?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
perf tools: Fix find tids routine by excluding "." and ".."
[cris-mirror.git] / net / ipv4 / route.c
blob560acc677ce485cce87a0161fa709f5e50b8e56f
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
112 #define RT_FL_TOS(oldflp) \
113 ((u32)(oldflp->fl4_tos & (IPTOS_RT_MASK | RTO_ONLINK)))
114
115 #define IP_MAX_MTU 0xFFF0
116
117 #define RT_GC_TIMEOUT (300*HZ)
118
119 static int ip_rt_max_size;
120 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
121 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
122 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
123 static int ip_rt_redirect_number __read_mostly = 9;
124 static int ip_rt_redirect_load __read_mostly = HZ / 50;
125 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
126 static int ip_rt_error_cost __read_mostly = HZ;
127 static int ip_rt_error_burst __read_mostly = 5 * HZ;
128 static int ip_rt_gc_elasticity __read_mostly = 8;
129 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
130 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
131 static int ip_rt_min_advmss __read_mostly = 256;
132 static int rt_chain_length_max __read_mostly = 20;
133
134 static struct delayed_work expires_work;
135 static unsigned long expires_ljiffies;
136
137 /*
138 * Interface to generic destination cache.
139 */
140
141 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
142 static void ipv4_dst_destroy(struct dst_entry *dst);
143 static void ipv4_dst_ifdown(struct dst_entry *dst,
144 struct net_device *dev, int how);
145 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
146 static void ipv4_link_failure(struct sk_buff *skb);
147 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
148 static int rt_garbage_collect(struct dst_ops *ops);
149
150
151 static struct dst_ops ipv4_dst_ops = {
152 .family = AF_INET,
153 .protocol = cpu_to_be16(ETH_P_IP),
154 .gc = rt_garbage_collect,
155 .check = ipv4_dst_check,
156 .destroy = ipv4_dst_destroy,
157 .ifdown = ipv4_dst_ifdown,
158 .negative_advice = ipv4_negative_advice,
159 .link_failure = ipv4_link_failure,
160 .update_pmtu = ip_rt_update_pmtu,
161 .local_out = __ip_local_out,
162 .entries = ATOMIC_INIT(0),
163 };
164
165 #define ECN_OR_COST(class) TC_PRIO_##class
166
167 const __u8 ip_tos2prio[16] = {
168 TC_PRIO_BESTEFFORT,
169 ECN_OR_COST(FILLER),
170 TC_PRIO_BESTEFFORT,
171 ECN_OR_COST(BESTEFFORT),
172 TC_PRIO_BULK,
173 ECN_OR_COST(BULK),
174 TC_PRIO_BULK,
175 ECN_OR_COST(BULK),
176 TC_PRIO_INTERACTIVE,
177 ECN_OR_COST(INTERACTIVE),
178 TC_PRIO_INTERACTIVE,
179 ECN_OR_COST(INTERACTIVE),
180 TC_PRIO_INTERACTIVE_BULK,
181 ECN_OR_COST(INTERACTIVE_BULK),
182 TC_PRIO_INTERACTIVE_BULK,
183 ECN_OR_COST(INTERACTIVE_BULK)
184 };
185
186
187 /*
188 * Route cache.
189 */
190
191 /* The locking scheme is rather straight forward:
192 *
193 * 1) Read-Copy Update protects the buckets of the central route hash.
194 * 2) Only writers remove entries, and they hold the lock
195 * as they look at rtable reference counts.
196 * 3) Only readers acquire references to rtable entries,
197 * they do so with atomic increments and with the
198 * lock held.
199 */
200
201 struct rt_hash_bucket {
202 struct rtable *chain;
203 };
204
205 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
206 defined(CONFIG_PROVE_LOCKING)
207 /*
208 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
209 * The size of this table is a power of two and depends on the number of CPUS.
210 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
211 */
212 #ifdef CONFIG_LOCKDEP
213 # define RT_HASH_LOCK_SZ 256
214 #else
215 # if NR_CPUS >= 32
216 # define RT_HASH_LOCK_SZ 4096
217 # elif NR_CPUS >= 16
218 # define RT_HASH_LOCK_SZ 2048
219 # elif NR_CPUS >= 8
220 # define RT_HASH_LOCK_SZ 1024
221 # elif NR_CPUS >= 4
222 # define RT_HASH_LOCK_SZ 512
223 # else
224 # define RT_HASH_LOCK_SZ 256
225 # endif
226 #endif
227
228 static spinlock_t *rt_hash_locks;
229 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
230
231 static __init void rt_hash_lock_init(void)
232 {
233 int i;
234
235 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
236 GFP_KERNEL);
237 if (!rt_hash_locks)
238 panic("IP: failed to allocate rt_hash_locks\n");
239
240 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
241 spin_lock_init(&rt_hash_locks[i]);
242 }
243 #else
244 # define rt_hash_lock_addr(slot) NULL
245
246 static inline void rt_hash_lock_init(void)
247 {
248 }
249 #endif
250
251 static struct rt_hash_bucket *rt_hash_table __read_mostly;
252 static unsigned rt_hash_mask __read_mostly;
253 static unsigned int rt_hash_log __read_mostly;
254
255 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
256 #define RT_CACHE_STAT_INC(field) \
257 (__raw_get_cpu_var(rt_cache_stat).field++)
258
259 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
260 int genid)
261 {
262 return jhash_3words((__force u32)daddr, (__force u32)saddr,
263 idx, genid)
264 & rt_hash_mask;
265 }
266
267 static inline int rt_genid(struct net *net)
268 {
269 return atomic_read(&net->ipv4.rt_genid);
270 }
271
272 #ifdef CONFIG_PROC_FS
273 struct rt_cache_iter_state {
274 struct seq_net_private p;
275 int bucket;
276 int genid;
277 };
278
279 static struct rtable *rt_cache_get_first(struct seq_file *seq)
280 {
281 struct rt_cache_iter_state *st = seq->private;
282 struct rtable *r = NULL;
283
284 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
285 if (!rt_hash_table[st->bucket].chain)
286 continue;
287 rcu_read_lock_bh();
288 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
289 while (r) {
290 if (dev_net(r->u.dst.dev) == seq_file_net(seq) &&
291 r->rt_genid == st->genid)
292 return r;
293 r = rcu_dereference_bh(r->u.dst.rt_next);
294 }
295 rcu_read_unlock_bh();
296 }
297 return r;
298 }
299
300 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
301 struct rtable *r)
302 {
303 struct rt_cache_iter_state *st = seq->private;
304
305 r = r->u.dst.rt_next;
306 while (!r) {
307 rcu_read_unlock_bh();
308 do {
309 if (--st->bucket < 0)
310 return NULL;
311 } while (!rt_hash_table[st->bucket].chain);
312 rcu_read_lock_bh();
313 r = rt_hash_table[st->bucket].chain;
314 }
315 return rcu_dereference_bh(r);
316 }
317
318 static struct rtable *rt_cache_get_next(struct seq_file *seq,
319 struct rtable *r)
320 {
321 struct rt_cache_iter_state *st = seq->private;
322 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
323 if (dev_net(r->u.dst.dev) != seq_file_net(seq))
324 continue;
325 if (r->rt_genid == st->genid)
326 break;
327 }
328 return r;
329 }
330
331 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
332 {
333 struct rtable *r = rt_cache_get_first(seq);
334
335 if (r)
336 while (pos && (r = rt_cache_get_next(seq, r)))
337 --pos;
338 return pos ? NULL : r;
339 }
340
341 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
342 {
343 struct rt_cache_iter_state *st = seq->private;
344 if (*pos)
345 return rt_cache_get_idx(seq, *pos - 1);
346 st->genid = rt_genid(seq_file_net(seq));
347 return SEQ_START_TOKEN;
348 }
349
350 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
351 {
352 struct rtable *r;
353
354 if (v == SEQ_START_TOKEN)
355 r = rt_cache_get_first(seq);
356 else
357 r = rt_cache_get_next(seq, v);
358 ++*pos;
359 return r;
360 }
361
362 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
363 {
364 if (v && v != SEQ_START_TOKEN)
365 rcu_read_unlock_bh();
366 }
367
368 static int rt_cache_seq_show(struct seq_file *seq, void *v)
369 {
370 if (v == SEQ_START_TOKEN)
371 seq_printf(seq, "%-127s\n",
372 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
373 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
374 "HHUptod\tSpecDst");
375 else {
376 struct rtable *r = v;
377 int len;
378
379 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
380 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
381 r->u.dst.dev ? r->u.dst.dev->name : "*",
382 (__force u32)r->rt_dst,
383 (__force u32)r->rt_gateway,
384 r->rt_flags, atomic_read(&r->u.dst.__refcnt),
385 r->u.dst.__use, 0, (__force u32)r->rt_src,
386 (dst_metric(&r->u.dst, RTAX_ADVMSS) ?
387 (int)dst_metric(&r->u.dst, RTAX_ADVMSS) + 40 : 0),
388 dst_metric(&r->u.dst, RTAX_WINDOW),
389 (int)((dst_metric(&r->u.dst, RTAX_RTT) >> 3) +
390 dst_metric(&r->u.dst, RTAX_RTTVAR)),
391 r->fl.fl4_tos,
392 r->u.dst.hh ? atomic_read(&r->u.dst.hh->hh_refcnt) : -1,
393 r->u.dst.hh ? (r->u.dst.hh->hh_output ==
394 dev_queue_xmit) : 0,
395 r->rt_spec_dst, &len);
396
397 seq_printf(seq, "%*s\n", 127 - len, "");
398 }
399 return 0;
400 }
401
402 static const struct seq_operations rt_cache_seq_ops = {
403 .start = rt_cache_seq_start,
404 .next = rt_cache_seq_next,
405 .stop = rt_cache_seq_stop,
406 .show = rt_cache_seq_show,
407 };
408
409 static int rt_cache_seq_open(struct inode *inode, struct file *file)
410 {
411 return seq_open_net(inode, file, &rt_cache_seq_ops,
412 sizeof(struct rt_cache_iter_state));
413 }
414
415 static const struct file_operations rt_cache_seq_fops = {
416 .owner = THIS_MODULE,
417 .open = rt_cache_seq_open,
418 .read = seq_read,
419 .llseek = seq_lseek,
420 .release = seq_release_net,
421 };
422
423
424 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
425 {
426 int cpu;
427
428 if (*pos == 0)
429 return SEQ_START_TOKEN;
430
431 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
432 if (!cpu_possible(cpu))
433 continue;
434 *pos = cpu+1;
435 return &per_cpu(rt_cache_stat, cpu);
436 }
437 return NULL;
438 }
439
440 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
441 {
442 int cpu;
443
444 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
445 if (!cpu_possible(cpu))
446 continue;
447 *pos = cpu+1;
448 return &per_cpu(rt_cache_stat, cpu);
449 }
450 return NULL;
451
452 }
453
454 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
455 {
456
457 }
458
459 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
460 {
461 struct rt_cache_stat *st = v;
462
463 if (v == SEQ_START_TOKEN) {
464 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");
465 return 0;
466 }
467
468 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
469 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
470 atomic_read(&ipv4_dst_ops.entries),
471 st->in_hit,
472 st->in_slow_tot,
473 st->in_slow_mc,
474 st->in_no_route,
475 st->in_brd,
476 st->in_martian_dst,
477 st->in_martian_src,
478
479 st->out_hit,
480 st->out_slow_tot,
481 st->out_slow_mc,
482
483 st->gc_total,
484 st->gc_ignored,
485 st->gc_goal_miss,
486 st->gc_dst_overflow,
487 st->in_hlist_search,
488 st->out_hlist_search
489 );
490 return 0;
491 }
492
493 static const struct seq_operations rt_cpu_seq_ops = {
494 .start = rt_cpu_seq_start,
495 .next = rt_cpu_seq_next,
496 .stop = rt_cpu_seq_stop,
497 .show = rt_cpu_seq_show,
498 };
499
500
501 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
502 {
503 return seq_open(file, &rt_cpu_seq_ops);
504 }
505
506 static const struct file_operations rt_cpu_seq_fops = {
507 .owner = THIS_MODULE,
508 .open = rt_cpu_seq_open,
509 .read = seq_read,
510 .llseek = seq_lseek,
511 .release = seq_release,
512 };
513
514 #ifdef CONFIG_NET_CLS_ROUTE
515 static int rt_acct_proc_show(struct seq_file *m, void *v)
516 {
517 struct ip_rt_acct *dst, *src;
518 unsigned int i, j;
519
520 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
521 if (!dst)
522 return -ENOMEM;
523
524 for_each_possible_cpu(i) {
525 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
526 for (j = 0; j < 256; j++) {
527 dst[j].o_bytes += src[j].o_bytes;
528 dst[j].o_packets += src[j].o_packets;
529 dst[j].i_bytes += src[j].i_bytes;
530 dst[j].i_packets += src[j].i_packets;
531 }
532 }
533
534 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
535 kfree(dst);
536 return 0;
537 }
538
539 static int rt_acct_proc_open(struct inode *inode, struct file *file)
540 {
