xref: /linux/net/ipv4/route.c (revision 424f0750edd5af866f80f5e65998e0610503cb5c)
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/atmclip.h>
112 #include <net/secure_seq.h>
113 
114 #define RT_FL_TOS(oldflp4) \
115     ((u32)(oldflp4->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK)))
116 
117 #define IP_MAX_MTU	0xFFF0
118 
119 #define RT_GC_TIMEOUT (300*HZ)
120 
121 static int ip_rt_max_size;
122 static int ip_rt_gc_timeout __read_mostly	= RT_GC_TIMEOUT;
123 static int ip_rt_gc_min_interval __read_mostly	= HZ / 2;
124 static int ip_rt_redirect_number __read_mostly	= 9;
125 static int ip_rt_redirect_load __read_mostly	= HZ / 50;
126 static int ip_rt_redirect_silence __read_mostly	= ((HZ / 50) << (9 + 1));
127 static int ip_rt_error_cost __read_mostly	= HZ;
128 static int ip_rt_error_burst __read_mostly	= 5 * HZ;
129 static int ip_rt_gc_elasticity __read_mostly	= 8;
130 static int ip_rt_mtu_expires __read_mostly	= 10 * 60 * HZ;
131 static int ip_rt_min_pmtu __read_mostly		= 512 + 20 + 20;
132 static int ip_rt_min_advmss __read_mostly	= 256;
133 static int rt_chain_length_max __read_mostly	= 20;
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_default_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 	.default_mtu =		ipv4_default_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;
420 
421 		n = dst_get_neighbour(&r->dst);
422 		seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
423 			      "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
424 			r->dst.dev ? r->dst.dev->name : "*",
425 			(__force u32)r->rt_dst,
426 			(__force u32)r->rt_gateway,
427 			r->rt_flags, atomic_read(&r->dst.__refcnt),
428 			r->dst.__use, 0, (__force u32)r->rt_src,
429 			dst_metric_advmss(&r->dst) + 40,
430 			dst_metric(&r->dst, RTAX_WINDOW),
431 			(int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
432 			      dst_metric(&r->dst, RTAX_RTTVAR)),
433 			r->rt_key_tos,
434 			-1,
435 			(n && (n->nud_state & NUD_CONNECTED)) ? 1 : 0,
436 			r->rt_spec_dst, &len);
437 
438 		seq_printf(seq, "%*s\n", 127 - len, "");
439 	}
440 	return 0;
441 }
442 
443 static const struct seq_operations rt_cache_seq_ops = {
444 	.start  = rt_cache_seq_start,
445 	.next   = rt_cache_seq_next,
446 	.stop   = rt_cache_seq_stop,
447 	.show   = rt_cache_seq_show,
448 };
449 
450 static int rt_cache_seq_open(struct inode *inode, struct file *file)
451 {
452 	return seq_open_net(inode, file, &rt_cache_seq_ops,
453 			sizeof(struct rt_cache_iter_state));
454 }
455 
456 static const struct file_operations rt_cache_seq_fops = {
457 	.owner	 = THIS_MODULE,
458 	.open	 = rt_cache_seq_open,
459 	.read	 = seq_read,
460 	.llseek	 = seq_lseek,
461 	.release = seq_release_net,
462 };
463 
464 
465 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
466 {
467 	int cpu;
468 
469 	if (*pos == 0)
470 		return SEQ_START_TOKEN;
471 
472 	for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
473 		if (!cpu_possible(cpu))
474 			continue;
475 		*pos = cpu+1;
476 		return &per_cpu(rt_cache_stat, cpu);
477 	}
478 	return NULL;
479 }
480 
481 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
482 {
483 	int cpu;
484 
485 	for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
486 		if (!cpu_possible(cpu))
487 			continue;
488 		*pos = cpu+1;
489 		return &per_cpu(rt_cache_stat, cpu);
490 	}
491 	return NULL;
492 
493 }
494 
495 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
496 {
497 
498 }
499 
500 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
501 {
502 	struct rt_cache_stat *st = v;
503 
504 	if (v == SEQ_START_TOKEN) {
505 		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");
506 		return 0;
507 	}
508 
509 	seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
510 		   " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
511 		   dst_entries_get_slow(&ipv4_dst_ops),
512 		   st->in_hit,
513 		   st->in_slow_tot,
514 		   st->in_slow_mc,
515 		   st->in_no_route,
516 		   st->in_brd,
517 		   st->in_martian_dst,
518 		   st->in_martian_src,
519 
520 		   st->out_hit,
521 		   st->out_slow_tot,
522 		   st->out_slow_mc,
523 
524 		   st->gc_total,
525 		   st->gc_ignored,
526 		   st->gc_goal_miss,
527 		   st->gc_dst_overflow,
528 		   st->in_hlist_search,
529 		   st->out_hlist_search
530 		);
531 	return 0;
532 }
533 
534 static const struct seq_operations rt_cpu_seq_ops = {
535 	.start  = rt_cpu_seq_start,
536 	.next   = rt_cpu_seq_next,
537 	.stop   = rt_cpu_seq_stop,
538 	.show   = rt_cpu_seq_show,
539 };
540 
541 
542 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
543 {
544 	return seq_open(file, &rt_cpu_seq_ops);
545 }
546 
547 static const struct file_operations rt_cpu_seq_fops = {
548 	.owner	 = THIS_MODULE,
549 	.open	 = rt_cpu_seq_open,
550 	.read	 = seq_read,
551 	.llseek	 = seq_lseek,
552 	.release = seq_release,
553 };
554 
555 #ifdef CONFIG_IP_ROUTE_CLASSID
556 static int rt_acct_proc_show(struct seq_file *m, void *v)
557 {
558 	struct ip_rt_acct *dst, *src;
559 	unsigned int i, j;
560 
561 	dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
562 	if (!dst)
563 		return -ENOMEM;
564 
565 	for_each_possible_cpu(i) {
566 		src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
567 		for (j = 0; j < 256; j++) {
568 			dst[j].o_bytes   += src[j].o_bytes;
569 			dst[j].o_packets += src[j].o_packets;
570 			dst[j].i_bytes   += src[j].i_bytes;
571 			dst[j].i_packets += src[j].i_packets;
572 		}
573 	}
574 
575 	seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
576 	kfree(dst);
577 	return 0;
578 }
579 
580 static int rt_acct_proc_open(struct inode *inode, struct file *file)
581 {
582 	return single_open(file, rt_acct_proc_show, NULL);
583 }
584 
585 static const struct file_operations rt_acct_proc_fops = {
586 	.owner		= THIS_MODULE,
587 	.open		= rt_acct_proc_open,
588 	.read		= seq_read,
589 	.llseek		= seq_lseek,
590 	.release	= single_release,
591 };
592 #endif
593 
594 static int __net_init ip_rt_do_proc_init(struct net *net)
595 {
596 	struct proc_dir_entry *pde;
597 
598 	pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
599 			&rt_cache_seq_fops);
600 	if (!pde)
601 		goto err1;
602 
603 	pde = proc_create("rt_cache", S_IRUGO,
604 			  net->proc_net_stat, &rt_cpu_seq_fops);
605 	if (!pde)
606 		goto err2;
607 
608 #ifdef CONFIG_IP_ROUTE_CLASSID
609 	pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
610 	if (!pde)
611 		goto err3;
612 #endif
613 	return 0;
614 
615 #ifdef CONFIG_IP_ROUTE_CLASSID
616 err3:
617 	remove_proc_entry("rt_cache", net->proc_net_stat);
618 #endif
619 err2:
620 	remove_proc_entry("rt_cache", net->proc_net);
621 err1:
622 	return -ENOMEM;
623 }
624 
625 static void __net_exit ip_rt_do_proc_exit(struct net *net)
626 {
627 	remove_proc_entry("rt_cache", net->proc_net_stat);
628 	remove_proc_entry("rt_cache", net->proc_net);
629 #ifdef CONFIG_IP_ROUTE_CLASSID
630 	remove_proc_entry("rt_acct", net->proc_net);
631 #endif
632 }
633 
634 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
635 	.init = ip_rt_do_proc_init,
636 	.exit = ip_rt_do_proc_exit,
637 };
638 
639 static int __init ip_rt_proc_init(void)
640 {
641 	return register_pernet_subsys(&ip_rt_proc_ops);
642 }
643 
644 #else
645 static inline int ip_rt_proc_init(void)
646 {
647 	return 0;
648 }
649 #endif /* CONFIG_PROC_FS */
650 
651 static inline void rt_free(struct rtable *rt)
652 {
653 	call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
654 }
655 
656 static inline void rt_drop(struct rtable *rt)
657 {
658 	ip_rt_put(rt);
659 	call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
660 }
661 
662 static inline int rt_fast_clean(struct rtable *rth)
663 {
664 	/* Kill broadcast/multicast entries very aggresively, if they
665 	   collide in hash table with more useful entries */
666 	return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
667 		rt_is_input_route(rth) && rth->dst.rt_next;
668 }
669 
670 static inline int rt_valuable(struct rtable *rth)
671 {
672 	return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
673 		(rth->peer && rth->peer->pmtu_expires);
674 }
675 
676 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
677 {
678 	unsigned long age;
679 	int ret = 0;
680 
681 	if (atomic_read(&rth->dst.__refcnt))
682 		goto out;
683 
684 	age = jiffies - rth->dst.lastuse;
685 	if ((age <= tmo1 && !rt_fast_clean(rth)) ||
686 	    (age <= tmo2 && rt_valuable(rth)))
687 		goto out;
688 	ret = 1;
689 out:	return ret;
690 }
691 
692 /* Bits of score are:
693  * 31: very valuable
694  * 30: not quite useless
695  * 29..0: usage counter
696  */
697 static inline u32 rt_score(struct rtable *rt)
698 {
699 	u32 score = jiffies - rt->dst.lastuse;
700 
701 	score = ~score & ~(3<<30);
702 
703 	if (rt_valuable(rt))
704 		score |= (1<<31);
705 
706 	if (rt_is_output_route(rt) ||
707 	    !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
708 		score |= (1<<30);
709 
710 	return score;
711 }
712 
713 static inline bool rt_caching(const struct net *net)
714 {
715 	return net->ipv4.current_rt_cache_rebuild_count <=
716 		net->ipv4.sysctl_rt_cache_rebuild_count;
717 }
718 
719 static inline bool compare_hash_inputs(const struct rtable *rt1,
720 				       const struct rtable *rt2)
721 {
722 	return ((((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
723 		((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
724 		(rt1->rt_route_iif ^ rt2->rt_route_iif)) == 0);
725 }
726 
727 static inline int compare_keys(struct rtable *rt1, struct rtable *rt2)
728 {
729 	return (((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
730 		((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
731 		(rt1->rt_mark ^ rt2->rt_mark) |
732 		(rt1->rt_key_tos ^ rt2->rt_key_tos) |
733 		(rt1->rt_route_iif ^ rt2->rt_route_iif) |
734 		(rt1->rt_oif ^ rt2->rt_oif)) == 0;
735 }
736 
737 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
738 {
739 	return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
740 }
741 
742 static inline int rt_is_expired(struct rtable *rth)
743 {
744 	return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
745 }
746 
747 /*
748  * Perform a full scan of hash table and free all entries.
749  * Can be called by a softirq or a process.
