							flowi4_to_flowi(flp4),
2588 							sk, 0);
2589 
2590 	return rt;
2591 }
2592 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2593 
2594 /* called with rcu_read_lock held */
2595 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2596 			struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2597 			struct sk_buff *skb, u32 portid, u32 seq)
2598 {
2599 	struct rtmsg *r;
2600 	struct nlmsghdr *nlh;
2601 	unsigned long expires = 0;
2602 	u32 error;
2603 	u32 metrics[RTAX_MAX];
2604 
2605 	nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0);
2606 	if (!nlh)
2607 		return -EMSGSIZE;
2608 
2609 	r = nlmsg_data(nlh);
2610 	r->rtm_family	 = AF_INET;
2611 	r->rtm_dst_len	= 32;
2612 	r->rtm_src_len	= 0;
2613 	r->rtm_tos	= fl4->flowi4_tos;
2614 	r->rtm_table	= table_id < 256 ? table_id : RT_TABLE_COMPAT;
2615 	if (nla_put_u32(skb, RTA_TABLE, table_id))
2616 		goto nla_put_failure;
2617 	r->rtm_type	= rt->rt_type;
2618 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2619 	r->rtm_protocol = RTPROT_UNSPEC;
2620 	r->rtm_flags	= (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2621 	if (rt->rt_flags & RTCF_NOTIFY)
2622 		r->rtm_flags |= RTM_F_NOTIFY;
2623 	if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2624 		r->rtm_flags |= RTCF_DOREDIRECT;
2625 
2626 	if (nla_put_in_addr(skb, RTA_DST, dst))
2627 		goto nla_put_failure;
2628 	if (src) {
2629 		r->rtm_src_len = 32;
2630 		if (nla_put_in_addr(skb, RTA_SRC, src))
2631 			goto nla_put_failure;
2632 	}
2633 	if (rt->dst.dev &&
2634 	    nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2635 		goto nla_put_failure;
2636 #ifdef CONFIG_IP_ROUTE_CLASSID
2637 	if (rt->dst.tclassid &&
2638 	    nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2639 		goto nla_put_failure;
2640 #endif
2641 	if (!rt_is_input_route(rt) &&
2642 	    fl4->saddr != src) {
2643 		if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2644 			goto nla_put_failure;
2645 	}
2646 	if (rt->rt_uses_gateway &&
2647 	    nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2648 		goto nla_put_failure;
2649 
2650 	expires = rt->dst.expires;
2651 	if (expires) {
2652 		unsigned long now = jiffies;
2653 
2654 		if (time_before(now, expires))
2655 			expires -= now;
2656 		else
2657 			expires = 0;
2658 	}
2659 
2660 	memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2661 	if (rt->rt_pmtu && expires)
2662 		metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2663 	if (rt->rt_mtu_locked && expires)
2664 		metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2665 	if (rtnetlink_put_metrics(skb, metrics) < 0)
2666 		goto nla_put_failure;
2667 
2668 	if (fl4->flowi4_mark &&
2669 	    nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2670 		goto nla_put_failure;
2671 
2672 	if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2673 	    nla_put_u32(skb, RTA_UID,
2674 			from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2675 		goto nla_put_failure;
2676 
2677 	error = rt->dst.error;
2678 
2679 	if (rt_is_input_route(rt)) {
2680 #ifdef CONFIG_IP_MROUTE
2681 		if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2682 		    IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2683 			int err = ipmr_get_route(net, skb,
2684 						 fl4->saddr, fl4->daddr,
2685 						 r, portid);
2686 
2687 			if (err <= 0) {
2688 				if (err == 0)
2689 					return 0;
2690 				goto nla_put_failure;
2691 			}
2692 		} else
2693 #endif
2694 			if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
2695 				goto nla_put_failure;
2696 	}
2697 
2698 	if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2699 		goto nla_put_failure;
2700 
2701 	nlmsg_end(skb, nlh);
2702 	return 0;
2703 
2704 nla_put_failure:
2705 	nlmsg_cancel(skb, nlh);
2706 	return -EMSGSIZE;
2707 }
2708 
2709 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
2710 						   u8 ip_proto, __be16 sport,
2711 						   __be16 dport)
2712 {
2713 	struct sk_buff *skb;
2714 	struct iphdr *iph;
2715 
2716 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2717 	if (!skb)
2718 		return NULL;
2719 
2720 	/* Reserve room for dummy headers, this skb can pass
2721 	 * through good chunk of routing engine.