541 return single_open(file, rt_acct_proc_show, NULL);
542 }
543
544 static const struct file_operations rt_acct_proc_fops = {
545 .owner = THIS_MODULE,
546 .open = rt_acct_proc_open,
547 .read = seq_read,
548 .llseek = seq_lseek,
549 .release = single_release,
550 };
551 #endif
552
553 static int __net_init ip_rt_do_proc_init(struct net *net)
554 {
555 struct proc_dir_entry *pde;
556
557 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
558 &rt_cache_seq_fops);
559 if (!pde)
560 goto err1;
561
562 pde = proc_create("rt_cache", S_IRUGO,
563 net->proc_net_stat, &rt_cpu_seq_fops);
564 if (!pde)
565 goto err2;
566
567 #ifdef CONFIG_NET_CLS_ROUTE
568 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
569 if (!pde)
570 goto err3;
571 #endif
572 return 0;
573
574 #ifdef CONFIG_NET_CLS_ROUTE
575 err3:
576 remove_proc_entry("rt_cache", net->proc_net_stat);
577 #endif
578 err2:
579 remove_proc_entry("rt_cache", net->proc_net);
580 err1:
581 return -ENOMEM;
582 }
583
584 static void __net_exit ip_rt_do_proc_exit(struct net *net)
585 {
586 remove_proc_entry("rt_cache", net->proc_net_stat);
587 remove_proc_entry("rt_cache", net->proc_net);
588 #ifdef CONFIG_NET_CLS_ROUTE
589 remove_proc_entry("rt_acct", net->proc_net);
590 #endif
591 }
592
593 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
594 .init = ip_rt_do_proc_init,
595 .exit = ip_rt_do_proc_exit,
596 };
597
598 static int __init ip_rt_proc_init(void)
599 {
600 return register_pernet_subsys(&ip_rt_proc_ops);
601 }
602
603 #else
604 static inline int ip_rt_proc_init(void)
605 {
606 return 0;
607 }
608 #endif /* CONFIG_PROC_FS */
609
610 static inline void rt_free(struct rtable *rt)
611 {
612 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
613 }
614
615 static inline void rt_drop(struct rtable *rt)
616 {
617 ip_rt_put(rt);
618 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
619 }
620
621 static inline int rt_fast_clean(struct rtable *rth)
622 {
623 /* Kill broadcast/multicast entries very aggresively, if they
624 collide in hash table with more useful entries */
625 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
626 rth->fl.iif && rth->u.dst.rt_next;
627 }
628
629 static inline int rt_valuable(struct rtable *rth)
630 {
631 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
632 rth->u.dst.expires;
633 }
634
635 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
636 {
637 unsigned long age;
638 int ret = 0;
639
640 if (atomic_read(&rth->u.dst.__refcnt))
641 goto out;
642
643 ret = 1;
644 if (rth->u.dst.expires &&
645 time_after_eq(jiffies, rth->u.dst.expires))
646 goto out;
647
648 age = jiffies - rth->u.dst.lastuse;
649 ret = 0;
650 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
651 (age <= tmo2 && rt_valuable(rth)))
652 goto out;
653 ret = 1;
654 out: return ret;
655 }
656
657 /* Bits of score are:
658 * 31: very valuable
659 * 30: not quite useless
660 * 29..0: usage counter
661 */
662 static inline u32 rt_score(struct rtable *rt)
663 {
664 u32 score = jiffies - rt->u.dst.lastuse;
665
666 score = ~score & ~(3<<30);
667
668 if (rt_valuable(rt))
669 score |= (1<<31);
670
671 if (!rt->fl.iif ||
672 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
673 score |= (1<<30);
674
675 return score;
676 }
677
678 static inline bool rt_caching(const struct net *net)
679 {
680 return net->ipv4.current_rt_cache_rebuild_count <=
681 net->ipv4.sysctl_rt_cache_rebuild_count;
682 }
683
684 static inline bool compare_hash_inputs(const struct flowi *fl1,
685 const struct flowi *fl2)
686 {
687 return ((((__force u32)fl1->nl_u.ip4_u.daddr ^ (__force u32)fl2->nl_u.ip4_u.daddr) |
688 ((__force u32)fl1->nl_u.ip4_u.saddr ^ (__force u32)fl2->nl_u.ip4_u.saddr) |
689 (fl1->iif ^ fl2->iif)) == 0);
690 }
691
692 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
693 {
694 return (((__force u32)fl1->nl_u.ip4_u.daddr ^ (__force u32)fl2->nl_u.ip4_u.daddr) |
695 ((__force u32)fl1->nl_u.ip4_u.saddr ^ (__force u32)fl2->nl_u.ip4_u.saddr) |
696 (fl1->mark ^ fl2->mark) |
697 (*(u16 *)&fl1->nl_u.ip4_u.tos ^ *(u16 *)&fl2->nl_u.ip4_u.tos) |
698 (fl1->oif ^ fl2->oif) |
699 (fl1->iif ^ fl2->iif)) == 0;
700 }
701
702 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
703 {
704 return net_eq(dev_net(rt1->u.dst.dev), dev_net(rt2->u.dst.dev));
705 }
706
707 static inline int rt_is_expired(struct rtable *rth)
708 {
709 return rth->rt_genid != rt_genid(dev_net(rth->u.dst.dev));
710 }
711
712 /*
713 * Perform a full scan of hash table and free all entries.
714 * Can be called by a softirq or a process.
715 * In the later case, we want to be reschedule if necessary
716 */
717 static void rt_do_flush(int process_context)
718 {
719 unsigned int i;
720 struct rtable *rth, *next;
721 struct rtable * tail;
722
723 for (i = 0; i <= rt_hash_mask; i++) {
724 if (process_context && need_resched())
725 cond_resched();
726 rth = rt_hash_table[i].chain;
727 if (!rth)
728 continue;
729
730 spin_lock_bh(rt_hash_lock_addr(i));
731 #ifdef CONFIG_NET_NS
732 {
733 struct rtable ** prev, * p;
734
735 rth = rt_hash_table[i].chain;
736
737 /* defer releasing the head of the list after spin_unlock */
738 for (tail = rth; tail; tail = tail->u.dst.rt_next)
739 if (!rt_is_expired(tail))
740 break;
741 if (rth != tail)
742 rt_hash_table[i].chain = tail;
743
744 /* call rt_free on entries after the tail requiring flush */
745 prev = &rt_hash_table[i].chain;
746 for (p = *prev; p; p = next) {
747 next = p->u.dst.rt_next;
748 if (!rt_is_expired(p)) {
749 prev = &p->u.dst.rt_next;
750 } else {
751 *prev = next;
752 rt_free(p);
753 }
754 }
755 }
756 #else
757 rth = rt_hash_table[i].chain;
758 rt_hash_table[i].chain = NULL;
759 tail = NULL;
760 #endif
761 spin_unlock_bh(rt_hash_lock_addr(i));
762
763 for (; rth != tail; rth = next) {
764 next = rth->u.dst.rt_next;
765 rt_free(rth);
766 }
767 }
768 }
769
770 /*
771 * While freeing expired entries, we compute average chain length
772 * and standard deviation, using fixed-point arithmetic.
773 * This to have an estimation of rt_chain_length_max
774 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
775 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
776 */
777
778 #define FRACT_BITS 3
779 #define ONE (1UL << FRACT_BITS)
780
781 /*
782 * Given a hash chain and an item in this hash chain,
783 * find if a previous entry has the same hash_inputs
784 * (but differs on tos, mark or oif)
785 * Returns 0 if an alias is found.
786 * Returns ONE if rth has no alias before itself.
787 */
788 static int has_noalias(const struct rtable *head, const struct rtable *rth)
789 {
790 const struct rtable *aux = head;
791
792 while (aux != rth) {
793 if (compare_hash_inputs(&aux->fl, &rth->fl))
794 return 0;
795 aux = aux->u.dst.rt_next;
796 }
797 return ONE;
798 }
799
800 static void rt_check_expire(void)
801 {
802 static unsigned int rover;
803 unsigned int i = rover, goal;
804 struct rtable *rth, **rthp;
805 unsigned long samples = 0;
806 unsigned long sum = 0, sum2 = 0;
807 unsigned long delta;
808 u64 mult;
809
810 delta = jiffies - expires_ljiffies;
811 expires_ljiffies = jiffies;
812 mult = ((u64)delta) << rt_hash_log;
813 if (ip_rt_gc_timeout > 1)
814 do_div(mult, ip_rt_gc_timeout);
815 goal = (unsigned int)mult;
816 if (goal > rt_hash_mask)
817 goal = rt_hash_mask + 1;
818 for (; goal > 0; goal--) {
819 unsigned long tmo = ip_rt_gc_timeout;
820 unsigned long length;
821
822 i = (i + 1) & rt_hash_mask;
823 rthp = &rt_hash_table[i].chain;
824
825 if (need_resched())
826 cond_resched();
827
828 samples++;
829
830 if (*rthp == NULL)
831 continue;
832 length = 0;
833 spin_lock_bh(rt_hash_lock_addr(i));
834 while ((rth = *rthp) != NULL) {
835 prefetch(rth->u.dst.rt_next);
836 if (rt_is_expired(rth)) {
837 *rthp = rth->u.dst.rt_next;
838 rt_free(rth);
839 continue;
840 }
841 if (rth->u.dst.expires) {
842 /* Entry is expired even if it is in use */
843 if (time_before_eq(jiffies, rth->u.dst.expires)) {
844 nofree:
845 tmo >>= 1;
846 rthp = &rth->u.dst.rt_next;
847 /*
848 * We only count entries on
849 * a chain with equal hash inputs once
850 * so that entries for different QOS
851 * levels, and other non-hash input
852 * attributes don't unfairly skew
853 * the length computation
854 */
855 length += has_noalias(rt_hash_table[i].chain, rth);
856 continue;
857 }
858 } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout))
859 goto nofree;
860
861 /* Cleanup aged off entries. */
862 *rthp = rth->u.dst.rt_next;
863 rt_free(rth);
864 }
865 spin_unlock_bh(rt_hash_lock_addr(i));
866 sum += length;
867 sum2 += length*length;
868 }
869 if (samples) {
870 unsigned long avg = sum / samples;
871 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
872 rt_chain_length_max = max_t(unsigned long,
873 ip_rt_gc_elasticity,
874 (avg + 4*sd) >> FRACT_BITS);
875 }
876 rover = i;
877 }
878
879 /*
880 * rt_worker_func() is run in process context.
881 * we call rt_check_expire() to scan part of the hash table
882 */
883 static void rt_worker_func(struct work_struct *work)
884 {
885 rt_check_expire();
886 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
887 }
888
889 /*
890 * Pertubation of rt_genid by a small quantity [1..256]
891 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
892 * many times (2^24) without giving recent rt_genid.
893 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
894 */
895 static void rt_cache_invalidate(struct net *net)
896 {
897 unsigned char shuffle;
898
899 get_random_bytes(&shuffle, sizeof(shuffle));
900 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
901 }
902
903 /*
904 * delay < 0 : invalidate cache (fast : entries will be deleted later)
905 * delay >= 0 : invalidate & flush cache (can be long)
906 */
907 void rt_cache_flush(struct net *net, int delay)
908 {
909 rt_cache_invalidate(net);
910 if (delay >= 0)
911 rt_do_flush(!in_softirq());
912 }
913
914 /* Flush previous cache invalidated entries from the cache */
915 void rt_cache_flush_batch(void)
916 {
917 rt_do_flush(!in_softirq());
918 }
919
920 static void rt_emergency_hash_rebuild(struct net *net)
921 {
922 if (net_ratelimit())
923 printk(KERN_WARNING "Route hash chain too long!\n");
924 rt_cache_invalidate(net);
925 }
926
927 /*
928 Short description of GC goals.
929
930 We want to build algorithm, which will keep routing cache
931 at some equilibrium point, when number of aged off entries
932 is kept approximately equal to newly generated ones.
933
934 Current expiration strength is variable "expire".
935 We try to adjust it dynamically, so that if networking
936 is idle expires is large enough to keep enough of warm entries,
937 and when load increases it reduces to limit cache size.
938 */
939
940 static int rt_garbage_collect(struct dst_ops *ops)
941 {
942 static unsigned long expire = RT_GC_TIMEOUT;
943 static unsigned long last_gc;
944 static int rover;
945 static int equilibrium;
946 struct rtable *rth, **rthp;
947 unsigned long now = jiffies;
948 int goal;
949
950 /*
951 * Garbage collection is pretty expensive,
952 * do not make it too frequently.
953 */
954
955 RT_CACHE_STAT_INC(gc_total);
956
957 if (now - last_gc < ip_rt_gc_min_interval &&
958 atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size) {
959 RT_CACHE_STAT_INC(gc_ignored);
960 goto out;
961 }
962
963 /* Calculate number of entries, which we want to expire now. */
964 goal = atomic_read(&ipv4_dst_ops.entries) -
965 (ip_rt_gc_elasticity << rt_hash_log);
966 if (goal <= 0) {
967 if (equilibrium < ipv4_dst_ops.gc_thresh)
968 equilibrium = ipv4_dst_ops.gc_thresh;
969 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
970 if (goal > 0) {
971 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
972 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
973 }
974 } else {
975 /* We are in dangerous area. Try to reduce cache really
976 * aggressively.