750  * In the later case, we want to be reschedule if necessary
751  */
752 static void rt_do_flush(struct net *net, int process_context)
753 {
754 	unsigned int i;
755 	struct rtable *rth, *next;
756 
757 	for (i = 0; i <= rt_hash_mask; i++) {
758 		struct rtable __rcu **pprev;
759 		struct rtable *list;
760 
761 		if (process_context && need_resched())
762 			cond_resched();
763 		rth = rcu_access_pointer(rt_hash_table[i].chain);
764 		if (!rth)
765 			continue;
766 
767 		spin_lock_bh(rt_hash_lock_addr(i));
768 
769 		list = NULL;
770 		pprev = &rt_hash_table[i].chain;
771 		rth = rcu_dereference_protected(*pprev,
772 			lockdep_is_held(rt_hash_lock_addr(i)));
773 
774 		while (rth) {
775 			next = rcu_dereference_protected(rth->dst.rt_next,
776 				lockdep_is_held(rt_hash_lock_addr(i)));
777 
778 			if (!net ||
779 			    net_eq(dev_net(rth->dst.dev), net)) {
780 				rcu_assign_pointer(*pprev, next);
781 				rcu_assign_pointer(rth->dst.rt_next, list);
782 				list = rth;
783 			} else {
784 				pprev = &rth->dst.rt_next;
785 			}
786 			rth = next;
787 		}
788 
789 		spin_unlock_bh(rt_hash_lock_addr(i));
790 
791 		for (; list; list = next) {
792 			next = rcu_dereference_protected(list->dst.rt_next, 1);
793 			rt_free(list);
794 		}
795 	}
796 }
797 
798 /*
799  * While freeing expired entries, we compute average chain length
800  * and standard deviation, using fixed-point arithmetic.
801  * This to have an estimation of rt_chain_length_max
802  *  rt_chain_length_max = max(elasticity, AVG + 4*SD)
803  * We use 3 bits for frational part, and 29 (or 61) for magnitude.
804  */
805 
806 #define FRACT_BITS 3
807 #define ONE (1UL << FRACT_BITS)
808 
809 /*
810  * Given a hash chain and an item in this hash chain,
811  * find if a previous entry has the same hash_inputs
812  * (but differs on tos, mark or oif)
813  * Returns 0 if an alias is found.
814  * Returns ONE if rth has no alias before itself.
815  */
816 static int has_noalias(const struct rtable *head, const struct rtable *rth)
817 {
818 	const struct rtable *aux = head;
819 
820 	while (aux != rth) {
821 		if (compare_hash_inputs(aux, rth))
822 			return 0;
823 		aux = rcu_dereference_protected(aux->dst.rt_next, 1);
824 	}
825 	return ONE;
826 }
827 
828 /*
829  * Perturbation of rt_genid by a small quantity [1..256]
830  * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
831  * many times (2^24) without giving recent rt_genid.
832  * Jenkins hash is strong enough that litle changes of rt_genid are OK.
833  */
834 static void rt_cache_invalidate(struct net *net)
835 {
836 	unsigned char shuffle;
837 
838 	get_random_bytes(&shuffle, sizeof(shuffle));
839 	atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
840 }
841 
842 /*
843  * delay < 0  : invalidate cache (fast : entries will be deleted later)
844  * delay >= 0 : invalidate & flush cache (can be long)
845  */
846 void rt_cache_flush(struct net *net, int delay)
847 {
848 	rt_cache_invalidate(net);
849 	if (delay >= 0)
850 		rt_do_flush(net, !in_softirq());
851 }
852 
853 /* Flush previous cache invalidated entries from the cache */
854 void rt_cache_flush_batch(struct net *net)
855 {
856 	rt_do_flush(net, !in_softirq());
857 }
858 
859 static void rt_emergency_hash_rebuild(struct net *net)
860 {
861 	if (net_ratelimit())
862 		printk(KERN_WARNING "Route hash chain too long!\n");
863 	rt_cache_invalidate(net);
864 }
865 
866 /*
867    Short description of GC goals.
868 
869    We want to build algorithm, which will keep routing cache
870    at some equilibrium point, when number of aged off entries
871    is kept approximately equal to newly generated ones.
872 
873    Current expiration strength is variable "expire".
874    We try to adjust it dynamically, so that if networking
875    is idle expires is large enough to keep enough of warm entries,
876    and when load increases it reduces to limit cache size.
877  */
878 
879 static int rt_garbage_collect(struct dst_ops *ops)
880 {
881 	static unsigned long expire = RT_GC_TIMEOUT;
882 	static unsigned long last_gc;
883 	static int rover;
884 	static int equilibrium;
885 	struct rtable *rth;
886 	struct rtable __rcu **rthp;
887 	unsigned long now = jiffies;
888 	int goal;
889 	int entries = dst_entries_get_fast(&ipv4_dst_ops);
890 
891 	/*
892 	 * Garbage collection is pretty expensive,
893 	 * do not make it too frequently.
894 	 */
895 
896 	RT_CACHE_STAT_INC(gc_total);
897 
898 	if (now - last_gc < ip_rt_gc_min_interval &&
899 	    entries < ip_rt_max_size) {
900 		RT_CACHE_STAT_INC(gc_ignored);
901 		goto out;
902 	}
903 
904 	entries = dst_entries_get_slow(&ipv4_dst_ops);
905 	/* Calculate number of entries, which we want to expire now. */
906 	goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
907 	if (goal <= 0) {
908 		if (equilibrium < ipv4_dst_ops.gc_thresh)
909 			equilibrium = ipv4_dst_ops.gc_thresh;
910 		goal = entries - equilibrium;
911 		if (goal > 0) {
912 			equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
913 			goal = entries - equilibrium;
914 		}
915 	} else {
916 		/* We are in dangerous area. Try to reduce cache really
917 		 * aggressively.
918 		 */
919 		goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
920 		equilibrium = entries - goal;
921 	}
922 
923 	if (now - last_gc >= ip_rt_gc_min_interval)
924 		last_gc = now;
925 
926 	if (goal <= 0) {
927 		equilibrium += goal;
928 		goto work_done;
929 	}
930 
931 	do {
932 		int i, k;
933 
934 		for (i = rt_hash_mask, k = rover; i >= 0; i--) {
935 			unsigned long tmo = expire;
936 
937 			k = (k + 1) & rt_hash_mask;
938 			rthp = &rt_hash_table[k].chain;
939 			spin_lock_bh(rt_hash_lock_addr(k));
940 			while ((rth = rcu_dereference_protected(*rthp,
941 					lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
942 				if (!rt_is_expired(rth) &&
943 					!rt_may_expire(rth, tmo, expire)) {
944 					tmo >>= 1;
945 					rthp = &rth->dst.rt_next;
946 					continue;
947 				}
948 				*rthp = rth->dst.rt_next;
949 				rt_free(rth);
950 				goal--;
951 			}
952 			spin_unlock_bh(rt_hash_lock_addr(k));
953 			if (goal <= 0)
954 				break;
955 		}
956 		rover = k;
957 
958 		if (goal <= 0)
959 			goto work_done;
960 
961 		/* Goal is not achieved. We stop process if:
962 
963 		   - if expire reduced to zero. Otherwise, expire is halfed.
964 		   - if table is not full.
965 		   - if we are called from interrupt.
966 		   - jiffies check is just fallback/debug loop breaker.
967 		     We will not spin here for long time in any case.
968 		 */
969 
970 		RT_CACHE_STAT_INC(gc_goal_miss);
971 
972 		if (expire == 0)
973 			break;
974 
975 		expire >>= 1;
976 
977 		if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
978 			goto out;
979 	} while (!in_softirq() && time_before_eq(jiffies, now));
980 
981 	if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
982 		goto out;
983 	if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
984 		goto out;
985 	if (net_ratelimit())
986 		printk(KERN_WARNING "dst cache overflow\n");
987 	RT_CACHE_STAT_INC(gc_dst_overflow);
988 	return 1;
989 
990 work_done:
991 	expire += ip_rt_gc_min_interval;
992 	if (expire > ip_rt_gc_timeout ||
993 	    dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
994 	    dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
995 		expire = ip_rt_gc_timeout;
996 out:	return 0;
997 }
998 
999 /*
1000  * Returns number of entries in a hash chain that have different hash_inputs
1001  */
1002 static int slow_chain_length(const struct rtable *head)
1003 {
1004 	int length = 0;
1005 	const struct rtable *rth = head;
1006 
1007 	while (rth) {
1008 		length += has_noalias(head, rth);
1009 		rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1010 	}
1011 	return length >> FRACT_BITS;
1012 }
1013 
1014 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, const void *daddr)
1015 {
1016 	struct neigh_table *tbl = &arp_tbl;
1017 	static const __be32 inaddr_any = 0;
1018 	struct net_device *dev = dst->dev;
1019 	const __be32 *pkey = daddr;
1020 	struct neighbour *n;
1021 
1022 #if defined(CONFIG_ATM_CLIP) || defined(CONFIG_ATM_CLIP_MODULE)
1023 	if (dev->type == ARPHRD_ATM)
1024 		tbl = clip_tbl_hook;
1025 #endif
1026 	if (dev->flags & (IFF_LOOPBACK | IFF_POINTOPOINT))
1027 		pkey = &inaddr_any;
1028 
1029 	n = __ipv4_neigh_lookup(tbl, dev, *(__force u32 *)pkey);
1030 	if (n)
1031 		return n;
1032 	return neigh_create(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 /* called in rcu_read_lock() section */
1308 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1309 		    __be32 saddr, struct net_device *dev)
1310 {
1311 	int s, i;
1312 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1313 	struct rtable *rt;
1314 	__be32 skeys[2] = { saddr, 0 };
1315 	int    ikeys[2] = { dev->ifindex, 0 };
1316 	struct flowi4 fl4;
1317 	struct inet_peer *peer;
1318 	struct net *net;
1319 
1320 	if (!in_dev)
1321 		return;
1322 
1323 	net = dev_net(dev);
1324 	if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1325 	    ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1326 	    ipv4_is_zeronet(new_gw))
1327 		goto reject_redirect;
1328 
1329 	if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1330 		if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1331 			goto reject_redirect;
1332 		if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1333 			goto reject_redirect;
1334 	} else {
1335 		if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1336 			goto reject_redirect;
1337 	}
1338 
1339 	memset(&fl4, 0, sizeof(fl4));
1340 	fl4.daddr = daddr;
1341 	for (s = 0; s < 2; s++) {
1342 		for (i = 0; i < 2; i++) {
1343 			fl4.flowi4_oif = ikeys[i];
1344 			fl4.saddr = skeys[s];
1345 			rt = __ip_route_output_key(net, &fl4);
1346 			if (IS_ERR(rt))
1347 				continue;
1348 
1349 			if (rt->dst.error || rt->dst.dev != dev ||
1350 			    rt->rt_gateway != old_gw) {
1351 				ip_rt_put(rt);
1352 				continue;
1353 			}
1354 
1355 			if (!rt->peer)
1356 				rt_bind_peer(rt, rt->rt_dst, 1);
1357 
1358 			peer = rt->peer;
1359 			if (peer) {
1360 				peer->redirect_learned.a4 = new_gw;
1361 				atomic_inc(&__rt_peer_genid);
1362 			}
1363 
1364 			ip_rt_put(rt);
1365 			return;
1366 		}
1367 	}
1368 	return;
1369 
1370 reject_redirect:
1371 #ifdef CONFIG_IP_ROUTE_VERBOSE
1372 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1373 		printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1374 			"  Advised path = %pI4 -> %pI4\n",
1375 		       &old_gw, dev->name, &new_gw,
1376 		       &saddr, &daddr);
1377 #endif
1378 	;
1379 }
1380 
1381 static bool peer_pmtu_expired(struct inet_peer *peer)
1382 {
1383 	unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1384 
1385 	return orig &&
1386 	       time_after_eq(jiffies, orig) &&
1387 	       cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1388 }
1389 
1390 static bool peer_pmtu_cleaned(struct inet_peer *peer)
1391 {
1392 	unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1393 
1394 	return orig &&
1395 	       cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1396 }
1397 
1398 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1399 {
1400 	struct rtable *rt = (struct rtable *)dst;
1401 	struct dst_entry *ret = dst;
1402 
1403 	if (rt) {
1404 		if (dst->obsolete > 0) {
1405 			ip_rt_put(rt);
1406 			ret = NULL;
1407 		} else if (rt->rt_flags & RTCF_REDIRECTED) {
1408 			unsigned hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1409 						rt->rt_oif,
1410 						rt_genid(dev_net(dst->dev)));
1411 			rt_del(hash, rt);
1412 			ret = NULL;
1413 		} else if (rt->peer && peer_pmtu_expired(rt->peer)) {
1414 			dst_metric_set(dst, RTAX_MTU, rt->peer->pmtu_orig);
1415 		}
1416 	}
1417 	return ret;
1418 }
1419 
1420 /*
1421  * Algorithm:
1422  *	1. The first ip_rt_redirect_number redirects are sent
1423  *	   with exponential backoff, then we stop sending them at all,
1424  *	   assuming that the host ignores our redirects.