2722 	 */
2723 	skb_reset_mac_header(skb);
2724 	skb_reset_network_header(skb);
2725 	skb->protocol = htons(ETH_P_IP);
2726 	iph = skb_put(skb, sizeof(struct iphdr));
2727 	iph->protocol = ip_proto;
2728 	iph->saddr = src;
2729 	iph->daddr = dst;
2730 	iph->version = 0x4;
2731 	iph->frag_off = 0;
2732 	iph->ihl = 0x5;
2733 	skb_set_transport_header(skb, skb->len);
2734 
2735 	switch (iph->protocol) {
2736 	case IPPROTO_UDP: {
2737 		struct udphdr *udph;
2738 
2739 		udph = skb_put_zero(skb, sizeof(struct udphdr));
2740 		udph->source = sport;
2741 		udph->dest = dport;
2742 		udph->len = sizeof(struct udphdr);
2743 		udph->check = 0;
2744 		break;
2745 	}
2746 	case IPPROTO_TCP: {
2747 		struct tcphdr *tcph;
2748 
2749 		tcph = skb_put_zero(skb, sizeof(struct tcphdr));
2750 		tcph->source	= sport;
2751 		tcph->dest	= dport;
2752 		tcph->doff	= sizeof(struct tcphdr) / 4;
2753 		tcph->rst = 1;
2754 		tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
2755 					    src, dst, 0);
2756 		break;
2757 	}
2758 	case IPPROTO_ICMP: {
2759 		struct icmphdr *icmph;
2760 
2761 		icmph = skb_put_zero(skb, sizeof(struct icmphdr));
2762 		icmph->type = ICMP_ECHO;
2763 		icmph->code = 0;
2764 	}
2765 	}
2766 
2767 	return skb;
2768 }
2769 
2770 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
2771 				       const struct nlmsghdr *nlh,
2772 				       struct nlattr **tb,
2773 				       struct netlink_ext_ack *extack)
2774 {
2775 	struct rtmsg *rtm;
2776 	int i, err;
2777 
2778 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2779 		NL_SET_ERR_MSG(extack,
2780 			       "ipv4: Invalid header for route get request");
2781 		return -EINVAL;
2782 	}
2783 
2784 	if (!netlink_strict_get_check(skb))
2785 		return nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX,
2786 				   rtm_ipv4_policy, extack);
2787 
2788 	rtm = nlmsg_data(nlh);
2789 	if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
2790 	    (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
2791 	    rtm->rtm_table || rtm->rtm_protocol ||
2792 	    rtm->rtm_scope || rtm->rtm_type) {
2793 		NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
2794 		return -EINVAL;
2795 	}
2796 
2797 	if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
2798 			       RTM_F_LOOKUP_TABLE |
2799 			       RTM_F_FIB_MATCH)) {
2800 		NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
2801 		return -EINVAL;
2802 	}
2803 
2804 	err = nlmsg_parse_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2805 				 rtm_ipv4_policy, extack);
2806 	if (err)
2807 		return err;
2808 
2809 	if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
2810 	    (tb[RTA_DST] && !rtm->rtm_dst_len)) {
2811 		NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
2812 		return -EINVAL;
2813 	}
2814 
2815 	for (i = 0; i <= RTA_MAX; i++) {
2816 		if (!tb[i])
2817 			continue;
2818 
2819 		switch (i) {
2820 		case RTA_IIF:
2821 		case RTA_OIF:
2822 		case RTA_SRC:
2823 		case RTA_DST:
2824 		case RTA_IP_PROTO:
2825 		case RTA_SPORT:
2826 		case RTA_DPORT:
2827 		case RTA_MARK:
2828 		case RTA_UID:
2829 			break;
2830 		default:
