xref: /linux/net/ipv4/route.c (revision 9e8ba5f3ec35cba4fd8a8bebda548c4db2651e40)
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 #include <linux/module.h>
66 #include <asm/uaccess.h>
67 #include <asm/system.h>
68 #include <linux/bitops.h>
69 #include <linux/types.h>
70 #include <linux/kernel.h>
71 #include <linux/mm.h>
72 #include <linux/bootmem.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/workqueue.h>
83 #include <linux/skbuff.h>
84 #include <linux/inetdevice.h>
85 #include <linux/igmp.h>
86 #include <linux/pkt_sched.h>
87 #include <linux/mroute.h>
88 #include <linux/netfilter_ipv4.h>
89 #include <linux/random.h>
90 #include <linux/jhash.h>
91 #include <linux/rcupdate.h>
92 #include <linux/times.h>
93 #include <linux/slab.h>
94 #include <net/dst.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/netevent.h>
107 #include <net/rtnetlink.h>
108 #ifdef CONFIG_SYSCTL
109 #include <linux/sysctl.h>
110 #endif
111 #include <net/secure_seq.h>
112 
113 #define RT_FL_TOS(oldflp4) \
114 	((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
115 
116 #define IP_MAX_MTU	0xFFF0
117 
118 #define RT_GC_TIMEOUT (300*HZ)
119 
120 static int ip_rt_max_size;
121 static int ip_rt_gc_timeout __read_mostly	= RT_GC_TIMEOUT;
122 static int ip_rt_gc_min_interval __read_mostly	= HZ / 2;
123 static int ip_rt_redirect_number __read_mostly	= 9;
124 static int ip_rt_redirect_load __read_mostly	= HZ / 50;
125 static int ip_rt_redirect_silence __read_mostly	= ((HZ / 50) << (9 + 1));
126 static int ip_rt_error_cost __read_mostly	= HZ;
127 static int ip_rt_error_burst __read_mostly	= 5 * HZ;
128 static int ip_rt_gc_elasticity __read_mostly	= 8;
129 static int ip_rt_mtu_expires __read_mostly	= 10 * 60 * HZ;
130 static int ip_rt_min_pmtu __read_mostly		= 512 + 20 + 20;
131 static int ip_rt_min_advmss __read_mostly	= 256;
132 static int rt_chain_length_max __read_mostly	= 20;
133 static int redirect_genid;
134 
135 /*
136  *	Interface to generic destination cache.
137  */
138 
139 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
140 static unsigned int	 ipv4_default_advmss(const struct dst_entry *dst);
141 static unsigned int	 ipv4_mtu(const struct dst_entry *dst);
142 static void		 ipv4_dst_destroy(struct dst_entry *dst);
143 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
144 static void		 ipv4_link_failure(struct sk_buff *skb);
145 static void		 ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
146 static int rt_garbage_collect(struct dst_ops *ops);
147 
148 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
149 			    int how)
150 {
151 }
152 
153 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
154 {
155 	struct rtable *rt = (struct rtable *) dst;
156 	struct inet_peer *peer;
157 	u32 *p = NULL;
158 
159 	if (!rt->peer)
160 		rt_bind_peer(rt, rt->rt_dst, 1);
161 
162 	peer = rt->peer;
163 	if (peer) {
164 		u32 *old_p = __DST_METRICS_PTR(old);
165 		unsigned long prev, new;
166 
167 		p = peer->metrics;
168 		if (inet_metrics_new(peer))
169 			memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
170 
171 		new = (unsigned long) p;
172 		prev = cmpxchg(&dst->_metrics, old, new);
173 
174 		if (prev != old) {
175 			p = __DST_METRICS_PTR(prev);
176 			if (prev & DST_METRICS_READ_ONLY)
177 				p = NULL;
178 		} else {
179 			if (rt->fi) {
180 				fib_info_put(rt->fi);
181 				rt->fi = NULL;
182 			}
183 		}
184 	}
185 	return p;
186 }
187 
188 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, const void *daddr);
189 
190 static struct dst_ops ipv4_dst_ops = {
191 	.family =		AF_INET,
192 	.protocol =		cpu_to_be16(ETH_P_IP),
193 	.gc =			rt_garbage_collect,
194 	.check =		ipv4_dst_check,
195 	.default_advmss =	ipv4_default_advmss,
196 	.mtu =			ipv4_mtu,
197 	.cow_metrics =		ipv4_cow_metrics,
198 	.destroy =		ipv4_dst_destroy,
199 	.ifdown =		ipv4_dst_ifdown,
200 	.negative_advice =	ipv4_negative_advice,
201 	.link_failure =		ipv4_link_failure,
202 	.update_pmtu =		ip_rt_update_pmtu,
203 	.local_out =		__ip_local_out,
204 	.neigh_lookup =		ipv4_neigh_lookup,
205 };
206 
207 #define ECN_OR_COST(class)	TC_PRIO_##class
208 
209 const __u8 ip_tos2prio[16] = {
210 	TC_PRIO_BESTEFFORT,
211 	ECN_OR_COST(BESTEFFORT),
212 	TC_PRIO_BESTEFFORT,
213 	ECN_OR_COST(BESTEFFORT),
214 	TC_PRIO_BULK,
215 	ECN_OR_COST(BULK),
216 	TC_PRIO_BULK,
217 	ECN_OR_COST(BULK),
218 	TC_PRIO_INTERACTIVE,
219 	ECN_OR_COST(INTERACTIVE),
220 	TC_PRIO_INTERACTIVE,
221 	ECN_OR_COST(INTERACTIVE),
222 	TC_PRIO_INTERACTIVE_BULK,
223 	ECN_OR_COST(INTERACTIVE_BULK),
224 	TC_PRIO_INTERACTIVE_BULK,
225 	ECN_OR_COST(INTERACTIVE_BULK)
226 };
227 
228 
229 /*
230  * Route cache.
231  */
232 
233 /* The locking scheme is rather straight forward:
234  *
235  * 1) Read-Copy Update protects the buckets of the central route hash.
236  * 2) Only writers remove entries, and they hold the lock
237  *    as they look at rtable reference counts.
238  * 3) Only readers acquire references to rtable entries,
239  *    they do so with atomic increments and with the
240  *    lock held.
241  */
242 
243 struct rt_hash_bucket {
244 	struct rtable __rcu	*chain;
245 };
246 
247 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
248 	defined(CONFIG_PROVE_LOCKING)
249 /*
250  * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
251  * The size of this table is a power of two and depends on the number of CPUS.
252  * (on lockdep we have a quite big spinlock_t, so keep the size down there)
253  */
254 #ifdef CONFIG_LOCKDEP
255 # define RT_HASH_LOCK_SZ	256
256 #else
257 # if NR_CPUS >= 32
258 #  define RT_HASH_LOCK_SZ	4096
259 # elif NR_CPUS >= 16
260 #  define RT_HASH_LOCK_SZ	2048
261 # elif NR_CPUS >= 8
262 #  define RT_HASH_LOCK_SZ	1024
263 # elif NR_CPUS >= 4
264 #  define RT_HASH_LOCK_SZ	512
265 # else
266 #  define RT_HASH_LOCK_SZ	256
267 # endif
268 #endif
269 
270 static spinlock_t	*rt_hash_locks;
271 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
272 
273 static __init void rt_hash_lock_init(void)
274 {
275 	int i;
276 
277 	rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
278 			GFP_KERNEL);
279 	if (!rt_hash_locks)
280 		panic("IP: failed to allocate rt_hash_locks\n");
281 
282 	for (i = 0; i < RT_HASH_LOCK_SZ; i++)
283 		spin_lock_init(&rt_hash_locks[i]);
284 }
285 #else
286 # define rt_hash_lock_addr(slot) NULL
287 
288 static inline void rt_hash_lock_init(void)
289 {
290 }
291 #endif
292 
293 static struct rt_hash_bucket 	*rt_hash_table __read_mostly;
294 static unsigned			rt_hash_mask __read_mostly;
295 static unsigned int		rt_hash_log  __read_mostly;
296 
297 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
298 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
299 
300 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
301 				   int genid)
302 {
303 	return jhash_3words((__force u32)daddr, (__force u32)saddr,
304 			    idx, genid)
305 		& rt_hash_mask;
306 }
307 
308 static inline int rt_genid(struct net *net)
309 {
310 	return atomic_read(&net->ipv4.rt_genid);
311 }
312 
313 #ifdef CONFIG_PROC_FS
314 struct rt_cache_iter_state {
315 	struct seq_net_private p;
316 	int bucket;
317 	int genid;
318 };
319 
320 static struct rtable *rt_cache_get_first(struct seq_file *seq)
321 {
322 	struct rt_cache_iter_state *st = seq->private;
323 	struct rtable *r = NULL;
324 
325 	for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
326 		if (!rcu_access_pointer(rt_hash_table[st->bucket].chain))
327 			continue;
328 		rcu_read_lock_bh();
329 		r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
330 		while (r) {
331 			if (dev_net(r->dst.dev) == seq_file_net(seq) &&
332 			    r->rt_genid == st->genid)
333 				return r;
334 			r = rcu_dereference_bh(r->dst.rt_next);
335 		}
336 		rcu_read_unlock_bh();
337 	}
338 	return r;
339 }
340 
341 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
342 					  struct rtable *r)
343 {
344 	struct rt_cache_iter_state *st = seq->private;
345 
346 	r = rcu_dereference_bh(r->dst.rt_next);
347 	while (!r) {
348 		rcu_read_unlock_bh();
349 		do {
350 			if (--st->bucket < 0)
351 				return NULL;
352 		} while (!rcu_access_pointer(rt_hash_table[st->bucket].chain));
353 		rcu_read_lock_bh();
354 		r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
355 	}
356 	return r;
357 }
358 
359 static struct rtable *rt_cache_get_next(struct seq_file *seq,
360 					struct rtable *r)
361 {
362 	struct rt_cache_iter_state *st = seq->private;
363 	while ((r = __rt_cache_get_next(seq, r)) != NULL) {
364 		if (dev_net(r->dst.dev) != seq_file_net(seq))
365 			continue;
366 		if (r->rt_genid == st->genid)
367 			break;
368 	}
369 	return r;
370 }
371 
372 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
373 {
374 	struct rtable *r = rt_cache_get_first(seq);
375 
376 	if (r)
377 		while (pos && (r = rt_cache_get_next(seq, r)))
378 			--pos;
379 	return pos ? NULL : r;
380 }
381 
382 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
383 {
384 	struct rt_cache_iter_state *st = seq->private;
385 	if (*pos)
386 		return rt_cache_get_idx(seq, *pos - 1);
387 	st->genid = rt_genid(seq_file_net(seq));
388 	return SEQ_START_TOKEN;
389 }
390 
391 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
392 {
393 	struct rtable *r;
394 
395 	if (v == SEQ_START_TOKEN)
396 		r = rt_cache_get_first(seq);
397 	else
398 		r = rt_cache_get_next(seq, v);
399 	++*pos;
400 	return r;
401 }
402 
403 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
404 {
405 	if (v && v != SEQ_START_TOKEN)
406 		rcu_read_unlock_bh();
407 }
408 
409 static int rt_cache_seq_show(struct seq_file *seq, void *v)
410 {
411 	if (v == SEQ_START_TOKEN)
412 		seq_printf(seq, "%-127s\n",
413 			   "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
414 			   "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
415 			   "HHUptod\tSpecDst");
416 	else {
417 		struct rtable *r = v;
418 		struct neighbour *n;
419 		int len, HHUptod;
420 
421 		rcu_read_lock();
422 		n = dst_get_neighbour_noref(&r->dst);
423 		HHUptod = (n && (n->nud_state & NUD_CONNECTED)) ? 1 : 0;
424 		rcu_read_unlock();
425 
426 		seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
427 			      "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
428 			r->dst.dev ? r->dst.dev->name : "*",
429 			(__force u32)r->rt_dst,
430 			(__force u32)r->rt_gateway,
431 			r->rt_flags, atomic_read(&r->dst.__refcnt),
432 			r->dst.__use, 0, (__force u32)r->rt_src,
433 			dst_metric_advmss(&r->dst) + 40,
434 			dst_metric(&r->dst, RTAX_WINDOW),
435 			(int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
436 			      dst_metric(&r->dst, RTAX_RTTVAR)),
437 			r->rt_key_tos,
438 			-1,
439 			HHUptod,
440 			r->rt_spec_dst, &len);
441 
442 		seq_printf(seq, "%*s\n", 127 - len, "");
443 	}
444 	return 0;
445 }
446 
447 static const struct seq_operations rt_cache_seq_ops = {
448 	.start  = rt_cache_seq_start,
449 	.next   = rt_cache_seq_next,
450 	.stop   = rt_cache_seq_stop,
451 	.show   = rt_cache_seq_show,
452 };
453 
454 static int rt_cache_seq_open(struct inode *inode, struct file *file)
455 {
456 	return seq_open_net(inode, file, &rt_cache_seq_ops,
457 			sizeof(struct rt_cache_iter_state));
458 }
459 
460 static const struct file_operations rt_cache_seq_fops = {
461 	.owner	 = THIS_MODULE,
462 	.open	 = rt_cache_seq_open,
463 	.read	 = seq_read,
464 	.llseek	 = seq_lseek,
465 	.release = seq_release_net,
466 };
467 
468 
469 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
470 {
471 	int cpu;
472 
473 	if (*pos == 0)
474 		return SEQ_START_TOKEN;
475 
476 	for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
477 		if (!cpu_possible(cpu))
478 			continue;
479 		*pos = cpu+1;
480 		return &per_cpu(rt_cache_stat, cpu);
481 	}
482 	return NULL;
483 }
484 
485 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
486 {
487 	int cpu;
488 
489 	for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
490 		if (!cpu_possible(cpu))
491 			continue;
492 		*pos = cpu+1;
493 		return &per_cpu(rt_cache_stat, cpu);
494 	}
495 	return NULL;
496 
497 }
498 
499 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
500 {
501 
502 }
503 
504 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
505 {
506 	struct rt_cache_stat *st = v;
507 
508 	if (v == SEQ_START_TOKEN) {
509 		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");
510 		return 0;
511 	}
512 
513 	seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
514 		   " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
515 		   dst_entries_get_slow(&ipv4_dst_ops),
516 		   st->in_hit,
517 		   st->in_slow_tot,
518 		   st->in_slow_mc,
519 		   st->in_no_route,
520 		   st->in_brd,
521 		   st->in_martian_dst,
522 		   st->in_martian_src,