1425  *	2. If we did not see packets requiring redirects
1426  *	   during ip_rt_redirect_silence, we assume that the host
1427  *	   forgot redirected route and start to send redirects again.
1428  *
1429  * This algorithm is much cheaper and more intelligent than dumb load limiting
1430  * in icmp.c.
1431  *
1432  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1433  * and "frag. need" (breaks PMTU discovery) in icmp.c.
1434  */
1435 
1436 void ip_rt_send_redirect(struct sk_buff *skb)
1437 {
1438 	struct rtable *rt = skb_rtable(skb);
1439 	struct in_device *in_dev;
1440 	struct inet_peer *peer;
1441 	int log_martians;
1442 
1443 	rcu_read_lock();
1444 	in_dev = __in_dev_get_rcu(rt->dst.dev);
1445 	if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1446 		rcu_read_unlock();
1447 		return;
1448 	}
1449 	log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1450 	rcu_read_unlock();
1451 
1452 	if (!rt->peer)
1453 		rt_bind_peer(rt, rt->rt_dst, 1);
1454 	peer = rt->peer;
1455 	if (!peer) {
1456 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1457 		return;
1458 	}
1459 
1460 	/* No redirected packets during ip_rt_redirect_silence;
1461 	 * reset the algorithm.
1462 	 */
1463 	if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1464 		peer->rate_tokens = 0;
1465 
1466 	/* Too many ignored redirects; do not send anything
1467 	 * set dst.rate_last to the last seen redirected packet.
1468 	 */
1469 	if (peer->rate_tokens >= ip_rt_redirect_number) {
1470 		peer->rate_last = jiffies;
1471 		return;
1472 	}
1473 
1474 	/* Check for load limit; set rate_last to the latest sent
1475 	 * redirect.
1476 	 */
1477 	if (peer->rate_tokens == 0 ||
1478 	    time_after(jiffies,
1479 		       (peer->rate_last +
1480 			(ip_rt_redirect_load << peer->rate_tokens)))) {
1481 		icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1482 		peer->rate_last = jiffies;
1483 		++peer->rate_tokens;
1484 #ifdef CONFIG_IP_ROUTE_VERBOSE
1485 		if (log_martians &&
1486 		    peer->rate_tokens == ip_rt_redirect_number &&
1487 		    net_ratelimit())
1488 			printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1489 			       &ip_hdr(skb)->saddr, rt->rt_iif,
1490 				&rt->rt_dst, &rt->rt_gateway);
1491 #endif
1492 	}
1493 }
1494 
1495 static int ip_error(struct sk_buff *skb)
1496 {
1497 	struct rtable *rt = skb_rtable(skb);
1498 	struct inet_peer *peer;
1499 	unsigned long now;
1500 	bool send;
1501 	int code;
1502 
1503 	switch (rt->dst.error) {
1504 	case EINVAL:
1505 	default:
1506 		goto out;
1507 	case EHOSTUNREACH:
1508 		code = ICMP_HOST_UNREACH;
1509 		break;
1510 	case ENETUNREACH:
1511 		code = ICMP_NET_UNREACH;
1512 		IP_INC_STATS_BH(dev_net(rt->dst.dev),
1513 				IPSTATS_MIB_INNOROUTES);
1514 		break;
1515 	case EACCES:
1516 		code = ICMP_PKT_FILTERED;
1517 		break;
1518 	}
1519 
1520 	if (!rt->peer)
1521 		rt_bind_peer(rt, rt->rt_dst, 1);
1522 	peer = rt->peer;
1523 
1524 	send = true;
1525 	if (peer) {
1526 		now = jiffies;
1527 		peer->rate_tokens += now - peer->rate_last;
1528 		if (peer->rate_tokens > ip_rt_error_burst)
1529 			peer->rate_tokens = ip_rt_error_burst;
1530 		peer->rate_last = now;
1531 		if (peer->rate_tokens >= ip_rt_error_cost)
1532 			peer->rate_tokens -= ip_rt_error_cost;
1533 		else
1534 			send = false;
1535 	}
1536 	if (send)
1537 		icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1538 
1539 out:	kfree_skb(skb);
1540 	return 0;
1541 }
1542 
1543 /*
1544  *	The last two values are not from the RFC but
1545  *	are needed for AMPRnet AX.25 paths.
1546  */
1547 
1548 static const unsigned short mtu_plateau[] =
1549 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1550 
1551 static inline unsigned short guess_mtu(unsigned short old_mtu)
1552 {
1553 	int i;
1554 
1555 	for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1556 		if (old_mtu > mtu_plateau[i])
1557 			return mtu_plateau[i];
1558 	return 68;
1559 }
1560 
1561 unsigned short ip_rt_frag_needed(struct net *net, const struct iphdr *iph,
1562 				 unsigned short new_mtu,
1563 				 struct net_device *dev)
1564 {
1565 	unsigned short old_mtu = ntohs(iph->tot_len);
1566 	unsigned short est_mtu = 0;
1567 	struct inet_peer *peer;
1568 
1569 	peer = inet_getpeer_v4(iph->daddr, 1);
1570 	if (peer) {
1571 		unsigned short mtu = new_mtu;
1572 
1573 		if (new_mtu < 68 || new_mtu >= old_mtu) {
1574 			/* BSD 4.2 derived systems incorrectly adjust
1575 			 * tot_len by the IP header length, and report
1576 			 * a zero MTU in the ICMP message.
1577 			 */
1578 			if (mtu == 0 &&
1579 			    old_mtu >= 68 + (iph->ihl << 2))
1580 				old_mtu -= iph->ihl << 2;
1581 			mtu = guess_mtu(old_mtu);
1582 		}
1583 
1584 		if (mtu < ip_rt_min_pmtu)
1585 			mtu = ip_rt_min_pmtu;
1586 		if (!peer->pmtu_expires || mtu < peer->pmtu_learned) {
1587 			unsigned long pmtu_expires;
1588 
1589 			pmtu_expires = jiffies + ip_rt_mtu_expires;
1590 			if (!pmtu_expires)
1591 				pmtu_expires = 1UL;
1592 
1593 			est_mtu = mtu;
1594 			peer->pmtu_learned = mtu;
1595 			peer->pmtu_expires = pmtu_expires;
1596 			atomic_inc(&__rt_peer_genid);
1597 		}
1598 
1599 		inet_putpeer(peer);
1600 	}
1601 	return est_mtu ? : new_mtu;
1602 }
1603 
1604 static void check_peer_pmtu(struct dst_entry *dst, struct inet_peer *peer)
1605 {
1606 	unsigned long expires = ACCESS_ONCE(peer->pmtu_expires);
1607 
1608 	if (!expires)
1609 		return;
1610 	if (time_before(jiffies, expires)) {
1611 		u32 orig_dst_mtu = dst_mtu(dst);
1612 		if (peer->pmtu_learned < orig_dst_mtu) {
1613 			if (!peer->pmtu_orig)
1614 				peer->pmtu_orig = dst_metric_raw(dst, RTAX_MTU);
1615 			dst_metric_set(dst, RTAX_MTU, peer->pmtu_learned);
1616 		}
1617 	} else if (cmpxchg(&peer->pmtu_expires, expires, 0) == expires)
1618 		dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1619 }
1620 
1621 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1622 {
1623 	struct rtable *rt = (struct rtable *) dst;
1624 	struct inet_peer *peer;
1625 
1626 	dst_confirm(dst);
1627 
1628 	if (!rt->peer)
1629 		rt_bind_peer(rt, rt->rt_dst, 1);
1630 	peer = rt->peer;
1631 	if (peer) {
1632 		unsigned long pmtu_expires = ACCESS_ONCE(peer->pmtu_expires);
1633 
1634 		if (mtu < ip_rt_min_pmtu)
1635 			mtu = ip_rt_min_pmtu;
1636 		if (!pmtu_expires || mtu < peer->pmtu_learned) {
1637 
1638 			pmtu_expires = jiffies + ip_rt_mtu_expires;
1639 			if (!pmtu_expires)
1640 				pmtu_expires = 1UL;
1641 
1642 			peer->pmtu_learned = mtu;
1643 			peer->pmtu_expires = pmtu_expires;
1644 
1645 			atomic_inc(&__rt_peer_genid);
1646 			rt->rt_peer_genid = rt_peer_genid();
1647 		}
1648 		check_peer_pmtu(dst, peer);
1649 	}
1650 }
1651 
1652 static int check_peer_redir(struct dst_entry *dst, struct inet_peer *peer)
1653 {
1654 	struct rtable *rt = (struct rtable *) dst;
1655 	__be32 orig_gw = rt->rt_gateway;
1656 	struct neighbour *n, *old_n;
1657 
1658 	dst_confirm(&rt->dst);
1659 
1660 	rt->rt_gateway = peer->redirect_learned.a4;
1661 
1662 	n = ipv4_neigh_lookup(&rt->dst, &rt->rt_gateway);
1663 	if (IS_ERR(n))
1664 		return PTR_ERR(n);
1665 	old_n = xchg(&rt->dst._neighbour, n);
1666 	if (old_n)
1667 		neigh_release(old_n);
1668 	if (!n || !(n->nud_state & NUD_VALID)) {
1669 		if (n)
1670 			neigh_event_send(n, NULL);
1671 		rt->rt_gateway = orig_gw;
1672 		return -EAGAIN;
1673 	} else {
1674 		rt->rt_flags |= RTCF_REDIRECTED;
1675 		call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
1676 	}
1677 	return 0;
1678 }
1679 
1680 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1681 {
1682 	struct rtable *rt = (struct rtable *) dst;
1683 
1684 	if (rt_is_expired(rt))
1685 		return NULL;
1686 	if (rt->rt_peer_genid != rt_peer_genid()) {
1687 		struct inet_peer *peer;
1688 
1689 		if (!rt->peer)
1690 			rt_bind_peer(rt, rt->rt_dst, 0);
1691 
1692 		peer = rt->peer;
1693 		if (peer) {
1694 			check_peer_pmtu(dst, peer);
1695 
1696 			if (peer->redirect_learned.a4 &&
1697 			    peer->redirect_learned.a4 != rt->rt_gateway) {
1698 				if (check_peer_redir(dst, peer))
1699 					return NULL;
1700 			}
1701 		}
1702 
1703 		rt->rt_peer_genid = rt_peer_genid();
1704 	}
1705 	return dst;
1706 }
1707 
1708 static void ipv4_dst_destroy(struct dst_entry *dst)
1709 {
1710 	struct rtable *rt = (struct rtable *) dst;
1711 	struct inet_peer *peer = rt->peer;
1712 
1713 	if (rt->fi) {
1714 		fib_info_put(rt->fi);
1715 		rt->fi = NULL;
1716 	}
1717 	if (peer) {
1718 		rt->peer = NULL;
1719 		inet_putpeer(peer);
1720 	}
1721 }
1722 
1723 
1724 static void ipv4_link_failure(struct sk_buff *skb)
1725 {
1726 	struct rtable *rt;
1727 
1728 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1729 
1730 	rt = skb_rtable(skb);
1731 	if (rt && rt->peer && peer_pmtu_cleaned(rt->peer))
1732 		dst_metric_set(&rt->dst, RTAX_MTU, rt->peer->pmtu_orig);
1733 }
1734 
1735 static int ip_rt_bug(struct sk_buff *skb)
1736 {
1737 	printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1738 		&ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1739 		skb->dev ? skb->dev->name : "?");
1740 	kfree_skb(skb);
1741 	WARN_ON(1);
1742 	return 0;
1743 }
1744 
1745 /*
1746    We do not cache source address of outgoing interface,
1747    because it is used only by IP RR, TS and SRR options,
1748    so that it out of fast path.