2831 			NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
2832 			return -EINVAL;
2833 		}
2834 	}
2835 
2836 	return 0;
2837 }
2838 
2839 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2840 			     struct netlink_ext_ack *extack)
2841 {
2842 	struct net *net = sock_net(in_skb->sk);
2843 	struct nlattr *tb[RTA_MAX+1];
2844 	u32 table_id = RT_TABLE_MAIN;
2845 	__be16 sport = 0, dport = 0;
2846 	struct fib_result res = {};
2847 	u8 ip_proto = IPPROTO_UDP;
2848 	struct rtable *rt = NULL;
2849 	struct sk_buff *skb;
2850 	struct rtmsg *rtm;
2851 	struct flowi4 fl4 = {};
2852 	__be32 dst = 0;
2853 	__be32 src = 0;
2854 	kuid_t uid;
2855 	u32 iif;
2856 	int err;
2857 	int mark;
2858 
2859 	err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
2860 	if (err < 0)
2861 		return err;
2862 
2863 	rtm = nlmsg_data(nlh);
2864 	src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2865 	dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2866 	iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2867 	mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2868 	if (tb[RTA_UID])
2869 		uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2870 	else
2871 		uid = (iif ? INVALID_UID : current_uid());
2872 
2873 	if (tb[RTA_IP_PROTO]) {
2874 		err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
2875 						  &ip_proto, extack);
2876 		if (err)
2877 			return err;
2878 	}
2879 
2880 	if (tb[RTA_SPORT])
2881 		sport = nla_get_be16(tb[RTA_SPORT]);
2882 
2883 	if (tb[RTA_DPORT])
2884 		dport = nla_get_be16(tb[RTA_DPORT]);
2885 
2886 	skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
2887 	if (!skb)
2888 		return -ENOBUFS;
2889 
2890 	fl4.daddr = dst;
2891 	fl4.saddr = src;
2892 	fl4.flowi4_tos = rtm->rtm_tos;
2893 	fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2894 	fl4.flowi4_mark = mark;
2895 	fl4.flowi4_uid = uid;
2896 	if (sport)
2897 		fl4.fl4_sport = sport;
2898 	if (dport)
2899 		fl4.fl4_dport = dport;
2900 	fl4.flowi4_proto = ip_proto;
2901 
2902 	rcu_read_lock();
2903 
2904 	if (iif) {
2905 		struct net_device *dev;
2906 
2907 		dev = dev_get_by_index_rcu(net, iif);
2908 		if (!dev) {
2909 			err = -ENODEV;
2910 			goto errout_rcu;
2911 		}
2912 
2913 		fl4.flowi4_iif = iif; /* for rt_fill_info */
2914 		skb->dev	= dev;
2915 		skb->mark	= mark;
2916 		err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos,
2917 					 dev, &res);
2918 
2919 		rt = skb_rtable(skb);
2920 		if (err == 0 && rt->dst.error)
2921 			err = -rt->dst.error;
2922 	} else {
2923 		fl4.flowi4_iif = LOOPBACK_IFINDEX;
2924 		skb->dev = net->loopback_dev;
2925 		rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
2926 		err = 0;
2927 		if (IS_ERR(rt))
2928 			err = PTR_ERR(rt);
2929 		else
2930 			skb_dst_set(skb, &rt->dst);
2931 	}
2932 
2933 	if (err)
2934 		goto errout_rcu;
2935 
2936 	if (rtm->rtm_flags & RTM_F_NOTIFY)
2937 		rt->rt_flags |= RTCF_NOTIFY;
2938 
2939 	if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2940 		table_id = res.table ? res.table->tb_id : 0;
2941 
2942 	/* reset skb for netlink reply msg */
2943 	skb_trim(skb, 0);