523 
524 		   st->out_hit,
525 		   st->out_slow_tot,
526 		   st->out_slow_mc,
527 
528 		   st->gc_total,
529 		   st->gc_ignored,
530 		   st->gc_goal_miss,
531 		   st->gc_dst_overflow,
532 		   st->in_hlist_search,
533 		   st->out_hlist_search
534 		);
535 	return 0;
536 }
537 
538 static const struct seq_operations rt_cpu_seq_ops = {
539 	.start  = rt_cpu_seq_start,
540 	.next   = rt_cpu_seq_next,
541 	.stop   = rt_cpu_seq_stop,
542 	.show   = rt_cpu_seq_show,
543 };
544 
545 
546 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
547 {
548 	return seq_open(file, &rt_cpu_seq_ops);
549 }
550 
551 static const struct file_operations rt_cpu_seq_fops = {
552 	.owner	 = THIS_MODULE,
553 	.open	 = rt_cpu_seq_open,
554 	.read	 = seq_read,
555 	.llseek	 = seq_lseek,
556 	.release = seq_release,
557 };
558 
559 #ifdef CONFIG_IP_ROUTE_CLASSID
560 static int rt_acct_proc_show(struct seq_file *m, void *v)
561 {
562 	struct ip_rt_acct *dst, *src;
563 	unsigned int i, j;
564 
565 	dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
566 	if (!dst)
567 		return -ENOMEM;
568 
569 	for_each_possible_cpu(i) {
570 		src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
571 		for (j = 0; j < 256; j++) {
572 			dst[j].o_bytes   += src[j].o_bytes;
573 			dst[j].o_packets += src[j].o_packets;
574 			dst[j].i_bytes   += src[j].i_bytes;
575 			dst[j].i_packets += src[j].i_packets;
576 		}
577 	}
578 
579 	seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
580 	kfree(dst);
581 	return 0;
582 }
583 
584 static int rt_acct_proc_open(struct inode *inode, struct file *file)
585 {
586 	return single_open(file, rt_acct_proc_show, NULL);
587 }
588 
589 static const struct file_operations rt_acct_proc_fops = {
590 	.owner		= THIS_MODULE,
591 	.open		= rt_acct_proc_open,
592 	.read		= seq_read,
593 	.llseek		= seq_lseek,
594 	.release	= single_release,
595 };
596 #endif
597 
598 static int __net_init ip_rt_do_proc_init(struct net *net)
599 {
600 	struct proc_dir_entry *pde;
601 
602 	pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
603 			&rt_cache_seq_fops);
604 	if (!pde)
605 		goto err1;
606 
607 	pde = proc_create("rt_cache", S_IRUGO,
608 			  net->proc_net_stat, &rt_cpu_seq_fops);
609 	if (!pde)
610 		goto err2;
611 
612 #ifdef CONFIG_IP_ROUTE_CLASSID
613 	pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
614 	if (!pde)
615 		goto err3;
616 #endif
617 	return 0;
618 
619 #ifdef CONFIG_IP_ROUTE_CLASSID
620 err3:
621 	remove_proc_entry("rt_cache", net->proc_net_stat);
622 #endif
623 err2:
624 	remove_proc_entry("rt_cache", net->proc_net);
625 err1:
626 	return -ENOMEM;
627 }
628 
629 static void __net_exit ip_rt_do_proc_exit(struct net *net)
630 {
631 	remove_proc_entry("rt_cache", net->proc_net_stat);
632 	remove_proc_entry("rt_cache", net->proc_net);
633 #ifdef CONFIG_IP_ROUTE_CLASSID
634 	remove_proc_entry("rt_acct", net->proc_net);
635 #endif
636 }
637 
638 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
639 	.init = ip_rt_do_proc_init,
640 	.exit = ip_rt_do_proc_exit,
641 };
642 
643 static int __init ip_rt_proc_init(void)
644 {
645 	return register_pernet_subsys(&ip_rt_proc_ops);
646 }
647 
648 #else
649 static inline int ip_rt_proc_init(void)
650 {
651 	return 0;
652 }
653 #endif /* CONFIG_PROC_FS */
654 
655 static inline void rt_free(struct rtable *rt)
656 {
657 	call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
658 }
659 
660 static inline void rt_drop(struct rtable *rt)
661 {
662 	ip_rt_put(rt);
663 	call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
664 }
665 
666 static inline int rt_fast_clean(struct rtable *rth)
667 {
668 	/* Kill broadcast/multicast entries very aggresively, if they
669 	   collide in hash table with more useful entries */
670 	return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
671 		rt_is_input_route(rth) && rth->dst.rt_next;
672 }
673 
674 static inline int rt_valuable(struct rtable *rth)
675 {
676 	return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
677 		(rth->peer && rth->peer->pmtu_expires);
678 }
679 
680 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
681 {
682 	unsigned long age;
683 	int ret = 0;
684 
685 	if (atomic_read(&rth->dst.__refcnt))
686 		goto out;
687 
688 	age = jiffies - rth->dst.lastuse;
689 	if ((age <= tmo1 && !rt_fast_clean(rth)) ||
690 	    (age <= tmo2 && rt_valuable(rth)))
691 		goto out;
692 	ret = 1;
693 out:	return ret;
694 }
695 
696 /* Bits of score are:
697  * 31: very valuable
698  * 30: not quite useless
699  * 29..0: usage counter
700  */
701 static inline u32 rt_score(struct rtable *rt)
702 {
703 	u32 score = jiffies - rt->dst.lastuse;
704 
705 	score = ~score & ~(3<<30);
706 
707 	if (rt_valuable(rt))
708 		score |= (1<<31);
709 
710 	if (rt_is_output_route(rt) ||
711 	    !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
712 		score |= (1<<30);
713 
714 	return score;
715 }
716 
717 static inline bool rt_caching(const struct net *net)
718 {
719 	return net->ipv4.current_rt_cache_rebuild_count <=
720 		net->ipv4.sysctl_rt_cache_rebuild_count;
721 }
722 
723 static inline bool compare_hash_inputs(const struct rtable *rt1,
724 				       const struct rtable *rt2)
725 {
726 	return ((((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
727 		((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
728 		(rt1->rt_route_iif ^ rt2->rt_route_iif)) == 0);
729 }
730 
731 static inline int compare_keys(struct rtable *rt1, struct rtable *rt2)
732 {
733 	return (((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
734 		((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
735 		(rt1->rt_mark ^ rt2->rt_mark) |
736 		(rt1->rt_key_tos ^ rt2->rt_key_tos) |
737 		(rt1->rt_route_iif ^ rt2->rt_route_iif) |
738 		(rt1->rt_oif ^ rt2->rt_oif)) == 0;
739 }
740 
741 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
742 {
743 	return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
744 }
745 
746 static inline int rt_is_expired(struct rtable *rth)
747 {
748 	return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
749 }
750 
751 /*
752  * Perform a full scan of hash table and free all entries.
753  * Can be called by a softirq or a process.
754  * In the later case, we want to be reschedule if necessary
755  */
756 static void rt_do_flush(struct net *net, int process_context)
757 {
758 	unsigned int i;
759 	struct rtable *rth, *next;
760 
761 	for (i = 0; i <= rt_hash_mask; i++) {
762 		struct rtable __rcu **pprev;
763 		struct rtable *list;
764 
765 		if (process_context && need_resched())
766 			cond_resched();
767 		rth = rcu_access_pointer(rt_hash_table[i].chain);
768 		if (!rth)
769 			continue;
770 
771 		spin_lock_bh(rt_hash_lock_addr(i));
772 
773 		list = NULL;
774 		pprev = &rt_hash_table[i].chain;
775 		rth = rcu_dereference_protected(*pprev,
776 			lockdep_is_held(rt_hash_lock_addr(i)));
777 
778 		while (rth) {
779 			next = rcu_dereference_protected(rth->dst.rt_next,
780 				lockdep_is_held(rt_hash_lock_addr(i)));
781 
782 			if (!net ||
783 			    net_eq(dev_net(rth->dst.dev), net)) {
784 				rcu_assign_pointer(*pprev, next);
785 				rcu_assign_pointer(rth->dst.rt_next, list);
786 				list = rth;
787 			} else {
788 				pprev = &rth->dst.rt_next;
789 			}
790 			rth = next;
791 		}
792 
793 		spin_unlock_bh(rt_hash_lock_addr(i));
794 
795 		for (; list; list = next) {
796 			next = rcu_dereference_protected(list->dst.rt_next, 1);
797 			rt_free(list);
798 		}
799 	}
800 }
801 
802 /*
803  * While freeing expired entries, we compute average chain length
804  * and standard deviation, using fixed-point arithmetic.
805  * This to have an estimation of rt_chain_length_max
806  *  rt_chain_length_max = max(elasticity, AVG + 4*SD)
807  * We use 3 bits for frational part, and 29 (or 61) for magnitude.
808  */
809 
810 #define FRACT_BITS 3
811 #define ONE (1UL << FRACT_BITS)
812 
813 /*
814  * Given a hash chain and an item in this hash chain,
815  * find if a previous entry has the same hash_inputs
816  * (but differs on tos, mark or oif)
817  * Returns 0 if an alias is found.
818  * Returns ONE if rth has no alias before itself.
819  */
820 static int has_noalias(const struct rtable *head, const struct rtable *rth)
821 {
822 	const struct rtable *aux = head;
823 
824 	while (aux != rth) {
825 		if (compare_hash_inputs(aux, rth))
826 			return 0;
827 		aux = rcu_dereference_protected(aux->dst.rt_next, 1);
828 	}
829 	return ONE;
830 }
831 
832 /*
833  * Perturbation of rt_genid by a small quantity [1..256]
834  * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
835  * many times (2^24) without giving recent rt_genid.
836  * Jenkins hash is strong enough that litle changes of rt_genid are OK.
837  */
838 static void rt_cache_invalidate(struct net *net)
839 {
840 	unsigned char shuffle;
841 
842 	get_random_bytes(&shuffle, sizeof(shuffle));
843 	atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
844 	redirect_genid++;
845 }
846 
847 /*
848  * delay < 0  : invalidate cache (fast : entries will be deleted later)
849  * delay >= 0 : invalidate & flush cache (can be long)
850  */
851 void rt_cache_flush(struct net *net, int delay)
852 {
853 	rt_cache_invalidate(net);
854 	if (delay >= 0)
855 		rt_do_flush(net, !in_softirq());
856 }
857 
858 /* Flush previous cache invalidated entries from the cache */
859 void rt_cache_flush_batch(struct net *net)
860 {
861 	rt_do_flush(net, !in_softirq());
862 }
863 
864 static void rt_emergency_hash_rebuild(struct net *net)
865 {
866 	if (net_ratelimit())
867 		printk(KERN_WARNING "Route hash chain too long!\n");
868 	rt_cache_invalidate(net);
869 }
870 
871 /*
872    Short description of GC goals.
873 
874    We want to build algorithm, which will keep routing cache
875    at some equilibrium point, when number of aged off entries
876    is kept approximately equal to newly generated ones.
877 
878    Current expiration strength is variable "expire".
879    We try to adjust it dynamically, so that if networking
880    is idle expires is large enough to keep enough of warm entries,
881    and when load increases it reduces to limit cache size.
882  */
883 
884 static int rt_garbage_collect(struct dst_ops *ops)
885 {
886 	static unsigned long expire = RT_GC_TIMEOUT;
887 	static unsigned long last_gc;
888 	static int rover;
889 	static int equilibrium;
890 	struct rtable *rth;
891 	struct rtable __rcu **rthp;
892 	unsigned long now = jiffies;
893 	int goal;
894 	int entries = dst_entries_get_fast(&ipv4_dst_ops);
895 
896 	/*
897 	 * Garbage collection is pretty expensive,
898 	 * do not make it too frequently.
899 	 */
900 
901 	RT_CACHE_STAT_INC(gc_total);
902 
903 	if (now - last_gc < ip_rt_gc_min_interval &&
904 	    entries < ip_rt_max_size) {
905 		RT_CACHE_STAT_INC(gc_ignored);
906 		goto out;
907 	}
908 
909 	entries = dst_entries_get_slow(&ipv4_dst_ops);
910 	/* Calculate number of entries, which we want to expire now. */
911 	goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
912 	if (goal <= 0) {
913 		if (equilibrium < ipv4_dst_ops.gc_thresh)
914 			equilibrium = ipv4_dst_ops.gc_thresh;
915 		goal = entries - equilibrium;
916 		if (goal > 0) {
917 			equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
918 			goal = entries - equilibrium;
919 		}
920 	} else {
921 		/* We are in dangerous area. Try to reduce cache really
922 		 * aggressively.
923 		 */
924 		goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
925 		equilibrium = entries - goal;
926 	}
927 
928 	if (now - last_gc >= ip_rt_gc_min_interval)
929 		last_gc = now;
930 
931 	if (goal <= 0) {
932 		equilibrium += goal;
933 		goto work_done;
934 	}
935 
936 	do {
937 		int i, k;
938 
939 		for (i = rt_hash_mask, k = rover; i >= 0; i--) {
940 			unsigned long tmo = expire;
941 
942 			k = (k + 1) & rt_hash_mask;
943 			rthp = &rt_hash_table[k].chain;
944 			spin_lock_bh(rt_hash_lock_addr(k));
945 			while ((rth = rcu_dereference_protected(*rthp,
946 					lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
947 				if (!rt_is_expired(rth) &&
948 					!rt_may_expire(rth, tmo, expire)) {
949 					tmo >>= 1;
950 					rthp = &rth->dst.rt_next;
951 					continue;
952 				}
953 				*rthp = rth->dst.rt_next;
954 				rt_free(rth);
955 				goal--;
956 			}
957 			spin_unlock_bh(rt_hash_lock_addr(k));
958 			if (goal <= 0)
959 				break;
960 		}
961 		rover = k;
962 
963 		if (goal <= 0)
964 			goto work_done;
965 
966 		/* Goal is not achieved. We stop process if:
967 
968 		   - if expire reduced to zero. Otherwise, expire is halfed.
969 		   - if table is not full.
970 		   - if we are called from interrupt.
971 		   - jiffies check is just fallback/debug loop breaker.