1749 
1750    BTW remember: "addr" is allowed to be not aligned
1751    in IP options!
1752  */
1753 
1754 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1755 {
1756 	__be32 src;
1757 
1758 	if (rt_is_output_route(rt))
1759 		src = ip_hdr(skb)->saddr;
1760 	else {
1761 		struct fib_result res;
1762 		struct flowi4 fl4;
1763 		struct iphdr *iph;
1764 
1765 		iph = ip_hdr(skb);
1766 
1767 		memset(&fl4, 0, sizeof(fl4));
1768 		fl4.daddr = iph->daddr;
1769 		fl4.saddr = iph->saddr;
1770 		fl4.flowi4_tos = RT_TOS(iph->tos);
1771 		fl4.flowi4_oif = rt->dst.dev->ifindex;
1772 		fl4.flowi4_iif = skb->dev->ifindex;
1773 		fl4.flowi4_mark = skb->mark;
1774 
1775 		rcu_read_lock();
1776 		if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1777 			src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1778 		else
1779 			src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1780 					RT_SCOPE_UNIVERSE);
1781 		rcu_read_unlock();
1782 	}
1783 	memcpy(addr, &src, 4);
1784 }
1785 
1786 #ifdef CONFIG_IP_ROUTE_CLASSID
1787 static void set_class_tag(struct rtable *rt, u32 tag)
1788 {
1789 	if (!(rt->dst.tclassid & 0xFFFF))
1790 		rt->dst.tclassid |= tag & 0xFFFF;
1791 	if (!(rt->dst.tclassid & 0xFFFF0000))
1792 		rt->dst.tclassid |= tag & 0xFFFF0000;
1793 }
1794 #endif
1795 
1796 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1797 {
1798 	unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1799 
1800 	if (advmss == 0) {
1801 		advmss = max_t(unsigned int, dst->dev->mtu - 40,
1802 			       ip_rt_min_advmss);
1803 		if (advmss > 65535 - 40)
1804 			advmss = 65535 - 40;
1805 	}
1806 	return advmss;
1807 }
1808 
1809 static unsigned int ipv4_default_mtu(const struct dst_entry *dst)
1810 {
1811 	unsigned int mtu = dst->dev->mtu;
1812 
1813 	if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1814 		const struct rtable *rt = (const struct rtable *) dst;
1815 
1816 		if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1817 			mtu = 576;
1818 	}
1819 
1820 	if (mtu > IP_MAX_MTU)
1821 		mtu = IP_MAX_MTU;
1822 
1823 	return mtu;
1824 }
1825 
1826 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1827 			    struct fib_info *fi)
1828 {
1829 	struct inet_peer *peer;
1830 	int create = 0;
1831 
1832 	/* If a peer entry exists for this destination, we must hook
1833 	 * it up in order to get at cached metrics.
1834 	 */
1835 	if (fl4 && (fl4->flowi4_flags & FLOWI_FLAG_PRECOW_METRICS))
1836 		create = 1;
1837 
1838 	rt->peer = peer = inet_getpeer_v4(rt->rt_dst, create);
1839 	if (peer) {
1840 		rt->rt_peer_genid = rt_peer_genid();
1841 		if (inet_metrics_new(peer))
1842 			memcpy(peer->metrics, fi->fib_metrics,
1843 			       sizeof(u32) * RTAX_MAX);
1844 		dst_init_metrics(&rt->dst, peer->metrics, false);
1845 
1846 		check_peer_pmtu(&rt->dst, peer);
1847 		if (peer->redirect_learned.a4 &&
1848 		    peer->redirect_learned.a4 != rt->rt_gateway) {
1849 			rt->rt_gateway = peer->redirect_learned.a4;
1850 			rt->rt_flags |= RTCF_REDIRECTED;
1851 		}
1852 	} else {
1853 		if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1854 			rt->fi = fi;
1855 			atomic_inc(&fi->fib_clntref);
1856 		}
1857 		dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1858 	}
1859 }
1860 
1861 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1862 			   const struct fib_result *res,
1863 			   struct fib_info *fi, u16 type, u32 itag)
1864 {
1865 	struct dst_entry *dst = &rt->dst;
1866 
1867 	if (fi) {
1868 		if (FIB_RES_GW(*res) &&
1869 		    FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1870 			rt->rt_gateway = FIB_RES_GW(*res);
1871 		rt_init_metrics(rt, fl4, fi);
1872 #ifdef CONFIG_IP_ROUTE_CLASSID
1873 		dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
1874 #endif
1875 	}
1876 
1877 	if (dst_mtu(dst) > IP_MAX_MTU)
1878 		dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU);
1879 	if (dst_metric_raw(dst, RTAX_ADVMSS) > 65535 - 40)
1880 		dst_metric_set(dst, RTAX_ADVMSS, 65535 - 40);
1881 
1882 #ifdef CONFIG_IP_ROUTE_CLASSID
1883 #ifdef CONFIG_IP_MULTIPLE_TABLES
1884 	set_class_tag(rt, fib_rules_tclass(res));
1885 #endif
1886 	set_class_tag(rt, itag);
1887 #endif
1888 }
1889 
1890 static struct rtable *rt_dst_alloc(struct net_device *dev,
1891 				   bool nopolicy, bool noxfrm)
1892 {
1893 	return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
1894 			 DST_HOST |
1895 			 (nopolicy ? DST_NOPOLICY : 0) |
1896 			 (noxfrm ? DST_NOXFRM : 0));
1897 }
1898 
1899 /* called in rcu_read_lock() section */
1900 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1901 				u8 tos, struct net_device *dev, int our)
1902 {
1903 	unsigned int hash;
1904 	struct rtable *rth;
1905 	__be32 spec_dst;
1906 	struct in_device *in_dev = __in_dev_get_rcu(dev);
1907 	u32 itag = 0;
1908 	int err;
1909 
1910 	/* Primary sanity checks. */
1911 
1912 	if (in_dev == NULL)
1913 		return -EINVAL;
1914 
1915 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1916 	    ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1917 		goto e_inval;
1918 
1919 	if (ipv4_is_zeronet(saddr)) {
1920 		if (!ipv4_is_local_multicast(daddr))
1921 			goto e_inval;
1922 		spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1923 	} else {
1924 		err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
1925 					  &itag);
1926 		if (err < 0)
1927 			goto e_err;
1928 	}
1929 	rth = rt_dst_alloc(init_net.loopback_dev,
1930 			   IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1931 	if (!rth)
1932 		goto e_nobufs;
1933 
1934 #ifdef CONFIG_IP_ROUTE_CLASSID
1935 	rth->dst.tclassid = itag;
1936 #endif
1937 	rth->dst.output = ip_rt_bug;
1938 
1939 	rth->rt_key_dst	= daddr;
1940 	rth->rt_key_src	= saddr;
1941 	rth->rt_genid	= rt_genid(dev_net(dev));
1942 	rth->rt_flags	= RTCF_MULTICAST;
1943 	rth->rt_type	= RTN_MULTICAST;
1944 	rth->rt_key_tos	= tos;
1945 	rth->rt_dst	= daddr;
1946 	rth->rt_src	= saddr;
1947 	rth->rt_route_iif = dev->ifindex;
1948 	rth->rt_iif	= dev->ifindex;
1949 	rth->rt_oif	= 0;
1950 	rth->rt_mark    = skb->mark;
1951 	rth->rt_gateway	= daddr;
1952 	rth->rt_spec_dst= spec_dst;
1953 	rth->rt_peer_genid = 0;
1954 	rth->peer = NULL;
1955 	rth->fi = NULL;
1956 	if (our) {
1957 		rth->dst.input= ip_local_deliver;
1958 		rth->rt_flags |= RTCF_LOCAL;
1959 	}
1960 
1961 #ifdef CONFIG_IP_MROUTE
1962 	if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1963 		rth->dst.input = ip_mr_input;
1964 #endif
1965 	RT_CACHE_STAT_INC(in_slow_mc);
1966 
1967 	hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1968 	rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
1969 	return IS_ERR(rth) ? PTR_ERR(rth) : 0;
1970 
1971 e_nobufs:
1972 	return -ENOBUFS;
1973 e_inval:
1974 	return -EINVAL;
1975 e_err:
1976 	return err;
1977 }
1978 
1979 
1980 static void ip_handle_martian_source(struct net_device *dev,
1981 				     struct in_device *in_dev,
1982 				     struct sk_buff *skb,
1983 				     __be32 daddr,
1984 				     __be32 saddr)
1985 {
1986 	RT_CACHE_STAT_INC(in_martian_src);
1987 #ifdef CONFIG_IP_ROUTE_VERBOSE
1988 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1989 		/*
1990 		 *	RFC1812 recommendation, if source is martian,
1991 		 *	the only hint is MAC header.
1992 		 */
1993 		printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1994 			&daddr, &saddr, dev->name);
1995 		if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1996 			int i;
1997 			const unsigned char *p = skb_mac_header(skb);
1998 			printk(KERN_WARNING "ll header: ");
1999 			for (i = 0; i < dev->hard_header_len; i++, p++) {
2000 				printk("%02x", *p);
2001 				if (i < (dev->hard_header_len - 1))
2002 					printk(":");
2003 			}
2004 			printk("\n");
2005 		}
2006 	}
2007 #endif
2008 }
2009 
2010 /* called in rcu_read_lock() section */
2011 static int __mkroute_input(struct sk_buff *skb,
2012 			   const struct fib_result *res,
2013 			   struct in_device *in_dev,
2014 			   __be32 daddr, __be32 saddr, u32 tos,
2015 			   struct rtable **result)
2016 {
2017 	struct rtable *rth;
2018 	int err;
2019 	struct in_device *out_dev;
2020 	unsigned int flags = 0;
2021 	__be32 spec_dst;
2022 	u32 itag;
2023 
2024 	/* get a working reference to the output device */
2025 	out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
2026 	if (out_dev == NULL) {
2027 		if (net_ratelimit())
2028 			printk(KERN_CRIT "Bug in ip_route_input" \
2029 			       "_slow(). Please, report\n");
2030 		return -EINVAL;
2031 	}
2032 
2033 
2034 	err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
2035 				  in_dev->dev, &spec_dst, &itag);
2036 	if (err < 0) {
2037 		ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
2038 					 saddr);
2039 
2040 		goto cleanup;
2041 	}
2042 
2043 	if (err)
2044 		flags |= RTCF_DIRECTSRC;
2045 
2046 	if (out_dev == in_dev && err &&
2047 	    (IN_DEV_SHARED_MEDIA(out_dev) ||
2048 	     inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2049 		flags |= RTCF_DOREDIRECT;
2050 
2051 	if (skb->protocol != htons(ETH_P_IP)) {
2052 		/* Not IP (i.e. ARP). Do not create route, if it is
2053 		 * invalid for proxy arp. DNAT routes are always valid.
2054 		 *
2055 		 * Proxy arp feature have been extended to allow, ARP
2056 		 * replies back to the same interface, to support
2057 		 * Private VLAN switch technologies. See arp.c.