2944 	skb_reset_network_header(skb);
2945 	skb_reset_transport_header(skb);
2946 	skb_reset_mac_header(skb);
2947 
2948 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2949 		if (!res.fi) {
2950 			err = fib_props[res.type].error;
2951 			if (!err)
2952 				err = -EHOSTUNREACH;
2953 			goto errout_rcu;
2954 		}
2955 		err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
2956 				    nlh->nlmsg_seq, RTM_NEWROUTE, table_id,
2957 				    rt->rt_type, res.prefix, res.prefixlen,
2958 				    fl4.flowi4_tos, res.fi, 0);
2959 	} else {
2960 		err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
2961 				   NETLINK_CB(in_skb).portid, nlh->nlmsg_seq);
2962 	}
2963 	if (err < 0)
2964 		goto errout_rcu;
2965 
2966 	rcu_read_unlock();
2967 
2968 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2969 
2970 errout_free:
2971 	return err;
2972 errout_rcu:
2973 	rcu_read_unlock();
2974 	kfree_skb(skb);
2975 	goto errout_free;
2976 }
2977 
2978 void ip_rt_multicast_event(struct in_device *in_dev)
2979 {
2980 	rt_cache_flush(dev_net(in_dev->dev));
2981 }
2982 
2983 #ifdef CONFIG_SYSCTL
2984 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
2985 static int ip_rt_gc_min_interval __read_mostly	= HZ / 2;
2986 static int ip_rt_gc_elasticity __read_mostly	= 8;
2987 static int ip_min_valid_pmtu __read_mostly	= IPV4_MIN_MTU;
2988 
2989 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2990 					void __user *buffer,
2991 					size_t *lenp, loff_t *ppos)
2992 {
2993 	struct net *net = (struct net *)__ctl->extra1;
2994 
2995 	if (write) {
2996 		rt_cache_flush(net);
2997 		fnhe_genid_bump(net);
2998 		return 0;
2999 	}
3000 
3001 	return -EINVAL;
3002 }
3003 
3004 static struct ctl_table ipv4_route_table[] = {
3005 	{
3006 		.procname	= "gc_thresh",
3007 		.data		= &ipv4_dst_ops.gc_thresh,
3008 		.maxlen		= sizeof(int),
3009 		.mode		= 0644,
3010 		.proc_handler	= proc_dointvec,
3011 	},
3012 	{
3013 		.procname	= "max_size",
3014 		.data		= &ip_rt_max_size,
3015 		.maxlen		= sizeof(int),
3016 		.mode		= 0644,
3017 		.proc_handler	= proc_dointvec,
3018 	},
3019 	{
3020 		/*  Deprecated. Use gc_min_interval_ms */
3021 
3022 		.procname	= "gc_min_interval",
3023 		.data		= &ip_rt_gc_min_interval,
3024 		.maxlen		= sizeof(int),
3025 		.mode		= 0644,
3026 		.proc_handler	= proc_dointvec_jiffies,
3027 	},
3028 	{
3029 		.procname	= "gc_min_interval_ms",
3030 		.data		= &ip_rt_gc_min_interval,
3031 		.maxlen		= sizeof(int),
3032 		.mode		= 0644,
3033 		.proc_handler	= proc_dointvec_ms_jiffies,
3034 	},
3035 	{
3036 		.procname	= "gc_timeout",
3037 		.data		= &ip_rt_gc_timeout,
3038 		.maxlen		= sizeof(int),
3039 		.mode		= 0644,
3040 		.proc_handler	= proc_dointvec_jiffies,
3041 	},
3042 	{
3043 		.procname	= "gc_interval",
3044 		.data		= &ip_rt_gc_interval,
3045 		.maxlen		= sizeof(int),
3046 		.mode		= 0644,
3047 		.proc_handler	= proc_dointvec_jiffies,
3048 	},
3049 	{
3050 		.procname	= "redirect_load",
3051 		.data		= &ip_rt_redirect_load,
3052 		.maxlen		= sizeof(int),
3053 		.mode		= 0644,
3054 		.proc_handler	= proc_dointvec,
3055 	},
3056 	{