972 		     We will not spin here for long time in any case.
973 		 */
974 
975 		RT_CACHE_STAT_INC(gc_goal_miss);
976 
977 		if (expire == 0)
978 			break;
979 
980 		expire >>= 1;
981 
982 		if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
983 			goto out;
984 	} while (!in_softirq() && time_before_eq(jiffies, now));
985 
986 	if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
987 		goto out;
988 	if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
989 		goto out;
990 	if (net_ratelimit())
991 		printk(KERN_WARNING "dst cache overflow\n");
992 	RT_CACHE_STAT_INC(gc_dst_overflow);
993 	return 1;
994 
995 work_done:
996 	expire += ip_rt_gc_min_interval;
997 	if (expire > ip_rt_gc_timeout ||
998 	    dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
999 	    dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
1000 		expire = ip_rt_gc_timeout;
1001 out:	return 0;
1002 }
1003 
1004 /*
1005  * Returns number of entries in a hash chain that have different hash_inputs
1006  */
1007 static int slow_chain_length(const struct rtable *head)
1008 {
1009 	int length = 0;
1010 	const struct rtable *rth = head;
1011 
1012 	while (rth) {
1013 		length += has_noalias(head, rth);
1014 		rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1015 	}
1016 	return length >> FRACT_BITS;
1017 }
1018 
1019 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, const void *daddr)
1020 {
1021 	static const __be32 inaddr_any = 0;
1022 	struct net_device *dev = dst->dev;
1023 	const __be32 *pkey = daddr;
1024 	struct neighbour *n;
1025 
1026 	if (dev->flags & (IFF_LOOPBACK | IFF_POINTOPOINT))
1027 		pkey = &inaddr_any;
1028 
1029 	n = __ipv4_neigh_lookup(&arp_tbl, dev, *(__force u32 *)pkey);
1030 	if (n)
1031 		return n;
1032 	return neigh_create(&arp_tbl, pkey, dev);
1033 }
1034 
1035 static int rt_bind_neighbour(struct rtable *rt)
1036 {
1037 	struct neighbour *n = ipv4_neigh_lookup(&rt->dst, &rt->rt_gateway);
1038 	if (IS_ERR(n))
1039 		return PTR_ERR(n);
1040 	dst_set_neighbour(&rt->dst, n);
1041 
1042 	return 0;
1043 }
1044 
1045 static struct rtable *rt_intern_hash(unsigned hash, struct rtable *rt,
1046 				     struct sk_buff *skb, int ifindex)
1047 {
1048 	struct rtable	*rth, *cand;
1049 	struct rtable __rcu **rthp, **candp;
1050 	unsigned long	now;
1051 	u32 		min_score;
1052 	int		chain_length;
1053 	int attempts = !in_softirq();
1054 
1055 restart:
1056 	chain_length = 0;
1057 	min_score = ~(u32)0;
1058 	cand = NULL;
1059 	candp = NULL;
1060 	now = jiffies;
1061 
1062 	if (!rt_caching(dev_net(rt->dst.dev))) {
1063 		/*
1064 		 * If we're not caching, just tell the caller we
1065 		 * were successful and don't touch the route.  The
1066 		 * caller hold the sole reference to the cache entry, and
1067 		 * it will be released when the caller is done with it.
1068 		 * If we drop it here, the callers have no way to resolve routes
1069 		 * when we're not caching.  Instead, just point *rp at rt, so
1070 		 * the caller gets a single use out of the route
1071 		 * Note that we do rt_free on this new route entry, so that
1072 		 * once its refcount hits zero, we are still able to reap it
1073 		 * (Thanks Alexey)
1074 		 * Note: To avoid expensive rcu stuff for this uncached dst,
1075 		 * we set DST_NOCACHE so that dst_release() can free dst without
1076 		 * waiting a grace period.
1077 		 */
1078 
1079 		rt->dst.flags |= DST_NOCACHE;
1080 		if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1081 			int err = rt_bind_neighbour(rt);
1082 			if (err) {
1083 				if (net_ratelimit())
1084 					printk(KERN_WARNING
1085 					    "Neighbour table failure & not caching routes.\n");
1086 				ip_rt_put(rt);
1087 				return ERR_PTR(err);
1088 			}
1089 		}
1090 
1091 		goto skip_hashing;
1092 	}
1093 
1094 	rthp = &rt_hash_table[hash].chain;
1095 
1096 	spin_lock_bh(rt_hash_lock_addr(hash));
1097 	while ((rth = rcu_dereference_protected(*rthp,
1098 			lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1099 		if (rt_is_expired(rth)) {
1100 			*rthp = rth->dst.rt_next;
1101 			rt_free(rth);
1102 			continue;
1103 		}
1104 		if (compare_keys(rth, rt) && compare_netns(rth, rt)) {
1105 			/* Put it first */
1106 			*rthp = rth->dst.rt_next;
1107 			/*
1108 			 * Since lookup is lockfree, the deletion
1109 			 * must be visible to another weakly ordered CPU before
1110 			 * the insertion at the start of the hash chain.
1111 			 */
1112 			rcu_assign_pointer(rth->dst.rt_next,
1113 					   rt_hash_table[hash].chain);
1114 			/*
1115 			 * Since lookup is lockfree, the update writes
1116 			 * must be ordered for consistency on SMP.
1117 			 */
1118 			rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1119 
1120 			dst_use(&rth->dst, now);
1121 			spin_unlock_bh(rt_hash_lock_addr(hash));
1122 
1123 			rt_drop(rt);
1124 			if (skb)
1125 				skb_dst_set(skb, &rth->dst);
1126 			return rth;
1127 		}
1128 
1129 		if (!atomic_read(&rth->dst.__refcnt)) {
1130 			u32 score = rt_score(rth);
1131 
1132 			if (score <= min_score) {
1133 				cand = rth;
1134 				candp = rthp;
1135 				min_score = score;
1136 			}
1137 		}
1138 
1139 		chain_length++;
1140 
1141 		rthp = &rth->dst.rt_next;
1142 	}
1143 
1144 	if (cand) {
1145 		/* ip_rt_gc_elasticity used to be average length of chain
1146 		 * length, when exceeded gc becomes really aggressive.
1147 		 *
1148 		 * The second limit is less certain. At the moment it allows
1149 		 * only 2 entries per bucket. We will see.
1150 		 */
1151 		if (chain_length > ip_rt_gc_elasticity) {
1152 			*candp = cand->dst.rt_next;
1153 			rt_free(cand);
1154 		}
1155 	} else {
1156 		if (chain_length > rt_chain_length_max &&
1157 		    slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1158 			struct net *net = dev_net(rt->dst.dev);
1159 			int num = ++net->ipv4.current_rt_cache_rebuild_count;
1160 			if (!rt_caching(net)) {
1161 				printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1162 					rt->dst.dev->name, num);
1163 			}
1164 			rt_emergency_hash_rebuild(net);
1165 			spin_unlock_bh(rt_hash_lock_addr(hash));
1166 
1167 			hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1168 					ifindex, rt_genid(net));
1169 			goto restart;
1170 		}
1171 	}
1172 
1173 	/* Try to bind route to arp only if it is output
1174 	   route or unicast forwarding path.
1175 	 */
1176 	if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1177 		int err = rt_bind_neighbour(rt);
1178 		if (err) {
1179 			spin_unlock_bh(rt_hash_lock_addr(hash));
1180 
1181 			if (err != -ENOBUFS) {
1182 				rt_drop(rt);
1183 				return ERR_PTR(err);
1184 			}
1185 
1186 			/* Neighbour tables are full and nothing
1187 			   can be released. Try to shrink route cache,
1188 			   it is most likely it holds some neighbour records.
1189 			 */
1190 			if (attempts-- > 0) {
1191 				int saved_elasticity = ip_rt_gc_elasticity;
1192 				int saved_int = ip_rt_gc_min_interval;
1193 				ip_rt_gc_elasticity	= 1;
1194 				ip_rt_gc_min_interval	= 0;
1195 				rt_garbage_collect(&ipv4_dst_ops);
1196 				ip_rt_gc_min_interval	= saved_int;
1197 				ip_rt_gc_elasticity	= saved_elasticity;
1198 				goto restart;
1199 			}
1200 
1201 			if (net_ratelimit())
1202 				printk(KERN_WARNING "ipv4: Neighbour table overflow.\n");
1203 			rt_drop(rt);
1204 			return ERR_PTR(-ENOBUFS);
1205 		}
1206 	}
1207 
1208 	rt->dst.rt_next = rt_hash_table[hash].chain;
1209 
1210 	/*
1211 	 * Since lookup is lockfree, we must make sure
1212 	 * previous writes to rt are committed to memory
1213 	 * before making rt visible to other CPUS.
1214 	 */
1215 	rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1216 
1217 	spin_unlock_bh(rt_hash_lock_addr(hash));
1218 
1219 skip_hashing:
1220 	if (skb)
1221 		skb_dst_set(skb, &rt->dst);
1222 	return rt;
1223 }
1224 
1225 static atomic_t __rt_peer_genid = ATOMIC_INIT(0);
1226 
1227 static u32 rt_peer_genid(void)
1228 {
1229 	return atomic_read(&__rt_peer_genid);
1230 }
1231 
1232 void rt_bind_peer(struct rtable *rt, __be32 daddr, int create)
1233 {
1234 	struct inet_peer *peer;
1235 
1236 	peer = inet_getpeer_v4(daddr, create);
1237 
1238 	if (peer && cmpxchg(&rt->peer, NULL, peer) != NULL)
1239 		inet_putpeer(peer);
1240 	else
1241 		rt->rt_peer_genid = rt_peer_genid();
1242 }
1243 
1244 /*
1245  * Peer allocation may fail only in serious out-of-memory conditions.  However
1246  * we still can generate some output.
1247  * Random ID selection looks a bit dangerous because we have no chances to
1248  * select ID being unique in a reasonable period of time.
1249  * But broken packet identifier may be better than no packet at all.
1250  */
1251 static void ip_select_fb_ident(struct iphdr *iph)
1252 {
1253 	static DEFINE_SPINLOCK(ip_fb_id_lock);
1254 	static u32 ip_fallback_id;
1255 	u32 salt;
1256 
1257 	spin_lock_bh(&ip_fb_id_lock);
1258 	salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1259 	iph->id = htons(salt & 0xFFFF);
1260 	ip_fallback_id = salt;
1261 	spin_unlock_bh(&ip_fb_id_lock);
1262 }
1263 
1264 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1265 {
1266 	struct rtable *rt = (struct rtable *) dst;
1267 
1268 	if (rt) {
1269 		if (rt->peer == NULL)
1270 			rt_bind_peer(rt, rt->rt_dst, 1);
1271 
1272 		/* If peer is attached to destination, it is never detached,
1273 		   so that we need not to grab a lock to dereference it.
1274 		 */
1275 		if (rt->peer) {
1276 			iph->id = htons(inet_getid(rt->peer, more));
1277 			return;
1278 		}
1279 	} else
1280 		printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1281 		       __builtin_return_address(0));
1282 
1283 	ip_select_fb_ident(iph);
1284 }
1285 EXPORT_SYMBOL(__ip_select_ident);
1286 
1287 static void rt_del(unsigned hash, struct rtable *rt)
1288 {
1289 	struct rtable __rcu **rthp;
1290 	struct rtable *aux;
1291 
1292 	rthp = &rt_hash_table[hash].chain;
1293 	spin_lock_bh(rt_hash_lock_addr(hash));
1294 	ip_rt_put(rt);
1295 	while ((aux = rcu_dereference_protected(*rthp,
1296 			lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1297 		if (aux == rt || rt_is_expired(aux)) {
1298 			*rthp = aux->dst.rt_next;
1299 			rt_free(aux);
1300 			continue;
1301 		}
1302 		rthp = &aux->dst.rt_next;
1303 	}
1304 	spin_unlock_bh(rt_hash_lock_addr(hash));
1305 }
1306 
1307 static void check_peer_redir(struct dst_entry *dst, struct inet_peer *peer)
1308 {
1309 	struct rtable *rt = (struct rtable *) dst;
1310 	__be32 orig_gw = rt->rt_gateway;
1311 	struct neighbour *n, *old_n;
1312 
1313 	dst_confirm(&rt->dst);
1314 
1315 	rt->rt_gateway = peer->redirect_learned.a4;
1316 
1317 	n = ipv4_neigh_lookup(&rt->dst, &rt->rt_gateway);
1318 	if (IS_ERR(n)) {
1319 		rt->rt_gateway = orig_gw;
1320 		return;
1321 	}
1322 	old_n = xchg(&rt->dst._neighbour, n);
1323 	if (old_n)
1324 		neigh_release(old_n);
1325 	if (!(n->nud_state & NUD_VALID)) {
1326 		neigh_event_send(n, NULL);
1327 	} else {
1328 		rt->rt_flags |= RTCF_REDIRECTED;
1329 		call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
1330 	}
1331 }
1332 
1333 /* called in rcu_read_lock() section */
1334 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1335 		    __be32 saddr, struct net_device *dev)
1336 {
1337 	int s, i;
1338 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1339 	__be32 skeys[2] = { saddr, 0 };
1340 	int    ikeys[2] = { dev->ifindex, 0 };
1341 	struct inet_peer *peer;
1342 	struct net *net;
1343 
1344 	if (!in_dev)
1345 		return;
1346 
1347 	net = dev_net(dev);
1348 	if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1349 	    ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1350 	    ipv4_is_zeronet(new_gw))
1351 		goto reject_redirect;
1352 
1353 	if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1354 		if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1355 			goto reject_redirect;
1356 		if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1357 			goto reject_redirect;
1358 	} else {
1359 		if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1360 			goto reject_redirect;
1361 	}
1362 
1363 	for (s = 0; s < 2; s++) {
1364 		for (i = 0; i < 2; i++) {
1365 			unsigned int hash;
1366 			struct rtable __rcu **rthp;
1367 			struct rtable *rt;
1368 
1369 			hash = rt_hash(daddr, skeys[s], ikeys[i], rt_genid(net));
1370 
1371 			rthp = &rt_hash_table[hash].chain;
1372 
1373 			while ((rt = rcu_dereference(*rthp)) != NULL) {
1374 				rthp = &rt->dst.rt_next;
1375 
1376 				if (rt->rt_key_dst != daddr ||
1377 				    rt->rt_key_src != skeys[s] ||
1378 				    rt->rt_oif != ikeys[i] ||
1379 				    rt_is_input_route(rt) ||
1380 				    rt_is_expired(rt) ||
1381 				    !net_eq(dev_net(rt->dst.dev), net) ||
1382 				    rt->dst.error ||
1383 				    rt->dst.dev != dev ||
1384 				    rt->rt_gateway != old_gw)
1385 					continue;
1386 
1387 				if (!rt->peer)
1388 					rt_bind_peer(rt, rt->rt_dst, 1);
1389 
1390 				peer = rt->peer;
1391 				if (peer) {
1392 					if (peer->redirect_learned.a4 != new_gw ||
1393 					    peer->redirect_genid != redirect_genid) {
1394 						peer->redirect_learned.a4 = new_gw;
1395 						peer->redirect_genid = redirect_genid;
1396 						atomic_inc(&__rt_peer_genid);
1397 					}
1398 					check_peer_redir(&rt->dst, peer);
1399 				}
1400 			}
1401 		}
1402 	}
1403 	return;
1404 
1405 reject_redirect:
1406 #ifdef CONFIG_IP_ROUTE_VERBOSE
1407 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1408 		printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1409 			"  Advised path = %pI4 -> %pI4\n",
1410 		       &old_gw, dev->name, &new_gw,
1411 		       &saddr, &daddr);
1412 #endif
1413 	;
1414 }
1415 
1416 static bool peer_pmtu_expired(struct inet_peer *peer)
1417 {
1418 	unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1419 
1420 	return orig &&
1421 	       time_after_eq(jiffies, orig) &&
1422 	       cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1423 }
1424 
1425 static bool peer_pmtu_cleaned(struct inet_peer *peer)
1426 {
1427 	unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1428 
1429 	return orig &&
1430 	       cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1431 }
1432 
1433 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1434 {
1435 	struct rtable *rt = (struct rtable *)dst;
1436 	struct dst_entry *ret = dst;
1437 
1438 	if (rt) {
1439 		if (dst->obsolete > 0) {
1440 			ip_rt_put(rt);
1441 			ret = NULL;
1442 		} else if (rt->rt_flags & RTCF_REDIRECTED) {
1443 			unsigned hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1444 						rt->rt_oif,
1445 						rt_genid(dev_net(dst->dev)));
1446 			rt_del(hash, rt);
1447 			ret = NULL;
1448 		} else if (rt->peer && peer_pmtu_expired(rt->peer)) {
1449 			dst_metric_set(dst, RTAX_MTU, rt->peer->pmtu_orig);
1450 		}
1451 	}
1452 	return ret;
1453 }
1454 
1455 /*
1456  * Algorithm:
1457  *	1. The first ip_rt_redirect_number redirects are sent
1458  *	   with exponential backoff, then we stop sending them at all,
1459  *	   assuming that the host ignores our redirects.