2058 		 */
2059 		if (out_dev == in_dev &&
2060 		    IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2061 			err = -EINVAL;
2062 			goto cleanup;
2063 		}
2064 	}
2065 
2066 	rth = rt_dst_alloc(out_dev->dev,
2067 			   IN_DEV_CONF_GET(in_dev, NOPOLICY),
2068 			   IN_DEV_CONF_GET(out_dev, NOXFRM));
2069 	if (!rth) {
2070 		err = -ENOBUFS;
2071 		goto cleanup;
2072 	}
2073 
2074 	rth->rt_key_dst	= daddr;
2075 	rth->rt_key_src	= saddr;
2076 	rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2077 	rth->rt_flags = flags;
2078 	rth->rt_type = res->type;
2079 	rth->rt_key_tos	= tos;
2080 	rth->rt_dst	= daddr;
2081 	rth->rt_src	= saddr;
2082 	rth->rt_route_iif = in_dev->dev->ifindex;
2083 	rth->rt_iif 	= in_dev->dev->ifindex;
2084 	rth->rt_oif 	= 0;
2085 	rth->rt_mark    = skb->mark;
2086 	rth->rt_gateway	= daddr;
2087 	rth->rt_spec_dst= spec_dst;
2088 	rth->rt_peer_genid = 0;
2089 	rth->peer = NULL;
2090 	rth->fi = NULL;
2091 
2092 	rth->dst.input = ip_forward;
2093 	rth->dst.output = ip_output;
2094 
2095 	rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
2096 
2097 	*result = rth;
2098 	err = 0;
2099  cleanup:
2100 	return err;
2101 }
2102 
2103 static int ip_mkroute_input(struct sk_buff *skb,
2104 			    struct fib_result *res,
2105 			    const struct flowi4 *fl4,
2106 			    struct in_device *in_dev,
2107 			    __be32 daddr, __be32 saddr, u32 tos)
2108 {
2109 	struct rtable* rth = NULL;
2110 	int err;
2111 	unsigned hash;
2112 
2113 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2114 	if (res->fi && res->fi->fib_nhs > 1)
2115 		fib_select_multipath(res);
2116 #endif
2117 
2118 	/* create a routing cache entry */
2119 	err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2120 	if (err)
2121 		return err;
2122 
2123 	/* put it into the cache */
2124 	hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
2125 		       rt_genid(dev_net(rth->dst.dev)));
2126 	rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
2127 	if (IS_ERR(rth))
2128 		return PTR_ERR(rth);
2129 	return 0;
2130 }
2131 
2132 /*
2133  *	NOTE. We drop all the packets that has local source
2134  *	addresses, because every properly looped back packet
2135  *	must have correct destination already attached by output routine.
2136  *
2137  *	Such approach solves two big problems:
2138  *	1. Not simplex devices are handled properly.
2139  *	2. IP spoofing attempts are filtered with 100% of guarantee.
2140  *	called with rcu_read_lock()
2141  */
2142 
2143 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2144 			       u8 tos, struct net_device *dev)
2145 {
2146 	struct fib_result res;
2147 	struct in_device *in_dev = __in_dev_get_rcu(dev);
2148 	struct flowi4	fl4;
2149 	unsigned	flags = 0;
2150 	u32		itag = 0;
2151 	struct rtable * rth;
2152 	unsigned	hash;
2153 	__be32		spec_dst;
2154 	int		err = -EINVAL;
2155 	struct net    * net = dev_net(dev);
2156 
2157 	/* IP on this device is disabled. */
2158 
2159 	if (!in_dev)
2160 		goto out;
2161 
2162 	/* Check for the most weird martians, which can be not detected
2163 	   by fib_lookup.
2164 	 */
2165 
2166 	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2167 	    ipv4_is_loopback(saddr))
2168 		goto martian_source;
2169 
2170 	if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2171 		goto brd_input;
2172 
2173 	/* Accept zero addresses only to limited broadcast;
2174 	 * I even do not know to fix it or not. Waiting for complains :-)
2175 	 */
2176 	if (ipv4_is_zeronet(saddr))
2177 		goto martian_source;
2178 
2179 	if (ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr))
2180 		goto martian_destination;
2181 
2182 	/*
2183 	 *	Now we are ready to route packet.
2184 	 */
2185 	fl4.flowi4_oif = 0;
2186 	fl4.flowi4_iif = dev->ifindex;
2187 	fl4.flowi4_mark = skb->mark;
2188 	fl4.flowi4_tos = tos;
2189 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2190 	fl4.daddr = daddr;
2191 	fl4.saddr = saddr;
2192 	err = fib_lookup(net, &fl4, &res);
2193 	if (err != 0) {
2194 		if (!IN_DEV_FORWARD(in_dev))
2195 			goto e_hostunreach;
2196 		goto no_route;
2197 	}
2198 
2199 	RT_CACHE_STAT_INC(in_slow_tot);
2200 
2201 	if (res.type == RTN_BROADCAST)
2202 		goto brd_input;
2203 
2204 	if (res.type == RTN_LOCAL) {
2205 		err = fib_validate_source(skb, saddr, daddr, tos,
2206 					  net->loopback_dev->ifindex,
2207 					  dev, &spec_dst, &itag);
2208 		if (err < 0)
2209 			goto martian_source_keep_err;
2210 		if (err)
2211 			flags |= RTCF_DIRECTSRC;
2212 		spec_dst = daddr;
2213 		goto local_input;
2214 	}
2215 
2216 	if (!IN_DEV_FORWARD(in_dev))
2217 		goto e_hostunreach;
2218 	if (res.type != RTN_UNICAST)
2219 		goto martian_destination;
2220 
2221 	err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2222 out:	return err;
2223 
2224 brd_input:
2225 	if (skb->protocol != htons(ETH_P_IP))
2226 		goto e_inval;
2227 
2228 	if (ipv4_is_zeronet(saddr))
2229 		spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2230 	else {
2231 		err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
2232 					  &itag);
2233 		if (err < 0)
2234 			goto martian_source_keep_err;
2235 		if (err)
2236 			flags |= RTCF_DIRECTSRC;
2237 	}
2238 	flags |= RTCF_BROADCAST;
2239 	res.type = RTN_BROADCAST;
2240 	RT_CACHE_STAT_INC(in_brd);
2241 
2242 local_input:
2243 	rth = rt_dst_alloc(net->loopback_dev,
2244 			   IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2245 	if (!rth)
2246 		goto e_nobufs;
2247 
2248 	rth->dst.input= ip_local_deliver;
2249 	rth->dst.output= ip_rt_bug;
2250 #ifdef CONFIG_IP_ROUTE_CLASSID
2251 	rth->dst.tclassid = itag;
2252 #endif
2253 
2254 	rth->rt_key_dst	= daddr;
2255 	rth->rt_key_src	= saddr;
2256 	rth->rt_genid = rt_genid(net);
2257 	rth->rt_flags 	= flags|RTCF_LOCAL;
2258 	rth->rt_type	= res.type;
2259 	rth->rt_key_tos	= tos;
2260 	rth->rt_dst	= daddr;
2261 	rth->rt_src	= saddr;
2262 #ifdef CONFIG_IP_ROUTE_CLASSID
2263 	rth->dst.tclassid = itag;
2264 #endif
2265 	rth->rt_route_iif = dev->ifindex;
2266 	rth->rt_iif	= dev->ifindex;
2267 	rth->rt_oif	= 0;
2268 	rth->rt_mark    = skb->mark;
2269 	rth->rt_gateway	= daddr;
2270 	rth->rt_spec_dst= spec_dst;
2271 	rth->rt_peer_genid = 0;
2272 	rth->peer = NULL;
2273 	rth->fi = NULL;
2274 	if (res.type == RTN_UNREACHABLE) {
2275 		rth->dst.input= ip_error;
2276 		rth->dst.error= -err;
2277 		rth->rt_flags 	&= ~RTCF_LOCAL;
2278 	}
2279 	hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2280 	rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2281 	err = 0;
2282 	if (IS_ERR(rth))
2283 		err = PTR_ERR(rth);
2284 	goto out;
2285 
2286 no_route:
2287 	RT_CACHE_STAT_INC(in_no_route);
2288 	spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2289 	res.type = RTN_UNREACHABLE;
2290 	if (err == -ESRCH)
2291 		err = -ENETUNREACH;
2292 	goto local_input;
2293 
2294 	/*
2295 	 *	Do not cache martian addresses: they should be logged (RFC1812)
2296 	 */
2297 martian_destination:
2298 	RT_CACHE_STAT_INC(in_martian_dst);
2299 #ifdef CONFIG_IP_ROUTE_VERBOSE
2300 	if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2301 		printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2302 			&daddr, &saddr, dev->name);
2303 #endif
2304 
2305 e_hostunreach:
2306 	err = -EHOSTUNREACH;
2307 	goto out;
2308 
2309 e_inval:
2310 	err = -EINVAL;
2311 	goto out;
2312 
2313 e_nobufs:
2314 	err = -ENOBUFS;
2315 	goto out;
2316 
2317 martian_source:
2318 	err = -EINVAL;
2319 martian_source_keep_err:
2320 	ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2321 	goto out;
2322 }
2323 
2324 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2325 			   u8 tos, struct net_device *dev, bool noref)
2326 {
2327 	struct rtable * rth;
2328 	unsigned	hash;
2329 	int iif = dev->ifindex;
2330 	struct net *net;
2331 	int res;
2332 
2333 	net = dev_net(dev);
2334 
2335 	rcu_read_lock();
2336 
2337 	if (!rt_caching(net))
2338 		goto skip_cache;
2339 
2340 	tos &= IPTOS_RT_MASK;
2341 	hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2342 
2343 	for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2344 	     rth = rcu_dereference(rth->dst.rt_next)) {
2345 		if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2346 		     ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2347 		     (rth->rt_route_iif ^ iif) |
2348 		     (rth->rt_key_tos ^ tos)) == 0 &&
2349 		    rth->rt_mark == skb->mark &&
2350 		    net_eq(dev_net(rth->dst.dev), net) &&
2351 		    !rt_is_expired(rth)) {
2352 			if (noref) {
2353 				dst_use_noref(&rth->dst, jiffies);
2354 				skb_dst_set_noref(skb, &rth->dst);
2355 			} else {
2356 				dst_use(&rth->dst, jiffies);
2357 				skb_dst_set(skb, &rth->dst);
2358 			}
2359 			RT_CACHE_STAT_INC(in_hit);
2360 			rcu_read_unlock();
2361 			return 0;
2362 		}
2363 		RT_CACHE_STAT_INC(in_hlist_search);
2364 	}
2365 
2366 skip_cache:
2367 	/* Multicast recognition logic is moved from route cache to here.
2368 	   The problem was that too many Ethernet cards have broken/missing
2369 	   hardware multicast filters :-( As result the host on multicasting
2370 	   network acquires a lot of useless route cache entries, sort of
2371 	   SDR messages from all the world. Now we try to get rid of them.
2372 	   Really, provided software IP multicast filter is organized
2373 	   reasonably (at least, hashed), it does not result in a slowdown
2374 	   comparing with route cache reject entries.
2375 	   Note, that multicast routers are not affected, because
2376 	   route cache entry is created eventually.