977 */
978 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
979 equilibrium = atomic_read(&ipv4_dst_ops.entries) - goal;
980 }
981
982 if (now - last_gc >= ip_rt_gc_min_interval)
983 last_gc = now;
984
985 if (goal <= 0) {
986 equilibrium += goal;
987 goto work_done;
988 }
989
990 do {
991 int i, k;
992
993 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
994 unsigned long tmo = expire;
995
996 k = (k + 1) & rt_hash_mask;
997 rthp = &rt_hash_table[k].chain;
998 spin_lock_bh(rt_hash_lock_addr(k));
999 while ((rth = *rthp) != NULL) {
1000 if (!rt_is_expired(rth) &&
1001 !rt_may_expire(rth, tmo, expire)) {
1002 tmo >>= 1;
1003 rthp = &rth->u.dst.rt_next;
1004 continue;
1005 }
1006 *rthp = rth->u.dst.rt_next;
1007 rt_free(rth);
1008 goal--;
1009 }
1010 spin_unlock_bh(rt_hash_lock_addr(k));
1011 if (goal <= 0)
1012 break;
1013 }
1014 rover = k;
1015
1016 if (goal <= 0)
1017 goto work_done;
1018
1019 /* Goal is not achieved. We stop process if:
1020
1021 - if expire reduced to zero. Otherwise, expire is halfed.
1022 - if table is not full.
1023 - if we are called from interrupt.
1024 - jiffies check is just fallback/debug loop breaker.
1025 We will not spin here for long time in any case.
1026 */
1027
1028 RT_CACHE_STAT_INC(gc_goal_miss);
1029
1030 if (expire == 0)
1031 break;
1032
1033 expire >>= 1;
1034 #if RT_CACHE_DEBUG >= 2
1035 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
1036 atomic_read(&ipv4_dst_ops.entries), goal, i);
1037 #endif
1038
1039 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
1040 goto out;
1041 } while (!in_softirq() && time_before_eq(jiffies, now));
1042
1043 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
1044 goto out;
1045 if (net_ratelimit())
1046 printk(KERN_WARNING "dst cache overflow\n");
1047 RT_CACHE_STAT_INC(gc_dst_overflow);
1048 return 1;
1049
1050 work_done:
1051 expire += ip_rt_gc_min_interval;
1052 if (expire > ip_rt_gc_timeout ||
1053 atomic_read(&ipv4_dst_ops.entries) < ipv4_dst_ops.gc_thresh)
1054 expire = ip_rt_gc_timeout;
1055 #if RT_CACHE_DEBUG >= 2
1056 printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
1057 atomic_read(&ipv4_dst_ops.entries), goal, rover);
1058 #endif
1059 out: return 0;
1060 }
1061
1062 /*
1063 * Returns number of entries in a hash chain that have different hash_inputs
1064 */
1065 static int slow_chain_length(const struct rtable *head)
1066 {
1067 int length = 0;
1068 const struct rtable *rth = head;
1069
1070 while (rth) {
1071 length += has_noalias(head, rth);
1072 rth = rth->u.dst.rt_next;
1073 }
1074 return length >> FRACT_BITS;
1075 }
1076
1077 static int rt_intern_hash(unsigned hash, struct rtable *rt,
1078 struct rtable **rp, struct sk_buff *skb, int ifindex)
1079 {
1080 struct rtable *rth, **rthp;
1081 unsigned long now;
1082 struct rtable *cand, **candp;
1083 u32 min_score;
1084 int chain_length;
1085 int attempts = !in_softirq();
1086
1087 restart:
1088 chain_length = 0;
1089 min_score = ~(u32)0;
1090 cand = NULL;
1091 candp = NULL;
1092 now = jiffies;
1093
1094 if (!rt_caching(dev_net(rt->u.dst.dev))) {
1095 /*
1096 * If we're not caching, just tell the caller we
1097 * were successful and don't touch the route. The
1098 * caller hold the sole reference to the cache entry, and
1099 * it will be released when the caller is done with it.
1100 * If we drop it here, the callers have no way to resolve routes
1101 * when we're not caching. Instead, just point *rp at rt, so
1102 * the caller gets a single use out of the route
1103 * Note that we do rt_free on this new route entry, so that
1104 * once its refcount hits zero, we are still able to reap it
1105 * (Thanks Alexey)
1106 * Note also the rt_free uses call_rcu. We don't actually
1107 * need rcu protection here, this is just our path to get
1108 * on the route gc list.
1109 */
1110
1111 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1112 int err = arp_bind_neighbour(&rt->u.dst);
1113 if (err) {
1114 if (net_ratelimit())
1115 printk(KERN_WARNING
1116 "Neighbour table failure & not caching routes.\n");
1117 rt_drop(rt);
1118 return err;
1119 }
1120 }
1121
1122 rt_free(rt);
1123 goto skip_hashing;
1124 }
1125
1126 rthp = &rt_hash_table[hash].chain;
1127
1128 spin_lock_bh(rt_hash_lock_addr(hash));
1129 while ((rth = *rthp) != NULL) {
1130 if (rt_is_expired(rth)) {
1131 *rthp = rth->u.dst.rt_next;
1132 rt_free(rth);
1133 continue;
1134 }
1135 if (compare_keys(&rth->fl, &rt->fl) && compare_netns(rth, rt)) {
1136 /* Put it first */
1137 *rthp = rth->u.dst.rt_next;
1138 /*
1139 * Since lookup is lockfree, the deletion
1140 * must be visible to another weakly ordered CPU before
1141 * the insertion at the start of the hash chain.
1142 */
1143 rcu_assign_pointer(rth->u.dst.rt_next,
1144 rt_hash_table[hash].chain);
1145 /*
1146 * Since lookup is lockfree, the update writes
1147 * must be ordered for consistency on SMP.
1148 */
1149 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1150
1151 dst_use(&rth->u.dst, now);
1152 spin_unlock_bh(rt_hash_lock_addr(hash));
1153
1154 rt_drop(rt);
1155 if (rp)
1156 *rp = rth;
1157 else
1158 skb_dst_set(skb, &rth->u.dst);
1159 return 0;
1160 }
1161
1162 if (!atomic_read(&rth->u.dst.__refcnt)) {
1163 u32 score = rt_score(rth);
1164
1165 if (score <= min_score) {
1166 cand = rth;
1167 candp = rthp;
1168 min_score = score;
1169 }
1170 }
1171
1172 chain_length++;
1173
1174 rthp = &rth->u.dst.rt_next;
1175 }
1176
1177 if (cand) {
1178 /* ip_rt_gc_elasticity used to be average length of chain
1179 * length, when exceeded gc becomes really aggressive.
1180 *
1181 * The second limit is less certain. At the moment it allows
1182 * only 2 entries per bucket. We will see.
1183 */
1184 if (chain_length > ip_rt_gc_elasticity) {
1185 *candp = cand->u.dst.rt_next;
1186 rt_free(cand);
1187 }
1188 } else {
1189 if (chain_length > rt_chain_length_max &&
1190 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1191 struct net *net = dev_net(rt->u.dst.dev);
1192 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1193 if (!rt_caching(net)) {
1194 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1195 rt->u.dst.dev->name, num);
1196 }
1197 rt_emergency_hash_rebuild(net);
1198 spin_unlock_bh(rt_hash_lock_addr(hash));
1199
1200 hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1201 ifindex, rt_genid(net));
1202 goto restart;
1203 }
1204 }
1205
1206 /* Try to bind route to arp only if it is output
1207 route or unicast forwarding path.
1208 */
1209 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1210 int err = arp_bind_neighbour(&rt->u.dst);
1211 if (err) {
1212 spin_unlock_bh(rt_hash_lock_addr(hash));
1213
1214 if (err != -ENOBUFS) {
1215 rt_drop(rt);
1216 return err;
1217 }
1218
1219 /* Neighbour tables are full and nothing
1220 can be released. Try to shrink route cache,
1221 it is most likely it holds some neighbour records.
1222 */
1223 if (attempts-- > 0) {
1224 int saved_elasticity = ip_rt_gc_elasticity;
1225 int saved_int = ip_rt_gc_min_interval;
1226 ip_rt_gc_elasticity = 1;
1227 ip_rt_gc_min_interval = 0;
1228 rt_garbage_collect(&ipv4_dst_ops);
1229 ip_rt_gc_min_interval = saved_int;
1230 ip_rt_gc_elasticity = saved_elasticity;
1231 goto restart;
1232 }
1233
1234 if (net_ratelimit())
1235 printk(KERN_WARNING "Neighbour table overflow.\n");
1236 rt_drop(rt);
1237 return -ENOBUFS;
1238 }
1239 }
1240
1241 rt->u.dst.rt_next = rt_hash_table[hash].chain;
1242
1243 #if RT_CACHE_DEBUG >= 2
1244 if (rt->u.dst.rt_next) {
1245 struct rtable *trt;
1246 printk(KERN_DEBUG "rt_cache @%02x: %pI4",
1247 hash, &rt->rt_dst);
1248 for (trt = rt->u.dst.rt_next; trt; trt = trt->u.dst.rt_next)
1249 printk(" . %pI4", &trt->rt_dst);
1250 printk("\n");
1251 }
1252 #endif
1253 /*
1254 * Since lookup is lockfree, we must make sure
1255 * previous writes to rt are comitted to memory
1256 * before making rt visible to other CPUS.
1257 */
1258 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1259
1260 spin_unlock_bh(rt_hash_lock_addr(hash));
1261
1262 skip_hashing:
1263 if (rp)
1264 *rp = rt;
1265 else
1266 skb_dst_set(skb, &rt->u.dst);
1267 return 0;
1268 }
1269
1270 void rt_bind_peer(struct rtable *rt, int create)
1271 {
1272 static DEFINE_SPINLOCK(rt_peer_lock);
1273 struct inet_peer *peer;
1274
1275 peer = inet_getpeer(rt->rt_dst, create);
1276
1277 spin_lock_bh(&rt_peer_lock);
1278 if (rt->peer == NULL) {
1279 rt->peer = peer;
1280 peer = NULL;
1281 }
1282 spin_unlock_bh(&rt_peer_lock);
1283 if (peer)
1284 inet_putpeer(peer);
1285 }
1286
1287 /*
1288 * Peer allocation may fail only in serious out-of-memory conditions. However
1289 * we still can generate some output.
1290 * Random ID selection looks a bit dangerous because we have no chances to
1291 * select ID being unique in a reasonable period of time.
1292 * But broken packet identifier may be better than no packet at all.
1293 */
1294 static void ip_select_fb_ident(struct iphdr *iph)
1295 {
1296 static DEFINE_SPINLOCK(ip_fb_id_lock);
1297 static u32 ip_fallback_id;
1298 u32 salt;
1299
1300 spin_lock_bh(&ip_fb_id_lock);
1301 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1302 iph->id = htons(salt & 0xFFFF);
1303 ip_fallback_id = salt;
1304 spin_unlock_bh(&ip_fb_id_lock);
1305 }
1306
1307 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1308 {
1309 struct rtable *rt = (struct rtable *) dst;
1310
1311 if (rt) {
1312 if (rt->peer == NULL)
1313 rt_bind_peer(rt, 1);
1314
1315 /* If peer is attached to destination, it is never detached,
1316 so that we need not to grab a lock to dereference it.