1460  *	2. If we did not see packets requiring redirects
1461  *	   during ip_rt_redirect_silence, we assume that the host
1462  *	   forgot redirected route and start to send redirects again.
1463  *
1464  * This algorithm is much cheaper and more intelligent than dumb load limiting
1465  * in icmp.c.
1466  *
1467  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1468  * and "frag. need" (breaks PMTU discovery) in icmp.c.
1469  */
1470 
1471 void ip_rt_send_redirect(struct sk_buff *skb)
1472 {
1473 	struct rtable *rt = skb_rtable(skb);
1474 	struct in_device *in_dev;
1475 	struct inet_peer *peer;
1476 	int log_martians;
1477 
1478 	rcu_read_lock();
1479 	in_dev = __in_dev_get_rcu(rt->dst.dev);
1480 	if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1481 		rcu_read_unlock();
1482 		return;
1483 	}
1484 	log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1485 	rcu_read_unlock();
1486 
1487 	if (!rt->peer)
1488 		rt_bind_peer(rt, rt->rt_dst, 1);
1489 	peer = rt->peer;
1490 	if (!peer) {
1491 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1492 		return;
1493 	}
1494 
1495 	/* No redirected packets during ip_rt_redirect_silence;
1496 	 * reset the algorithm.
1497 	 */
1498 	if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1499 		peer->rate_tokens = 0;
1500 
1501 	/* Too many ignored redirects; do not send anything
1502 	 * set dst.rate_last to the last seen redirected packet.
1503 	 */
1504 	if (peer->rate_tokens >= ip_rt_redirect_number) {
1505 		peer->rate_last = jiffies;
1506 		return;
1507 	}
1508 
1509 	/* Check for load limit; set rate_last to the latest sent
1510 	 * redirect.
1511 	 */
1512 	if (peer->rate_tokens == 0 ||
1513 	    time_after(jiffies,
1514 		       (peer->rate_last +
1515 			(ip_rt_redirect_load << peer->rate_tokens)))) {
1516 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1517 		peer->rate_last = jiffies;
1518 		++peer->rate_tokens;
1519 #ifdef CONFIG_IP_ROUTE_VERBOSE
1520 		if (log_martians &&
1521 		    peer->rate_tokens == ip_rt_redirect_number &&
1522 		    net_ratelimit())
1523 			printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1524 			       &ip_hdr(skb)->saddr, rt->rt_iif,
1525 				&rt->rt_dst, &rt->rt_gateway);
1526 #endif
1527 	}
1528 }
1529 
1530 static int ip_error(struct sk_buff *skb)
1531 {
1532 	struct rtable *rt = skb_rtable(skb);
1533 	struct inet_peer *peer;
1534 	unsigned long now;
1535 	bool send;
1536 	int code;
1537 
1538 	switch (rt->dst.error) {
1539 	case EINVAL:
1540 	default:
1541 		goto out;
1542 	case EHOSTUNREACH:
1543 		code = ICMP_HOST_UNREACH;
1544 		break;
1545 	case ENETUNREACH:
1546 		code = ICMP_NET_UNREACH;
1547 		IP_INC_STATS_BH(dev_net(rt->dst.dev),
1548 				IPSTATS_MIB_INNOROUTES);
1549 		break;
1550 	case EACCES:
1551 		code = ICMP_PKT_FILTERED;
1552 		break;
1553 	}
1554 
1555 	if (!rt->peer)
1556 		rt_bind_peer(rt, rt->rt_dst, 1);
1557 	peer = rt->peer;
1558 
1559 	send = true;
1560 	if (peer) {
1561 		now = jiffies;
1562 		peer->rate_tokens += now - peer->rate_last;
1563 		if (peer->rate_tokens > ip_rt_error_burst)
1564 			peer->rate_tokens = ip_rt_error_burst;
1565 		peer->rate_last = now;
1566 		if (peer->rate_tokens >= ip_rt_error_cost)
1567 			peer->rate_tokens -= ip_rt_error_cost;
1568 		else
1569 			send = false;
1570 	}
1571 	if (send)
1572 		icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1573 
1574 out:	kfree_skb(skb);
1575 	return 0;
1576 }
1577 
1578 /*
1579  *	The last two values are not from the RFC but
1580  *	are needed for AMPRnet AX.25 paths.
1581  */
1582 
1583 static const unsigned short mtu_plateau[] =
1584 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1585 
1586 static inline unsigned short guess_mtu(unsigned short old_mtu)
1587 {
1588 	int i;
1589 
1590 	for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1591 		if (old_mtu > mtu_plateau[i])
1592 			return mtu_plateau[i];
1593 	return 68;
1594 }
1595 
1596 unsigned short ip_rt_frag_needed(struct net *net, const struct iphdr *iph,
1597 				 unsigned short new_mtu,
1598 				 struct net_device *dev)
1599 {
1600 	unsigned short old_mtu = ntohs(iph->tot_len);
1601 	unsigned short est_mtu = 0;
1602 	struct inet_peer *peer;
1603 
1604 	peer = inet_getpeer_v4(iph->daddr, 1);
1605 	if (peer) {
1606 		unsigned short mtu = new_mtu;
1607 
1608 		if (new_mtu < 68 || new_mtu >= old_mtu) {
1609 			/* BSD 4.2 derived systems incorrectly adjust
1610 			 * tot_len by the IP header length, and report
1611 			 * a zero MTU in the ICMP message.
1612 			 */
1613 			if (mtu == 0 &&
1614 			    old_mtu >= 68 + (iph->ihl << 2))
1615 				old_mtu -= iph->ihl << 2;
1616 			mtu = guess_mtu(old_mtu);
1617 		}
1618 
1619 		if (mtu < ip_rt_min_pmtu)
1620 			mtu = ip_rt_min_pmtu;
1621 		if (!peer->pmtu_expires || mtu < peer->pmtu_learned) {
1622 			unsigned long pmtu_expires;
1623 
1624 			pmtu_expires = jiffies + ip_rt_mtu_expires;
1625 			if (!pmtu_expires)
1626 				pmtu_expires = 1UL;
1627 
1628 			est_mtu = mtu;
1629 			peer->pmtu_learned = mtu;
1630 			peer->pmtu_expires = pmtu_expires;
1631 			atomic_inc(&__rt_peer_genid);
1632 		}
1633 
1634 		inet_putpeer(peer);
1635 	}
1636 	return est_mtu ? : new_mtu;
1637 }
1638 
1639 static void check_peer_pmtu(struct dst_entry *dst, struct inet_peer *peer)
1640 {
1641 	unsigned long expires = ACCESS_ONCE(peer->pmtu_expires);
1642 
1643 	if (!expires)
1644 		return;
1645 	if (time_before(jiffies, expires)) {
1646 		u32 orig_dst_mtu = dst_mtu(dst);
1647 		if (peer->pmtu_learned < orig_dst_mtu) {
1648 			if (!peer->pmtu_orig)
1649 				peer->pmtu_orig = dst_metric_raw(dst, RTAX_MTU);
1650 			dst_metric_set(dst, RTAX_MTU, peer->pmtu_learned);
1651 		}
1652 	} else if (cmpxchg(&peer->pmtu_expires, expires, 0) == expires)
1653 		dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1654 }
1655 
1656 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1657 {
1658 	struct rtable *rt = (struct rtable *) dst;
1659 	struct inet_peer *peer;
1660 
1661 	dst_confirm(dst);
1662 
1663 	if (!rt->peer)
1664 		rt_bind_peer(rt, rt->rt_dst, 1);
1665 	peer = rt->peer;
1666 	if (peer) {
1667 		unsigned long pmtu_expires = ACCESS_ONCE(peer->pmtu_expires);
1668 
1669 		if (mtu < ip_rt_min_pmtu)
1670 			mtu = ip_rt_min_pmtu;
1671 		if (!pmtu_expires || mtu < peer->pmtu_learned) {
1672 
1673 			pmtu_expires = jiffies + ip_rt_mtu_expires;
1674 			if (!pmtu_expires)
1675 				pmtu_expires = 1UL;
1676 
1677 			peer->pmtu_learned = mtu;
1678 			peer->pmtu_expires = pmtu_expires;
1679 
1680 			atomic_inc(&__rt_peer_genid);
1681 			rt->rt_peer_genid = rt_peer_genid();
1682 		}
1683 		check_peer_pmtu(dst, peer);
1684 	}
1685 }
1686 
1687 
1688 static void ipv4_validate_peer(struct rtable *rt)
1689 {
1690 	if (rt->rt_peer_genid != rt_peer_genid()) {
1691 		struct inet_peer *peer;
1692 
1693 		if (!rt->peer)
1694 			rt_bind_peer(rt, rt->rt_dst, 0);
1695 
1696 		peer = rt->peer;
1697 		if (peer) {
1698 			check_peer_pmtu(&rt->dst, peer);
1699 
1700 			if (peer->redirect_genid != redirect_genid)
1701 				peer->redirect_learned.a4 = 0;
1702 			if (peer->redirect_learned.a4 &&
1703 			    peer->redirect_learned.a4 != rt->rt_gateway)
1704 				check_peer_redir(&rt->dst, peer);
1705 		}
1706 
1707 		rt->rt_peer_genid = rt_peer_genid();
1708 	}
1709 }
1710 
1711 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1712 {
1713 	struct rtable *rt = (struct rtable *) dst;
1714 
1715 	if (rt_is_expired(rt))
1716 		return NULL;
1717 	ipv4_validate_peer(rt);
1718 	return dst;
1719 }
1720 
1721 static void ipv4_dst_destroy(struct dst_entry *dst)
1722 {
1723 	struct rtable *rt = (struct rtable *) dst;
1724 	struct inet_peer *peer = rt->peer;
1725 
1726 	if (rt->fi) {
1727 		fib_info_put(rt->fi);
1728 		rt->fi = NULL;
1729 	}
1730 	if (peer) {
1731 		rt->peer = NULL;
1732 		inet_putpeer(peer);
1733 	}
1734 }
1735 
1736 
1737 static void ipv4_link_failure(struct sk_buff *skb)
1738 {
1739 	struct rtable *rt;
1740 
1741 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1742 
1743 	rt = skb_rtable(skb);
1744 	if (rt && rt->peer && peer_pmtu_cleaned(rt->peer))
1745 		dst_metric_set(&rt->dst, RTAX_MTU, rt->peer->pmtu_orig);
1746 }
1747 
1748 static int ip_rt_bug(struct sk_buff *skb)
1749 {
1750 	printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1751 		&ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1752 		skb->dev ? skb->dev->name : "?");
1753 	kfree_skb(skb);
1754 	WARN_ON(1);
1755 	return 0;
1756 }
1757 
1758 /*
1759    We do not cache source address of outgoing interface,
1760    because it is used only by IP RR, TS and SRR options,
1761    so that it out of fast path.
1762 
1763    BTW remember: "addr" is allowed to be not aligned
1764    in IP options!