2377 	 */
2378 	if (ipv4_is_multicast(daddr)) {
2379 		struct in_device *in_dev = __in_dev_get_rcu(dev);
2380 
2381 		if (in_dev) {
2382 			int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2383 						  ip_hdr(skb)->protocol);
2384 			if (our
2385 #ifdef CONFIG_IP_MROUTE
2386 				||
2387 			    (!ipv4_is_local_multicast(daddr) &&
2388 			     IN_DEV_MFORWARD(in_dev))
2389 #endif
2390 			   ) {
2391 				int res = ip_route_input_mc(skb, daddr, saddr,
2392 							    tos, dev, our);
2393 				rcu_read_unlock();
2394 				return res;
2395 			}
2396 		}
2397 		rcu_read_unlock();
2398 		return -EINVAL;
2399 	}
2400 	res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2401 	rcu_read_unlock();
2402 	return res;
2403 }
2404 EXPORT_SYMBOL(ip_route_input_common);
2405 
2406 /* called with rcu_read_lock() */
2407 static struct rtable *__mkroute_output(const struct fib_result *res,
2408 				       const struct flowi4 *fl4,
2409 				       __be32 orig_daddr, __be32 orig_saddr,
2410 				       int orig_oif, struct net_device *dev_out,
2411 				       unsigned int flags)
2412 {
2413 	struct fib_info *fi = res->fi;
2414 	u32 tos = RT_FL_TOS(fl4);
2415 	struct in_device *in_dev;
2416 	u16 type = res->type;
2417 	struct rtable *rth;
2418 
2419 	if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2420 		return ERR_PTR(-EINVAL);
2421 
2422 	if (ipv4_is_lbcast(fl4->daddr))
2423 		type = RTN_BROADCAST;
2424 	else if (ipv4_is_multicast(fl4->daddr))
2425 		type = RTN_MULTICAST;
2426 	else if (ipv4_is_zeronet(fl4->daddr))
2427 		return ERR_PTR(-EINVAL);
2428 
2429 	if (dev_out->flags & IFF_LOOPBACK)
2430 		flags |= RTCF_LOCAL;
2431 
2432 	in_dev = __in_dev_get_rcu(dev_out);
2433 	if (!in_dev)
2434 		return ERR_PTR(-EINVAL);
2435 
2436 	if (type == RTN_BROADCAST) {
2437 		flags |= RTCF_BROADCAST | RTCF_LOCAL;
2438 		fi = NULL;
2439 	} else if (type == RTN_MULTICAST) {
2440 		flags |= RTCF_MULTICAST | RTCF_LOCAL;
2441 		if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2442 				     fl4->flowi4_proto))
2443 			flags &= ~RTCF_LOCAL;
2444 		/* If multicast route do not exist use
2445 		 * default one, but do not gateway in this case.
2446 		 * Yes, it is hack.
2447 		 */
2448 		if (fi && res->prefixlen < 4)
2449 			fi = NULL;
2450 	}
2451 
2452 	rth = rt_dst_alloc(dev_out,
2453 			   IN_DEV_CONF_GET(in_dev, NOPOLICY),
2454 			   IN_DEV_CONF_GET(in_dev, NOXFRM));
2455 	if (!rth)
2456 		return ERR_PTR(-ENOBUFS);
2457 
2458 	rth->dst.output = ip_output;
2459 
2460 	rth->rt_key_dst	= orig_daddr;
2461 	rth->rt_key_src	= orig_saddr;
2462 	rth->rt_genid = rt_genid(dev_net(dev_out));
2463 	rth->rt_flags	= flags;
2464 	rth->rt_type	= type;
2465 	rth->rt_key_tos	= tos;
2466 	rth->rt_dst	= fl4->daddr;
2467 	rth->rt_src	= fl4->saddr;
2468 	rth->rt_route_iif = 0;
2469 	rth->rt_iif	= orig_oif ? : dev_out->ifindex;
2470 	rth->rt_oif	= orig_oif;
2471 	rth->rt_mark    = fl4->flowi4_mark;
2472 	rth->rt_gateway = fl4->daddr;
2473 	rth->rt_spec_dst= fl4->saddr;
2474 	rth->rt_peer_genid = 0;
2475 	rth->peer = NULL;
2476 	rth->fi = NULL;
2477 
2478 	RT_CACHE_STAT_INC(out_slow_tot);
2479 
2480 	if (flags & RTCF_LOCAL) {
2481 		rth->dst.input = ip_local_deliver;
2482 		rth->rt_spec_dst = fl4->daddr;
2483 	}
2484 	if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2485 		rth->rt_spec_dst = fl4->saddr;
2486 		if (flags & RTCF_LOCAL &&
2487 		    !(dev_out->flags & IFF_LOOPBACK)) {
2488 			rth->dst.output = ip_mc_output;
2489 			RT_CACHE_STAT_INC(out_slow_mc);
2490 		}
2491 #ifdef CONFIG_IP_MROUTE
2492 		if (type == RTN_MULTICAST) {
2493 			if (IN_DEV_MFORWARD(in_dev) &&
2494 			    !ipv4_is_local_multicast(fl4->daddr)) {
2495 				rth->dst.input = ip_mr_input;
2496 				rth->dst.output = ip_mc_output;
2497 			}
2498 		}
2499 #endif
2500 	}
2501 
2502 	rt_set_nexthop(rth, fl4, res, fi, type, 0);
2503 
2504 	return rth;
2505 }
2506 
2507 /*
2508  * Major route resolver routine.
2509  * called with rcu_read_lock();
2510  */
2511 
2512 static struct rtable *ip_route_output_slow(struct net *net, struct flowi4 *fl4)
2513 {
2514 	struct net_device *dev_out = NULL;
2515 	u32 tos	= RT_FL_TOS(fl4);
2516 	unsigned int flags = 0;
2517 	struct fib_result res;
2518 	struct rtable *rth;
2519 	__be32 orig_daddr;
2520 	__be32 orig_saddr;
2521 	int orig_oif;
2522 
2523 	res.fi		= NULL;
2524 #ifdef CONFIG_IP_MULTIPLE_TABLES
2525 	res.r		= NULL;
2526 #endif
2527 
2528 	orig_daddr = fl4->daddr;
2529 	orig_saddr = fl4->saddr;
2530 	orig_oif = fl4->flowi4_oif;
2531 
2532 	fl4->flowi4_iif = net->loopback_dev->ifindex;
2533 	fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2534 	fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2535 			 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2536 
2537 	rcu_read_lock();
2538 	if (fl4->saddr) {
2539 		rth = ERR_PTR(-EINVAL);
2540 		if (ipv4_is_multicast(fl4->saddr) ||
2541 		    ipv4_is_lbcast(fl4->saddr) ||
2542 		    ipv4_is_zeronet(fl4->saddr))
2543 			goto out;
2544 
2545 		/* I removed check for oif == dev_out->oif here.
2546 		   It was wrong for two reasons:
2547 		   1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2548 		      is assigned to multiple interfaces.
2549 		   2. Moreover, we are allowed to send packets with saddr
2550 		      of another iface. --ANK
2551 		 */
2552 
2553 		if (fl4->flowi4_oif == 0 &&
2554 		    (ipv4_is_multicast(fl4->daddr) ||
2555 		     ipv4_is_lbcast(fl4->daddr))) {
2556 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2557 			dev_out = __ip_dev_find(net, fl4->saddr, false);
2558 			if (dev_out == NULL)
2559 				goto out;
2560 
2561 			/* Special hack: user can direct multicasts
2562 			   and limited broadcast via necessary interface
2563 			   without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2564 			   This hack is not just for fun, it allows
2565 			   vic,vat and friends to work.
2566 			   They bind socket to loopback, set ttl to zero
2567 			   and expect that it will work.
2568 			   From the viewpoint of routing cache they are broken,
2569 			   because we are not allowed to build multicast path
2570 			   with loopback source addr (look, routing cache
2571 			   cannot know, that ttl is zero, so that packet
2572 			   will not leave this host and route is valid).
2573 			   Luckily, this hack is good workaround.
2574 			 */
2575 
2576 			fl4->flowi4_oif = dev_out->ifindex;
2577 			goto make_route;
2578 		}
2579 
2580 		if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2581 			/* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2582 			if (!__ip_dev_find(net, fl4->saddr, false))
2583 				goto out;
2584 		}
2585 	}
2586 
2587 
2588 	if (fl4->flowi4_oif) {
2589 		dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2590 		rth = ERR_PTR(-ENODEV);
2591 		if (dev_out == NULL)
2592 			goto out;
2593 
2594 		/* RACE: Check return value of inet_select_addr instead. */
2595 		if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2596 			rth = ERR_PTR(-ENETUNREACH);
2597 			goto out;
2598 		}
2599 		if (ipv4_is_local_multicast(fl4->daddr) ||
2600 		    ipv4_is_lbcast(fl4->daddr)) {
2601 			if (!fl4->saddr)
2602 				fl4->saddr = inet_select_addr(dev_out, 0,
2603 							      RT_SCOPE_LINK);
2604 			goto make_route;
2605 		}
2606 		if (fl4->saddr) {
2607 			if (ipv4_is_multicast(fl4->daddr))
2608 				fl4->saddr = inet_select_addr(dev_out, 0,
2609 							      fl4->flowi4_scope);
2610 			else if (!fl4->daddr)
2611 				fl4->saddr = inet_select_addr(dev_out, 0,
2612 							      RT_SCOPE_HOST);
2613 		}
2614 	}
2615 
2616 	if (!fl4->daddr) {
2617 		fl4->daddr = fl4->saddr;
2618 		if (!fl4->daddr)
2619 			fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2620 		dev_out = net->loopback_dev;
2621 		fl4->flowi4_oif = net->loopback_dev->ifindex;
2622 		res.type = RTN_LOCAL;
2623 		flags |= RTCF_LOCAL;
2624 		goto make_route;
2625 	}
2626 
2627 	if (fib_lookup(net, fl4, &res)) {
2628 		res.fi = NULL;
2629 		if (fl4->flowi4_oif) {
2630 			/* Apparently, routing tables are wrong. Assume,
2631 			   that the destination is on link.
2632 
2633 			   WHY? DW.
2634 			   Because we are allowed to send to iface
2635 			   even if it has NO routes and NO assigned
2636 			   addresses. When oif is specified, routing
2637 			   tables are looked up with only one purpose:
2638 			   to catch if destination is gatewayed, rather than
2639 			   direct. Moreover, if MSG_DONTROUTE is set,
2640 			   we send packet, ignoring both routing tables
2641 			   and ifaddr state. --ANK
2642 
2643 
2644 			   We could make it even if oif is unknown,
2645 			   likely IPv6, but we do not.