1317 */
1318 if (rt->peer) {
1319 iph->id = htons(inet_getid(rt->peer, more));
1320 return;
1321 }
1322 } else
1323 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1324 __builtin_return_address(0));
1325
1326 ip_select_fb_ident(iph);
1327 }
1328
1329 static void rt_del(unsigned hash, struct rtable *rt)
1330 {
1331 struct rtable **rthp, *aux;
1332
1333 rthp = &rt_hash_table[hash].chain;
1334 spin_lock_bh(rt_hash_lock_addr(hash));
1335 ip_rt_put(rt);
1336 while ((aux = *rthp) != NULL) {
1337 if (aux == rt || rt_is_expired(aux)) {
1338 *rthp = aux->u.dst.rt_next;
1339 rt_free(aux);
1340 continue;
1341 }
1342 rthp = &aux->u.dst.rt_next;
1343 }
1344 spin_unlock_bh(rt_hash_lock_addr(hash));
1345 }
1346
1347 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1348 __be32 saddr, struct net_device *dev)
1349 {
1350 int i, k;
1351 struct in_device *in_dev = in_dev_get(dev);
1352 struct rtable *rth, **rthp;
1353 __be32 skeys[2] = { saddr, 0 };
1354 int ikeys[2] = { dev->ifindex, 0 };
1355 struct netevent_redirect netevent;
1356 struct net *net;
1357
1358 if (!in_dev)
1359 return;
1360
1361 net = dev_net(dev);
1362 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1363 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1364 ipv4_is_zeronet(new_gw))
1365 goto reject_redirect;
1366
1367 if (!rt_caching(net))
1368 goto reject_redirect;
1369
1370 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1371 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1372 goto reject_redirect;
1373 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1374 goto reject_redirect;
1375 } else {
1376 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1377 goto reject_redirect;
1378 }
1379
1380 for (i = 0; i < 2; i++) {
1381 for (k = 0; k < 2; k++) {
1382 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1383 rt_genid(net));
1384
1385 rthp=&rt_hash_table[hash].chain;
1386
1387 rcu_read_lock();
1388 while ((rth = rcu_dereference(*rthp)) != NULL) {
1389 struct rtable *rt;
1390
1391 if (rth->fl.fl4_dst != daddr ||
1392 rth->fl.fl4_src != skeys[i] ||
1393 rth->fl.oif != ikeys[k] ||
1394 rth->fl.iif != 0 ||
1395 rt_is_expired(rth) ||
1396 !net_eq(dev_net(rth->u.dst.dev), net)) {
1397 rthp = &rth->u.dst.rt_next;
1398 continue;
1399 }
1400
1401 if (rth->rt_dst != daddr ||
1402 rth->rt_src != saddr ||
1403 rth->u.dst.error ||
1404 rth->rt_gateway != old_gw ||
1405 rth->u.dst.dev != dev)
1406 break;
1407
1408 dst_hold(&rth->u.dst);
1409 rcu_read_unlock();
1410
1411 rt = dst_alloc(&ipv4_dst_ops);
1412 if (rt == NULL) {
1413 ip_rt_put(rth);
1414 in_dev_put(in_dev);
1415 return;
1416 }
1417
1418 /* Copy all the information. */
1419 *rt = *rth;
1420 rt->u.dst.__use = 1;
1421 atomic_set(&rt->u.dst.__refcnt, 1);
1422 rt->u.dst.child = NULL;
1423 if (rt->u.dst.dev)
1424 dev_hold(rt->u.dst.dev);
1425 if (rt->idev)
1426 in_dev_hold(rt->idev);
1427 rt->u.dst.obsolete = -1;
1428 rt->u.dst.lastuse = jiffies;
1429 rt->u.dst.path = &rt->u.dst;
1430 rt->u.dst.neighbour = NULL;
1431 rt->u.dst.hh = NULL;
1432 #ifdef CONFIG_XFRM
1433 rt->u.dst.xfrm = NULL;
1434 #endif
1435 rt->rt_genid = rt_genid(net);
1436 rt->rt_flags |= RTCF_REDIRECTED;
1437
1438 /* Gateway is different ... */
1439 rt->rt_gateway = new_gw;
1440
1441 /* Redirect received -> path was valid */
1442 dst_confirm(&rth->u.dst);
1443
1444 if (rt->peer)
1445 atomic_inc(&rt->peer->refcnt);
1446
1447 if (arp_bind_neighbour(&rt->u.dst) ||
1448 !(rt->u.dst.neighbour->nud_state &
1449 NUD_VALID)) {
1450 if (rt->u.dst.neighbour)
1451 neigh_event_send(rt->u.dst.neighbour, NULL);
1452 ip_rt_put(rth);
1453 rt_drop(rt);
1454 goto do_next;
1455 }
1456
1457 netevent.old = &rth->u.dst;
1458 netevent.new = &rt->u.dst;
1459 call_netevent_notifiers(NETEVENT_REDIRECT,
1460 &netevent);
1461
1462 rt_del(hash, rth);
1463 if (!rt_intern_hash(hash, rt, &rt, NULL, rt->fl.oif))
1464 ip_rt_put(rt);
1465 goto do_next;
1466 }
1467 rcu_read_unlock();
1468 do_next:
1469 ;
1470 }
1471 }
1472 in_dev_put(in_dev);
1473 return;
1474
1475 reject_redirect:
1476 #ifdef CONFIG_IP_ROUTE_VERBOSE
1477 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1478 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1479 " Advised path = %pI4 -> %pI4\n",
1480 &old_gw, dev->name, &new_gw,
1481 &saddr, &daddr);
1482 #endif
1483 in_dev_put(in_dev);
1484 }
1485
1486 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1487 {
1488 struct rtable *rt = (struct rtable *)dst;
1489 struct dst_entry *ret = dst;
1490
1491 if (rt) {
1492 if (dst->obsolete > 0) {
1493 ip_rt_put(rt);
1494 ret = NULL;
1495 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1496 (rt->u.dst.expires &&
1497 time_after_eq(jiffies, rt->u.dst.expires))) {
1498 unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1499 rt->fl.oif,
1500 rt_genid(dev_net(dst->dev)));
1501 #if RT_CACHE_DEBUG >= 1
1502 printk(KERN_DEBUG "ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1503 &rt->rt_dst, rt->fl.fl4_tos);
1504 #endif
1505 rt_del(hash, rt);
1506 ret = NULL;
1507 }
1508 }
1509 return ret;
1510 }
1511
1512 /*
1513 * Algorithm:
1514 * 1. The first ip_rt_redirect_number redirects are sent
1515 * with exponential backoff, then we stop sending them at all,
1516 * assuming that the host ignores our redirects.
1517 * 2. If we did not see packets requiring redirects
1518 * during ip_rt_redirect_silence, we assume that the host
1519 * forgot redirected route and start to send redirects again.
1520 *
1521 * This algorithm is much cheaper and more intelligent than dumb load limiting
1522 * in icmp.c.
1523 *
1524 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1525 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1526 */
1527
1528 void ip_rt_send_redirect(struct sk_buff *skb)
1529 {
1530 struct rtable *rt = skb_rtable(skb);
1531 struct in_device *in_dev;
1532 int log_martians;
1533
1534 rcu_read_lock();
1535 in_dev = __in_dev_get_rcu(rt->u.dst.dev);
1536 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1537 rcu_read_unlock();
1538 return;
1539 }
1540 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1541 rcu_read_unlock();
1542
1543 /* No redirected packets during ip_rt_redirect_silence;
1544 * reset the algorithm.
1545 */
1546 if (time_after(jiffies, rt->u.dst.rate_last + ip_rt_redirect_silence))
1547 rt->u.dst.rate_tokens = 0;
1548
1549 /* Too many ignored redirects; do not send anything
1550 * set u.dst.rate_last to the last seen redirected packet.
1551 */
1552 if (rt->u.dst.rate_tokens >= ip_rt_redirect_number) {
1553 rt->u.dst.rate_last = jiffies;
1554 return;
1555 }
1556
1557 /* Check for load limit; set rate_last to the latest sent
1558 * redirect.
1559 */
1560 if (rt->u.dst.rate_tokens == 0 ||
1561 time_after(jiffies,
1562 (rt->u.dst.rate_last +
1563 (ip_rt_redirect_load << rt->u.dst.rate_tokens)))) {
1564 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1565 rt->u.dst.rate_last = jiffies;
1566 ++rt->u.dst.rate_tokens;
1567 #ifdef CONFIG_IP_ROUTE_VERBOSE
1568 if (log_martians &&
1569 rt->u.dst.rate_tokens == ip_rt_redirect_number &&
1570 net_ratelimit())
1571 printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1572 &rt->rt_src, rt->rt_iif,
1573 &rt->rt_dst, &rt->rt_gateway);
1574 #endif
1575 }
1576 }
1577
1578 static int ip_error(struct sk_buff *skb)
1579 {
1580 struct rtable *rt = skb_rtable(skb);
1581 unsigned long now;
1582 int code;
1583
1584 switch (rt->u.dst.error) {
1585 case EINVAL:
1586 default:
1587 goto out;
1588 case EHOSTUNREACH:
1589 code = ICMP_HOST_UNREACH;
1590 break;
1591 case ENETUNREACH:
1592 code = ICMP_NET_UNREACH;
1593 IP_INC_STATS_BH(dev_net(rt->u.dst.dev),
1594 IPSTATS_MIB_INNOROUTES);
1595 break;
1596 case EACCES:
1597 code = ICMP_PKT_FILTERED;
1598 break;
1599 }
1600
1601 now = jiffies;
1602 rt->u.dst.rate_tokens += now - rt->u.dst.rate_last;
1603 if (rt->u.dst.rate_tokens > ip_rt_error_burst)
1604 rt->u.dst.rate_tokens = ip_rt_error_burst;
1605 rt->u.dst.rate_last = now;
1606 if (rt->u.dst.rate_tokens >= ip_rt_error_cost) {
1607 rt->u.dst.rate_tokens -= ip_rt_error_cost;
1608 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1609 }
1610
1611 out: kfree_skb(skb);
1612 return 0;
1613 }
1614
1615 /*
1616 * The last two values are not from the RFC but
1617 * are needed for AMPRnet AX.25 paths.
1618 */
1619
1620 static const unsigned short mtu_plateau[] =
1621 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1622
1623 static inline unsigned short guess_mtu(unsigned short old_mtu)
1624 {
1625 int i;
1626
1627 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1628 if (old_mtu > mtu_plateau[i])
1629 return mtu_plateau[i];
1630 return 68;
1631 }
1632
1633 unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1634 unsigned short new_mtu,
1635 struct net_device *dev)
1636 {
1637 int i, k;
1638 unsigned short old_mtu = ntohs(iph->tot_len);
1639 struct rtable *rth;
1640 int ikeys[2] = { dev->ifindex, 0 };
1641 __be32 skeys[2] = { iph->saddr, 0, };
1642 __be32 daddr = iph->daddr;
1643 unsigned short est_mtu = 0;
1644
1645 for (k = 0; k < 2; k++) {
1646 for (i = 0; i < 2; i++) {
1647 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1648 rt_genid(net));
1649
1650 rcu_read_lock();
1651 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1652 rth = rcu_dereference(rth->u.dst.rt_next)) {
1653 unsigned short mtu = new_mtu;
1654
1655 if (rth->fl.fl4_dst != daddr ||
1656 rth->fl.fl4_src != skeys[i] ||
1657 rth->rt_dst != daddr ||
1658 rth->rt_src != iph->saddr ||
1659 rth->fl.oif != ikeys[k] ||
1660 rth->fl.iif != 0 ||
1661 dst_metric_locked(&rth->u.dst, RTAX_MTU) ||
1662 !net_eq(dev_net(rth->u.dst.dev), net) ||
1663 rt_is_expired(rth))
1664 continue;
1665
1666 if (new_mtu < 68 || new_mtu >= old_mtu) {
1667
1668 /* BSD 4.2 compatibility hack :-( */
1669 if (mtu == 0 &&
1670 old_mtu >= dst_mtu(&rth->u.dst) &&
1671 old_mtu >= 68 + (iph->ihl << 2))
1672 old_mtu -= iph->ihl << 2;
1673
1674 mtu = guess_mtu(old_mtu);
1675 }
1676 if (mtu <= dst_mtu(&rth->u.dst)) {
1677 if (mtu < dst_mtu(&rth->u.dst)) {
1678 dst_confirm(&rth->u.dst);
1679 if (mtu < ip_rt_min_pmtu) {
1680 mtu = ip_rt_min_pmtu;
1681 rth->u.dst.metrics[RTAX_LOCK-1] |=
1682 (1 << RTAX_MTU);
1683 }
1684 rth->u.dst.metrics[RTAX_MTU-1] = mtu;
1685 dst_set_expires(&rth->u.dst,
1686 ip_rt_mtu_expires);
1687 }
1688 est_mtu = mtu;
1689 }
1690 }
1691 rcu_read_unlock();
1692 }
1693 }
1694 return est_mtu ? : new_mtu;
1695 }
1696
1697 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1698 {
1699 if (dst_mtu(dst) > mtu && mtu >= 68 &&
1700 !(dst_metric_locked(dst, RTAX_MTU))) {
1701 if (mtu < ip_rt_min_pmtu) {
1702 mtu = ip_rt_min_pmtu;
1703 dst->metrics[RTAX_LOCK-1] |= (1 << RTAX_MTU);
1704 }
1705 dst->metrics[RTAX_MTU-1] = mtu;
1706 dst_set_expires(dst, ip_rt_mtu_expires);
1707 call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
1708 }
1709 }
1710
1711 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1712 {
1713 if (rt_is_expired((struct rtable *)dst))
1714 return NULL;
1715 return dst;
1716 }
1717
1718 static void ipv4_dst_destroy(struct dst_entry *dst)
1719 {
1720 struct rtable *rt = (struct rtable *) dst;
1721 struct inet_peer *peer = rt->peer;
1722 struct in_device *idev = rt->idev;
1723
1724 if (peer) {
1725 rt->peer = NULL;
1726 inet_putpeer(peer);
1727 }
1728
1729 if (idev) {
1730 rt->idev = NULL;
1731 in_dev_put(idev);
1732 }
1733 }
1734
1735 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
1736 int how)
1737 {
1738 struct rtable *rt = (struct rtable *) dst;
1739 struct in_device *idev = rt->idev;
1740 if (dev != dev_net(dev)->loopback_dev && idev && idev->dev == dev) {
1741 struct in_device *loopback_idev =
1742 in_dev_get(dev_net(dev)->loopback_dev);
1743 if (loopback_idev) {
1744 rt->idev = loopback_idev;
1745 in_dev_put(idev);
1746 }
1747 }
1748 }
1749
1750 static void ipv4_link_failure(struct sk_buff *skb)
1751 {
1752 struct rtable *rt;
1753
1754 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1755
1756 rt = skb_rtable(skb);
1757 if (rt)
1758 dst_set_expires(&rt->u.dst, 0);
1759 }
1760
1761 static int ip_rt_bug(struct sk_buff *skb)
1762 {
1763 printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1764 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1765 skb->dev ? skb->dev->name : "?");
1766 kfree_skb(skb);
1767 return 0;
1768 }
1769
1770 /*
1771 We do not cache source address of outgoing interface,
1772 because it is used only by IP RR, TS and SRR options,
1773 so that it out of fast path.
1774
1775 BTW remember: "addr" is allowed to be not aligned
1776 in IP options!
