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