3057 		.procname	= "redirect_number",
3058 		.data		= &ip_rt_redirect_number,
3059 		.maxlen		= sizeof(int),
3060 		.mode		= 0644,
3061 		.proc_handler	= proc_dointvec,
3062 	},
3063 	{
3064 		.procname	= "redirect_silence",
3065 		.data		= &ip_rt_redirect_silence,
3066 		.maxlen		= sizeof(int),
3067 		.mode		= 0644,
3068 		.proc_handler	= proc_dointvec,
3069 	},
3070 	{
3071 		.procname	= "error_cost",
3072 		.data		= &ip_rt_error_cost,
3073 		.maxlen		= sizeof(int),
3074 		.mode		= 0644,
3075 		.proc_handler	= proc_dointvec,
3076 	},
3077 	{
3078 		.procname	= "error_burst",
3079 		.data		= &ip_rt_error_burst,
3080 		.maxlen		= sizeof(int),
3081 		.mode		= 0644,
3082 		.proc_handler	= proc_dointvec,
3083 	},
3084 	{
3085 		.procname	= "gc_elasticity",
3086 		.data		= &ip_rt_gc_elasticity,
3087 		.maxlen		= sizeof(int),
3088 		.mode		= 0644,
3089 		.proc_handler	= proc_dointvec,
3090 	},
3091 	{
3092 		.procname	= "mtu_expires",
3093 		.data		= &ip_rt_mtu_expires,
3094 		.maxlen		= sizeof(int),
3095 		.mode		= 0644,
3096 		.proc_handler	= proc_dointvec_jiffies,
3097 	},
3098 	{
3099 		.procname	= "min_pmtu",
3100 		.data		= &ip_rt_min_pmtu,
3101 		.maxlen		= sizeof(int),
3102 		.mode		= 0644,
3103 		.proc_handler	= proc_dointvec_minmax,
3104 		.extra1		= &ip_min_valid_pmtu,
3105 	},
3106 	{
3107 		.procname	= "min_adv_mss",
3108 		.data		= &ip_rt_min_advmss,
3109 		.maxlen		= sizeof(int),
3110 		.mode		= 0644,
3111 		.proc_handler	= proc_dointvec,
3112 	},
3113 	{ }
3114 };
3115 
3116 static struct ctl_table ipv4_route_flush_table[] = {
3117 	{
3118 		.procname	= "flush",
3119 		.maxlen		= sizeof(int),
3120 		.mode		= 0200,
3121 		.proc_handler	= ipv4_sysctl_rtcache_flush,
3122 	},
3123 	{ },
3124 };
3125 
3126 static __net_init int sysctl_route_net_init(struct net *net)
3127 {
3128 	struct ctl_table *tbl;
3129 
3130 	tbl = ipv4_route_flush_table;
3131 	if (!net_eq(net, &init_net)) {
3132 		tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3133 		if (!tbl)
3134 			goto err_dup;
3135 
3136 		/* Don't export sysctls to unprivileged users */
3137 		if (net->user_ns != &init_user_ns)
3138 			tbl[0].procname = NULL;
3139 	}
3140 	tbl[0].extra1 = net;
3141 
3142 	net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3143 	if (!net->ipv4.route_hdr)
3144 		goto err_reg;
3145 	return 0;
3146 
3147 err_reg:
3148 	if (tbl != ipv4_route_flush_table)
3149 		kfree(tbl);
3150 err_dup:
3151 	return -ENOMEM;
3152 }
3153 
3154 static __net_exit void sysctl_route_net_exit(struct net *net)
3155 {
3156 	struct ctl_table *tbl;
3157 
3158 	tbl = net->ipv4.route_hdr->ctl_table_arg;
3159 	unregister_net_sysctl_table(net->ipv4.route_hdr);
3160 	BUG_ON(tbl == ipv4_route_flush_table);
3161 	kfree(tbl);
3162 }
3163 
3164 static __net_initdata struct pernet_operations sysctl_route_ops = {
3165 	.init = sysctl_route_net_init,
3166 	.exit = sysctl_route_net_exit,
3167 };
3168 #endif
3169 
3170 static __net_init int rt_genid_init(struct net *net)
3171 {
3172 	atomic_set(&net->ipv4.rt_genid, 0);
3173 	atomic_set(&net->fnhe_genid, 0);
3174 	atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3175 	return 0;