1777 */
1778
1779 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1780 {
1781 __be32 src;
1782 struct fib_result res;
1783
1784 if (rt->fl.iif == 0)
1785 src = rt->rt_src;
1786 else if (fib_lookup(dev_net(rt->u.dst.dev), &rt->fl, &res) == 0) {
1787 src = FIB_RES_PREFSRC(res);
1788 fib_res_put(&res);
1789 } else
1790 src = inet_select_addr(rt->u.dst.dev, rt->rt_gateway,
1791 RT_SCOPE_UNIVERSE);
1792 memcpy(addr, &src, 4);
1793 }
1794
1795 #ifdef CONFIG_NET_CLS_ROUTE
1796 static void set_class_tag(struct rtable *rt, u32 tag)
1797 {
1798 if (!(rt->u.dst.tclassid & 0xFFFF))
1799 rt->u.dst.tclassid |= tag & 0xFFFF;
1800 if (!(rt->u.dst.tclassid & 0xFFFF0000))
1801 rt->u.dst.tclassid |= tag & 0xFFFF0000;
1802 }
1803 #endif
1804
1805 static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1806 {
1807 struct fib_info *fi = res->fi;
1808
1809 if (fi) {
1810 if (FIB_RES_GW(*res) &&
1811 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1812 rt->rt_gateway = FIB_RES_GW(*res);
1813 memcpy(rt->u.dst.metrics, fi->fib_metrics,
1814 sizeof(rt->u.dst.metrics));
1815 if (fi->fib_mtu == 0) {
1816 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
1817 if (dst_metric_locked(&rt->u.dst, RTAX_MTU) &&
1818 rt->rt_gateway != rt->rt_dst &&
1819 rt->u.dst.dev->mtu > 576)
1820 rt->u.dst.metrics[RTAX_MTU-1] = 576;
1821 }
1822 #ifdef CONFIG_NET_CLS_ROUTE
1823 rt->u.dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1824 #endif
1825 } else
1826 rt->u.dst.metrics[RTAX_MTU-1]= rt->u.dst.dev->mtu;
1827
1828 if (dst_metric(&rt->u.dst, RTAX_HOPLIMIT) == 0)
1829 rt->u.dst.metrics[RTAX_HOPLIMIT-1] = sysctl_ip_default_ttl;
1830 if (dst_mtu(&rt->u.dst) > IP_MAX_MTU)
1831 rt->u.dst.metrics[RTAX_MTU-1] = IP_MAX_MTU;
1832 if (dst_metric(&rt->u.dst, RTAX_ADVMSS) == 0)
1833 rt->u.dst.metrics[RTAX_ADVMSS-1] = max_t(unsigned int, rt->u.dst.dev->mtu - 40,
1834 ip_rt_min_advmss);
1835 if (dst_metric(&rt->u.dst, RTAX_ADVMSS) > 65535 - 40)
1836 rt->u.dst.metrics[RTAX_ADVMSS-1] = 65535 - 40;
1837
1838 #ifdef CONFIG_NET_CLS_ROUTE
1839 #ifdef CONFIG_IP_MULTIPLE_TABLES
1840 set_class_tag(rt, fib_rules_tclass(res));
1841 #endif
1842 set_class_tag(rt, itag);
1843 #endif
1844 rt->rt_type = res->type;
1845 }
1846
1847 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1848 u8 tos, struct net_device *dev, int our)
1849 {
1850 unsigned hash;
1851 struct rtable *rth;
1852 __be32 spec_dst;
1853 struct in_device *in_dev = in_dev_get(dev);
1854 u32 itag = 0;
1855
1856 /* Primary sanity checks. */
1857
1858 if (in_dev == NULL)
1859 return -EINVAL;
1860
1861 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1862 ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1863 goto e_inval;
1864
1865 if (ipv4_is_zeronet(saddr)) {
1866 if (!ipv4_is_local_multicast(daddr))
1867 goto e_inval;
1868 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1869 } else if (fib_validate_source(saddr, 0, tos, 0,
1870 dev, &spec_dst, &itag, 0) < 0)
1871 goto e_inval;
1872
1873 rth = dst_alloc(&ipv4_dst_ops);
1874 if (!rth)
1875 goto e_nobufs;
1876
1877 rth->u.dst.output = ip_rt_bug;
1878 rth->u.dst.obsolete = -1;
1879
1880 atomic_set(&rth->u.dst.__refcnt, 1);
1881 rth->u.dst.flags= DST_HOST;
1882 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1883 rth->u.dst.flags |= DST_NOPOLICY;
1884 rth->fl.fl4_dst = daddr;
1885 rth->rt_dst = daddr;
1886 rth->fl.fl4_tos = tos;
1887 rth->fl.mark = skb->mark;
1888 rth->fl.fl4_src = saddr;
1889 rth->rt_src = saddr;
1890 #ifdef CONFIG_NET_CLS_ROUTE
1891 rth->u.dst.tclassid = itag;
1892 #endif
1893 rth->rt_iif =
1894 rth->fl.iif = dev->ifindex;
1895 rth->u.dst.dev = init_net.loopback_dev;
1896 dev_hold(rth->u.dst.dev);
1897 rth->idev = in_dev_get(rth->u.dst.dev);
1898 rth->fl.oif = 0;
1899 rth->rt_gateway = daddr;
1900 rth->rt_spec_dst= spec_dst;
1901 rth->rt_genid = rt_genid(dev_net(dev));
1902 rth->rt_flags = RTCF_MULTICAST;
1903 rth->rt_type = RTN_MULTICAST;
1904 if (our) {
1905 rth->u.dst.input= ip_local_deliver;
1906 rth->rt_flags |= RTCF_LOCAL;
1907 }
1908
1909 #ifdef CONFIG_IP_MROUTE
1910 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1911 rth->u.dst.input = ip_mr_input;
1912 #endif
1913 RT_CACHE_STAT_INC(in_slow_mc);
1914
1915 in_dev_put(in_dev);
1916 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1917 return rt_intern_hash(hash, rth, NULL, skb, dev->ifindex);
1918
1919 e_nobufs:
1920 in_dev_put(in_dev);
1921 return -ENOBUFS;
1922
1923 e_inval:
1924 in_dev_put(in_dev);
1925 return -EINVAL;
1926 }
1927
1928
1929 static void ip_handle_martian_source(struct net_device *dev,
1930 struct in_device *in_dev,
1931 struct sk_buff *skb,
1932 __be32 daddr,
1933 __be32 saddr)
1934 {
1935 RT_CACHE_STAT_INC(in_martian_src);
1936 #ifdef CONFIG_IP_ROUTE_VERBOSE
1937 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1938 /*
1939 * RFC1812 recommendation, if source is martian,
1940 * the only hint is MAC header.
1941 */
1942 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1943 &daddr, &saddr, dev->name);
1944 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1945 int i;
1946 const unsigned char *p = skb_mac_header(skb);
1947 printk(KERN_WARNING "ll header: ");
1948 for (i = 0; i < dev->hard_header_len; i++, p++) {
1949 printk("%02x", *p);
1950 if (i < (dev->hard_header_len - 1))
1951 printk(":");
1952 }
1953 printk("\n");
1954 }
1955 }
1956 #endif
1957 }
1958
1959 static int __mkroute_input(struct sk_buff *skb,
1960 struct fib_result *res,
1961 struct in_device *in_dev,
1962 __be32 daddr, __be32 saddr, u32 tos,
1963 struct rtable **result)
1964 {
1965
1966 struct rtable *rth;
1967 int err;
1968 struct in_device *out_dev;
1969 unsigned flags = 0;
1970 __be32 spec_dst;
1971 u32 itag;
1972
1973 /* get a working reference to the output device */
1974 out_dev = in_dev_get(FIB_RES_DEV(*res));
1975 if (out_dev == NULL) {
1976 if (net_ratelimit())
1977 printk(KERN_CRIT "Bug in ip_route_input" \
1978 "_slow(). Please, report\n");
1979 return -EINVAL;
1980 }
1981
1982
1983 err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1984 in_dev->dev, &spec_dst, &itag, skb->mark);
1985 if (err < 0) {
1986 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1987 saddr);
1988
1989 err = -EINVAL;
1990 goto cleanup;
1991 }
1992
1993 if (err)
1994 flags |= RTCF_DIRECTSRC;
1995
1996 if (out_dev == in_dev && err &&
1997 (IN_DEV_SHARED_MEDIA(out_dev) ||
1998 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1999 flags |= RTCF_DOREDIRECT;
2000
2001 if (skb->protocol != htons(ETH_P_IP)) {
2002 /* Not IP (i.e. ARP). Do not create route, if it is
2003 * invalid for proxy arp. DNAT routes are always valid.
2004 *
2005 * Proxy arp feature have been extended to allow, ARP
2006 * replies back to the same interface, to support
2007 * Private VLAN switch technologies. See arp.c.
2008 */
2009 if (out_dev == in_dev &&
2010 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2011 err = -EINVAL;
2012 goto cleanup;
2013 }
2014 }
2015
2016
2017 rth = dst_alloc(&ipv4_dst_ops);
2018 if (!rth) {
2019 err = -ENOBUFS;
2020 goto cleanup;
2021 }
2022
2023 atomic_set(&rth->u.dst.__refcnt, 1);
2024 rth->u.dst.flags= DST_HOST;
2025 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2026 rth->u.dst.flags |= DST_NOPOLICY;
2027 if (IN_DEV_CONF_GET(out_dev, NOXFRM))
2028 rth->u.dst.flags |= DST_NOXFRM;
2029 rth->fl.fl4_dst = daddr;
2030 rth->rt_dst = daddr;
2031 rth->fl.fl4_tos = tos;
2032 rth->fl.mark = skb->mark;
2033 rth->fl.fl4_src = saddr;
2034 rth->rt_src = saddr;
2035 rth->rt_gateway = daddr;
2036 rth->rt_iif =
2037 rth->fl.iif = in_dev->dev->ifindex;
2038 rth->u.dst.dev = (out_dev)->dev;
2039 dev_hold(rth->u.dst.dev);
2040 rth->idev = in_dev_get(rth->u.dst.dev);
2041 rth->fl.oif = 0;
2042 rth->rt_spec_dst= spec_dst;
2043
2044 rth->u.dst.obsolete = -1;
2045 rth->u.dst.input = ip_forward;
2046 rth->u.dst.output = ip_output;
2047 rth->rt_genid = rt_genid(dev_net(rth->u.dst.dev));
2048
2049 rt_set_nexthop(rth, res, itag);
2050
2051 rth->rt_flags = flags;
2052
2053 *result = rth;
2054 err = 0;
2055 cleanup:
2056 /* release the working reference to the output device */
2057 in_dev_put(out_dev);
2058 return err;
2059 }
2060
2061 static int ip_mkroute_input(struct sk_buff *skb,
2062 struct fib_result *res,
2063 const struct flowi *fl,
2064 struct in_device *in_dev,
2065 __be32 daddr, __be32 saddr, u32 tos)
2066 {
2067 struct rtable* rth = NULL;
2068 int err;
2069 unsigned hash;
2070
2071 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2072 if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
2073 fib_select_multipath(fl, res);
2074 #endif
2075
2076 /* create a routing cache entry */
2077 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2078 if (err)
2079 return err;
2080
2081 /* put it into the cache */
2082 hash = rt_hash(daddr, saddr, fl->iif,
2083 rt_genid(dev_net(rth->u.dst.dev)));
2084 return rt_intern_hash(hash, rth, NULL, skb, fl->iif);
2085 }
2086
2087 /*
2088 * NOTE. We drop all the packets that has local source
2089 * addresses, because every properly looped back packet
2090 * must have correct destination already attached by output routine.
2091 *
2092 * Such approach solves two big problems:
2093 * 1. Not simplex devices are handled properly.
2094 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2095 */
2096
2097 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2098 u8 tos, struct net_device *dev)
2099 {
2100 struct fib_result res;
2101 struct in_device *in_dev = in_dev_get(dev);
2102 struct flowi fl = { .nl_u = { .ip4_u =
2103 { .daddr = daddr,
2104 .saddr = saddr,
2105 .tos = tos,
2106 .scope = RT_SCOPE_UNIVERSE,
2107 } },
2108 .mark = skb->mark,
2109 .iif = dev->ifindex };
2110 unsigned flags = 0;
2111 u32 itag = 0;
2112 struct rtable * rth;
2113 unsigned hash;
2114 __be32 spec_dst;
2115 int err = -EINVAL;
2116 int free_res = 0;
2117 struct net * net = dev_net(dev);
2118
2119 /* IP on this device is disabled. */
2120
2121 if (!in_dev)
2122 goto out;
2123
2124 /* Check for the most weird martians, which can be not detected
2125 by fib_lookup.
2126 */
2127
2128 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2129 ipv4_is_loopback(saddr))
2130 goto martian_source;
2131
2132 if (daddr == htonl(0xFFFFFFFF) || (saddr == 0 && daddr == 0))
2133 goto brd_input;
2134
2135 /* Accept zero addresses only to limited broadcast;
2136 * I even do not know to fix it or not. Waiting for complains :-)
2137 */
2138 if (ipv4_is_zeronet(saddr))
2139 goto martian_source;
2140
2141 if (ipv4_is_lbcast(daddr) || ipv4_is_zeronet(daddr) ||
2142 ipv4_is_loopback(daddr))
2143 goto martian_destination;
2144
2145 /*
2146 * Now we are ready to route packet.