2646 			 */
2647 
2648 			if (fl4->saddr == 0)
2649 				fl4->saddr = inet_select_addr(dev_out, 0,
2650 							      RT_SCOPE_LINK);
2651 			res.type = RTN_UNICAST;
2652 			goto make_route;
2653 		}
2654 		rth = ERR_PTR(-ENETUNREACH);
2655 		goto out;
2656 	}
2657 
2658 	if (res.type == RTN_LOCAL) {
2659 		if (!fl4->saddr) {
2660 			if (res.fi->fib_prefsrc)
2661 				fl4->saddr = res.fi->fib_prefsrc;
2662 			else
2663 				fl4->saddr = fl4->daddr;
2664 		}
2665 		dev_out = net->loopback_dev;
2666 		fl4->flowi4_oif = dev_out->ifindex;
2667 		res.fi = NULL;
2668 		flags |= RTCF_LOCAL;
2669 		goto make_route;
2670 	}
2671 
2672 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2673 	if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2674 		fib_select_multipath(&res);
2675 	else
2676 #endif
2677 	if (!res.prefixlen &&
2678 	    res.table->tb_num_default > 1 &&
2679 	    res.type == RTN_UNICAST && !fl4->flowi4_oif)
2680 		fib_select_default(&res);
2681 
2682 	if (!fl4->saddr)
2683 		fl4->saddr = FIB_RES_PREFSRC(net, res);
2684 
2685 	dev_out = FIB_RES_DEV(res);
2686 	fl4->flowi4_oif = dev_out->ifindex;
2687 
2688 
2689 make_route:
2690 	rth = __mkroute_output(&res, fl4, orig_daddr, orig_saddr, orig_oif,
2691 			       dev_out, flags);
2692 	if (!IS_ERR(rth)) {
2693 		unsigned int hash;
2694 
2695 		hash = rt_hash(orig_daddr, orig_saddr, orig_oif,
2696 			       rt_genid(dev_net(dev_out)));
2697 		rth = rt_intern_hash(hash, rth, NULL, orig_oif);
2698 	}
2699 
2700 out:
2701 	rcu_read_unlock();
2702 	return rth;
2703 }
2704 
2705 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *flp4)
2706 {
2707 	struct rtable *rth;
2708 	unsigned int hash;
2709 
2710 	if (!rt_caching(net))
2711 		goto slow_output;
2712 
2713 	hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2714 
2715 	rcu_read_lock_bh();
2716 	for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2717 		rth = rcu_dereference_bh(rth->dst.rt_next)) {
2718 		if (rth->rt_key_dst == flp4->daddr &&
2719 		    rth->rt_key_src == flp4->saddr &&
2720 		    rt_is_output_route(rth) &&
2721 		    rth->rt_oif == flp4->flowi4_oif &&
2722 		    rth->rt_mark == flp4->flowi4_mark &&
2723 		    !((rth->rt_key_tos ^ flp4->flowi4_tos) &
2724 			    (IPTOS_RT_MASK | RTO_ONLINK)) &&
2725 		    net_eq(dev_net(rth->dst.dev), net) &&
2726 		    !rt_is_expired(rth)) {
2727 			dst_use(&rth->dst, jiffies);
2728 			RT_CACHE_STAT_INC(out_hit);
2729 			rcu_read_unlock_bh();
2730 			if (!flp4->saddr)
2731 				flp4->saddr = rth->rt_src;
2732 			if (!flp4->daddr)
2733 				flp4->daddr = rth->rt_dst;
2734 			return rth;
2735 		}
2736 		RT_CACHE_STAT_INC(out_hlist_search);
2737 	}
2738 	rcu_read_unlock_bh();
2739 
2740 slow_output:
2741 	return ip_route_output_slow(net, flp4);
2742 }
2743 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2744 
2745 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2746 {
2747 	return NULL;
2748 }
2749 
2750 static unsigned int ipv4_blackhole_default_mtu(const struct dst_entry *dst)
2751 {
2752 	return 0;
2753 }
2754 
2755 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2756 {
2757 }
2758 
2759 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2760 					  unsigned long old)
2761 {
2762 	return NULL;
2763 }
2764 
2765 static struct dst_ops ipv4_dst_blackhole_ops = {
2766 	.family			=	AF_INET,
2767 	.protocol		=	cpu_to_be16(ETH_P_IP),
2768 	.destroy		=	ipv4_dst_destroy,
2769 	.check			=	ipv4_blackhole_dst_check,
2770 	.default_mtu		=	ipv4_blackhole_default_mtu,
2771 	.default_advmss		=	ipv4_default_advmss,
2772 	.update_pmtu		=	ipv4_rt_blackhole_update_pmtu,
2773 	.cow_metrics		=	ipv4_rt_blackhole_cow_metrics,
2774 	.neigh_lookup		=	ipv4_neigh_lookup,
2775 };
2776 
2777 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2778 {
2779 	struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
2780 	struct rtable *ort = (struct rtable *) dst_orig;
2781 
2782 	if (rt) {
2783 		struct dst_entry *new = &rt->dst;
2784 
2785 		new->__use = 1;
2786 		new->input = dst_discard;
2787 		new->output = dst_discard;
2788 		dst_copy_metrics(new, &ort->dst);
2789 
2790 		new->dev = ort->dst.dev;
2791 		if (new->dev)
2792 			dev_hold(new->dev);
2793 
2794 		rt->rt_key_dst = ort->rt_key_dst;
2795 		rt->rt_key_src = ort->rt_key_src;
2796 		rt->rt_key_tos = ort->rt_key_tos;
2797 		rt->rt_route_iif = ort->rt_route_iif;
2798 		rt->rt_iif = ort->rt_iif;
2799 		rt->rt_oif = ort->rt_oif;
2800 		rt->rt_mark = ort->rt_mark;
2801 
2802 		rt->rt_genid = rt_genid(net);
2803 		rt->rt_flags = ort->rt_flags;
2804 		rt->rt_type = ort->rt_type;
2805 		rt->rt_dst = ort->rt_dst;
2806 		rt->rt_src = ort->rt_src;
2807 		rt->rt_gateway = ort->rt_gateway;
2808 		rt->rt_spec_dst = ort->rt_spec_dst;
2809 		rt->peer = ort->peer;
2810 		if (rt->peer)
2811 			atomic_inc(&rt->peer->refcnt);
2812 		rt->fi = ort->fi;
2813 		if (rt->fi)
2814 			atomic_inc(&rt->fi->fib_clntref);
2815 
2816 		dst_free(new);
2817 	}
2818 
2819 	dst_release(dst_orig);
2820 
2821 	return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2822 }
2823 
2824 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2825 				    struct sock *sk)
2826 {
2827 	struct rtable *rt = __ip_route_output_key(net, flp4);
2828 
2829 	if (IS_ERR(rt))
2830 		return rt;
2831 
2832 	if (flp4->flowi4_proto)
2833 		rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2834 						   flowi4_to_flowi(flp4),
2835 						   sk, 0);
2836 
2837 	return rt;
2838 }
2839 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2840 
2841 static int rt_fill_info(struct net *net,
2842 			struct sk_buff *skb, u32 pid, u32 seq, int event,
2843 			int nowait, unsigned int flags)
2844 {
2845 	struct rtable *rt = skb_rtable(skb);
2846 	struct rtmsg *r;
2847 	struct nlmsghdr *nlh;
2848 	long expires = 0;
2849 	const struct inet_peer *peer = rt->peer;
2850 	u32 id = 0, ts = 0, tsage = 0, error;
2851 
2852 	nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2853 	if (nlh == NULL)
2854 		return -EMSGSIZE;
2855 
2856 	r = nlmsg_data(nlh);
2857 	r->rtm_family	 = AF_INET;
2858 	r->rtm_dst_len	= 32;
2859 	r->rtm_src_len	= 0;
2860 	r->rtm_tos	= rt->rt_key_tos;
2861 	r->rtm_table	= RT_TABLE_MAIN;
2862 	NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2863 	r->rtm_type	= rt->rt_type;
2864 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2865 	r->rtm_protocol = RTPROT_UNSPEC;
2866 	r->rtm_flags	= (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2867 	if (rt->rt_flags & RTCF_NOTIFY)
2868 		r->rtm_flags |= RTM_F_NOTIFY;
2869 
2870 	NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2871 
2872 	if (rt->rt_key_src) {
2873 		r->rtm_src_len = 32;
2874 		NLA_PUT_BE32(skb, RTA_SRC, rt->rt_key_src);
2875 	}
2876 	if (rt->dst.dev)
2877 		NLA_PUT_U32(skb, RTA_OIF, rt->dst.dev->ifindex);
2878 #ifdef CONFIG_IP_ROUTE_CLASSID
2879 	if (rt->dst.tclassid)
2880 		NLA_PUT_U32(skb, RTA_FLOW, rt->dst.tclassid);
2881 #endif
2882 	if (rt_is_input_route(rt))
2883 		NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2884 	else if (rt->rt_src != rt->rt_key_src)
2885 		NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2886 
2887 	if (rt->rt_dst != rt->rt_gateway)
2888 		NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2889 
2890 	if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2891 		goto nla_put_failure;
2892 
2893 	if (rt->rt_mark)
2894 		NLA_PUT_BE32(skb, RTA_MARK, rt->rt_mark);
2895 
2896 	error = rt->dst.error;
2897 	if (peer) {
2898 		inet_peer_refcheck(rt->peer);
2899 		id = atomic_read(&peer->ip_id_count) & 0xffff;
2900 		if (peer->tcp_ts_stamp) {
2901 			ts = peer->tcp_ts;
2902 			tsage = get_seconds() - peer->tcp_ts_stamp;
2903 		}
2904 		expires = ACCESS_ONCE(peer->pmtu_expires);
2905 		if (expires)
2906 			expires -= jiffies;
2907 	}
2908 
2909 	if (rt_is_input_route(rt)) {
2910 #ifdef CONFIG_IP_MROUTE
2911 		__be32 dst = rt->rt_dst;
2912 
2913 		if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2914 		    IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2915 			int err = ipmr_get_route(net, skb,
2916 						 rt->rt_src, rt->rt_dst,
2917 						 r, nowait);
2918 			if (err <= 0) {
2919 				if (!nowait) {
2920 					if (err == 0)
2921 						return 0;
2922 					goto nla_put_failure;
2923 				} else {
2924 					if (err == -EMSGSIZE)
2925 						goto nla_put_failure;
2926 					error = err;
2927 				}
2928 			}
2929 		} else
2930 #endif
2931 			NLA_PUT_U32(skb, RTA_IIF, rt->rt_iif);
2932 	}
2933 
2934 	if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
2935 			       expires, error) < 0)
2936 		goto nla_put_failure;
2937 
2938 	return nlmsg_end(skb, nlh);
2939 
2940 nla_put_failure:
2941 	nlmsg_cancel(skb, nlh);
2942 	return -EMSGSIZE;
2943 }
2944 
2945 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2946 {
2947 	struct net *net = sock_net(in_skb->sk);
2948 	struct rtmsg *rtm;
2949 	struct nlattr *tb[RTA_MAX+1];
2950 	struct rtable *rt = NULL;
2951 	__be32 dst = 0;
2952 	__be32 src = 0;
2953 	u32 iif;
2954 	int err;
2955 	int mark;
2956 	struct sk_buff *skb;
2957 
2958 	err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2959 	if (err < 0)
2960 		goto errout;
2961 
2962 	rtm = nlmsg_data(nlh);
2963 
2964 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2965 	if (skb == NULL) {
2966 		err = -ENOBUFS;
2967 		goto errout;
2968 	}
2969 
2970 	/* Reserve room for dummy headers, this skb can pass
2971 	   through good chunk of routing engine.