1765  */
1766 
1767 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1768 {
1769 	__be32 src;
1770 
1771 	if (rt_is_output_route(rt))
1772 		src = ip_hdr(skb)->saddr;
1773 	else {
1774 		struct fib_result res;
1775 		struct flowi4 fl4;
1776 		struct iphdr *iph;
1777 
1778 		iph = ip_hdr(skb);
1779 
1780 		memset(&fl4, 0, sizeof(fl4));
1781 		fl4.daddr = iph->daddr;
1782 		fl4.saddr = iph->saddr;
1783 		fl4.flowi4_tos = RT_TOS(iph->tos);
1784 		fl4.flowi4_oif = rt->dst.dev->ifindex;
1785 		fl4.flowi4_iif = skb->dev->ifindex;
1786 		fl4.flowi4_mark = skb->mark;
1787 
1788 		rcu_read_lock();
1789 		if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1790 			src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1791 		else
1792 			src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1793 					RT_SCOPE_UNIVERSE);
1794 		rcu_read_unlock();
1795 	}
1796 	memcpy(addr, &src, 4);
1797 }
1798 
1799 #ifdef CONFIG_IP_ROUTE_CLASSID
1800 static void set_class_tag(struct rtable *rt, u32 tag)
1801 {
1802 	if (!(rt->dst.tclassid & 0xFFFF))
1803 		rt->dst.tclassid |= tag & 0xFFFF;
1804 	if (!(rt->dst.tclassid & 0xFFFF0000))
1805 		rt->dst.tclassid |= tag & 0xFFFF0000;
1806 }
1807 #endif
1808 
1809 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1810 {
1811 	unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1812 
1813 	if (advmss == 0) {
1814 		advmss = max_t(unsigned int, dst->dev->mtu - 40,
1815 			       ip_rt_min_advmss);
1816 		if (advmss > 65535 - 40)
1817 			advmss = 65535 - 40;
1818 	}
1819 	return advmss;
1820 }
1821 
1822 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1823 {
1824 	const struct rtable *rt = (const struct rtable *) dst;
1825 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
1826 
1827 	if (mtu && rt_is_output_route(rt))
1828 		return mtu;
1829 
1830 	mtu = dst->dev->mtu;
1831 
1832 	if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1833 
1834 		if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1835 			mtu = 576;
1836 	}
1837 
1838 	if (mtu > IP_MAX_MTU)
1839 		mtu = IP_MAX_MTU;
1840 
1841 	return mtu;
1842 }
1843 
1844 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1845 			    struct fib_info *fi)
1846 {
1847 	struct inet_peer *peer;
1848 	int create = 0;
1849 
1850 	/* If a peer entry exists for this destination, we must hook
1851 	 * it up in order to get at cached metrics.
1852 	 */
1853 	if (fl4 && (fl4->flowi4_flags & FLOWI_FLAG_PRECOW_METRICS))
1854 		create = 1;
1855 
1856 	rt->peer = peer = inet_getpeer_v4(rt->rt_dst, create);
1857 	if (peer) {
1858 		rt->rt_peer_genid = rt_peer_genid();
1859 		if (inet_metrics_new(peer))
1860 			memcpy(peer->metrics, fi->fib_metrics,
1861 			       sizeof(u32) * RTAX_MAX);
1862 		dst_init_metrics(&rt->dst, peer->metrics, false);
1863 
1864 		check_peer_pmtu(&rt->dst, peer);
1865 		if (peer->redirect_genid != redirect_genid)
1866 			peer->redirect_learned.a4 = 0;
1867 		if (peer->redirect_learned.a4 &&
1868 		    peer->redirect_learned.a4 != rt->rt_gateway) {
1869 			rt->rt_gateway = peer->redirect_learned.a4;
1870 			rt->rt_flags |= RTCF_REDIRECTED;
1871 		}
1872 	} else {
1873 		if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1874 			rt->fi = fi;
1875 			atomic_inc(&fi->fib_clntref);
1876 		}
1877 		dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1878 	}
1879 }
1880 
1881 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1882 			   const struct fib_result *res,
1883 			   struct fib_info *fi, u16 type, u32 itag)
1884 {
1885 	struct dst_entry *dst = &rt->dst;
1886 
1887 	if (fi) {
1888 		if (FIB_RES_GW(*res) &&
1889 		    FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1890 			rt->rt_gateway = FIB_RES_GW(*res);
1891 		rt_init_metrics(rt, fl4, fi);
1892 #ifdef CONFIG_IP_ROUTE_CLASSID
1893 		dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
1894 #endif
1895 	}
1896 
1897 	if (dst_mtu(dst) > IP_MAX_MTU)
1898 		dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU);
1899 	if (dst_metric_raw(dst, RTAX_ADVMSS) > 65535 - 40)
1900 		dst_metric_set(dst, RTAX_ADVMSS, 65535 - 40);
1901 
1902 #ifdef CONFIG_IP_ROUTE_CLASSID
1903 #ifdef CONFIG_IP_MULTIPLE_TABLES
1904 	set_class_tag(rt, fib_rules_tclass(res));
1905 #endif
1906 	set_class_tag(rt, itag);
1907 #endif
1908 }
1909 
1910 static struct rtable *rt_dst_alloc(struct net_device *dev,
1911 				   bool nopolicy, bool noxfrm)
1912 {
1913 	return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
1914 			 DST_HOST |
1915 			 (nopolicy ? DST_NOPOLICY : 0) |
1916 			 (noxfrm ? DST_NOXFRM : 0));
1917 }
1918 
1919 /* called in rcu_read_lock() section */
1920 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1921 				u8 tos, struct net_device *dev, int our)
1922 {
1923 	unsigned int hash;
1924 	struct rtable *rth;
1925 	__be32 spec_dst;
1926 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1927 	u32 itag = 0;
1928 	int err;
1929 
1930 	/* Primary sanity checks. */
1931 
1932 	if (in_dev == NULL)
1933 		return -EINVAL;
1934 
1935 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1936 	    ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1937 		goto e_inval;
1938 
1939 	if (ipv4_is_zeronet(saddr)) {
1940 		if (!ipv4_is_local_multicast(daddr))
1941 			goto e_inval;
1942 		spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1943 	} else {
1944 		err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
1945 					  &itag);
1946 		if (err < 0)
1947 			goto e_err;
1948 	}
1949 	rth = rt_dst_alloc(init_net.loopback_dev,
1950 			   IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1951 	if (!rth)
1952 		goto e_nobufs;
1953 
1954 #ifdef CONFIG_IP_ROUTE_CLASSID
1955 	rth->dst.tclassid = itag;
1956 #endif
1957 	rth->dst.output = ip_rt_bug;
1958 
1959 	rth->rt_key_dst	= daddr;
1960 	rth->rt_key_src	= saddr;
1961 	rth->rt_genid	= rt_genid(dev_net(dev));
1962 	rth->rt_flags	= RTCF_MULTICAST;
1963 	rth->rt_type	= RTN_MULTICAST;
1964 	rth->rt_key_tos	= tos;
1965 	rth->rt_dst	= daddr;
1966 	rth->rt_src	= saddr;
1967 	rth->rt_route_iif = dev->ifindex;
1968 	rth->rt_iif	= dev->ifindex;
1969 	rth->rt_oif	= 0;
1970 	rth->rt_mark    = skb->mark;
1971 	rth->rt_gateway	= daddr;
1972 	rth->rt_spec_dst= spec_dst;
1973 	rth->rt_peer_genid = 0;
1974 	rth->peer = NULL;
1975 	rth->fi = NULL;
1976 	if (our) {
1977 		rth->dst.input= ip_local_deliver;
1978 		rth->rt_flags |= RTCF_LOCAL;
1979 	}
1980 
1981 #ifdef CONFIG_IP_MROUTE
1982 	if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1983 		rth->dst.input = ip_mr_input;
1984 #endif
1985 	RT_CACHE_STAT_INC(in_slow_mc);
1986 
1987 	hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1988 	rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
1989 	return IS_ERR(rth) ? PTR_ERR(rth) : 0;
1990 
1991 e_nobufs:
1992 	return -ENOBUFS;
1993 e_inval:
1994 	return -EINVAL;
1995 e_err:
1996 	return err;
1997 }
1998 
1999 
2000 static void ip_handle_martian_source(struct net_device *dev,
2001 				     struct in_device *in_dev,
2002 				     struct sk_buff *skb,
2003 				     __be32 daddr,
2004 				     __be32 saddr)
2005 {
2006 	RT_CACHE_STAT_INC(in_martian_src);
2007 #ifdef CONFIG_IP_ROUTE_VERBOSE
2008 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
2009 		/*
2010 		 *	RFC1812 recommendation, if source is martian,
2011 		 *	the only hint is MAC header.
2012 		 */
2013 		printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
2014 			&daddr, &saddr, dev->name);
2015 		if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
2016 			int i;
2017 			const unsigned char *p = skb_mac_header(skb);
2018 			printk(KERN_WARNING "ll header: ");
2019 			for (i = 0; i < dev->hard_header_len; i++, p++) {
2020 				printk("%02x", *p);
2021 				if (i < (dev->hard_header_len - 1))
2022 					printk(":");
2023 			}
2024 			printk("\n");
2025 		}
2026 	}
2027 #endif
2028 }
2029 
2030 /* called in rcu_read_lock() section */
2031 static int __mkroute_input(struct sk_buff *skb,
2032 			   const struct fib_result *res,
2033 			   struct in_device *in_dev,
2034 			   __be32 daddr, __be32 saddr, u32 tos,
2035 			   struct rtable **result)
2036 {
2037 	struct rtable *rth;
2038 	int err;
2039 	struct in_device *out_dev;
2040 	unsigned int flags = 0;
2041 	__be32 spec_dst;
2042 	u32 itag;
2043 
2044 	/* get a working reference to the output device */
2045 	out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
2046 	if (out_dev == NULL) {
2047 		if (net_ratelimit())
2048 			printk(KERN_CRIT "Bug in ip_route_input" \
2049 			       "_slow(). Please, report\n");
2050 		return -EINVAL;
2051 	}
2052 
2053 
2054 	err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
2055 				  in_dev->dev, &spec_dst, &itag);
2056 	if (err < 0) {
2057 		ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
2058 					 saddr);
2059 
2060 		goto cleanup;
2061 	}
2062 
2063 	if (err)
2064 		flags |= RTCF_DIRECTSRC;
2065 
2066 	if (out_dev == in_dev && err &&
2067 	    (IN_DEV_SHARED_MEDIA(out_dev) ||
2068 	     inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2069 		flags |= RTCF_DOREDIRECT;
2070 
2071 	if (skb->protocol != htons(ETH_P_IP)) {
2072 		/* Not IP (i.e. ARP). Do not create route, if it is
2073 		 * invalid for proxy arp. DNAT routes are always valid.
2074 		 *
2075 		 * Proxy arp feature have been extended to allow, ARP
2076 		 * replies back to the same interface, to support
2077 		 * Private VLAN switch technologies. See arp.c.
2078 		 */
2079 		if (out_dev == in_dev &&
2080 		    IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2081 			err = -EINVAL;
2082 			goto cleanup;
2083 		}
2084 	}
2085 
2086 	rth = rt_dst_alloc(out_dev->dev,
2087 			   IN_DEV_CONF_GET(in_dev, NOPOLICY),
2088 			   IN_DEV_CONF_GET(out_dev, NOXFRM));
2089 	if (!rth) {
2090 		err = -ENOBUFS;
2091 		goto cleanup;
2092 	}
2093 
2094 	rth->rt_key_dst	= daddr;
2095 	rth->rt_key_src	= saddr;
2096 	rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2097 	rth->rt_flags = flags;
2098 	rth->rt_type = res->type;
2099 	rth->rt_key_tos	= tos;
2100 	rth->rt_dst	= daddr;
2101 	rth->rt_src	= saddr;
2102 	rth->rt_route_iif = in_dev->dev->ifindex;
2103 	rth->rt_iif 	= in_dev->dev->ifindex;
2104 	rth->rt_oif 	= 0;
2105 	rth->rt_mark    = skb->mark;
2106 	rth->rt_gateway	= daddr;
2107 	rth->rt_spec_dst= spec_dst;
2108 	rth->rt_peer_genid = 0;
2109 	rth->peer = NULL;
2110 	rth->fi = NULL;
2111 
2112 	rth->dst.input = ip_forward;
2113 	rth->dst.output = ip_output;
2114 
2115 	rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
2116 
2117 	*result = rth;
2118 	err = 0;
2119  cleanup:
2120 	return err;
2121 }
2122 
2123 static int ip_mkroute_input(struct sk_buff *skb,
2124 			    struct fib_result *res,
2125 			    const struct flowi4 *fl4,
2126 			    struct in_device *in_dev,
2127 			    __be32 daddr, __be32 saddr, u32 tos)
2128 {
2129 	struct rtable* rth = NULL;
2130 	int err;
2131 	unsigned hash;
2132 
2133 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2134 	if (res->fi && res->fi->fib_nhs > 1)
2135 		fib_select_multipath(res);
2136 #endif
2137 
2138 	/* create a routing cache entry */
2139 	err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2140 	if (err)
2141 		return err;
2142 
2143 	/* put it into the cache */
2144 	hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
2145 		       rt_genid(dev_net(rth->dst.dev)));
2146 	rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
2147 	if (IS_ERR(rth))
2148 		return PTR_ERR(rth);
2149 	return 0;
2150 }
2151 
2152 /*
2153  *	NOTE. We drop all the packets that has local source
2154  *	addresses, because every properly looped back packet
2155  *	must have correct destination already attached by output routine.
2156  *
2157  *	Such approach solves two big problems:
2158  *	1. Not simplex devices are handled properly.
2159  *	2. IP spoofing attempts are filtered with 100% of guarantee.
2160  *	called with rcu_read_lock()
2161  */
2162 
2163 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2164 			       u8 tos, struct net_device *dev)
2165 {
2166 	struct fib_result res;
2167 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2168 	struct flowi4	fl4;
2169 	unsigned	flags = 0;
2170 	u32		itag = 0;
2171 	struct rtable * rth;
2172 	unsigned	hash;
2173 	__be32		spec_dst;
2174 	int		err = -EINVAL;
2175 	struct net    * net = dev_net(dev);
2176 
2177 	/* IP on this device is disabled. */
2178 
2179 	if (!in_dev)
2180 		goto out;
2181 
2182 	/* Check for the most weird martians, which can be not detected
2183 	   by fib_lookup.
2184 	 */
2185 
2186 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2187 	    ipv4_is_loopback(saddr))
2188 		goto martian_source;
2189 
2190 	if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2191 		goto brd_input;
2192 
2193 	/* Accept zero addresses only to limited broadcast;
2194 	 * I even do not know to fix it or not. Waiting for complains :-)
2195 	 */
2196 	if (ipv4_is_zeronet(saddr))
2197 		goto martian_source;
2198 
2199 	if (ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr))
2200 		goto martian_destination;
2201 
2202 	/*
2203 	 *	Now we are ready to route packet.