2147 */
2148 if ((err = fib_lookup(net, &fl, &res)) != 0) {
2149 if (!IN_DEV_FORWARD(in_dev))
2150 goto e_hostunreach;
2151 goto no_route;
2152 }
2153 free_res = 1;
2154
2155 RT_CACHE_STAT_INC(in_slow_tot);
2156
2157 if (res.type == RTN_BROADCAST)
2158 goto brd_input;
2159
2160 if (res.type == RTN_LOCAL) {
2161 int result;
2162 result = fib_validate_source(saddr, daddr, tos,
2163 net->loopback_dev->ifindex,
2164 dev, &spec_dst, &itag, skb->mark);
2165 if (result < 0)
2166 goto martian_source;
2167 if (result)
2168 flags |= RTCF_DIRECTSRC;
2169 spec_dst = daddr;
2170 goto local_input;
2171 }
2172
2173 if (!IN_DEV_FORWARD(in_dev))
2174 goto e_hostunreach;
2175 if (res.type != RTN_UNICAST)
2176 goto martian_destination;
2177
2178 err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
2179 done:
2180 in_dev_put(in_dev);
2181 if (free_res)
2182 fib_res_put(&res);
2183 out: return err;
2184
2185 brd_input:
2186 if (skb->protocol != htons(ETH_P_IP))
2187 goto e_inval;
2188
2189 if (ipv4_is_zeronet(saddr))
2190 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2191 else {
2192 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
2193 &itag, skb->mark);
2194 if (err < 0)
2195 goto martian_source;
2196 if (err)
2197 flags |= RTCF_DIRECTSRC;
2198 }
2199 flags |= RTCF_BROADCAST;
2200 res.type = RTN_BROADCAST;
2201 RT_CACHE_STAT_INC(in_brd);
2202
2203 local_input:
2204 rth = dst_alloc(&ipv4_dst_ops);
2205 if (!rth)
2206 goto e_nobufs;
2207
2208 rth->u.dst.output= ip_rt_bug;
2209 rth->u.dst.obsolete = -1;
2210 rth->rt_genid = rt_genid(net);
2211
2212 atomic_set(&rth->u.dst.__refcnt, 1);
2213 rth->u.dst.flags= DST_HOST;
2214 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2215 rth->u.dst.flags |= DST_NOPOLICY;
2216 rth->fl.fl4_dst = daddr;
2217 rth->rt_dst = daddr;
2218 rth->fl.fl4_tos = tos;
2219 rth->fl.mark = skb->mark;
2220 rth->fl.fl4_src = saddr;
2221 rth->rt_src = saddr;
2222 #ifdef CONFIG_NET_CLS_ROUTE
2223 rth->u.dst.tclassid = itag;
2224 #endif
2225 rth->rt_iif =
2226 rth->fl.iif = dev->ifindex;
2227 rth->u.dst.dev = net->loopback_dev;
2228 dev_hold(rth->u.dst.dev);
2229 rth->idev = in_dev_get(rth->u.dst.dev);
2230 rth->rt_gateway = daddr;
2231 rth->rt_spec_dst= spec_dst;
2232 rth->u.dst.input= ip_local_deliver;
2233 rth->rt_flags = flags|RTCF_LOCAL;
2234 if (res.type == RTN_UNREACHABLE) {
2235 rth->u.dst.input= ip_error;
2236 rth->u.dst.error= -err;
2237 rth->rt_flags &= ~RTCF_LOCAL;
2238 }
2239 rth->rt_type = res.type;
2240 hash = rt_hash(daddr, saddr, fl.iif, rt_genid(net));
2241 err = rt_intern_hash(hash, rth, NULL, skb, fl.iif);
2242 goto done;
2243
2244 no_route:
2245 RT_CACHE_STAT_INC(in_no_route);
2246 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2247 res.type = RTN_UNREACHABLE;
2248 if (err == -ESRCH)
2249 err = -ENETUNREACH;
2250 goto local_input;
2251
2252 /*
2253 * Do not cache martian addresses: they should be logged (RFC1812)
2254 */
2255 martian_destination:
2256 RT_CACHE_STAT_INC(in_martian_dst);
2257 #ifdef CONFIG_IP_ROUTE_VERBOSE
2258 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2259 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2260 &daddr, &saddr, dev->name);
2261 #endif
2262
2263 e_hostunreach:
2264 err = -EHOSTUNREACH;
2265 goto done;
2266
2267 e_inval:
2268 err = -EINVAL;
2269 goto done;
2270
2271 e_nobufs:
2272 err = -ENOBUFS;
2273 goto done;
2274
2275 martian_source:
2276 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2277 goto e_inval;
2278 }
2279
2280 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2281 u8 tos, struct net_device *dev, bool noref)
2282 {
2283 struct rtable * rth;
2284 unsigned hash;
2285 int iif = dev->ifindex;
2286 struct net *net;
2287
2288 net = dev_net(dev);
2289
2290 if (!rt_caching(net))
2291 goto skip_cache;
2292
2293 tos &= IPTOS_RT_MASK;
2294 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2295
2296 rcu_read_lock();
2297 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2298 rth = rcu_dereference(rth->u.dst.rt_next)) {
2299 if ((((__force u32)rth->fl.fl4_dst ^ (__force u32)daddr) |
2300 ((__force u32)rth->fl.fl4_src ^ (__force u32)saddr) |
2301 (rth->fl.iif ^ iif) |
2302 rth->fl.oif |
2303 (rth->fl.fl4_tos ^ tos)) == 0 &&
2304 rth->fl.mark == skb->mark &&
2305 net_eq(dev_net(rth->u.dst.dev), net) &&
2306 !rt_is_expired(rth)) {
2307 if (noref) {
2308 dst_use_noref(&rth->u.dst, jiffies);
2309 skb_dst_set_noref(skb, &rth->u.dst);
2310 } else {
2311 dst_use(&rth->u.dst, jiffies);
2312 skb_dst_set(skb, &rth->u.dst);
2313 }
2314 RT_CACHE_STAT_INC(in_hit);
2315 rcu_read_unlock();
2316 return 0;
2317 }
2318 RT_CACHE_STAT_INC(in_hlist_search);
2319 }
2320 rcu_read_unlock();
2321
2322 skip_cache:
2323 /* Multicast recognition logic is moved from route cache to here.
2324 The problem was that too many Ethernet cards have broken/missing
2325 hardware multicast filters :-( As result the host on multicasting
2326 network acquires a lot of useless route cache entries, sort of
2327 SDR messages from all the world. Now we try to get rid of them.
2328 Really, provided software IP multicast filter is organized
2329 reasonably (at least, hashed), it does not result in a slowdown
2330 comparing with route cache reject entries.
2331 Note, that multicast routers are not affected, because
2332 route cache entry is created eventually.
2333 */
2334 if (ipv4_is_multicast(daddr)) {
2335 struct in_device *in_dev;
2336
2337 rcu_read_lock();
2338 if ((in_dev = __in_dev_get_rcu(dev)) != NULL) {
2339 int our = ip_check_mc(in_dev, daddr, saddr,
2340 ip_hdr(skb)->protocol);
2341 if (our
2342 #ifdef CONFIG_IP_MROUTE
2343 ||
2344 (!ipv4_is_local_multicast(daddr) &&
2345 IN_DEV_MFORWARD(in_dev))
2346 #endif
2347 ) {
2348 rcu_read_unlock();
2349 return ip_route_input_mc(skb, daddr, saddr,
2350 tos, dev, our);
2351 }
2352 }
2353 rcu_read_unlock();
2354 return -EINVAL;
2355 }
2356 return ip_route_input_slow(skb, daddr, saddr, tos, dev);
2357 }
2358 EXPORT_SYMBOL(ip_route_input_common);
2359
2360 static int __mkroute_output(struct rtable **result,
2361 struct fib_result *res,
2362 const struct flowi *fl,
2363 const struct flowi *oldflp,
2364 struct net_device *dev_out,
2365 unsigned flags)
2366 {
2367 struct rtable *rth;
2368 struct in_device *in_dev;
2369 u32 tos = RT_FL_TOS(oldflp);
2370 int err = 0;
2371
2372 if (ipv4_is_loopback(fl->fl4_src) && !(dev_out->flags&IFF_LOOPBACK))
2373 return -EINVAL;
2374
2375 if (fl->fl4_dst == htonl(0xFFFFFFFF))
2376 res->type = RTN_BROADCAST;
2377 else if (ipv4_is_multicast(fl->fl4_dst))
2378 res->type = RTN_MULTICAST;
2379 else if (ipv4_is_lbcast(fl->fl4_dst) || ipv4_is_zeronet(fl->fl4_dst))
2380 return -EINVAL;
2381
2382 if (dev_out->flags & IFF_LOOPBACK)
2383 flags |= RTCF_LOCAL;
2384
2385 /* get work reference to inet device */
2386 in_dev = in_dev_get(dev_out);
2387 if (!in_dev)
2388 return -EINVAL;
2389
2390 if (res->type == RTN_BROADCAST) {
2391 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2392 if (res->fi) {
2393 fib_info_put(res->fi);
2394 res->fi = NULL;
2395 }
2396 } else if (res->type == RTN_MULTICAST) {
2397 flags |= RTCF_MULTICAST|RTCF_LOCAL;
2398 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2399 oldflp->proto))
2400 flags &= ~RTCF_LOCAL;
2401 /* If multicast route do not exist use
2402 default one, but do not gateway in this case.
2403 Yes, it is hack.
2404 */
2405 if (res->fi && res->prefixlen < 4) {
2406 fib_info_put(res->fi);
2407 res->fi = NULL;
2408 }
2409 }
2410
2411
2412 rth = dst_alloc(&ipv4_dst_ops);
2413 if (!rth) {
2414 err = -ENOBUFS;
2415 goto cleanup;
2416 }
2417
2418 atomic_set(&rth->u.dst.__refcnt, 1);
2419 rth->u.dst.flags= DST_HOST;
2420 if (IN_DEV_CONF_GET(in_dev, NOXFRM))
2421 rth->u.dst.flags |= DST_NOXFRM;
2422 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2423 rth->u.dst.flags |= DST_NOPOLICY;
2424
2425 rth->fl.fl4_dst = oldflp->fl4_dst;
2426 rth->fl.fl4_tos = tos;
2427 rth->fl.fl4_src = oldflp->fl4_src;
2428 rth->fl.oif = oldflp->oif;
2429 rth->fl.mark = oldflp->mark;
2430 rth->rt_dst = fl->fl4_dst;
2431 rth->rt_src = fl->fl4_src;
2432 rth->rt_iif = oldflp->oif ? : dev_out->ifindex;
2433 /* get references to the devices that are to be hold by the routing
2434 cache entry */
2435 rth->u.dst.dev = dev_out;
2436 dev_hold(dev_out);
2437 rth->idev = in_dev_get(dev_out);
2438 rth->rt_gateway = fl->fl4_dst;
2439 rth->rt_spec_dst= fl->fl4_src;
2440
2441 rth->u.dst.output=ip_output;
2442 rth->u.dst.obsolete = -1;
2443 rth->rt_genid = rt_genid(dev_net(dev_out));
2444
2445 RT_CACHE_STAT_INC(out_slow_tot);
2446
2447 if (flags & RTCF_LOCAL) {
2448 rth->u.dst.input = ip_local_deliver;
2449 rth->rt_spec_dst = fl->fl4_dst;
2450 }
2451 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2452 rth->rt_spec_dst = fl->fl4_src;
2453 if (flags & RTCF_LOCAL &&
2454 !(dev_out->flags & IFF_LOOPBACK)) {
2455 rth->u.dst.output = ip_mc_output;
2456 RT_CACHE_STAT_INC(out_slow_mc);
2457 }
2458 #ifdef CONFIG_IP_MROUTE
2459 if (res->type == RTN_MULTICAST) {
2460 if (IN_DEV_MFORWARD(in_dev) &&
2461 !ipv4_is_local_multicast(oldflp->fl4_dst)) {
2462 rth->u.dst.input = ip_mr_input;
2463 rth->u.dst.output = ip_mc_output;
2464 }
2465 }
2466 #endif
2467 }
2468
2469 rt_set_nexthop(rth, res, 0);
2470
2471 rth->rt_flags = flags;
2472
2473 *result = rth;
2474 cleanup:
2475 /* release work reference to inet device */
2476 in_dev_put(in_dev);
2477
2478 return err;
2479 }
2480
2481 static int ip_mkroute_output(struct rtable **rp,
2482 struct fib_result *res,
2483 const struct flowi *fl,
2484 const struct flowi *oldflp,
2485 struct net_device *dev_out,
2486 unsigned flags)
2487 {
2488 struct rtable *rth = NULL;
2489 int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2490 unsigned hash;
2491 if (err == 0) {
2492 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif,
2493 rt_genid(dev_net(dev_out)));
2494 err = rt_intern_hash(hash, rth, rp, NULL, oldflp->oif);
2495 }
2496
2497 return err;
2498 }
2499
2500 /*
2501 * Major route resolver routine.
2502 */
2503
2504 static int ip_route_output_slow(struct net *net, struct rtable **rp,
2505 const struct flowi *oldflp)
2506 {
2507 u32 tos = RT_FL_TOS(oldflp);
2508 struct flowi fl = { .nl_u = { .ip4_u =
2509 { .daddr = oldflp->fl4_dst,
2510 .saddr = oldflp->fl4_src,
2511 .tos = tos & IPTOS_RT_MASK,
2512 .scope = ((tos & RTO_ONLINK) ?