2972 	 */
2973 	skb_reset_mac_header(skb);
2974 	skb_reset_network_header(skb);
2975 
2976 	/* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2977 	ip_hdr(skb)->protocol = IPPROTO_ICMP;
2978 	skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2979 
2980 	src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2981 	dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2982 	iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2983 	mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2984 
2985 	if (iif) {
2986 		struct net_device *dev;
2987 
2988 		dev = __dev_get_by_index(net, iif);
2989 		if (dev == NULL) {
2990 			err = -ENODEV;
2991 			goto errout_free;
2992 		}
2993 
2994 		skb->protocol	= htons(ETH_P_IP);
2995 		skb->dev	= dev;
2996 		skb->mark	= mark;
2997 		local_bh_disable();
2998 		err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2999 		local_bh_enable();
3000 
3001 		rt = skb_rtable(skb);
3002 		if (err == 0 && rt->dst.error)
3003 			err = -rt->dst.error;
3004 	} else {
3005 		struct flowi4 fl4 = {
3006 			.daddr = dst,
3007 			.saddr = src,
3008 			.flowi4_tos = rtm->rtm_tos,
3009 			.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
3010 			.flowi4_mark = mark,
3011 		};
3012 		rt = ip_route_output_key(net, &fl4);
3013 
3014 		err = 0;
3015 		if (IS_ERR(rt))
3016 			err = PTR_ERR(rt);
3017 	}
3018 
3019 	if (err)
3020 		goto errout_free;
3021 
3022 	skb_dst_set(skb, &rt->dst);
3023 	if (rtm->rtm_flags & RTM_F_NOTIFY)
3024 		rt->rt_flags |= RTCF_NOTIFY;
3025 
3026 	err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3027 			   RTM_NEWROUTE, 0, 0);
3028 	if (err <= 0)
3029 		goto errout_free;
3030 
3031 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3032 errout:
3033 	return err;
3034 
3035 errout_free:
3036 	kfree_skb(skb);
3037 	goto errout;
3038 }
3039 
3040 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
3041 {
3042 	struct rtable *rt;
3043 	int h, s_h;
3044 	int idx, s_idx;
3045 	struct net *net;
3046 
3047 	net = sock_net(skb->sk);
3048 
3049 	s_h = cb->args[0];
3050 	if (s_h < 0)
3051 		s_h = 0;
3052 	s_idx = idx = cb->args[1];
3053 	for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3054 		if (!rt_hash_table[h].chain)
3055 			continue;
3056 		rcu_read_lock_bh();
3057 		for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3058 		     rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3059 			if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3060 				continue;
3061 			if (rt_is_expired(rt))
3062 				continue;
3063 			skb_dst_set_noref(skb, &rt->dst);
3064 			if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3065 					 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3066 					 1, NLM_F_MULTI) <= 0) {
3067 				skb_dst_drop(skb);
3068 				rcu_read_unlock_bh();
3069 				goto done;
3070 			}
3071 			skb_dst_drop(skb);
3072 		}
3073 		rcu_read_unlock_bh();
3074 	}
3075 
3076 done:
3077 	cb->args[0] = h;
3078 	cb->args[1] = idx;
3079 	return skb->len;
3080 }
3081 
3082 void ip_rt_multicast_event(struct in_device *in_dev)
3083 {
3084 	rt_cache_flush(dev_net(in_dev->dev), 0);
3085 }
3086 
3087 #ifdef CONFIG_SYSCTL
3088 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3089 					void __user *buffer,
3090 					size_t *lenp, loff_t *ppos)
3091 {
3092 	if (write) {
3093 		int flush_delay;
3094 		ctl_table ctl;
3095 		struct net *net;
3096 
3097 		memcpy(&ctl, __ctl, sizeof(ctl));
3098 		ctl.data = &flush_delay;
3099 		proc_dointvec(&ctl, write, buffer, lenp, ppos);
3100 
3101 		net = (struct net *)__ctl->extra1;
3102 		rt_cache_flush(net, flush_delay);
3103 		return 0;
3104 	}
3105 
3106 	return -EINVAL;
3107 }
3108 
3109 static ctl_table ipv4_route_table[] = {
3110 	{
3111 		.procname	= "gc_thresh",
3112 		.data		= &ipv4_dst_ops.gc_thresh,
3113 		.maxlen		= sizeof(int),
3114 		.mode		= 0644,
3115 		.proc_handler	= proc_dointvec,
3116 	},
3117 	{
3118 		.procname	= "max_size",
3119 		.data		= &ip_rt_max_size,
3120 		.maxlen		= sizeof(int),
3121 		.mode		= 0644,
3122 		.proc_handler	= proc_dointvec,
3123 	},
3124 	{
3125 		/*  Deprecated. Use gc_min_interval_ms */
3126 
3127 		.procname	= "gc_min_interval",
3128 		.data		= &ip_rt_gc_min_interval,
3129 		.maxlen		= sizeof(int),
3130 		.mode		= 0644,
3131 		.proc_handler	= proc_dointvec_jiffies,
3132 	},
3133 	{
3134 		.procname	= "gc_min_interval_ms",
3135 		.data		= &ip_rt_gc_min_interval,
3136 		.maxlen		= sizeof(int),
3137 		.mode		= 0644,
3138 		.proc_handler	= proc_dointvec_ms_jiffies,
3139 	},
3140 	{
3141 		.procname	= "gc_timeout",
3142 		.data		= &ip_rt_gc_timeout,
3143 		.maxlen		= sizeof(int),
3144 		.mode		= 0644,
3145 		.proc_handler	= proc_dointvec_jiffies,
3146 	},
3147 	{
3148 		.procname	= "redirect_load",
3149 		.data		= &ip_rt_redirect_load,
3150 		.maxlen		= sizeof(int),
3151 		.mode		= 0644,
3152 		.proc_handler	= proc_dointvec,
3153 	},
3154 	{
3155 		.procname	= "redirect_number",
3156 		.data		= &ip_rt_redirect_number,
3157 		.maxlen		= sizeof(int),
3158 		.mode		= 0644,
3159 		.proc_handler	= proc_dointvec,
3160 	},
3161 	{
3162 		.procname	= "redirect_silence",
3163 		.data		= &ip_rt_redirect_silence,
3164 		.maxlen		= sizeof(int),
3165 		.mode		= 0644,
3166 		.proc_handler	= proc_dointvec,
3167 	},
3168 	{
3169 		.procname	= "error_cost",
3170 		.data		= &ip_rt_error_cost,
3171 		.maxlen		= sizeof(int),
3172 		.mode		= 0644,
3173 		.proc_handler	= proc_dointvec,
3174 	},
3175 	{
3176 		.procname	= "error_burst",
3177 		.data		= &ip_rt_error_burst,
3178 		.maxlen		= sizeof(int),
3179 		.mode		= 0644,
3180 		.proc_handler	= proc_dointvec,
3181 	},
3182 	{
3183 		.procname	= "gc_elasticity",
3184 		.data		= &ip_rt_gc_elasticity,
3185 		.maxlen		= sizeof(int),
3186 		.mode		= 0644,
3187 		.proc_handler	= proc_dointvec,
3188 	},
3189 	{
3190 		.procname	= "mtu_expires",
3191 		.data		= &ip_rt_mtu_expires,
3192 		.maxlen		= sizeof(int),
3193 		.mode		= 0644,
3194 		.proc_handler	= proc_dointvec_jiffies,
3195 	},
3196 	{
3197 		.procname	= "min_pmtu",
3198 		.data		= &ip_rt_min_pmtu,
3199 		.maxlen		= sizeof(int),
3200 		.mode		= 0644,
3201 		.proc_handler	= proc_dointvec,
3202 	},
3203 	{
3204 		.procname	= "min_adv_mss",
3205 		.data		= &ip_rt_min_advmss,
3206 		.maxlen		= sizeof(int),
3207 		.mode		= 0644,
3208 		.proc_handler	= proc_dointvec,
3209 	},
3210 	{ }
3211 };
3212 
3213 static struct ctl_table empty[1];
3214 
3215 static struct ctl_table ipv4_skeleton[] =
3216 {
3217 	{ .procname = "route",
3218 	  .mode = 0555, .child = ipv4_route_table},
3219 	{ .procname = "neigh",
3220 	  .mode = 0555, .child = empty},
3221 	{ }
3222 };
3223 
3224 static __net_initdata struct ctl_path ipv4_path[] = {
3225 	{ .procname = "net", },
3226 	{ .procname = "ipv4", },
3227 	{ },
3228 };
3229 
3230 static struct ctl_table ipv4_route_flush_table[] = {
3231 	{
3232 		.procname	= "flush",
3233 		.maxlen		= sizeof(int),
3234 		.mode		= 0200,
3235 		.proc_handler	= ipv4_sysctl_rtcache_flush,
3236 	},
3237 	{ },
3238 };
3239 
3240 static __net_initdata struct ctl_path ipv4_route_path[] = {
3241 	{ .procname = "net", },
3242 	{ .procname = "ipv4", },
3243 	{ .procname = "route", },
3244 	{ },
3245 };
3246 
3247 static __net_init int sysctl_route_net_init(struct net *net)
3248 {
3249 	struct ctl_table *tbl;
3250 
3251 	tbl = ipv4_route_flush_table;
3252 	if (!net_eq(net, &init_net)) {
3253 		tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3254 		if (tbl == NULL)
3255 			goto err_dup;
3256 	}
3257 	tbl[0].extra1 = net;
3258 
3259 	net->ipv4.route_hdr =
3260 		register_net_sysctl_table(net, ipv4_route_path, tbl);
3261 	if (net->ipv4.route_hdr == NULL)
3262 		goto err_reg;
3263 	return 0;
3264 
3265 err_reg:
3266 	if (tbl != ipv4_route_flush_table)
3267 		kfree(tbl);
3268 err_dup:
3269 	return -ENOMEM;
3270 }
3271 
3272 static __net_exit void sysctl_route_net_exit(struct net *net)
3273 {
3274 	struct ctl_table *tbl;
3275 
3276 	tbl = net->ipv4.route_hdr->ctl_table_arg;
3277 	unregister_net_sysctl_table(net->ipv4.route_hdr);
3278 	BUG_ON(tbl == ipv4_route_flush_table);
3279 	kfree(tbl);
3280 }
3281 
3282 static __net_initdata struct pernet_operations sysctl_route_ops = {
3283 	.init = sysctl_route_net_init,
3284 	.exit = sysctl_route_net_exit,
3285 };
3286 #endif
3287 
3288 static __net_init int rt_genid_init(struct net *net)
3289 {
3290 	get_random_bytes(&net->ipv4.rt_genid,
3291 			 sizeof(net->ipv4.rt_genid));
3292 	get_random_bytes(&net->ipv4.dev_addr_genid,
3293 			 sizeof(net->ipv4.dev_addr_genid));
3294 	return 0;
3295 }
3296 
3297 static __net_initdata struct pernet_operations rt_genid_ops = {
3298 	.init = rt_genid_init,
3299 };
3300 
3301 
3302 #ifdef CONFIG_IP_ROUTE_CLASSID
3303 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3304 #endif /* CONFIG_IP_ROUTE_CLASSID */
3305 
3306 static __initdata unsigned long rhash_entries;
3307 static int __init set_rhash_entries(char *str)
3308 {
3309 	if (!str)
3310 		return 0;
3311 	rhash_entries = simple_strtoul(str, &str, 0);
3312 	return 1;
3313 }
3314 __setup("rhash_entries=", set_rhash_entries);
3315 
3316 int __init ip_rt_init(void)
3317 {
3318 	int rc = 0;
3319 
3320 #ifdef CONFIG_IP_ROUTE_CLASSID
3321 	ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3322 	if (!ip_rt_acct)
3323 		panic("IP: failed to allocate ip_rt_acct\n");
3324 #endif
3325 
3326 	ipv4_dst_ops.kmem_cachep =
3327 		kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3328 				  SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3329 
3330 	ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3331 
3332 	if (dst_entries_init(&ipv4_dst_ops) < 0)
3333 		panic("IP: failed to allocate ipv4_dst_ops counter\n");
3334 
3335 	if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3336 		panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3337 
3338 	rt_hash_table = (struct rt_hash_bucket *)
3339 		alloc_large_system_hash("IP route cache",
3340 					sizeof(struct rt_hash_bucket),
3341 					rhash_entries,
3342 					(totalram_pages >= 128 * 1024) ?
3343 					15 : 17,
3344 					0,
3345 					&rt_hash_log,
3346 					&rt_hash_mask,
3347 					rhash_entries ? 0 : 512 * 1024);
3348 	memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3349 	rt_hash_lock_init();
3350 
3351 	ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3352 	ip_rt_max_size = (rt_hash_mask + 1) * 16;
3353 
3354 	devinet_init();
3355 	ip_fib_init();
3356 
3357 	if (ip_rt_proc_init())
3358 		printk(KERN_ERR "Unable to create route proc files\n");
3359 #ifdef CONFIG_XFRM
3360 	xfrm_init();
3361 	xfrm4_init(ip_rt_max_size);
3362 #endif
3363 	rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
3364 
3365 #ifdef CONFIG_SYSCTL
3366 	register_pernet_subsys(&sysctl_route_ops);
3367 #endif
3368 	register_pernet_subsys(&rt_genid_ops);
3369 	return rc;
3370 }
3371 
3372 #ifdef CONFIG_SYSCTL
3373 /*
3374  * We really need to sanitize the damn ipv4 init order, then all
3375  * this nonsense will go away.
3376  */
3377 void __init ip_static_sysctl_init(void)
3378 {
3379 	register_sysctl_paths(ipv4_path, ipv4_skeleton);
3380 }
3381 #endif
3382