3176 }
3177 
3178 static __net_initdata struct pernet_operations rt_genid_ops = {
3179 	.init = rt_genid_init,
3180 };
3181 
3182 static int __net_init ipv4_inetpeer_init(struct net *net)
3183 {
3184 	struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3185 
3186 	if (!bp)
3187 		return -ENOMEM;
3188 	inet_peer_base_init(bp);
3189 	net->ipv4.peers = bp;
3190 	return 0;
3191 }
3192 
3193 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3194 {
3195 	struct inet_peer_base *bp = net->ipv4.peers;
3196 
3197 	net->ipv4.peers = NULL;
3198 	inetpeer_invalidate_tree(bp);
3199 	kfree(bp);
3200 }
3201 
3202 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3203 	.init	=	ipv4_inetpeer_init,
3204 	.exit	=	ipv4_inetpeer_exit,
3205 };
3206 
3207 #ifdef CONFIG_IP_ROUTE_CLASSID
3208 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3209 #endif /* CONFIG_IP_ROUTE_CLASSID */
3210 
3211 int __init ip_rt_init(void)
3212 {
3213 	int cpu;
3214 
3215 	ip_idents = kmalloc_array(IP_IDENTS_SZ, sizeof(*ip_idents),
3216 				  GFP_KERNEL);
3217 	if (!ip_idents)
3218 		panic("IP: failed to allocate ip_idents\n");
3219 
3220 	prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
3221 
3222 	ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
3223 	if (!ip_tstamps)
3224 		panic("IP: failed to allocate ip_tstamps\n");
3225 
3226 	for_each_possible_cpu(cpu) {
3227 		struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3228 
3229 		INIT_LIST_HEAD(&ul->head);
3230 		spin_lock_init(&ul->lock);
3231 	}
3232 #ifdef CONFIG_IP_ROUTE_CLASSID
3233 	ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3234 	if (!ip_rt_acct)
3235 		panic("IP: failed to allocate ip_rt_acct\n");
3236 #endif
3237 
3238 	ipv4_dst_ops.kmem_cachep =
3239 		kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3240 				  SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3241 
3242 	ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3243 
3244 	if (dst_entries_init(&ipv4_dst_ops) < 0)
3245 		panic("IP: failed to allocate ipv4_dst_ops counter\n");
3246 
3247 	if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3248 		panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3249 
3250 	ipv4_dst_ops.gc_thresh = ~0;
3251 	ip_rt_max_size = INT_MAX;
3252 
3253 	devinet_init();
3254 	ip_fib_init();
3255 
3256 	if (ip_rt_proc_init())
3257 		pr_err("Unable to create route proc files\n");
3258 #ifdef CONFIG_XFRM
3259 	xfrm_init();
3260 	xfrm4_init();
3261 #endif
3262 	rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3263 		      RTNL_FLAG_DOIT_UNLOCKED);
3264 
3265 #ifdef CONFIG_SYSCTL
3266 	register_pernet_subsys(&sysctl_route_ops);
3267 #endif
3268 	register_pernet_subsys(&rt_genid_ops);
3269 	register_pernet_subsys(&ipv4_inetpeer_ops);
3270 	return 0;
3271 }
3272 
3273 #ifdef CONFIG_SYSCTL
3274 /*
3275  * We really need to sanitize the damn ipv4 init order, then all
3276  * this nonsense will go away.
3277  */
3278 void __init ip_static_sysctl_init(void)
3279 {
3280 	register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3281 }
3282 #endif
3283