2513 RT_SCOPE_LINK :
2514 RT_SCOPE_UNIVERSE),
2515 } },
2516 .mark = oldflp->mark,
2517 .iif = net->loopback_dev->ifindex,
2518 .oif = oldflp->oif };
2519 struct fib_result res;
2520 unsigned flags = 0;
2521 struct net_device *dev_out = NULL;
2522 int free_res = 0;
2523 int err;
2524
2525
2526 res.fi = NULL;
2527 #ifdef CONFIG_IP_MULTIPLE_TABLES
2528 res.r = NULL;
2529 #endif
2530
2531 if (oldflp->fl4_src) {
2532 err = -EINVAL;
2533 if (ipv4_is_multicast(oldflp->fl4_src) ||
2534 ipv4_is_lbcast(oldflp->fl4_src) ||
2535 ipv4_is_zeronet(oldflp->fl4_src))
2536 goto out;
2537
2538 /* I removed check for oif == dev_out->oif here.
2539 It was wrong for two reasons:
2540 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2541 is assigned to multiple interfaces.
2542 2. Moreover, we are allowed to send packets with saddr
2543 of another iface. --ANK
2544 */
2545
2546 if (oldflp->oif == 0 &&
2547 (ipv4_is_multicast(oldflp->fl4_dst) ||
2548 oldflp->fl4_dst == htonl(0xFFFFFFFF))) {
2549 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2550 dev_out = ip_dev_find(net, oldflp->fl4_src);
2551 if (dev_out == NULL)
2552 goto out;
2553
2554 /* Special hack: user can direct multicasts
2555 and limited broadcast via necessary interface
2556 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2557 This hack is not just for fun, it allows
2558 vic,vat and friends to work.
2559 They bind socket to loopback, set ttl to zero
2560 and expect that it will work.
2561 From the viewpoint of routing cache they are broken,
2562 because we are not allowed to build multicast path
2563 with loopback source addr (look, routing cache
2564 cannot know, that ttl is zero, so that packet
2565 will not leave this host and route is valid).
2566 Luckily, this hack is good workaround.
2567 */
2568
2569 fl.oif = dev_out->ifindex;
2570 goto make_route;
2571 }
2572
2573 if (!(oldflp->flags & FLOWI_FLAG_ANYSRC)) {
2574 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2575 dev_out = ip_dev_find(net, oldflp->fl4_src);
2576 if (dev_out == NULL)
2577 goto out;
2578 dev_put(dev_out);
2579 dev_out = NULL;
2580 }
2581 }
2582
2583
2584 if (oldflp->oif) {
2585 dev_out = dev_get_by_index(net, oldflp->oif);
2586 err = -ENODEV;
2587 if (dev_out == NULL)
2588 goto out;
2589
2590 /* RACE: Check return value of inet_select_addr instead. */
2591 if (__in_dev_get_rtnl(dev_out) == NULL) {
2592 dev_put(dev_out);
2593 goto out; /* Wrong error code */
2594 }
2595
2596 if (ipv4_is_local_multicast(oldflp->fl4_dst) ||
2597 oldflp->fl4_dst == htonl(0xFFFFFFFF)) {
2598 if (!fl.fl4_src)
2599 fl.fl4_src = inet_select_addr(dev_out, 0,
2600 RT_SCOPE_LINK);
2601 goto make_route;
2602 }
2603 if (!fl.fl4_src) {
2604 if (ipv4_is_multicast(oldflp->fl4_dst))
2605 fl.fl4_src = inet_select_addr(dev_out, 0,
2606 fl.fl4_scope);
2607 else if (!oldflp->fl4_dst)
2608 fl.fl4_src = inet_select_addr(dev_out, 0,
2609 RT_SCOPE_HOST);
2610 }
2611 }
2612
2613 if (!fl.fl4_dst) {
2614 fl.fl4_dst = fl.fl4_src;
2615 if (!fl.fl4_dst)
2616 fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2617 if (dev_out)
2618 dev_put(dev_out);
2619 dev_out = net->loopback_dev;
2620 dev_hold(dev_out);
2621 fl.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, &fl, &res)) {
2628 res.fi = NULL;
2629 if (oldflp->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 (fl.fl4_src == 0)
2649 fl.fl4_src = inet_select_addr(dev_out, 0,
2650 RT_SCOPE_LINK);
2651 res.type = RTN_UNICAST;
2652 goto make_route;
2653 }
2654 if (dev_out)
2655 dev_put(dev_out);
2656 err = -ENETUNREACH;
2657 goto out;
2658 }
2659 free_res = 1;
2660
2661 if (res.type == RTN_LOCAL) {
2662 if (!fl.fl4_src)
2663 fl.fl4_src = fl.fl4_dst;
2664 if (dev_out)
2665 dev_put(dev_out);
2666 dev_out = net->loopback_dev;
2667 dev_hold(dev_out);
2668 fl.oif = dev_out->ifindex;
2669 if (res.fi)
2670 fib_info_put(res.fi);
2671 res.fi = NULL;
2672 flags |= RTCF_LOCAL;
2673 goto make_route;
2674 }
2675
2676 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2677 if (res.fi->fib_nhs > 1 && fl.oif == 0)
2678 fib_select_multipath(&fl, &res);
2679 else
2680 #endif
2681 if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2682 fib_select_default(net, &fl, &res);
2683
2684 if (!fl.fl4_src)
2685 fl.fl4_src = FIB_RES_PREFSRC(res);
2686
2687 if (dev_out)
2688 dev_put(dev_out);
2689 dev_out = FIB_RES_DEV(res);
2690 dev_hold(dev_out);
2691 fl.oif = dev_out->ifindex;
2692
2693
2694 make_route:
2695 err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2696
2697
2698 if (free_res)
2699 fib_res_put(&res);
2700 if (dev_out)
2701 dev_put(dev_out);
2702 out: return err;
2703 }
2704
2705 int __ip_route_output_key(struct net *net, struct rtable **rp,
2706 const struct flowi *flp)
2707 {
2708 unsigned hash;
2709 struct rtable *rth;
2710
2711 if (!rt_caching(net))
2712 goto slow_output;
2713
2714 hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif, rt_genid(net));
2715
2716 rcu_read_lock_bh();
2717 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2718 rth = rcu_dereference_bh(rth->u.dst.rt_next)) {
2719 if (rth->fl.fl4_dst == flp->fl4_dst &&
2720 rth->fl.fl4_src == flp->fl4_src &&
2721 rth->fl.iif == 0 &&
2722 rth->fl.oif == flp->oif &&
2723 rth->fl.mark == flp->mark &&
2724 !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2725 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2726 net_eq(dev_net(rth->u.dst.dev), net) &&
2727 !rt_is_expired(rth)) {
2728 dst_use(&rth->u.dst, jiffies);
2729 RT_CACHE_STAT_INC(out_hit);
2730 rcu_read_unlock_bh();
2731 *rp = rth;
2732 return 0;
2733 }
2734 RT_CACHE_STAT_INC(out_hlist_search);
2735 }
2736 rcu_read_unlock_bh();
2737
2738 slow_output:
2739 return ip_route_output_slow(net, rp, flp);
2740 }
2741
2742 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2743
2744 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2745 {
2746 }
2747
2748 static struct dst_ops ipv4_dst_blackhole_ops = {
2749 .family = AF_INET,
2750 .protocol = cpu_to_be16(ETH_P_IP),
2751 .destroy = ipv4_dst_destroy,
2752 .check = ipv4_dst_check,
2753 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2754 .entries = ATOMIC_INIT(0),
2755 };
2756
2757
2758 static int ipv4_dst_blackhole(struct net *net, struct rtable **rp, struct flowi *flp)
2759 {
2760 struct rtable *ort = *rp;
2761 struct rtable *rt = (struct rtable *)
2762 dst_alloc(&ipv4_dst_blackhole_ops);
2763
2764 if (rt) {
2765 struct dst_entry *new = &rt->u.dst;
2766
2767 atomic_set(&new->__refcnt, 1);
2768 new->__use = 1;
2769 new->input = dst_discard;
2770 new->output = dst_discard;
2771 memcpy(new->metrics, ort->u.dst.metrics, RTAX_MAX*sizeof(u32));
2772
2773 new->dev = ort->u.dst.dev;
2774 if (new->dev)
2775 dev_hold(new->dev);
2776
2777 rt->fl = ort->fl;
2778
2779 rt->idev = ort->idev;
2780 if (rt->idev)
2781 in_dev_hold(rt->idev);
2782 rt->rt_genid = rt_genid(net);
2783 rt->rt_flags = ort->rt_flags;
2784 rt->rt_type = ort->rt_type;
2785 rt->rt_dst = ort->rt_dst;
2786 rt->rt_src = ort->rt_src;
2787 rt->rt_iif = ort->rt_iif;
2788 rt->rt_gateway = ort->rt_gateway;
2789 rt->rt_spec_dst = ort->rt_spec_dst;
2790 rt->peer = ort->peer;
2791 if (rt->peer)
2792 atomic_inc(&rt->peer->refcnt);
2793
2794 dst_free(new);
2795 }
2796
2797 dst_release(&(*rp)->u.dst);
2798 *rp = rt;
2799 return (rt ? 0 : -ENOMEM);
2800 }
2801
2802 int ip_route_output_flow(struct net *net, struct rtable **rp, struct flowi *flp,
2803 struct sock *sk, int flags)
2804 {
2805 int err;
2806
2807 if ((err = __ip_route_output_key(net, rp, flp)) != 0)
2808 return err;
2809
2810 if (flp->proto) {
2811 if (!flp->fl4_src)
2812 flp->fl4_src = (*rp)->rt_src;
2813 if (!flp->fl4_dst)
2814 flp->fl4_dst = (*rp)->rt_dst;
2815 err = __xfrm_lookup(net, (struct dst_entry **)rp, flp, sk,
2816 flags ? XFRM_LOOKUP_WAIT : 0);
2817 if (err == -EREMOTE)
2818 err = ipv4_dst_blackhole(net, rp, flp);
2819
2820 return err;
2821 }
2822
2823 return 0;
2824 }
2825
2826 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2827
2828 int ip_route_output_key(struct net *net, struct rtable **rp, struct flowi *flp)
2829 {
2830 return ip_route_output_flow(net, rp, flp, NULL, 0);
2831 }
2832
2833 static int rt_fill_info(struct net *net,
2834 struct sk_buff *skb, u32 pid, u32 seq, int event,
2835 int nowait, unsigned int flags)
2836 {
2837 struct rtable *rt = skb_rtable(skb);
2838 struct rtmsg *r;
2839 struct nlmsghdr *nlh;
2840 long expires;
2841 u32 id = 0, ts = 0, tsage = 0, error;
2842
2843 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2844 if (nlh == NULL)
2845 return -EMSGSIZE;
2846
2847 r = nlmsg_data(nlh);
2848 r->rtm_family = AF_INET;
2849 r->rtm_dst_len = 32;
2850 r->rtm_src_len = 0;
2851 r->rtm_tos = rt->fl.fl4_tos;
2852 r->rtm_table = RT_TABLE_MAIN;
2853 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2854 r->rtm_type = rt->rt_type;
2855 r->rtm_scope = RT_SCOPE_UNIVERSE;
2856 r->rtm_protocol = RTPROT_UNSPEC;
2857 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2858 if (rt->rt_flags & RTCF_NOTIFY)
2859 r->rtm_flags |= RTM_F_NOTIFY;
2860
2861 NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2862
2863 if (rt->fl.fl4_src) {
2864 r->rtm_src_len = 32;
2865 NLA_PUT_BE32(skb, RTA_SRC, rt->fl.fl4_src);
2866 }
2867 if (rt->u.dst.dev)
2868 NLA_PUT_U32(skb, RTA_OIF, rt->u.dst.dev->ifindex);
2869 #ifdef CONFIG_NET_CLS_ROUTE
2870 if (rt->u.dst.tclassid)
2871 NLA_PUT_U32(skb, RTA_FLOW, rt->u.dst.tclassid);
2872 #endif
2873 if (rt->fl.iif)
2874 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2875 else if (rt->rt_src != rt->fl.fl4_src)
2876 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2877
2878 if (rt->rt_dst != rt->rt_gateway)
2879 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2880
2881 if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
2882 goto nla_put_failure;
2883
2884 error = rt->u.dst.error;
2885 expires = rt->u.dst.expires ? rt->u.dst.expires - jiffies : 0;
2886 if (rt->peer) {
2887 id = atomic_read(&rt->peer->ip_id_count) & 0xffff;
2888 if (rt->peer->tcp_ts_stamp) {
2889 ts = rt->peer->tcp_ts;
2890 tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2891 }
2892 }
2893
2894 if (rt->fl.iif) {
2895 #ifdef CONFIG_IP_MROUTE
2896 __be32 dst = rt->rt_dst;
2897
2898 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2899 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2900 int err = ipmr_get_route(net, skb, r, nowait);
2901 if (err <= 0) {
2902 if (!nowait) {
2903 if (err == 0)
2904 return 0;
2905 goto nla_put_failure;
2906 } else {
2907 if (err == -EMSGSIZE)
2908 goto nla_put_failure;
2909 error = err;
2910 }
2911 }
2912 } else
2913 #endif
2914 NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2915 }
2916
2917 if (rtnl_put_cacheinfo(skb, &rt->u.dst, id, ts, tsage,
2918 expires, error) < 0)
2919 goto nla_put_failure;
2920
2921 return nlmsg_end(skb, nlh);
2922
2923 nla_put_failure:
2924 nlmsg_cancel(skb, nlh);
2925 return -EMSGSIZE;
2926 }
2927
2928 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2929 {
2930 struct net *net = sock_net(in_skb->sk);
2931 struct rtmsg *rtm;
2932 struct nlattr *tb[RTA_MAX+1];
2933 struct rtable *rt = NULL;
2934 __be32 dst = 0;
2935 __be32 src = 0;
2936 u32 iif;
2937 int err;
2938 struct sk_buff *skb;
2939
2940 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2941 if (err < 0)
2942 goto errout;
2943
2944 rtm = nlmsg_data(nlh);
2945
2946 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2947 if (skb == NULL) {
2948 err = -ENOBUFS;
2949 goto errout;
2950 }
2951
2952 /* Reserve room for dummy headers, this skb can pass
2953 through good chunk of routing engine.