2204 	 */
2205 	fl4.flowi4_oif = 0;
2206 	fl4.flowi4_iif = dev->ifindex;
2207 	fl4.flowi4_mark = skb->mark;
2208 	fl4.flowi4_tos = tos;
2209 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2210 	fl4.daddr = daddr;
2211 	fl4.saddr = saddr;
2212 	err = fib_lookup(net, &fl4, &res);
2213 	if (err != 0) {
2214 		if (!IN_DEV_FORWARD(in_dev))
2215 			goto e_hostunreach;
2216 		goto no_route;
2217 	}
2218 
2219 	RT_CACHE_STAT_INC(in_slow_tot);
2220 
2221 	if (res.type == RTN_BROADCAST)
2222 		goto brd_input;
2223 
2224 	if (res.type == RTN_LOCAL) {
2225 		err = fib_validate_source(skb, saddr, daddr, tos,
2226 					  net->loopback_dev->ifindex,
2227 					  dev, &spec_dst, &itag);
2228 		if (err < 0)
2229 			goto martian_source_keep_err;
2230 		if (err)
2231 			flags |= RTCF_DIRECTSRC;
2232 		spec_dst = daddr;
2233 		goto local_input;
2234 	}
2235 
2236 	if (!IN_DEV_FORWARD(in_dev))
2237 		goto e_hostunreach;
2238 	if (res.type != RTN_UNICAST)
2239 		goto martian_destination;
2240 
2241 	err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2242 out:	return err;
2243 
2244 brd_input:
2245 	if (skb->protocol != htons(ETH_P_IP))
2246 		goto e_inval;
2247 
2248 	if (ipv4_is_zeronet(saddr))
2249 		spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2250 	else {
2251 		err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
2252 					  &itag);
2253 		if (err < 0)
2254 			goto martian_source_keep_err;
2255 		if (err)
2256 			flags |= RTCF_DIRECTSRC;
2257 	}
2258 	flags |= RTCF_BROADCAST;
2259 	res.type = RTN_BROADCAST;
2260 	RT_CACHE_STAT_INC(in_brd);
2261 
2262 local_input:
2263 	rth = rt_dst_alloc(net->loopback_dev,
2264 			   IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2265 	if (!rth)
2266 		goto e_nobufs;
2267 
2268 	rth->dst.input= ip_local_deliver;
2269 	rth->dst.output= ip_rt_bug;
2270 #ifdef CONFIG_IP_ROUTE_CLASSID
2271 	rth->dst.tclassid = itag;
2272 #endif
2273 
2274 	rth->rt_key_dst	= daddr;
2275 	rth->rt_key_src	= saddr;
2276 	rth->rt_genid = rt_genid(net);
2277 	rth->rt_flags 	= flags|RTCF_LOCAL;
2278 	rth->rt_type	= res.type;
2279 	rth->rt_key_tos	= tos;
2280 	rth->rt_dst	= daddr;
2281 	rth->rt_src	= saddr;
2282 #ifdef CONFIG_IP_ROUTE_CLASSID
2283 	rth->dst.tclassid = itag;
2284 #endif
2285 	rth->rt_route_iif = dev->ifindex;
2286 	rth->rt_iif	= dev->ifindex;
2287 	rth->rt_oif	= 0;
2288 	rth->rt_mark    = skb->mark;
2289 	rth->rt_gateway	= daddr;
2290 	rth->rt_spec_dst= spec_dst;
2291 	rth->rt_peer_genid = 0;
2292 	rth->peer = NULL;
2293 	rth->fi = NULL;
2294 	if (res.type == RTN_UNREACHABLE) {
2295 		rth->dst.input= ip_error;
2296 		rth->dst.error= -err;
2297 		rth->rt_flags 	&= ~RTCF_LOCAL;
2298 	}
2299 	hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2300 	rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2301 	err = 0;
2302 	if (IS_ERR(rth))
2303 		err = PTR_ERR(rth);
2304 	goto out;
2305 
2306 no_route:
2307 	RT_CACHE_STAT_INC(in_no_route);
2308 	spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2309 	res.type = RTN_UNREACHABLE;
2310 	if (err == -ESRCH)
2311 		err = -ENETUNREACH;
2312 	goto local_input;
2313 
2314 	/*
2315 	 *	Do not cache martian addresses: they should be logged (RFC1812)
2316 	 */
2317 martian_destination:
2318 	RT_CACHE_STAT_INC(in_martian_dst);
2319 #ifdef CONFIG_IP_ROUTE_VERBOSE
2320 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2321 		printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2322 			&daddr, &saddr, dev->name);
2323 #endif
2324 
2325 e_hostunreach:
2326 	err = -EHOSTUNREACH;
2327 	goto out;
2328 
2329 e_inval:
2330 	err = -EINVAL;
2331 	goto out;
2332 
2333 e_nobufs:
2334 	err = -ENOBUFS;
2335 	goto out;
2336 
2337 martian_source:
2338 	err = -EINVAL;
2339 martian_source_keep_err:
2340 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2341 	goto out;
2342 }
2343 
2344 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2345 			   u8 tos, struct net_device *dev, bool noref)
2346 {
2347 	struct rtable * rth;
2348 	unsigned	hash;
2349 	int iif = dev->ifindex;
2350 	struct net *net;
2351 	int res;
2352 
2353 	net = dev_net(dev);
2354 
2355 	rcu_read_lock();
2356 
2357 	if (!rt_caching(net))
2358 		goto skip_cache;
2359 
2360 	tos &= IPTOS_RT_MASK;
2361 	hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2362 
2363 	for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2364 	     rth = rcu_dereference(rth->dst.rt_next)) {
2365 		if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2366 		     ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2367 		     (rth->rt_route_iif ^ iif) |
2368 		     (rth->rt_key_tos ^ tos)) == 0 &&
2369 		    rth->rt_mark == skb->mark &&
2370 		    net_eq(dev_net(rth->dst.dev), net) &&
2371 		    !rt_is_expired(rth)) {
2372 			ipv4_validate_peer(rth);
2373 			if (noref) {
2374 				dst_use_noref(&rth->dst, jiffies);
2375 				skb_dst_set_noref(skb, &rth->dst);
2376 			} else {
2377 				dst_use(&rth->dst, jiffies);
2378 				skb_dst_set(skb, &rth->dst);
2379 			}
2380 			RT_CACHE_STAT_INC(in_hit);
2381 			rcu_read_unlock();
2382 			return 0;
2383 		}
2384 		RT_CACHE_STAT_INC(in_hlist_search);
2385 	}
2386 
2387 skip_cache:
2388 	/* Multicast recognition logic is moved from route cache to here.
2389 	   The problem was that too many Ethernet cards have broken/missing
2390 	   hardware multicast filters :-( As result the host on multicasting
2391 	   network acquires a lot of useless route cache entries, sort of
2392 	   SDR messages from all the world. Now we try to get rid of them.
2393 	   Really, provided software IP multicast filter is organized
2394 	   reasonably (at least, hashed), it does not result in a slowdown
2395 	   comparing with route cache reject entries.
2396 	   Note, that multicast routers are not affected, because
2397 	   route cache entry is created eventually.
2398 	 */
2399 	if (ipv4_is_multicast(daddr)) {
2400 		struct in_device *in_dev = __in_dev_get_rcu(dev);
2401 
2402 		if (in_dev) {
2403 			int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2404 						  ip_hdr(skb)->protocol);
2405 			if (our
2406 #ifdef CONFIG_IP_MROUTE
2407 				||
2408 			    (!ipv4_is_local_multicast(daddr) &&
2409 			     IN_DEV_MFORWARD(in_dev))
2410 #endif
2411 			   ) {
2412 				int res = ip_route_input_mc(skb, daddr, saddr,
2413 							    tos, dev, our);
2414 				rcu_read_unlock();
2415 				return res;
2416 			}
2417 		}
2418 		rcu_read_unlock();
2419 		return -EINVAL;
2420 	}
2421 	res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2422 	rcu_read_unlock();
2423 	return res;
2424 }
2425 EXPORT_SYMBOL(ip_route_input_common);
2426 
2427 /* called with rcu_read_lock() */
2428 static struct rtable *__mkroute_output(const struct fib_result *res,
2429 				       const struct flowi4 *fl4,
2430 				       __be32 orig_daddr, __be32 orig_saddr,
2431 				       int orig_oif, __u8 orig_rtos,
2432 				       struct net_device *dev_out,
2433 				       unsigned int flags)
2434 {
2435 	struct fib_info *fi = res->fi;
2436 	struct in_device *in_dev;
2437 	u16 type = res->type;
2438 	struct rtable *rth;
2439 
2440 	if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2441 		return ERR_PTR(-EINVAL);
2442 
2443 	if (ipv4_is_lbcast(fl4->daddr))
2444 		type = RTN_BROADCAST;
2445 	else if (ipv4_is_multicast(fl4->daddr))
2446 		type = RTN_MULTICAST;
2447 	else if (ipv4_is_zeronet(fl4->daddr))
2448 		return ERR_PTR(-EINVAL);
2449 
2450 	if (dev_out->flags & IFF_LOOPBACK)
2451 		flags |= RTCF_LOCAL;
2452 
2453 	in_dev = __in_dev_get_rcu(dev_out);
2454 	if (!in_dev)
2455 		return ERR_PTR(-EINVAL);
2456 
2457 	if (type == RTN_BROADCAST) {
2458 		flags |= RTCF_BROADCAST | RTCF_LOCAL;
2459 		fi = NULL;
2460 	} else if (type == RTN_MULTICAST) {
2461 		flags |= RTCF_MULTICAST | RTCF_LOCAL;
2462 		if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2463 				     fl4->flowi4_proto))
2464 			flags &= ~RTCF_LOCAL;
2465 		/* If multicast route do not exist use
2466 		 * default one, but do not gateway in this case.
2467 		 * Yes, it is hack.
2468 		 */
2469 		if (fi && res->prefixlen < 4)
2470 			fi = NULL;
2471 	}
2472 
2473 	rth = rt_dst_alloc(dev_out,
2474 			   IN_DEV_CONF_GET(in_dev, NOPOLICY),
2475 			   IN_DEV_CONF_GET(in_dev, NOXFRM));
2476 	if (!rth)
2477 		return ERR_PTR(-ENOBUFS);
2478 
2479 	rth->dst.output = ip_output;
2480 
2481 	rth->rt_key_dst	= orig_daddr;
2482 	rth->rt_key_src	= orig_saddr;
2483 	rth->rt_genid = rt_genid(dev_net(dev_out));
2484 	rth->rt_flags	= flags;
2485 	rth->rt_type	= type;
2486 	rth->rt_key_tos	= orig_rtos;
2487 	rth->rt_dst	= fl4->daddr;
2488 	rth->rt_src	= fl4->saddr;
2489 	rth->rt_route_iif = 0;
2490 	rth->rt_iif	= orig_oif ? : dev_out->ifindex;
2491 	rth->rt_oif	= orig_oif;
2492 	rth->rt_mark    = fl4->flowi4_mark;
2493 	rth->rt_gateway = fl4->daddr;
2494 	rth->rt_spec_dst= fl4->saddr;
2495 	rth->rt_peer_genid = 0;
2496 	rth->peer = NULL;
2497 	rth->fi = NULL;
2498 
2499 	RT_CACHE_STAT_INC(out_slow_tot);
2500 
2501 	if (flags & RTCF_LOCAL) {
2502 		rth->dst.input = ip_local_deliver;
2503 		rth->rt_spec_dst = fl4->daddr;
2504 	}
2505 	if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2506 		rth->rt_spec_dst = fl4->saddr;
2507 		if (flags & RTCF_LOCAL &&
2508 		    !(dev_out->flags & IFF_LOOPBACK)) {
2509 			rth->dst.output = ip_mc_output;
2510 			RT_CACHE_STAT_INC(out_slow_mc);
2511 		}
2512 #ifdef CONFIG_IP_MROUTE
2513 		if (type == RTN_MULTICAST) {
2514 			if (IN_DEV_MFORWARD(in_dev) &&
2515 			    !ipv4_is_local_multicast(fl4->daddr)) {
2516 				rth->dst.input = ip_mr_input;
2517 				rth->dst.output = ip_mc_output;
2518 			}
2519 		}
2520 #endif
2521 	}
2522 
2523 	rt_set_nexthop(rth, fl4, res, fi, type, 0);
2524 
2525 	return rth;
2526 }
2527 
2528 /*
2529  * Major route resolver routine.
2530  * called with rcu_read_lock();
2531  */
2532 
2533 static struct rtable *ip_route_output_slow(struct net *net, struct flowi4 *fl4)
2534 {
2535 	struct net_device *dev_out = NULL;
2536 	__u8 tos = RT_FL_TOS(fl4);
2537 	unsigned int flags = 0;
2538 	struct fib_result res;
2539 	struct rtable *rth;
2540 	__be32 orig_daddr;
2541 	__be32 orig_saddr;
2542 	int orig_oif;
2543 
2544 	res.fi		= NULL;
2545 #ifdef CONFIG_IP_MULTIPLE_TABLES
2546 	res.r		= NULL;
2547 #endif
2548 
2549 	orig_daddr = fl4->daddr;
2550 	orig_saddr = fl4->saddr;
2551 	orig_oif = fl4->flowi4_oif;
2552 
2553 	fl4->flowi4_iif = net->loopback_dev->ifindex;
2554 	fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2555 	fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2556 			 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2557 
2558 	rcu_read_lock();
2559 	if (fl4->saddr) {
2560 		rth = ERR_PTR(-EINVAL);
2561 		if (ipv4_is_multicast(fl4->saddr) ||
2562 		    ipv4_is_lbcast(fl4->saddr) ||
2563 		    ipv4_is_zeronet(fl4->saddr))
2564 			goto out;
2565 
2566 		/* I removed check for oif == dev_out->oif here.
2567 		   It was wrong for two reasons:
2568 		   1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2569 		      is assigned to multiple interfaces.
2570 		   2. Moreover, we are allowed to send packets with saddr
2571 		      of another iface. --ANK
2572 		 */
2573 
2574 		if (fl4->flowi4_oif == 0 &&
2575 		    (ipv4_is_multicast(fl4->daddr) ||
2576 		     ipv4_is_lbcast(fl4->daddr))) {
2577 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2578 			dev_out = __ip_dev_find(net, fl4->saddr, false);
2579 			if (dev_out == NULL)
2580 				goto out;
2581 
2582 			/* Special hack: user can direct multicasts
2583 			   and limited broadcast via necessary interface
2584 			   without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2585 			   This hack is not just for fun, it allows
2586 			   vic,vat and friends to work.