2954 */
2955 skb_reset_mac_header(skb);
2956 skb_reset_network_header(skb);
2957
2958 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2959 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2960 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2961
2962 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2963 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2964 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2965
2966 if (iif) {
2967 struct net_device *dev;
2968
2969 dev = __dev_get_by_index(net, iif);
2970 if (dev == NULL) {
2971 err = -ENODEV;
2972 goto errout_free;
2973 }
2974
2975 skb->protocol = htons(ETH_P_IP);
2976 skb->dev = dev;
2977 local_bh_disable();
2978 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2979 local_bh_enable();
2980
2981 rt = skb_rtable(skb);
2982 if (err == 0 && rt->u.dst.error)
2983 err = -rt->u.dst.error;
2984 } else {
2985 struct flowi fl = {
2986 .nl_u = {
2987 .ip4_u = {
2988 .daddr = dst,
2989 .saddr = src,
2990 .tos = rtm->rtm_tos,
2991 },
2992 },
2993 .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2994 };
2995 err = ip_route_output_key(net, &rt, &fl);
2996 }
2997
2998 if (err)
2999 goto errout_free;
3000
3001 skb_dst_set(skb, &rt->u.dst);
3002 if (rtm->rtm_flags & RTM_F_NOTIFY)
3003 rt->rt_flags |= RTCF_NOTIFY;
3004
3005 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3006 RTM_NEWROUTE, 0, 0);
3007 if (err <= 0)
3008 goto errout_free;
3009
3010 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3011 errout:
3012 return err;
3013
3014 errout_free:
3015 kfree_skb(skb);
3016 goto errout;
3017 }
3018
3019 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
3020 {
3021 struct rtable *rt;
3022 int h, s_h;
3023 int idx, s_idx;
3024 struct net *net;
3025
3026 net = sock_net(skb->sk);
3027
3028 s_h = cb->args[0];
3029 if (s_h < 0)
3030 s_h = 0;
3031 s_idx = idx = cb->args[1];
3032 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3033 if (!rt_hash_table[h].chain)
3034 continue;
3035 rcu_read_lock_bh();
3036 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3037 rt = rcu_dereference_bh(rt->u.dst.rt_next), idx++) {
3038 if (!net_eq(dev_net(rt->u.dst.dev), net) || idx < s_idx)
3039 continue;
3040 if (rt_is_expired(rt))
3041 continue;
3042 skb_dst_set_noref(skb, &rt->u.dst);
3043 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3044 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3045 1, NLM_F_MULTI) <= 0) {
3046 skb_dst_drop(skb);
3047 rcu_read_unlock_bh();
3048 goto done;
3049 }
3050 skb_dst_drop(skb);
3051 }
3052 rcu_read_unlock_bh();
3053 }
3054
3055 done:
3056 cb->args[0] = h;
3057 cb->args[1] = idx;
3058 return skb->len;
3059 }
3060
3061 void ip_rt_multicast_event(struct in_device *in_dev)
3062 {
3063 rt_cache_flush(dev_net(in_dev->dev), 0);
3064 }
3065
3066 #ifdef CONFIG_SYSCTL
3067 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3068 void __user *buffer,
3069 size_t *lenp, loff_t *ppos)
3070 {
3071 if (write) {
3072 int flush_delay;
3073 ctl_table ctl;
3074 struct net *net;
3075
3076 memcpy(&ctl, __ctl, sizeof(ctl));
3077 ctl.data = &flush_delay;
3078 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3079
3080 net = (struct net *)__ctl->extra1;
3081 rt_cache_flush(net, flush_delay);
3082 return 0;
3083 }
3084
3085 return -EINVAL;
3086 }
3087
3088 static ctl_table ipv4_route_table[] = {
3089 {
3090 .procname = "gc_thresh",
3091 .data = &ipv4_dst_ops.gc_thresh,
3092 .maxlen = sizeof(int),
3093 .mode = 0644,
3094 .proc_handler = proc_dointvec,
3095 },
3096 {
3097 .procname = "max_size",
3098 .data = &ip_rt_max_size,
3099 .maxlen = sizeof(int),
3100 .mode = 0644,
3101 .proc_handler = proc_dointvec,
3102 },
3103 {
3104 /* Deprecated. Use gc_min_interval_ms */
3105
3106 .procname = "gc_min_interval",
3107 .data = &ip_rt_gc_min_interval,
3108 .maxlen = sizeof(int),
3109 .mode = 0644,
3110 .proc_handler = proc_dointvec_jiffies,
3111 },
3112 {
3113 .procname = "gc_min_interval_ms",
3114 .data = &ip_rt_gc_min_interval,
3115 .maxlen = sizeof(int),
3116 .mode = 0644,
3117 .proc_handler = proc_dointvec_ms_jiffies,
3118 },
3119 {
3120 .procname = "gc_timeout",
3121 .data = &ip_rt_gc_timeout,
3122 .maxlen = sizeof(int),
3123 .mode = 0644,
3124 .proc_handler = proc_dointvec_jiffies,
3125 },
3126 {
3127 .procname = "gc_interval",
3128 .data = &ip_rt_gc_interval,
3129 .maxlen = sizeof(int),
3130 .mode = 0644,
3131 .proc_handler = proc_dointvec_jiffies,
3132 },
3133 {
3134 .procname = "redirect_load",
3135 .data = &ip_rt_redirect_load,
3136 .maxlen = sizeof(int),
3137 .mode = 0644,
3138 .proc_handler = proc_dointvec,
3139 },
3140 {
3141 .procname = "redirect_number",
3142 .data = &ip_rt_redirect_number,
3143 .maxlen = sizeof(int),
3144 .mode = 0644,
3145 .proc_handler = proc_dointvec,
3146 },
3147 {
3148 .procname = "redirect_silence",
3149 .data = &ip_rt_redirect_silence,
3150 .maxlen = sizeof(int),
3151 .mode = 0644,
3152 .proc_handler = proc_dointvec,
3153 },
3154 {
3155 .procname = "error_cost",
3156 .data = &ip_rt_error_cost,
3157 .maxlen = sizeof(int),
3158 .mode = 0644,
3159 .proc_handler = proc_dointvec,
3160 },
3161 {
3162 .procname = "error_burst",
3163 .data = &ip_rt_error_burst,
3164 .maxlen = sizeof(int),
3165 .mode = 0644,
3166 .proc_handler = proc_dointvec,
3167 },
3168 {
3169 .procname = "gc_elasticity",
3170 .data = &ip_rt_gc_elasticity,
3171 .maxlen = sizeof(int),
3172 .mode = 0644,
3173 .proc_handler = proc_dointvec,
3174 },
3175 {
3176 .procname = "mtu_expires",
3177 .data = &ip_rt_mtu_expires,
3178 .maxlen = sizeof(int),
3179 .mode = 0644,
3180 .proc_handler = proc_dointvec_jiffies,
3181 },
3182 {
3183 .procname = "min_pmtu",
3184 .data = &ip_rt_min_pmtu,
3185 .maxlen = sizeof(int),
3186 .mode = 0644,
3187 .proc_handler = proc_dointvec,
3188 },
3189 {
3190 .procname = "min_adv_mss",
3191 .data = &ip_rt_min_advmss,
3192 .maxlen = sizeof(int),
3193 .mode = 0644,
3194 .proc_handler = proc_dointvec,
3195 },
3196 { }
3197 };
3198
3199 static struct ctl_table empty[1];
3200
3201 static struct ctl_table ipv4_skeleton[] =
3202 {
3203 { .procname = "route",
3204 .mode = 0555, .child = ipv4_route_table},
3205 { .procname = "neigh",
3206 .mode = 0555, .child = empty},
3207 { }
3208 };
3209
3210 static __net_initdata struct ctl_path ipv4_path[] = {
3211 { .procname = "net", },
3212 { .procname = "ipv4", },
3213 { },
3214 };
3215
3216 static struct ctl_table ipv4_route_flush_table[] = {
3217 {
3218 .procname = "flush",
3219 .maxlen = sizeof(int),
3220 .mode = 0200,
3221 .proc_handler = ipv4_sysctl_rtcache_flush,
3222 },
3223 { },
3224 };
3225
3226 static __net_initdata struct ctl_path ipv4_route_path[] = {
3227 { .procname = "net", },
3228 { .procname = "ipv4", },
3229 { .procname = "route", },
3230 { },
3231 };
3232
3233 static __net_init int sysctl_route_net_init(struct net *net)
3234 {
3235 struct ctl_table *tbl;
3236
3237 tbl = ipv4_route_flush_table;
3238 if (!net_eq(net, &init_net)) {
3239 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3240 if (tbl == NULL)
3241 goto err_dup;
3242 }
3243 tbl[0].extra1 = net;
3244
3245 net->ipv4.route_hdr =
3246 register_net_sysctl_table(net, ipv4_route_path, tbl);
3247 if (net->ipv4.route_hdr == NULL)
3248 goto err_reg;
3249 return 0;
3250
3251 err_reg:
3252 if (tbl != ipv4_route_flush_table)
3253 kfree(tbl);
3254 err_dup:
3255 return -ENOMEM;
3256 }
3257
3258 static __net_exit void sysctl_route_net_exit(struct net *net)
3259 {
3260 struct ctl_table *tbl;
3261
3262 tbl = net->ipv4.route_hdr->ctl_table_arg;
3263 unregister_net_sysctl_table(net->ipv4.route_hdr);
3264 BUG_ON(tbl == ipv4_route_flush_table);
3265 kfree(tbl);
3266 }
3267
3268 static __net_initdata struct pernet_operations sysctl_route_ops = {
3269 .init = sysctl_route_net_init,
3270 .exit = sysctl_route_net_exit,
3271 };
3272 #endif
3273
3274 static __net_init int rt_genid_init(struct net *net)
3275 {
3276 get_random_bytes(&net->ipv4.rt_genid,
3277 sizeof(net->ipv4.rt_genid));
3278 return 0;
3279 }
3280
3281 static __net_initdata struct pernet_operations rt_genid_ops = {
3282 .init = rt_genid_init,
3283 };
3284
3285
3286 #ifdef CONFIG_NET_CLS_ROUTE
3287 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3288 #endif /* CONFIG_NET_CLS_ROUTE */
3289
3290 static __initdata unsigned long rhash_entries;
3291 static int __init set_rhash_entries(char *str)
3292 {
3293 if (!str)
3294 return 0;
3295 rhash_entries = simple_strtoul(str, &str, 0);
3296 return 1;
3297 }
3298 __setup("rhash_entries=", set_rhash_entries);
3299
3300 int __init ip_rt_init(void)
3301 {
3302 int rc = 0;
3303
3304 #ifdef CONFIG_NET_CLS_ROUTE
3305 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3306 if (!ip_rt_acct)
3307 panic("IP: failed to allocate ip_rt_acct\n");
3308 #endif
3309
3310 ipv4_dst_ops.kmem_cachep =
3311 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3312 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3313
3314 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3315
3316 rt_hash_table = (struct rt_hash_bucket *)
3317 alloc_large_system_hash("IP route cache",
3318 sizeof(struct rt_hash_bucket),
3319 rhash_entries,
3320 (totalram_pages >= 128 * 1024) ?
3321 15 : 17,
3322 0,
3323 &rt_hash_log,
3324 &rt_hash_mask,
3325 rhash_entries ? 0 : 512 * 1024);
3326 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3327 rt_hash_lock_init();
3328
3329 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3330 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3331
3332 devinet_init();
3333 ip_fib_init();
3334
3335 /* All the timers, started at system startup tend
3336 to synchronize. Perturb it a bit.
3337 */
3338 INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3339 expires_ljiffies = jiffies;
3340 schedule_delayed_work(&expires_work,
3341 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3342
3343 if (ip_rt_proc_init())
3344 printk(KERN_ERR "Unable to create route proc files\n");
3345 #ifdef CONFIG_XFRM
3346 xfrm_init();
3347 xfrm4_init(ip_rt_max_size);
3348 #endif
3349 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3350
3351 #ifdef CONFIG_SYSCTL
3352 register_pernet_subsys(&sysctl_route_ops);
3353 #endif
3354 register_pernet_subsys(&rt_genid_ops);
3355 return rc;
3356 }
3357
3358 #ifdef CONFIG_SYSCTL
3359 /*
3360 * We really need to sanitize the damn ipv4 init order, then all
3361 * this nonsense will go away.
3362 */
3363 void __init ip_static_sysctl_init(void)
3364 {
3365 register_sysctl_paths(ipv4_path, ipv4_skeleton);
3366 }
3367 #endif
3368
3369 EXPORT_SYMBOL(__ip_select_ident);
3370 EXPORT_SYMBOL(ip_route_output_key);
CRIS linux kernel tree