2587 			   They bind socket to loopback, set ttl to zero
2588 			   and expect that it will work.
2589 			   From the viewpoint of routing cache they are broken,
2590 			   because we are not allowed to build multicast path
2591 			   with loopback source addr (look, routing cache
2592 			   cannot know, that ttl is zero, so that packet
2593 			   will not leave this host and route is valid).
2594 			   Luckily, this hack is good workaround.
2595 			 */
2596 
2597 			fl4->flowi4_oif = dev_out->ifindex;
2598 			goto make_route;
2599 		}
2600 
2601 		if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2602 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2603 			if (!__ip_dev_find(net, fl4->saddr, false))
2604 				goto out;
2605 		}
2606 	}
2607 
2608 
2609 	if (fl4->flowi4_oif) {
2610 		dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2611 		rth = ERR_PTR(-ENODEV);
2612 		if (dev_out == NULL)
2613 			goto out;
2614 
2615 		/* RACE: Check return value of inet_select_addr instead. */
2616 		if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2617 			rth = ERR_PTR(-ENETUNREACH);
2618 			goto out;
2619 		}
2620 		if (ipv4_is_local_multicast(fl4->daddr) ||
2621 		    ipv4_is_lbcast(fl4->daddr)) {
2622 			if (!fl4->saddr)
2623 				fl4->saddr = inet_select_addr(dev_out, 0,
2624 							      RT_SCOPE_LINK);
2625 			goto make_route;
2626 		}
2627 		if (fl4->saddr) {
2628 			if (ipv4_is_multicast(fl4->daddr))
2629 				fl4->saddr = inet_select_addr(dev_out, 0,
2630 							      fl4->flowi4_scope);
2631 			else if (!fl4->daddr)
2632 				fl4->saddr = inet_select_addr(dev_out, 0,
2633 							      RT_SCOPE_HOST);
2634 		}
2635 	}
2636 
2637 	if (!fl4->daddr) {
2638 		fl4->daddr = fl4->saddr;
2639 		if (!fl4->daddr)
2640 			fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2641 		dev_out = net->loopback_dev;
2642 		fl4->flowi4_oif = net->loopback_dev->ifindex;
2643 		res.type = RTN_LOCAL;
2644 		flags |= RTCF_LOCAL;
2645 		goto make_route;
2646 	}
2647 
2648 	if (fib_lookup(net, fl4, &res)) {
2649 		res.fi = NULL;
2650 		if (fl4->flowi4_oif) {
2651 			/* Apparently, routing tables are wrong. Assume,
2652 			   that the destination is on link.
2653 
2654 			   WHY? DW.
2655 			   Because we are allowed to send to iface
2656 			   even if it has NO routes and NO assigned
2657 			   addresses. When oif is specified, routing
2658 			   tables are looked up with only one purpose:
2659 			   to catch if destination is gatewayed, rather than
2660 			   direct. Moreover, if MSG_DONTROUTE is set,
2661 			   we send packet, ignoring both routing tables
2662 			   and ifaddr state. --ANK
2663 
2664 
2665 			   We could make it even if oif is unknown,
2666 			   likely IPv6, but we do not.
2667 			 */
2668 
2669 			if (fl4->saddr == 0)
2670 				fl4->saddr = inet_select_addr(dev_out, 0,
2671 							      RT_SCOPE_LINK);
2672 			res.type = RTN_UNICAST;
2673 			goto make_route;
2674 		}
2675 		rth = ERR_PTR(-ENETUNREACH);
2676 		goto out;
2677 	}
2678 
2679 	if (res.type == RTN_LOCAL) {
2680 		if (!fl4->saddr) {
2681 			if (res.fi->fib_prefsrc)
2682 				fl4->saddr = res.fi->fib_prefsrc;
2683 			else
2684 				fl4->saddr = fl4->daddr;
2685 		}
2686 		dev_out = net->loopback_dev;
2687 		fl4->flowi4_oif = dev_out->ifindex;
2688 		res.fi = NULL;
2689 		flags |= RTCF_LOCAL;
2690 		goto make_route;
2691 	}
2692 
2693 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2694 	if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2695 		fib_select_multipath(&res);
2696 	else
2697 #endif
2698 	if (!res.prefixlen &&
2699 	    res.table->tb_num_default > 1 &&
2700 	    res.type == RTN_UNICAST && !fl4->flowi4_oif)
2701 		fib_select_default(&res);
2702 
2703 	if (!fl4->saddr)
2704 		fl4->saddr = FIB_RES_PREFSRC(net, res);
2705 
2706 	dev_out = FIB_RES_DEV(res);
2707 	fl4->flowi4_oif = dev_out->ifindex;
2708 
2709 
2710 make_route:
2711 	rth = __mkroute_output(&res, fl4, orig_daddr, orig_saddr, orig_oif,
2712 			       tos, dev_out, flags);
2713 	if (!IS_ERR(rth)) {
2714 		unsigned int hash;
2715 
2716 		hash = rt_hash(orig_daddr, orig_saddr, orig_oif,
2717 			       rt_genid(dev_net(dev_out)));
2718 		rth = rt_intern_hash(hash, rth, NULL, orig_oif);
2719 	}
2720 
2721 out:
2722 	rcu_read_unlock();
2723 	return rth;
2724 }
2725 
2726 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *flp4)
2727 {
2728 	struct rtable *rth;
2729 	unsigned int hash;
2730 
2731 	if (!rt_caching(net))
2732 		goto slow_output;
2733 
2734 	hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2735 
2736 	rcu_read_lock_bh();
2737 	for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2738 		rth = rcu_dereference_bh(rth->dst.rt_next)) {
2739 		if (rth->rt_key_dst == flp4->daddr &&
2740 		    rth->rt_key_src == flp4->saddr &&
2741 		    rt_is_output_route(rth) &&
2742 		    rth->rt_oif == flp4->flowi4_oif &&
2743 		    rth->rt_mark == flp4->flowi4_mark &&
2744 		    !((rth->rt_key_tos ^ flp4->flowi4_tos) &
2745 			    (IPTOS_RT_MASK | RTO_ONLINK)) &&
2746 		    net_eq(dev_net(rth->dst.dev), net) &&
2747 		    !rt_is_expired(rth)) {
2748 			ipv4_validate_peer(rth);
2749 			dst_use(&rth->dst, jiffies);
2750 			RT_CACHE_STAT_INC(out_hit);
2751 			rcu_read_unlock_bh();
2752 			if (!flp4->saddr)
2753 				flp4->saddr = rth->rt_src;
2754 			if (!flp4->daddr)
2755 				flp4->daddr = rth->rt_dst;
2756 			return rth;
2757 		}
2758 		RT_CACHE_STAT_INC(out_hlist_search);
2759 	}
2760 	rcu_read_unlock_bh();
2761 
2762 slow_output:
2763 	return ip_route_output_slow(net, flp4);
2764 }
2765 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2766 
2767 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2768 {
2769 	return NULL;
2770 }
2771 
2772 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2773 {
2774 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2775 
2776 	return mtu ? : dst->dev->mtu;
2777 }
2778 
2779 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2780 {
2781 }
2782 
2783 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2784 					  unsigned long old)
2785 {
2786 	return NULL;
2787 }
2788 
2789 static struct dst_ops ipv4_dst_blackhole_ops = {
2790 	.family			=	AF_INET,
2791 	.protocol		=	cpu_to_be16(ETH_P_IP),
2792 	.destroy		=	ipv4_dst_destroy,
2793 	.check			=	ipv4_blackhole_dst_check,
2794 	.mtu			=	ipv4_blackhole_mtu,
2795 	.default_advmss		=	ipv4_default_advmss,
2796 	.update_pmtu		=	ipv4_rt_blackhole_update_pmtu,
2797 	.cow_metrics		=	ipv4_rt_blackhole_cow_metrics,
2798 	.neigh_lookup		=	ipv4_neigh_lookup,
2799 };
2800 
2801 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2802 {
2803 	struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
2804 	struct rtable *ort = (struct rtable *) dst_orig;
2805 
2806 	if (rt) {
2807 		struct dst_entry *new = &rt->dst;
2808 
2809 		new->__use = 1;
2810 		new->input = dst_discard;
2811 		new->output = dst_discard;
2812 		dst_copy_metrics(new, &ort->dst);
2813 
2814 		new->dev = ort->dst.dev;
2815 		if (new->dev)
2816 			dev_hold(new->dev);
2817 
2818 		rt->rt_key_dst = ort->rt_key_dst;
2819 		rt->rt_key_src = ort->rt_key_src;
2820 		rt->rt_key_tos = ort->rt_key_tos;
2821 		rt->rt_route_iif = ort->rt_route_iif;
2822 		rt->rt_iif = ort->rt_iif;
2823 		rt->rt_oif = ort->rt_oif;
2824 		rt->rt_mark = ort->rt_mark;
2825 
2826 		rt->rt_genid = rt_genid(net);
2827 		rt->rt_flags = ort->rt_flags;
2828 		rt->rt_type = ort->rt_type;
2829 		rt->rt_dst = ort->rt_dst;
2830 		rt->rt_src = ort->rt_src;
2831 		rt->rt_gateway = ort->rt_gateway;
2832 		rt->rt_spec_dst = ort->rt_spec_dst;
2833 		rt->peer = ort->peer;
2834 		if (rt->peer)
2835 			atomic_inc(&rt->peer->refcnt);
2836 		rt->fi = ort->fi;
2837 		if (rt->fi)
2838 			atomic_inc(&rt->fi->fib_clntref);
2839 
2840 		dst_free(new);
2841 	}
2842 
2843 	dst_release(dst_orig);
2844 
2845 	return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2846 }
2847 
2848 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2849 				    struct sock *sk)
2850 {
2851 	struct rtable *rt = __ip_route_output_key(net, flp4);
2852 
2853 	if (IS_ERR(rt))
2854 		return rt;
2855 
2856 	if (flp4->flowi4_proto)
2857 		rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2858 						   flowi4_to_flowi(flp4),
2859 						   sk, 0);
2860 
2861 	return rt;
2862 }
2863 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2864 
2865 static int rt_fill_info(struct net *net,
2866 			struct sk_buff *skb, u32 pid, u32 seq, int event,
2867 			int nowait, unsigned int flags)
2868 {
2869 	struct rtable *rt = skb_rtable(skb);
2870 	struct rtmsg *r;
2871 	struct nlmsghdr *nlh;
2872 	unsigned long expires = 0;
2873 	const struct inet_peer *peer = rt->peer;
2874 	u32 id = 0, ts = 0, tsage = 0, error;
2875 
2876 	nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2877 	if (nlh == NULL)
2878 		return -EMSGSIZE;
2879 
2880 	r = nlmsg_data(nlh);
2881 	r->rtm_family	 = AF_INET;
2882 	r->rtm_dst_len	= 32;
2883 	r->rtm_src_len	= 0;
2884 	r->rtm_tos	= rt->rt_key_tos;
2885 	r->rtm_table	= RT_TABLE_MAIN;
2886 	NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2887 	r->rtm_type	= rt->rt_type;
2888 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2889 	r->rtm_protocol = RTPROT_UNSPEC;
2890 	r->rtm_flags	= (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2891 	if (rt->rt_flags & RTCF_NOTIFY)
2892 		r->rtm_flags |= RTM_F_NOTIFY;
2893 
2894 	NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2895 
2896 	if (rt->rt_key_src) {
2897 		r->rtm_src_len = 32;
2898 		NLA_PUT_BE32(skb, RTA_SRC, rt->rt_key_src);
2899 	}
2900 	if (rt->dst.dev)
2901 		NLA_PUT_U32(skb, RTA_OIF, rt->dst.dev->ifindex);
2902 #ifdef CONFIG_IP_ROUTE_CLASSID
2903 	if (rt->dst.tclassid)
2904 		NLA_PUT_U32(skb, RTA_FLOW, rt->dst.tclassid);
2905 #endif
2906 	if (rt_is_input_route(rt))
2907 		NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2908 	else if (rt->rt_src != rt->rt_key_src)
2909 		NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2910 
2911 	if (rt->rt_dst != rt->rt_gateway)
2912 		NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2913 
2914 	if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2915 		goto nla_put_failure;
2916 
2917 	if (rt->rt_mark)
2918 		NLA_PUT_BE32(skb, RTA_MARK, rt->rt_mark);
2919 
2920 	error = rt->dst.error;
2921 	if (peer) {
2922 		inet_peer_refcheck(rt->peer);
2923 		id = atomic_read(&peer->ip_id_count) & 0xffff;
2924 		if (peer->tcp_ts_stamp) {
2925 			ts = peer->tcp_ts;
2926 			tsage = get_seconds() - peer->tcp_ts_stamp;
2927 		}
2928 		expires = ACCESS_ONCE(peer->pmtu_expires);
2929 		if (expires) {
2930 			if (time_before(jiffies, expires))
2931 				expires -= jiffies;
2932 			else
2933 				expires = 0;
2934 		}
2935 	}
2936 
2937 	if (rt_is_input_route(rt)) {
2938 #ifdef CONFIG_IP_MROUTE
2939 		__be32 dst = rt->rt_dst;
2940 
2941 		if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2942 		    IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2943 			int err = ipmr_get_route(net, skb,
2944 						 rt->rt_src, rt->rt_dst,
2945 						 r, nowait);
2946 			if (err <= 0) {
2947 				if (!nowait) {
2948 					if (err == 0)
2949 						return 0;
2950 					goto nla_put_failure;
2951 				} else {
2952 					if (err == -EMSGSIZE)
2953 						goto nla_put_failure;
2954 					error = err;
2955 				}
2956 			}
2957 		} else
2958 #endif
2959 			NLA_PUT_U32(skb, RTA_IIF, rt->rt_iif);
2960 	}
2961 
2962 	if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
2963 			       expires, error) < 0)
2964 		goto nla_put_failure;
2965 
2966 	return nlmsg_end(skb, nlh);
2967 
2968 nla_put_failure:
2969 	nlmsg_cancel(skb, nlh);
2970 	return -EMSGSIZE;
2971 }
2972 
2973 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2974 {
2975 	struct net *net = sock_net(in_skb->sk);
2976 	struct rtmsg *rtm;
2977 	struct nlattr *tb[RTA_MAX+1];
2978 	struct rtable *rt = NULL;
2979 	__be32 dst = 0;
2980 	__be32 src = 0;
2981 	u32 iif;
2982 	int err;
2983 	int mark;
2984 	struct sk_buff *skb;
2985 
2986 	err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2987 	if (err < 0)
2988 		goto errout;
2989 
2990 	rtm = nlmsg_data(nlh);
2991 
2992 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2993 	if (skb == NULL) {
2994 		err = -ENOBUFS;
2995 		goto errout;
2996 	}
2997 
2998 	/* Reserve room for dummy headers, this skb can pass
2999 	   through good chunk of routing engine.
3000 	 */
3001 	skb_reset_mac_header(skb);
3002 	skb_reset_network_header(skb);
3003 
3004 	/* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
3005 	ip_hdr(skb)->protocol = IPPROTO_ICMP;
3006 	skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
3007 
3008 	src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
3009 	dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
3010 	iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3011 	mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3012 
3013 	if (iif) {
3014 		struct net_device *dev;
3015 
3016 		dev = __dev_get_by_index(net, iif);
3017 		if (dev == NULL) {
3018 			err = -ENODEV;
3019 			goto errout_free;
3020 		}
3021 
3022 		skb->protocol	= htons(ETH_P_IP);
3023 		skb->dev	= dev;
3024 		skb->mark	= mark;
3025 		local_bh_disable();
3026 		err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
3027 		local_bh_enable();
3028 
3029 		rt = skb_rtable(skb);
3030 		if (err == 0 && rt->dst.error)
3031 			err = -rt->dst.error;
3032 	} else {
3033 		struct flowi4 fl4 = {
3034 			.daddr = dst,
3035 			.saddr = src,
3036 			.flowi4_tos = rtm->rtm_tos,
3037 			.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
3038 			.flowi4_mark = mark,
3039 		};
3040 		rt = ip_route_output_key(net, &fl4);
3041 
3042 		err = 0;
3043 		if (IS_ERR(rt))
3044 			err = PTR_ERR(rt);
3045 	}
3046 
3047 	if (err)
3048 		goto errout_free;
3049 
3050 	skb_dst_set(skb, &rt->dst);
3051 	if (rtm->rtm_flags & RTM_F_NOTIFY)
3052 		rt->rt_flags |= RTCF_NOTIFY;
3053 
3054 	err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3055 			   RTM_NEWROUTE, 0, 0);
3056 	if (err <= 0)
3057 		goto errout_free;
3058 
3059 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3060 errout:
3061 	return err;
3062 
3063 errout_free:
3064 	kfree_skb(skb);
3065 	goto errout;
3066 }
3067 
3068 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
3069 {
3070 	struct rtable *rt;
3071 	int h, s_h;
3072 	int idx, s_idx;
3073 	struct net *net;
3074 
3075 	net = sock_net(skb->sk);
3076 
3077 	s_h = cb->args[0];
3078 	if (s_h < 0)
3079 		s_h = 0;
3080 	s_idx = idx = cb->args[1];
3081 	for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3082 		if (!rt_hash_table[h].chain)
3083 			continue;
3084 		rcu_read_lock_bh();
3085 		for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3086 		     rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3087 			if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3088 				continue;
3089 			if (rt_is_expired(rt))
3090 				continue;
3091 			skb_dst_set_noref(skb, &rt->dst);
3092 			if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3093 					 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3094 					 1, NLM_F_MULTI) <= 0) {
3095 				skb_dst_drop(skb);
3096 				rcu_read_unlock_bh();
3097 				goto done;
3098 			}
3099 			skb_dst_drop(skb);
3100 		}
3101 		rcu_read_unlock_bh();
3102 	}
3103 
3104 done:
3105 	cb->args[0] = h;
3106 	cb->args[1] = idx;
3107 	return skb->len;
3108 }
3109 
3110 void ip_rt_multicast_event(struct in_device *in_dev)
3111 {
3112 	rt_cache_flush(dev_net(in_dev->dev), 0);
3113 }
3114 
3115 #ifdef CONFIG_SYSCTL
3116 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3117 					void __user *buffer,
3118 					size_t *lenp, loff_t *ppos)
3119 {
3120 	if (write) {
3121 		int flush_delay;
3122 		ctl_table ctl;
3123 		struct net *net;
3124 
3125 		memcpy(&ctl, __ctl, sizeof(ctl));
3126 		ctl.data = &flush_delay;
3127 		proc_dointvec(&ctl, write, buffer, lenp, ppos);
3128 
3129 		net = (struct net *)__ctl->extra1;
3130 		rt_cache_flush(net, flush_delay);
3131 		return 0;
3132 	}
3133 
3134 	return -EINVAL;
3135 }
3136 
3137 static ctl_table ipv4_route_table[] = {
3138 	{
3139 		.procname	= "gc_thresh",
3140 		.data		= &ipv4_dst_ops.gc_thresh,
3141 		.maxlen		= sizeof(int),
3142 		.mode		= 0644,
3143 		.proc_handler	= proc_dointvec,
3144 	},
3145 	{
3146 		.procname	= "max_size",
3147 		.data		= &ip_rt_max_size,
3148 		.maxlen		= sizeof(int),
3149 		.mode		= 0644,
3150 		.proc_handler	= proc_dointvec,
3151 	},
3152 	{
3153 		/*  Deprecated. Use gc_min_interval_ms */
3154 
3155 		.procname	= "gc_min_interval",
3156 		.data		= &ip_rt_gc_min_interval,
3157 		.maxlen		= sizeof(int),
3158 		.mode		= 0644,
3159 		.proc_handler	= proc_dointvec_jiffies,
3160 	},
3161 	{
3162 		.procname	= "gc_min_interval_ms",
3163 		.data		= &ip_rt_gc_min_interval,
3164 		.maxlen		= sizeof(int),
3165 		.mode		= 0644,
3166 		.proc_handler	= proc_dointvec_ms_jiffies,
3167 	},
3168 	{
3169 		.procname	= "gc_timeout",
3170 		.data		= &ip_rt_gc_timeout,
3171 		.maxlen		= sizeof(int),
3172 		.mode		= 0644,
3173 		.proc_handler	= proc_dointvec_jiffies,
3174 	},
3175 	{
3176 		.procname	= "redirect_load",
3177 		.data		= &ip_rt_redirect_load,
3178 		.maxlen		= sizeof(int),
3179 		.mode		= 0644,
3180 		.proc_handler	= proc_dointvec,
3181 	},
3182 	{
3183 		.procname	= "redirect_number",
3184 		.data		= &ip_rt_redirect_number,
3185 		.maxlen		= sizeof(int),
3186 		.mode		= 0644,
3187 		.proc_handler	= proc_dointvec,
3188 	},
3189 	{
3190 		.procname	= "redirect_silence",
3191 		.data		= &ip_rt_redirect_silence,
3192 		.maxlen		= sizeof(int),
3193 		.mode		= 0644,
3194 		.proc_handler	= proc_dointvec,
3195 	},
3196 	{
3197 		.procname	= "error_cost",
3198 		.data		= &ip_rt_error_cost,
3199 		.maxlen		= sizeof(int),
3200 		.mode		= 0644,
3201 		.proc_handler	= proc_dointvec,
3202 	},
3203 	{
3204 		.procname	= "error_burst",
3205 		.data		= &ip_rt_error_burst,
3206 		.maxlen		= sizeof(int),
3207 		.mode		= 0644,
3208 		.proc_handler	= proc_dointvec,
3209 	},
3210 	{
3211 		.procname	= "gc_elasticity",
3212 		.data		= &ip_rt_gc_elasticity,
3213 		.maxlen		= sizeof(int),
3214 		.mode		= 0644,
3215 		.proc_handler	= proc_dointvec,
3216 	},
3217 	{
3218 		.procname	= "mtu_expires",
3219 		.data		= &ip_rt_mtu_expires,
3220 		.maxlen		= sizeof(int),
3221 		.mode		= 0644,
3222 		.proc_handler	= proc_dointvec_jiffies,
3223 	},
3224 	{
3225 		.procname	= "min_pmtu",
3226 		.data		= &ip_rt_min_pmtu,
3227 		.maxlen		= sizeof(int),
3228 		.mode		= 0644,
3229 		.proc_handler	= proc_dointvec,
3230 	},
3231 	{
3232 		.procname	= "min_adv_mss",
3233 		.data		= &ip_rt_min_advmss,
3234 		.maxlen		= sizeof(int),
3235 		.mode		= 0644,
3236 		.proc_handler	= proc_dointvec,
3237 	},
3238 	{ }
3239 };
3240 
3241 static struct ctl_table empty[1];
3242 
3243 static struct ctl_table ipv4_skeleton[] =
3244 {
3245 	{ .procname = "route",
3246 	  .mode = 0555, .child = ipv4_route_table},
3247 	{ .procname = "neigh",
3248 	  .mode = 0555, .child = empty},
3249 	{ }
3250 };
3251 
3252 static __net_initdata struct ctl_path ipv4_path[] = {
3253 	{ .procname = "net", },
3254 	{ .procname = "ipv4", },
3255 	{ },
3256 };
3257 
3258 static struct ctl_table ipv4_route_flush_table[] = {
3259 	{
3260 		.procname	= "flush",
3261 		.maxlen		= sizeof(int),
3262 		.mode		= 0200,
3263 		.proc_handler	= ipv4_sysctl_rtcache_flush,
3264 	},
3265 	{ },
3266 };
3267 
3268 static __net_initdata struct ctl_path ipv4_route_path[] = {
3269 	{ .procname = "net", },
3270 	{ .procname = "ipv4", },
3271 	{ .procname = "route", },
3272 	{ },
3273 };
3274 
3275 static __net_init int sysctl_route_net_init(struct net *net)
3276 {
3277 	struct ctl_table *tbl;
3278 
3279 	tbl = ipv4_route_flush_table;
3280 	if (!net_eq(net, &init_net)) {
3281 		tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3282 		if (tbl == NULL)
3283 			goto err_dup;
3284 	}
3285 	tbl[0].extra1 = net;
3286 
3287 	net->ipv4.route_hdr =
3288 		register_net_sysctl_table(net, ipv4_route_path, tbl);
3289 	if (net->ipv4.route_hdr == NULL)
3290 		goto err_reg;
3291 	return 0;
3292 
3293 err_reg:
3294 	if (tbl != ipv4_route_flush_table)
3295 		kfree(tbl);
3296 err_dup:
3297 	return -ENOMEM;
3298 }
3299 
3300 static __net_exit void sysctl_route_net_exit(struct net *net)
3301 {
3302 	struct ctl_table *tbl;
3303 
3304 	tbl = net->ipv4.route_hdr->ctl_table_arg;
3305 	unregister_net_sysctl_table(net->ipv4.route_hdr);
3306 	BUG_ON(tbl == ipv4_route_flush_table);
3307 	kfree(tbl);
3308 }
3309 
3310 static __net_initdata struct pernet_operations sysctl_route_ops = {
3311 	.init = sysctl_route_net_init,
3312 	.exit = sysctl_route_net_exit,
3313 };
3314 #endif
3315 
3316 static __net_init int rt_genid_init(struct net *net)
3317 {
3318 	get_random_bytes(&net->ipv4.rt_genid,
3319 			 sizeof(net->ipv4.rt_genid));
3320 	get_random_bytes(&net->ipv4.dev_addr_genid,
3321 			 sizeof(net->ipv4.dev_addr_genid));
3322 	return 0;
3323 }
3324 
3325 static __net_initdata struct pernet_operations rt_genid_ops = {
3326 	.init = rt_genid_init,
3327 };
3328 
3329 
3330 #ifdef CONFIG_IP_ROUTE_CLASSID
3331 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3332 #endif /* CONFIG_IP_ROUTE_CLASSID */
3333 
3334 static __initdata unsigned long rhash_entries;
3335 static int __init set_rhash_entries(char *str)
3336 {
3337 	if (!str)
3338 		return 0;
3339 	rhash_entries = simple_strtoul(str, &str, 0);
3340 	return 1;
3341 }
3342 __setup("rhash_entries=", set_rhash_entries);
3343 
3344 int __init ip_rt_init(void)
3345 {
3346 	int rc = 0;
3347 
3348 #ifdef CONFIG_IP_ROUTE_CLASSID
3349 	ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3350 	if (!ip_rt_acct)
3351 		panic("IP: failed to allocate ip_rt_acct\n");
3352 #endif
3353 
3354 	ipv4_dst_ops.kmem_cachep =
3355 		kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3356 				  SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3357 
3358 	ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3359 
3360 	if (dst_entries_init(&ipv4_dst_ops) < 0)
3361 		panic("IP: failed to allocate ipv4_dst_ops counter\n");
3362 
3363 	if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3364 		panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3365 
3366 	rt_hash_table = (struct rt_hash_bucket *)
3367 		alloc_large_system_hash("IP route cache",
3368 					sizeof(struct rt_hash_bucket),
3369 					rhash_entries,
3370 					(totalram_pages >= 128 * 1024) ?
3371 					15 : 17,
3372 					0,
3373 					&rt_hash_log,
3374 					&rt_hash_mask,
3375 					rhash_entries ? 0 : 512 * 1024);
3376 	memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3377 	rt_hash_lock_init();
3378 
3379 	ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3380 	ip_rt_max_size = (rt_hash_mask + 1) * 16;
3381 
3382 	devinet_init();
3383 	ip_fib_init();
3384 
3385 	if (ip_rt_proc_init())
3386 		printk(KERN_ERR "Unable to create route proc files\n");
3387 #ifdef CONFIG_XFRM
3388 	xfrm_init();
3389 	xfrm4_init(ip_rt_max_size);
3390 #endif
3391 	rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
3392 
3393 #ifdef CONFIG_SYSCTL
3394 	register_pernet_subsys(&sysctl_route_ops);
3395 #endif
3396 	register_pernet_subsys(&rt_genid_ops);
3397 	return rc;
3398 }
3399 
3400 #ifdef CONFIG_SYSCTL
3401 /*
3402  * We really need to sanitize the damn ipv4 init order, then all
3403  * this nonsense will go away.
3404  */
3405 void __init ip_static_sysctl_init(void)
3406 {
3407 	register_sysctl_paths(ipv4_path, ipv4_skeleton);
3408 }
3409 #endif
3410